## Abstract

A striking feature of some field potential recordings in visual cortex is a rhythmic oscillation within the gamma band (30–80 Hz). These oscillations have been proposed to underlie computations in perception, attention, and information transmission. Recent studies of cortical field potentials, including human electrocorticography (ECoG), have emphasized another signal within the gamma band, a nonoscillatory, broadband signal, spanning 80–200 Hz. It remains unclear under what conditions gamma oscillations are elicited in visual cortex, whether they are necessary and ubiquitous in visual encoding, and what relationship they have to nonoscillatory, broadband field potentials. We demonstrate that ECoG responses in human visual cortex (V1/V2/V3) can include robust narrowband gamma oscillations, and that these oscillations are reliably elicited by some spatial contrast patterns (luminance gratings) but not by others (noise patterns and many natural images). The gamma oscillations can be conspicuous and robust, but because they are absent for many stimuli, which observers can see and recognize, the oscillations are not necessary for seeing. In contrast, all visual stimuli induced broadband spectral changes in ECoG responses. Asynchronous neural signals in visual cortex, reflected in the broadband ECoG response, can support transmission of information for perception and recognition in the absence of pronounced gamma oscillations.

## Introduction

Recordings of stimulus induced field potentials in visual cortex sometimes show a narrowband increase in spectral power within the gamma range typically peaking ∼40–60 Hz (reviewed by Fries et al. 2008). This narrowband response has attracted a great deal of attention because of the suggestion that the oscillations play a key role in perception and cognition (Eckhorn et al. 1988; Gray et al. 1989, 1999; Fries et al. 2007), and because of the related hypothesis that the precise timing of these neural oscillations is critical for information transfer between remote cortical sites (Fries 2005).

Some researchers have made vigorous counterarguments against the functional significance of these oscillations for perception (Shadlen and Movshon 1999; Thiele and Stoner 2003; Roelfsema et al. 2004). Recent studies in nonhuman primates show that the characteristics of narrowband gamma responses depend on the spatial structure of the visual stimulus. For example, the amplitude (Henrie and Shapley 2005) and peak frequency (Ray and Maunsell 2010) of the gamma oscillations increase with grating contrast. Large gamma oscillations can be recorded in response to a large grating stimulus, but the response amplitude is reduced when a perpendicular grating is superimposed to create a plaid pattern (Lima et al. 2010; Bartolo et al. 2011), when noise is superimposed on the grating (Zhou et al. 2008; Jia et al. 2011), when the grating is relatively small (Bauer et al. 1995; Gieselmann and Thiele 2008; Jia et al. 2011; Ray and Maunsell 2011; Jia Xing et al. 2013), or when natural movies are viewed rather than a grating (Kayser et al. 2003). Some researchers have argued that these stimulus-dependent variations in frequency and amplitude of the oscillatory gamma response preclude a role for oscillations in perceptual binding or communication (Ray and Maunsell 2010).

Despite the clear stimulus dependencies of gamma oscillations, hypotheses about a fundamental role of oscillations for vision have not been rejected. Several reasons are offered. First, the fact that parameters of the gamma oscillation depend on stimulus properties might not preclude narrowband gamma oscillations from playing an important role in information transfer (Bosman et al. 2012; Roberts et al. 2013). Second, the absence of gamma oscillations when viewing natural stimuli (Kayser et al. 2003), was explained as a consequence of motion signals in the stimuli masking gamma oscillations rather than the oscillations being sensitive to the spatial structure of the image. Finally, a recent study in nonhuman primates claims that gamma oscillations are elicited in macaque during free viewing of all static natural images (Brunet et al. 2015).

If narrowband gamma oscillations are essential for seeing and recognition (Melloni et al. 2007) and for feedforward information transfer (Bosman et al. 2012), then we expect these oscillations to be elicited by all static, visible stimuli, not just a restricted set—say stimuli with a periodic spatial or temporal structure. Moreover, it is important to analyze the data using methods that distinguish narrowband gamma oscillations, which are indicated by a peak in the spectral response (Lopes da Silva 2013), from broadband signals, which span the gamma range but do not have a distinct peak (Miller, Sorensen et al. 2009). Recent reports have claimed that these 2 signals may originate from different neural circuits (Manning et al. 2009; Ray and Maunsell 2011). Prior reports of gamma oscillations in human intracranial recordings did not always separate these 2 responses (Tallon-Baudry et al. 2005; Crone et al. 2011).

We specifically looked for narrowband gamma oscillations using intracranial recordings in human visual cortex, and we analyzed the dependence of these oscillations on the visual stimulus. We found that static grating stimuli elicited both narrowband gamma and broadband high frequency responses in a large area of visual cortex. The amplitude of the oscillations was large and the narrowband response could be measured on single trials. However, highly visible static noise patterns failed to elicit a narrowband response. Instead, these patterns produced a substantial broadband response. Only a small subset of face and house picture stimuli elicited narrowband oscillations, and these responses were present in only a restricted region within early visual areas. We could find no gamma oscillations on the fusiform or parahippocampal gyri, regions that appear to be important for perceiving face and house stimuli (Allison et al. 1994; Puce et al. 1995; Parvizi et al. 2012).

The amplitude of the narrowband oscillations depends substantially on the spatial structure of the image, so much so that many plainly visible stimuli do not give rise to narrowband responses. The cortical locations of the oscillations to face and house stimuli are not located in regions thought to be responsible for perception of these objects. We conclude that gamma oscillations do not play an essential role for seeing many types of visual stimuli.

## Materials and Methods

### Subject and Procedure

ECoG data were measured from 2 subjects who were implanted with subdural electrodes (2.3 mm diameter, AdTech Medical Instrument Corp.) for clinical purposes. All subjects gave informed consent to participate in the study and the study was approved by the Stanford University IRB. During ECoG measurements stimuli were shown on a 15 in. MacBook Pro laptop using Psychtoolbox (http://psychtoolbox.org/). The laptop was placed 60 cm from the subject's eyes at chest level. Screen resolution was 1280 × 800 pixels (33 × 21 cm). In Subject 1, a functional magnetic resonance imaging (fMRI) scan was performed for retinotopic mapping before the implantation of the electrodes and the same experiment was repeated during ECoG (data presented in Winawer et al. 2013, see Supplementary Fig. S1).

