Gamma-band resonance of visual cortex to optogenetic stimulation

Activated visual cortex typically engages in neuronal synchronization in the gamma-frequency band (30-90 Hz). Gamma-band synchronization is related to cognitive functioning, and its mechanisms have been extensively investigated, predominantly through in-vitro studies. To further elucidate its mechanisms in-vivo, we performed simultaneous optogenetic stimulation and electrophysiological recordings of visual cortical areas 17 and 21a in the anesthetized cat. Viral transfection with AAV1 or AAV9 under a CamKIIα promoter led to robust Channelrhodopsin-2 (ChR2) expression. Immunohistochemical analysis showed that all ChR2-expressing neurons were negative for Parvalbumin, consistent with predominant or exclusive expression in excitatory neurons. Optogenetic stimulation used primarily surface illumination directly above the transfected and recorded cells. Stimulation with constant light led to strong and sustained gamma-band synchronization with strength and bandwidth similar to visually induced gamma. Rhythmic stimulation with light-pulse trains or sinusoidal light modulation revealed strongest resonance for gamma-band frequencies. Gamma resonance was confirmed by optogenetic white-noise stimulation. White-noise stimulation allowed the quantification of the transfer function between the optogenetic stimulation and the local field potential response. This transfer function showed a dominant peak in the gamma band. Thus, we find that visual cortical circuits resonate most strongly to gamma-band components in their input. This resonance renders both the sensitivity to input, and the output of these circuits, selectively tuned to gamma. Significance Statement Activated groups of cortical neurons often display rhythmic synchronization in the gamma-frequency band (30-90 Hz). Gamma-band synchronization is particularly well studied in visual cortex. We used optogenetics to control visual cortex neurons with light. Different optogenetic stimulation protocols, using constant light, rhythmically modulated light or white-noise modulated light, all demonstrated that the investigated circuits predominantly resonate to stimulation in the gamma band. The observed gamma-band resonance renders visual cortical circuits most sensitive to gamma-rhythmic synaptic inputs. This in turn renders their spike output and the ensuing interareal synchronization gamma rhythmic. This work was supported by DFG (SPP 1665, FOR 1847, FR2557/5-1-CORNET to P.F.; EXC 1086, DI 1908/5-1, DI 1908/6-1 to I.D.), BMBF (01GQ1301 to I.D.), EU (HEALTH-F2-2008-200728-BrainSynch, FP7-604102-HBP, FP7-600730-Magnetrodes to P.F.; ERC Starting Grant OptoMotorPath to I.D.), a European Young Investigator Award to P.F., the FENS-Kavli Network of Excellence to I.D., National Institutes of Health (1U54MH091657-WU-Minn-Consortium-HCP to P.F.), the LOEWE program (NeFF to P.F. and I.D.). Present address of I.D.: Optophysiology, Bernstein Center and BrainLinks-BrainTools, University of Freiburg, Albertstrase 23, 79104 Freiburg, Germany. Author contributions J.N, C.M.L., T.W., P.F. designed research; J.N, C.M.L., T.W., P.J., I.D., P.F. performed experiments; J.N., C.M.L., T.W. analyzed data; J.N., P.F. wrote the paper.


Abstract 28
Activated visual cortex typically engages in neuronal synchronization in the gamma-29 frequency band (30-90 Hz). Gamma-band synchronization is related to cognitive 30 functioning, and its mechanisms have been extensively investigated, predominantly 31 through in-vitro studies. To further elucidate its mechanisms in-vivo, we performed 32 simultaneous optogenetic stimulation and electrophysiological recordings of visual 33 cortical areas 17 and 21a in the anesthetized cat. Viral transfection with AAV1 or AAV9 34 under a CamKIIα promoter led to robust Channelrhodopsin-2 (ChR2) expression. 35 Immunohistochemical analysis showed that all ChR2-expressing neurons were 36 negative for Parvalbumin, consistent with predominant or exclusive expression in 37 excitatory neurons. Optogenetic stimulation used primarily surface illumination directly 38 above the transfected and recorded cells. Stimulation with constant light led to strong 39 and sustained gamma-band synchronization with strength and bandwidth similar to 40 visually induced gamma. Rhythmic stimulation with light-pulse trains or sinusoidal light 41 modulation revealed strongest resonance for gamma-band frequencies. Gamma 42 resonance was confirmed by optogenetic white-noise stimulation. White-noise 43 stimulation allowed the quantification of the transfer function between the optogenetic 44 stimulation and the local field potential response. This transfer function showed a 45 dominant peak in the gamma band. Thus, we find that visual cortical circuits resonate 46 most strongly to gamma-band components in their input. This resonance renders both 47 the sensitivity to input, and the output of these circuits, selectively tuned to gamma. 48

