Electroencephalography is increasingly being used to probe the functional organization of auditory cortex. Modulation of the electroencephalographic (EEG) signal by tones was examined in primary auditory cortex (A1) of awake monkeys. EEG data were measured at 4 laminar depths defined by current source density profiles evoked by best frequency (BF) tones. Midlaminar multiunit activity was used to define the tuning characteristics of A1 sites. Presentation of BF tones increased EEG power across the range of frequencies examined (4–290 Hz), with maximal effects evident within the first 100 ms after stimulus onset. The largest relative increases in EEG power generally occurred at very high gamma frequency bands (130–210 Hz). Increases in EEG power for frequencies less than 70 Hz primarily represented changes in phase-locked activity, whereas increases at higher frequencies primarily represented changes in non–phase-locked activity. Power increases in higher gamma bands were better correlated with the A1 tonotopic organization than power increases in lower frequency bands. Results were similar across the 4 laminar depths examined. These findings highlight the value of examining high-frequency EEG components in exploring the functional organization of auditory cortex and may enhance interpretation of related studies in humans.
Intracranially recorded auditory evoked potentials (AEPs) are capable of probing the organization of human auditory cortex at high spatial and temporal resolution and have served as a “gold standard” against which noninvasive AEPs and neuromagnetic responses can be compared (e.g., Liégeois-Chauvel et al. 1994; Halgren et al. 1995; Howard et al. 2000; Brugge et al. 2003; Trébuchon-Da Fonseca et al. 2005). Even with their superior temporal and spatial resolution, it is uncertain whether these responses are optimal indices of auditory cortical organization. This concern is exemplified by the finding that field potentials are less specific than unit responses in defining the tonotopic organization of primary auditory cortex (A1) (Galván et al. 2002; Noreña and Eggermont 2002). Furthermore, averaging procedures used to generate AEPs maximize phase-locked activity at the expense of non–phase-locked (NPL) activity. Generally, phase-locked activity includes neural responses within lower frequency bands, whereas NPL activity is dominated by neural signals in the gamma (>30 Hz) frequency range (e.g., Crone et al. 2001). Electroencephalographic (EEG) activity in the gamma band reflects important processes in sensory perception (Freiwald et al. 2001; Ward 2003; Herrmann et al. 2004) and correlates better with functional magnetic resonance imaging changes than the evoked potential (Mukamel et al. 2005; Niessing et al. 2005). Thus, a central goal of this study is to compare the relative sensitivity of EEG responses within different frequency bands with basic organizational properties of auditory cortex. We investigate this question by examining the relationship between EEG responses and the tonotopy of A1 in awake monkeys (Recanzone et al. 2000). These findings can thus build upon earlier observations regarding the correlation between increased gamma activity and A1 tonotopy in anesthetized primates (Brosch et al. 2002).
Although earlier investigations focused on lower frequency gamma-band responses (∼40 Hz), more recent work has begun exploring the potential relevance of higher gamma-band frequencies in defining auditory cortical organization (Crone et al. 2001; Kaiser et al. 2002; Palva et al. 2002). These types of investigations may be especially important, given the accumulating evidence that different EEG rhythms may index discrete neural circuitry components in cortex (e.g., Salinas and Sejnowski 2001; Jones and Barth 2002; Olufsen et al. 2003; Whittington and Traub 2003). Thus, understanding the timing, strength, and relationship to A1 tonotopy of stimulus-evoked activity in distinct EEG frequency bands may help probe-specific types of cellular interactions and their relationship to the functional organization of auditory cortex.
Finally, the laminar distribution of EEG changes in various frequency bands needs to be elucidated. This consideration takes on greater importance when one considers that the most common form of intracranial recordings in humans is through grid electrodes placed over the cortical surface. Without prior information, it is impossible to infer response patterns occurring in deeper cortical laminae from frequency-specific activity profiles recorded at the cortical surface. Therefore, this study examines lamina-specific EEG signals physiologically defined by current source density (CSD) profiles to clarify spatiotemporal patterns of stimulus-modulated EEG activity in monkey A1.
Materials and Methods
Five male macaque monkeys (Macaca fascicularis) weighing between 2.5 and 3.5 kg were studied following the approval by the Animal Care and Use Committee of Albert Einstein College of Medicine. Institutional and federal guidelines governing the use of primates were strictly followed. Animals were housed in our Association for Assessment and Accreditation of Laboratory Animal Care-accredited Animal Institute. Other experiments were performed in parallel with this study to minimize the overall number of monkeys used. General methods have been extensively outlined in previous publications (e.g., Steinschneider et al. 2005). Briefly, animals were trained to sit quietly in customized primate chairs with hands restrained. Surgery was then performed using sterile techniques and general anesthesia (sodium pentobarbital). Epidural matrices were placed over holes drilled into the skull to allow electrode penetrations into the brain. Matrices consisted of 18-gauge stainless steel tubes glued together into a honeycomb form and were shaped to approximate the contour of the cortical convexity. They were stereotaxically positioned to target A1 at an angle of 30 degrees from \normal to approximate the anterior–posterior tilt of the superior temporal gyrus. This angle oriented electrode penetrations so that they were approximately orthogonal to the surface of A1. Matrices and Plexiglas bars permitting painless head fixation were embedded in dental acrylic secured to the skull with inverted bolts keyed into the bone. Peri- and postoperative care included use of anti-inflammatory, antibiotic, and analgesic medications. Recordings began 2 weeks after surgery.