In Experiment 1, the subjects viewed static images of gratings and noise patterns for 500 ms each, with 500 ms of zero-contrast (mean luminance) between successive stimuli. Stimuli came from 7 classes (30 exemplars per class, 25 × 25°), including high contrast vertical gratings (0.16, 0.33, 0.65, or 1.3 duty cycles per degree square wave) and noise patterns (spectral power distributions of $k/f4,k/f2,$ and $k/f0$). Subjects fixated on a dot in the center of the screen that alternated between red and green, changing colors at random times. Subject 1 pressed a button when the fixation dot changed color. Subject 2 fixated the dot but did not make manual responses because these responses were found to interfere with visual fixation.

In Experiment 2, the subject viewed static images of faces or houses (50 each, 10 × 10°) in random order with a small amount of stationary (not moving) white noise superimposed on the image (70% gray-scale image + 30% noise in pixel intensities). Images were presented for durations ranging from 293 to 2625 ms, with a mean ISI of 1838 ms. Only stimuli with durations of at least 460 ms were included in analyses (n = 99). The subject indicated whether the image was a face or a house by pressing a button.

### ECoG Recording

ECoG data were recorded at 3052/1528 Hz (Subject 1/Subject 2) from 118/96 electrodes through a 128-channel Tucker Davis Technologies recording system (http://www.tdt.com). To localize electrodes, a computed tomography (CT) scan was acquired after electrode implantation and co-registered with a preoperative structural MRI scan. Electrodes were localized from the CT scan and co-registered to the MRI (Hermes et al. 2010). Electrodes that showed large artifacts or showed epileptic activity, as determined by the patient's neurologist (JP) were excluded from analysis (7/35 electrodes were excluded in Subject 1/Subject 2 in Experiment 1 and 13 electrodes were excluded in Experiment 2). Off-line, data were re-referenced to the common average, low pass filtered and resampled at 1000 Hz for computational purposes using the Matlab resample function. Line noise was removed at 60, 120 and 180 Hz using a third order Butterworth filter. Results from the manuscript were unchanged if data were pairwise re-referenced to a “silent” electrode rather than to the common average.

### ECoG Analyses

#### Time Frequency Analysis

To visualize spectral power changes during stimulus presentation, a time frequency analysis was performed around stimulus onset (−200 to 1000 ms) with a multitaper approach (Percival and Walden 1993) using chronux (http://www.chronux.org/ (Mitra and Bokil 2008)). A moving window of 200 ms (with overlap of 50 ms) and the use of 5 tapers result in a frequency resolution of 5 Hz, with a spectral smoothing of ±15 Hz. To normalize the responses to baseline, the average spectrum from all inter trial intervals 250–500 ms after stimulus offset was computed and divided from every time bin. The base 10 log was then computed on this normalized power and plotted (Figs 1 and 4).

Figure 1.

Time–frequency power estimates in V1/V2 for gratings and noise stimuli. (A) ECoG electrode locations in Subjects 1 and 2. (B) Grating stimuli were presented for 500 ms, insets in circles show a magnified portion of the stimuli for visibility. Time–frequency power estimates (“spectrograms”) from an electrode in V1/V2 from Subject 1 (top row) and Subject 2 (bottom row). (C) Noise patterns were presented for 500 ms. Spectrograms from the same early visual electrode in Subject 1 (top row) and Subject 2 (bottom row). All spectrograms are normalized with respect to the same baseline: the inter-stimulus interval between all trials (from 750–1000 ms after stimulus onset). Spectrograms are cut off at a maximum of ±1.3 log10 units. The multitaper approach results in a temporal smoothing of 200 ms and a frequency smoothing of ±15 Hz. Spectrograms represent averages across all trials of a given type.

Figure 1.

Time–frequency power estimates in V1/V2 for gratings and noise stimuli. (A) ECoG electrode locations in Subjects 1 and 2. (B) Grating stimuli were presented for 500 ms, insets in circles show a magnified portion of the stimuli for visibility. Time–frequency power estimates (“spectrograms”) from an electrode in V1/V2 from Subject 1 (top row) and Subject 2 (bottom row). (C) Noise patterns were presented for 500 ms. Spectrograms from the same early visual electrode in Subject 1 (top row) and Subject 2 (bottom row). All spectrograms are normalized with respect to the same baseline: the inter-stimulus interval between all trials (from 750–1000 ms after stimulus onset). Spectrograms are cut off at a maximum of ±1.3 log10 units. The multitaper approach results in a temporal smoothing of 200 ms and a frequency smoothing of ±15 Hz. Spectrograms represent averages across all trials of a given type.

#### Spectral Analysis

To quantify the responses, we calculated power spectra and separated ECoG responses into broadband and narrowband gamma band spectral power increases. To control for the influence of evoked activity on the spectrum, event related potentials (ERPs) were calculated per condition and the condition-specific ERP was regressed from each trial. This procedure makes sure that the broadband increase is not due to a sharp edge in the ERP; the same pattern of results is obtained if this step is omitted. For each condition, the average power spectral density was calculated every 1 Hz by Welch's (1967) method with a 500 ms window (0–500 ms after stimulus onset, and 0–500 ms after stimulus offset for the baseline) and a Hamming window to attenuate edge effects (Fig. 2 and see Supplementary Fig. S2). ECoG data are known to obey a power law and to capture broadband and narrowband gamma increases separately the following function (F) was fitted to the average log spectrum from 35 to 200 Hz (leaving out 60 Hz line noise and harmonics) from each condition:

(1)
$P(x)=(βbroadband−nx)+βnarrowbandG(x|μ,σ)$
Figure 2.

Power spectra in V1/V2 for gratings and noise stimuli. Power spectra to the grating stimuli for the same electrodes shown in Figure 1 on a log–log plot (top, Subject 1; bottom, Subject 2). The black line shows the power spectrum from the grating stimuli (A) and noise stimuli (B). The dashed line shows the power spectrum from the baseline periods. A line plus Gaussian (model) was fitted to the average power spectra of each condition (data) in log–log space, resulting in one weight for broadband and one weight for narrowband increases relative to baseline. The fit from this model is plotted in gray.

Figure 2.