49
Activated groups of cortical neurons often display rhythmic synchronization in the 50 gamma-frequency band (30-90 Hz). Gamma-band synchronization is particularly well 51 studied in visual cortex. We used optogenetics to control visual cortex neurons with 52 light. Different optogenetic stimulation protocols, using constant light, rhythmically 53 modulated light or white-noise modulated light, all demonstrated that the investigated 54 circuits predominantly resonate to stimulation in the gamma band. The observed 55 gamma-band resonance renders visual cortical circuits most sensitive to gamma-56 rhythmic synaptic inputs. This in turn renders their spike output and the ensuing 57 interareal synchronization gamma rhythmic. 58

Introduction 59
When visual cortex of an awake or lightly anesthetized subject is activated by 60 appropriate stimuli, its neurons typically synchronize their activity in the gamma-61 frequency band, between 30 and 90 Hz (Gray et al., 1989;Kreiter and Singer, 1996; 62 Hoogenboom et al., 2006). Very similar gamma-band synchronization has also been 63 found outside visual cortex, e.g. in somatosensory and auditory cortex (Brosch et  Experimental studies on the mechanisms underlying gamma-band synchronization 76 have partly used in-vivo approaches. For example, intracellular recordings in 77 anesthetized cat visual cortex revealed a type of cell, denoted "chattering cell", that 78 intrinsically generates gamma-rhythmic bursts when depolarized by current injection 79 and that exhibits pronounced oscillations when visually stimulated (Gray and 80 McCormick, 1996). Also, several in-vivo studies demonstrated that excitatory neurons 81 lead inhibitory neurons during the gamma cycle by a few milliseconds (Csicsvari et  We used optogenetics to investigate the resonance properties of visual cortex to 103 provide further insights into mechanisms behind gamma-band synchronization among 104 visual cortical neurons in vivo. We used visual cortex of the lightly anesthetized cat, a 105 classical model system for research on vision and gamma-band synchronization. First, 106 we tried three viral vectors and found that AAV5 fails to provide expression in the cat, 107 whereas both AAV1 and AAV9 lead to robust expression. Constant optogenetic 108 stimulation induced strong and sustained gamma-band activity. Rhythmic stimulation 109 with pulse trains or sine waves at frequencies between 5 and 80 Hz revealed network 110 resonance at 40 Hz or above. To investigate this resonance with greater spectral 111 resolution, we applied optogenetic white noise stimulation, which confirmed resonance 112 with a peak at 40-60 Hz. 113

114
Eight adult domestic cats (felis catus; four females) were used in this study. All 115 procedures complied with the German law for the protection of animals and were 116 approved by the regional authority (Regierungspräsidium Darmstadt). After an initial 117 surgery for the injection of viral vectors and a 4-6 week period for opsin expression, 118 recordings were obtained during a terminal experiment under general anesthesia. 119

Viral vector injection 120
For the injection surgery, anesthesia was induced by intramuscular injection of 121 ketamine (10 mg/kg) and dexmedetomidine (0.02 mg/kg), cats were intubated, and 122 anesthesia was maintained with N2O:O2 (60/40%), isoflurane (~1.5%) and remifentanil 123 (0.3 µg/kg/min). Four cats were injected in area 17 and another four cats in area 21a. with the use of a micromanipulator and under visual inspection to a cortical depth of 131 1 mm below the pia mater. Subsequently, 2 µl of viral vector dispersion was injected 132 at a rate of 150 nl/min. After each injection, the needle was left in place for 10 min 133 before withdrawal, to avoid reflux. Upon completion of injections, the dura opening 134 was covered with silicone foil and a thin layer of silicone gel, the trepanation was filled 135 with dental acrylic, and the scalp was sutured. 136 In one cat, area 17 in the left hemisphere was injected with AAV1-CamKIIα-137 hChR2(H134R)-eYFP (titer 8.97*10 12 GC/ml) and area 17 in the right hemisphere with 138 AAV9-CamKIIα-ChR2-eYFP (titer 1.06*10 13 GC/ml). In two cats, area 17 of the left 139 hemisphere was injected with AAV1-CamKIIα-hChR2(H134R)-eYFP (titer: 140 1.22*10 13 GC/ml). In one cat, area 17 of the left hemisphere was injected with AAV5-141 CamKIIα-ChR2-eYFP (titer 4*10 13 GC/ml coverslipped and imaged with a Zeiss CLSM, using a 25X water immersion objective. 199

