Representation of Spectral and Temporal Sound Features in Three
Cortical Fields of the Cat. Similarities Outweigh Differences
JOS J. EGGERMONT
Departments of Physiology and Biophysics, and Psychology, The University of Calgary, Calgary,
Alberta T2N 1N4, Canada
Eggermont, Jos J. Representation of spectral and temporal sound
features in three cortical fields of the cat. Similarities outweigh differences. J. Neurophysiol. 80: 2743–2764, 1998. This study investigates
the degree of similarity of three different auditory cortical areas with
respect to the coding of periodic stimuli. Simultaneous single- and
multiunit recordings in response to periodic stimuli were made from
primary auditory cortex (AI), anterior auditory field (AAF), and
secondary auditory cortex (AII) in the cat to addresses the following
questions: is there, within each cortical area, a difference in the
temporal coding of periodic click trains, amplitude-modulated (AM)
noise bursts, and AM tone bursts? Is there a difference in this coding
between the three cortical fields? Is the coding based on the temporal
modulation transfer function (tMTF) and on the all-order interspikeinterval (ISI) histogram the same? Is the perceptual distinction between rhythm and roughness for AM stimuli related to a temporal
versus spatial representation of AM frequency in auditory cortex?
Are interarea differences in temporal response properties related to
differences in frequency tuning? The results showed that: 1) AM
stimuli produce much higher best modulation frequencies (BMFs)
and limiting rates than periodic click trains. 2) For periodic click
trains and AM noise, the BMFs and limiting rates were not significantly different for the three areas. However, for AM tones the BMF
and limiting rates were about a factor 2 lower in AAF compared
with the other areas. 3) The representation of stimulus periodicity in
ISIs resulted in significantly lower mean BMFs and limiting rates
compared with those estimated from the tMTFs. The difference was
relatively small for periodic click trains but quite large for both AM
stimuli, especially in AI and AII. 4) Modulation frequencies õ20
Hz were represented in the ISIs, suggesting that rhythm is coded in
auditory cortex in temporal fashion. 5) In general only a modest
interdependence of spectral- and temporal-response properties in AI
and AII was found. The BMFs were correlated positively with characteristic frequency in AAF. The limiting rate was positively correlated
with the frequency-tuning curve bandwidth in AI and AII but not in
AAF. Only in AAF was a correlation between BMF and minimum
latency was found. Thus whereas differences were found in the frequency-tuning curve bandwidth and minimum response latencies
among the three areas, the coding of periodic stimuli in these areas
was fairly similar with the exception of the very poor representation
of AM tones in AII. This suggests a strong parallel processing organization in auditory cortex.
INTRODUCTION
Differences and similarities in the representation of both
spectral and temporal aspects of sound have been demonstrated in about half a dozen auditory cortical areas. DifferThe costs of publication of this article were defrayed in part by the
payment of page charges. The article must therefore be hereby marked
‘‘advertisement’’ in accordance with 18 U.S.C. Section 1734 solely to
indicate this fact.
ences support the idea that auditory cortex is organized functionally in a segregated manner, i.e., where different areas
respond to different aspects of sound. As a result of such
differences, two segregated pathways, similar to the ‘‘what’’
and ‘‘where’’ pathways in the visual system, have been proposed for monkey auditory cortex (Rauschecker 1997).
Similarities emphasize that multiple cortical representations
of certain stimulus properties are preserved thereby allowing
a synthesis of spatially distributed responses. In addition,
local integration of the property that is preserved in each
area with other sound attributes that are mapped differently
in these areas can be performed independently.
Spectral stimulus properties
In the cat, the primary auditory cortex (AI) and the anterior auditory field (AAF) show very similar single-unit frequency tuning (Knight 1977; Phillips and Irvine 1982). In
contrast, the depth-recorded local field potential in AI of the
cat was more sharply tuned than in AAF (Knight 1977). In
AI and AAF, the percentages of monotonic and nonmonotonic units was also very similar (Kowalski et al. 1995;
Phillips and Irvine 1982) but AAF featured multipeaked
frequency response areas that rarely were observed in the
central and ventral parts of AI but frequently were seen in
the dorsal part (Knight 1977; Sutter and Schreiner 1991;
Tian and Rauschecker 1994). In the ferret, the single-unit
frequency-tuning curve bandwidth at 20 dB above threshold
in AAF (mean 2.5 octaves) was approximately twice as
large as that in AI (mean 1.3 octaves) (Kowalski et al.
1995). In the Mongolian gerbil, no differences were found
in the responses to tone bursts and slow FM sweeps in AI
and AAF (Schulze et al. 1997).
Secondary auditory cortex (AII) in cat appeared not as
well organized tonotopically as AI, and the units showed
broader frequency-tuning and higher thresholds to tone-burst
stimulation than in AI. In parts of AII, responses were more
sustained than in AI and latencies could be ú100 ms; however, phasic responses with relatively short latencies ( õ50
ms) were also common (Schreiner and Cynader 1984). Recently a double representation of the cochlea within the classical boundaries of AII in cat was demonstrated (Volkov and
Galazyuk 1998). One tonotopic representation was found in
the dorsocaudal region (2.6–3.2 mm long) with a spatial
orientation similar to that described in AI (low frequencies
caudal, high frequencies rostral), the second ventrorostral
region was smaller (1.4–2.5 mm) and had the opposite tonotopic orientation (i.e., as in AAF). The dorsocaudal region
0022-3077/98 $5.00 Copyright q 1998 The American Physiological Society
/ 9k2e$$no40 J364-8
10-27-98 16:59:01
neupa
LP-Neurophys
2743
2744
J. J. EGGERMONT
may be close to the transition zone between AI and AII as
defined by Schreiner and Cynader (1988). Thus AII, not
unlike AI, may consist of several specialized subregions.
Latencies
Conflicting reports exist about the minimum latencies in
AI and AAF. Whereas Knight (1977) and Phillips and Irvine
(1982) found the same latencies in the cat, Kowalski et al.
(1995) reported a mean AAF latency of 16.8 ms that was
substantially shorter than the mean latency in AI of 19.4 ms.
These latencies were, however, several milliseconds longer
than found in cat AI (mean 12.5 ms) by Raggio and
Schreiner (1994) and Mendelson et al. (1997) but comparable with those (mean 17.7 ms) reported for free-field studies
by Eggermont (1996). In cat, AI and AAF receive different
anatomic projections from the auditory thalamus: AI receives the heavier projections from the ventral medial geniculate body (MGB), whereas AAF receives the heavier projections from the posterior group of thalamic nuclei, PO
(Morel and Imig 1987). AII receives its input largely from
the caudal dorsal nucleus of the MGB (Andersen et al. 1980)
and is part of the extra lemniscal pathway. AI, AAF, and
AII also receive input in layer I from the nontopographical
organized medial part of the MGB (Winer 1992). This different innervation may in part be responsible for the observed differences in latencies and frequency tuning.
Temporal stimulus properties
Periodic amplitude modulated (AM) sounds such as sineor square-wave modulated tone bursts also are represented
differently in separate cortical fields in cat (Schreiner and
Urbas 1988). The average best modulation frequency
(BMF) in AAF of the paralyzed and lightly barbiturate anesthetized cat was higher than in AI. This was largely the result
of higher BMFs ( °100 Hz) for units with characteristic
frequencies (CFs) ú10 kHz, whereas for lower CFs, all
BMFs were õ20 Hz and similar to those in AI. In AII, a
range of ‘‘normal’’ BMFs close to 10 Hz was found but
supplemented by a large group of BMFs with values õ5
Hz. Such low BMFs were found less frequently in AI or
AAF (Schreiner and Urbas 1988).
Comparison of the effect of various periodic stimuli such
as sinusoidal AM and rectangular AM of tones at the CF
(Schreiner and Urbas 1986, 1988) with periodic click trains
(Eggermont 1991, 1993; Schreiner and Raggio 1996) in
cat AI suggests that the BMFs are highest for sinusoidal
modulation than for rectangular modulation and lowest for
periodic click trains. Phillips et al. (1989) used continuous
repetitive tone pips and analyzed them in terms of the response per tone pip, i.e., in terms of entrainment. Eggermont
(1991) showed a similar analysis for periodic click stimulation and Schreiner and Raggio (1996) also presented their
data in this alternative format in addition to using the temporal modulation transfer function (tMTFs). These entrainment functions are low-pass functions of click (tone pip)
repetition rate and were very similar in these three studies
(Eggermont 1997).
Internal neural representations
Regardless of whether entrainment functions or tMTFs
are used, these measures are constructed on basis of full
/ 9k2e$$no40 J364-8
knowledge of the stimulus presented to the animal. The stimulus is, of course, unknown to the animal, and its nervous
system cannot extract temporal information from aspects
such as the stimulus onset and repetition rate. As a consequence, period histograms, as used in the construction of the
tMTFs, cannot be computed by the nervous system. However, information similar to that in period histograms is
available in the interspike interval (ISI) histograms (Horst et
al. 1986) or autocorrelograms (Cariani and Delgutte 1996a).
These internal neural representations can be used to extract
stimulus periodicity information and may even be preferred.
As Cariani and Delgutte (1996b) remarked ‘‘period histograms or their derived measures (synchronization and modulation indexes), for example, detect only stimulus-locked
time patterns, and will miss more subtle, asynchronous codes
. . . . Until better methods are used to study responses to
complex stimuli in more central auditory stations, complex
temporal pattern codes cannot be ruled out.’’ In response to
harmonic stimuli, the most frequent ISI present in a population of auditory nerve fibers appears to correspond to the
perceived pitch (Cariani and Delgutte 1996a). This ISI calculation can be done for individual neurons and then pooled
across the entire population or across a local group of neurons recorded on the same electrode. A direct comparison
between tMTFs and ISI histograms for cortical coding of
periodic stimuli has so far not been made. It is one of the
purposes of this study to compare this for three sets of stimuli: periodic click trains, amplitude-modulated noise bursts
(AM noise), and amplitude-modulated tone bursts (AM
tone) in three areas: AI, AAF, and AII.
Anesthesia, cortical rhythms, and temporal coding
Recently, we (Kenmochi and Eggermont 1997) reported
a strong correlation between the dominant oscillation (spindle) frequency in the spontaneous local field potential (LFP)
and the BMF for periodic click train stimulation in primary
auditory cortex. The mean spontaneous rhythm was generally in the frequency range of 8–14 Hz but could be as high
as 30 Hz, and one expects this LFP rhythm to be similar in
other auditory cortical fields. This suggests that the BMFs
in AI, AAF, and AII will be very similar when recorded
under identical anesthesia levels. Spontaneous EEG rhythms
are strongly dependent on the level of arousal or drowsiness,
thus one expects the tMTFs to be similarly effected by
changes in anesthesia level. However, Schulze and Langner
(1997) did not find substantial effects of anesthesia on coding of AM tones in AI in the gerbil.
In awake squirrel monkeys, Bieser and Müller-Preuss
(1996) investigated periodicity coding for AM sounds in
eight auditory cortical areas and found a broad range of
BMFs ranging from 2 to 128 Hz. Low modulation frequencies (2–64 Hz) produced mostly phase-locked neural responses whereas higher AM (128–512 Hz) sounds showed
only a distinction in overall-spike-rate variations. For instance in field Pi, the highest firing rates were found for
modulations of 128 Hz, whereas in field T1 they were highest
for 256 Hz. Specifically in fields AI, Pi, and T1 the sensitivity to AM frequency was large enough to encode envelope
fluctuations (modulation rates between 4 and 64 Hz) found
in monkey calls. Steinschneider et al. (1980, 1982) found
10-27-98 16:59:01
neupa
LP-Neurophys
TEMPORAL CODING IN THREE AUDITORY CORTICAL AREAS
phase-locked activity in the depth-recorded local field potentials in auditory cortex of awake macaques up to 250 Hz
and for multiunit spike activity °100 Hz.