Power spectra in V1/V2 for gratings and noise stimuli. Power spectra to the grating stimuli for the same electrodes shown in Figure 1 on a log–log plot (top, Subject 1; bottom, Subject 2). The black line shows the power spectrum from the grating stimuli (A) and noise stimuli (B). The dashed line shows the power spectrum from the baseline periods. A line plus Gaussian (model) was fitted to the average power spectra of each condition (data) in log–log space, resulting in one weight for broadband and one weight for narrowband increases relative to baseline. The fit from this model is plotted in gray.

In which,

$x=log10(frequency)$

$G(x|μ,σ)=e−(x−μ)2/2σ2$
with 10σ = 1.1 Hz and 35 Hz < 10μ < 80 Hz.

The slope of the log–log spectral power function (n) was fixed for each electrode by fitting it based on the average power spectrum of the baseline. To estimate whether broadband and gamma increases were significant, confidence intervals were calculated by a bootstrap procedure. For each condition C with Nc trials, Nc trials were drawn randomly with replacement and power spectra were averaged. The function P is the average log-power spectrum from these trials and the parameters β were fitted. This was repeated 100 times, resulting in a distribution of broadband and narrowband weights. Effects are reported as significant if the 95% overlapping confidence intervals were not overlapping with the 95% confidence interval from the baseline period, and are not further corrected for multiple comparisons across electrodes.

### Example Datasets and Software Code

In the interest of reproducible scientific computation (Gavish and Donoho 2012; LeVeque et al. 2012), the full Matlab code and sample data used to create Figures 1, 2, 3, 4B and 5 are made available in the Supplementary Materials.

Figure 3.

Spatial distribution of narrowband (gamma) and broadband weights. (A) Spectral power changes during visual stimulation compared with baseline were separated in increases in narrowband gamma rhythms and broadband increases by fitting a power law shape with a Gaussian. (B) Spatial distribution of narrowband gamma (top) and broadband weights (bottom) in Subject 1 during grating stimuli. (C) Spatial distribution of narrowband gamma (top) and broadband weights (bottom) in Subject 1 during noise stimuli. Electrodes that showed a significant increase compared with baseline are plotted in color on the rendered brain surface (P < 0.05 uncorrected). (D) ECoG electrodes labeled by visual field mapping experiment (Winawer et al. 2013). CalS = calcarine sulcus, PHG = parahippocampal gyrus, CoS = collateral sulcus, FG = fusiform gyrus, and OTS = occipital-temporal sulcus.

Figure 3.

Spatial distribution of narrowband (gamma) and broadband weights. (A) Spectral power changes during visual stimulation compared with baseline were separated in increases in narrowband gamma rhythms and broadband increases by fitting a power law shape with a Gaussian. (B) Spatial distribution of narrowband gamma (top) and broadband weights (bottom) in Subject 1 during grating stimuli. (C) Spatial distribution of narrowband gamma (top) and broadband weights (bottom) in Subject 1 during noise stimuli. Electrodes that showed a significant increase compared with baseline are plotted in color on the rendered brain surface (P < 0.05 uncorrected). (D) ECoG electrodes labeled by visual field mapping experiment (Winawer et al. 2013). CalS = calcarine sulcus, PHG = parahippocampal gyrus, CoS = collateral sulcus, FG = fusiform gyrus, and OTS = occipital-temporal sulcus.

Figure 4.

Spectral power changes during face and house viewing. (A) Examples of face and house stimuli presented during experiment 2. (B) Spectrograms from an early visual electrode in Subject 1 (Fig. 1A) while seeing faces (left) and houses (right). (C) Spatial distribution of significant increases in narrowband and broadband weights (P < 0.05 uncorrected). Early visual areas show little or no narrowband gamma increases while viewing faces (top left) and houses (top right), but large increases in broadband power (bottom). The fusiform gyrus and parahippocampal gyrus show significant broadband increases, but no gamma increases, while viewing faces and houses, respectively.

Figure 4.

Spectral power changes during face and house viewing. (A) Examples of face and house stimuli presented during experiment 2. (B) Spectrograms from an early visual electrode in Subject 1 (Fig. 1A) while seeing faces (left) and houses (right). (C) Spatial distribution of significant increases in narrowband and broadband weights (P < 0.05 uncorrected). Early visual areas show little or no narrowband gamma increases while viewing faces (top left) and houses (top right), but large increases in broadband power (bottom). The fusiform gyrus and parahippocampal gyrus show significant broadband increases, but no gamma increases, while viewing faces and houses, respectively.

Figure 5.

Narrowband (gamma) and broadband power in individual trials. (A) Increases in narrowband gamma (top row) and broadband (bottom row) in a V1/V2 electrode in Subject 1 during all individual trials of Experiment 1. Horizontal bars indicate the median (solid) and quartiles (dotted) for each condition. (B) The same for Experiment 2. This electrode shows increased gamma oscillations for only some face and house stimuli, it shows a broadband increase for almost all stimuli.

Figure 5.

Narrowband (gamma) and broadband power in individual trials. (A) Increases in narrowband gamma (top row) and broadband (bottom row) in a V1/V2 electrode in Subject 1 during all individual trials of Experiment 1. Horizontal bars indicate the median (solid) and quartiles (dotted) for each condition. (B) The same for Experiment 2. This electrode shows increased gamma oscillations for only some face and house stimuli, it shows a broadband increase for almost all stimuli.

## Results

In one experiment, we measured responses to square wave gratings of different spatial frequencies and noise patterns with several amplitude spectra. In a second experiment, we measured responses to images of faces and houses. For all stimuli in both experiments, the spectral responses were modeled as the sum of 3 components: a baseline, a narrowband gamma signal, and a broadband signal (Winawer et al. 2013). Quantitatively modeling the responses enables us to test for stimulus specificity of both types of signals across several areas in visual cortex.

### Gratings, but Not Noise Patterns, Elicit Narrowband Gamma

Measurements were made in visual cortex in 2 subjects. To illustrate the principles, we first consider responses in 2 electrodes (one from each subject) located near the foveal representation at the V1/V2v border (Fig. 1A). We then discuss a more extensive set of data.