Recordings were conducted in a sound-attenuated chamber covered with sound insulating foam. Animals maintained an awake state during the experiment, facilitated by frequent human contact and delivery of juice reinforcements in between recording blocks. However, brief time periods when the animals may have become drowsy are likely. Recordings were performed with multicontact electrodes constructed in our laboratory. They contained 14 recording contacts arranged in a linear array and evenly spaced at 150-μm intervals (<10% error), permitting simultaneous recording across multiple cortical laminae. Contacts were 25-μm stainless steel wires insulated except at the tip and were fixed in place within the sharpened distal portion of a 30-gauge tube. Impedance of each contact was maintained at 0.1–0.4 MΩ at 1 kHz. The reference was an occipital epidural electrode. Headstage preamplification was followed by amplification (5000×) with differential amplifiers (Grass P5, down 3 dB at 3 Hz and 3 kHz). Unprocessed data were stored on a digital tape recorder (DT-1600, MicroData Instrument, Inc., South Plainfield, NJ, sample rate 6 kHz) for later processing of neural signals.
Electrodes were positioned with a microdrive whose movements were guided by online inspection of AEPs and multiunit activity (MUA) evoked by 80-dB clicks. Test stimuli were presented when the recording contacts of the linear array electrode straddled the inversion of early cortical AEP components, and the largest evoked MUA was maximal in the middle electrode contacts. Online response averages were generated from 50–75 stimulus presentations. MUA was extracted by high-pass filtering the raw input at 500 Hz (roll-off 24 dB/octave), followed by further amplification (8×), full-wave rectification, low-pass filtering at 600 Hz (RP2 modules, Tucker Davis Technologies, Inc., Alachua, FL), and computer averaging at a digitization rate of 3400 Hz. MUA measures the envelope of action potential activity generated by neuronal aggregates, weighted by neuronal location and size. MUA is similar to cluster activity but has greater response stability than either cluster or single-unit responses (Nelken et al. 1994: Supèr and Roelfsema 2005). While reducing response variability, however, pooling single-unit activity into summed aggregate MUA measures limits the informational capacity of neuronal populations (Reich et al. 2001; Montani et al. 2007).
One-dimensional CSD analysis was used to physiologically identify the laminar location of recording sites. This information, in turn, determined the selection of the 4 electrode channels used for EEG frequency analysis. CSD was calculated from AEP laminar profiles using an algorithm that approximated the second spatial derivative of the field potentials recorded at 3 adjacent depths (Freeman and Nicholson 1975). Depths of the earliest click-evoked and tone-evoked current sinks were used to locate lamina 4 and lower lamina 3 (e.g., Müller-Preuss and Mitzdorf 1984; Steinschneider et al. 1992; Cruikshank et al. 2002). The location of the initial sink served as one of the 4 electrode channels used for EEG analysis. The second electrode site chosen for EEG analysis corresponded to the location of a slightly later current sink in upper lamina 3 (superficial sink). The third and fourth electrode sites corresponded to locations where a concurrent source located more superficially than the superficial sink (superficial source), and a deeper current source located below the initial sink (deep source) were maximal. This profile of current sources and sinks has been repeatedly identified in previous studies and serves as a reliable template for identifying the approximate laminar location of recording sites (e.g., Müller-Preuss and Mitzdorf 1984; Steinschneider et al. 1992, 2003; Cruikshank et al. 2002). The physiological procedure for laminar identification was later checked by correlation with measured widths of A1 and its laminae at select electrode penetrations obtained from histological data (see below).
Stimuli were pure tones generated and delivered at a sample rate of 100 kHz by a PC-based system using RP2 modules (Tucker Davis Technologies, Inc.). Tones ranged from 0.2 to 17.0 kHz and were 175 ms in duration with linear rise/decay times of 10 ms. All stimuli were monaurally delivered to the ear contralateral to the recorded hemisphere via a dynamic headphone (MDR-7502, Sony, Inc., Tokyo, Japan) with a stimulus-onset asynchrony of 658 ms. Sounds were presented to the ear through a 3-inch long, 60-cc plastic tube attached to the headphone. Pure tone intensity was 60-dB sound pressure level measured with a “Bruel and Kjaer” sound level meter (type 2236) positioned at the opening of the plastic tube. The frequency response of the headphone was flattened (±3 dB) from 0.2 to 17.0 kHz by a graphic equalizer (GE-60, Rane, Inc., Mukilteo, WA).
EEG signals recorded on the digital tape recorder were low-pass filtered at 400 Hz and digitized at a rate of 2000 Hz. Data from 40 single trials and the resultant AEPs were stored for each tone stimulus using Neuroscan software. Analysis time was 500 ms. Methods of EEG frequency analysis were derived from published procedures (Crone et al. 2001). This analysis is illustrated in Figure 1. Single-trial data from designated electrode depths were band-pass filtered into 16 frequency bands, 3 of which are depicted in Figure 1A. This example illustrates single-trial data from an electrode site located at the depth of the superficial sink evoked by a best frequency (BF) tone of 2.0 kHz. The AEP derived from the 40 stimulus presentations is shown in Figure 1B. Frequency bands included theta (4–8 Hz), alpha (8–14 Hz), beta (14–30 Hz), and 13 successive 20-Hz-wide bands ranging from 30 to 290 Hz. This method centers 60-Hz line voltage and its harmonics at the middle of a 20-Hz frequency band. Voltages were squared and averaged to derive the power within each frequency band. This total power (TP), which includes both phase-locked power (i.e., comprising the AEP) and NPL power, is shown in Figure 1C. To estimate the NPL power, the AEP was first subtracted from each single-trial waveform and the resultant signal band-pass filtered, squared, and averaged across single trials (Fig. 1D). Digital filtering was implemented in Matlab with finite impulse response (FIR) filters using 100 filter coefficients. FIR filters yield linear phase responses without group delay distortion across the examined frequency bands (Lyons 1997). Waveforms were shifted by the common group delay.
