INTRODUCTION

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

In anesthetized cat auditory cortex, the overall multi- or single-unit firing rate is little dependent on the AM frequency or click repetition rate (Eggermont 1994, 1998). However, in awake squirrel monkey, most units showed a rate dependence on modulation frequency (Bieser and Müller-Preuss 1996), and in anesthetized cat, anterior field such units were frequent as well (Schreiner and Urbas 1986). The vector strength (VS) and synchronized firing rate are always strongly affected by modulation frequency or click repetition rate (Bieser and Müller-Preuss 1996; Eggermont 1991, 1994, 1998; Kilgard and Merzenich 1999; Lu and Wang 2000; Schreiner and Raggio 1996; Schreiner and Urbas 1986, 1988).

Different modulating waveforms produce different synchronized firing rates and VS (Eggermont 1994; Schreiner and Urbas 1986). This difference can already be noted in the response to the first period of the modulating waveform. Poor responses to the initial stimulus period generally result in poor temporal modulation transfer functions (tMTFs). We have previously modeled tMTFs for periodic click trains purely as the result of synaptic depression and facilitation (Eggermont 1999). However, this model does not well describe tMTFs for AM stimuli without changing the depression time constant and facilitation strength (unpublished results). Thus one would have to invoke stimulus-dependent synaptic mechanisms to explain the results for AM tones and AM noise. An alternative is to assume that the different tMTFs result from differences in the degree of spike synchronization with the modulation waveform. For instance, a sharp transient such as a click or a tone pip will produce better synchronization, i.e., less spike jitter, compared with a low-frequency sinusoidal modulation of a tone. If this difference plays a dominant role in the synchronization of the spikes to subsequent AM in the tone or noise burst, one could normalize for this effect. It would then be possible to separate stimulus waveform effects from, potentially stimulus-invariant, synaptic mechanisms underlying the temporal response properties of auditory cortical neurons.

Factors influencing spike timing and its effect on encoding stimulus periodicity have been addresses previously (Phillips 1989; Phillips and Hall 1990) and suggest that timing accuracy of neurons in primary auditory cortex of the anesthetized cat are sufficient to support spike entrainment to repetition frequencies of 60-100 Hz. However, this conclusion was based on the magnitude of first-spike latency jitter only. Calculation of the VS is based on first and subsequent spikes elicited by the stimulus. In addition, one has to take into account that in calculations of the VS the distribution of those spikes over one modulation (or repetition) period is the determining factor. Even if the distribution of spike jitter remained constant in absolute terms, it would still change the VS for changing modulation frequency. In considering VS calculations for AM and FM stimuli, the varying rate of change of amplitude or carrier frequency with modulation frequency has also to be taken into account. Both first-spike latency, and thus likely spike jitter, and response strength are determined by the rate of change of the sound level (Heil 1997a,b) or potentially the integrated SPL (Heil and Neubauer 2001). Thus a normalization procedure would need to take all these factors into account.

The mean tMTFs for multiunit (MU) recordings in response to click trains were comparable to those for local field potential (LFP) triggers (Eggermont 1998; Eggermont and Smith 1995). For AM stimuli, the LFPs were less pronounced than for click trains. The MU recordings showed on average lower best modulation frequencies (BMFs) and limiting rates than the LFP triggers, suggesting that a more detailed analysis into the origin of these differences is warranted. This could also be the result of differences in onset synchrony for clicks and AM stimuli. LFPs are the compound extracellular representation of synchronized excitatory postsynaptic potentials (EPSP), whereas the resulting MU activity is also determined by inhibition and postactivation suppression. By comparing the tMTFs for LFPs and MU activity recorded on the same electrode, one can, in principle, partition the effect of presynaptic and postsynaptic mechanism on the tMTFs. This could indicate the origin of the rate-limiting process that determines the highest modulation frequencies with significant envelope locked activity.

To study the effect of spike synchronization to the modulating waveform, we used trains of gamma tones, asymmetrical tone pips with a gamma function envelope (Hermes et al. 1982). This asymmetry suggests the use of trains of time-reversed gamma tones that have a slower rate of rise than gamma tones but have the same spectrum. Asymmetrical stimuli, such as ramped and damped sinusoids, have been used in marmoset auditory cortex (Lu et al. 2001) where, as judged by firing rate, most neurons strongly preferred either one or the other type of stimulus.

We will present here a comparison between the LFP and MU responses of neurons in cat primary auditory cortex (AI) to periodic click trains, gamma-tone and time-reversed gamma-tone trains, AM noise, AM tones, and frequency-modulated (FM) tones. In a large number of cases, the response to all six stimuli was obtained for the same neurons. This investigation provides an extension to a previous study (Eggermont 1994) where at most three stimulus types were presented to the same units. We investigate whether cortical neurons are likely to respond to all types of repetitive transients and modulated stimuli and whether a dependence on modulating waveform, or tone or noise carrier, exists. We will show that the observed differences in the tMTFs are to a large extent indeed produced by the different degree of phase locking of the neuronal firings to the first stimulus envelope. A normalization procedure, based on these synchronization differences, unified the tMTFs for all stimuli except clicks and allowed the identification of a stimulus-invariant low-pass temporal filter function that reflects the properties of synaptic depression.

	METHODS

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

The care and the use of animals reported on in this study was approved by the Life and Environmental Sciences Animal Care Committee of the University of Calgary (P88095).

Animal preparation

Cats were premedicated with 0.25 ml/kg body wt of a mixture of 0.2 ml acepromazine (0.25 mg/ml) and 0.8 ml of atropine methyl nitrate (25 mg/ml) subcutaneously. After about one-half hour, they received an intramuscular injection of 30 mg/kg of ketamine (100 mg/ml) and 20 mg/kg of pentobarbital sodium (65 mg/ml). Lidocain (20 mg/ml) was injected subcutaneously and rubbed in gently, then a skin flap was removed and the skull cleared from overlying muscle tissue. A large screw was cemented upside down on the skull with dental acrylic. An 8-mm-diam hole was trephined over the right temporal cortex so as to expose most of AI. The dura was opened, and the brain was covered with light mineral oil. Then the cat was placed in a sound-treated room on a vibration isolation frame, and the head was secured with the screw. Additional acepromazine/atropine mixture was administered every 2 h. Light anesthesia was maintained with intramuscular injections of 2-5 mg · kg¹ · h¹ of ketamine. The wound margins were infused every 2 h with lidocain, and also every 2 h new mineral oil was added if needed. The temperature of the cat was maintained at 37°C. At the end of the experiment, the animals were killed with an overdose of pentobarbital sodium.