The ECoG responses differ strikingly as a function of the spatial pattern in the stimuli. Here, we show the complete response spectrograms from both subjects to square wave gratings at several spatial frequencies (Fig. 1B) and several types of noise patterns (Fig. 1C). The power spectra are shown in Figure 2. In all cases, the square wave stimuli elicited robust narrowband gamma responses, peaking near 50 Hz (Figs 1B and 2A). This response duration is sustained for the entire stimulus duration (0.5 s). These ECoG oscillations are consistent with local field potential (LFP) recordings in animal and magnetoencephalography (MEG) and electroencephalography (EEG) recordings in human (Fries et al. 2008).

There is also a spectrally broadband response elevation. The broadband response spans a wide range of frequencies, from 50–200 Hz (upper limit due to the noise floor). This response begins slightly earlier in time (50–100 ms) than the narrowband response. It is larger at stimulus onset and then continues at a lower level for the stimulus duration. The amplitude of the broadband response increases as the fundamental spatial frequency of the grating stimuli rises from 0.16 to 1.28 cycles per degree.

The responses to the spatial noise patterns are very different from the responses to the gratings (Figs 1C and 2B). The noise stimuli produce a broadband elevation in spectral power, but there are no narrowband gamma oscillations. The broadband response to the noise stimuli is similar to the broadband response to the gratings; the amplitude is largest at stimulus onset and then continues throughout the stimulus presentation but at a lower level (Fig. 2C).

The response pattern in the foveal V1/V2 electrodes in the 2 subjects shown in Figures 1 and 2 is replicable across >30 other cortical locations (Fig. 3). Across visual sites, the response is well approximated as a mixture of a narrowband gamma oscillation and a broadband response. To summarize the measurements, we modeled the power spectrum as the weighted sum of 2 components in log-power by log-frequency plot: (1) a straight line from 35 to 200 Hz (broadband) and (2) a Gaussian centered between 35 and 80 Hz (gamma band oscillation) (Equation 1; see Supplementary Fig. S2). This allowed us to visualize the spatial distribution of the 2 signal components across the electrode arrays.

Nearly all electrodes (33/34) in visual areas V1/V2/V3 show a significant narrowband gamma response to at least one of the square wave gratings (Fig. 3B, top). Most (31/34) also have a significant broadband response (Fig. 3B, bottom see Supplementary Fig. S2). For the noise patterns, however, only 2 electrodes on V1/V2/V3 have measurable gamma responses (Fig. 3C, top, pink noise condition). These 2 electrodes were not located near the foveal representation.

Most V1/V2/V3 electrodes (28/34) have a significant broadband response to at least one of the 3 types of noise stimuli (Fig. 3C, bottom; see Supplementary Fig. S2). Broadband responses in the fovea are larger for white and pink noise than for Brown noise patterns, consistent with the observation that foveal V1/V2/V3 is less sensitive to low spatial frequencies (Wandell 1995), which dominate the Brown noise stimuli. Peripheral electrodes show the opposite pattern: a larger broadband response to the Brown noise stimuli than the white noise stimuli. This spatial pattern is consistent with the fact that the periphery is more sensitive to low spatial frequencies. The ventral occipital-temporal electrodes responded occasionally, but weakly, to both grating and noise stimuli.

### The Spatial Distribution of Broadband and Narrowband Responses to Object Stimuli Differ

Next, we measured responses to face and house picture stimuli (Subject 1; Fig. 4). During the ECoG recording, the subject correctly indicated for each image whether it was a face or a house (n = 99, mean reaction time ± standard deviation (SD): 682 ± 160 ms for faces and 726 ± 202 ms for houses). There was little narrowband gamma evoked in electrodes near the foveal representation in V1/V2. Spectrograms from the V1/V2 electrode with the strongest narrowband gamma response are shown in Figure 4B. In foveal V1/V2, 0 of 2 electrodes showed significant narrowband gamma responses for faces, 1 of 2 for houses (Fig. 4C). Narrowband gamma responses were not observed in the fusiform nor parahippocampal gyri. Some weak but significant gamma oscillations were observed in peripheral V1/V2/V3, in cortical regions that represent the visual field position at the stimulus edge.

The spatial pattern of broadband responses is quite different from the pattern of narrowband responses (Fig. 4C). The distribution of this activity is similar to the pattern observed in functional MRI experiments. The largest response is near the occipital pole (in the 2 electrodes on foveal V1/V2). This is expected because these stimuli subtended a relatively small visual angle (10° diameter). In addition, significant broadband responses were measured on the fusiform gyrus for face stimuli and on the parahippocampal gyrus for house stimuli, consistent with the location of face- and place-selective regions of ventral temporal cortex (Allison et al. 1994; Puce et al. 1995; Kanwisher et al. 1997; Aguirre et al. 1998; Epstein and Kanwisher 1998).

### Individual Trial Analysis

Noise patterns did not elicit reliable gamma oscillations on single trials; but gratings did. For an electrode on the boundary of foveal V1/V2, the narrowband gamma responses to the noise stimuli were beyond the 50% confidence interval defined by the baseline period response for only 4 of 90 trials (Fig. 5A, top). For the highest spatial frequency grating stimuli, the narrowband gamma responses were beyond the 50% confidence interval of the baseline periods on 28 of 30 trials (Fig. 5A, top).

A broadband response was measured on nearly every trial with stimuli containing relatively high spatial frequency patterns (pink and white noise patterns; gratings of 1.3 cpd) (Fig. 5A, bottom). The relative insensitivity of a foveal V1/V2 electrode to very low spatial frequencies (0.16 cpd grating and brown noise patterns) is expected (Wandell 1995).

For face and house stimuli, the narrowband gamma responses were mostly in the baseline range (Fig. 5B, top). For a few trials, however, narrowband gamma responses exceeded baseline. The broadband responses to faces and houses were reliably outside the baseline range (Fig. 5B, bottom).

## Discussion

Narrowband gamma responses to certain stimuli can be reliably measured with ECoG in human visual cortex. In this general sense, there is agreement between ECoG and MEG/EEG in human, and LFP measurements in cat and macaque (reviewed by Fries et al. 2008). When present, the ECoG narrowband oscillations are substantially (∼1.5 log10 units) above baseline and they can be observed on individual trials.

However, many types of static, visible stimuli fail to elicit narrowband gamma oscillations. Three different types of noise patterns fail to elicit narrowband gamma oscillations. Many pictures of faces and houses tested do not elicit gamma oscillations.