For periodic click trains, Eggermont (1993) found for
ketamine-anesthetized cats a best repetition rate of 7.9 {
2.1 (SD) Hz, and in barbiturate-anesthetized cats Schreiner
and Raggio (1996) obtained a similar value of 6.5 { 3.8
Hz. A comparison of BMFs for AM sound in cat (Eggermont
1994, 1997; Schreiner and Urbas 1988) with those from
awake squirrel monkeys (Bieser and Müller-Preuss 1996)
and awake macaques (Steinschneider et al. 1980, 1982) suggests that anesthesia affects the BMF. This conclusion also
was drawn by Goldstein et al. (1959) and Eggermont and
Smith (1995a) on the basis of evoked potential (EP) studies.
Under barbiturate anesthesia the limiting rate (50% of maximum EP amplitude) in cats was 6–8 Hz, down from 15–
20 Hz in awake cats (Goldstein et al. 1959), whereas for
light ketamine anesthesia the limiting rate was only slightly
õ15 Hz (Eggermont and Smith 1995a).
Comparing the results in awake and anesthetized animals
suggests that level and type of anesthesia are likely having
an effect, potentially by influencing the frequency range of
the dominant autonomous cortical rhythms. It is thus desirable to eliminate potential effects of differences in anesthesia
level between recordings when comparing spectral and temporal coding properties in various cortical fields. For this
purpose, simultaneous recordings were made from AI, AAF,
and AII in lightly ketamine-anesthetized cats.
Perceptual attributes of periodic sounds
Amplitude-modulated tones or noise produce various
hearing sensations depending on the modulation frequency.
These are rhythm and fluctuation strength for AM frequencies below Ç20 Hz, and roughness and pitch for AM frequencies ú20 Hz. The sensation of roughness disappears
ú300 Hz and is strongest at 70 Hz, whereas pitch related
to periodicity in the stimulus envelope loses much of its
perceptual strength ú3 kHz (Zwicker and Fastl 1990). The
click repetition rates and AM frequencies used in this study
straddle the boundary between rhythm and roughness, and
the limiting rate of phase-locked cortical responses may be
related to that perceptual boundary. All results obtained for
AM stimuli in auditory cortex so far point to a potential
temporal code up to frequencies of Ç20–30 Hz (Schulze
and Langner 1997). However, no direct evidence for an
internal temporal code based on ISIs so far has been demonstrated in auditory cortex nor has its AM frequency range
been determined.
This study addresses the following questions: 1) is there,
within each cortical area, a difference in the temporal coding
of periodic click trains, AM noise bursts, and AM tone
bursts? 2) Is there a difference in this coding among the
three cortical fields? 3) Is the repetition rate range of temporal coding as inferred from stimulus referenced measures
such as the tMTF and internal neural representations such
as the all-order ISI histogram the same? 4) Is the perceptual
distinction between rhythm and roughness for AM stimuli
related to the upper limit of a temporal representation of
AM frequency in auditory cortex? And 5) are interarea differences in temporal response properties related to differences in frequency tuning and response latencies?
/ 9k2e$$no40 J364-8
2745
METHODS
The care and the use of animals reported on in this study was
approved (No. P88095) by the Life and Environmental Sciences
Animal Care Committee of the University of Calgary.
Animal preparation
Cats were premedicated with 0.25 ml/kg body wt of a mixture
of 0.1 ml acepromazine (0.25 mg/ml) and 0.9 ml of atropine
methyl nitrate (5 mg/ml) subcutaneously. After Ç0.5 h, they received an intramuscular injection of 25 mg/kg of ketamine (100
mg/ml) and 20 mg/kg of pentobarbital sodium (65 mg/ml). Durocain (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 parts of AI and AII.
A 4-mm hole was drilled over the AAF. The dura was left intact,
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 secured with the screw. Additional acepromazine/atropine
mixture was administered every 2 h. Light anesthesia was maintained with intramuscular injections of 2–5 mgrkg 01rh 01 of ketamine. The wound margins were infused every 2 h with durocain,
and also every 2 h new mineral oil was added if needed. The
temperature of the cat was maintained at 377C. 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 in front of the cat’s head. The sound-treated
room was made anechoic for frequencies ú625 Hz by covering
walls and ceiling with acoustic wedges (Sonex 3’’) 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
bursts, and clicks was used to locate units. Characteristic frequency
(CF) and tuning curve of the individual neurons were determined
with 50-ms-duration, gamma-shape-envelope, tone pips 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. After the frequencytuning properties of the cells at each electrode were determined,
periodic click trains (1-s duration followed by 2 s of silence) and
amplitude-modulated noise or tone bursts (0.5-s duration followed
by 2.5 s of silence) were presented once per 3 s. The modulation
frequencies were between 1 and 32 (click trains) or 2 and 64 Hz
(AM sounds) at logarithmically equal distance with four values
per octave and were presented randomly. The AM wave form was
an exponentially transformed sine wave (Epping and Eggermont
1986) with a maximum modulation depth of 17.4 dB so that the
envelope, when expressed in dB, was sinusoidally modulated. The
sequences of 21 click trains or 21 AM noise or AM tone bursts were
repeated 10 times resulting in a total stimulus ensemble duration of
630 s per stimulus type. The click trains and AM sounds were
presented at peak intensities of 35, 55, and 75 dB SPL, and results
for the intensity where the firing rate was the highest are presented.
Clicks consisted of single polarity electric pulses 0.1 ms in duration.
10-27-98 16:59:01
neupa
LP-Neurophys
2746
J. J. EGGERMONT
Recording and spike separation procedure
Three tungsten micro electrodes (Micro Probe) with impedances
between 1.5 and 2.5 MV were advanced independently perpendicular to the AI, AAF, and AII surfaces using remotely controlled
motorized hydraulic microdrives (Trent-Wells Mark III). The electrode signals were amplified using extracellular preamplifiers (Dagan 2400) and filtered between 200 Hz (Kemo VBF8, high-pass,
24 dB/oct) and 3 kHz (6 dB/oct, Dagan roll-off) to remove local
field potentials. The signals were sampled through 12 bit A/D
converters (Data Translation, DT 2752) into a PDP 11/53 microcomputer, together with timing signals from three Schmitt-triggers.
In general the recorded signal on each electrode contained activity
of two to four neural units. The PDP was programmed to separate
these multiunit spike trains into single-unit spike trains using a
maximum variance algorithm (Eggermont 1992). The spikes from
well-separated wave-form classes, each assumed to represent a
particular neuron, were stored and coded for display. The multiunit
data presented in this paper represent only well-separated single
units that, because of their regular spike wave form, likely are
dominantly from pyramidal cells (Eggermont 1996). Thus contrary
to the common use of the term multiunit as a cluster of not well
separable units, in this analysis the separable single-unit spike
trains extracted from the multiunit recording were added again to
a form a multiunit spike train with better statistics and that likely
consists of contributions from only one type of neurons.
In addition, the electrode signals were band-pass filtered between
10 and 100 Hz to obtain spike-free signals of ongoing local field
potentials (LFP). These signals also were passed through Schmitttriggers set at Ç2 SD (i.e., at about 0100 mV) below the mean
value of the ongoing signal during silence. The ‘‘spikes’’ of these
LFPs were processed in the same way as single-unit spike data.
We have shown previously that these level crossings represent
most of the temporal (Eggermont and Smith 1995a) and spectral
(Eggermont 1996) response properties of the single units recorded
at the same electrode.
The boundary between AI and AAF was explored by taking a
series of LFP and multiunit measures (with the high-pass filter set
at 10 Hz with 6 dB/oct) from caudal to rostral and assuring that
there was a gradual increase in CF, which reversed in direction
when advancing to the AAF. The AII was identified anatomically
and electrophysiologically based on the broader tuning curves and
different response patterns compared with those in the central and
ventral parts of AI. Recordings in AII were generally made from
the ventrorostral part. Recording electrode positions in the three
cortical areas were chosen such that recordings with approximately
similar best frequencies (within 0.5 octave) at 50–70 dB SPL were
obtained. Recordings were made between 600 and 1,200 mm below
the cortex surface.
Data analysis
The number of action potentials in the first 100 ms after each
tone-pip presentation were counted for each intensity. 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 5 octaves so
that the final resolution was Ç0.2 octaves. The results per stimulus
intensity were combined into a rate-frequency-intensity profile
from which tuning curves, rate-intensity functions, and iso-intensity-rate contours could be derived (Eggermont 1996). The frequency-tuning curve was defined for a firing rate at 25% of the
maximum firing rate. The threshold was determined as 2.5 dB
below the intensity that produced visible time locked responses to
the tone pip, i.e., midway between the stimulus that produced a
response and the one that did not. The tuning curve bandwidth was
measured at 20 dB above threshold and expressed in octaves.
/ 9k2e$$no40 J364-8
Latency measures were corrected for the acoustic travel time
from the speaker to the cat’s ears by subtracting 1.6 ms (speaker
distance 550 mm divided by the speed of sound 340 mm/ms).
The following capacity of the neurons for click repetition rate
and AM frequency was estimated from the temporal modulation
transfer functions and ISI histograms. The temporal modulation
transfer functions were obtained by Fourier transformation of the
period histograms (Eggermont 1991). Each modulation period was
divided into 16 bins, and only recordings with at least five counts
in the maximum bin per 10 stimulus presentations at a rate of 8 Hz
were analyzed further. The rate-tMTF was defined as the number of
spikes per 10 presentations of the 1-s duration click train or 0.5s-duration modulated noise or tone burst, as a function of click
repetition rate or modulation frequency. The synchronization-tMTF
was defined as the amplitude of the first harmonic of the period
histogram, as a function of the click-repetition rate or AM frequency. The BMF was defined as the click or AM rate for which
the synchronization-tMTF was maximal. The limiting rate was
defined as the highest rate at which the response was 50% of
that at the BMF. The significance of the vector strength for each
modulation frequency (defined as P õ 0.005) was calculated using
the Raleigh test (Mardia 1972).
ISI histograms and autocorrelograms for lags up to three periods
were constructed with 16 bins per stimulus period for each individual click rate or AM frequency. The accuracy was thus proportional
to the stimulus period, and the capacity to represent click repetition
rates or AM rates in an interval code is reflected in a peak in the
same position in the histograms regardless of the click repetition
rate or AM frequency. Population autocorrelograms and interval
histograms were obtained by adding all the individual single-unit
ones. Multiunit (MU) autocorrelograms per recording site were
also calculated, they reflect both single-unit autocorrelograms as
well as cross-correlograms between units on the same electrode.
The autocorrelograms is formally identical to the all-order ISI histogram, i.e., formed by adding the interval histogram for adjacent
spikes, for one but adjacent spikes, etc., (Cariani and Delgutte
1996a) and I use the terms interchangeably.
All statistical analyses were performed using Statview 4.5. Illustrations were made with Horizon and Powerpoint Software.
RESULTS
Results presented for periodic click trains are from 62
simultaneous recordings with an electrode in each of AI,
AAF, and AII in 14 cats, resulting in 186 MU records and
the same number of LFP records. The MU records could be
separated into 552 single-unit (SU) spike trains of which
250 had large enough firing rates to permit analysis of the
tMTF (as defined in METHODS ). In nine of these cats, responses also were obtained for AM noise and AM tones for
a total of 102 MU records (separable into 156 SU spike trains
with sufficient firing rates). We also made 23 simultaneous
recordings for click train stimulation with two electrodes in
AI and one in AAF resulting in 69 MU records (88 single
units with sufficient firing rates) and 69 LFP recordings in
seven additional cats. In 27 of these recordings, we also
stimulated with AM noise and AM tones (27 MU records
resulting in 33 single-unit spike trains with sufficient firing
rates). The total number of single units presented in this
study is 338 (of 740 single units recorded from) in 21 cats.
