INTRODUCTION

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

Human speech and musical sounds contain prominent temporal modulations in both amplitude and frequency. Low-frequency (<50 Hz) modulations are important for speech perception and melody recognition, whereas modulations at higher frequencies produce other types of sensations such as pitch and roughness (Houtgast and Steeneken 1973; Rosen 1992). Amplitude and frequency modulations (AM and FM) are also important components of communication sounds of animals and are found in a wide range of species-specific vocalizations. The neural representation of amplitude- and frequency-modulated sounds begins at the auditory periphery, where auditory-nerve fibers faithfully represent both fine and coarse temporal structures of complex sounds in their temporal discharge patterns (Johnson 1980; Joris and Yin 1992; Palmer 1982). At subsequent brain stem nuclei along the ascending auditory pathway, the precision of the temporal representation degrades gradually, due to the biophysical properties of neurons along the ascending pathway and temporal integration of converging inputs from one station to the next (Blackburn and Sachs 1989; Creutzfeldt et al. 1980; de Ribaupierre et al. 1980; Frisina et al. 1990; Langner and Schreiner 1988). In a modeling study of the transformation of temporal discharge patterns from the auditory-nerve to the cochlear nucleus, Wang and Sachs (1995) showed that the reduction of phase-locking in stellate cells can result from three mechanisms: convergence of subthreshold inputs on the soma, inhibition, and the well-known dendritic low-pass filtering (Rall and Agmon-Snir 1998). These basic mechanisms may also operate at successive nuclei leading to the auditory cortex, progressively reducing the temporal limit of stimulus-synchronized responses.

It has long been known that neurons in the auditory cortex have a limited capacity to represent temporally modulated signals (Goldstein et al. 1959; deRibaupierre et al. 1972; Whitfield and Evans 1965). In contrast to subcortical neurons, neurons in the auditory cortex can only synchronize to temporally modulated signals at modulation rates of up to tens of Hertz (Eggermont 1991, 1994; Gaese and Ostwald 1995; Schreiner and Urbas 1988) compared with hundreds or thousands of Hertz subcortically (Creutzfeldt et al. 1980; Frisina et al. 1990; Joris and Yin 1992). Because most of the studies in the past three decades on this subject were conducted in anesthetized animals, with a few exceptions (Bieser and Müller-Preuss 1996; Goldstein et al. 1959; deRibaupierre et al. 1972), it has been suspected that the low temporal response rate reported in the auditory cortex might partially be caused by anesthetics, which have been shown to alter temporal responses properties of the auditory cortex (Gaese and Ostwald 2001; Goldstein et al. 1959). It is therefore important to obtain measurements of cortical responses to temporally modulated signals under unanesthetized conditions, which could provide a better correlation with the perception of these signals.

While mechanisms based on stimulus-synchronized discharges have long been assumed to be the predominant means for the cortex to represent temporal modulations (see review by Langner 1992), the significance of discharge rate-based mechanisms in representing temporal features of complex sounds has gained little attention. This is perhaps due to the fact that sustained discharges are not commonly observed under anesthetized conditions. Under the awake condition, however, neurons in the auditory cortex often respond with sustained discharges throughout the entire stimulus duration (Bieser and Müller-Preuss 1996; Evans and Whitfield 1964; Lu et al. 2001a,b; Recanzone et al. 2000). Our recent study of the auditory cortex in awake primates using sequential stimuli has provided clear evidence to support a two-stage temporal processing mechanism that suggests temporal coding for slowly changing acoustic events and rate coding for rapidly changing acoustic events (Lu et al. 2001b). The present study using continuously modulated signals provides further supporting evidence for such a mechanism.

Another important issue regarding the cortical processing of temporal modulations is how cortical neurons represent similar temporal features that are introduced by different means. In the spectral domain, the notion of frequency filtering has been well established on the basis of the response area obtained from pure tones or other types of stimuli. It is not clear, however, whether a common temporal processing mechanism, or a temporal filter, is applied by cortical neurons to a variety of time-varying signals. To answer these questions, it is necessary to comprehensively test the temporal response properties of cortical neurons with a variety of temporally modulated signals. In this study, we systematically characterized cortical responses to two representative classes of temporally modulated signals, sinusoidally amplitude-modulated (sAM) and frequency-modulated (sFM) tones, in a large number of single-units in awake marmoset monkeys (Callithrix jacchus), a highly vocal primate species (Wang 2000). Preliminary observations from the present study were presented at two conferences (Liang et al. 1999; Wang et al. 2001).

	METHODS

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

Animal preparation and recording procedures

Details on animal preparation and recording procedures were described in a previous study (Lu et al. 2001a) and are only briefly described here. Marmosets were adapted to sit quietly during recording sessions in an apparatus specially designed for this species. The auditory cortex was accessed laterally using a single tungsten microelectrode of impedance typically ranging from 2 to 5 M at 1 kHz (A-M Systems) through a small hole (diameter, ~1.0 mm) in the skull. Only one opening in the skull existed at any given time during the recording sessions. Each hole was sealed by dental cement after several days of recordings. Necessary steps were taken to ensure sterility during all recording sessions. Daily recording sessions (3-5 h) were carried out for several months in each animal. The advantage of this procedure was that it only left a very small portion of the cortex exposed, which greatly increased the recording stability, avoided excess tissue growth and reduced the chance of infections through the opening. All recording sessions were conducted in a double-walled, sound-proof chamber (IAC-1024). The interior of the chamber was covered by 3-in acoustic absorption foam (Sonex, Illbruck). The experimental procedures were approved by the Animal Care and Use Committee at The Johns Hopkins University.

Densely positioned recording holes were made covering the auditory cortex. The data presented were mainly obtained from the primary auditory cortex (A1) and may include a few neurons from the immediately adjacent areas that responded to sAM and/or sFM stimuli. The location of A1 was determined by its tonotopic organization, its relationship to the lateral belt area (which was more responsive to noises than tones), and by its response properties (e.g., highly responsive to tonal stimuli). Electrode penetrations, perpendicular to the cortical surface, were made within each recording hole under visual guidance via an operating microscope. This gave good control and estimates of recording depths. Single-units were encountered at all cortical layers, but the majority of the recorded units was from upper layers, judging by the depths and response characteristics. On average, one to three well-isolated single-units were studied in each daily session. A representative example of raw recordings is shown in Fig. 1. Signal-to-noise ratio was typically >10:1 in our recordings. Spike waveforms were filtered, digitized, and detected using a template-matching discriminator (MSD, Alpha-Omega Engineering) and were closely and constantly monitored by an experimenter as the recording proceeded. The template matching method prevented any unwanted noises (e.g., due to the animal's movement) from triggering false spikes.

