Content uploaded by Agrita Dubey
Author content
All content in this area was uploaded by Agrita Dubey on Apr 16, 2020
Content may be subject to copyright.
1
Title: Comparison of tuning properties of gamma and high-gamma power in local field
1
potential (LFP) versus electrocorticogram (ECoG) in visual cortex
2
3
Running title: Comparison of gamma and hi-gamma tuning in LFP versus ECoG
4
5
Authors
6
Agrita Dubey1,2 and Supratim Ray1*
7
8
Affiliations
9
1Centre for Neuroscience, Indian Institute of Science, Bangalore, India, 560012
10
Telephone +91 80 2293 3437, Facsimile +91 80 2360 3323
11
2Center for Neural Science, New York University, New York, USA, 10003
12
13
Corresponding author:
14
*Supratim Ray: sray@iisc.ac.in
15
16
Number of Pages: 35
17
Number of Figures: 7 (all color)
18
Keywords ECoG, LFP, Gamma oscillations, High-gamma activity
19
author/funder. All rights reserved. No reuse allowed without permission.
The copyright holder for this preprint (which was not peer-reviewed) is the. https://doi.org/10.1101/803429doi: bioRxiv preprint
2
Abstract
20
Electrocorticogram (ECoG), obtained from macroelectrodes placed on the cortex, is typically
21
used in drug-resistant epilepsy patients, and is increasingly being used to study cognition in
22
humans. These studies often use power in gamma (30-70 Hz) or high-gamma (>80 Hz) ranges
23
to make inferences about neural processing. However, while the stimulus tuning properties of
24
gamma/high-gamma power have been well characterized in local field potential (LFP; obtained
25
from microelectrodes), analogous characterization has not been done for ECoG. Using a hybrid
26
array containing both micro and ECoG electrodes implanted in the primary visual cortex of
27
two female macaques, we compared the stimulus tuning preferences of gamma/high-gamma
28
power in LFP versus ECoG and found them to be surprisingly similar. High-gamma power,
29
thought to index the average firing rate around the electrode, was highest for the smallest
30
stimulus (0.3 radius), and decreased with increasing size in both LFP and ECoG, suggesting
31
local origins of both signals. Further, gamma oscillations were similarly tuned in LFP and
32
ECoG to stimulus orientation, contrast and spatial frequency. This tuning was significantly
33
weaker in electroencephalogram (EEG), suggesting that ECoG is more like LFP than EEG.
34
Overall, our results validate the use of ECoG in clinical and basic cognitive research.
35
author/funder. All rights reserved. No reuse allowed without permission.
The copyright holder for this preprint (which was not peer-reviewed) is the. https://doi.org/10.1101/803429doi: bioRxiv preprint
3
Introduction
36
Electrocorticography (ECoG), also known as intracranial electroencephalography (iEEG), is
37
obtained from macroelectrodes placed subdurally on the pial surface of cortex and is widely
38
used in drug-resistant epilepsy patients. The patients are often monitored for weeks for
39
localization of the seizure focus, allowing (with patient’s consent) researchers to conduct
40
cognitive and neuroscience studies1–9.
41
42
These studies often use power in gamma (30-70 Hz) and high-gamma (>80 Hz) ranges to make
43
inferences about the underlying neural processing10. High-gamma activity (>80 Hz) refers to
44
power over a broad range of frequencies above the gamma band that, in ECoG, is modulated
45
by stimulus presentation as well as the behavioral state4,5,10–13. High-gamma activity is also
46
observed in local field potential (LFP) obtained by inserting microelectrodes in the cortex of
47
animals, where it is tightly correlated with the spiking activity of neurons in the vicinity of the
48
microelectrode13–17.
49
50
Gamma rhythm (30-70 Hz), which is different from high-gamma activity17, has been
51
extensively studied in electroencephalogram (EEG) in humans and LFP in animals, and has
52
been associated with high level cognitive functions such as attention, memory and
53
perception18–24. Further, gamma is known to be strongly induced by stimuli such as
54
bars/gratings and depends on stimulus properties such as size, orientation, spatial frequency,
55
contrast and temporal frequency16,17,25–29. Stimulus dependence of gamma has also been
56
characterized in EEG/MEG studies30–35. However, only a few studies have characterized the
57
stimulus preference of gamma in ECoG30,36. No study, to our knowledge, has done a direct
58
comparison of stimulus preferences of gamma/high-gamma in LFP versus ECoG.
59
60
author/funder. All rights reserved. No reuse allowed without permission.
The copyright holder for this preprint (which was not peer-reviewed) is the. https://doi.org/10.1101/803429doi: bioRxiv preprint
4
Apart from providing clues about the neural correlates of gamma/high-gamma activity in
61
ECoG, such a comparison allows us to determine the spatial spread (the cortical area around
62
the electrode that contributes to the signal that is recorded from that electrode) of ECoG, which
63
we have recently shown to be very local37. For example, both the firing rates and LFP high-
64
gamma power reduce with increasing stimulus size because of larger surround suppression17.
65
However, since a larger stimulus activates a larger cortical area, we might observe an increase
66
in ECoG high-gamma (despite a reduction in firing rate) if ECoG spatial spread is much larger
67
than LFP. Similarly, we have recently shown that gamma power recorded using EEG has much
68
weaker tuning preferences (for stimulus orientation, size and contrast) compared to LFP29. A
69
comparison of analogous gamma tuning preferences for ECoG versus LFP will provide clues
70
about their similarity. Recording from a unique hybrid grid which consists of both micro and
71
macro-electrodes, implanted in the primary visual cortex of the same two female macaques for
72
which we had earlier compared LFP versus EEG tuning29 and LFP versus ECoG spatial
73
spreads37, we compared the strength of ECoG and LFP gamma/high-gamma power for
74
different stimulus properties such as size, orientation, spatial frequency and contrast.
75
author/funder. All rights reserved. No reuse allowed without permission.
The copyright holder for this preprint (which was not peer-reviewed) is the. https://doi.org/10.1101/803429doi: bioRxiv preprint
5
Results
76
We simultaneously recorded LFP and ECoG signals using a special custom-made hybrid grid
77
electrode array implanted in the left primary visual cortex (V1) of two monkeys (Monkeys 3
78
and 4), trained to perform a fixation task, while visual gratings that varied in size, orientation,
79
contrast or spatial frequency were presented on a screen. This hybrid grid consisted of 9 (3x3)
80
ECoG electrodes and 81 (9x9) microelectrodes, both attached to the same connector and
81
referenced to same wire. The microelectrode array was placed between four ECoG electrodes
82
in V1 (see Figure 1 of Ref 37). For the variable stimulus size condition (Figures 1-4), data
83
from two additional monkeys (Monkeys 1 and 2) was used, for which microelectrode and
84
ECoG recordings were conducted separately (see Methods for details). All spectral analyses
85
were performed using the multi-taper method38,39.
86
87
High-gamma activity in ECoG is maximum for a small stimulus size (radius of 0.3)
88
Figure 1A shows the raster plot and multiunit firing rate of an example recording site from
89
Monkey 3 when gratings of six different radii (0.3º, 0.6º, 1.2º, 2.4º, 4.8º and 9.6º) were
90
presented between 0 and 800 ms. The peristimulus histogram averaged across trials is overlaid
91
on each of the raster plots. Consistent with our previous results17, increasing the stimulus size
92
decreased the firing rate. Similar trends were observed for the population dataset of 15, 107,
93
24 and 22 recordings sites from the four monkeys (Figure 1B). Note that the stimulus radii for
94
Monkeys 1 and 2 were different from Monkeys 3 and 4.
95
96
Next, we studied the LFP and ECoG signals for varying stimulus sizes. Figure 2A shows the
97
change in LFP power relative to the baseline period (defined as 500 to 0 ms before stimulus
98
onset) for the same example site as Figure 1A from Monkey 3 for six different sizes. These
99
time-frequency energy difference spectra showed a prominent gamma rhythm (red horizontal
100
author/funder. All rights reserved. No reuse allowed without permission.
The copyright holder for this preprint (which was not peer-reviewed) is the. https://doi.org/10.1101/803429doi: bioRxiv preprint
6
band) at ~50 Hz for stimulus size of 0.6 and above, which appeared after the initial transient
101
and remained present throughout the stimulus duration (up to 0.8 s). Consistent with previous
102
studies16,17,26,29, strength of LFP gamma rhythm increased with an increase in stimulus size
103
while the gamma peak frequency decreased. Further, the smallest stimulus (radius 0.3)
104
showed a prominent increase in power over a broad frequency range above the gamma band.