Data analysis 200
All data analysis was performed using custom code and the The spike autocorrelation histogram (ACH) was calculated at 1 ms resolution with 212 maximum time lag of 250 ms. Subsequently, the ACH was normalized by the triangle 213 function tri(t) and the MUA rate, such that the ACH is expressed in units of 214 coincidence/spike: 215 The triangle function is defined as 217 , with being the spike train length in seconds, and being the mean spike rate in 219 Hertz. 220 The ACH was smoothed with a Gaussian (SD = 0.5 ms, truncated at ±1.5 SD), and 221 the F1 component of the ACH was calculated and normalized as for the MUA spike 222 train. 223 time series driving the laser and the simultaneously acquired local field potential was 249 determined. The transfer function was estimated by Welch's average periodogram 250 method, separately per recording site and trial. It is the ratio of the cross spectral 251 density between the input (laser) and the output (LFP) time series, and the power 252 spectral density of the input (laser). To determine the white-noise driven resonance 253 spectrum, the magnitude of the transfer function was computed for each recording 254 site. The values from one such estimate demonstrate the transfer function for a single 255 example recording site (Fig. 9D). In order to estimate the average transfer function 256 across all recording sites, these magnitudes were normalized to equalize the total 257 power. The normalized values across all recording sites were averaged to calculate 258 the average spectrum (Fig. 9E). 259 Statistical testing. High-resolution spectra of LFP power changes and MUA-LFP PPC 260 were compared between stimulation with blue light and control stimulation with yellow 261 light (Fig. 5E,F). We calculated paired t-tests between spectra obtained with blue and 262 yellow light, across recording sites. Statistical inference was not based directly on the 263 t-tests (and therefore corresponding assumptions will not limit our inference), but the 264 resulting t-values were merely used as a difference metric for the subsequent cluster-265 based non-parametric permutation test. For each of 10,000 permutations, we did the 266 following: 1) We made a random decision per recording site to either exchange the 267 spectrum obtained with blue light and the spectrum obtained with yellow light or not; 268 2) We performed the t-test; 3) Clusters of adjacent frequencies with significant t-values 269 (p<0.05) were detected, and t-values were summed over all frequencies in the cluster 270 to form the cluster-level test statistic. 4) The maximum and the minimum cluster-level 271 statistic were placed into maximum and minimum randomization distributions, 272 respectively. For the observed data, clusters were derived as for the randomized data. 273 Observed clusters were considered significant if they fell below the 2.5 th percentile of 274 the minimum randomization distribution or above the 97.5 th percentile of the maximum 275 randomization distribution (Maris and Oostenveld, 2007 the cat homologue of primate area V4 (Payne, 1993). 290 AAV5 was injected into area 17 of one cat and this did not result in visible ChR2-eYFP 291 expression ( Figure 1A). This failure of AAV5 expression is consistent with a previous 292 AAV5 transduction study in cat cerebral cortex (Vite et al., 2003). By contrast, AAV1 293 and AAV9 injections into area 17, and AAV9 injections into area 21a resulted in robust 294 ChR2-eYFP expression ( Figure 1B-D). For both, AAV1 and AAV9, fluorescence 295 showed a dependence on cortical depth, being strong in superficial layers, of medium 296 strength in deep layers and relatively weak in middle layers ( Figure 1B,C). Higher 297 magnification revealed labeling of individual neurons ( Figure 1B-D, right panels). 298 As described below, we find that optogenetic stimulation of the transfected neurons 299 leads to network resonance in the gamma-band range. The generation of gamma-300 band activity has been linked to Parvalbumin-positive (PV+) interneurons. We 301 therefore investigated, whether ChR2 was expressed in PV+ neurons. In two cats, we 302 stained histological slices with fluorescence-marked antibodies against parvalbumin 303 ( Fig. 2A-F). One cat had been injected with AAV9-CamKIIα-ChR2-eYFP into 304 area 21a, the other had been injected with AAV1-CamKIIα-hChR2(H134R)-eYFP into 305 area 17. Across several slices and imaging windows of area 21a, we identified 182 306 unequivocally labeled neurons, which showed ChR2-eYFP expression or PV+ anti-307 body staining ( Fig. 2A-C); of those, 73 were PV+ and 109 were ChR2-eYFP neurons, 308 and there was zero overlap between these groups (Fig. 2G). Across several slices and 309 imaging windows of area 17, we identified 282 unequivocally labeled neurons, which 310 showed ChR2-eYFP expression or PV+ anti-body staining ( Fig. 2D-F); of those, 154 311 were PV+ and 128 were ChR2-eYFP neurons, and again there was zero overlap 312 between these groups (Fig. 2G). 313