Filters yielded high-frequency specificity within gamma frequency bands (>30 Hz) that became progressively less specific in lower frequency bands. Frequency specificity of the filters was quantified by measuring the amplitude of a filtered test signal whose frequency was centered in each of the 16 frequency bands. Signals centered in a 20-Hz gamma frequency band (e.g., 60 Hz) were 42 dB down in adjacent 20-Hz bands and greater than 60 dB down within the next more distant bands. These latter values were essentially at the limit of our recording sensitivity. The remaining spectral filtering characteristics are shown in the Table 1 as signal decrement in decibel.
|Frequency band||Filter characteristics|
|Test signal and response (dB down)|
|6 Hz||11 Hz||22 Hz||40 Hz||60 Hz|
|Theta (4–8 Hz)||0||6||48||>60||>60|
|Alpha (8–14 Hz)||3||0||18||>60||>60|
|Beta (14–30 Hz)||>30||>12||0||39||>60|
|Frequency band||Filter characteristics|
|Test signal and response (dB down)|
|6 Hz||11 Hz||22 Hz||40 Hz||60 Hz|
|Theta (4–8 Hz)||0||6||48||>60||>60|
|Alpha (8–14 Hz)||3||0||18||>60||>60|
|Beta (14–30 Hz)||>30||>12||0||39||>60|
Note: Amplitude decrement of responses (dB down) in specific frequency bands after test signals whose frequencies were centered in other bands were processed through the filtering software. Response characteristics of the higher 20-Hz-wide frequency bands were identical to those of the lower 20-Hz-wide bands shown in the Table.
After completion of a recording series, animals were deeply anesthetized with sodium pentobarbital and perfused through the heart with physiological saline and 10% buffered formalin. A1 was delineated physiologically by its typically large amplitude responses and by a BF map that was organized with low BFs located anterolaterally and higher BFs posteromedially (e.g., Merzenich and Brugge 1973; Morel et al. 1993). Electrode tracks were reconstructed from coronal sections stained with cresyl violet and for acetylcholinesterase, and A1 was anatomically identified using published criteria (e.g., Morel et al. 1993).
Four adjacent channels of MUA located in lamina 4 and lamina 3 were averaged together for analysis of responses to pure tones. BF was defined as the tone frequency eliciting the largest amplitude MUA within the first 20 ms after stimulus onset. Analysis of band-pass filtered EEG data was performed on the responses to the BF and 8 other tones of different frequency at each recording site. The other tones were chosen to represent a wide range of MUA response strengths relative to the BF.
The modulatory effects of BF tone stimulation on the EEG were examined by comparing the logarithm of the ratio of stimulus-elicited and baseline power. This method was deemed superior to comparing absolute power amplitudes or percent power changes for the following reasons. In the case of comparing absolute amplitudes, the method was hampered by the variability in absolute values of lower frequency components across penetrations, minimizing any statistical difference that might arise in these frequency bands induced by stimulation. In the case of comparing nonlogarithmic ratios (e.g., % change), stimulus-related changes obtained in lower frequency bands may be underestimated relative to higher frequency bands due to their much larger absolute values.
Recording Site Characteristics
Results are based on the analysis of 20 electrode penetrations into A1. Criteria for study inclusion included that neural signals had minimal line voltage in the raw data, that penetrations were clearly localized to A1 and yielded large amplitude AEP, CSD, and MUA responses, and that data represented a sample of activity from multiple animals and a wide range of BFs. EEG frequency composition was unknown at the time of study inclusion. BFs of the recording sites ranged from 0.7 to 12.5 kHz. Four sites had a BF less than 1.0 kHz, 5 sites had a BF of 1.0–2.0 kHz, 7 sites had a BF of 2.0–5 kHz, and 4 sites had a BF greater than 5 kHz. Figure 2 illustrates the average CSD and MUA evoked by the BF tones from these 20 penetrations and at the 4 depths chosen for EEG analysis. Mean CSD activity in A1 begins 9 ms after stimulus onset at the location of the initial sink, whereas MUA begins several hundred microseconds later at the same depth. Peak of this sink occurs at 17 ms, whereas the colocated MUA peaks at 19 ms. MUA falls off markedly in amplitude at surrounding depths and is nearly absent at the depth of the superficial source. The superficial sink begins at 12 ms and peaks at 26 ms. It is mirrored by a more superficial source with nearly identical timing. A deeper source below the initial sink completes the early CSD profile evoked by stimulus onset.
The averaged colocated AEPs evoked by the BF tones (±standard error of the mean [SEM]) at these 4 depths are shown in Figure 3. Polarity of response components inverts between the superficial source and the deep source depths. At the intervening 2 depths, the early activity is dominated by large amplitude negativities and following positive waves. The average separation between the 4 depths is shown at the right of the figure. On average, there was a separation of 1140 μm between the superficial source and deep source depths. Dotted lines in Figure 3 demarcate the EEG interval used to calculate baseline values. By this interval of 350–450 ms poststimulus onset (175–275 ms poststimulus offset), the AEP in A1 has dissipated.