105
The power in this broadband showed the opposite trend and decreased with an increase in
106
stimulus size. Figure 2B shows the time-frequency difference spectra for an example ECoG
107
electrode from the same monkey. Similar to LFP, the power of ECoG gamma increased with
108
increasing stimulus size. Surprisingly, even though the ECoG electrode was much larger than
109
LFP, the smallest stimulus produced the largest high-gamma power even in ECoG. The
110
increase in ECoG high-gamma power was more prominent up to ~250 Hz, unlike LFP high-
111
gamma that remained prominent up to 400 Hz and beyond. Similar results were obtained from
112
the population average of 24 LFP sites and 5 ECoG sites (Figure 2C and 2D).
113
114
Figure 3 A, C, E and G show mean change in power from the baseline (obtained by subtracting
115
log of baseline power from the log of stimulus power, see Methods for details) across recording
116
sites, as a function of frequency for Monkeys 1, 2, 3 and 4. In all monkeys, the largest stimulus
117
produced the strongest but slowest gamma, visible as a prominent peak at ~45-60 Hz (orange
118
traces). In all monkeys except Monkey 4, a prominent harmonic of gamma was also visible
119
between 80-120 Hz. However, there were interesting differences between Monkeys 1, 2 and
120
Monkeys 3, 4, because much larger stimulus sizes were used for the latter two monkeys. For
121
example, in Monkey 4, a second gamma peak was clearly visible at ~30 Hz for the largest
122
stimulus size, which is the ‘slow’ gamma as described in our previous study29. Also, the LFP
123
gamma in Monkey 4 was weaker than Monkey 3 (this was also observed in our previous
124
study29, in which recordings were done from a different hemisphere using a different array);
125
author/funder. All rights reserved. No reuse allowed without permission.
The copyright holder for this preprint (which was not peer-reviewed) is the. https://doi.org/10.1101/803429doi: bioRxiv preprint
7
we discuss this in more detail in the Discussion. Importantly, in spite of the differences in the
126
strength of gamma and high-gamma band across monkeys, the overall trends remained similar:
127
the strength of gamma rhythm increased with an increase in stimulus size whereas high-gamma
128
power decreased. Importantly, similar trends were also observed in the ECoG signals. To
129
compare the changes in power with stimulus size for LFP and ECoG, we computed the power
130
in two frequency bands: 30-65 Hz for gamma and 150-250 Hz for high-gamma, as shown in
131
Figure 3 B, D, F and H. The gamma range was chosen to avoid the ‘slow’ gamma, while the
132
high-gamma range was chosen to avoid the harmonic of gamma between 80-120 Hz. As
133
observed in PSD plots, the power in gamma band increased with size for both LFP and ECoG
134
(the only exception was the ECoG of Monkey 2 for which only a single electrode was
135
available), whereas high-gamma power showed opposite trends. Interestingly, high-gamma
136
power was maximum for the smallest stimulus (radius of 0.3) for both LFP and ECoG for all
137
the four monkeys. This suggests local origins of ECoG in primary visual cortex, similar to our
138
previous study37, since high-gamma would have been expected to be higher for a larger
139
stimulus if spatial summation occurred over a large cortical area for ECoG. However, unlike
140
our previous approaches37, this approach did not provide a quantitative estimate of the spatial
141
spread. We discuss this in more detail in the Discussion.
142
143
A comparison of the shape of the change in power spectra for LFP (Figure 3A, C, E, G, top
144
row) versus ECoG (bottom row) revealed an interesting difference. Beyond ~100 Hz, the traces
145
were almost parallel to the x-axis in the case of LFP (in all except Monkey 4) but showed a
146
negative slope for ECoG in all monkeys. This suggested that the slope of the PSD in the high-
147
gamma range during stimulus and baseline periods were comparable in case of LFP (such that
148
the difference produced a zero-slope line), but stimulus PSD had a steeper slope than baseline
149
in case of ECoG. Indeed, we have previously observed that while increase in high-gamma
150
author/funder. All rights reserved. No reuse allowed without permission.
The copyright holder for this preprint (which was not peer-reviewed) is the. https://doi.org/10.1101/803429doi: bioRxiv preprint
8
power could be observed up to at least ~400 Hz in LFP40, it was prominent only up to ~150 Hz
151
in human ECoG13. We further quantified this by plotting the slopes of high-gamma range
152
during stimulus period versus baseline (Figure 4). The LFP slopes for stimulus and baseline
153
period were comparable (mean slope during stimulus: 1.31, baseline: 1.22, p=0.15, paired t-
154
test (two sample t-test)), whereas the ECoG slopes for stimulus period were greater than
155
baseline period (mean slope during stimulus: 2.92, baseline: 1.87, p=0.00035).
156
157
Stimulus tuning of gamma oscillations
158
We first compared the orientation tuning (both preferred angle and selectivity; equations 3 and
159
4) between LFP and ECoG, for two reasons. First, while it is well established that different
160
neurons prefer different orientations in V1 such that the distribution of orientation preferences
161
of MUA is more or less uniform41–43, several studies have shown that the stimulus orientation
162
that generates the strongest gamma in microelectrode recordings is remarkably similar across
163
all the recording sites16,29,44. However, since these microelectrode arrays span only ~4x4 mm2
164
patch of cortex, it is possible that different patches of cortex prefer different orientations (the
165
preferred orientation for gamma is location specific, but not monkey specific). Because ECoGs
166
record from brain areas separated by 10 mm or more, comparison of orientation preferences
167
across ECoG sites could provide clues about the specificity of orientation tuning in the gamma
168
band. Second, we have recently shown that the orientation selectivity (measure of the strength
169
of orientation tuning) for gamma was much weaker in EEG compared to LFP29. This could be
170
because EEG records activity from a much larger part of the brain than LFP, and these parts
171
may not be as well tuned for a particular orientation. A comparison of the orientation selectivity
172
of ECoG and LFP could therefore provide clues about their similarity.
173
174
author/funder. All rights reserved. No reuse allowed without permission.
The copyright holder for this preprint (which was not peer-reviewed) is the. https://doi.org/10.1101/803429doi: bioRxiv preprint
9
Figure 5A shows the population average of the change in LFP and ECoG power as a function
175
of frequency, across 77 LFP (top) and 5 ECoG (bottom) recording sites for Monkey 3. The
176
change in power was computed between 250 ms to 750 ms relative to baseline period (0 ms to
177
500 ms before stimulus onset) and then averaged across sites on a log scale. The eight colored
178
traces represent the change in power spectrum for eight stimulus orientations. We observed
179
that the mean LFP gamma between 45 to 70 Hz was strongest and fastest at a stimulus
180
orientation of 90. Surprisingly, mean ECoG gamma showed similar trends as LFP gamma
181
with the strongest and fastest gamma for 90 orientation (Figure 5B, top panel).
182
183
To examine the preferred orientation of gamma at different cortical locations we computed the
184
preferred orientation of gamma in 45 to 70 Hz frequency range for each of the recording sites.
185
Figure 5C shows ECoG (diamonds) and LFP (circles) electrodes, plotted at their receptive field
186
centers and color-coded based on preferred orientation for Monkey 3. Consistent to previous
187
studies16,29,44, we observed that preferred orientation of LFP gamma was similar across sites
188
(Figure 5B, bottom panel, magenta bars). Interestingly, all the five ECoG electrodes which
189
covered ~20 x 20 mm in the cortex, showed a remarkably similar preference for stimulus
190
orientation. Although we observed small variations in preferred orientation from the electrode
191
to electrode, the distribution of ECoG (ranging from 70 to 100) was similar to the LFP
192
(ranging from 80 to 100; Figure 5B, bottom panel). Further, the strength of orientation tuning
193
measured by orientation selectivity was on average comparable for ECoG and LFP (Figure
194
5D). The ECoG electrodes which showed a deviation from 90 had low orientation selectivity
195
values, represented by the smaller marker size in Figure 5C. Similar results were observed for
196
Monkey 4 across 18 LFP and 4 ECoG recording sites (Figure 5E - 5H). Thus, the orientation
197
preference of gamma is monkey specific but not location specific.
198
199
author/funder. All rights reserved. No reuse allowed without permission.
The copyright holder for this preprint (which was not peer-reviewed) is the. https://doi.org/10.1101/803429doi: bioRxiv preprint
10
The orientation preference and selectivity depended on the choice of the frequency band. In
200
particular, for Monkey 3, gamma peak frequency was below our lower cutoff of 45 Hz for
201
some orientations. We used this gamma range to be in congruence with our previous study29,
202
in which we had recorded from the same monkeys but used a microelectrode array implanted
203
in the other hemisphere, and had also collected simultaneous EEG data. Since the orientation
204
preferences for LFPs were similar for the two arrays, having the same frequency range allowed
205
us to better compare the LFP, ECoG and EEG gamma tuning. Further, the low frequency cutoff
206
could not be lowered due to the presence of ‘slow gamma’ (see Ref 29), which peaked between
207
30-35 Hz for the two monkeys. As discussed in more detail later, tuning properties critically
208
depend on the choice of the lower frequency cutoff. Nonetheless, visual inspection of Figures
209
5A and 5E reveals that the gamma peaks were remarkably similar for LFP and ECoG for both
210
monkeys, such that choosing a different frequency range changed the tuning parameters in
211
similar ways.