View larger version (34K): [in this window] [in a new window]   Fig. 1. Examples of digitized traces of extracellular recordings from the auditory cortex of an awake marmoset in response to sinusoidally amplitude-modulated (sAM) stimuli. Action potentials from a single-unit are present in the recordings. A complete display of this unit's responses to sAM and sinusoidally frequency-modulated (sFM) stimuli is given in Fig. 2B. A: digitized traces are ordered vertically according to modulation frequency (from 1 to 512 Hz as indicated along ordinate). Stimulus duration was 1,000 ms (onset at 500 ms). One repetition for each modulation frequency is shown. Signal-to-noise ratio was typically >10:1 in our recordings. B: a portion of the response at 64-Hz modulation frequency is shown in an enlarged view. C: average waveform of all action potentials (n = 3,359) recorded during the presentation of the entire set of sAM stimuli from this unit (10 repetitions at each of 10 modulation frequencies). , the mean value; - - - , 1 SD away from the mean. Sampling rate for the data shown in this figure was 30 kHz. Quantitative analyses of responses recorded from this unit are subsequently shown in Figs. 2-4B and 7B.

Acoustic stimuli

Two types of temporally modulated sounds were used as the experimental stimuli in this study: sAM and sFM sounds. For sAM stimuli, the carrier frequency, set at a unit's characteristic frequency (CF), was held constant while its amplitude was modulated by a sinusoid. For sFM sounds, the amplitude remained constant while the carrier frequency, centered at a unit's CF, was sinusoidally modulated. Acoustic stimuli were delivered in free-field through a loudspeaker located ~1 meter in front of the animal. Frequency tuning was obtained using a series of randomly presented tone bursts (50-100 ms in duration), from which the CF of a unit was determined as the frequency at which the strongest discharge rate was evoked. A rate-level function was obtained at CF. Most units in the awake auditory cortex exhibited nonmonotonic discharge rate versus sound level functions (Pfingst and O'Connor 1981; Wang et al. 1999), for which a preferred sound level can be determined. After these initial characterizations, multiple repetitions of sAM and sFM stimuli were delivered at the preferred sound level for nonmonotonic units (or 30 dB above threshold if otherwise). In a subset of units, sAM and/or sFM stimuli were tested at multiple sound levels. Several stimulus parameters were varied to probe neural responses. For every unit included in the analysis, modulation frequency was typically varied between 1 and 512 Hz in a base-2 logarithmic scale; finer steps were used in testing some units. The modulation depth of sAM stimuli was generally set at 100%; in a subset of sampled units, a range of depth (0-100%) was tested. The modulation depth of sFM stimuli was usually set at an optimal (in terms of maximum firing rate) or a nearly optimal depth centered at a unit's CF. Multiple FM depths were tested in some units. Unlike sAM stimuli, the firing rate resulting from a sFM stimulus was not monotonically related to modulation depth. The optimal depth, at which the maximum firing rate was achieved, varied from unit to unit. It was therefore not feasible to choose a fixed FM modulation depth for all units. Using the optimal FM depth for a unit allowed measurement of modulation selectivity to be made at its maximum discharge level, thus increasing the robustness of the measurement. Because firing rates of a unit at a given modulation frequency generally increased with increasing AM depth for a sAM stimulus, it was possible to used a fixed AM depth (100%) to test all units. The duration of each sAM and sFM stimulus was 1,000 ms. Neural activities prior to and following stimulus presentation were also recorded to estimate spontaneous discharges and to reveal any long-lasting effects. Ten to 20 repetitions of a sAM or sFM stimulus were presented at each modulation frequency at a given sound level. Stimuli of all modulation frequencies were presented randomly. Inter-stimulus intervals were >1 s. Stimuli were synthesized at 100-kHz sampling rate and low-pass filtered at 50 kHz. The spike times were digitized at 50-kHz sampling rate.

Data analysis

The results reported were based on 211 single-units recorded from the left hemispheres of three awake marmoset monkeys. Responses to sAM and sFM stimuli were recorded in 200 and 142 units, respectively. Some units responded to both types of stimuli and others responded to either sAM or sFM stimuli. All units were recorded indiscriminately, provided they could be driven by either sAM or sFM stimuli. Given the diversity of cortical responses in the awake preparation, we did not expect all recorded units to respond to both types of stimuli. Discharge rates of the majority of units varied with changing modulation frequency. We separated sAM and sFM responses, respectively, into two groups in the analyses according to the characteristics of the discharge rate versus modulation frequency profile of a unit. A rate profile was considered having a "band-pass" shape if its peak value was higher than values at both the lower and higher frequency sides. Some units with band-pass rate profiles also exhibited increased discharge rates with increasing modulation frequency at frequencies much higher than the frequency corresponding to the peak of the rate profile. This portion of responses was not included in the analysis of rate-based modulation selectivity because they were more likely to be influenced by spectral effects at such high modulation frequencies. Responses with band-pass rate profiles were further screened on the basis of a d' value, d' = |µ_Rd  µ_Rs|/_Rs, where µ_Rd is the mean discharge rate during a stimulus presentation, µ_Rs is the mean spontaneous discharge rate, and _Rs is the SD of the spontaneous discharge rate. Values of d' were calculated for responses to each modulation frequency tested. If a unit had a band-pass rate profile and a maximum d' value 1.0, it was classified into the band-pass (BP) group. The rest of the units were referred to as the nonband-pass (non-BP) group, which included units whose maximum d' values were <1.0 (indicating weak responses to sAM or sFM stimuli) as well as those units whose rate profiles did not exhibit "band-pass" shapes. A unit belonged to either the BP or non-BP group. This classification process was separately applied to sAM and sFM responses. For sAM responses, 146/200 (73%) units belonged to the BP group and 54/200 (27%) units belonged to the non-BP group. For sFM responses, the numbers were 93/142 (65%, BP group) and 49/142 (35%, non-BP group), respectively. Units in the BP group responded to the change in modulation frequency by their firing rates and were analyzed for discharge rate-based modulation selectivity. Units in both BP and non-BP groups were analyzed for discharge synchrony-based modulation selectivity. Population averages were computed for each group of units as well as for all the units.