Acoustic stimulus presentation

Acoustic stimuli were presented in an anechoic room from a speaker placed 55 cm from the cat's head and 45° from the midline to the left, i.e., in the contralateral hemifield of the cortex recorded from. The sound-treated room was made anechoic for frequencies 625 Hz by covering walls and ceiling with acoustic wedges (Sonex 3-in) and by covering exposed parts of the vibration isolation frame, equipment, and floor with wedge-material as well. Calibration and monitoring of the sound field was done using a B&K (type 4134) microphone placed above the animal's head and facing the loudspeaker. A search stimulus consisting of random-frequency tone pips, noise burst, and clicks was used to locate units.

The characteristic frequency (CF) and tuning curve of the individual neurons or multi-unit clusters were determined with 100-ms total duration (duration at 50% amplitude was 15 ms) tone pips with gamma-shape envelope (gamma tones) presented randomly in frequency once per second (Eggermont 1996). The 81 different frequencies used were equally spaced logarithmically between 625 Hz and 20 kHz (or between 1.25 and 40 kHz) so that 16 frequencies were present per octave. Thus the ratio between subsequent frequencies was 1.0443.

The envelope of the gamma tones is given by

(1)

with t in milliseconds. Figure 1 shows the envelope of the forward and time-reversed gamma tones (/- - -).

View larger version (17K): [in this window] [in a new window]   Fig. 1. Envelope waveforms for gamma tones and time-reversed gamma tones. Both stimuli last 100 ms, but after 50 ms, the gamma tone amplitude has decreased to 0.1% of its maximum value. The effective duration (at half-amplitude) is 15 ms.

After the frequency tuning properties of the MU activity recorded at each electrode were determined, 1-s duration gamma-tone trains (and their time-reversed versions) followed by 2 s of silence were presented 20 times each. In these trains, the 100-ms duration gamma tones were presented with repetition rates of 1, 2, 3, 4, 6, 8, 12, 16, and 20 Hz. The gamma tones were windowed at a duration of the interstimulus interval for the highest three rates.

Subsequently, periodic click trains (1-s duration followed by 2 s of silence), AM noise or AM tone bursts, and FM tone bursts (1-s duration followed by 2 s of silence) were presented. The repetition rates or modulation frequencies (MF) were from 2 to 64 Hz for click trains, 2 to 64 Hz for AM sounds, and 1 to 32 Hz for FM sounds. MFs were selected at logarithmically equal distance with four values per octave, i.e., spaced at a ratio of 1.19. The click trains, AM and FM stimuli thus consisted of 21 different repetition rates or modulation frequencies that were randomly presented. The sequences of 21 click trains, AM noise bursts, AM tone bursts, and FM tone bursts were repeated 10 times, resulting in stimulus ensemble durations of 630 s. The click trains and AM noise sounds were presented at peak intensities of 65 dB SPL. The gamma-tone trains and the AM tone and FM tone sounds were presented at peak intensities of 55 dB SPL. The carrier frequency was equal to the CF of one of the recording sites.

The AM waveform was an exponentially transformed sine wave (Epping and Eggermont 1986b) with a maximum modulation depth of 17.4 dB (i.e., modulation index 0.93) so that the envelope, when expressed in dB, was sinusoidally modulated. The onset and offset of the AM and FM stimuli were linearly ramped from 0 to maximum amplitude over 100 ms. The FM waveform was sinusoidal with a frequency-sweep range between 0.5 octave above and 0.5 octave below center frequency. The modulation always started from 0.5 octave above the center frequency in the downward direction.

Recording and spike separation procedure

Recordings were made between 600 and 1,200 µM below the cortex surface. Two recording methods were used. In the first method, three tungsten microelectrodes (Micro Probe) with impedances between 1.5 and 2.5 M were independently advanced perpendicular to the AI surface using remotely controlled motorized hydraulic microdrives (Trent-Wells Mark III). The electrode signals were amplified using extracellular preamplifiers (Dagan 2400) and filtered between 10 Hz and 3 kHz (Dagan, 6 dB/octave roll-off). Subsequent high-pass filtering at 200 Hz (Kemo VBF8, high-pass, 24 dB/octave) was used to remove local field potentials. For this data set only, the raw electrode signals were also low-pass filtered at 100 Hz to obtain spike-free signals of ongoing LFPs. These LFPs were passed through Schmitt-triggers set at ~2 SDs (i.e., at about 100 µV) below the mean value of the ongoing signal during silence (Eggermont and Smith 1995). The "spikes" of these LFPs were processed in the same way as single-unit spike data.

In the second method, an array of eight electrodes (Frederic Haer) with impedance similar to those for the individual electrodes described in the preceding text was used. The electrodes were arranged in a 4 × 2 configuration with interelectrode distance within rows and columns equal to 0.5 mm. The electrode array was oriented so that all electrodes were touching the cortical surface and then were manually advanced using a Narishige M101 hydraulic microdrive. The signals were amplified 10,000 times using a Frederic Haer HiZx8 set of amplifiers with filter cutoff frequencies set at 300 Hz and 5 kHz.

The amplified signals were processed by a DataWave multichannel data-acquisition system. Spike sorting was done off-line using a semi-automated procedure based on principal component analysis (Eggermont 1990) implemented in MATLAB. The spike times and waveforms were stored. The MU data presented in this paper represent only separated single units. Thus contrary to the common use of the term MU as a cluster of not well separable units, in this analysis, the separable single-unit spike trains extracted from the MU recording that all had similar response properties were added again to a form a MU spike train. Thus the term MU in this study refers strictly to a cluster of separated single units that, because of their regular spike wave form, likely are dominantly from pyramidal cells (Eggermont 1996).