Absolute Values of Baseline EEG
TP of the EEG in the baseline interval is shown for each of the 4 recording depths in Figure 4. There is a pronounced decrease in power above the beta frequency range and an exponential decrease in power within sequentially higher gamma-band frequency ranges. Maximal power in the alpha frequency range is consistent with a predominantly awake state of the animals during recording. Importantly, noise induced by 60-cycle line voltage and its harmonics do not significantly contaminate the activity in our chosen sample. To diminish the unwieldy nature of examining and depicting activity in 16 frequency bands, various gamma frequency bands are grouped together. Gamma activity is operationally defined as occurring within the frequency band of 30–70 Hz, high gamma (HG) activity as occurring within 70–130 Hz, very high gamma (VHG) as occurring within 130–210 Hz, and ultrahigh gamma (UHG) as occurring within 210–290 Hz. Prior to these groupings, responses in each 20-Hz band were examined to ensure that peaks of activity in restricted frequency ranges were not overlooked.
Stimulus-Modulated Power: Frequency and Timing Characteristics
The modulatory effects of BF tone stimulation on the EEG are examined by comparing the logarithm of the ratio of stimulus-elicited and baseline power (see Materials and Methods). Largest increases in power relative to baseline occur within the first 100 ms poststimulus onset. Increases between 100 and 200 ms poststimulus onset decrease toward baseline, whereas activity from 200 to 300 ms is not statistically different than baseline values.
Power increases within the first 100 ms poststimulus onset are not uniform across laminae or across frequency bands (Fig. 5). At the depth of the superficial source, largest increases from baseline occur in the gamma and VHG frequency bands (repeated-measure analysis of variance [ANOVA]; F6,19 = 4.219; P < 0.001). The results of post hoc analyses (Newman–Keuls multiple comparisons test) are partially represented in the shading of the response bars shown in the figure. Solid black bars denote frequency bands where relative power is greater than at least one other band depicted by unshaded bars at a value of P < 0.01, whereas solid gray bars denote the same relationship but at a value of P < 0.05. Numbers above each frequency band denote the number of other frequency bands that display statistically smaller power increases (P < 0.05). The pattern changes at the depth of the superficial sink, where maximum increases are restricted to the VHG band (repeated-measure ANOVA; F6,19 = 3.544; P = 0.003). Although responses are significantly different at the depth of the initial sink (repeated-measure ANOVA; F6,19 = 2.294; P = 0.040), with the VHG band yielding the largest mean change from baseline, post hoc tests do not yield significant differences among the frequency bands. Frequency-specific responses at the level of the deep source are not different from each other (repeated-measure ANOVA; F6,19 = 2.153; P = 0.053).
The reliability of BF tone–evoked changes in the EEG across the various frequency bands was assessed by examining the coefficient of variation (CV = σ/μ) for each frequency band at each of the 4 recording depths (Fig. 6). Coefficient values less than one are generally considered low variance, whereas those greater than one are considered high variance. At all recording depths, the lowest frequency bands show the least reliable increases, whereas gamma, HG, and VHG uniformly show the most reliable increases. UHG activity is most reliable in deeper laminae and less reliable in superficial laminae. VHG increases are highly reliable in the deeper laminae, even though its increases at the depths of the initial sink and deep source are not significantly greater than power increases in the other frequency bands (Fig. 5).
Stimulus-Modulated Power: Total versus NPL Power
Stimulus-elicited power increases in lower frequency bands primarily represent phase-locked activity manifested in the AEP. In contrast, power increases in higher frequency bands mainly represent stimulus-induced activity (i.e., NPL power). These findings are illustrated in Figure 7, which depicts TP increases in each frequency band as solid bars, whereas increases remaining after subtraction of the AEP from each response trial (see Materials and Methods) are shown as crosshatched bars. In lower frequency bands, including the gamma band between 30 and 70 Hz, changes in NPL power are minimal despite increases in TP. In higher bands, most of the TP increases are attributed to NPL power.
The relative absence of NPL power in the lower gamma frequency band is somewhat surprising, given its reported prominence in other studies (e.g., Brosch et al. 2002). We investigated whether our negative finding was due to combining activity between 30 and 50 Hz with that between 50 and 70 Hz or due to the coarseness of the time intervals examined (100 ms). Figure 8 depicts total (solid bars) and NPL (crosshatched bars) stimulus-modulated power in the more restricted 30- to 50-Hz gamma band range and examined in more discrete 25-ms time bins. Despite this spectrally and temporally more refined examination, NPL power increases are minimal and are markedly diminished relative to the TP. Indeed, an apparent decrement below baseline values in both total and NPL power is observed after initial increases at all but the depth of the superficial source. For comparison, a similar analysis performed in the VHG range of 130–150 Hz demonstrates a large increase in NPL power from 25 to 100 ms poststimulus onset (Fig. 9). Activity in the first 25 ms after stimulus onset is primarily phase locked and represents the high-frequency content of the rapid-onset A1 AEP. The greater relative increase in NPL power compared with TP at several time points is due to the use of baseline NPL power in the calculations.