Unless specified, average discharge rates were calculated over a window including the stimulus duration and 100 ms after stimulus offset: (t_onset, t_offset + 100 ms), where t_onset and t_offset are stimulus onset and offset times. Spontaneous discharge rates were estimated from activities prior to stimulus onset (500 ms) and subtracted from raw discharge rates. Several response measures, described in the following text, were used to quantify the modulation selectivity of cortical neurons. Comparison between response measures resulting from sAM and sFM stimuli were made between populations of units that responded to each stimulus type as well as on a unit-by-unit basis in individual units in which both measures could be obtained. Statistical comparisons between distributions of response measures were made using the Wilcoxon rank-sum test (Rice 1988). Unit-by-unit comparisons between response measures recorded from the same unit were made using the paired t-test. P < 0.01 was considered statistically significant for these analyses.

RATE MODULATION TRANSFER FUNCTION (rMTF). The relationship between average discharge rate and modulation frequency is referred to as the discharge rate-based modulation transfer function (rMTF). The discharge rate-based best modulation frequency (rBMF) was calculated in the following steps for each unit belonging to the BP-group. For units with peaks in their rMTFs that consisted of two or more points, an estimate of rBMF was first obtained as the modulation frequency corresponding to the largest discharge rate of a rMTF. The rBMF was then calculated by weighting those modulation frequencies that were continuous and adjacent to the estimated rBMF and whose discharge rates were not significantly different from the estimated rBMF (P > 0.05, Wilcoxon rank-sum test). A geometric mean was used to average the modulation frequencies by their discharge rates. The rBMF would be equal to the estimated rBMF if discharge rates at other frequencies were all significantly different from that at the estimated rBMF. This method has an advantage over simply assigning the rBMF to one of tested modulation frequencies corresponding to the largest discharge rate, as commonly used in most previous studies. For example, a rMTF that had a broad peak centered on two modulation frequencies with similar discharge rates would have a calculated rBMF near their geometric mean but weighted closer to the modulation frequency that produced a stronger discharge.

SYNCHRONY MODULATION TRANSFER FUNCTION (tMTF). Stimulus-synchronized discharges were characterized first by the vector strength (VS) (Goldberg and Brown 1969) and then converted to the Rayleigh statistics (2nVS², where n is the total number of spikes) (Mardia and Jupp 2000) to assess their statistical significance. VS was calculated with a time window beginning 100 ms after stimulus onset to the end of the stimulus. The values of the Rayleigh measure >13.8 were considered as statistically significant (P < 0.001) (Mardia and Jupp 2000). The relationship between Rayleigh statistics and modulation frequency was referred to as the temporal modulation transfer function (tMTF). In some units, tMTF had a band-pass shape. A response measure called discharge synchrony-based best modulation frequency (tBMF) was calculated from each tMTF. For units whose tMTFs consisted of two or more significant values (Rayleigh statistic > 13.8), the tBMF was obtained by weighting modulation frequencies with significant VS that were adjacent, continuous and surrounding the modulation frequency corresponding to the maximum VS. A geometric mean was used to average the modulation frequencies by their VS. To calculate maximum synchronization frequency (f_max), we first determined the highest modulation frequency at which significant discharge synchrony was found. A linear interpolation was made between this frequency and the adjacent, higher, tested modulation frequency with nonsignificant Rayleigh statistics. The modulation frequency where the interpolated Rayleigh statistic line crossed 13.8 was taken to be f_max.

CALCULATION OF THE DOUBLING OF SYNCHRONIZATION FREQUENCY. A subset of units exhibited discharge patterns that were synchronized to twice the modulation frequency (2f_m) under certain stimulus conditions. For such units, tMTFs based on Rayleigh statistics were calculated at f_m and 2f_m, respectively [i.e., 2n(VS_fm)² and 2n(VS_2fm)²]. A unit was considered having the doubling of synchronization frequency if the peak of the tMTF(2f_m) was higher than the magnitude of the tMTF(f_m) at the corresponding modulation frequency. For this type of unit, f_max based on both tMTF(f_m) and tMTF(2f_m) were computed in the same manner as described in the preceding text.

MINIMUM RESPONSE LATENCY. Minimum response latency to sAM or sFM stimuli was determined, respectively, on the basis of a composite PSTH of a unit's responses at all tested modulation frequencies. A cumulative post-stimulus histogram (PSTH) was then constructed by integrating the PSTH over time. The time after stimulus onset at which the spike count in a bin of the cumulative PSTH exceeded twice the largest SD of spike counts prior to stimulus onset was calculated as the minimum response latency. The binwidth used in the calculation was 1 ms. The minimum response latency defined here could be different from the first spike latency measured from CF tones in each unit. We used this latency measure instead of the first spike latency because it was a more direct indicator of onset timing of a neuron in response to sAM or sFM stimuli. The first spike latency was not always available as some neurons did not respond well to unmodulated CF tones.

	RESULTS

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

General observations

The majority of the sampled neurons responded to CF tones as well sAM or sFM stimuli. However, neurons generally responded more strongly to sAM and sFM stimuli, often with sustained firing, than to tones as judged by number of spikes evoked over the duration of the sAM and sFM stimuli (1 s). Some of the neurons could only be driven by sAM or sFM stimuli with proper parameters (e.g., modulation frequency and depth). Representative examples of responses to sAM and sFM stimuli are shown in Fig. 2. Overall discharge patterns can be seen from the dot raster and PSTH. Temporal discharge patterns are further illustrated by period histograms computed from the same group of units in Fig. 3. These examples show that responses of units generally varied as a function of modulation frequency. They also show that some, but not all, recorded units exhibited stimulus-synchronized discharges at low modulation frequencies that gradually disappeared with increasing modulation frequency (Fig. 2, A-C). Responses to sAM and sFM stimuli often diminished at high modulation frequencies for the frequency range tested (Fig. 2Db). Not all units responded at low modulation frequencies (Fig. 2, Da and Ea). In general, sustained discharges were limited to a narrower range of modulation frequencies than onset discharges (Fig. 2, D and E). These and other response characteristics are quantitatively analyzed in the following sections.