Data analysis

FREQUENCY-TUNING PROPERTIES. To assess frequency-tuning properties, the peak number of action potentials in the poststimulus time histogram (PSTH) over the first 100 ms for each frequency presentation (5-ms bins) was estimated. The counts for three adjacent frequencies were combined, to reduce variability, and divided by number of stimuli and presented as a firing rate per stimulus. This resulted in 27 frequencies covering five octaves so that the final frequency resolution was approximately 0.2 octaves. The results were calculated per stimulus intensity and were combined into a intensity-frequency rate profile from which tuning curves, rate-intensity functions and iso-intensity rate-frequency contours could be derived (Eggermont 1996) using routines implemented in MATLAB. The frequency-tuning curve was defined for a firing rate at 25% of the maximum peak-firing rate. This was typically ~20% above the background firing-rate, but as the latter was dependent on the level of stimulus-induced suppression, the criterion based on peak firing rate was preferred. Using the spike count in 100-ms poststimulus onset was a less sensitive criterion in determining the tuning curve in case of high background activity because the poststimulus suppression of spontaneous activity could entirely compensate for the stimulus induced spikes. The tuning curve bandwidth (BW20) was measured at 20 dB above threshold at CF and expressed in octaves.

TEMPORAL PROPERTIES. The stimulus-following capacity of the neurons for click and gamma-tone repetition-rate and AM/FM frequency was estimated from the modulation transfer functions and autocorrelation histograms using routines written in MATLAB (Fig. 2). The top left part shows the dot display for periodic click trains, with time after click train onset on the horizontal axis and repetition rate (logarithmic scaling) on the vertical axis. The top right part shows the corresponding PSTH. The temporal modulation transfer functions were obtained by Fourier transformation of the period histograms shown at the far bottom left (Eggermont 1991; Epping and Eggermont 1986a,b). They are shown in the bottom middle, for overall number of spikes per train, rate modulation transfer function (rMTF) (top), number of synchronized spikes per train, tMTF (middle), and VS (bottom). Each modulation period was divided into 16 bins, so that the relative timing accuracy of the spikes per period was constant. Only recordings with at least two adjacent MFs with significant (P < 0.001) VS, based on the Rayleigh test (Fisher 1993), were further considered. The BMF was defined as that repetition rate, or modulation frequency, for which the synchronized firing rate was maximal. In this example, the BMF was estimated at 11.28 Hz. The limiting rate was defined as the highest rate at which the response showed visible envelope locking in the PSTH or autocorrelogram. The autocorrelogram shows multiple peaks for most repetition rates, indicating preferred firings at the period of the repetition rate and its multiples. For the highest rate at which there was locking to the period, only one peak is visible at the reciprocal of the limiting rate (45.12 Hz, midway between the 32- and 64-Hz mark). In case of weak responses, the limiting rate was determined on the basis of the response to the highest modulation frequency with significant VS.

View larger version (70K): [in this window] [in a new window]   Fig. 2. Basic method of analysis for temporal response properties. Top left: a multi-unit (comprising 3 single units) raster dot display for periodic click train stimulation. The click repetition rate is shown vertically and ranges between 2 and 64 Hz. The time since click train onset is on the horizontal axis; the click trains last 1 s. For click rates that can be expressed in natural numbers, a click is present at the 1-s mark. Top right: the corresponding poststimulus time (PST) histogram. Bottom right: the population autocorrelogram. The axes are the same as for the dot display and the PST histogram. The autocorrelation function is computed for each single unit individually, and the normalized autocorrelograms are added. The sharp peaks are the result of 1 single unit with very little spike jitter; it is the same unit that produces the short latency responses in the PSTH and dot display. The set of 4 small graphs at the bottom left represent the period histogram, the rate modulation transfer function (rMTF) (number of spikes per click train), the temporal modulation transfer function (tMTF, number of synchronized spikes per click train), and the vector strength (VS, the ratio of tMTF and rMTF).

	RESULTS

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

Data are presented for 174 recording sites from AI in 31 cats that showed significant modulation rate following for at least one of the six stimulus types. Thus although the number of recording sites per cat is small, the experiments formed only a small part of the investigations in each cat, the number of cats used allows an appreciation of the variety and combination of response types across neurons. Some of the data have been published previously in abstract form (Eggermont 2001). Typically, each recording site resulted in two or three separable single units for an overall yield of 401 single units.

In 97% of the recordings, a tMTF was obtained for gamma-tone trains, in 92% for periodic click trains, in 83% for time-reversed gamma-tone trains, in 82% for AM noise, in 71% for FM tones, and only in 53% for AM tones. In 31% of the cases, the units responded to all six stimuli in an envelope-following way. This distribution shows the preference of AI units for stimuli with short rise times such as clicks and gamma tones. It also shows a clear asymmetry in the ability to respond to AM noise and AM tones, i.e., stimuli with identical envelopes but different carriers. Gamma-tone trains more commonly evoked a following response than did time-reversed gamma-tone trains. FM tones were more effective than AM tones with the same carrier frequency. There was no dependence on the difference (in octaves) between carrier tone frequency and CF of the recording site with respect to the ability to respond in an envelope following way. Neither were BMF nor limiting rate dependent on the difference (in octaves) between carrier tone frequency and CF of the recording site.

Forward vs. time-reversed gamma tones

The comparison between the responses to trains of gamma tones and trains of time-reversed gamma tones tests explicitly for the effect of the rising part of the modulation envelope on the tMTFs. Ninety recordings were made for stimulation with both gamma tones and time-reversed gamma tones. In 87 cases (97%), a well-defined BMF could be obtained for gamma-tone trains and in 83% for time-reversed gamma tones. An example of responses to forward and time-reversed gamma tones with nearly equal firing patterns is shown in Fig. 3, top, and an example where the time-reversed gamma tones evoked distinctly less synchronization than the gamma tones is shown in Fig. 3, bottom. The envelope waveforms of the stimuli are shown in gray. In the first example clear following responses are found for repetition rates 8 Hz for both stimuli and to the first three stimuli presented at 12 Hz, albeit that the adaptation is somewhat stronger for the time-reversed gamma tones. In the second example, the gamma tone stimuli show following responses up 6 Hz, and for the first three stimuli in the 8-Hz train. In contrast, the time-reversed gamma tones show response following only 4 Hz. Note that the synchronization of the firings to the first stimulus is much stronger (i.e., the jitter is less) for gamma tones than for the time-reversed gamma tones.

View larger version (79K): [in this window] [in a new window]   Fig. 3. Dot raster displays for gamma tone trains (left) and time-reversed gamma tone trains (right) superimposed on the stimulus envelope. Top: an example of a recording where the response is nearly the same for both stimuli. Bottom: an example where the time-reversed gamma tones produce much less synchronization.