In the MU data, all units with well-defined spike wave form
were included regardless of their firing rate.
Representative examples for periodicity coding
Figure 1 shows dot displays from electrode sites in AI (left),
AAF (middle), and AII (right) with similar CF of Ç5 kHz in
10-27-98 16:59:01
neupa
LP-Neurophys
TEMPORAL CODING IN THREE AUDITORY CORTICAL AREAS
2747
FIG . 1. Dot displays for simultaneous
local field potential (LFP) and single-unit
(SU) activity in primary auditory cortex
(AI), anterior auditory field (AAF), and
secondary auditory cortex (AII) in response
to periodic click trains, and amplitude-modulated noise and tone bursts (5 kHz). LFP
triggers are indicated by the orange dots,
SU spikes by blue, green, or black dots.
Click repetition rate (1–32 Hz) and the AM
frequencies (2–64 Hz) are represented on a
logarithmic axis. First 600 ms after stimulus
(click train or AM burst) is shown. characteristic frequencies (CFs) of the recording
sites were 5.2 kHz (AI), 6 kHz (AII), and
6 kHz (AAF) This example illustrates the
consistent modulation following of the LFP
triggers in all areas. SU responses besides
AM following include OFF responses (AAF
for AM noise at high AM frequencies) and
rebound activity related to spindle rhythms
(AII, AM noise).
response to periodic click trains (top), AM noise bursts (center
row), and AM tone bursts (carrier 5 kHz; bottom). Click
repetition rates were between 1 and 32 Hz, whereas AM frequencies were between 2 and 64 Hz. Both LFP triggers (orange
dots) and single units (green, blue, or black dots) are shown
for a time base of 600 ms after stimulus onset. As a consequence only the response to the first half of the click train is
shown. The recording site in AI did not show strong singleunit responses to click trains (and is therefore not representative
of our sample of AI units) but good following for the LFP.
For AM stimuli, however, three single units were responding
up to AM rates of 16 Hz, with the stronger responses for AM
noise. No onset unit responses coinciding with the LFP triggers
were noted for AM noise bursts and unit onset responses were
limited to modulation frequencies between 8 and 32 Hz for
AM tone bursts. For AAF, the LFP triggers showed similar
following as in AI, whereas the unit activity was much more
pronounced especially for the AM stimuli. The best following
was found for AM tone bursts ( °32 Hz) albeit that the more
vigorous responses were obtained for AM noise burst stimulation. A pronounced OFF response is noted for noise with AM
frequencies of 16–64 Hz. In AII, LFP activity showed weak
following as did most unit activity, and the strong long latency
unit activity for noise bursts was largely independent of AM
rates and likely a rebound response after initial suppression of
firing.
The second example (Fig. 2) from recording sites, with
a best frequency of Ç2.5 kHz in AI and AAF and 3–7 kHz
(broadly tuned) in AII, illustrates preferential locking of the
unit firing to the modulation frequency for AM sounds in
AI, to the broadband sounds (click train and AM noise) in
AAF and to the AM tones (2.5 kHz) for AII. The LFP
triggers in AI and AAF are found to follow repetition rates
for clicks up to Ç10 Hz and with skipping one period °16
Hz. AM rate following was slightly higher: up to Ç16 and
20 Hz with skipping one period. OFF responses again were
noted for AM noise stimulation but this time in AI.
/ 9k2e$$no40 J364-8
From these dot displays the next step is to generate period
histograms and from these the rate tMTF (mean number of
spikes per 10 stimuli), the vector strength (Goldberg and
Brown 1969), and the synchronization tMTF [the product of
rate and vector strength (VS)]. Figure 3 presents this for three
MU conditions shown in Fig. 2. Figure 3A present results
for the AAF recording in response to click trains, the period
histogram (the 1st 2 panels show this twice, useful to track
peak phase changes that exceed 1 period as in Fig. 3C) illustrates the gradual shift of the peak response in its relative
position in the click-repetition period (i.e., its phase). The
third panel shows the rate tMTF, and an abrupt drop in the
response is noted ú9.56 Hz. The VS (4th panel) appears
significant for all repetition rates except at 11.3 and 32 Hz.
Finally, the synchrony tMTF peaks at 4.76 Hz (its BMF)
although it is only marginally larger than at 8 Hz. The limiting
rate (50% of the value at BMF) is found for 9.52 Hz and
coincides with the abrupt drop in firing rate. Figure 3B shows
the same analysis for AM noise in the AI recording. The nearly
constant phase of the peak response in the period histograms
for AM frequencies õ8 Hz forms a clear exception to the
normal monotonic increase of phase. The abrupt drop in mean
firing rate at 8 Hz results in both a BMF and limiting rate at
6.72 Hz. Figure 3C shows one of the infrequent instances
where a recording in AII showed good locking to the period
of AM tones. The mean number of spikes per 10 stimuli
decreases monotonically with AM frequency, the VS shows
a band-pass dependence on AM frequency, the BMF is 9.52
Hz, and the limiting rate is 11.3 Hz.
Group data
DISTRIBUTION OF BEST MODULATION FREQUENCIES AND LIMITING RATES FOR PERIODIC CLICK TRAINS. The BMFs for
multiunits in response to periodic click train stimulation were
approximately normally distributed (Fig. 4A). No significant effect of area was noted (Table 1, paired t-test, P ú
10-27-98 16:59:01
neupa
LP-Neurophys
2748
J. J. EGGERMONT
FIG . 2. Dot displays for simultaneous
LFP and SU activity in AI, AAF, and AII
in response to periodic click trains, and
amplitude-modulated noise and tone bursts
(2.5 kHz). Format is similar to Fig. 1. CFs
of the recording sites were Ç2.5 kHz in
AI and AAF and 3–7 kHz in AII (broadly
tuned).
0.08). For individual single units, the distribution of BMFs
generally was comparable with that for the MU data (Fig.
4B) but there were relatively more values in the 2- to 3-Hz
range. There were no significant differences between the
BMF values for single units with those for the MU data
(Table 1). The BMF value for the tMTF based on LFP
triggers (Table 1) was significantly larger (paired t-test, P Å
0.008) than that for MU spikes in AI and also in AII (paired
t-test, P Å 0.04) but not in AAF. The BMFs for LFP triggers
and MU spikes were correlated positively (R Å 0.205; Fisher’s r to z test, P Å 0.0025).
The distribution for the limiting rates for multiunit is
shown in Fig. 5A. No effect of area was found on the mean
values (Table 1, paired t-test, P ú 0.5). For single units,
the distribution of limiting rates was generally more predominant in the 9- to 12-Hz range than for the MU data (Fig.
5B). Again higher mean values were found for LFP triggers
but these were not significantly different between areas. For
all areas, the limiting rates were higher for the LFPs than
for the corresponding MUs (paired t-test, P õ 0.05). The
limiting rates for both measures were correlated positively
(R Å 0.139; Fisher’s r to z test, P Å 0.034).
Besides taking all LFP triggers, a separate analysis was
run on only the first LFP trigger after a click (LFP1) to
avoid effects of secondary LFP triggers (within 50 ms of
the first, as shown in Fig. 2) that are never accompanied by
spikes. The BMFs for these LFP1 triggers were not different
from those where all triggers were used, however, the limiting rate for LFP1 triggers was significantly higher in AI
(19.3 vs. 17.2 Hz; paired t-test, P õ 0.0001).
DISTRIBUTION OF BEST MODULATION FREQUENCIES AND LIMITING RATES FOR AM STIMULI. Best modulation frequencies
and limiting rates for MU recordings were not significantly
different for AM noise and AM tone stimulation (Table 1;
analysis of variance with post hoc Scheffe test, at the P õ
0.05 level) and therefore the data were combined in the
frequency distributions shown in Fig. 6. Most BMFs (Fig.
/ 9k2e$$no40 J364-8
6A) were found for AM frequencies õ20 Hz in all three
areas but in contrast to click train stimulation, BMFs ú20
Hz were found as well. As a result, the standard deviation in
the BMFs is very large and the modal BMFs are significantly
smaller than the mean (Table 1). Limiting rates (Fig. 6B)
were relatively uniformly distributed between 2 and 64 Hz.
As shown in Table 1, in each cortical area, the BMFs
obtained for LFP triggers were generally similar to those for
MU spikes, and the only significant difference was that the
LFP measure was significantly higher than that for MUs in
AAF for AM tone stimulation (paired t-test, P õ 0.05). The
limiting rates were significantly higher for LFP triggers than
for MU spikes for AAF (both AM noise and AM tone, P õ
0.05) and AII (AM noise only, P õ 0.05). In AAF the
BMFs and limiting rates were significantly (P õ 0.05) lower
than those for AI and AII, whereas there was no difference
between AI and AII.
Limiting rates between AM stimuli were not significantly
different within a cortical area, but for both AM stimuli they
were significantly (P õ 0.0001) higher than for click train
stimulation. This was found for all cortical areas, both for
MU spikes and LFP triggers. The BMFs for LFP triggers
were significantly higher for AM stimuli than for clicks in
AI and AII but not in AAF. For MU spikes, the BMFs were
significantly higher for AM stimuli than for clicks in AI and
AII, but again not in AAF.
MEAN TEMPORAL MODULATION TRANSFER FUNCTIONS FOR PERIODIC CLICK TRAINS. Across all recordings, the mean num-
ber of spikes during the click trains was computed for LFP
triggers, LFP1 triggers, and MU spikes and shown in Fig. 7.
This figure thus represents the rate-tMTFs. The LFP triggers
show a weak band-pass dependence on click rate, the LFP1
triggers show a high-pass dependence (as a consequence of
only taking at most 1 trigger per click) and the MU spike
rate is also weakly band-pass. The highest number of spikes
per click train on average was obtained in AAF.
The mean VS for the three response measures is shown
10-27-98 16:59:01
neupa
LP-Neurophys
TEMPORAL CODING IN THREE AUDITORY CORTICAL AREAS
2749
FIG . 4. Distributions of the BMFs for multiunit (MU) and SU activity
in response to periodic click trains in the 3 areas. A: BMFs are very similarly
distributed for MUs. B: same for single units.
per cortical area in Fig. 8. The general trend is the same for
each area and shows a very high VS for LFP1 triggers at
low click rates, because of the large binwidth (1/16th of the
period) at these rates. At click rates ú16 Hz, the result for
FIG . 3. A: period histograms, rate modulation transfer function, vector
strength, and synchrony modulation transfer function for the response to
clicks in AAF as shown in Fig. 2. Period histogram is shown twice. Preferred
phase of the response shifts from the beginning to little over half of the
period for increasing click rates. Rate temporal modulation transfer function
(tMTF, middle panel) shows an abrupt reduction at a click rate of 11.28
Hz. Vector strength (VS) is relatively strong throughout the entire range
of click rates. Synchrony tMTF peaked at 4.76 and at 8 Hz, the best
modulation frequencies (BMF) was estimated at 8 Hz. Limiting rate, at
50% of the value at the BMF, was at 11.28 Hz. B: following the same
format, the result for AM noise in AI, for data shown in Fig. 2, is presented.
Note that AM rates now are between 2 and 64 Hz. Firing rate and VS as
well as number of synchronized spikes drops substantially at 8 Hz (limiting
rate) for a BMF of 6.72 Hz. C: again in the same format, shows results
for AM tone stimulation in AII (data same as in Fig. 2). Number of spikes
per tone burst is a slowly decreasing function of AM frequency, the VS is
high only between 6.72 Hz and 9.52 Hz. BMF is 9.52 Hz, and the limiting
rate is 11.28 Hz.