View larger version (68K): [in this window] [in a new window]   Fig. 2. Examples of cortical responses to sAM and sFM stimuli from 5 representative single-units. A-E: each row shows responses to both sAM (a) and sFM (b) stimuli recorded from each unit in the form of dot raster (left) and poststimulus time histogram (PSTH; right). The displays are arranged according to modulation frequency along the ordinate. Stimulus duration was 1,000 ms (onset at 500 ms), as indicated by a horizontal bar below time axis (E). Stimulus parameters for each example are as follows fc, carrier frequency; dsAM, sAM modulation depth; dsFM, sFM modulation depth. A: fc = 7.47 kHz, 20 dB SPL, dsAM = 100%, dsFM = 256 Hz. B: fc = 16.07 kHz, 80 dB SPL, dsAM = 100%, dsFM = 1,024 Hz. C: fc = 0.62 kHz, 50 dB SPL, dsAM = 100%, dsFM = 32 Hz. D: fc = 6.94 kHz, 60 dB SPL, dsAM = 100%, dsFM = 3566 Hz. E: fc = 2.38 kHz, 40 dB SPL, dsAM = 100%, dsFM = 362 Hz.

View larger version (34K): [in this window] [in a new window]   Fig. 3. Period histograms calculated from the responses shown in Fig. 2. A-E correspond to those in Fig. 2, A-E, respectively. Responses to sAM and sFM are shown in a and b, respectively. Two stimulus periods are shown in each histogram. The periods are shown in units of radians.

Discharge rate-based modulation frequency selectivity

PROPERTY OF INDIVIDUAL NEURONS. The majority of recorded units displayed selectivity for a particular modulation frequency when measured by average discharge rate. Figure 2 shows responses to both sAM (a) and sFM (b) stimuli recorded from the same units across a range of modulation frequencies. rMTF from all five units shown in Fig. 2 are plotted in Fig. 4. For example, the unit shown in Fig. 2C responded most strongly near 32-Hz modulation frequency. PSTHs in Fig. 2C show sustained firing at a modulation frequency of 32 Hz for both sAM (Fig. 2Ca) and sFM (Fig. 2Cb) stimuli. rMTFs produced by sAM and sFM stimuli had similar band-pass shapes and peaked at a modulation frequency of 32 Hz (Fig. 4C). The peak in a rMTF is conventionally referred to as the rBMF. In this study, we used a quantitative method to calculate the rBMF instead of using a particular tested modulation frequency (see METHODS). Other examples in Fig. 2 illustrate the typical range of modulation frequency selectivity observed in the recorded units. A prominent feature in these examples, representative of our large samples, is the sustained firing for the entire stimulus duration at modulation frequencies near rBMF (Fig. 2, D and E). Neurons generally responded more weakly at modulation frequencies lower than rBMF. In some cases, there were no responses or only brief onset responses at these lower modulation frequencies (e.g., Fig. 2E). The disappearance of sustained discharges at modulation frequencies higher than rBMF was commonly observed (e.g., Fig. 2, B-E). The lack of responses at high modulation frequencies appeared to result from inhibition in many cases (e.g., Fig. 2, C and D). In general, rMTFs derived from responses of a unit to sAM and sFM stimuli had similar shapes and closely matched rBMF.

View larger version (15K): [in this window] [in a new window]   Fig. 4. Examples of discharge rate-based modulation transfer functions (rMTF). Average discharge rates were calculated over a period including the stimulus duration and 100 ms after stimulus offset, with spontaneous discharge rates subtracted (see METHODS). Units in A-E correspond to those shown in Fig. 2, A-E, respectively. For each unit, rMTFs produced by sAM () and sFM (××) stimuli are shown. The best modulation frequencies (rBMF) calculated from rMTF (see METHODS) for the 5 examples shown are as follows (rBMFsAM, rBMFsFM). A: 11.07, 4.16 Hz. B: 16.0, 16.0 Hz. C: 23.38, 32.0 Hz. D: 87.36, 64.0 Hz. E: 128.0, 128.0 Hz. The half-height bandwidth (BW) and its corresponding low- and high-frequency boundaries (see METHODS) are listed in the following order: BWsAM (low-frequency, high-frequency), BWsFM (low-frequency, high-frequency). A: 25.19 (1.65, 26.84) Hz, none (none, 16.98) Hz. B: 153.23 (2.85, 156.08) Hz, 57.71 (3.19,60.9) Hz. C: 54.39 (6.88, 61.27) Hz, 51.74 (9.87, 61.61) Hz. D: 166.94 (44.77, 211.71) Hz, 106.02 (23.86, 129.88) Hz. E: 138.12 (59.07 197.19) Hz, 137.34 (59.7 197.04) Hz. The Q measure, defined as rBMF/BW, is listed in the following order: QsAM, QsFM. A: 0.44, none. B: 0.1, 0.28. C: 0.43, 0.62. D: 0.52, 0.6. E: 0.93, 0.93.

POPULATION PROPERTIES. Using the procedures described in METHODS, we have computed rBMF from the units that responded reliably to sAM and/or sFM stimuli (see METHODS). Figure 5A shows distributions of rBMF_sAM and rBMF_sFM, respectively. Both distributions are centered between 16 and 32 Hz (rBMF_sAM: median = 22.6 Hz, rBMF_sFM: median = 18.1 Hz, see Table 1) and are statistically indistinguishable (Wilcoxon rank-sum test, P = 0.1). Among the 211 single-units that we studied, rBMF_sAM could be determined in 146 units (69%), whereas as rBMF_sFM could be determined in 93 units (44%). Figure 5B shows the relationship between rBMF_sAM and rBMF_sFM in 76 units where both were determined in the same unit. rBMF_sAM and rBMF_sFM were highly correlated (correlation coefficient r = 0.7, Table 1). A paired t-test showed that there was no significant difference (P = 0.02) between rBMF_sAM and rBMF_sFM when compared on a unit-by-unit basis. A closer examination revealed that units in the upper 50th percentile of rBMF_sAM (rBMF_sAM > median rBMF_sAM = 20.7 Hz) appeared to have significantly higher rBMF_sAM than rBMF_sFM values (paired t-test, P < 0.01). There was no statistically significant difference (paired t-test, P = 0.09) between rBMF_sAM and rBMF_sFM for the units in the lower 50th percentile of rBMF_sAM (Table 1). The distribution of the difference between rBMF_sAM and rBMF_sFM pairs is shown in Fig. 5C. The close match between rBMF_sAM and rBMF_sFM was found in a substantial proportion of this population of units. The median of the distribution was 0.46 octaves, with 50 of 76 units (66%) having closely matched rBMF_sAM and rBMF_sFM (differences within 1.0 octave). These data show that there was a great degree of similarity in the responses to sAM and sFM stimuli, as reflected in mean firing rate, both at the level of single neurons and the level of populations of neurons in the auditory cortex. In general, the match between the rBMF_sAM and rBMF_sFM was independent of whether discharges were synchronized to the modulation waveform or not.