Across all recordings, the average spike count per stimulus as a function of repetition rate was the same for gamma tones and their time-reversed version (Fig. 4A). However, for time-reversed gamma tones, there was a significantly lower VS (paired comparisons, t-test, P < 0.0001 for 3-12 Hz; P < 0.01 for 16 Hz) for repetition rates between 3 and 16 Hz (Fig. 4B). A pair-wise comparison for average spikes per train at a repetition rate of 8 Hz between the two stimuli is shown in Fig. 5A; the correlation between the two firing rate measures is high. The correlation between the VS values at 8 Hz (Fig. 5B) is smaller. There was no dependence on either CF of the recording site (slope of linear regression line not different from 0, P > 0.67) nor on the carrier tone frequency for either BMF, VS or limiting rate (slope of linear regression not different from 0, P > 0.07).

View larger version (32K): [in this window] [in a new window]   Fig. 4. Number of spikes per stimulus train and vector strength for gamma tones and time-reversed gamma tones. Individual data points and the LOWESS functions are shown. The rMTFs (A) are nearly identical for the 2 stimuli, whereas the mean VS (B) is larger for gamma tones from 3 to 16 Hz.

View larger version (29K): [in this window] [in a new window]   Fig. 5. Scatterplots comparing the rMTF (A) and VS (B) at a repetition rate of 8 Hz for gamma tones and time-reversed gamma tones. Regression lines are drawn in and squared correlation coefficients are indicated in the top left of each graph.

Responses to click trains, AM and FM stimuli

Figure 6 shows a set of four MU responses from the same recording electrode to periodic click trains (Fig. 6A), FM tones (Fig. 6B), AM noise (Fig. 6C), and AM tones (Fig. 6D). For the AM and FM stimuli, the modulating waveforms are plotted as well. The click waveforms are not shown. The responses to click trains (Fig. 6A) show good locking throughout the entire train for repetition rates 13.44 Hz, but only for the first two clicks at a click repetition rate of 16 Hz. For higher repetition rates, 32 Hz, the responses skip the second click and respond again to the third click in the train.

View larger version (93K): [in this window] [in a new window]   Fig. 6. Raster dot displays for a multiunit (MU) recording for clicks, FM tones, AM noise, and AM tones. Except for the click trains, the modulating waveforms are drawn in. Two single units are present in each recording, 1 is indicated as , the other as .

As can be seen in Fig. 6B, the responses to the FM tone appear both on the downward sweep and on the upward sweep for modulation frequencies 4 Hz. For higher modulation frequencies, only the initial three responses follow the down-up-down trend, after that only responses to upward sweeps are found. The response to AM noise (Fig. 6C) follow the first three modulation periods 26.88 Hz, but consistent locking throughout the noise burst is limited to modulation frequencies <16 Hz. For the AM tone (Fig. 6D), the response is similar to that for the AM noise but the synchronization appears to be less. There was no dependence on either CF of the recording site (slope of linear regression line not different from 0, P > 0.67) or on the carrier tone frequency for BMF, VS, or limiting rate (slope of linear regression line not different from 0, P > 0.05).

Pair-wise comparison of the responses to AM noise and AM tones and to AM tones and FM tones will be done in separate sections.

AM noise and AM tones

For 133 recording sites, a comparison was made between the MU responses to AM noise and AM tones. In 71 cases (53%), the tMTF to AM tones was well defined, in the sense that it showed significant VS for at least two adjacent MFs, compared with 110 cases (82%) for AM noise. In 17 recordings, the units did not respond to either stimulus, and for 65 recording sites, they responded to both stimuli. Examples of responses to AM tones and AM noise stimuli, where both show good envelope following, are shown in Fig. 7, A and B. Good envelope following for at least four periods is found 16 Hz for the AM tone and 26.88 Hz for AM noise. For modulation frequencies 32 Hz, the response consists of an on-response followed after ~100 ms by a rebound. This rebound response is stronger for AM noise and tends to be rhythmic. At modulation frequencies 32 Hz, one also notes an off-response after the 1-s mark. An example where the response to AM tones was much weaker than to AM noise is shown in Fig. 7, C and D. The response to AM noise is consistently following the modulation envelope 8 Hz. Between 8 and 26.88 Hz, an on response followed by a rebound is observed. For the AM tone stimulation, an on response is present followed by a rebound, but there is no sign of locking to the modulating waveform.

View larger version (68K): [in this window] [in a new window]   Fig. 7. Raster dot displays for AM tones (left) and AM noise (right). Shown are recordings for which the responses show envelope following for both stimuli (A and B) and where there is only a very weak following response to AM tones (C and D).

Figure 8, A and B, shows a comparison between the mean spike count per stimulus and the VS as a function of modulation frequency for AM noise and AM tones (as well as for FM tones, to be discussed in the next section). For clarity, a randomly drawn (10% of the data, but still maintaining recordings from the same sites for the 3 stimuli, using a Statview routine) sample of the 65 recordings where the units responded to all three stimuli is plotted. The mean curves calculated for all the data are drawn in. One observes (Fig. 8A) that the spike count for the 1-s-duration stimulus is highest for FM tones, then for AM noise, and smallest for AM tones. These differences were significant at all MFs (paired comparisons, t-test, P < 0.005). The mean VS-rate functions (Fig. 8B) are very similar (paired comparisons, t-test, P > 0.3) for these three stimuli and tend to level off at VS = 0.3 for modulation rates <10 Hz.

View larger version (35K): [in this window] [in a new window]   Fig. 8. Number of spikes per stimulus and vector strength for FM tone, AM tones, and AM noise. Individual data points and the LOWESS functions are shown for a randomly drawn subset of 10%. The rMTFs (A) for FM tones are weakly low-pass, those for AM stimuli are independent of modulation frequency. The VS functions (B) are identical for the 3 stimuli, showing a low-pass function with a corner frequency of 10 Hz.