The broadband response amplitude and position is in much better agreement with excitations expected from visible stimuli. With the exception of spatial frequency inappropriate stimuli (low frequencies in the fovea and high frequencies in the periphery), all the stimuli used in this study produce a general (broadband) response that spanned all frequencies from ∼80 Hz up to the noise floor. The broadband responses to face and house stimuli were clearly present in ventral occipital regions that are thought to be important for recognizing these images.

### Gamma Oscillations are Stimulus Specific

The failure to observe narrowband gamma responses to certain classes of stimuli cannot be explained by the experimental setting or signal contamination. In electrodes where grating stimuli generate clear narrowband gamma as well as broadband responses, other types of stimuli in the same visual field positions produce robust single trial broadband responses in the absence of gamma oscillations. It is unlikely that the stimuli that did not elicit gamma rhythms were not perceived, because all stimuli were large (at least 10 × 10°), had an abrupt onset during central fixation, and remained on the screen for at least 500 ms, sufficiently long to allow for conscious perception (Dehaene and Changeux 2011). Moreover, the grating stimuli (which reliably elicited gamma oscillations) and the noise stimuli (which did not) were randomly interleaved and presented at a rapid rate (1 stimulus per second), ruling out the possibility that the subjects' attentional state differed between the stimulus classes. Finally, the behavioral performance during the face and house experiment showed that the subject correctly labeled all stimuli (n = 99 trials) as a face or a house.

A previous report showed that in macaque LFP recordings, narrowband gamma oscillations are observed in response to high contrast gratings but not low contrast gratings (Henrie and Shapley 2005). In our experiments, the noise stimuli and the gratings had the same luminance range, and the binary white noise stimuli and the gratings had the same luminance variance. Hence the absence of gamma oscillations in response to the noise patterns is not a result of low spatial contrast. Some spatial patterns do not elicit gamma oscillations.

The lack of narrowband oscillations to the noise patterns cannot be due to a lack of coherent spatial structure or recognizable content in the images. Images containing faces and houses with moderate white noise superimposed did not lead to consistent gamma band oscillations in the expected locations in visual cortex, despite the fact that the subject attended to and accurately classified the images as faces or houses.

Exactly which stimulus features are necessary to elicit gamma oscillations is unclear. As we show, noise patterns do not have the necessary features. It appears that gratings and bars reliably elicit gamma oscillations in humans (Figs 1–3, 5 and Hoogenboom et al. 2006) as well as animals (Gray and Singer 1989; Gray et al. 1989). Important features of the oscillation such as its peak frequency and amplitude depend on specific stimulus features such as color (Swettenham et al. 2013), contrast (Henrie and Shapley 2005; Ray and Maunsell 2010), spatial area (Bauer et al. 1995; Gieselmann and Thiele 2008; Jia et al. 2011; Ray and Maunsell 2011; Jia Xing et al. 2013), noise masking (Zhou et al. 2008; Jia and Kohn 2011; Jia Tanabe et al. 2013), and the number of orientations (Lima et al. 2010; Bartolo et al. 2011).

A recent ECoG measurement in macaque showed that gamma oscillations were elicited during the free viewing of natural images (Brunet et al. 2015). The gamma response was present for several of the images shown and in the mean response across all images. In a re-analysis of a dataset used for another study (Parvizi et al. 2012; Jacques et al. 2013), we quantified the narrowband gamma response to 72 images of faces, houses, limbs, and cars (see Supplementary Fig. S3). It is likely that our results are consistent with those of Brunet et al. (2015) in one sense: some natural images elicit narrowband gamma oscillations in visual cortex. Consequently, the mean across all stimuli is above zero.

However, many images do not elicit significant gamma oscillations in V1 (32 and 33 of 72 images for 2 V1 electrodes). The lack of gamma oscillations observed for nearly half the stimuli is unlikely to be due to measurement artifacts, as nearly all stimuli elicited a significant broadband response (69 and 68 of 72 images for the same 2 electrodes). Figure 6 shows 2 example images, one that produces a large gamma oscillation and one that does not. These images were viewed 6 times each and the amplitude of the broadband and narrowband responses was reliable across the repetitions, but different between stimuli. All of the images contained recognizable objects or scenes; the fact that many failed to elicit clear gamma oscillations indicates that this response depends on the spatial structure of the image, and is not likely to be necessary for perception, attention, or recognition. We cannot rule out the possibility that for stimuli near detection or recognition threshold, the presence or absence of gamma oscillations will modulate an observer's judgments. However, since very large differences in the gamma response can be observed across different stimuli, it is unlikely that the gamma amplitude is the principal determinant of stimulus detection or recognition.

Figure 6.

Two example stimuli that induced different levels of narrowband gamma power in V1. (A) Power spectra from a foveal V1 electrode for 2 example stimuli (from a set of 72 stimuli, see Supplementary Fig. S3) that were shown to a separate subject in a prior study (Parvizi et al. 2012). Each stimulus was shown 6 times in random order. The 95% confidence intervals are shown in black and the baseline periods are shown in light gray. The baseline is identical for all images. The model fits (power law plus Gaussian) are plotted as a dark gray line. (B) The 2 images with the population receptive field (pRF) from the measured electrode indicated by a white circle (2 SD of the pRF Gaussian). The pRF models were obtained from a prior study (Winawer et al. 2013). The car image on the left elicited only a broadband response but no clear gamma oscillation. The car image on the right elicited both a broadband response and a large gamma oscillation.

Figure 6.

Two example stimuli that induced different levels of narrowband gamma power in V1. (A) Power spectra from a foveal V1 electrode for 2 example stimuli (from a set of 72 stimuli, see Supplementary Fig. S3) that were shown to a separate subject in a prior study (Parvizi et al. 2012). Each stimulus was shown 6 times in random order. The 95% confidence intervals are shown in black and the baseline periods are shown in light gray. The baseline is identical for all images. The model fits (power law plus Gaussian) are plotted as a dark gray line. (B) The 2 images with the population receptive field (pRF) from the measured electrode indicated by a white circle (2 SD of the pRF Gaussian). The pRF models were obtained from a prior study (Winawer et al. 2013). The car image on the left elicited only a broadband response but no clear gamma oscillation. The car image on the right elicited both a broadband response and a large gamma oscillation.