View larger version (21K): [in this window] [in a new window]   Fig. 5. Population properties for discharge rBMF selectivity. A: overlapping histograms showing distributions of rBMF derived from sAM () and sFM () stimuli, respectively. The binwidths of the histograms are on a base-2 logarithmic scale. The distributions of rBMFsAM and rBMFsFM are not statistically different from each other (Wilcoxon rank-sum test, P = 0.1). The means of the 2 distributions are 25.9 Hz (rBMFsAM) and 19.2 Hz (rBMFsFM) on the base-2 logarithmic scale and 48.8 Hz (rBMFsAM) and 35.1 Hz (rBMFsFM) on the linear scale, respectively (Table 1, columns 1-3). The medians of the 2 distributions are 22.6 Hz (rBMFsAM) and 18.1 Hz (rBMFsFM), respectively. B and C: unit-by-unit comparison between rBMFsFM and rBMFsAM recorded from the same unit (Table 1, column 4). This group of 76 units was a subset of the units analyzed in A. rBMFsAM and rBMFsFM were not statistically different from each other in the unit-by-unit comparison (paired t-test, P = 0.02). The correlation coefficient (r) between rBMFsAM and rBMFsFM (B) was 0.7. The difference between rBMFsAM and rBMFsFM was calculated in octaves as log2(rBMFsAM/rBMFsFM) (C, median = 0.46 octaves). - - -, slope of 1.0.

BANDWIDTH OF rMTF. In Fig. 6 we analyzed and compared half-height BW and sharpness of tuning (Q) of rMTF for both sAM and sFM stimuli (see METHODS). The distributions of BW across populations of the neurons were similar for both types of stimuli (Fig. 6A) and were not statistically different (Wilcoxon rank-sum test, P = 0.16). The distributions of Q values (Fig. 6C) were also similar between sAM and sFM stimuli (Wilcoxon rank-sum test, P = 0.77). The medians of BW distributions were between 32 and 64 Hz (BW_sAM: 53.9 Hz, BW_sFM: 49.3 Hz, see Table 1), approximately one octave greater than the medians of rBMF (Fig. 5), which resulted in Q distributions centered around 0.5 octaves (Q_sAM: 0.45, Q_sFM: 0.47). In a subpopulation of units, BW and Q could be measured for both sAM and sFM stimuli in the same units. Figure 6, B and D, shows that, when compared on a unit-by-unit basis, the two measures of the tuning width of rMTF were not statistically different (paired t-test, BW: P = 0.27, Q: P = 0.1) between sAM and sFM stimuli (Table 1). These results showed that in addition to the similarity in rBMF, there was also a similarity in the sharpness of tuning of rMTF produced by sAM and sFM stimuli, both at the level of single neurons and across populations of neurons.

View larger version (42K): [in this window] [in a new window]   Fig. 6. The tuning width of rMTF. A: distributions of half-height BW for sAM () and sFM () stimuli, respectively. Two distributions are not statistically different from each other (Wilcoxon rank-sum test, P = 0.16). B: unit-by-unit comparison between BWsAM and BWsFM (paired t-test, P = 0.27). C: distributions of the sharpness of tuning, Q = rBMF/BW, for sAM () and sFM () stimuli, respectively (Wilcoxon rank-sum test, P = 0.77). D: unit-by-unit comparison between QsAM and QsFM (paired t-test, P = 0.1). - - -, slope of 1.0.

Discharge synchrony-based modulation frequency selectivity

PROPERTY OF INDIVIDUAL NEURONS. The examples given in Figs. 2 and 3 also show that discharges of cortical neurons in response to sAM and sFM stimuli could exhibit stimulus-synchronized temporal patterns. Period histograms shown in Fig. 3, A-E, corresponding to the units shown in Fig. 2, A-E, further illustrate temporal discharge patterns evoked by the modulated sounds. Discharges synchronized to modulation waveform should show peaks in both of the two periods plotted (Fig. 3). Discharges registered in the first but not the second period of a histogram indicate that they were synchronized to stimulus onset but not to the modulation waveform. The unit in Fig. 3B responded to sAM stimuli with well-synchronized discharges at modulation frequencies 128 Hz as can been seen from the period histogram. The phase delay increased with increasing modulation frequency. At a 32-Hz modulation frequency, discharges from preceding period began to appear (Fig. 3Ba). The temporal discharge patterns in response to sFM stimuli (Fig. 3Bb) in the same unit differed markedly from those to sAM stimuli (Fig. 3Ba) in that there were two clusters of firings within each modulation period. This was because both the upward and downward trajectory of the modulation waveform excited this unit. In general, when sAM stimuli were used, discharges could be synchronized at a rate approximately equal to the modulation frequency, whereas for sFM stimuli, response synchronization could occur at a rate twice as large as the modulation frequency. Moreover, stimulus-induced synchronization was sometimes produced by one type of the modulated sounds but not by another type in an individual unit. For example, the unit in Figs. 2D and 3D responded with synchronized discharges to sFM but not sAM stimuli.

View larger version (22K): [in this window] [in a new window]   Fig. 7. A-E, a: examples of temporal modulation transfer function (tMTF). · · · (equal to 13.8), the threshold for statistically significant stimulus-synchronized responses (Rayleigh test, P < 0.001). tMTFs due to sAM (  ) and sFM (××) stimuli are calculated at fm, except in B where an additional tMTF ( - ) is calculated at the frequency equal to twice the modulation frequency (2fm). The best modulation frequencies (tBMF) calculated from tMTF (see METHODS) for the 5 examples shown are as follows (tBMFsAM, tBMFsFM). A: 6.19, 4.99 Hz. B: 13.75, 12.17 Hz. C: 16.0, 10.7 Hz. D: none, 8.51 Hz. E: 64.0, none Hz. A-E, b: examples of vector strength (VS) vs. modulation frequency (fm) profiles for the same group of units shown in a. Nonsignificant VS values were set to 0. Units in A-E correspond to those shown in Figs. 2 and 3, A-E, respectively.