A comparison between the mean BMFs for the recordings that gave reliable tMTFs, for each of the stimuli considered so far is presented in Table 1. Included in the table are the number of recordings obtained and the number of recordings that gave well-defined tMTFs for each stimulus and that resulted in the various measures represented in this table. This table shows that the mean BMF is highest for periodic click trains, then for AM noise, then for AM tones, gamma tones, FM tones, and finally for time-reversed gamma tones. The highest limiting rate is found for the AM noise. Significant differences were present between the BMFs for time-reversed gamma tones and all other stimuli (paired comparisons, t-test, P < 0.01) and between click trains and gamma-tones (paired comparisons, t-test, P < 0.0001). The differences between limiting rates were significant except for those between click trains and AM tones (paired comparisons, t-test, P = 0.4) and between gamma tones and time-reversed gamma tones (paired comparisons, t-test, P = 0.4). The peak VSs decreased in the approximately same order as the BMFs for the various stimuli. All peak VSs were significantly different from each other (paired comparisons, t-test, P < 0.005) except between AM tones and gamma tones. The peak VS was significantly, and positively, correlated with BMF for AM noise (r² = 0.10, P = 0.0007) but not for the other stimuli.

The temporal response properties, BMF, peak VS, and limiting rate, were independent of the frequency response properties, CF and BW20, for all stimuli. The correlation of BMF, peak VS, or limiting rate with CF and BW20 was not significantly different from zero.

Comparison of responses to FM and AM tones

For 169 recordings, MU responses to AM and FM tones were obtained. In 76 recordings, both stimuli produced responses with well-defined tMTFs. In 29 recordings, neither of the stimuli produced more than an onset response. In 9 recordings, there was a tMTF for AM tones but not for FM tones, and in 55 recordings, a tMTF to a FM tone was obtained but none for AM tones. Combined, 72% of recordings produced a clear following response to FM, whereas only 53% showed modulation following for AM. Figure 9, A and B, shows an example for a recording with similar modulation following responses to the FM (Fig. 9A) and the AM (Fig. 9B). In the other example, the response to the FM stimulus (Fig. 9C) shows clear following of the modulation for both downward and upward sweeps for modulation frequencies <8 Hz. For higher modulation frequencies, the response occurs for downward sweeps only. In contrast, the response to the AM stimulus (Fig. 9D) shows, for modulation rates <32 Hz, a strong onset and only some weak following of the modulation.

View larger version (75K): [in this window] [in a new window]   Fig. 9. Raster dot displays for FM tones (left) and AM tones (right). Shown are recordings for which the responses show envelope following for both stimuli (A and B) and where there is only a very weak following response to AM tones (C and D). Two single units are present in each recording, one is indicated as , the other as .

The response to the FM tones may contain a response only to the downward part of the sweep (Fig. 9A) or to both down and upward sweeps (Fig. 9C). Responses limited to upward FM sweeps throughout the modulation frequency range were not found. Because the effective, sweep-through-the-tuning-curve, modulation frequency for the unit differs in these cases, we calculated tMTFs both for a 2- to 64-Hz range (applicable if units respond both to up and down sweeps) and a 1- to 32-Hz range in case the response showed a downward sweep direction only. This procedure affected the values of the VS and the synchronized-spike tMTFs but only shifted the position of the rMTFs. In case the response mode was not clear from visualizing the responses superimposed on the modulating waveform (cf. Fig. 6B), we assigned the response to that modulation frequency range (i.e., 1-32 or 2-64 Hz) that produced the highest VS values. In 116 recordings, the directional preference for FM could be unambiguously determined from visual inspection of the dot displays or from the VS values. In 70/116 (60%) recordings, the units responded only to the downward sweep, in 46/116 recordings the response was to both downward and upward sweeps.

Figure 10 shows a comparison between the rMTFs and VS as a function of modulation frequency for units that synchronized to both up and down (bidirectional) FM sweeps (Fig. 10A) and for units that responded only to downward sweeps (Fig. 10B). Bidirectional responses were plotted over a FM range of 2-64 Hz, whereas unidirectional responses were plotted over a FM range of 1-32 Hz. One notices that the number of spikes/stimulus is the same for uni- and bidirectional units. The VS function for the bidirectional units was very similar to that for the unidirectional units. BMFs were 6.1 ± 3.5 (SD) Hz for unidirectional units and 7.2 ± 4.1 Hz for bidirectional units and were not significantly different (P = 0.13). Peak VS values were also similar for unidirectional (0.56 ± 0.16) and bidirectional units (0.54 ± 0.17).

View larger version (41K): [in this window] [in a new window]   Fig. 10. Number of spikes per stimulus and vector strength for FM tones for units with responses to both upward and downward sweeps (A) and for units that only respond to downward sweeps (B). For bidirectional units, the effective modulation frequency runs from 2 to 64 Hz (twice the frequency of the modulating sine wave).

For those units that showed a modulation following response to both AM and FM stimuli, the mean rMTFs were nearly identical between AM tones and FM tones. The VS functions were also similar for AM and FM units. Pair-wise comparison showed that BMFs (P = 0.46) and limiting rates (P = 0.81) were not significantly different for AM and FM tones.

Modulation envelope-following capacity and frequency-tuning properties

A comparison was made between the distributions of BMF across CF and across the frequency-tuning curve bandwidth at 20 dB above threshold. As earlier observed for gamma tones and time-reversed gamma tones, no differences in the distributions were observed for the presence or absence of envelope-following responses for AM tones and FM tones. Similarly, the presence or absence of a following response to periodic click trains and AM noise was independent of CF or BW20.

The carrier frequency of the AM and FM tones was generally not the same as the CF largely because not all of the simultaneous recording sites would have a CF equal to the carrier frequency. The dependence on the carrier frequency and on the difference (in octaves) between CF and carrier frequency was investigated. Again, the distribution of the envelope-following responses and that of the units that did not respond, or only with an onset response, was similar. So this mismatch of carrier frequency and CF cannot explain why an AM or FM following response, at the 55 dB SPL level tested, is present or absent.

We also investigated whether the position of the recording electrodes in the medial-dorsal plane, i.e., along the isofrequency contours, was a criterion to separate periodicity following units from nonfollowers. We could not demonstrate any effect of recording position likely because the number of recording sites per animal was either limited or confined to a relatively small recording area, particularly when using the electrode array.

Clustering of response properties to all six stimuli was investigated for 55 recordings done with the 4 × 2-electrode array. Clustering was not observed. For instance, in a recording where six electrodes produced sufficient spikes to determine tMTFs, three recordings showed envelope following only to clicks and gamma tones, whereas the other three showed an envelope following response to all six stimuli. The electrodes that responded only to clicks and gamma tones were interleaved with those that responded to all stimuli.