Test of Frequency-Specific Responses
The present data suggest that: 1) BF tones evoke the greatest relative increases in EEG power within the VHG range, 2) smaller increases occur in the classic gamma band of about 40 Hz except at the depth of the superficial source where large increases are observed, 3) the smallest relative power increases occur in the lower frequency bands, and 4) increases in lower frequency bands, including the lowest gamma band, primarily reflect phase-locked power, whereas increases in higher gamma bands primarily reflect NPL power. These findings, however, may have been biased by the specific filtering parameters used in this study. For instance, filtering in the theta, alpha, and beta bands was not sharp, leading to significant spillover of power in one frequency band into adjacent frequency bands. Additionally, less pronounced increases in power within the 40-Hz gamma band may have been due to spillover from higher beta frequencies or due to setting the filtering bandwidth at 20 Hz. To test for these possible contaminants, we reexamined the data set using 5 different bands. The first band was set at 4–20 Hz, thus minimizing spillover of energy from one low-frequency band into another. The second band examined lower gamma frequencies in the more restricted range of 35–45 Hz. The remaining 3 bands were all 30-Hz wide with cutoff frequencies separated from line voltage or its harmonics by 15 Hz. Thus, these bands included activity in the 75–105, 135–165, and 195–225 Hz ranges.
Results are similar to previous findings with the exception that at all depths one-way ANOVA with repeated measures yielded significant differences among the frequency bands (P < 0.05; Fig. 10, see Fig. 5). At all 4 cortical depths, the VHG frequency band of 135–165 Hz displayed the largest relative increases in power within the first 100 ms poststimulus onset, whereas the lowest frequency band of 4–20 Hz showed the smallest relative power increase. At the depths of the superficial and deep sources, large increases were also present in the classic gamma band of 35–45 Hz that were similar to those in the 135- to 165-Hz band. Furthermore, TP in both the 4- to 20-Hz and the 35- to 45-Hz bands was predominantly due to phase-locked activity, whereas power in higher frequency bands was predominantly due to NPL activity (crosshatched bars; see Fig. 7). Patterns of variability as represented by values of the CV were nearly identical to those depicted in Figure 6.
Three additional concerns may be raised by our methods. First, artificially exaggerated power increases in HG bands relative to lower frequency bands could occur if there was extensive jitter in the onset of the AEP. The higher frequency components in the AEP, present primarily at signal onset, would be diminished in the average due to temporal jitter, leading to an apparent increase in NPL power relative to phase-locked power. Second, the baseline period from which power ratio calculations were derived (350–450 ms poststimulus onset) may have been suboptimal if EEG frequency patterns had an insufficient time to stabilize following the offset of sound stimulation. Finally, rapidly occurring changes in the EEG frequency profile might be lost by the previous analyses that used 100- and 25-ms time bins.
To examine whether NPL responses in higher gamma frequencies were exaggerated due to response jitter, the AEPs were subjected to the same filtering procedures as previously performed on the single trials and compared with the TP increases. If there was extensive jitter in the onset of the AEP, then TP increases in the higher gamma frequencies should be larger than those in the AEP within the same time range but with somewhat greater temporal dispersion. The stability of the baseline period was tested by using as baseline the 5-ms period immediately preceding sound stimulation (653–658 vs. 350–450 ms poststimulus onset). Finally, temporal resolution was enhanced by examining the neural responses in 5-ms increments.
Results of this combined analysis are shown in Figure 11, which depicts contour plots of the frequency-specific response over time in 5-ms increments at all 4 recording depths for the AEP, TP, and their difference in the left, center, and right columns, respectively. Spectral power for the AEP and the TP were normalized to their separate baselines, summed in 5-ms bins, and independently normalized to the maximal bin for each frequency band. The difference (TP minus AEP) was normalized to the maximum power bin in the TP for each frequency band.
Maximum power in the AEP for all but the lowest frequency bands is restricted to a narrow time period occurring soon after stimulus onset and falls off dramatically at later time periods (Fig. 11, left column). In contrast, TP profiles show maximum power in the high frequencies that is temporally delayed relative to the AEP power at all recording depths (Fig. 11, center column). This pattern is inconsistent with NPL high-frequency components being induced by jitter of the field potential, which would have revealed maximum power at earlier time points that slowly decayed over time. Difference waveforms also support the conclusion that NPL power in higher frequencies is not primarily due to jitter in the onset of the AEP as maximal differences between the 2 power measures occur after the high-frequency power increases in the AEP have dissipated (Fig. 11, right column).
Stimulus-Modulated Power: Tonotopic Organization
Spectral sensitivity curves determined from the early MUA (<20 ms after tone onset) are roughly paralleled by the spectral sensitivity of EEG in higher gamma bands occurring within the first 100 ms. This relationship is illustrated in Figure 12, which depicts superimposed curves plotting MUA amplitude (solid lines) and VHG band activity (dotted lines) as a function of stimulus frequency. Four examples with a wide BF range at the 4 examined laminar depths are shown. In each example, there is good concordance between the frequency sensitivity of VHG band activity and that of the middle laminae MUA. Peaks of activity in the MUA are associated with similar maxima in VHG activity. Fall off in MUA amplitude as tone frequencies become progressively distant from the BF is roughly paralleled by decreases in VHG responses.
Correlations between the tone frequency tuning of middle laminae MUA and that of activity in each of the EEG frequency bands are quantified and illustrated in Figure 13. The early MUA evoked by 9 tones at each of the 20 recording sites was ranked and compared with ranks at each recording depth of the corresponding EEG power within each of the 7 frequency bands during the first 100 ms poststimulus onset. The figure depicts linear regressions showing the relationship between MUA spectral sensitivity and that of EEG within each of the frequency bands examined. Spearman correlation coefficients for the maximum and minimum correlations are also shown. Frequency bands displaying the second largest and smallest correlations with the MUA are indicated at the right of each graph.