POPULATION PROPERTIES. Figure 8A shows distributions of tBMF_sAM and tBMF_sFM that were analyzed based on calculations at the modulation frequency. The two distributions were statistically indistinguishable (Wilcoxon rank-sum test, P = 0.82). An important property is that the median tBMF of the population is 9.6 Hz for sAM stimuli and 10.0 Hz for sFM stimuli, respectively, which are nearly one octave lower than their counterparts derived from rMTF (median rBMF_sAM: 22.6 Hz, median rBMF_sFM: 18.1 Hz, see Table 1). Direct comparison between tBMF_sAM and tBMF_sFM in the same unit is shown in Fig. 8, B and C. Similar to the discharge rate-based analysis, a large proportion of recorded units had closely matched tBMFs produced by sAM and sFM stimuli (Fig. 8B). The difference between tBMF_sAM and tBMF_sFM was smaller than the difference between rBMF_sAM and rBMF_sFM (tBMF: median difference 0.21 octave; rBMF: median difference 0.46 octave; see Table 1). A paired t-test showed that there was no significant difference (P = 0.95) between tBMF_sAM and tBMF_sFM when both were measured in the same units. These data show that both at the level of single neurons and across populations of neurons, there was a large degree of similarity in the preferred modulation frequency as measured by stimulus-synchronized discharges. It should be noted that statistically significant stimulus-synchronized discharges were not detected in a substantial number of units studied (sAM responses: 66/200, 33%; sFM response: 67/142, 47%).

View larger version (20K): [in this window] [in a new window]   Fig. 8. Population properties for discharge synchrony-based modulation frequency selectivity. A: overlapping histograms showing distributions of tBMF derived from sAM () and sFM () stimuli, respectively. The binwidths of the histograms are on a base-2 logarithmic scale. The distributions of tBMFsAM and tBMFsFM are not statistically different from each other (Wilcoxon rank-sum test, P = 0.82). The means of the 2 distributions are 9.7 Hz (tBMFsAM) and 9.2 Hz (tBMFsFM) on the base-2 logarithmic scale and 15.6 Hz (tBMFsAM) and 14.2 Hz (tBMFsFM) on the linear scale, respectively (Table 1, columns 1-3). The medians of the 2 distributions are 9.6 Hz (tBMFsAM) and 10.0 Hz (tBMFsFM), respectively. B and C: unit-by-unit comparison between tBMFsFM and tBMFsAM recorded from the same units (Table 1, column 4). This group of 59 units was a subset of the units analyzed in A. tBMFsAM and tBMFsFM were not statistically different from each other in the unit-by-unit comparison (paired t-test, P = 0.95). The correlation coefficient (r) between tBMFsAM and tBMFsFM (B) was 0.5. The difference between tBMFsAM and tBMFsFM was calculated in octaves as log2(tBMFsAM/tBMFsFM) (C, median = 0.21 octaves). - - -, slope of 1.0.

LIMIT ON STIMULUS SYNCHRONIZED DISCHARGES. Another important measure of stimulus-synchronized discharges is the maximum synchronization frequency (f_max). This measure indicates the upper limit of stimulus-synchronized discharges in each unit, whereas tBMF defines the modulation frequency at which the strongest discharge synchronization could be induced. Figure 9 shows the distributions of f_max for both sAM and sFM stimuli, respectively. The two distributions were statistically indistinguishable (Wilcoxon rank-sum test, P = 0.97) and had medians of 34.2 Hz (sAM) and 39.4 Hz (sFM), respectively, which were much higher than their counterparts in tBMF (median tBMF_sAM: 9.6 Hz, median tBMF_sFM: 10.0 Hz, see Table 1). Figure 9B shows the cumulative distributions of f_max for both types of stimuli, which characterizes the upper boundary of stimulus-synchronized activities for the population of recorded units. These curves show how well the A1, as a whole, can represent temporal modulations by temporal discharge patterns. The cumulative distributions of f_max for both sAM and sFM stimuli are nearly identical (Fig. 9B), indicating the similarity in stimulus-synchronized discharges that resulted from these two classes of stimuli. The curves have low-pass shapes and begin to drop more rapidly above ~16 Hz. The medians of cumulative f_max distributions are between 32 Hz and 64 Hz for responses to both types of stimuli (Fig. 9B). There were <10% of units that were able to synchronize to modulation waveform at 256 Hz. Further comparison on a unit-by-unit basis between f_max measured from sAM and sFM responses is shown in Fig. 9, C and D. There was a high degree of correlation between f_max in individual units as well (r = 0.7, Fig. 9C). Many units had closely matched f_max (Fig. 9D). A paired t-test showed that there was no significant difference (P = 0.11) between f_max for sAM and sFM stimuli when both were measured in the same units.

View larger version (35K): [in this window] [in a new window]   Fig. 9. Population properties for maximum synchronization frequency (fmax). A: overlapping histograms showing distributions of fmax derived from sAM () and sFM () stimuli, respectively. The binwidths of the histograms are on a base-2 logarithmic scale. The distributions of fmax(sAM) and fmax(sFM) are not statistically different from each other (Wilcoxon rank-sum test, P = 0.97). The means of the 2 distributions are 34.2 Hz (sAM) and 32.9 Hz (sFM) on the base-2 logarithmic scale and 58.9 Hz (sAM) and 57.4 Hz (sFM) on the linear scale, respectively (Table 1, columns 1-3). The medians of the 2 distributions are 34.2 Hz (sAM) and 39.4 Hz (sFM), respectively. B: cumulative distributions of fmax based on the same populations of units shown in A. C and D: unit-by-unit comparison between fmax(sAM) and fmax(sFM) recorded from the same units (Table 1, column 4). This group of 59 units was a subset of the units analyzed in A. fmax(sAM) and fmax(sFM) were not statistically different from each other in the unit-by-unit comparison (paired t-test, P = 0.11). The correlation coefficient (r) between fmax(sAM) and fmax(sFM) (C) was 0.7. The difference between fmax(sAM) and fmax(sFM) was calculated in octaves as log2(fmax(sAM)/fmax(sFM)) (D, median = 0.14 octaves). - - -, slope of 1.0.