212
213
Like orientation, gamma tuning of LFP and ECoG were similar for spatial frequency (Figure
214
6) and contrast (Figure 7). In particular, ECoG gamma peak frequency increased with contrast
215
and was similar to LFP peak frequency in both monkeys (for contrasts above 25% that
216
generated salient gamma peaks; Figure 7B, D), unlike EEG gamma peak frequency that did
217
not show a substantial increase with contrast29. Overall, our results suggest that ECoG is more
218
similar to LFP than EEG.
219
author/funder. All rights reserved. No reuse allowed without permission.
The copyright holder for this preprint (which was not peer-reviewed) is the. https://doi.org/10.1101/803429doi: bioRxiv preprint
11
Discussion
220
We compared the stimulus tuning properties of gamma/high-gamma in LFP and ECoG by
221
simultaneously recording these signals using a custom-made hybrid grid and found them to be
222
surprisingly similar. The smallest stimulus size tested (radius of 0.3), which has been earlier
223
shown to produce largest high-gamma power in LFP17, produced the largest high-gamma
224
power in ECoG as well. Further, tuning preferences of gamma oscillations for stimulus size,
225
orientation, spatial frequency and contrast were very similar for LFP and ECoG. Overall, these
226
results suggest that ECoG is an excellent signal to study gamma oscillations.
227
228
These results are consistent with our recent study37, in which we used a receptive field (RF)
229
mapping approach to show that the spatial spread of ECoG was surprisingly local (SD of ~1.5
230
mm or 2SD of ~3mm), not much larger than the diameter of the ECoG electrode (2.3 mm), and
231
only ~3 times the spread of LFP (2SD of ~ 1mm). These results are also consistent with the
232
observation that the RFs of ECoGs recorded in humans are very small3, although in that study
233
the RFs (measured in degrees) were not converted to cortical spreads (measured in mm).
234
235
Unfortunately, this approach did not yield a quantitative estimate of the ECoG spread, for two
236
reasons. First, it is possible that ECoG preferentially samples neurons in the upper layers of the
237
cortex that may prefer smaller stimulus sizes, so it is difficult to deduce spatial spread from
238
size tuning. Second, the range of stimulus sizes that we used was not wide enough to
239
quantitatively compare the spreads of LFP and ECoG. Use of even smaller stimuli (for
240
example, radius of 0.1º) would have yielded a better estimate of the ‘optimal’ stimulus size for
241
LFP high-gamma power, and comparison of optimal stimulus sizes for LFP and ECoG would
242
have yielded a quantitative estimate of their respective spatial spreads. However, when
243
extremely small stimuli are used, appropriate comparison is possible only in the absence of eye
244
author/funder. All rights reserved. No reuse allowed without permission.
The copyright holder for this preprint (which was not peer-reviewed) is the. https://doi.org/10.1101/803429doi: bioRxiv preprint
12
jitters. Given that the monkey had to maintain fixation only within 1º or more around the
245
fixation spot, it is possible that a very small stimulus would occasionally miss the receptive
246
field completely if the monkey’s gaze was away from the fixation spot, increasing the
247
variability of the estimate of high-gamma power for very small stimuli. The method used in
248
our previous study37, which is originally based on the model proposed by Xing and
249
colleagues45, partially addressed this concern because the inflation in the estimate of the RF
250
size due to several factors (including eye jitters) is similar for different measures (MUA, LFP
251
and ECoG), and therefore a model that estimates the spatial spreads based on the differences
252
in RF sizes between measures (such as MUA versus LFP and LFP versus ECoG) can cancel
253
out these common terms (see Refs 37,45 for details). We had also used another approach that
254
involved the comparison of the PSDs of ECoG and LFP during spontaneous periods to show
255
that the ECoG spread was local. The present approach, obtained by simply comparing the high-
256
gamma power as a function of stimulus size, provides a third, albeit weaker line of evidence
257
that ECoG is a local signal. Further, this result is obtained without any model or additional
258
assumptions and is complementary to the previous two approaches that used either very small
259
stimuli to map RFs or compared the PSDs during spontaneous periods.
260
261
What are the origins of high-gamma activity in ECoG? High-gamma activity was initially
262
interpreted in the same conceptual framework as gamma oscillations, just operating at a higher
263
frequency46–48. More recently, high-gamma in the LFP has been shown to be tightly correlated
264
with the multiunit firing rate13–17. ECoG high-gamma power has been proposed to reflect the
265
synchrony in neural population13, although direct experimental evidence, to our knowledge, is
266
lacking. In the size study, we observed that upper range of ECoG high-gamma was limited to
267
200-250 Hz compared to at least 400 Hz in LFP (see Figure 2B vs 2A for stimulus radius of
268
0.3). This was consistent across electrodes (Figure 2D vs 2C) and monkeys (Figure 3A, C, E
269
author/funder. All rights reserved. No reuse allowed without permission.
The copyright holder for this preprint (which was not peer-reviewed) is the. https://doi.org/10.1101/803429doi: bioRxiv preprint
13
and G; bottom vs top panel), and was further quantified by comparing the slopes in stimulus
270
period with baseline period (Figure 4). This could be because the PSD of the ECoG was much
271
steeper than LFP at low frequencies (see Ref 37), and therefore the overall power of the ECoG
272
at high frequencies was much lower than LFP. Thus, the noise (either in the device or the brain)
273
could have affected the ECoG signal more than LFP at high frequencies. It appears that even
274
the LFP for Monkey 4 was more affected by noise, since the PSD slopes in this monkey were
275
shallower during both baseline and stimulus periods compared to other monkeys (Figure 4).
276
The differences in PSD slopes for ECoG compared to LFP could be due to its larger size, lower
277
impedance or position.
278
279
We observed that the tuning preferences of gamma were similar for ECoG and LFP for all the
280
four stimulus manipulations (size, orientation, spatial frequency and contrast), while previously
281
we had observed considerable differences between LFP and EEG tuning29. Note that while
282
these recordings were done on the same monkeys, we did not record all three signals
283
simultaneously because of technical difficulties (see Methods). Nonetheless, the weak tuning
284
of EEG gamma was observed in humans also29, and is therefore likely to be a general feature
285
of EEG signals. However, note that the similarity in tuning profile of LFP and ECoG gamma
286
rhythms for different stimulus manipulations could be because of a coherent network because
287
of the use of full screen gratings at full contrast which are known to produce strong and
288
coherent gamma rhythms16,17 over a large brain area. Both the microelectrodes and
289
macroelectrodes captured the activity of this network and therefore showed similar tuning
290
preferences. Interestingly, ECoG electrodes which were on the surface of cortex captured this
291
activity as reliably as microelectrodes which were presumably in the superficial layers of the
292
cortex. Apart from the stimulus, another factor that could have influenced our results is volume
293
conduction49,50. In a previous study50, in which we recorded from microelectrodes implanted
294
author/funder. All rights reserved. No reuse allowed without permission.
The copyright holder for this preprint (which was not peer-reviewed) is the. https://doi.org/10.1101/803429doi: bioRxiv preprint
14
in Monkeys 1 and 2, we showed that the LFP-LFP phase coherence almost becomes flat for
295
CSD (current source density, a double spatial derivative of potential, obtained by subtracting
296
the potential of an electrode from the potentials of four neighboring electrodes; see Fig 4A of
297
Ref 50). Since, we had only five (Monkey 3) and four (Monkey 4) ECoG electrodes, the CSD
298
analysis could not be performed for ECoG in the current setup.
299
300
As described earlier in Results section, the tuning parameters depended critically on the low
301
frequency limit of the gamma band. This is because the actual power (not change in power
302
which is displayed in the figures) falls off rapidly with frequency and displays a prominent
303
“1/f” structure. The total power in a band is therefore dominated by the lower frequencies that
304
have larger absolute power. For example, in the orientation tuning experiment, gamma peak
305
was strongest for the stimulus orientation of 90º but also the fastest (peak around ~55 Hz) for
306
Monkey 3 (Figure 5A). Orientation of 0º produced a smaller bump, but since it was around 40
307
Hz, the power between 35-40 Hz was more for 0º stimulus than 90º. However, if we had chosen
308
the gamma band between 35-70 Hz, the preferred orientation would have shifted towards 0º
309
just because the absolute power between 35-40 Hz far exceeds the power between 50-60 Hz.
310
This issue can be partially addressed by using the normalized instead of absolute power while
311
computing the power in a band, but in general, it is difficult to compare gamma power across
312
stimulus conditions when the peak frequency itself shifts with stimulus.
313
314
In our case, the choice of frequency band is of less relevance because the actual power spectra
315
for LFP and ECoG were remarkably similar for every stimulus condition: if the gamma peak
316
did not fall in a specified range for LFP, it invariably fell outside the range for ECoG as well.