Modulating waveform and phase-locking of on responses

The rising part of the modulating waveform for gamma tones and time-reversed gamma tones is different (cf. Fig. 1). The differing slopes of the waveforms produce different degrees of envelope phase-locking of the spikes. This is already visible for the response to the first stimulus in the train (cf. Fig. 3, C and D). This initial synchrony difference will affect the VS and tMTF for the entire stimulus train. It was observed that the VS-repetition rate function was significantly lower for the time-reversed gamma tones (cf. Fig. 4B).

A method was devised to compensate for the different degree of envelope locking for the various stimuli used in this study. This was based on the assumption that the difference in envelope locking was already present in the response to the first stimulus in the train. Thus the VS was computed, only over the first stimulus period, for the first gamma tone or time-reversed gamma tone in the various trains. Then the original VS-repetition rate function was divided by the VS function for first stimuli only. The procedure is illustrated in Fig. 11 in a linear systems analysis analogy for click trains as stimuli. The normalized VS here is portrayed as a temporal filter (middle) that takes the VS for the response to the first click in the train as its input (left) and produces the overall VS-rate function for the entire click train as its output (right). The temporal filter frequency response function is thus obtained by dividing the VS for the entire train by the VS for the first click in each train. This temporal filter obtained is a low-pass filter with a corner frequency ~10 Hz.

View larger version (21K): [in this window] [in a new window]   Fig. 11. Normalization procedure for VS functions (top), and its linear systems equivalent (bottom). The normalized VS function (top middle) is obtained by dividing the VS for the entire train (top right) by the VS for the first stimulus only (top left). The normalized VS can be interpreted as the gain function of a temporal filter.

An overview of average VS-MTFs for the first period to AM and FM stimuli is shown in Fig. 12A and for the first stimulus period for clicks, gamma tones, and time-reversed gamma tones in Fig. 12B. Note again that the period of analysis depends on the repetition rate or modulation frequency of the stimuli that follow in the entire train. The three modulated stimuli show very similar VS(1)'s, peaking ~13 Hz. The VS(1) functions for the transient stimuli show peak values at 8 Hz and a less pronounced decrease for higher repetition rates. For all stimuli, the VS for modulation frequencies or stimulus repetition rates <4-6 Hz is generally low and increases gradually to a peak value ~12 Hz. This reflects initially the decrease in stimulus period so that spontaneous activity, suppression, and rebound effects have increasingly less effect. The decrease in VS after the peak value is likely the result of a decreasing modulation period so that the envelope-locked firings occupy a relatively large part of the analysis window. This effect is less pronounced for the relatively well-synchronized responses to gamma tones and click. The highest onset synchrony is found for clicks and AM noise, then for gamma tones and AM tones, and finally for time-reversed gamma tones and FM tones. The differences in onset synchronization between gamma tones and the time-reversed ones are evident between 8 and 20 Hz.

View larger version (22K): [in this window] [in a new window]   Fig. 12. Average VS functions for the 1st period of stimulation for modulated waveforms (A) or the 1st stimulus in periodic stimulus trains (B).

The result of the normalization procedure on the VS for gamma tones and time-reversed gamma tones is shown in Fig. 13, and one observes an, on average, low-pass normalized VS function with the mean value approaching unity at low modulation rates. One also observes that the normalized VS functions for gamma tones and time-reversed gamma tones are much more similar than the original ones (Fig. 4B). In fact there is no modulation rate for which a difference in the VS can be demonstrated at the P < 0.05 level.

View larger version (26K): [in this window] [in a new window]   Fig. 13. The normalized VS functions for gamma tones and time-reversed gamma tones are nearly identical.

The normalized VS can be interpreted as the modulation gain of the frequency response function of a putative temporal filter. All average modulation-gain functions are plotted in Fig. 14A. Because those for gamma tones and time-reversed gamma tones were nearly identical, they were averaged together. The average modulation gain at low modulation frequencies is between 2.5 and 8 dB with the smallest gain for the click trains and the highest for AM noise. The gain functions for AM tones and FM tones are similar <13 Hz, but for higher modulation frequencies, the function for FM tones is consistently lower than for AM tones. The modulation gain for AM noise is highest of all for modulation frequencies <8 Hz then follows the course of that for FM tones. The modulation gain function for clicks is distinctly different from all other gain functions for repetition rates 8 Hz. The LFP and model functions will be described in the next section.

View larger version (26K): [in this window] [in a new window]   Fig. 14. Average modulation gain for MU responses and local field potential (LFP) triggers to the 6 stimuli used in this study compared with model results. The average curves for clicks, AM tones, AM noise, and FM tones are shown in thin drawn lines and symbols. The average curves for gamma tone and time-reversed gamma tones are shown as one here (long thin dashed line). The average functions obtained for LFP triggers are shown in fat, full or dashed, lines. The model results are shown as a dotted line for the membrane filter and as full line for the combination of synaptic depression and membrane filter.

The same filter estimation procedure applied to a single-unit recording is shown in Fig. 14B. For comparison purposes, a first order low-pass filter (fat full line) was fitted through the average filter function for all stimuli (fat dashed line). It is noted that at modulation frequencies 25 Hz, deviations from the low-pass filter shape are found for all stimuli. The different filter estimates all follow a similar course for modulation frequencies <25 Hz but with the roll off frequencies distributed in the range of 5-10 Hz. This unit shows the strongest low-pass filtering for AM noise and the weakest for AM tones.

Local field potentials

To locate the origin of the putative, stimulus-invariant, temporal filter, the same normalization procedures were applied to a set of data (55 recording sites) where in addition to the MU responses, LFPs were also recorded. These results were obtained for click train stimulation, AM noise, AM tones, and FM tones. As an example, the results for click trains are shown. The rMTFs for MU and LFP in response to click trains were similar in shape and largely independent of click repetition rate (Fig. 15A), whereas the VS peaked at about a factor 2 lower repetition rate for MU than for LFPs (Fig. 15B).

View larger version (32K): [in this window] [in a new window]   Fig. 15. Number of spikes per stimulus and vector strength for MU activity and LFP triggers from the same recording sites in response to periodic click trains. The rMTFs have a similar shape, whereas the VS show a lower best modulation frequency (BMF) and smaller limiting rate for the MU results.