DOUBLING OF SYNCHRONIZATION FREQUENCY. As the example in Figs. 2B and 7B showed, some units exhibited discharge patterns that were synchronized to twice the modulation frequency (^2f_m). This was more commonly observed in the responses of sFM stimuli when the frequency component of a stimulus shifted into and out of a unit's excitatory response area during each modulation cycle. In these cases, the synchronization index calculated at 2f_m was greater than that calculated at f_m (e.g., Fig. 7B). Figure 10 shows the analysis of such cases for both classes of stimuli. There were 36 units (36/93, ~39% of samples) that exhibited synchronization frequency doubling due to sFM stimuli. In contrast, only a small number of units (7/146, ~5% of samples) were found to show this property with sAM stimuli. In the latter case, the doubling was likely caused by on and off responses to each modulation cycle. For most of the units shown in Fig. 10, f_max can be measured using either f_m or 2f_m. A higher f_max value was obtained for most of these units when 2f_m was used in the calculation (Fig. 10). In four units, f_max could only be measured using 2f_m but not f_m in their responses to sAM stimuli, indicating a nearly complete doubling of synchronization frequency (Fig. 10, larger pluses). The doubling of the synchronization frequency did not result in a significant shift of the f_max distribution when the entire population of neurons was considered.

View larger version (17K): [in this window] [in a new window]   Fig. 10. Unit-by-unit comparison between fmax(fm) and fmax(2fm) calculated from tMTF based on fm and 2fm, respectively, for a subpopulation of units that had doubling of synchronization frequency (see METHODS). fmax calculated from sAM and sFM responses are indicated by circles and pluses, respectively. Diagonal dotted line, slope of 1; dashed line, slope of 2. Units with significantly synchronized discharges at 2fm (as judged by Rayleigh statistic calculated at 2fm) but not at fm are shown as larger pluses alongside the ordinate.

Comparison between rate- and synchrony-based modulation frequency selectivity

As illustrated in Figs. 4 and 7, neurons were typically tuned to a higher modulation frequency when measured by average discharge rate than by synchronized discharges. In Fig. 11, we compared rate- and synchrony-based modulation frequency selectivity on a unit-by-unit basis when both rBMF and tBMF could be measured in the same units. For the vast majority of units, rBMF was greater than tBMF, for both sAM (Fig. 11A) and sFM (Fig. 11B) stimuli. On average, rBMF is more than twice higher than tBMF (Table 2). This means that the average discharge rate reaches the maximum at modulation frequencies as high as where the strongest stimulus-synchronized discharges could be observed. The difference between rBMF and tBMF is statistically significant (sAM: paired t-test, P < 0.001; sFM: paired t-test, P < 0.001; Table 2). The correlation coefficient between rBMF and tBMF was small (sAM: 0.31, sFM: 0.09; Table 2). In some units, f_max was found at frequencies higher than rBMF as shown in Fig. 11, C and D. Direct comparisons showed that f_max did not differ significantly from rBMF for sAM responses (paired t-test, P = 0.03, Table 2), although a significant difference was found for sFM responses (paired t-test, P <0.01, Table 2). Correlation between f_max and rBMF was poor (sAM: 0.20, sFM: 0.01; Table 2). These comparisons showed that rBMF, a discharge rate-based measurement, of a unit was not significantly correlated with discharge synchrony-based measurements (tBMF and f_max). In contrast, tBMF and f_max were highly correlated while they differed significantly (sAM: paired t-test, P < 0.001, r = 0.65; sFM: paired t-test, P < 0.001, r = 0.70; Table 2). f_max is more than three times higher than tBMF (Fig. 11, E and F, Table 2). The distributions in Fig. 11 again showed the similarity between responses to sAM and sFM stimuli.

View larger version (30K): [in this window] [in a new window]   Fig. 11. Comparison between rate- and synchrony-based modulation selectivity evaluated on the basis of individual units. A and B: tBMF is plotted vs. rBMF for sAM stimuli (A: n = 112) and sFM stimuli (B: n = 65). tBMF differed significantly from rBMF for both sAM and sFM stimuli (paired t-test, P < 0.001; Table 2). C and D: comparison between fmax and rBMF for sAM stimuli (C: n = 112, paired t-test, P = 0.03) and sFM stimuli (D: n = 65, paired t-test, P < 0.01), respectively. E and F: comparison between fmax and tBMF for sAM stimuli (E: n = 134) and sFM stimuli (F: n = 75). fmax differed significantly from tBMF for both sAM and sFM stimuli (paired t-test, P < 0.001; Table 2). - - -, slope of 1 in all plots.

Comparisons between neural populations

In Fig. 12, rate- and synchrony-based response measures were compared between different populations of units. The recorded units were partitioned into BP and non-BP groups in our analyses (see METHODS). Averaged discharge rates are plotted versus modulation frequency for both of these groups as well as for all units in Fig. 12, A (sAM) and B (sFM). The population-averaged discharge rate profiles of the BP-group showed a maximum between 16 and 32 Hz of modulation frequency (Fig. 12, A, and B, ), similar to that observed in the distributions of rBMF measured from individual units (Fig. 5A). This feature can also be seen when the responses of the entire population are averaged (Fig. 12, A and B, - - -). These observations indicate that not only were there more units tuned to modulation frequencies in the range of 16-32 Hz, the A1 responded collectively more strongly to this range of modulation frequencies than to lower or higher modulation frequencies. The profiles of the non-BP group, however, were flat between 4 and 64 Hz and showed an increase in discharge rate at higher modulation frequencies (Fig. 12, A and B,  - ).

View larger version (34K): [in this window] [in a new window]   Fig. 12. Comparisons of rate- and synchrony-based response measures between populations of units. A and B: discharge rates averaged over 3 groups of units, all (- - -), band-pass (BP, ) and non-BP () group, respectively, are plotted vs. modulation frequency for sAM (A) and sFM (B) stimuli. Spontaneous rates were not subtracted in the calculations. C and D: percent of units with statistically significant Rayleigh statistic (>13.8) is plotted vs. modulation frequency. Three groups of units, all (- - -), BP (), and non-BP () group, respectively, are shown for sAM (C) and sFM (D) stimuli.