317
Therefore, our main result that LFP and ECoG gamma tuning is remarkably similar holds
318
irrespective of the choice of the frequency band.
319
author/funder. All rights reserved. No reuse allowed without permission.
The copyright holder for this preprint (which was not peer-reviewed) is the. https://doi.org/10.1101/803429doi: bioRxiv preprint
15
Although the overall trends were similar for Monkeys 3 and 4, the strength of tuning was
320
different. For example, orientation selectivity was different for the two monkeys for LFP
321
gamma whereas ECoG gamma showed comparable selectivity (Figure 5D and 5H). One reason
322
could be because the LFP receptive field locations were very foveal in case of Monkey 4
323
(Figure 5G), although the foveal ECoG electrodes in both the monkeys showed strong
324
orientation tuning (Figure 5C and 5G). Moreover, Xu and colleagues 51 found no difference in
325
orientation selectivity as a function of eccentricity in V1. We suspect that the main reason
326
behind weaker LFP gamma in Monkey 4 is because the microelectrode array had earlier been
327
explanted (see Methods for details), although it is unlikely that this affected any of the major
328
results.
329
To conclude, our findings highlight the presence of gamma oscillations in ECoG which shows
330
similar tuning preference to gamma oscillations observed in LFP recordings, even though the
331
size of the ECoG electrode is several hundred times larger than the microelectrode. Therefore,
332
ECoG gamma can act as a potent marker for the diagnosis of brain disorders such as autism
333
and schizophrenia which have been associated with abnormal gamma rhythms52,53. Further,
334
comparing the high-gamma activity between ECoG and LFP we showed that ECoG has local
335
origins in V1. Together, our results validate the use of ECoG in brain-machine interface
336
applications and basic science research.
337
author/funder. All rights reserved. No reuse allowed without permission.
The copyright holder for this preprint (which was not peer-reviewed) is the. https://doi.org/10.1101/803429doi: bioRxiv preprint
16
Methods
338
Animal preparation and Recording
339
All animal experiments and protocols performed in this study are in strict accordance with the
340
relevant guidelines and regulations approved by the Institutional Animal Care and Use
341
Committee of Harvard Medical School (for Monkeys 1, 2) and Institutional Animal Ethics
342
Committee (IAEC) of the Indian Institute of Science and the Committee for the Purpose of
343
Control and Supervision of Experiments on Animals (CPCSEA) (for Monkeys 3 and 4). The
344
details of our experiment design and data collection have been described in detail in our
345
previous study37; here we explain them briefly. The microelectrode and ECoG data used in this
346
study were collected in two separate set of experiments. The first set was conducted on two
347
male monkeys (Macaca mulatta; 11 and 14 Kg); animal protocols approved by the Institutional
348
Animal Care and Use Committee of Harvard Medical School. For this set of experiments,
349
microelectrode and ECoG recordings were performed separately and are described in detail
350
elsewhere 17,27,54. Briefly, after monkeys learned the behavioral task, a 10x10 microelectrode
351
grid (96 active channels, Blackrock Microsystems) was implanted in the right primary visual
352
cortex (~15 mm anterior to the occipital ridge and ~15 mm lateral to the midline). The
353
microelectrodes were 1 mm long separated by 400 m. After microelectrode recordings, a
354
second surgery was performed to implant the custom-made array having 2 ECoG contacts (2.3
355
mm in diameter and 10 mm apart, Ad-Tech Medical Instrument) on the left primary visual
356
cortex of the same monkeys (see Materials and Methods of Ref 37, for details). One ECoG
357
electrode in Monkey 2 did not show any stimulus evoked response and thus was excluded,
358
yielding 3 ECoG electrodes from these two monkeys. Note that ECoG and microelectrode
359
recordings were non-simultaneous for these two monkeys.
360
361
author/funder. All rights reserved. No reuse allowed without permission.
The copyright holder for this preprint (which was not peer-reviewed) is the. https://doi.org/10.1101/803429doi: bioRxiv preprint
17
The second set of experiments involved simultaneous recordings of spikes, LFP and ECoG
362
signals from two female adult monkeys (Macaca radiata; 3.3 and 4 Kg); animal protocols
363
approved by the Institutional Animal Ethics Committee (IAEC) of the Indian Institute of
364
Science and the Committee for the Purpose of Control and Supervision of Experiments on
365
Animals (CPCSEA). Once the monkey had learned the fixation task, a custom-made hybrid
366
array (see Figure 1 of Ref 37) was implanted in the left cerebral hemispheres. This hybrid array
367
had 3x3 ECoG electrodes (Ad-tech Medical Instrument) and 9x9 microelectrodes, both
368
attached to the same connector made by Blackrock Microsystems. The ECoG electrodes were
369
platinum discs of exposed diameter of 2.3 mm and inter-electrode center- to-center distance of
370
10 mm. The microelectrodes were 1 mm long, 400 m apart. The electrode array was implanted
371
under general anesthesia; first a large craniotomy and a smaller durotomy were performed,
372
subsequent to which the ECoG sheet was inserted subdurally such that the previously made
373
silastic gap between four ECoG electrodes was in alignment with the durotomy (see Ref 37 for
374
details). The microelectrode array was finally inserted into the gap, ~10 – 15 mm from the
375
occipital ridge and ~10-15 mm from the midline. In Monkey 3, out of six ECoG electrodes
376
which were posterior to lunate sulcus, one had noisy receptive field estimate, yielding 5 ECoG
377
electrodes for further analysis. For Monkey 4, the ECoG grid did not slide smoothly on the
378
cortex and one column (electrodes 1-3) had to be removed, yielding 4 ECoG electrodes in V1.
379
Two reference wires, common for both microelectrode and ECoG grid were either inserted
380
near the edge of the craniotomy or wounded over the titanium screws on the metal strap which
381
was used to secure the bone on the craniotomy. Other findings based on data recorded from
382
Monkeys 3 and 4 but from a different microelectrode array (implanted in the right hemisphere)
383
have been reported elsewhere29,55.
384
385
author/funder. All rights reserved. No reuse allowed without permission.
The copyright holder for this preprint (which was not peer-reviewed) is the. https://doi.org/10.1101/803429doi: bioRxiv preprint
18
In case of Monkey 4, we used a hybrid array that had been implanted on a different monkey,
386
but it had to be explanted after 2 days due to complications related to the surgery. One reference
387
wire was lost during the process, and the insulation was removed from the other one (in
388
Monkey 3, insulation from only the tip of the reference wires were removed). This could have
389
led to higher noise in the LFP data collected from Monkey 4 at frequencies above 250 Hz,
390
because the power spectral density appeared to be shallow than other monkeys. It is unlikely
391
that this affected any of the results, since clear gamma rhythm and high-gamma activity were
392
observed in the LFP, which were generally similar to the recordings done earlier using a fresh
393
array implanted in the other hemisphere29. Further, ECoG electrodes that were simply placed
394
on the cortex were unaffected by the explantation and showed strong gamma peaks.
395
396
All signals were recorded using Blackrock Microsystems data acquisition system (Cerebus
397
Neural Signal Processor). Local field potential (LFP) and multi-unit activity (MUA) were
398
recorded from microelectrode array. LFP and ECoG were obtained by band-pass filtering the
399
raw data between 0.3 Hz (Butterworth filter, first order, analog) and 500 Hz (Butterworth filter,
400
fourth order, digital), sampled at 2 kHz and digitized at 16-bit resolution. MUA was derived
401
by filtering the raw signal between 250 Hz (Butterworth filter, fourth order, digital) and 7,500
402
Hz (Butterworth filter, third order, analog), followed by an amplitude threshold (set at ~6.25
403
(Monkey 1), ~4.25 (Monkey 2) and ~5 (Monkeys 3 and 4) of the SDs of the signal).
404
405
The data acquisition system has provisions to measure both the impedance of the electrodes as
406
well as potential cross-talk across pairs of electrodes. The similarity in the gamma oscillations
407
recorded in LFP and ECoG signals was not due to potential crosstalk between LFP and ECoG
408
electrodes, which we could measure explicitly. Further, RF centers for LFP and ECoG
409
electrodes were far apart (Figure 5C and 5G), and small stimuli that covered the RF of only
410
author/funder. All rights reserved. No reuse allowed without permission.
The copyright holder for this preprint (which was not peer-reviewed) is the. https://doi.org/10.1101/803429doi: bioRxiv preprint
19
one signal produced salient gamma oscillations in that signal but virtually no response in the
411
other, ruling out potential cross-talk influencing our results.
412
413
Previously we had also recorded EEG data from Monkeys 3 and 4 simultaneously with the
414
LFP29. In this study, EEG signals were found to be extremely noisy. This was because a much
415
larger craniotomy was needed to insert the ECoG array, and consequently a larger titanium
416
mesh, longer plates and more screws were required to secure the bone flap. Further, as this was
417
the second surgery on these monkeys, there was considerable hardware present on the other
418
hemisphere from the first surgery as well. Consequently, there was hardly enough space to put
419
EEG electrodes on the occipital areas, and those signals were noisy.