Neuronal responses to optogenetic stimulation after transfection with AAV1, 314
AAV5 and AAV9 315 Between 4 and 6 weeks after virus injection, we performed terminal experiments under 316 general anesthesia. The injected part of cortex was illuminated with blue light while 317 neuronal spike and LFP activity was recorded from the optogenetically stimulated 318 region. As mentioned above, one injection used AAV5 in area 17 of one cat and failed 319 to show transfected neurons in the later histology. Correspondingly, the recordings in 320 this case also failed to show any neuronal response to light application (Fig. 3A, B; 321 pulses of 18 mW strength and 2 ms duration, applied in a regular 40 Hz pulse train). 322 Firing rates following light pulses (in a window from 2 to 10 ms after light onset) did 323 not differ significantly from rates immediately preceding the pulses (-10 to 0 ms) 324 (Wilcoxon rank-sum test = 2503, p = 0.88, n = 5 sites). This was in stark contrast to 325 responses in a cat injected with AAV1 and AAV9. In one cat, AAV1 was injected into 326 area 17 in the left hemisphere (Fig. 3C,D), and AAV9 was injected into area 17 in the 327 right hemisphere (Fig. 3E,F). Both injections led to strong optogenetic responses. 328 Pulse trains of 20 Hz resulted in strong firing rate enhancements with a clear 20 Hz 329 modulation. The peri-stimulus time histograms showed response latencies after light 330 pulses of 3.9 ms (AAV1, Fig. 3D) and 3.6 ms (AAV9, Fig. 3F). onset. Figure 4B shows the spike responses of this recording site for many interleaved 345 trials of stimulation with blue or yellow light, confirming the selective optogenetic 346 stimulation by blue light. Figure 4C shows the spike-triggered average of the LFP, 347 demonstrating that spikes were locked to the LFP gamma-band component. The time-348 frequency analysis of both, LFP power (Fig. 4D, E) and MUA-LFP locking (Fig. 4F, G) 349 showed a strong and sustained gamma-band peak for stimulation with blue light, that 350 was absent for stimulation with yellow light. optogenetic stimulation also caused a power reduction between 6 and 12 Hz (Fig. 5E  364 left panel for lower frequencies, note the scale is much smaller than for the higher 365 frequencies shown in the right panel; Fig.5B inset). At the same time, it caused a 366 reduction in MUA-LFP locking between 10 and 12 Hz ( Fig. 5F and Fig.5C inset). 367

Neuronal responses to optogenetic pulse train stimulation at different 368 frequencies 369
To characterize the temporal response properties of the network to optogenetic 370 stimulation of the transfected neurons, we applied pulse trains of different frequencies.

371
Pulses always had a duration of 2 ms, and were repeated at frequencies of 5, 10, 20, 372 40, and 80 Hz. Pulse intensity was adjusted per recording site (see Materials and 373 Methods) and was kept constant for a given site across the different pulse train 374 frequencies. The analysis was limited to spike trains and excluded LFP, to avoid LFP 375 artifacts caused by light stimulation. Pulse trains of all employed frequencies resulted 376 in clear increases in firing rate, with strong rhythmicity at the pulse train frequency 377 (Fig. 6). We calculated spike density functions, subtracted the baseline values and 378 averaged them across recordings sites. Figure 6A-C shows those average spike 379 densities for 10 Hz, 40 Hz and 80 Hz. We quantified their rhythmicity by calculating 380 the Fourier transforms at the pulse train frequency (F1, see Materials and Methods). 381 Figure 6D shows The results so far suggest that the stimulated circuits resonate most strongly in the 391 gamma-frequency band. Yet, for a fixed stimulation epoch, higher pulse train 392 frequency imposed higher total light power onto the brain tissue. Therefore, we also responses aligned to the light pulses (Fig. 8A) or to the peaks of the sine wave light 408 stimuli (Fig. 8B). Light pulses caused a small enhancement of MUA starting within 409 approximately 1 ms after light onset and lasting a fraction of a millisecond, which most 410 likely reflects light artifacts. The main response to the pulses followed later, starting at 411 latencies after pulse onset of approximately 3 ms and peaking at 5.4-6.9 ms (Fig. 8C).