For all 55 recording sites, a tMTF for LFP triggers was obtained for click trains and FM tones, in 86% for AM noise, but only in 62% of the cases for AM tones. Of the 48 recordings that showed a tMTF for LFP triggers to AM noise, only 24 showed a tMTF for LFP triggers to AM tones. Thus a large percentage of recordings to AM tones failed to show envelope locking for LFPs, whereas they did for AM noise. This suggests, that part of the low yield of the MU tMTFs for AM tones may result from subcortical rate limiting processes as reflected in the LFPs.

BMFs and limiting rates were obtained for click trains and AM and FM stimuli and were generally significantly lower for MU than for LFPs (Fig. 16, A and B). Paired t-tests showed that for all stimuli, the MU values were significantly different from the corresponding LFP values (P < 0.005).

View larger version (20K): [in this window] [in a new window]   Fig. 16. Comparison of BMFs and limiting rates obtained from MU activity and LFP triggers from the same recording site for various stimuli. The MU results are generally smaller than those for the corresponding LFP results. Limiting rates up to ~60 Hz are observed for MU activity and LFP triggers

The VS functions for LFP triggers were again normalized by the VS functions for the first stimulus period only. The resulting modulation-gain functions for click LFPs and AM tone and AM noise LFPs are shown in Fig. 17. The results for AM stimuli were combined because they did not show differences. The difference in the gain functions for clicks and AM stimuli was expressed as an enhanced response of the click LFPs for repetition rates between 10 and 40 Hz, i.e., in the beta and gamma band. The average temporal-filter gain functions for LFP triggers are shown in Fig. 14 for comparison with their MU equivalents. The click LFP gain function (fat full line) stands out by its enhancement in the 10- to 40-Hz range. The AM-LFP gain function (fat long-short dashes) overlaps with the click-MU gain function for modulation frequencies 8 Hz. A comparison with the gain functions for MU activity suggests the presence of additional limiting mechanisms for modulation frequency following beyond the excitatory postsynaptic potential (EPSP) level in cortical cells.

View larger version (28K): [in this window] [in a new window]   Fig. 17. Modulation gain functions calculated for LFP triggers for click train () and AM tone and AM noise stimulation. The results for the AM stimuli were combined. Except for the enhancement between 10 and 30 Hz, the gain functions for the clicks and AM stimuli are similar.

Onset response strength and modulation following capacity

For 16/55 recording sites, for which all six stimuli were presented, units were found that responded to all stimuli with an envelope-following response. For the remainder, 39 recording sites, the response was restricted to a subset of the six stimuli. Figure 18 shows the average onset response, calculated for AM and FM stimuli over half of the first modulation period or for 15 ms after the minimum spike latency, whatever is largest, and for clicks and gamma tones over a window of 15 ms after the minimum spike latency. One observes that the "all" units group has a much larger (P < 0.0001) onset response to all stimuli presented than the other group. This suggests that high onset responses are at least one aspect that determines the capacity to respond to all stimuli in an envelope-locking manner. The peak VS, however, was significantly correlated only with the onset response for click trains (r² = 0.24, P = 0.0002) and gamma-tones (r² = 0.19, P = 0.001), i.e., for fast rise-time stimuli. Thus other aspects than high onset responses determine the capacity for modulation following responses.

View larger version (31K): [in this window] [in a new window]   Fig. 18. Onset activity with SD in response to 6 different periodic stimuli. Those units that responded to all stimuli in envelope-following fashion had significantly higher onset firing rate than all other units.

No significant difference was found for the CFs and BW20 values between the units that responded to all stimuli and those that did not.

	DISCUSSION

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

We found that most cortical neurons responded to periodic click trains, AM noise bursts, and periodic gamma tones and time-reversed gamma tones but were less responsive to FM tones and, particularly, to AM tones. An envelope-following response to AM tones was only obtained in about half of the recordings, albeit that most units responded with an onset response to this stimulus. In contrast to AM tones, AM noise produced envelope following in 82% of the MU recordings despite the fact that the number of spikes in the onset responses was not significantly different from that to AM tones. This distinction was also present in tMTFs based on LFP triggers. This points to a strong effect of the frequency content of the carrier on the subcortical processing of AM stimuli. Yet all temporal response properties were independent of CF and frequency-tuning curve bandwidth.

Phillips et al. (1985) demonstrated that the responses to FM stimuli were proportional to the slope of the rate-frequency functions at the tested stimulus level. It was suggested by Phillips and Hall (1987) that strong AM responses were associated with the rising slope of the rate-intensity function. In the present study, the levels for the AM and FM tones were the same, but those for the click and noise stimuli were 10 dB higher. This was done to assure that the activity produced within a typical unit receptive field was about the same as for the tonal stimuli and indeed resulted in similar average onset responses. The differences observed for AM noise and AM tone stimuli would suggest that the rate-intensity function was much shallower for a tonal carrier than for a noise carrier. We did obtain rate-intensity functions for gamma tones presented at 1 Hz and found that monotonicity or nonmonotonicity of this rate intensity function was not a factor in determining the presence of a modulation following response to AM tone and AM noise. In addition, nearly all of the units responded to periodically presented gamma tones in a repetition-rate following fashion.

Time-reversed gamma tones produced responses much more often than the AM tones. The rising slope of the time-reversed gamma tones is about the same as that for a 20 Hz exponential sine modulated tone. So if slope is a critical point, one would at least expect that the responses at the high end of the AM range would be effective. This was not the case because most VS-MTFs functions were low-pass for AM tones. Similarly, for 20-Hz repetition rate the time-reverse gamma tones never produced an envelope-locking response. Thus the shallower rise times and the <100% modulation depth for the AM tones may further limit the capacity for the temporal representation of modulation frequency.