420
421
Behavioral task
422
Three separate datasets were used in this study. The first set was used to study the effect of size
423
(‘size study’, Figures 1, 2, 3 and 4) on LFP and ECoG power and were collected from all four
424
monkeys. The second and third data sets were collected from Monkeys 3 and 4 to study the
425
effect of orientation and spatial frequency (‘orientation and spatial frequency study’, Figures 5
426
and 6) and the effect of contrast (‘contrast study’, Figure 7) on LFP and ECoG power. The
427
behavioral task and stimuli used in these studies are described below in detail.
428
429
Size study
430
The data set and results from microelectrode recordings from the first two monkeys have been
431
reported previously17. The experimental design and behavioral task for ECoG recordings were
432
similar. Monkeys 1 and 2 performed an orientation change task, while two achromatic odd-
433
symmetric stimuli were presented synchronously for 400 ms with an inter-stimulus period of
434
600 ms. A Grating stimulus of variable size centered on the receptive field of one of the
435
author/funder. All rights reserved. No reuse allowed without permission.
The copyright holder for this preprint (which was not peer-reviewed) is the. https://doi.org/10.1101/803429doi: bioRxiv preprint
20
recording electrodes (new location for each session) was presented in the left hemifield for
436
microelectrode recordings and right hemifield for ECoG recordings. The monkeys were cued
437
to attend to a low-contrast Gabor stimulus outside of the receptive field (RF) and respond to a
438
change in the orientation of the Gabor stimulus by 90º in one of the presentations. Monkeys
439
responded by making a saccade within 500 ms of the orientation change. The Gratings were a
440
static stimulus with a spatial frequency of 4 cycles/degree (cpd), full contrast, located at the
441
center of the RF of one of the sites (different recording site each session), one of six different
442
orientations (0º, 30º, 60º, 90º, 120º and 150º) and six different radii (0.3º, 0.72º, 1.14º, 1.56º,
443
1.98º and 2.4º), chosen pseudo-randomly. For ECoG recordings in Monkey 2, only five radii
444
were presented (up to 1.98), since the RF center of the ECoG electrode was very fovial
445
(azimuth: 1.16, elevation: 1.83) and the largest stimulus (2.4º) covered the fixation spot. The
446
Gabor stimulus presented outside the RF was also static with an SD of 0.5, spatial frequency
447
4 cpd and an average contrast of ~6% and ~4.3% for Monkeys 1 and 2. Monkeys 1 and 2
448
performed the task in 10 and 24 recording sessions for microelectrode recordings (results
449
presented in Ref 17; and 2 and 1 recordings sessions for ECoG recordings (one session for each
450
ECoG electrode).
451
452
Monkeys 3 and 4 performed the fixation task while they were in a monkey chair, with their
453
head fixed by the headpost. The monkeys were required to hold their gaze within 2 of a small
454
central dot (0.10 diameter) located at the center of a monitor (BenQ XL2411, LCD, 1280x720
455
pixels, 100 Hz refresh rate, gamma corrected) and were rewarded with a juice pulse at the end
456
of the trial upon successful fixation. The stimulus was a Grating with a spatial frequency of 4
457
cpd, full contrast, one of eight different orientations (0º, 22.5º, 45, 67.5, 90º, 112.5º, 135º and
458
157.5º) and six different radii (0.3º, 0.6º, 1.2º, 2.4º, 4.8º and 9.6º), chosen pseudo-randomly,
459
presented for 800 ms with an inter-stimulus period of 700 ms at the RF of one of the recording
460
author/funder. All rights reserved. No reuse allowed without permission.
The copyright holder for this preprint (which was not peer-reviewed) is the. https://doi.org/10.1101/803429doi: bioRxiv preprint
21
sites (different recording site each session). The data were collected in 15 (Monkey 3) and 6
461
(Monkey 4) recording sessions for microelectrode recordings and 5 (Monkey 3) and 4 (Monkey
462
4) recording sessions for ECoG electrodes.
463
464
Only correct trials were used for analysis. For each stimulus size condition, the trials were
465
pooled across orientations to increase the statistical power. The average number of repetitions
466
for each size condition for LFP and ECoG were 182 (range 133 to 288) and 141 (range 129
467
153) for Monkey 1, 145 (range 106 to 196) and 176 (range 173 to 179) for Monkey 2, 79 (range
468
37 to 205) and 150 (range 92 to 189) for Monkey 3, and 91 (range 30 to 127) and 115 (range
469
87 to 153) for Monkey 4.
470
471
Orientation and Spatial frequency tuning study
472
A full-screen static Grating stimulus was presented for 800 ms with an inter-stimulus period of
473
700 ms while Monkeys 3 and 4 performed a fixation task. The Gratings were presented at full
474
contrast at one of five spatial frequencies (0.5, 1, 2, 4 and 8 cpd) and one of the eight
475
orientations (0º, 22.5º, 45, 67.5, 90º, 112.5º, 135º and 157.5º) chosen pseudo-randomly. The
476
effect of orientation was studied (Figure 5) at spatial frequency which produced highest power
477
in gamma range (4 and 2 cpd for Monkeys 3 and 4). The average number of repetitions for
478
each orientation condition and preferred spatial frequency were 33 (range 28 to 36) for Monkey
479
3 and 42 (range 37 to 45) for Monkey 4. Similarly, the effect of spatial frequency was studied
480
(Figure 6) at preferred orientation (~90) which produced highest gamma power. The average
481
number of repetitions were 33 (range 32 to 36) and 34 (range 15 to 45).
482
483
Contrast study
484
author/funder. All rights reserved. No reuse allowed without permission.
The copyright holder for this preprint (which was not peer-reviewed) is the. https://doi.org/10.1101/803429doi: bioRxiv preprint
22
The stimulus for Monkey 3 was a full-screen Grating at preferred spatial frequency (4 cpd),
485
preferred orientation (90), one of seven contrasts (100, 50, 25, 12.5, 6.25, 3.125 and 0%) and
486
one of eight different temporal frequencies (tf = 50, 32, 16, 8, 4, 2, 1and 0 cycle per second;
487
counterphase). We studied (Figure 7) the effect of contrast for the static grating (tf = 0 cps);
488
average number of repetitions was 17 (range 16 to 18). For Monkey 4, stimulus was a static
489
full screen Grating at preferred spatial frequency (2 cpd), one of the six contrasts (100, 50, 25,
490
12.5, 6.25 and 0%) and one of the eight orientations (0º, 22.5º, 45, 67.5, 90º, 112.5º, 135º and
491
157.5º). Contrast tuning was studied at preferred orientation (90º); average number of
492
repetitions was 27 (range 26 to 29). Both monkeys performed a fixation task and stimulus was
493
presented for 800 ms with an inter-stimulus period of 700 ms.
494
495
Electrode selection
496
Receptive fields were mapped by flashing small Gabor stimuli at various positions on the
497
screen, as described in detail in our previous studies37,54. As in our previous studies, only
498
electrodes for which the RF estimates were stable across days (SD less than 0.1) were used
499
for further analysis, yielding 27, 71, 77 and 18 microelectrodes and 2, 1, 5 and 4 ECoG
500
electrodes from Monkeys 1, 2, 3 and 4.
501
502
For the size study, the smallest stimulus was of radius 0.3, covering only a few
503
microelectrodes in the visual field. Therefore, for each recording session, we selected
504
electrodes whose RF centers were within 0.2 of the stimulus center. Since we recorded
505
multiple sessions, the same electrode was counted more than once, yielding 56 (24 unique),
506
141 (66 unique), 62 (40 unique) and 70 (18 unique) electrodes for Monkeys 1-4. Out of this
507
set, we selected electrodes for which the average firing rate was at least 1 spike/s (for an
508
analysis period of 200 to 400 ms for Monkeys 1 and 2 and 250 to 750 ms for Monkeys 3 and
509
author/funder. All rights reserved. No reuse allowed without permission.
The copyright holder for this preprint (which was not peer-reviewed) is the. https://doi.org/10.1101/803429doi: bioRxiv preprint
23
4) for all the stimulus sizes, and a signal-to-noise ratio56 greater than 1.5. This yielded 15 (11
510
unique), 107 (58 unique), 24 (20 unique) and 22 (13 unique) electrodes for further analysis for
511
the four monkeys.
512
513
For the orientation, spatial and contrast studies, full screen stimuli were used because that
514
condition produced the strongest gamma. Consequently, firing rates were weak for most sites29.
515
Since our primary interest was to compare gamma power, we used the full set of 77 (Monkey
516
3) and 18 (Monkey 4) microelectrodes and compared the power with 5 (Monkey 3) and 4
517
(Monkey 4) ECoG electrodes.