We conclude that narrowband gamma oscillations are stimulus specific. Principles of information transfer that are governed by gamma oscillations, such as increased coherence between sites for attended stimuli (Womelsdorf et al. 2007), may hold for specific stimulus patterns that produce gamma oscillations, but are not likely to hold generally across different types of stimuli (for an example, see Supplementary Fig. S4).

Predicting which stimuli will elicit a narrowband gamma response and predicting the parameters of that response will require a more complete model. There has been some progress modeling the local neural circuits that give rise to gamma oscillations in neocortex. For example, in vitro measurements show that gamma oscillations can be driven by the interactions between principal cells and interneurons (Llinas et al. 1991; Whittington et al. 1995; Cunningham et al. 2004), and computational models have been developed to capture these interactions (Wang 2010). One possibility is that regularized spatial patterns projected to visual cortex might be traded for a regularized “reverberation” timescale in recurrent feedback between principal cell and interneuron synaptic connections, reflected by the frequency of gamma oscillations (20–25 ms for feedback corresponding to a 40–50 Hz oscillation). More work is needed to define the stimulus-driven inputs that set up the conditions to produce these oscillations.

### Broadband Spectral Power Elevation is Observed for All Visual Stimuli Tested

The stimulus features that give rise to the broadband ECoG signals in visual cortex, unlike the features which give rise to gamma oscillations, are qualitatively well-matched to expectations from other functional measurements, such as fMRI and single unit measurements in animals. All stimuli with spatial contrast patterns induced robust broadband signals across V1/V2/V3, with the exception of stimuli confined to very low spatial frequencies that only elicited broadband signals in more peripheral electrodes, and stimuli dominated by high frequencies, which only elicited broadband signals in more foveal electrodes. Furthermore, broadband ECoG responses to faces and houses arise in the same regions known to be activated in fMRI measurements (Puce et al. 1995; Kanwisher et al. 1997; Aguirre et al. 1998; Epstein and Kanwisher 1998).

Although human MEG and EEG recordings sometimes show narrowband oscillations (Hoogenboom et al. 2006; Vidal et al. 2006; Muthukumaraswamy et al. 2009; Scheeringa et al. 2011; van Pelt and Fries 2013), human ECoG measurements typically show an increase in broadband amplitude (for a review see Crone et al. 2011 and for recordings in early visual cortex see Harvey et al. 2013; Winawer et al. 2013). Although these broadband responses often extend into the gamma range, they should not be considered oscillations, as the spectra are not peaked (Ray and Maunsell 2011; Lopes da Silva 2013). A previous human ECoG study reported gamma oscillations in visual cortex (Tallon-Baudry et al. 2005), but from analyses and figures these narrowband responses are often not separable from broadband responses. Because broadband increases can span frequencies in which oscillations are typically observed (30–80 Hz) (Miller, Sorensen et al. 2009; Harvey et al. 2013; Winawer et al. 2013), decomposing the signal into 2 frequency bins (e.g., 50–80 and 100–200 Hz) may not adequately separate the 2 responses. We fit a model to the spectrum to differentiate between these 2 phenomena; other methods have also been developed to separate narrowband and broadband changes in the ECoG power spectrum (Miller, Zanos et al. 2009; Winawer et al. 2013).

### Implications for Theories of Brain Function

Only certain stimuli, such as gratings, reliably produce neural responses that drive resonant (narrowband) gamma oscillations in visual cortex circuits. However, all the stimuli used in this study produce a general (broadband) response. Gamma oscillations and broadband responses do not just differ in the frequency range in which they occur; they also differ in that one has a narrow peak and one does not, reflecting different underlying circuit properties (Manning et al. 2009; Miller, Sorensen et al. 2009; Ray and Maunsell 2011). Understanding the relationship between the stimulus and the 2 responses is likely to clarify fundamental aspects of the circuitry in visual cortex.

For certain stimuli, the power in gamma oscillations is poorly matched to the amplitude of multi-unit activity, in that stimulus manipulations may increase one signal and decrease the other (Bartolo et al. 2011; Ray and Maunsell 2011). Spectrally broadband signals, in contrast, are likely better matched to the total neural activity in nearby regions, reflecting the rate of either input (Miller, Sorensen et al. 2009) or output (Ray and Maunsell 2011) action potentials. Hence when a large spectral peak is observed in a field potential recording, it does not imply that there is a large underlying neural response; it may be a relatively small response that is highly synchronized. In contrast, a large broadband response is more likely to indicate a large neural response. It is likely that perceptual phenomena and other physiological measures such as the BOLD signal will depend in different ways on these circuits.

Broadband and narrowband gamma signals thus indicate different neurophysiological processes. Gamma oscillations may serve a function of information transfer, but only for certain types of visual stimuli that produce a gamma oscillation. Theories about the role of gamma oscillations in visual information processing that do not incorporate stimulus properties are at best incomplete.

## Supplementary Material

Supplementary material can be found at: http://www.cercor.oxfordjournals.org/

## Notes

The authors thank Josef Parvizi for experimental environment, patient recruitment, and collaborative spirit; Vinitha Rangarajan for her help in data collection; Brett Foster for his help in data collection and helpful discussions; Kendrick Kay for helpful discussions; and Corentin Jacques, Kalanit Grill-Spector, and Nathan Witthoft for providing the data for Figure 6 and Supplementary Figure S3. Conflict of Interest: None declared.