Neural synchronization to the stimulus

We concentrated in this study on the VS of spikes and LFP triggers to the modulating waveform because, as we found in previous studies (Eggermont 1994, 1998), the overall firing rate was largely independent of the modulation frequency 64 Hz, the highest repetition rate used. In previous studies, Schreiner and Urbas (1986) found in the anterior auditory field of the anesthetized cat that firing rate was often band-pass tuned to modulation frequency just as VS. Bieser and Müller-Preuss (1996) found various firing rate dependencies in auditory cortex of awake squirrel monkeys, but rate independence was rare. In rat auditory cortex, 60% of rMTFs were flat, 30% were band-pass and the remainder low-pass (Gaese and Ostwald 1995). Lu and Wang (2000) found a steady increase of firing rate for click intervals shorter than 40 ms in ketamine-anesthetized cat AI. Looking at their Fig. 8, an appreciable increase is, however, only observed for intervals <10 ms. Because we used only three values in the range between 16 and 40 ms, such an increase could not be demonstrated in our data. Bieser and Müller-Preuss (1996) noted that the representation of AM sounds in awake squirrel monkeys between 2 and 64 Hz was by phase-locked responses, whereas higher AM frequencies seem to be represented by firing rate. It appears from a comparison of these studies that anesthesia plays only a minor role as the tMTFs in ketamine-anesthetized and awake animals are similar. However, the variable spontaneous firing rate in ketamine-anesthetized animals will affect the VS. This may explain some of the variability in the VS, as for instance shown in Figs. 4 and 11.

The definition of the VS (Goldberg and Brown 1968) implies that punctuate phase-locked responses separated by suppressed spontaneous activity result in greater VS than more tonic responses. For instance, in case of a perfect full-wave sinusoidal modulation of the period histogram one obtains a VS = 0.5. In cases where the period histogram resembles a half-wave rectified sinewave, i.e., no activity in one half of the histogram, the VS = 0.785. In cases where all the firings occur in one specific phase bin of the sinewave the VS = 1. Any form of postactivation suppression could cause strongly phase-locked patterns.

Previously, we attributed the dependence of gap-detection thresholds on leading burst duration to postactivation suppression resulting from after hyperpolarization (Eggermont 2000). Other possibilities, provided they have the same dynamics, are delayed inhibition resulting from feed forward or feedback connections. Both types of inhibition involve interneurons and the effects extend laterally to neighboring cells (Douglas and Martin 1998). Feedback inhibition would depend on the strength of the cortical neural response, whereas feed forward inhibition would be dependent on the strength of the thalamic input. Broadband stimuli such as clicks and AM noise likely produce stronger lateral inhibition than gamma tones and AM tones. The stronger postactivation suppression and rebound activity observed for the broadband stimuli provide evidence for this. As a result, the VS values for wideband stimuli may be larger and also reach significance at lower firing rates.

FM tones produced envelope-locked responses more effectively than AM tones, and this could also relate to the lateral inhibition produced by FM. This observation was also made by Gaese and Ostwald (1995) who reported phase-locked responses for FM in 80% and for AM in 55% of the recordings. Bieser and Müller-Preuss (1996) reported that 22% of neurons failed to lock to the AM envelope. In addition, the FM tone resides for less time within the receptive field of the neurons than does the corresponding AM tone. As a result, the latter would cause greater adaptation, and because of its dynamic range of only 17.4 dB, it would have a larger probability of continuously depressing transmitter release. However, when cortical neurons respond to both FM and AM tones, the resulting tMTFs are similar. Gaese and Ostwald (1995) also showed that in the auditory cortex of rats the best modulation frequencies, for sinusoidal AM and FM, were significantly correlated for the two stimuli.

We used a fixed number of bins for the period histograms that had lengths ranging from 15.6 ms (64 Hz) to 1 s. This was intended to produce the same relative accuracy across the range of modulation frequencies, i.e., one bin stands for a 20° phase angle. One could also opt for a fixed bin size, say of 1 ms, to retain the same resolution of 16 bins/period at a modulation frequency of 64 Hz (Schreiner and Raggio 1996; Schreiner and Urbas 1986). This will not change the overall characteristics of the tMTF but will reduce the VS progressively for the lower modulation frequencies compared with the approach using a fixed number of bins. As a result tMTFs (and VS) for fixed binwidth tend to be more band-pass like than tMTFs based on relative binwidth. A normalization procedure as described here would correct for this difference and result in the same modulation gain for the temporal filter based on either fixed or relative binwidth.

Comparison to previous findings

This report is a follow-up study on Eggermont (1994) where the effect of modulating waveform and tone or noise carrier was studied and is also comparable to the results for primary auditory cortex from Eggermont (1998). In general, the findings for those stimuli that were used in these three studies were very similar and show flat rMTFs and low- or band-pass VS functions. For all three datasets, the click tMTF was band-pass. For the present dataset, the VS function for AM noise was a low-pass function of MF (cf. Fig. 8) as it was for the AI recordings reported in Eggermont (1998). Those in the earlier report (Eggermont 1994) showed one set of data with a tMTF for AM noise similar to that for clicks, i.e., band-pass, whereas the other set showed a low-pass tMTF. Band-pass VS functions were also the dominant finding in rat AI for both AM tones and FM tones (Gaese and Ostwald 1995). We have seen this in the present study for a subset of neurons with high onset firing rates that tended to respond to all stimuli in a phase-locked manner. In the earlier study, responses were obtained at various intensity levels, and the tMTF with the highest response was used in the average results, whereas in the two later studies, responses were obtained at a fixed intensity level.

Schreiner et al. (1997) showed that repetition rate coding for click stimuli was related to onset latency (see also Lu and Wang 2000; Schreiner and Raggio 1996). Because onset latency is mapped systematically along the dorsoventral extent of the primary auditory cortex in cat (Mendelson et al. 1997), one expects that in our study neurons with different response type will be mapped differentially on the cortical surface. We could not find evidence for such a mapping. This could be because the number of recordings per animal was either relatively small or confined to small patches. In addition, we could not find a systematic mapping or clustering of response properties across the 4 × 2 electrode array. This suggests that a potential spatial clustering of temporal response properties is either confined to patches smaller than 0.5 mm in diameter (the interelectrode distance) or present diffusely on a much larger scale than the size of the array, which had an area of ~0.75 mm². Given the results of Schreiner et al. (1987), this latter option is the most likely.

Mendelson and Cynader (1985) and Heil et al. (1992) showed that two-thirds of neurons in cat auditory cortex showed a preference for linear downward FM sweeps, independent of the CF of the neuron. In our sample, 60% of the neurons was responding to downward sinusoidal sweeps only, whereas the other units responded to both up and down sweeps. In rat auditory cortex, 52% of neurons were selective for upward sweeps, and only 30% to downward and the remainder was bidirectional (Gaese and Ostwald 1995).

In the present study FM rates were between 1 and 32 octaves/s. The CFs of the units in this study ranged from ~1 to 20 kHz. Thus the range of modulation rate