518
519
Data analysis
520
All the data were analyzed using custom codes written in MATLAB (The MathWorks,
521
RRID:SCR_001622). Power spectral density (PSD) and the time-frequency spectra were
522
computed using the multi-taper method with three tapers, implemented in Chronux 2.0 (Bokil
523
et al., 2010, RRID:SCR_005547), an open-source, data analysis toolbox available at
524
http://chronux.org. The baseline period was chosen between -200 to 0 ms for Monkeys 1 and
525
2 and -500 to 0 ms for Monkeys 3 and 4, where 0 indicates stimulus onset. Stimulus period
526
was chosen between 200 to 400 ms for Monkeys 1 and 2 and 250 to 750 ms for Monkeys 3 and
527
4 to avoid the stimulus-onset related transients.
528
Time-frequency difference spectra shown in Figure 2 were obtained by first computing the
529
time-frequency power spectra using a moving window of size 250 ms and a step size of 25 ms
530
and then subtracting the baseline power:
531
(1)
532
533
author/funder. All rights reserved. No reuse allowed without permission.
The copyright holder for this preprint (which was not peer-reviewed) is the. https://doi.org/10.1101/803429doi: bioRxiv preprint
24
Where is the mean energy averaged over trials at time t and frequency w, and
534
is the baseline energy computed for 500 ms (-500 to 0 ms before stimulus onset). Since
535
subtraction is done on a log scale, this is essentially the log of the ratio of power at any time
536
and the baseline power and has units of decibel (dB). For population data (Figure 2C and 2D),
537
the values over recording sites were averaged. Note that the baseline energy was
538
calculated across all the stimulus conditions for each recording site.
539
540
For the size study, gamma range was chosen between 30 – 65 Hz for all the four monkeys
541
(Figure 3). This was done to accommodate the peak frequency for all stimulus sizes, as gamma
542
peak frequency decreases with an increase in stimulus size17,28,29. The high-gamma range (150
543
– 250 Hz) was chosen higher than usual (>80 Hz) to avoid the harmonic of gamma rhythm
544
(~100 Hz, see Figure 3). The gamma frequency range for orientation and spatial frequency
545
studies, in which a full-screen Grating was presented, was chosen to be 45 – 70 Hz for Monkeys
546
3 and 4. This was done in congruence with our previous study29 which used data from the same
547
two monkeys (but different hemispheres), and to avoid contamination from ‘slow gamma’29
548
which was prominent in Monkey 4. For the contrast study, gamma range was chosen between
549
20 – 75 Hz. This was done to accommodate peak frequency for all stimulus contrast values,
550
since gamma peak frequency has been to shown to decrease considerably with a reduction in
551
stimulus contrast27.
552
553
Power in gamma and high-gamma ranges were calculated by first averaging the power values
554
obtained from the PSDs in the corresponding frequency ranges, excluding line noise (60 Hz
555
for Monkeys 1, 2 and 50 Hz for Monkeys 3, 4) and their harmonics. Change in power for each
556
stimulus condition was then calculated as follows:
557
558
author/funder. All rights reserved. No reuse allowed without permission.
The copyright holder for this preprint (which was not peer-reviewed) is the. https://doi.org/10.1101/803429doi: bioRxiv preprint
25
(2)
559
560
where STi is the power summed across the frequency range of interest for stimulus condition
561
i, and BLave is the baseline power averaged across conditions ( .
562
563
Preferred orientation and orientation selectivity for each recording site were calculated using
564
the following equations:
565
566
(3)
567
568
(4)
569
570
where and are the orientations and sum of the power in gamma band. N is the total number
571
of orientation values (8).
572
573
The slopes (Figure 4) were calculated for stimulus (200 to 400 ms for Monkeys 1, 2 and 250
574
to 500 ms for Monkeys 3, 4) and baseline (-200 to 0 ms for Monkeys 1, 2 and -500 to 0 ms for
575
Monkeys 3 and 4) periods in high-gamma frequency range (150 – 250 Hz) by fitting the
576
function , where P is the PSD, f is the frequency, c is the constant
577
or noise floor and m is the slope40,57. In this frequency range, the amplifier roll off is negligible,
578
and therefore the slopes are similar with or without amplifier roll-off correction40. We also
579
tested the amplifier noise floor by shorting the inputs and found the power to be at least an
580
order of magnitude lower than the signal power. Therefore, the estimated slopes did not depend
581
on the characteristics of the amplifier.
582
author/funder. All rights reserved. No reuse allowed without permission.
The copyright holder for this preprint (which was not peer-reviewed) is the. https://doi.org/10.1101/803429doi: bioRxiv preprint
26
References
583
1. Engel, A. K., Moll, C. K. E., Fried, I. & Ojemann, G. A. Invasive recordings from the human
584
brain: clinical insights and beyond. Nat. Rev. Neurosci. 6, 35–47 (2005).
585
2. Mukamel, R. et al. Coupling between neuronal firing, field potentials, and FMRI in human
586
auditory cortex. Science 309, 951–954 (2005).
587
3. Yoshor, D., Bosking, W. H., Ghose, G. M. & Maunsell, J. H. R. Receptive Fields in Human
588
Visual Cortex Mapped with Surface Electrodes. Cereb. Cortex 17, 2293–2302 (2007).
589
4. Mukamel, R. & Fried, I. Human Intracranial Recordings and Cognitive Neuroscience. Annu.
590
Rev. Psychol. 63, 511–537 (2011).
591
5. Lachaux, J.-P., Axmacher, N., Mormann, F., Halgren, E. & Crone, N. E. High-frequency
592
neural activity and human cognition: past, present and possible future of intracranial EEG
593
research. Prog. Neurobiol. 98, 279–301 (2012).
594
6. Morshed, B. I. & Khan, A. A Brief Review of Brain Signal Monitoring Technologies for
595
BCI Applications: Challenges and Prospects. J. Bioeng. Biomed. Sci. 4, 1–10 (2014).
596
7. Yang, T., Hakimian, S. & Schwartz, T. H. Intraoperative ElectroCorticoGraphy (ECog):
597
indications, techniques, and utility in epilepsy surgery. Epileptic. Disord. 271–279 (2014).
598
doi:10.1684/epd.2014.0675
599
8. Winawer, J. & Parvizi, J. Linking Electrical Stimulation of Human Primary Visual Cortex,
600
Size of Affected Cortical Area, Neuronal Responses, and Subjective Experience. Neuron
601
92, 1213–1219 (2016).
602
9. Kucewicz, M. T. et al. Dissecting gamma frequency activity during human memory
603
processing. Brain J. Neurol. 140, 1337–1350 (2017).
604
10. Parvizi, J. & Kastner, S. Promises and limitations of human intracranial
605
electroencephalography. Nat. Neurosci. 21, 474–483 (2018).
606
author/funder. All rights reserved. No reuse allowed without permission.
The copyright holder for this preprint (which was not peer-reviewed) is the. https://doi.org/10.1101/803429doi: bioRxiv preprint
27
11. Crone, N. E., Sinai, A. & Korzeniewska, A. High-frequency gamma oscillations and
607
human brain mapping with electrocorticography. Prog. Brain Res. 159, 275–295 (2006).
608
12. Crone, N. E., Korzeniewska, A. & Franaszczuk, P. J. Cortical γ responses: searching high
609
and low. Int. J. Psychophysiol. Off. J. Int. Organ. Psychophysiol. 79, 9–15 (2011).
610
13. Ray, S., Crone, N. E., Niebur, E., Franaszczuk, P. J. & Hsiao, S. S. Neural Correlates of
611
High-Gamma Oscillations (60–200 Hz) in Macaque Local Field Potentials and Their
612
Potential Implications in Electrocorticography. J. Neurosci. 28, 11526–11536 (2008).
613
14. Liu, J. & Newsome, W. T. Local field potential in cortical area MT: stimulus tuning and
614
behavioral correlations. J. Neurosci. Off. J. Soc. Neurosci. 26, 7779–7790 (2006).
615
15. Manning, J. R., Jacobs, J., Fried, I. & Kahana, M. J. Broadband shifts in local field potential
616
power spectra are correlated with single-neuron spiking in humans. J. Neurosci. Off. J. Soc.
617
Neurosci. 29, 13613–13620 (2009).
618
16. Jia, X., Smith, M. A. & Kohn, A. Stimulus selectivity and spatial coherence of gamma
619
components of the local field potential. J. Neurosci. Off. J. Soc. Neurosci. 31, 9390–9403
620
(2011).
621
17. Ray, S. & Maunsell, J. H. R. Different Origins of Gamma Rhythm and High-Gamma
622
Activity in Macaque Visual Cortex. PLoS Biol 9, e1000610 (2011).
623
18. Rodriguez, E. et al. Perception’s shadow: long-distance synchronization of human brain
624
activity. Nature 397, 430–433 (1999).