## References

Aguirre
GK
Zarahn
E
D'Esposito
M
1998
.
An area within human ventral cortex sensitive to "building" stimuli: evidence and implications
.
Neuron
.
21
:
373
383
.
Allison
T
Ginter
H
McCarthy
G
Nobre
AC
Puce
A
Luby
M
Spencer
DD
1994
.
Face recognition in human extrastriate cortex
.
J Neurophysiol
.
71
:
821
825
.
Bartolo
MJ
Gieselmann
MA
Vuksanovic
V
Hunter
D
Sun
L
Chen
X
Delicato
LS
Thiele
A
2011
.
Stimulus-induced dissociation of neuronal firing rates and local field potential gamma power and its relationship to the resonance blood oxygen level-dependent signal in macaque primary visual cortex
.
Eur J Neurosci
.
34
:
1857
1870
.
Bauer
R
Brosch
M
Eckhorn
R
1995
.
Different rules of spatial summation from beyond the receptive field for spike rates and oscillation amplitudes in cat visual cortex
.
Brain Res
.
669
:
291
297
.
Bosman
CA
Schoffelen
JM
Brunet
N
Oostenveld
R
Bastos
AM
Womelsdorf
T
Rubehn
B
Stieglitz
T
De Weerd
P
Fries
P
2012
.
Attentional stimulus selection through selective synchronization between monkey visual areas
.
Neuron
.
75
:
875
888
.
Brunet
N
Bosman
CA
Roberts
M
Oostenveld
R
Womelsdorf
T
De Weerd
P
Fries
P
2015
.
Visual Cortical gamma-band activity during free viewing of natural images
.
Cereb Cortex
.
25
:
918
926
.
Crone
NE
Korzeniewska
A
Franaszczuk
PJ
2011
.
Cortical gamma responses: searching high and low
.
Int J Psychophysiol
.
79
:
9
15
.
Cunningham
MO
Whittington
MA
Bibbig
A
Roopun
A
LeBeau
FE
Vogt
A
Monyer
H
Buhl
EH
Traub
RD
2004
.
A role for fast rhythmic bursting neurons in cortical gamma oscillations in vitro
.
Proc Natl Acad Sci U S A
.
101
:
7152
7157
.
Dehaene
S
Changeux
JP
2011
.
Experimental and theoretical approaches to conscious processing
.
Neuron
.
70
:
200
227
.
Eckhorn
R
Bauer
R
Jordan
W
Brosch
M
Kruse
W
Munk
M
Reitboeck
HJ
1988
.
Coherent oscillations: a mechanism of feature linking in the visual cortex? Multiple electrode and correlation analyses in the cat
.
Biol Cybern
.
60
:
121
130
.
Epstein
R
Kanwisher
N
1998
.
A cortical representation of the local visual environment
.
Nature
.
392
:
598
601
.
Fries
P
2005
.
A mechanism for cognitive dynamics: neuronal communication through neuronal coherence
.
Trends Cogn Sci
.
9
:
474
480
.
Fries
P
Nikolic
D
Singer
W
2007
.
The gamma cycle
.
Trends Neurosci
.
30
:
309
316
.
Fries
P
Scheeringa
R
Oostenveld
R
2008
.
Finding gamma
.
Neuron
.
58
:
303
305
.
Gavish
M
Donoho
DL
2012
.
Three dream applications of verifiable computational results
.
Comput Sci Eng
.
14
:
26
31
.
Gieselmann
MA
Thiele
A
2008
.
Comparison of spatial integration and surround suppression characteristics in spiking activity and the local field potential in macaque V1
.
Eur J Neurosci
.
28
:
447
459
.
Gray
CM
1999
.
The temporal correlation hypothesis of visual feature integration: still alive and well
.
Neuron
.
24
:
31
47
.
111–125
.
Gray
CM
Konig
P
Engel
AK
Singer
W
1989
.
Oscillatory responses in cat visual cortex exhibit inter-columnar synchronization which reflects global stimulus properties
.
Nature
.
338
:
334
337
.
Gray
CM
Singer
W
1989
.
Stimulus-specific neuronal oscillations in orientation columns of cat visual cortex
.
Proc Natl Acad Sci U S A
.
86
:
1698
1702
.
Harvey
BM
Vansteensel
MJ
Ferrier
CH
Petridou
N
Zuiderbaan
W
Aarnoutse
EJ
Bleichner
MG
Dijkerman
HC
van Zandvoort
MJ
Leijten
FS
et al
2013
.
Frequency specific spatial interactions in human electrocorticography: V1 alpha oscillations reflect surround suppression
.
Neuroimage
.
65
:
424
432
.
Henrie
JA
Shapley
R
2005
.
LFP power spectra in V1 cortex: the graded effect of stimulus contrast
.
J Neurophysiol
.
94
:
479
490
.
Hermes
D
Miller
KJ
Noordmans
HJ
Vansteensel
MJ
Ramsey
NF
2010
.
Automated electrocorticographic electrode localization on individually rendered brain surfaces
.
J Neurosci Methods
.
185
:
293
298
.
Hoogenboom
N
Schoffelen
JM
Oostenveld
R
Parkes
LM
Fries
P
2006
.
Localizing human visual gamma-band activity in frequency, time and space
.
Neuroimage
.
29
:
764
773
.
Jacques
C
Witthoft
N
Weiner
KS
Foster
BL
Miller
KJ
Hermes
D
Parvizi
J
Grill-Spector
K
2013
.
Electrocorticography of category-selectivity in human ventral temporal cortex: spatial organization, responses to single images, and coupling with fMR
.
J Vis
.
13
:
495
.
Jia
X
Kohn
A
2011
.
Gamma rhythms in the brain
.
PLoS Biol
.
9
:
e1001045
.
Jia
X
Smith
MA
Kohn
A
2011
.
Stimulus selectivity and spatial coherence of gamma components of the local field potential
.
J Neurosci
.
31
:
9390
9403
.
Jia
X
Tanabe
S
Kohn
A
2013
.
Gamma and the coordination of spiking activity in early visual cortex
.
Neuron
.
77
:
762
774
.
Jia
X
Xing
D
Kohn
A
2013
.
No consistent relationship between gamma power and peak frequency in macaque primary visual cortex
.
J Neurosci
.
33
:
17
25
.
Kanwisher
N
McDermott
J
Chun
MM
1997
.
The fusiform face area: a module in human extrastriate cortex specialized for face perception
.
J Neurosci
.
17
:
4302
4311
.
Kayser
C
Salazar
RF
Konig
P
2003
.
Responses to natural scenes in cat V1
.
J Neurophysiol
.
90
:
1910
1920
.
LeVeque
RJ
Mitchell
IM
Stodden
V
2012
.