In all cases, the strongest correlations with the spectral tuning of MUA are displayed by activity within higher gamma frequency bands. At the laminar depths of the superficial and initial sinks, the VHG band shows the highest correlation with MUA tuning, whereas the HG and UHG bands have the strongest correlations at the superficial and deep source sites, respectively. Correlations between the early MUA tonotopy and the gamma and VHG bands are nearly equal to that of the UHG band at the depth of the deep source. In contrast, the weakest correlations are always observed for the lowest EEG frequency bands. Correlations between theta and alpha activity and the frequency tuning of MUA, however, still are statistically significant (P < 0.05).
One possible concern is that the reason high-frequency gamma bands correlate better with the tonotopic organization as defined by early MUA than lower frequency bands is that these frequencies incorporate leakage of lower frequency components of action potential spikes. If this was the case, then high temporal resolution analysis of the higher gamma activity should reveal a time course of spectral sensitivity to tones that parallels the MUA. To test this hypothesis, data were reanalyzed by baseline normalizing EEG power bands to a new baseline from 653 to 658 ms poststimulus onset, summing the frequency band responses in 5-ms bins, and normalizing data to the bin containing the maximum power within each frequency band.
Results of this analysis are shown in Figure 14, which depicts contour plots of the MUA-ranked tone responses (1 through 9) across time. MUA, power in the gamma band showing the best correlation with the MUA at each recording depth, and power in the frequency band that displays the worst correlation (i.e., theta) are shown in the left, center, and right-hand columns, respectively. As expected, MUA displays high-stimulus frequency specificity at the depth of the initial sink, weaker specificity at surrounding depths, and extremely poor tuning at the superficial source depth (left-hand column). The latter finding reflects in part the near absence of MUA at this depth (see Fig. 2). The frequency tuning at the other depths is strongest at the onset of the responses (<25 ms) and subsequently deteriorates at later intervals. Best gamma responses differ from the MUA in several important ways. First, a well-maintained frequency tuning is evident at the superficial source depth despite the absence of a similar organization in the MUA. Second, the time course of this tuning is later than that observed for the MUA at all 4 recording depths and is maximal at around 50 ms. This differential time course includes activity in the UHG band at the depth of the deep source. The UHG band includes frequencies up to 290 Hz and would be the most likely frequency band to incorporate low-frequency components of action potential spikes. Theta responses display weak frequency tuning at all but the depth of the initial sink. Maxima of frequency tuning for theta responses occur at slightly earlier time points than the best gamma response and are closer in time to the interval displaying the most pronounced MUA tonotopy.
As a further check on whether action potential spikes embedded within the highest gamma bands are responsible for the stronger relationship between these bands and the MUA-defined tonotopy, we compared correlations between the MUA and the lowest and highest 20-Hz bands included in the VHG and UHG activity at the 3 lowest depths that contained sizeable MUA. If lower frequency components of spikes were primarily responsible for the enhanced correlations, then the highest 20-Hz bands closest in spectral content to the MUA should yield better correlations. This was not the case at any of the 3 depths. At the depth of the superficial sink, the correlation between the 130- to 150-Hz component of the VHG band and the middle laminae MUA is greater than that of the 190- to 210-Hz component (rs = 0.58 and rs = 0.45, respectively). A similar pattern is observed at the depth of the initial sink for the 130- to 150-Hz bands and 190- to 210-Hz bands (rs = 0.71 and rs = 0.68, respectively). Because the UHG band displayed the strongest correlation with the MUA at the depth of the deep source, we compared its 210-to 230-Hz and 270- to 290-Hz components. Here, the correlations were rs = 0.53 and rs = 0.47, respectively. All these considerations therefore argue against the correlation of HG activity with the MUA-defined tonotopic organization as being just a manifestation of lower frequency components of action potential spikes.
This study identifies 6 characteristics of frequency-specific EEG changes elicited by tone stimulation in A1 of the awake monkey. 1) Power increases span the entire range of frequencies examined (4–290 Hz). 2) Power increases are maximal in the first 100 ms after stimulus onset. 3) Across multiple cortical laminae, the most pronounced relative increases in power elicited by BF tones occur at VHG frequencies of approximately 130–210 Hz. 4) Similar but more spatially restricted increases in relative power are observed in the classic gamma range of ∼40 Hz in superficial and deep laminae. 5) Power increases in higher frequency bands (>70 Hz) predominantly represent NPL activity, whereas increases in lower frequency bands, including the classic gamma band, predominantly reflect phase-locked activity. 6) Increases in power within higher gamma bands correlate better than increases in lower frequency bands with tone frequency tuning in A1 as defined by MUA.
The observation that BF tones elicit increases in EEG power across a wide range of frequencies is consistent with other intracranial studies of auditory cortex. BF tones evoke increases in EEG power at frequencies greater than 100 Hz in A1 of anesthetized monkeys (Brosch et al. 2002). Human studies have demonstrated activation at frequencies up to 200–300 Hz (Crone et al. 2001, 2006; Edwards et al. 2005; Trautner et al. 2006). Similar increases in power are seen in visual, somatosensory, and motor cortices, indicating that enhancement of high-frequency activity is a common feature of neocortex activation (Jones and Barth 2002; Kruse and Hoffman 2002; Kayser et al. 2003; Miller et al. 2007).