625
19. Fries, P., Reynolds, J. H., Rorie, A. E. & Desimone, R. Modulation of Oscillatory Neuronal
626
Synchronization by Selective Visual Attention. Science 291, 1560–1563 (2001).
627
20. Pesaran, B., Pezaris, J. S., Sahani, M., Mitra, P. P. & Andersen, R. A. Temporal structure
628
in neuronal activity during working memory in macaque parietal cortex. Nat. Neurosci. 5,
629
805–811 (2002).
630
author/funder. All rights reserved. No reuse allowed without permission.
The copyright holder for this preprint (which was not peer-reviewed) is the. https://doi.org/10.1101/803429doi: bioRxiv preprint
28
21. Womelsdorf, T., Fries, P., Mitra, P. P. & Desimone, R. Gamma-band synchronization in
631
visual cortex predicts speed of change detection. Nature 439, 733–736 (2006).
632
22. Melloni, L. et al. Synchronization of neural activity across cortical areas correlates with
633
conscious perception. J. Neurosci. Off. J. Soc. Neurosci. 27, 2858–2865 (2007).
634
23. Colgin, L. L. et al. Frequency of gamma oscillations routes flow of information in the
635
hippocampus. Nature 462, 353–357 (2009).
636
24. Gregoriou, G. G., Gotts, S. J., Zhou, H. & Desimone, R. High-frequency, long-range
637
coupling between prefrontal and visual cortex during attention. Science 324, 1207–1210
638
(2009).
639
25. Henrie, J. A. & Shapley, R. LFP Power Spectra in V1 Cortex: The Graded Effect of
640
Stimulus Contrast. J. Neurophysiol. 94, 479–490 (2005).
641
26. Gieselmann, M. A. & Thiele, A. Comparison of spatial integration and surround
642
suppression characteristics in spiking activity and the local field potential in macaque V1.
643
Eur. J. Neurosci. 28, 447–459 (2008).
644
27. Ray, S. & Maunsell, J. H. R. Differences in Gamma Frequencies across Visual Cortex
645
Restrict Their Possible Use in Computation. Neuron 67, 885–896 (2010).
646
28. Jia, X., Xing, D. & Kohn, A. No Consistent Relationship between Gamma Power and Peak
647
Frequency in Macaque Primary Visual Cortex. J. Neurosci. 33, 17–25 (2013).
648
29. Murty, D. V. P. S., Shirhatti, V., Ravishankar, P. & Ray, S. Large visual stimuli induce
649
two distinct gamma oscillations in primate visual cortex. J. Neurosci. 2270–17 (2018).
650
doi:10.1523/JNEUROSCI.2270-17.2017
651
30. Rols, G., Tallon-Baudry, C., Girard, P., Bertrand, O. & Bullier, J. Cortical mapping of
652
gamma oscillations in areas V1 and V4 of the macaque monkey. Vis. Neurosci. 18, 527–
653
540 (2001).
654
author/funder. All rights reserved. No reuse allowed without permission.
The copyright holder for this preprint (which was not peer-reviewed) is the. https://doi.org/10.1101/803429doi: bioRxiv preprint
29
31. Adjamian, P. et al. Induced visual illusions and gamma oscillations in human primary
655
visual cortex. Eur. J. Neurosci. 20, 587–592 (2004).
656
32. Hall, S. D. et al. The missing link: analogous human and primate cortical gamma
657
oscillations. NeuroImage 26, 13–17 (2005).
658
33. Swettenham, J. B., Muthukumaraswamy, S. D. & Singh, K. D. Spectral Properties of
659
Induced and Evoked Gamma Oscillations in Human Early Visual Cortex to Moving and
660
Stationary Stimuli. J. Neurophysiol. 102, 1241–1253 (2009).
661
34. Muthukumaraswamy, S. D. & Singh, K. D. Visual gamma oscillations: the effects of
662
stimulus type, visual field coverage and stimulus motion on MEG and EEG recordings.
663
NeuroImage 69, 223–230 (2013).
664
35. Perry, G., Hamandi, K., Brindley, L. M., Muthukumaraswamy, S. D. & Singh, K. D. The
665
properties of induced gamma oscillations in human visual cortex show individual
666
variability in their dependence on stimulus size. NeuroImage 68, 83–92 (2013).
667
36. Hermes, D., Miller, K. J., Wandell, B. A. & Winawer, J. Stimulus Dependence of Gamma
668
Oscillations in Human Visual Cortex. Cereb. Cortex N. Y. N 1991 25, 2951–2959 (2015).
669
37. Dubey, A. & Ray, S. Cortical Electrocorticogram (ECoG) Is a Local Signal. J. Neurosci.
670
39, 4299–4311 (2019).
671
38. Mitra, P. P. & Pesaran, B. Analysis of dynamic brain imaging data. Biophys. J. 76, 691–
672
708 (1999).
673
39. Bokil, H., Andrews, P., Kulkarni, J. E., Mehta, S. & Mitra, P. Chronux: A Platform for
674
Analyzing Neural Signals. J. Neurosci. Methods 192, 146–151 (2010).
675
40. Shirhatti, V., Borthakur, A. & Ray, S. Effect of Reference Scheme on Power and Phase of
676
the Local Field Potential. Neural Comput. 28, 882–913 (2016).
677
41. Hubel, D. H. & Wiesel, T. N. Receptive fields, binocular interaction and functional
678
architecture in the cat’s visual cortex. J. Physiol. 160, 106 (1962).
679
author/funder. All rights reserved. No reuse allowed without permission.
The copyright holder for this preprint (which was not peer-reviewed) is the. https://doi.org/10.1101/803429doi: bioRxiv preprint
30
42. Hubel, D. H. & Wiesel, T. N. Receptive fields and functional architecture of monkey striate
680
cortex. J. Physiol. 195, 215 (1968).
681
43. Hubel, D. H. & Wiesel, T. N. Ferrier Lecture: Functional Architecture of Macaque Monkey
682
Visual Cortex. Proc. R. Soc. Lond. B Biol. Sci. 198, 1–59 (1977).
683
44. Berens, P., Keliris, G. A., Ecker, A. S., Logothetis, N. K. & Tolias, A. S. Comparing the
684
Feature Selectivity of the Gamma-Band of the Local Field Potential and the Underlying
685
Spiking Activity in Primate Visual Cortex. Front. Syst. Neurosci. 2, (2008).
686
45. Xing, D., Yeh, C.-I. & Shapley, R. M. Spatial Spread of the Local Field Potential and its
687
Laminar Variation in Visual Cortex. J. Neurosci. 29, 11540–11549 (2009).
688
46. Canolty, R. T. et al. High gamma power is phase-locked to theta oscillations in human
689
neocortex. Science 313, 1626–1628 (2006).
690
47. Buschman, T. J. & Miller, E. K. Top-down versus bottom-up control of attention in the
691
prefrontal and posterior parietal cortices. Science 315, 1860–1862 (2007).
692
48. He, B. J., Zempel, J. M., Snyder, A. Z. & Raichle, M. E. The temporal structures and
693
functional significance of scale-free brain activity. Neuron 66, 353–369 (2010).
694
49. Pesaran, B. et al. Investigating large-scale brain dynamics using field potential recordings:
695
analysis and interpretation. Nat. Neurosci. 21, 903–919 (2018).
696
50. Srinath, R. & Ray, S. Effect of amplitude correlations on coherence in the local field
697
potential. J. Neurophysiol. 112, 741–751 (2014).
698
51. Xu, X., Anderson, T. J. & Casagrande, V. A. How do functional maps in primary visual
699
cortex vary with eccentricity? J. Comp. Neurol. 501, 741–755 (2007).
700
52. Uhlhaas, P. J. & Singer, W. Abnormal neural oscillations and synchrony in schizophrenia.
701
Nat. Rev. Neurosci. 11, 100–113 (2010).
702
53. Uhlhaas, P. J. & Singer, W. Neuronal dynamics and neuropsychiatric disorders: toward a
703
translational paradigm for dysfunctional large-scale networks. Neuron 75, 963–980 (2012).
704
author/funder. All rights reserved. No reuse allowed without permission.
The copyright holder for this preprint (which was not peer-reviewed) is the. https://doi.org/10.1101/803429doi: bioRxiv preprint
31
54. Dubey, A. & Ray, S. Spatial spread of local field potential is band-pass in the primary
705
visual cortex. J. Neurophysiol. 116, 1986–1999 (2016).
706
55. Shirhatti, V. & Ray, S. Long-wavelength (reddish) hues induce unusually large gamma
707
oscillations in the primate primary visual cortex. Proc. Natl. Acad. Sci. U. S. A. 115, 4489–
708
4494 (2018).
709
56. Kelly, R. C. et al. Comparison of Recordings from Microelectrode Arrays and Single
710
Electrodes in the Visual Cortex. J. Neurosci. 27, 261–264 (2007).