/ 9k2e$$no40 J364-8
10-27-98 16:59:01
neupa
LP-Neurophys
2750
TABLE
J. J. EGGERMONT
1.
Mean BMF and limiting rates for three stimuli and three cortical areas
AI
Stimulus
Measure
Click trains
LFP
MU
SU
ISI histogram
LFP
MU
ISI histogram
LFP
MU
ISI histogram
AM noise
AM tone
BMF, Hz
12.5
8.6
8.1
6.3
16.5
15.7
10.2
17.2
20.6
8.6
{
{
{
{
{
{
{
{
{
{
8.1
3.8
2.6
2.3
12.0
16.9
2.3
13.0
17.7
2.2
AAF
Limiting rate, Hz
20.8
10.9
11.8
11.7
43.2
40.4
25.4
48.9
40.3
29.4
{
{
{
{
{
{
{
{
{
{
10.4
4.5
4.9
1.9
18.4
18.7
10.3
17.3
16.8
10.6
BMF, Hz
9.6
6.7
7.3
6.1
10.2
11.9
8.5
14.5
8.9
8.3
{
{
{
{
{
{
{
{
{
{
AII
Limiting rate, Hz
3.6
3.0
4.5
2.6
5.9
15.3
2.4
10.0
11.0
1.6
16.7
8.6
13.7
12.0
42.3
28.9
18.5
44.2
27.8
17.1
{
{
{
{
{
{
{
{
{
{
6.1
5.4
7.2
3.6
19.3
19.3
8.4
15.2
16.0
7.0
BMF, Hz
8.1
8.5
7.4
7.0
17.5
15.3
9.4
19.0
15.7
9.3
{
{
{
{
{
{
{
{
{
{
3.9
3.7
3.2
2.6
14.1
16.2
1.9
14.3
17.0
1.2
Limiting rate, Hz
18.9
13.2
14.4
12.2
47.2
37.1
18.6
42.7
45.3
23.1
{
{
{
{
{
{
{
{
{
{
4.9
7.2
7.8
2.5
20.0
19.0
6.5
18.5
19.4
14.4
Values are means { SD. BMF, best modulation frequency; AI, primary auditory cortex; AAF, anterior auditory field; AII, secondary auditory cortex;
LFP, long field potential; MU, multiunit; SU, single unit; ISI, interspike interval.
LFP1 triggers approaches the VS values for all the LFPtrigger conditions and for MU spikes. This suggests that the
jitter in both LFP triggering and spike firing is limiting the
VS at high click rates. In AI, the VS for MU spikes was
equal to that for the LFP triggers õ4 Hz and ú16 Hz but
was higher in between. In AAF, the VS was very similar
for MU spikes and LFP triggers. For AII, the VS for MU
spikes was slightly higher than for LFP triggers at all click
rates. The highest VSs for MU recordings was obtained in
AI for click rates of 8 and 9.56 Hz.
The product of the number of spikes per click train and
the VS results in the synchronization tMTF (Fig. 9). All
the tMTFs were band-pass with the BMF at 8 or 9.52 Hz
regardless of cortical area or response measure. The maximum number of synchronized spikes or LFP triggers per
click train was highest in AI.
The ratio of the number of MU spikes and the number of
LFP triggers was largely independent of click rate (Fig.
10A), whereas the ratio for the synchronized rate dropped
somewhat ú10 Hz (Fig. 10B). This again suggests that LFP
triggers can be used to predict the unit response to periodic
click trains up to a scaling factor depending likely somewhat
on the setting of the trigger level.
Averaging across all units to obtain mean tMTFs such as
shown in Fig. 9 inevitably is dominated by units with the
highest firing rates. Thus we also normalized each singleunit tMTF on its value at the BMF and then averaged. The
so-obtained mean normalized-tMTF did not differ essentially
from the normalized mean-tMTF for AI but was slightly
different for AAF and AII (Fig. 11). The differences suggest
that in AAF and AII, units with BMFs õ8 Hz and ú16 Hz
had on average lower firing rates than units with BMFs in
between.
The VS and tMTF curves averaged across the single
units were similar to those for the MU data. The BMF of
the mean SU tMTFs was 9.52 Hz for AI and AII and 8
Hz for AAF.
Average synchronization
tMTFs based on LFP triggers evoked by AM noise and AM
tone stimuli are very similar, but they are distinctly different
from those for periodic click stimulation. Figure 12 shows
this for each area separately. The number of LFP triggers
for the click trains has been divided by two because the click
trains are twice as long as the AM bursts. For the higher
repetition and AM rates, the responses for all three stimuli
appear to asymptote to the same level, slightly below one
trigger per stimulus. The nearly linear rise in number of
triggers up to AM rates of 10 Hz suggests that adaptation
is not playing a big role for low modulation rates. Above
10 Hz, click responses appear to be affected stronger than
AM responses, likely because the click train stimuli last
longer and thus allow for more adaptation. The tMTFs to
AM stimuli that do not produce a large peak but have the
same response floor in these cases would tend to have a
shallower slope.
For MU spikes, similarly normalized, click stimulation is
again more effective, but now some interareal differences
between the responses to the two AM stimuli are emerging
(Fig. 12B). In AI, the tMTF to the AM tones shows a clear
peak at 9.5 Hz, the same frequency as for the click BMF,
whereas for AM noise, the tMTF is much shallower with a
broad maximum between 8 and 11 Hz. In AAF, both AM
stimuli show a clear peak in the 8- to 11-Hz range, similar
to that for click stimulation. In AII, the average tMTF for
AM tones does show a very weak response without any
preference for AM, whereas the AM noise stimulus is AM
rate sensitive. Across cortical areas, AM noise is the most
effective stimulus (in terms of number of synchronized
spikes at the BMF) in AAF followed by AII and AI. AM
tones evoke the same peak activity in AI and AAF, but at
higher AM rates the activity in AAF is the largest, whereas
AM tones are very ineffective in activating units in AII. The
click stimuli evoke qualitatively the same tMTFs but with
the number of spikes at the BMF largest in AI and smallest
in AII. Note that the BMF of the mean tMTFs for AM
stimuli is similar to the modal BMF and thus much smaller
than the mean BMF. This is because units with high BMFs
generally had lower firing rates.
ISI and autocorrelogram representation
MEAN TMTFS FOR AM STIMULI.
/ 9k2e$$no40 J364-8
First-order ISI histograms and autocorrelograms (all-order ISI histograms) were calculated
for all SU and separately for all MU spike trains. As an
example, we show the period histograms (triggered with
reference to the clicks and the basis for the tMTFs) and allPERIODIC CLICK TRAINS.
10-27-98 16:59:01
neupa
LP-Neurophys
TEMPORAL CODING IN THREE AUDITORY CORTICAL AREAS
2751
terspike intervals equal to one click period are for click
repetition rates of 13.44 Hz (AI) and potentially for 16 Hz
in both AAF and AII. Evidence for intervals equal to two
periods of the stimulus is found °19.04 Hz for all three
areas. Considering only first-order ISIs (Fig. 14B) also
shows that, especially for ‘‘optimal’’ click rates such as
between 8 and 11.28 Hz, ISIs equal to twice the click period
frequently occur. Such firings contribute to the tMTF at those
repetition rates but not to the ‘‘correct’’ ISI. This suggests
that the period histogram method resulting in tMTFs sets
the upper limit for a temporal representation of periodicity.
The mean values of the BMF and limiting rate calculated
for the all-order ISI histogram representation for single units
are very similar to corresponding values obtained from the
temporal modulation transfer functions (Table 1). For those
single-unit recordings where the all-order interval histogram
showed visible peaks at lag times equal to one stimulus period
(in 178 of the 338 units for which a tMTF was calculated),
scattergram plots of BMFs and limiting rates from tMTFs and
FIG . 5. Distributions of the limiting rates for MU and SU activity in
response to periodic click trains. A: limiting rates for MU data are very
similar for the 3 cortical areas. B: same for SU data.
order ISI histograms for simultaneous MU recordings in the
three areas with all CFs Ç5 kHz. In Fig. 13A, the period
histograms show visible click following °32 Hz with the
largest amplitudes in the 4- to 8-Hz range in each area.
In contrast, the all-order ISI histograms (Fig. 13B) show
preferred activity at ISIs equal to one period of the click
repetition rate only up to 9.52 Hz, possibly to 11.28 Hz with
an absence of ISIs for higher click rates. This suggests that
the firings locked to the clicks at these rates were isolated
spikes or spike pairs that were at least more than three click
periods apart. Note that the peak width at the one- and twoperiod marks are the same relative to the period duration,
suggesting an accuracy of firing that is proportional to the
time interval between clicks.
Population ISI histograms and autocorrelograms (all-order ISI histograms) were constructed by adding the 338
individual SU ISI histograms and autocorrelograms. For the
entire population, the pooled all-order ISI histogram is
shown in Fig. 14A. The highest detectable preferred in-
/ 9k2e$$no40 J364-8
FIG . 6. Distribution of BMFs and limiting rates for MU responses to
AM stimuli in 3 areas. A: most frequent BMF values found are õ10 Hz
but the range of values is much wider than for click stimulation (cf. Figs.
4 and 5), thus the mean value is considerably larger than the mode. B:
fairly uniform distribution of limiting rates for AM stimuli in all areas.
10-27-98 16:59:01
neupa
LP-Neurophys
2752
J. J. EGGERMONT
similar for the three cortical areas, clear interareal differences are found for AM noise and AM tone stimulation.
Figure 16A shows the population results for AM noise based
on individual SU all-order ISI histograms; in AI, ISIs equal
to the AM period are found for rates °38 Hz, whereas for
AAF and AII, this is at most °16 Hz. For AM tones (Fig.
16B), in AI ISIs equal to the AM period are again found
°32–38 Hz and in AAF at most °16 Hz, whereas in AII
hardly any preferred intervals are seen. This does not mean
that no single units can be found that show ISIs equal to a
stimulus period; in fact the mean and SD shown in Table 1
clearly indicates this presence. The difference is that only
FIG . 7. Comparison of mean rate modulation transfer functions in the
3 areas. Top: results based on LFP triggers. tMTF is a weakly band-pass
function of click rate and very similar for the three areas. Middle: same
but only taking the 1st LFP trigger per click into account. Three functions
are now high-pass functions of click rate. Bottom: results for MU data,
again a weakly band-pass function is obtained. AAF on average shows the
largest firing rates across all click rates.
from all-order ISI histograms (Fig. 15, A and B) show that
these measures are positively correlated in all three areas
(P õ 0.05). The regression lines are labeled with the area
name. For this subgroup of single units, the BMF for the allorder interval histogram was significantly lower ( Ç1.7 Hz;
P õ 0.0005) than the BMF from the tMTF in all areas. A
paired t-test showed no significant differences in the limitingrate measures in either of the three cortical areas. Thus the
mean values for BMF and limiting rate obtained with the
tMTF and the all-order ISI histograms are very similar.
AM STIMULI. Whereas the all-order population ISI histograms for click train stimulation (cf. Fig. 14A) are very
/ 9k2e$$no40 J364-8
FIG . 8. Comparison of the mean VS as a function of click rate in the
3 areas. Top: results for AI, the highest VS is found for the 1st LFP trigger
after a click (LFP1), where only 1 trigger per stimulus period is used, the
lowest is for the case where all LFP triggers are used and the MU data are
in between. At high click rates, all 3 measures converge to the same level.
For AAF (middle), one observes the same basic pattern but the difference
between MU and LFP triggers is now very small. For AII ( bottom), again
the same basic finding is presented.