Although very high EEG frequencies can routinely be observed using intracranial recording procedures, it is unclear if similar high frequencies can be identified using noninvasive techniques. With few exceptions (e.g., Hashimoto et al. 1996), the majority of noninvasive studies using both magnetic and electrophysiologic recordings have examined higher gamma frequencies up to 80–100 Hz (e.g., Kaiser et al. 2002; Palva et al. 2002; Fitzgibbon et al. 2004; Goffauxa et al. 2004; Cho et al. 2006; Melloni et al. 2007). It is uncertain from the literature whether these upper cutoffs in the examined frequencies reflect an arbitrary decision or one based on data, suggesting that even higher cutoffs would not yield further useful information. Regardless, this frequency range is within the HG band that we found to display highly reliable increases in total and NPL power and to have the strongest correlation at the depth of the superficial source with tonotopy of middle laminae MUA. Furthermore, it is unknown whether species differences might suggest equivalence between gamma responses in the range of 70–100 Hz in the human and VHG responses in the monkey. The finding that maximal stimulus-elicited EEG power changes in human auditory cortex occur at frequencies of approximately 100 Hz (Edwards et al. 2005) whereas those in the monkey occur at slightly higher frequencies lends support to this idea.
Our finding that NPL gamma activity in A1 was maximal in the first 100 ms after stimulus onset is at variance with previous observations in the anesthetized monkey (Brosch et al. 2002). We observed that higher gamma band activity is superimposed upon middle latency AEP components, whereas the previous study found that gamma activity began near the termination of the AEP (Brosch et al. 2002). This discrepancy may reflect the use of ketamine anesthesia in the previous study as ketamine markedly changes the strength, temporal pattern, and timing of neuronal activity in A1 (Zurita et al. 1994; Syka et al. 2005). Similar to our findings, HG activity in human auditory cortex has been observed as early as 35 ms after stimulus onset in one intracranial study (Edwards et al. 2005) and within the first 100 ms in another intracranial study (Trautner et al. 2006) and in neuromagnetic signal recordings (Palva et al. 2002). High-frequency gamma activity is superimposed upon the N20m component of the somatosensory evoked magnetic response in humans (Hashimoto et al. 1996), the earliest evoked potential components in rat S1 (Jones and Barth 2002) and occurs within 50 ms of stimulus onset in monkey V4 (Fries et al. 2001). These considerations indicate that high-frequency NPL gamma activity is a prominent feature of early sensory processing within cortex.
An unexpected finding was the relative paucity of increases in NPL activity in the lower gamma range (∼40 Hz). Instead, increases predominantly represented more rapid phase-locked components of the AEP. This dissociation between lower and higher gamma bands suggests that they index different features of auditory cortical activation, a point that will be expanded on below. Similar dissociations have been observed in human auditory cortex, where increases in gamma activity at ∼40 Hz were inconsistently observed whereas more reliable increases were observed at higher gamma frequencies (Crone et al. 2006; Trautner et al. 2006). Furthermore, higher gamma activity can occur in the absence of lower gamma increases and has a more widespread distribution (Crone et al. 2001, 2006; Edwards et al. 2005).
Increases in power within higher gamma bands correlate better than increases in lower frequency bands with the tonotopic organization of A1 as determined by MUA. These findings extend previous results also showing that gamma-band enhancement is generally maximal for BF tones (Brosch et al. 2002). The latter study, however, did not compare the relationship between the tonotopic organization and spectral modulation of the EEG both within and outside the gamma frequency range. Here, we observe that power increases in low frequencies that predominantly represent components of the AEP have the poorest correlation with this fundamental organizational feature. These observations support the interpretation that early components of the AEP index excitatory postsynaptic events evoked by a broad array of subcortical inputs that become sharpened by intracortical inhibitory mechanisms (Noreña and Eggermont 2002). The stronger correlations observed for higher gamma bands thus suggest that these frequencies reflect neural activity that, at least in part, is generated after this sharpening process has occurred.
The strong correlation observed between HG activity and A1 tonotopy extends similar findings observed in visual cortex. Gamma activity increases in monkey V1 are maximal when sites are presented with optimal visual stimuli and also show greater stimulus specificity than increases in lower frequency activity (Friedman-Hill et al. 2000; Frien et al. 2000). This relationship between the magnitude of HG activity and stimulus tuning extends beyond primary visual areas and has been identified in areas MT and MST of behaving monkeys (Kruse and Hoffman 2002). Similar correlations between spatially restricted increases in HG activity and the somatotopic organization of human motor cortex when subjects made specific movements have also been observed (Miller et al. 2007).
One could argue that the presence of lower frequency components of action potentials in the EEG is responsible for the better correlation of higher gamma bands with the MUA-defined tonotopy. Further, it must be acknowledged that these components are likely embedded within the highest gamma frequencies examined. However, multiple considerations strongly argue against this interpretation. First, if action potentials were the sole reason for the enhanced correlation between the higher gamma activity and the MUA, then the highest frequency band (i.e., UHG) would have shown the strongest correlation as its frequency content is closest to that of the MUA and to the spectral composition of extracellular spikes. Instead, at all recording depths except the location of the deep source in the CSD, the best correlations were observed for the lower frequency HG bands (Fig. 13). Even at the depth of the deep source, correlations with MUA-defined tonotopy are almost identical for the UHG, VHG, and gamma bands. Frequency bands lower than the UHG band are unlikely to have included significant contributions from action potentials given the high specificity of the filters within the gamma range (see Table 1). Second, the lowest 20-Hz component of the VHG band at the depths of the superficial and initial sink and the lowest 20-Hz component of the UHG band at the depth of the deep source display stronger correlations with the tonotopy than the highest 20-Hz component within its respective gamma band. Third, the temporal distributions of EEG showing the strongest tonotopy are later than those of the MUA (Fig. 14). This temporal dissociation would not be expected if lower frequency components of action potentials were driving the tonotopy in the EEG. Fourth, strong correlations between MUA-derived tonotopy defined from midlaminar depths and HG, VHG, and gamma bands are present at the depth of the superficial source that does not display a similar profile in the colocated MUA. These findings all suggest that processes other than action potentials are responsible for the strong relationship between the tonotopic organization of A1 and high-frequency components of the EEG.