711
57. Miller, K. J., Sorensen, L. B., Ojemann, J. G. & Nijs, M. den. Power-Law Scaling in the
712
Brain Surface Electric Potential. PLOS Comput. Biol. 5, e1000609 (2009).
713
714
author/funder. All rights reserved. No reuse allowed without permission.
The copyright holder for this preprint (which was not peer-reviewed) is the. https://doi.org/10.1101/803429doi: bioRxiv preprint
32
Acknowledgements
715
We thank Dr. John Maunsell for his help in experimental design and data collection from
716
Monkeys 1 and 2 and Steven Sleboda and Vivian Imamura for technical support. We thank
717
Ad-Tech Medical Instrument Corporation and Blackrock Microsystems for the hybrid
718
electrode grid. We also thank Dr. Sebastian Chandu for his assistance in surgeries. This work
719
was supported by Wellcome Trust/DBT India Alliance (500145/Z/09/Z; Intermediate
720
fellowship to SR), Tata Trusts Grant and DBT-IISc Partnership Programme.
721
722
Author contributions
723
A.D. and S.R. conceptualized the study. S.R. collected the data from Monkeys 1 and 2 and
724
A.D. is responsible for data from Monkeys 3 and 4. A.D. analyzed the data and wrote the first
725
draft of the manuscript. A.D. and S.R. were involved in editing of the manuscript.
726
727
Competing interests
728
The authors declare no competing interests.
729
730
Corresponding author
731
Correspondence to Supratim Ray (sray@iisc.ac.in).
732
733
Data availability
734
The datasets analyzed during the current study are available from the corresponding author on
735
reasonable request.
736
737
author/funder. All rights reserved. No reuse allowed without permission.
The copyright holder for this preprint (which was not peer-reviewed) is the. https://doi.org/10.1101/803429doi: bioRxiv preprint
33
Figure Legends
738
739
Figure 1: Spiking activity for different stimulus sizes. (A) Raster plots showing spiking
740
activity in individual trials for each stimulus size for an example unit from Monkey 3. Each
741
row represents a trial. The peristimulus histogram, averaged across trials is overlaid on the
742
raster plots. (B) Averaged firing rates for six stimulus sizes shown as different color traces for
743
Monkeys 1, 2, 3 and 4.
744
745
Figure 2: Gamma oscillations and high-gamma activity as a function of stimulus size in
746
LFP and ECoG for Monkey 3. (A) Time-frequency energy difference plots (in dB) showing
747
the difference in energy relative to baseline energy (-500 to 0 ms, 0 denotes the stimulus onset,
748
stimulus is presented from 0 to 800 ms) for six stimulus radii (labelled above the plots in
749
degrees) for an example LFP recording site (same as shown in Figure 1A). The gamma rhythm
750
at ~50 Hz increases with size, while the high-band activity above the gamma band decreases
751
with size. (B) same as A for an example ECoG recording site. (C–D) show the corresponding
752
population responses of 24 LFP and 5 ECoG recording sites.
753
754
Figure 3: Tuning of gamma oscillations and high-gamma activity for stimulus size. (A, C)
755
Average relative change in power spectra between 200 and 400 ms from baseline energy (-200
756
to 0 ms) for 15 and 107 LFP recordings sites (top panel), 2 and 1 ECoG recording sites (bottom
757
panel) for Monkeys 1 and 2. (E, G) same as A, C but for 24 and 22 LFP recordings sites (top
758
panel), 5 and 4 ECoG recording sites (bottom panel) for Monkeys 3 and 4. The change in power
759
is computed between 250 to 750 ms relative to baseline energy (-500 to 0 ms). (B, D, F and
760
H) Change in LFP (magenta) and ECoG (blue) for gamma (30 – 65 Hz) and high-gamma (150
761
author/funder. All rights reserved. No reuse allowed without permission.
The copyright holder for this preprint (which was not peer-reviewed) is the. https://doi.org/10.1101/803429doi: bioRxiv preprint
34
– 250 Hz) frequency bands as a function of stimulus size. Error bar indicates SEs of the mean.
762
Note that the stimulus radii for Monkeys 1 and 2 are different from Monkeys 3 and 4.
763
764
Figure 4: Slope of the high-gamma activity for 0.3 stimulus. The slope of LFP (magenta)
765
and ECoG (blue) electrodes computed for high-gamma frequency range (150 – 250 Hz) for
766
baseline period is plotted in x-axis and for stimulus period in y-axis. The four monkeys are
767
represented using four different marker types.
768
769
Figure 5: Orientation tuning of gamma oscillations in LFP and ECoG. (A) Average
770
relative change in power spectra between 250 and 750 ms from baseline energy (-500 to 0 ms)
771
for 77 LFP (top panel) and 5 ECoG recording sites (bottom panel) for Monkey 3. Eight colored
772
traces are for eight different orientation values (labelled at the centre of Figure). (B) Average
773
change in gamma power as a function of orientation (top panel) and the histogram of orientation
774
preference (bottom panel) across recording sites for LFP (magenta) and ECoG (blue). Error
775
bar indicates SEs of the mean. (C) Orientation preference of gamma rhythm across LFP (circle)
776
and ECoG (diamond) recording sites plotted at the respective RF centers. The color represents
777
the preferred orientation while the size of the marker represents the strength of tuning. (D)
778
Median orientation selectivity of LFP and ECoG across recording sites. Error bar indicates SEs
779
of the median, computed using bootstrapping. The orange circles are the five ECoG electrodes.
780
(E–H) same as A–D but for 18 LFP and 4 ECoG recording sites in Monkey 4.
781
782
Figure 6: Spatial frequency tuning of gamma oscillations in LFP and ECoG. (A, C) Mean
783
change in power spectra across 77 and 18 LFP recording sites (top panel), 5 and 4 ECoG
784
recording sites (bottom panel) for Monkeys 3 and 4 calculated at stimulus orientationss that
785
induce largest power change in gamma (90 for both monkeys). Five colored traces represent
786
author/funder. All rights reserved. No reuse allowed without permission.
The copyright holder for this preprint (which was not peer-reviewed) is the. https://doi.org/10.1101/803429doi: bioRxiv preprint
35
five different spatial frequency values. (B, D) left panel: Average change in gamma power as
787
a function of spatial frequency for LFP (magenta) and ECoG (blue). right panel: Average
788
gamma peak frequency as a function of spatial frequency. 8 cpd was ignored as the gamma
789
peak was out of the selected frequency range. Error bar indicates SEs of the mean.
790
791
Figure 7: Contrast tuning of gamma oscillations in LFP and ECoG. (A, C) Mean change
792
in power spectra across 77 and 18 LFP recording sites (top panel), 5 and 4 ECoG recording
793
sites (bottom panel) for Monkeys 3 and 4 calculated at stimulus orientations and spatial
794
frequencies that induce largest power change in gamma (90 and 4cpd for Monkey 3 and 90
795
and 2cpd for Monkey 4) . Seven colored traces represent seven different contrast values. Note
796
that for Monkey 4 there are only six traces. (B, D) left panel: Average change in gamma power
797
as a function of contrast for LFP (magenta) and ECoG (blue). right panel: Average gamma
798
peak frequency as a function of contrast. Error bar indicates SEs of the mean.
799
author/funder. All rights reserved. No reuse allowed without permission.
The copyright holder for this preprint (which was not peer-reviewed) is the. https://doi.org/10.1101/803429doi: bioRxiv preprint
A
B
0.60
0.30
0.72
1.14
1.56
1.98
2.40
0.30
0.60
1.20
2.40
4.80
9.60
0.30
0.60
1.20
2.40
4.80
9.60
0.30
0.72
1.14
1.56
1.98
2.40
0.30 1.20 2.40 4.80 9.60
author/funder. All rights reserved. No reuse allowed without permission.
The copyright holder for this preprint (which was not peer-reviewed) is the. https://doi.org/10.1101/803429doi: bioRxiv preprint
9.60
4.80
2.401.200.60
0.30
LFP
Example
ECoG
Example
LFP
Population
ECoG
Population
A
B
C
D
Change in power (dB)
0 0.4 0.8
0
200
400
Time (s)
Frequency (Hz)
author/funder. All rights reserved. No reuse allowed without permission.
The copyright holder for this preprint (which was not peer-reviewed) is the. https://doi.org/10.1101/803429doi: bioRxiv preprint
A
B
C
D
E
F
G
H
ECoG LFP
0.30
0.72
1.14
1.56
1.98
2.40
400 400
0.30
0.72
1.14
1.56
1.98
2.40
0.30
0.60
1.20
2.40
4.80
9.60
400
0.30
0.60
1.20
2.40
4.80
9.60
400
author/funder. All rights reserved. No reuse allowed without permission.
The copyright holder for this preprint (which was not peer-reviewed) is the. https://doi.org/10.1101/803429doi: bioRxiv preprint