Reproducible Research for Scientific Computing: Tools and Strategies for Changing the Culture
.
Comput Sci Eng
.
14
:
13
17
.
Lima
B
Singer
W
Chen
NH
Neuenschwander
S
2010
.
Synchronization dynamics in response to plaid stimuli in monkey V1
.
Cereb Cortex
.
20
:
1556
1573
.
Llinas
RR
Grace
AA
Yarom
Y
1991
.
In vitro neurons in mammalian cortical layer 4 exhibit intrinsic oscillatory activity in the 10- to 50-Hz frequency range
.
Proc Natl Acad Sci U S A
.
88
:
897
901
.
Lopes da Silva
F
2013
.
EEG and MEG: Relevance to Neuroscience
.
Neuron
.
80
:
1112
1128
.
Manning
JR
Jacobs
J
Fried
I
Kahana
MJ
2009
.
Broadband shifts in local field potential power spectra are correlated with single-neuron spiking in humans
.
J Neurosci
.
29
:
13613
13620
.
Melloni
L
Molina
C
Pena
M
Torres
D
Singer
W
Rodriguez
E
2007
.
Synchronization of neural activity across cortical areas correlates with conscious perception
.
J Neurosci
.
27
:
2858
2865
.
Miller
KJ
Sorensen
LB
Ojemann
JG
den Nijs
M
2009
.
Power-law scaling in the brain surface electric potential
.
PLoS Comput Biol
.
5
:
e1000609
.
Miller
KJ
Zanos
S
Fetz
EE
den Nijs
M
Ojemann
JG
2009
.
Decoupling the cortical power spectrum reveals real-time representation of individual finger movements in humans
.
J Neurosci
.
29
:
3132
3137
.
Mitra
P
Bokil
H
2008
.
Observed brain dynamics
.
New York
:
Oxford University Press
.
Muthukumaraswamy
SD
Edden
RA
Jones
DK
Swettenham
JB
Singh
KD
2009
.
Resting GABA concentration predicts peak gamma frequency and fMRI amplitude in response to visual stimulation in humans
.
Proc Natl Acad Sci U S A
.
106
:
8356
8361
.
Parvizi
J
Jacques
C
Foster
BL
Witthoft
N
Rangarajan
V
Weiner
KS
Grill-Spector
K
2012
.
Electrical stimulation of human fusiform face-selective regions distorts face perception
.
J Neurosci
.
32
:
14915
14920
.
Percival
DB
Walden
A
1993
.
Spectral analysis for physical applications, multitaper and conventional univariate techniques
.
Cambrigde, UK
:
Cambridge UP
.
Puce
A
Allison
T
Gore
JC
McCarthy
G
1995
.
Face-sensitive regions in human extrastriate cortex studied by functional MRI
.
J Neurophysiol
.
74
:
1192
1199
.
Ray
S
Maunsell
JH
2010
.
Differences in gamma frequencies across visual cortex restrict their possible use in computation
.
Neuron
.
67
:
885
896
.
Ray
S
Maunsell
JH
2011
.
Different origins of gamma rhythm and high-gamma activity in macaque visual cortex
.
PLoS Biol
.
9
:
e1000610
.
Roberts
MJ
Lowet
E
Brunet
NM
Ter Wal
M
Tiesinga
P
Fries
P
De Weerd
P
2013
.
Robust gamma coherence between macaque V1 and V2 by dynamic frequency matching
.
Neuron
.
78
:
523
536
.
Roelfsema
PR
Lamme
VA
Spekreijse
H
2004
.
Synchrony and covariation of firing rates in the primary visual cortex during contour grouping
.
Nat Neurosci
.
7
:
982
991
.
Scheeringa
R
Fries
P
KM
Oostenveld
R
Grothe
I
Norris
DG
Hagoort
P
Bastiaansen
MC
2011
.
Neuronal dynamics underlying high- and low-frequency EEG oscillations contribute independently to the human BOLD signal
.
Neuron
.
69
:
572
583
.
MN
Movshon
JA
1999
.
Synchrony unbound: a critical evaluation of the temporal binding hypothesis
.
Neuron
.
24
:
67
77
.
111–125
.
Swettenham
JB
Muthukumaraswamy
SD
Singh
KD
2013
.
BOLD Responses in Human Primary Visual Cortex are Insensitive to Substantial Changes in Neural Activity
.
Front Hum Neurosci
.
7
:
76
.
Tallon-Baudry
C
Bertrand
O
Henaff
MA
Isnard
J
Fischer
C
2005
.
Attention modulates gamma-band oscillations differently in the human lateral occipital cortex and fusiform gyrus
.
Cereb Cortex
.
15
:
654
662
.
Thiele
A
Stoner
G
2003
.
Neuronal synchrony does not correlate with motion coherence in cortical area MT
.
Nature
.
421
:
366
370
.
van Pelt
S
Fries
P
2013
.
Visual stimulus eccentricity affects human gamma peak frequency
.
Neuroimage
.
78
:
439
447
.
Vidal
JR
Chaumon
M
O'Regan
JK
Tallon-Baudry
C
2006
.
Visual grouping and the focusing of attention induce gamma-band oscillations at different frequencies in human magnetoencephalogram signals
.
J Cogn Neurosci
.
18
:
1850
1862
.
Wandell
BA
1995
.
Foundations of vision
.
Sunderland, MA
:
Sinauer Associates
.
Wang
XJ
2010
.
Neurophysiological and computational principles of cortical rhythms in cognition
.
Physiol Rev
.
90
:
1195
1268
.
Welch
P
1967
.
The use of fast Fourier transform for the estimation of power spectra: A method based on time averaging over short, modified periodograms
.
IEEE Trans Audio Electroacoustics
.
15
:
70
73
.
Whittington
MA
Traub
RD
Jefferys
JG
1995
.
Synchronized oscillations in interneuron networks driven by metabotropic glutamate receptor activation
.
Nature
.
373
:
612
615
.
Winawer
J
Kay
KN
Foster
BL
Rauschecker
AM
Parvizi
J
Wandell
BA
2013
.
Asynchronous Broadband Signals Are the Principal Source of the BOLD Response in Human Visual Cortex
.
Curr Biol
.
23
:
1145
1153
.
Womelsdorf
T
Schoffelen
JM
Oostenveld
R
Singer
W
Desimone
R
Engel
AK
Fries
P
2007
.
Modulation of neuronal interactions through neuronal synchronization
.
Science
.
316
:
1609
1612
.
Zhou
Z
Bernard
MR
Bonds
AB
2008
.
Deconstruction of spatial integrity in visual stimulus detected by modulation of synchronized activity in cat visual cortex
.
J Neurosci
.
28
:
3759
3768
.