Detailed in vitro and in vivo recordings offer insights into the potential processes underlying frequency-specific changes in the EEG. Lower frequency EEG components present in the evoked potential and generated by the initial and superficial current sinks primarily reflect stimulus-driven excitatory post-synaptic potentials mediated by glutamatergic synapses (Metherate and Cruikshank 1999). In contrast, higher frequency gamma activity reflects circuit interactions between pyramidal cells and local interneurons (Metherate and Cruikshank 1999; Cunningham et al. 2004; Traub et al. 2005). Gamma frequencies greater than 80 Hz are derived in part from synaptic and gap junction–mediated interactions between interneurons and gap junction coupling of pyramidal cells at the level of axonal plexes (Jones and Barth 2002; Brunel and Wang 2003; LeBeau et al. 2003; Traub et al. 2003, 2005; Cunningham et al. 2004).
Translational relevance of our study is enhanced by demonstrating that the tonotopic organization is reflected in HG activity recorded near the cortical surface. Most intracranial recordings in humans are performed using electrode grids placed adjacent to lamina 1. Results from these recordings could reflect organizational properties specific to upper laminae or could reflect more extensive properties that span multiple cortical laminae. The present findings indicate that at least some organizational schemes of sensory cortex originating within the cortical depths are reflected by higher gamma activity recorded near the cortical surface. Additional work will be required to determine whether other organizational schemes are reflected by high-frequency gamma activity recorded superficially both within A1 and secondary auditory areas. Human intracranial recordings do, however, support the reliability of high-frequency gamma activity recorded from grid electrodes as a specific index of cortical activation within multiple nonprimary sensory cortical areas and motor cortex (Crone et al. 2001, 2006; Sinai et al. 2005; Miller et al. 2007).
Our results should not be construed to indicate that the AEP does not reflect the tonotopic organization as components of the AEP do accurately reflect the spectral sensitivity of A1 (e.g., Fishman et al. 2000). The poorer correlation with tonotopy observed for low versus high EEG frequencies may be based on using power changes as our principal measure. Lower frequencies that dominate the AEP may represent in part a phase resetting of ongoing EEG rhythms (Shah et al. 2004; Lakatos et al. 2005), whereas higher frequencies likely represent new energy derived from cellular interactions.
Several methodological issues deserve comment. Results of subtracting the AEP from single trials to obtain an estimate of the power of NPL activity (Crone et al. 2001) must be interpreted with caution. Variability of single-trial responses will artificially inflate NPL activity in power calculations. This concern, however, does not affect measures of TP nor does it negate our qualitatively based conclusions regarding NPL power and conclusions derived from other measures that do not involve subtractions. For related reasons of intertrial variability and the possible introduction of spurious results, single-trial CSD analyses of the spatial distribution of gamma activity were not performed (see Jones and Barth 2002).
The patterns of A1 EEG activity identified in this report have significant limitations and will almost assuredly change with modifications of the experimental paradigm. These limitations may be a major contributor to the paucity of NPL gamma activity at ∼40 Hz that we observed. Use of more “preferred stimuli” that lead to enhanced sustained neuronal firing with a wider laminar distribution will likely modulate the dynamics of EEG activity within multiple spectral bands (Wang et al. 2005). EEG changes can also be expected for stimuli of greater biologic relevance and for those that require attention or discrimination coupled to behavioral responses. For instance, attention enhances both lower and higher frequency gamma activity in sensory cortex (Salinas and Sejnowski 2001). Further work will be required to identify these expected EEG changes and examine the degree of concordance with the patterns of activity observed in the present data set.
Gamma activity has been envisioned as a dynamic network filter with profound implications for cortical functions (Sejnowski and Paulsen 2006). Crucial for these functions is the postulated role of gamma activity in promoting locally coherent firing of pyramidal cells (Salinas and Sejnowski 2001; Sejnowski and Paulsen 2006), which in turn can facilitate synaptic plasticity within spatially restricted networks (Cunningham et al. 2004; Traub et al. 2005; Sejnowski and Paulsen 2006), modulate neuronal activation within downstream regions (deCharms and Merzenich 1996), and support binding of multiple stimulus attributes encoded in distributed networks (e.g., Freiwald et al. 2001). It is estimated that over 1500 patients per year undergo invasive EEG monitoring (Kahana 2006). These patients offer unique opportunities to examine human brain physiology in great detail (Crone et al. 2006). This report highlights the utility of high-frequency gamma activity to index features of auditory cortical organization in monkeys, which may enhance interpretation of translational work in humans.
National Institutes of Health (DC00657 and HD01799).
The authors wish to thank the excellent technical assistance of Ms Jeannie Hutagalung and Ms Shirley Seto. Conflict of Interest: None declared.