Figure 12, C and D, showed the proportion of units that exhibited statistically significant Rayleigh statistic at each modulation frequency for the three groups of units. The highest percentages of units with significant synchronized discharges were between 4- and 16-Hz modulation frequency and were centered near 8 Hz (sAM: 48%, sFM: 32%), consistent with the distribution of tBMF of individual units (Fig. 8A). Less than half of all sampled units showed stimulus-synchronized discharges at any tested modulation frequency. These profiles reflect the overall strength of stimulus-synchronized discharges across modulation frequency that are evoked by sAM and sFM stimuli. The low percentages of units with synchronized discharges in the non-BP group were partially explained by the relatively low response magnitudes of these units (Fig. 12, A and B). Data in Fig. 12 demonstrate that the units belonging to the BP group carry far more information than units of the non-BP group in terms of both rate- and synchrony based representations of modulation frequency.

Dependency of modulation-frequency selectivity on stimulus parameters

DEPENDENCE OF MTF ON SOUND LEVEL. Rate-level functions were nonmonotonic for narrowband stimuli (e.g., pure and modulated tones) in most units that we studied in awake marmosets (Wang et al. 1999). We tested in a subset of units the dependence of modulation selectivity on sound level and found that the shape of rMTF and, to a lesser extent, the shape of tBMF was relatively invariant across supra-threshold sound levels. Figure 13, A and B, shows examples of sAM responses from two units. In each case, while changing sound level resulted in changes in firing rates, both rBMF and tBMF remained largely unchanged across a wide range of sound levels (Fig. 13, A and B). Notice that the rate-level relationship was nonmonotonic for the units shown in Fig. 13, A and B. Figure 13C shows an analysis of sound-level dependence of rBMF (Fig. 13Ca) and tBMF (Fig. 13Cb) over a population of units. For most of the units tested, changing sound level did not result in large changes in rBMF or tBMF. Similar observations were obtained with sFM responses as shown in Fig. 14, in individual examples (Fig. 14, A and B) as well as in a population analysis (Fig.14C). The unit shown in Fig. 14B had large changes in tMTF but much smaller changes in rMTF across sound levels. In general, the shape of rMTF, and consequently rBMF, tended to be more resistant to changes in sound level than did tMTF and tBMF. These data showed that although changing sound level may change the peak firing rate and the sharpness of tuning to the preferred modulation frequency, it usually did not result in significant shift of MTF. There appeared to be, however, greater changes due to sound level in sFM responses than in sAM responses.

View larger version (29K): [in this window] [in a new window]   Fig. 13. Effects of sound level on rMTF (left, a) and tMTF (right, b) for sAM stimuli. A and B: examples of MTFs measured at multiple sound levels in 2 representative units. · · · (equal to 13.8), the statistical significance threshold for Rayleigh test (P < 0.001). Stimulus parameters are indicated on each plot. C: summary of population properties to changes in sound level for rBMF (a) and tBMF (b), respectively. Top: the mean () and the range of minimum and maximum (|) rBMF (a) or tBMF (b) measured in each unit. Units are ordered by their mean rBMF or tBMF, respectively (from low to high). Corresponding ranges of sound levels tested for each unit are shown in bottom plots. dsAM, modulation depth for sAM stimuli. At least two sound levels were tested for each data point shown in C.

View larger version (28K): [in this window] [in a new window]   Fig. 14. Effects of sound level on rMTF and tMTF for sFM stimuli. The format is the same as Fig. 13. dsFM, modulation depth for sFM stimuli.

DEPENDENCE OF MTF ON MODULATION DEPTH. Modulation depth was another important parameter that affected the responsiveness of a unit to modulated sounds. In general, increasing modulation depth of a sAM stimulus at or near rBMF always led to a monotonic change in discharge rate (increased or saturated). For the unit shown in Fig. 15A, the maximum firing rate of the rMTF increased from ~13 to ~38 spikes/s as modulation depth of the sAM stimuli was increased from 50 to 100%. However, the rBMF remained unchanged near 32 Hz. This was also true for the synchronization measure, with tBMF near 16 Hz (Fig. 15Ab). Similar properties can be seen in another example in Fig. 15B. Figure 15C shows a population analysis that further strengthens these observations regarding responses to sAM stimuli.

View larger version (27K): [in this window] [in a new window]   Fig. 15. Effects of modulation depth on rMTF and tMTF for sAM stimuli. The format is the same as Fig. 13. dsAM, modulation depth for sAM stimuli.

View larger version (32K): [in this window] [in a new window]   Fig. 16. Effects of modulation depth on rMTF and tMTF for sFM stimuli. The format is the same as Fig. 13. dsFM, modulation depth for sFM stimuli.

Temporal modulation is essential to drive many cortical neurons

Temporal modulation (in amplitude or frequency) was essential to drive many units that were unresponsive or only weakly responsive to pure tones. Figure 17, A and B, shows two representative examples of a class of units that required sufficient AM to fire. The unit in Fig. 17A gave only onset discharges at zero modulation depth (equivalent to a CF tone) but gave sustained discharges when the modulation depth of the sAM stimulus was raised to 50%. Figure 17B shows another example that had offset discharges at modulation depths <70-80% but fired continuously throughout the stimulus duration at greater modulation depths. Discharge rate versus modulation depth functions from a group of such units is shown in Fig. 17C. These units had weak or no responses to sAM stimuli at zero modulation depth, i.e., unmodulated CF tones. Some units did not respond to pure tones until they were sufficiently modulated in frequency, as shown by examples in Fig. 17, D-F. In contrast to sAM stimuli, sFM stimuli generally produced maximum firing rate at a particular modulation depth in each unit. Increasing modulation depth beyond this optimal depth could result in a reduction in firing rate (Fig. 17F), presumably due to the recruitment of flanking inhibitions by expanded spectral side bands.

View larger version (38K): [in this window] [in a new window]   Fig. 17. Dependence of cortical responses on amplitude or frequency modulation. Examples of units that responded weakly to unmodulated tones but strongly to sAM (left, A-C) or sFM (right, D-F) stimuli with sufficient modulation depths. A and B: 2 representative examples of units that responded to sAM stimuli during the stimulus duration only at sufficiently large modulation depths. Onset (A) or offset (B) responses were present at all modulation depths. C: normalized discharge rate is plotted vs. modulation depth for a group of units. Mean firing rate at 100% modulation depth was used for the normalization. Maximum firing rate in each unit was usually achieved at or near the largest modulation depth tested. D and E: 2 representative examples of units that responded to sFM stimuli during the stimulus duration only at sufficiently large modulation depths. F: firing rate, normalized by the maximum discharge