10-27-98 16:59:01
neupa
LP-Neurophys
TEMPORAL CODING IN THREE AUDITORY CORTICAL AREAS
2753
ALL-ORDER ISIS COMPARED FOR CLICKS AND AM STIMULI.
An illustration of the dependence of all-order interval histograms on stimulus type for three selected SU recordings is
shown in Fig. 17, A–C. For primary auditory cortex (Fig.
17A), the results for click trains (click rates from 1 to 32
Hz) show that rates between 1 and 11.3 Hz produce preferred
ISIs equal to the click interval, whereas for AM noise and
AM tones (AM rates from 2 to 64 Hz) this range is shifted
upward from Ç5–8 to 32 Hz. Clearly low AM rates for
noise and tones are not effective in generating a temporal
representation of the AM period. In contrast to high click
rates, high AM rates are much more effective likely because
AM stimuli do not induce the strong postactivation suppression that is commonly found for clicks (cf. Fig. 2). For
AAF (Fig. 17B), the picture is somewhat different. For
clicks, preferred ISIs equal to the interclick interval are again
found °11.3 Hz, but for AM noise and AM tones, the upper
boundary is only slightly higher at Ç16–19 Hz. Again a
strong band-pass dependency is found for AM noise, but
less so for AM tones. In AII (Fig. 17C), one of the limited
number of examples for which AM tones produced some
clear ISIs in single-unit activity at the AM period (mostly
between 8 and 16 Hz), the result was not overly clear for
the other stimuli: for AM noise and for click trains only
rates Ç8 Hz were effective.
To construct an analogue to the tMTF based on ISIs, the
maximum in the three bins surrounding and including the
AM period for the all-order ISI population histograms was
FIG . 9. Comparison of the mean synchronization modulation transfer
functions in the 3 areas. Top: results for LFP triggers. Number of synchronized triggers at low and high click rates is very similar in the 3 areas, but
the peak number of synchronized triggers is largest in AI and about the
same in AAF and AII. Taking only 1st triggers for the LFP shows basically
the same result with AII slightly better than AAF. For MU spikes (bottom),
again the results are qualitatively the same in the three areas but with AI
producing more synchronized spikes than AAF and AII. Note that the
average number of spikes per stimulus was highest in AAF (cf. Fig. 7).
10% of the units in AII showed detectable periodic ISIs for
AM tones, whereas nearly all did for AM noise. The mean
BMFs and limiting rates calculated from the individual allorder ISI histograms (Table 1) are strikingly lower than for
the values obtained from the tMTFs (except the BMFs for
AAF), but somewhat similar to those obtained from the
population ISI histogram. As can be seen, the SDs for the
limiting rates for AM stimuli are much larger (and also
relative to the mean) than for click stimulation.
/ 9k2e$$no40 J364-8
FIG . 10. Comparison of mean number of MU spikes and mean number
of LFP triggers. Similarity of the tMTFs shown in the previous 3 figures
is further illustrated in the nearly constant ratio of number of MU spikes
and number of LFP triggers in the 3 areas. This suggests a simple threshold
model of spike generation.
10-27-98 16:59:01
neupa
LP-Neurophys
2754
J. J. EGGERMONT
number of ISIs is highest at 9.52 Hz and its subharmonic of
4.76 Hz (for AI and AII), and slightly lower at 8 and 4 Hz
for AAF. This dip might be related to an interaction between
depression periods in ongoing spindle activity and the activation introduced by the clicks with rates between 4 and 9 Hz
(Eggermont and Smith 1995a).
For AM noise stimulation, the elicited maximum number
of periodic ISIs is largest in AAF than in AI and smallest
in AII. There is a rapid drop in the number of ISIs ú11.3
Hz for AAF and AII, whereas in AI the number only slowly
decreases for higher AM rates. Again a dip in the number
of ISIs is noted, this time at 4.76 Hz; however, there is no
indication of a harmonic structure in the dependency on
modulation rate. AM tone stimulation only produces significant numbers of periodic ISIs in AAF with very few intervals
ú13 Hz, whereas AI and AII both stay close to the background level. Thus with the exception of AM noise in AI,
the highest AM rate for a significant number of ISIs is Ç13
Hz. Comparison of, for instance, Figs. 11 and 12 (see also
Table 1) shows that the BMFs and the limiting rates are
generally higher for the tMTF-based characterization. This
suggests that individual stimulus-locked spikes are evoked
for rates ú13 Hz and are distributed across stimulus duration
but that there are no doublets with an ISI equal to the AM
period generated for these higher AM rates.
TEMPORAL MODULATION TRANSFER FUNCTIONS AND LATENCY. For single units, the first-click latencies obtained at
peak equivalent SPL levels of 75 dB were correlated positively with the minimum latency for tone pips (significant
in all 3 areas, P õ 0.001) and were on average 1.3 ms longer
than the minimum latency for pips (which typically was
found also at 75 dB SPL, the highest intensity used). Firstclick latencies were shortest (P õ 0.0001) in AAF and
about equal in AI and in AII (Table 3). Minimum tone pip
latencies were also shortest (P õ 0.0005) in AAF and again
about equal in AI and AII. The BMF for single units in
response to periodic click trains was correlated positively
with the first-click latency (Fig. 19A). The regression lines
are labeled with the area name. This correlation was significant for AAF but not for AI and AII. The limiting rate was
correlated positively with first-click latency for AI and AAF
but not for AII (Fig. 19B).
FIG . 11. Comparison of a normalized mean tMTF for MU data with the
mean of normalized MU tMTFs. For the normalized mean tMTFs, the same
graphs shown in Fig. 9 were normalized to a peak level of 1. In the 2nd
case, all individual MU tMTFs first were normalized, then averaged, and
then the peak value again was adjusted to 1. For AI, the 2 curves are
identical; for AAF and AII, some differences are noted both at low and
high click rates.
computed for all stimuli and areas. A correction for the background in the autocorrelogram was carried out by subtracting
the mean of the four bins midway between period 1 and period
2. The stimulus driven number of periodic intervals in the allorder ISIs then was plotted (Fig. 18) as a function of click or
AM rate. For click stimulation, the dependence on the repetition rate is qualitatively the same for all three areas with best
response in AI. There is a sharp decrease ú11.3 Hz and hardly
any periodic intervals for repetition rates ú13.4 Hz. In all
three areas, there is a dip in the number of ISIs around a
repetition rate of 6.5 Hz. Alternatively one could say that the
/ 9k2e$$no40 J364-8
CORRELATION
OF
SPECTRAL
AND
TEMPORAL
PROPERTIES.
For all recording sites, the LFP trigger and multiunit intensity-frequency response areas were measured. Multipeaked
response areas for MU recordings were found in 8% of
recording sites in AI, for 26% of recording sites in AAF,
and for 9% of recording sites in AII. The CF and bandwidth
(BW) at 20 dB above threshold at CF was determined for
both LFP and MU recordings. At the same recording sites,
the CF values were, with some exceptions, very similar for
LFP and MU (Fig. 20A). In 10 cases the differences in CF
were ú1 octave. However, the 20 dB BW was significantly
larger in AI and AAF for frequency tuning of LFP triggers
than for MU spikes recorded on the same electrode (Table
2, Fig. 20B). This amounted to 0.4 octaves (P õ 0.01) for
AI and to 0.8 octaves (P õ 0.0001) in AAF. In AII the
bandwidths were not significantly different for LFP-triggerand MU-spike-based tuning curves. Between cortical areas,
the MU tuning curve bandwidth in AAF and AII was not
10-27-98 16:59:01
neupa
LP-Neurophys
TEMPORAL CODING IN THREE AUDITORY CORTICAL AREAS
2755
FIG . 12. Comparison of LFP trigger and MU-spike based tMTFs for 3 stimulus types in 3 cortical areas. A: in all 3 areas,
click trains evoke ¢3 times more LFP triggers per stimulus (or half the stimulus in case of the click trains) for modulation
rates õ30 Hz. No noticeable difference was found between the 2 AM stimuli. For MU spikes, the difference in response
size was smaller (B). Also in this case the click stimuli produce considerably larger responses than the AM stimuli, especially
in AI and AAF. In AII, the mean number of synchronized spikes per AM tone was õ0.5 across the AM frequency range.
In AI, the response to AM noise was also very low.
significantly different (P Å 0.8). The BW in AAF was
slightly higher by 0.3 octaves than in AI (P Å 0.3). In
contrast, the BW in AII was Ç1 octave higher than in AI
(P õ 0.001). For the LFP measure, the tuning curve BW
for AAF and AII were the same. The BWs in AAF and AII
were 0.7 octaves higher than in AI (P õ 0.0005).
The BMFs and limiting rates for LFP recordings were independent of the CF and BW in all three areas. The BMF for
MU recordings was independent of CF in AI and AII but was
correlated positively with CF for AAF (P õ 0.05; Fig. 21A).
/ 9k2e$$no40 J364-8
This was largely the result of several high BMF values for
recording sites with CFs ú10 kHz. The limiting rates for MU
recordings were independent of CF in all three areas (Fig.
21B). The limiting rates for MU recordings, however, were
correlated positively with tuning curve BW in AI and AII (P õ
0.05) and independent of BW in AAF (Fig. 21C).
DISCUSSION
In answer to the research questions posited in the
we found the following:
DUCTION,
10-27-98 16:59:01
neupa
LP-Neurophys
INTRO-
2756
TABLE
J. J. EGGERMONT
2.
CF range and tuning curve bandwidths in three areas
AI
AAF
AII
Measure
CF range, kHz
BW, octaves
CF range, kHz
BW, octaves
CF range, kHz
BW, octaves
LFP
MU
1.0–17.0
1.0–15.5
1.9 { 0.8
1.5 { 0.8
1.2–20.0
1.2–15.5
2.6 { 1.2
1.8 { 1.0
3.8–25.8
2.6–17.6
2.6 { 1.4
2.5 { 1.3
Values are means { SD. CF, characteristic frequency; BW, bandwidth.
1 The mean BMFs for the temporal modulation transfer
functions in AI and AII were about a factor 2 higher for AM
stimuli than for periodic click trains, and limiting rates were a
factor 4 higher. In AAF, the mean BMFs for the AM stimuli
were Ç1.5 times larger than for periodic click trains and the
limiting rates were about a factor 3 larger. For all three areas,
the mean BMFs based on the all-order ISIs for AM stimuli
were Ç1.5 times those for clicks and limiting rates were less
than a factor 2 higher. Thus AM stimuli in general produce
much higher BMFs and limiting rates than periodic click trains.
FIG . 13. Examples of period histograms and all-order interspike-interval
(ISI) histograms for simultaneous MU recordings in AI, AAF, and AII.
CFs of the 3 recording sites were Ç5 kHz. A: period histograms with clear
synchronized responses °32 Hz in all areas. B: ISIs equal to the click
repetition period cease to exist for click rates ú9.52 Hz.
/ 9k2e$$no40 J364-8
2 Temporal modulation transfer functions for periodic
click stimulation were very similar in the three cortical areas,
hence the BMFs and limiting rates were not significantly
different. For AM noise, the BMFs and limiting rates were
also similar across areas, whereas for AM tones the BMF
and limiting rates were about a factor 2 lower in AAF compared with the other areas. However, the firing rates for AM
tones were by far the highest in AAF.
FIG . 14. Population ISI histogram for click train stimulation obtained
by adding SU ISI histograms. A : all-order ISI histograms; intervals with
periodicity equal to the click period are found in all 3 areas °14.33 Hz.
Evidence for intervals equal to 2 periods of the stimulus is found °19.04
Hz for all 3 areas. B : 1st-order ISIs only. It is clear that even at ‘‘optimum’’ click rates, a substantial portion of ISIs is equal to twice the click
period.