412
During sine wave stimulation, the light was modulated between the respective maximal 413 intensity and almost zero intensity. Thus, the light crosses the threshold for effective 414 neuronal stimulation at an unknown intensity, and it is not possible to calculate 415 response latencies in the same way as for the pulse trains. Therefore, we used a 416 technique of latency estimation that has been developed in the study of synchronized 417 oscillations, and that is based on the slope of the phase spectrum of the coherency 418 between two signals (Schoffelen et al., 2005), in our case the light intensity and the 419 MUA. Figure 8D shows this phase spectrum and reveals a strictly linear dependence 420 of phase on frequency. The slope of this linear relationship allows to infer a latency of 421 5.5 ms, in close agreement to the values obtained for the different pulse train 422 frequencies. 423

Neuronal responses to optogenetic white-noise stimulation 424
The described neuronal responses to pulse trains and sine waves of different 425 frequencies suggest that the network responds more strongly to rhythmic stimulation 426 with higher frequencies, potentially with a peak in the gamma-frequency range. It 427 would be interesting to characterize the spectrum of neuronal response to a large 428 number of driving frequencies. Yet, testing neuronal responses to optogenetic 429 stimulation at a sufficiently large number of individual frequencies to fully characterize 430 the spectrum would require excessively long recordings. We therefore employed 431 optogenetic stimulation with light intensities following a Gaussian random process 432 (sampled at ≈1000 Hz) with a flat power spectrum (Fig 9). This white-noise stimulus 433 contains the same energy at all frequencies up to 500 Hz. Recordings obtained during 434 white-noise stimulation allow the estimation of the transfer function, which specifies 435 for each frequency the strength of the neuronal circuit's response given optogenetic 436 stimulation at that frequency. Figure 9A shows an example LFP and MUA recording 437 for an example trial of white-noise stimulation. The time-frequency analyses of the 438 respective LFP power (Fig. 9B) and MUA-LFP locking (Fig. 9C), averaged over trials, 439 showed sustained responses that peaked in the gamma-frequency range. The 440 average transfer function between the white-noise stimuli and the example LFP 441 recording site is shown in Figure 9D and reveals a dominant peak in the gamma band. 442 Figure 9E shows The stimulation with light can cause artifacts in recordings with metal electrodes, as 464 used here. We found artifacts in the LFP that were sizeable, yet constrained to the first 465 few hundred milliseconds after light onset (data not shown). We also found artifacts in 466 some of the MUA recordings, which were always constrained to the first 2 ms after 467 light onset (Fig. 8A). The observation, that sustained optogenetic stimulation induces 468 sustained gamma-band oscillations, is not due to artifacts, because gamma is 469 sustained for the entire duration of optogenetic stimulation, long after the light-onset 470 related artifact has ceased. The analysis of signals recorded with pulsed optical 471 stimulation excluded LFPs, because under these conditions, our LFP recordings 472 contained substantial artifact components, which were difficult to separate from 473 neuronal components. By contrast, the analysis of MUA responses to pulsed light 474 showed light-related artifacts that were small compared to the optogenetically driven 475 neuronal responses (Fig. 8A). Finally, the analysis of signals recorded with white-noise 476 optical stimulation most likely includes some artifacts in the LFP, yet these artifacts 477 cannot explain the band-limited transfer function, because the power spectrum of light 478 stimulation was by construction white. Thus, our finding of resonance in the gamma-frequency band might be partly due to a 489 predominantly superficial localization of our optogenetic stimulation. Whether neuronal 490 circuits in other layers resonate in different ways will require layer-specific expression 491 and/or optical stimulation. 