10-27-98 16:59:01
neupa
LP-Neurophys
TEMPORAL CODING IN THREE AUDITORY CORTICAL AREAS
2757
dent of CF in AI and AII and positively correlated with CF
in AAF. The limiting rate for MU data were independent of
CF in all three areas. The limiting rate, however, was correlated positively with the frequency-tuning curve BW in AI
and AII but not in AAF. So in general there is only a modest
interdependence of spectral- and temporal-response properties in AI and AII. The BMF for single units in response to
periodic click trains was correlated positively with the firstclick latency in AAF. The limiting rate was correlated positively with first-click latency for AI and AAF but not for
AII. In AAF, a strong correlation between temporal response
properties and latency was found.
Synchronized firing combines firing rate and stimulus
synchrony
In the present study, as in previous ones from our lab
(Eggermont 1991; Eggermont and Smith 1995a), temporal
FIG . 15. Comparison of BMFs and limiting rates estimated from tMTFs
and autocorrelograms (ACH). Both measures are correlated positively, but
the BMF for the all-order interval histogram was Ç1.7 Hz lower than the
BMF from the tMTF across all areas. No significant differences in the
limiting rate measures in any of the 3 cortical areas were found.
3 The representation of stimulus periodicity in ISIs
showed significantly lower estimates of mean BMFs and
limiting rates in all three areas compared with those estimated from the temporal modulation transfer functions. The
difference was relatively small for periodic click trains in
all three areas and largest for AM stimuli in AI and AII.
4 Modulation frequencies õ20 Hz were represented
strongly in the ISIs. This is the repetition-rate and AMfrequency range where the sensation of rhythm prevails over
that of roughness. Thus rhythm may be coded in auditory
cortex in temporal fashion in the all-order ISIs of single
units across the entire CF range.
5 Frequency tuning for MU spikes is sharpest in AI followed by AAF and broadest in AII. Single-unit latencies to
high-intensity clicks are shortest in AAF and Ç4 ms longer
in both AI and AII. The BMFs for MU data were indepen-
/ 9k2e$$no40 J364-8
FIG . 16. Population ISI histogram for AM noise and AM tone stimulation obtained by adding SU ISI histograms. A: for AM noise in AI, detectable ISIs equal to the AM period can be found °38 Hz, whereas in AAF
and AII this upper limit is at most 16 Hz. B: for AM tone stimulation in
AI, detectable ISIs equal to the AM period again can be found °38 Hz,
whereas in AAF the upper limit is at most 16 Hz, and in AII there is no
clear representation in this population response.
10-27-98 16:59:01
neupa
LP-Neurophys
2758
J. J. EGGERMONT
modulation transfer functions based on firing rate, VS, and
synchronized firing were calculated. Rate tMTFs were generally nondistinctive and mostly low-pass functions of click
repetition rate or AM frequency. As a consequence, the synchronized firing tMTF, from which all our BMF and limiting-rate measures were obtained, to some extend represents
the dependence of the VS on AM frequency. The synchronized rate is preferred above the VS because low-firing-rate
neurons, which fire at most a single spike to a click or AM
peak, tend to have higher VSs than units that fire spike
clusters; yet for the nervous system, the latter may have
more impact in being able to fire other neurons. Furthermore
the VS is an idealization that requires the evaluation of relative timing and correction for overall firing rate. These stimulus aspects are available to the experimenter but not to the
animal, so this calculation is not likely to happen in the
nervous system. Consequently, synchronized firing rate is a
more appropriate measure to use than VS.
LFPs versus spike activity
This work demonstrates that LFP triggers and single- or
multiunit spikes result in similar estimates of the CF of
a recording site, thereby corroborating and extending the
conclusions of an earlier study that was confined to AI (Eggermont 1996). The 20-dB bandwidth of the frequencytuning curves for LFP triggers was substantially higher than
those for MU spikes in AI and AAF, whereas there was no
difference in AII. The difference in AI and AAF can be
explained on the assumption that the postsynaptic potential
contributions to the LFP can be recorded from a larger volume surrounding the electrode tip than action potentials.
This would require a larger dipole field for synaptic potentials than for action potentials, and this is not unreasonable
to assume (Mitzdorf 1985). The similarities in AII for LFP
tuning and MU tuning in AII then must be explained by an
extensive convergence of units with a broader range of CFs
to a single AII unit or to its projection site, the dorsal part
of the MGB. However, if this larger summing volume idea
is correct and the cell density and dipole orientation is the
same in AI and AAF, one would expect the BW for LFP
tuning curves to be more similar in AI and AAF because
the MU tuning curve bandwidths are also close in value.
This is not what is observed; hence to explain the difference
in LFP and MU bandwidths, other factors such as an effect of
inhibitory activity on spike initiation should be considered.
Because the depth-negative LFP component from which the
triggers are obtained represents dominantly excitatory postsynaptic events (Mitzdorf 1985) and single-unit firing also
is determined by inhibitory mechanisms, the sharper tuning
observed for single- and multiunit activity might be the result
of such inhibitory activity. Under this assumption, the simiFIG . 17. MU all-order ISI histograms for click and AM stimuli in 3 areas.
A: results for AI for the same set of units. Note the different repetition rate
range for clicks compared with the AM stimuli. Upper limits for periodlocked intervals for clicks are 13.44 Hz and for the AM stimuli 32 Hz. B:
results for AAF. Here the upper limit for click stimulation is 16 Hz and for AM
stimuli around 32 Hz. C: results for AII. One notices the poor representation of
stimulus periods in the histograms. For click stimulation, ISIs equal to 1
period stand out only for rates at 8 and 9.52 Hz; for AM noise the range is
limited from 6 to 11 Hz and for AM tone from 8 to 13 Hz.
/ 9k2e$$no40 J364-8
10-27-98 16:59:01
neupa
LP-Neurophys
TEMPORAL CODING IN THREE AUDITORY CORTICAL AREAS
TABLE
3.
2759
Click and tonepip latencies
75 dB SPL clicks
Minimum pip latency
AI, ms
AAF, ms
AII, ms
17.5 { 4.1
15.8 { 4.1
13.5 { 3.3
12.4 { 1.7
17.6 { 3.8
16.4 { 5.1
Values are means { SD.
BMF for clicks trains and AM stimuli, thereby extending
Eggermont and Smith’s (1995a) findings, which were limited to click trains in AI. Because for click stimulation,
BMFs were in the 8- to 14-Hz range for LFP and MU and
strongly correlated with the autonomous cortical rhythms
(Kenmochi and Eggermont 1997), this was expected. Cortical oscillations in the 7- to 14-Hz range have been suggested
previously to play a role in the magnitude and timing of
cortical evoked activity and behavioral responses. For in-
FIG . 18. Temporal modulation transfer functions derived from the population all-order ISI histograms. These functions were obtained by averaging
the 3 bins around and including the 1 period value and subtracting the mean
of 4 bins halfway in between period 1 and period 2. For click stimulation,
the response in all 3 cortical areas is very similar with hardly any intervals
ú13 Hz. For AM noise stimulation, intervals are detectable up to Ç30 Hz,
and all areas respond with AAF providing about twice the number of
ISIs than AII. For AM tone stimuli, AAF stands out, whereas only weak
representation of the modulation period is found in AI and AII. Upper limit
appears to be Ç13 Hz.
larities and differences in bandwidth for LFP and MU tuning
curves the three areas suggest that this lateral inhibitory
activity is most effective in AI, then in AAF, and least effective in AII.
LFP and MU data also resulted in similar estimates of the
/ 9k2e$$no40 J364-8
FIG . 19. Dependence of SU BMF and limiting rate on the latency of
the response to the 1st click and on the temporal filter delay. A and B:
BMFs (in AAF) and limiting rates (AI and AAF) are significantly positive
correlated with 1st-click latency.
10-27-98 16:59:01
neupa
LP-Neurophys
2760
J. J. EGGERMONT
higher for the LFP than for the MU estimate, whereas for
click train stimulation, there were no differences. The VS
of the synchronized response was significantly lower for AM
stimuli than for click stimulation, especially the ‘‘resonance’’ around 9 Hz was much reduced. This has the immediate consequence that the limiting rate, a relative measure
equal to the repetition rate at 50% of the strength of the
response at the BMF, will be higher for AM stimuli. The
use of such a relative measure therefore is more useful to
capture the shape of the tMTF than to estimate the upper
bound of phase locking to AM frequency. For the ISI measure, the limiting rate was estimated from the visibility of
interval peaks, so that measure poses a strict limit to interval
FIG . 20. Correlation of frequency-tuning curve properties obtained for
MU spikes and LFP triggers. Whereas the CF of the recording site extracted
from MU and LFP data are generally the same, the tuning curve bandwidth
at 20 dB above threshold is nearly always substantially larger for LFP
triggers.
stance, Ahissar et al. (1997) found that in the somatosensory
cortex of urethan-anesthetized rats and guinea pigs, Ç25%
of units in layer IV/V exhibited spontaneous oscillations in
the neighborhood of preferred whisking frequencies ( É10
Hz). They suggested that such independent neural oscillators
with intrinsic frequencies that match the rhythm of sensory
events can participate in decoding temporal sensory information through phase-locked thalamocortical loops. Combined
with our present and previous findings, this suggests that
synaptic and action potential activity in the cortex depends
as much on ongoing LFP rhythms as it does on stimulusinduced rhythmicity. Under certain conditions, the one can
entrain the other. We have shown previously that synchrony
in the firing of distant neurons within AI could be explained
largely by such global cortical activity (Eggermont and
Smith 1995b).
For AM stimuli, the limiting rates of the tMTFs were
/ 9k2e$$no40 J364-8
FIG . 21. Correlation of temporal response properties and frequency-tuning properties. A: there was no correlation between BMF and CF in AI and
AII and a significant positive correlation for AAF. B: limiting rates were
independent of CF. C: positive correlation in AI and AII between limiting
rate and frequency-tuning curve bandwidth.
10-27-98 16:59:01
neupa
LP-Neurophys
TEMPORAL CODING IN THREE AUDITORY CORTICAL AREAS
coding. The finding that these limiting rates were generally
smaller than for the corresponding measure based on tMTFs
suggests that either the tMTF measure overestimates the
capacity for repetition rate coding or that the tMTF for high
AM rates is based only on an average of preferred firing
times of isolated spikes or on a combination of these two
factors.
Similarities and differences between areas and
comparison with other studies
SPECTRAL RESPONSE PROPERTIES. The observation that the
frequency-tuning curve bandwidth for LFP triggers was
much higher in AAF than in AI confirms the earlier observation by Knight (1977). For MU data, the bandwidth was
lowest in AI and largest in AII with that for AAF midway
between the others. The finding that the MU frequencytuning curve bandwidth in AII is much higher than in AI
also agrees with previous findings (Schreiner and Cynader
1984). The MU bandwidth in AAF was somewhat higher
than that in AI but not as high as reported for the ferret
(Kowalski et al. 1995), which showed a 20-dB bandwidth
in AAF that was comparable with what I found in AII.
Knight (1977) had found that the frequency response characteristics of single units in AAF and AI in the cat were strikingly similar. The findings of broader single-unit tuning
curves in AAF may be related to the percentage of
multipeaked tuning curves found in this area. Tian and
Rauschecker (1994) found a substantial number of them,
and I did encounter such tuning curves in 26% of the recordings in AAF. In contrast, Knight (1977) found them
only in a few penetrations.
TEMPORAL RESPONSE PROPERTIES. For MU responses to periodic click train stimulation, the BMFs and limiting rates
of the temporal modulation transfer functions were very similar in AI, AAF, and AII. The BMFs were also similar for
AM noise and AM tones in all areas but, especially in AI
and AII, significantly higher than the BMFs for click trains.