492 We found that visual cortical circuits resonate in the typical gamma-frequency band. 493 Yet, across the different optogenetic stimulation protocols, we found variable peak 494 frequencies. For constant optogenetic stimulation, the average gamma-band peak 495 extended from 50-100 Hz. For rhythmically pulsed stimulation, only a limited set of 496 frequencies was tested and strongest resonance was found at 40 or 80 Hz. For white-497 noise stimulation, the average transfer function peaked between 30 and 60 Hz. This 498 variability in gamma peak frequency is likely due to a combination of factors. First, the 499 gamma peak frequency is partly genetically determined and thereby varies across 500 individuals (van Pelt et al., 2012). The different stimulation protocols analyzed here 501 were applied to different subsets of cats, such that inter-individual differences in 502 gamma peak frequency could lead to apparent differences between stimulation 503 protocols. Second, the gamma peak frequency is likely affected by state changes, that 504 can occur during anesthesia and might resemble changes in attention, which have 505 been shown to modulate gamma peak frequency (Bosman et al., 2012). Third, the 506 peak frequency of visually induced gamma-band activity is strongly affected by visual 507 stimulus parameters with stimuli of higher salience typically leading to higher gamma 508 peak frequency (Fries, 2015). Gamma peak frequency is e.g. reduced for stimuli that stimulation employed in the present study. For example, rhythmically pulsed 512 stimulation might be highly salient, leading to gamma resonance at the upper end of 513 the gamma-frequency range; by contrast, white-noise stimulation might be of lower 514 salience and similar to noisy visual stimulation, leading to lower gamma peak 515 frequencies. 516 The finding that constant optogenetic stimulation of cortex induces sustained gamma-517 band activity is consistent with previous reports. Several in-vitro studies applied slowly 518 ramping optogenetic stimulation to slices of somatosensory cortex or hippocampus 519 and found that this induces strong and narrowband gamma-band activity (Adesnik and 520 Scanziani found that slowly ramping and/or constant optogenetic stimulation induces strong 522 gamma-band activity that is less narrowband than in-vitro. When awake mouse frontal 523 cortex receives sustained activation by means of a step-function opsin, this enhances 524 power in a 50-90 Hz band and reduces power at 4-25 Hz (Yizhar et al., 2011). When 525 awake macaque motor cortex is transfected to express the C1V1 opsin and stimulated 526 with constant or slowly ramping light, it shows gamma-band activity of constant or 527 varying peak frequency (Lu et al., 2015). When anesthetized cat visual cortex is 528 transfected to express ChR2 and stimulated by constant light, it generates sustained 529 gamma-band activity (Ni et al., 2016). This latter study used partly data from the same 530 animals as used here, yet from different recording sessions. In the present study, we 531 confirm that constant optogenetic stimulation of cat visual cortex induces gamma-band 532 synchronization, and we add that it also reduces LFP power in the theta and alpha 533 bands and MUA-LFP locking in the alpha band (Fig. 5E, F). These reductions in low-534 frequency LFP power and MUA-LFP locking are very similar to effects of visual 535 stimulation and selective attention in awake macaque area V4 (Fries et al., 2008). 536 The findings with pulsed stimulation differ from earlier reports. A previous study used 537 mouse Cre-lines to express ChR2 selectively in either PV-positive fast-spiking (FS) 538 neurons or CamKIIα-positive regular-spiking (RS) cells (Cardin et al., 2009). Pulse-539 train stimulation of the PV-FS circuit led to LFP resonance in a 30-70 Hz frequency 540 band. By contrast, pulse-train stimulation of the CamKIIα-RS circuit led to resonance 541 for low frequencies, up to 24 Hz. The cells expressing ChR2 in the CamKIIα-Cre mice 542 were 100% PV negative, consistent with exclusive expression in excitatory neurons. 543 The present study in cat visual cortex cannot build on Cre lines to target expression to 544 specific cell types. Any cell-type specificity of opsin expression is likely due to the 545 employed CamKIIα promoter. Promoters generally control cell-type specific 546 expression less tightly than Cre-driver lines. Nevertheless, the immuno-histochemical The current results suggest that input to (superficial) visual cortical circuits will be most 562 effective if it is gamma rhythmic. If the input is spectrally broad, then resonance will be 563 strongest to the gamma-band components. Local gamma-band synchronization is 564 particularly strong in superficial layers (Buffalo et