The BMFs in AAF were not very different for AM stimuli
and click trains, largely because they were relatively low for
the AM stimuli. In general, click trains were much more
effective in generating responses with a high VS, and thus
a high number of synchronized spikes, than were AM stimuli. In this respect, AM noise was most effective in AAF,
then in AII, and least in AI. AM tones were also most effective in AAF, then in AI, and least in AII. This effectiveness
may be related to the fact that solely in AAF the BMFs were
relatively low, i.e., in the frequency range (8–14 Hz) of the
autonomous cortical spindle rhythm. This will enhance the
VS at that frequency and also enhance the resonance effect
because spikes produced with the (n / 1)th click or AM
peak will line up with those in the rebound to the nth click
or AM peak (Eggermont and Smith 1995a). Units with
BMFs in the 8- to 10-Hz range invariably had higher firing
rates than those with BMF outside this area. The drawback
of this strong resonance for click stimulation is that for repetition rates above the resonance frequency, the strong postactivation suppression will limit click following severely. The
weaker synchronization and suppression for AM stimuli
allows higher limiting rates. I did not observe a lack of
responsiveness to clicks for units with CFs Ç10 kHz (the
/ 9k2e$$no40 J364-8
2761
‘‘null’’ in the click spectrum), likely because the stimuli
were presented at a sufficient high level to allow strong
activation of the units.
For click trains, the dominance of BMFs Ç8 Hz was
similar to our earlier findings (Eggermont 1991; Eggermont
and Smith 1995a) and those of Schreiner and Raggio
(1996). The BMFs for AM tones in AI were also generally
similar to those reported by Schreiner and Urbas (1988),
but in the present study, the BMFs were distinctly higher in
AII and were lower in AAF than reported by Schreiner and
Urbas. The difference in the BMF values for AII may be
related to our sampling bias, which preferred short latency
neurons, and these will have higher BMFs than neurons that
respond with very long latencies. For instance, if the latencies are Ç100 ms, a frequent finding in AII (Schreiner and
Cynader 1988) and likely the result of a rebound from inhibition, then one does not expect the units to follow modulation
rates ú10 Hz. Because long-latency responses are not very
punctuate in onset and tend to be tonic, the VS will be very
low. Thus in general the BMF will be much lower than 10
Hz for long-latency neurons. In the few instances that we
did record from long-latency neurons, the BMF was õ4 Hz.
Our finding of a much lower mean BMF in AAF was limited
to the response to AM tones because for the same units an
AM noise stimulus resulted in a much higher BMF. The
BMF to AM tones was in the 8- to 14-Hz spindle frequency
range. If in Schreiner and Urbas’s (1988) recordings the
amount of spindling was reduced or the spindling frequency
was higher because of a very light anesthesia (the cats also
were paralyzed), it is likely that higher BMFs would be
found.
The BMF for single units in response to periodic click trains was correlated positively with the latency
of the response to first clicks in AAF but not in AI and AII.
Thus whenever a correlation was present, higher BMFs were
found in units with longer latencies. The limiting rate was
correlated positively with first-click latency for AI and AAF
but not for AII. Latencies are determined by conduction
delays and as such are sensitive to length and axon diameters
differences in major afferent projections. Latencies are also
sensitive to effects of inhibition that can delay the firstspike response to a stimulus substantially, and for high AM
frequencies or click rates, latencies also are affected by the
amount of synaptic depression. The minimum latency in AI
was on average 15.8 ms for tone pips, this is similar to the
Kowalski et al. (1995) finding of 16.8 ms and somewhat
shorter than my previous results of 17.7 ms (Eggermont
1996) but significantly longer than the value of 12.5 ms
found by Raggio and Schreiner (1994) and Mendelson et
al. (1997). One potential reason for this difference could be
the inclusion of more units from the dorsal part of AI, although the small percentage of multipeaked tuning curves
in this study argues against that (Heil et al. 1992). Another
potential reason may be found in the use of free-field stimulation in the present study with a speaker in the midline,
this adds 1–2 ms to the latency for free-field contralateral
stimulation, which is more comparable to closed-field contralateral stimulation (Eggermont 1998b). Minimum latencies in AAF were Ç4 ms shorter than in AI in the present
study, and this is in contrast to other studies in cat (Knight
LATENCIES.
10-27-98 16:59:01
neupa
LP-Neurophys
2762
J. J. EGGERMONT
1977; Phillips and Irvine 1982), which did not show any
significant difference. The finding is, however, in agreement
with observations in the ferret by Kowalski et al. (1995).
The difference in latency has a potential substrate in the
difference in dominant projections to AAF (from PO) and
AI (from MGBv).
Again, in partial contrast to previous findings (Schreiner
and Cynader 1984), we did not find the latencies in AII to
differ substantially from those in AI. My results likely are
biased by the fact that I searched for recording sites in AII
that produced short-latency responses. I usually found those
in tracks very close to tracks that produced only fuzzy responses; in general it was sufficient to raise the electrode
and insert it at nearly the same spot (within 100–200 mm).
Despite these short latencies, the frequency-tuning curves in
AII were substantially broader than in AI.
Schreiner and Raggio (1996) had observed for click train
stimulation in AI that the shorter the response latency of the
neuron the higher the BMF or limiting rate. This dependence
had been seen previously in the central nucleus of the inferior
colliculus (Langner et al. 1987). Because latencies are
shorter in the central parts of AI than in the dorsal and
ventral portions of AI, this suggested a topographic representation of, or at least regional differences in, repetition rate
coding (Schreiner et al. 1997). We found a positive correlation in AI, albeit that it was not significant for BMFs, and
thus our data are again different from those of Schreiner
and Raggio (1996), who showed a negative correlation. A
potential explanation may be found in sampling from a different part of AI; because our data span a larger latency range
(10–30 ms) than Schreiner and Raggio’s (10–20 ms), it
is likely that our sample included more units from dorsal
areas of AI. This area has units with longer latencies, broader
tuning curves, and better responses to broadband stimuli
such as clicks (Mendelson et al. 1997; Schreiner and Mendelson 1990; Schreiner and Sutter 1992). However, as mentioned before, we only found a modest percentage of
multipeaked response areas, and this argues against a large
proportion of units from the dorsal AI (Heil et al. 1992).
Another difference is that in Schreiner and Raggio’s study,
the contralateral ear was deafened and the acoustic stimuli
were delivered ipsilaterally. Compared with the sound presentation from a speaker in the midline, stimulating both
ears equally, as in the present study, also can account for
the latency differences.
RELATION BETWEEN TEMPORAL AND SPECTRAL RESPONSE
PROPERTIES. The positive correlation between BMF and CF
found in AAF, but not in AI and AII, corroborates similar
findings by Schreiner and Urbas (1988). A positive correlation between limiting rate and frequency-tuning curve bandwidth was found in AI and AII but not in AAF. This conforms to the intuitive notion that broader frequency tuning
should allow better temporal responses.
Temporal coding of rhythm in cortex
The amplitude-modulated stimuli and periodic click trains
used in this study produce a different hearing sensation below and above Ç16 Hz. Although they sound clearly rhythmic õ16 Hz, a sensation of roughness and pitch starts to
dominate for higher modulation frequencies. Psychophysical
/ 9k2e$$no40 J364-8
studies (reviewed in Zwicker and Fastl 1990) have shown
that the strongest sensation of rhythm is produced between
2 and 8 Hz with an optimum at 4 Hz, whereas the strongest
sensation of roughness is obtained for Ç70 Hz. The
roughness sensation disappears altogether above a modulation frequency of 300 Hz. Above this frequency AM produces a sensation of pitch only.
The boundary frequency between rhythm and roughness
may be related to a difference in neural coding above and
below that frequency. As demonstrated here, ISIs code periodic stimuli for click rates °14 Hz (from the all-order interspike interval population histogram) and for AM noise
rates °38 Hz (in AI). AM tones appear to be producing
about three times as many ISIs in AAF than in the other
areas but with a limiting rate of at most 16 Hz. Thus a fairly
strong temporal representation, residing in the ISIs is found
for low-modulation frequencies. This is the repetition rate
and AM frequency range where the sensation of rhythm
prevails over that of roughness. Thus rhythm may be coded
in auditory cortex in temporal fashion in the all-order ISIs
of single units, whereas roughness and likely also pitch
(Schulze and Langner 1997) may be represented as a rateplace code.
It is known that steadystate vowel discrimination is preserved in case of lesions in
human auditory cortex; however, stop-consonant discrimination, depending largely on precise voice-onset-time (VOT)
representation, is impaired (for review: Phillips 1993). Thus
auditory cortex is involved and necessary to integrate timing
information with spectral information as needed in word
perception. The crucial question is whether different cortical
areas have different roles in complex sound processing? I
have shown previously that VOT is coded similarly in single
units in AI (Eggermont 1995a) and in AAF and AII (Eggermont 1998a). Several other neural correlates of precise timing, such as required in gap-detection (Eggermont 1995b,
1998a) and the representation of rhythmicity (this study)
also seem to have a similar representation in AI, AAF,
and AII.
Some of the auditory primitives considered in psychophysical and model studies to bind components of the same
sound object perceptually include common onset and offset
of sound, common rates of AM and FM, harmonicity, and
common spatial origin (Bregman 1990; Cooke 1993). It is
useful for the subsequent discussion to group these auditory
primitives in contour and texture components of sound. Contour components are those temporal aspects of sound that
covary across frequency and are likely exclusively coded in
the temporal domain. Onsets and common rates of slow
( õ20–30 Hz) amplitude and FM, i.e., the region where
rhythm is dominant, are clearly contours that delineate, for
instance, sound duration and separation between noise bursts
and format transitions. Common higher rates of AM, in the
roughness and pitch range, as well as harmonicity are aspects
of sound texture. Texture aspects can be further characterized by pitch, timber, and roughness and slow changes
therein. These texture aspects of sound thus relate to, constant or slowly changing, spectral representations in a cortical rate-place code. One expects texture and contour components thus to have largely independent neural representations
COMPLEX SOUND REPRESENTATION.
10-27-98 16:59:01
neupa
LP-Neurophys
TEMPORAL CODING IN THREE AUDITORY CORTICAL AREAS
in auditory cortex or potentially also in subcortical areas.
Binding various sound features that are rate-place coded in
different cortical areas is likely done in the time domain and
for the auditory system based on the contour aspects of
sound.
We found that the temporal representation of rhythm as
well as sound onsets was strong in both primary cortical
areas, AI and AAF, whereas it was distinctly less punctuate
or absent (for AM tones) in AII. Thus the primary areas
may be more suited to represent all the contour aspects of
sound. Specifically, because of its strong response to clicks
and noise onsets, AI may be specialized in sound transients
such as onsets and offsets. In AII, low-frequency AM of
broadband signals, clicks, and AM noise are still effectively
represented in temporal code, but slow changes in the AM
of pure tones are not represented.
Thus it seems that the neural activity in AI, AAF, and
AII together, but in different proportions and in distinctly
different coding strategies, represents contour and texture.
If the auditory cortex would be segregated in two streams
like the visual system (Rauschecker 1997), then the three
areas studied here are likely all part of the ‘‘what’’ stream.
This also suggests a more extreme parallelism for the auditory cortex compared with the visual one. The ‘‘where’’
stream dealing with sound localization and movement should
then, if it exists in the cat, largely reside outside these areas,
for instance, in the anterior ectosylvian sulcus (Middlebrooks et al. 1994). The fact that these latter aspects of
sound are computed from input of both ears distinguishes
them from the, at least partially, monocular character of the
‘‘where’’ pathway in the visual system.
G. Smith provided valuable suggestions throughout the experiments.
K. Ochi and M. Kenmochi assisted with the data collection.
This investigation was supported by grants from the Alberta Heritage
Foundation for Medical Research and the Natural Sciences and Engineering
Research Council of Canada.
Address for reprint requests: Dept. of Psychology, The University of
Calgary, 2500 University Dr. N.W., Calgary, Alberta T2N 1N4, Canada.
Received 13 May 1998; accepted in final form 31 July 1998.
REFERENCES
AHISSAR, E., HAIDALIU, S., AND ZACKSENHOUSE, M. Decoding temporally
encoded sensory input by cortical oscillations and thalamic phase comparators. Proc. Natl. Acad. Sci. USA 94: 11633–11638, 1997.
ANDERSEN, R. A., KNIGHT, P. L., AND MERZENICH, M. M. The thalamocortical and corticocortical connections of AI, AII, and the anterior auditory
field (AAF) in the cat: evidence for two largely segregated systems of
connections. J. Comp. Neurol. 194: 663–701, 1980.
BIESER, A. AND MÜLLER-PREUSS, P. Auditory responsive cortex in the
squirrel monkey: neural responses to amplitude-modulated sounds. Exp.
Brain Res. 108: 273–284, 1996.
BREGMAN, A. S. Auditory Scene Analysis. Cambridge, MA: Mit Press, 1990.
CARIANI, P. AND DELGUTTE, B. Neural correlates of the pitch of complex
tones. I. Pitch and pitch salience. J. Neurophysiol. 76: 1698–1716, 1996a.
CARIANI, P. AND DELGUTTE, B. Neural correlates of the pitch of complex
tones. II. Pitch shift, pitch ambiguity, phase invariance, pitch circularity,
rate pitch, and the dominance region for pitch. J. Neurophysiol. 76:
1717–1734, 1996b.
COOKE, M. Modelling Auditory Processing and Organization. Cambridge,
UK: Cambridge Univ. Press, 1993.
EGGERMONT, J. J. Rate and synchronization measures of periodicity coding
in cat primary auditory cortex. Hear. Res. 56: 153–167, 1991.
EGGERMONT, J. J. Stimulus induced and spontaneous rhythmic firing of
single units in cat primary auditory cortex. Hear. Res. 61: 1–11, 1992.
/ 9k2e$$no40 J364-8
2763
EGGERMONT, J. J. Differential effects of age on click-rate and amplitude
modulation-frequency coding in primary auditory cortex of the cat. Hear
Res. 65: 175–192, 1993.
EGGERMONT, J. J. Temporal modulation transfer functions for AM and FM
stimuli in cat auditory cortex. Effects of carrier type, modulating waveform and intensity. Hear. Res. 74: 51–66, 1994.
EGGERMONT, J. J. Representation of a voice onset time continuum in primary
auditory cortex of the cat. J. Acoust. Soc. Am. 98: 911–920, 1995a.
EGGERMONT, J. J. Neural correlates of gap detection and auditory fusion in
cat auditory cortex. Neuroreport 6: 1645–1648, 1995b.
EGGERMONT, J. J. How homogeneous is cat primary auditory cortex? Evidence from simultaneous single-unit recordings. Auditory Neurosci. 2:
76–96, 1996.
EGGERMONT, J. J. Representation of amplitude modulated sounds in two
fields in auditory cortex of the cat. In: Acoustical Signal Processing in
the Central Auditory System, edited by J. Syka. New York: Plenum Press,
1997, p. 303–319.
EGGERMONT, J. J. Neural synchrony within and between three cortical fields
in cat for complex sounds. In: Psychophysical and Physiological Advances in Hearing, edited by A. R. Palmer, A. Rees, A. Q. Summerfield,
and R. Meddis. London: Whurr Publishers, 1998a, p. 497–503.
EGGERMONT, J. J. Azimuth coding in primary auditory cortex of the cat. II.
Relative latency and interspike interval represenation. J. Neurophysiol.
80: 2151–2161, 1998.
EGGERMONT, J. J. AND SMITH, G. M. Synchrony between single-unit activity
and local field potentials in relation to periodicity coding in primary
auditory cortex. J. Neurophysiol. 73: 227–245, 1995a.
EGGERMONT, J. J. AND SMITH, G. M. Separating local from global effects
in neural pair correlograms. Neuroreport 6: 2121–2124, 1995b.
EPPING, W.J.M. AND EGGERMONT, J. J. Sensitivity of neurons in the auditory
midbrain of the grassfrog to temporal characteristics of sound. II. Stimulation with amplitude modulated sound. Hear. Res. 24: 55–72, 1986.
GOLDBERG, J. M. AND BROWN, P. B. Responses of binaural neurons of dog
superior olivary complex to dichotic tonal stimuli: some physiological
mechanisms of sound localization. J. Neurophysiol. 32: 613–636, 1969.
GOLDSTEIN, M. H., KIANG, N.Y.-S., AND BROWN, R. M. Responses of the
auditory cortex to repetitive acoustic stimuli. J. Acoust. Soc. Am. 31:
356–364, 1959.
HEIL, P., RAJAN, R., AND IRVINE, D.R.F. Sensitivity of neurons in cat
primary auditory cortex to tones and frequency-modulated stimuli. II.
Organization of response properties along the ‘‘isofrequency’’ dimension.
Hear. Res. 63: 135–156, 1992.
HORST, J. W., JAVEL, E., AND FARLEY, G. R. Coding of spectral fine structure in the auditory nerve. I. Fourier analysis of period and interspike
interval histograms. J. Acoust. Soc. Am. 79: 398–416, 1986.
KENMOCHI, M., AND EGGERMONT, J. J. Autonomous cortical rhythms affect
temporal modulation transfer functions. Neuroreport 8: 1589–1593,
1997.
KNIGHT, P. L. Representation of the cochlea within the anterior auditory
field (AAF) of the cat. Brain Res. 130: 447–467, 1977.
KOWALSKI, N., VERSNEL, H., AND SHAMMA, S. A. Comparison of responses
in the anterior and primary auditory fields of the ferret cortex. J. Neurophysiol. 73: 1513–1523, 1995.
LANGNER, G., SCHREINER, C., AND MERZENICH, M. M. Covariation of latency and temporal resolution in the inferior colliculus of the cat. Hear.
Res. 31: 197–202, 1987.
MARDIA, K. V. Statistics of Directional Data. London: Academic Press,
1972.
MENDELSON, J. R., SCHREINER, C. E., AND SUTTER, M. L. Functional topography of cat primary auditory cortex: response latencies. J. Comp. Physiol
[A] 181: 615–633, 1997.
MIDDLEBROOKS, J. C., CLOCK, A. E., XU, L., AND GREEN, D. M. A panoramic code for sound location by cortical neurons. Science 264: 842–
844, 1994.
MITZDORF, U. Current source-density method and application in cat cerebral
cortex: investigation of evoked potential and EEG phenomena. Physiol.
Rev. 65: 37–99, 1985.
MOREL, A. AND IMIG, T. J. Thalamic projections to fields A, AI, P and VP
in the cat auditory cortex. J. Comp. Neurol. 265: 119–144, 1987.
PHILLIPS, D. P. Representation of acoustic events in the primary auditory
cortex. J. Exp. Psychol. HEP 19: 203–216, 1993.
PHILLIPS, D. P., HALL, S. E., AND HOLLETT, J. L. Repetition rate and signal
level effects on neuronal responses to brief tone pulses in cat auditory
cortex. J. Acoust. Soc. Am. 85: 2537–2549, 1989.
10-27-98 16:59:01
neupa
LP-Neurophys
2764
J. J. EGGERMONT
PHILLIPS, D. P. AND IRVINE, D.R.F. Properties of single neurons in the
anterior auditory field (AAF) of cat cerebral cortex. Brain Res. 248:
237–244, 1982.
RAGGIO, M. W. AND SCHREINER, C. E. Neuronal responses in cat primary
auditory cortex to electrical cochlear stimulation. I. Intensity dependence
of firing rate and response latency. J. Neurophysiol. 72: 2334–2359, 1994.
RAUSCHECKER, 1. P. Processing of complex sounds in the auditory cortex
of cat, monkey and man. Acta Otolaryngol. (Stockh.) Suppl. 532: 34–
38, 1997.
SCHREINER, C. E. AND CYNADER, M. S. Basic functional organization of
second auditory cortical field (AII) of the cat. J. Neurophysiol. 51: 1284–
1305, 1984.
SCHREINER, C. E. AND MENDELSON, J. R. Functional topography of cat primary auditory cortex: distribution of integrated excitation. J. Neurophysiol. 64: 1442–1459, 1990.
SCHREINER, C. E., MENDELSON, J., RAGGIO, M. W., BROSCH, M., AND
KRUEGER, K. Temporal processing in cat primary auditory cortex. Acta
Otolaryngol. Suppl. 532: 54–60, 1997.
SCHREINER, C. E. AND RAGGIO, M. W. Neuronal responses in cat primary
auditory cortex to electrical cochlear stimulation. II. Repetition rate coding. J. Neurophysiol. 75: 1283–1300, 1996.
SCHREINER, C. E. AND SUTTER, M. L. Topography of excitatory bandwidth
in cat primary auditory cortex: single-neuron versus multiple-neuron recordings. J. Neurophysiol. 68: 1487–1502, 1992.
SCHREINER, C. E. AND URBAS, J. V. Representation of amplitude modulation
in the auditory cortex of the cat. I. The anterior auditory field (AAF).
Hear. Res. 21: 227–241, 1986.
SCHREINER, C. E. AND URBAS, J. V. Representation of amplitude modulation
in the auditory cortex of the cat. II. Comparison between cortical fields.
Hear. Res. 32: 49–64, 1988.
/ 9k2e$$no40 J364-8
SCHULZE, H. AND LANGNER, G. Perioducity coding in the primary auditory
cortex of the Mongolian gerbil ( Meriones unguiculatus): two different
coding strategies for pitch and rhythm? J. Comp. Physiol. [A] 181: 651–
663, 1997.
SCHULZE, H., OHL, F. W., HEIL, P., AND SCHEICH, H. Field-specific responses in the auditory cortex of the unanesthetized Mongolian gerbil to
tones and slow frequency modulations. J. Comp. Physiol [A] 181: 573–
589, 1997.
STEINSCHNEIDER, M., AREZZO, J., AND VAUGHAN, H. G. Phase-locked cortical responses to a human speech sound and low-frequency tones in the
monkey. Brain Res. 198: 75–84, 1980.
STEINSCHNEIDER, M., AREZZO, J., AND VAUGHAN, H. G. Speech evoked
activity in the auditory radiations and cortex of the awake monkey. Brain
Res. 252: 353–365, 1982.
SUTTER, M. L. AND SCHREINER, C. E. Physiology and topography of neurons
with multipeaked tuning curves in cat primary auditory cortex. J. Neurophysiol. 65: 1207–1226, 1991.
TIAN, B. AND RAUSCHECKER, J. P. Processing of frequency-modulated
sounds in the cat’s anterior auditory field. J. Neurophysiol. 71: 1959–
1975, 1994.
VOLKOV, I. O. AND GALAZYUK, A. V. Cochleo- and tonotopic organization
of the second auditory cortical area in the cat. Neuroscience 82: 499–
509, 1998.
WINER, J. A. The functional architecture of the medial geniculate body
and the primary auditory cortex. In: The Mammalian Auditory Pathway:
Neuroanatomy, edited by D. R. Webster, A. N. Popper, and R. R. Fay.
New York: Springer-Verlag, 1992, p. 222–409.
ZWICKER, H., AND FASTL, H. Psychoacoustics. Facts and Models. Berlin:
Springer-Verlag, 1990.
10-27-98 16:59:01
neupa
LP-Neurophys