Thirty-two or 128 multichannel graphene arrays were built to precisely detect neural signals on the cortical surface in response to auditory and somatosensory stimuli, respectively (Fig. 1a and Figs. S1–2). The incorporation of thru-holes of the device substrate between all electrodes allowed significantly stronger contact with the brain surface and the simultaneous use with penetrating electrodes (Fig. 1b and Fig. S1). Atomically thin graphene, which is composed of sp2-bonded carbon groups, can achieve good bonding and conformal contact with the skin, resulting in low electrical impedance and a high SNR. In particular, four-layer (4 L) graphene was used because it could provide low impedance for better detection of ECoG signals.
a Schematic illustration of the graphene-based electrode array. b Optical image of the device on the cortical surface. c Nyquist plot of 4 L-doped graphene and gold electrodes; measurements were conducted using the three-electrode method with a platinum counter electrode and Ag/AgCl reference electrode in 0.01 M phosphorus buffered saline. The surface area of the electrode is 60 × 60 μm2. A graph and an image of the inset show the magnified plot of 4 L-doped graphene and gold electrodes and an equivalent circuit model of 4 L-doped graphene and gold, respectively. The results of the equivalent circuit model are noted in Table S1. d Histogram showing the impedance of 30 channels comprising the electrode array. Impedance was derived via electrochemical impedance spectroscopy (EIS) at 1 kHz. e Impedance of 4 L-doped graphene electrodes with respect to the bending radius. f Illustrations of rat auditory and somatosensory cortices. Two maps of auditory and somatosensory areas responding to auditory stimuli with various frequency tones and physical stimuli applied to each body part. g Thirty-channel (excluding the reference and ground electrodes) recording of the graphene-electrode array on the cortical surface. The colors of boxes correlate with the geographical location of the rat’s cortical surface.
Figure 1c, d shows a Nyquist plot and an impedance histogram, respectively, of 4 L-doped graphene and gold electrodes with the same electrode design. The calculated results of parameters in the equivalent circuit model were made using EIS measurements in phosphate-buffered saline (Table S1). The impedance of a graphene electrode is approximately 50 times lower than that of a gold electrode. The low impedance can suppress electrical noise, allowing the electrode size to be scaled down and increasing the detectability of neural signals through a high SNR. With these excellent characteristics of graphene multichannel electrodes for measuring neural signals, we studied cortical map plasticity over an area of several square millimeters. To investigate the mechanical stability of the graphene electrode array on a nonplanar surface, we measured the impedance changes of graphene electrodes with respect to the bending radius (Fig. 1e). With a small bending radius (5 mm), the graphene electrodes still maintained their impedance value below ~ 100 kΩ. These results indicated that the graphene electrode array could conduct uniform spatial brain mapping with low noise on a nonplanar cortical surface.
Figure 1f presents illustrations of a rat’s auditory and somatosensory cortices and the shape of the rat’s body, as presented in actual proportions on a somatosensory cortex map. Auditory stimuli with various frequency tones and physical stimuli to body parts, such as the whiskers, trunk, limbs, and paws, highlighted by different colors, induced location-dependent neural responses (evident on both auditory and somatosensory cortex maps). With the graphene array situated on the cortical surface, we recorded stimulus-specific responses on multiple channels (Fig. 1g). Each colored box highlighted in the 30 channel recording results is closely correlated with the geographical location of the somatosensory map of the rat.
We assessed the sensory maps by recording cortical responses to sensory stimuli using either a graphene-surface array or penetrating electrodes. We compared layer-specific cortical maps for two main reasons: first, to elucidate how information changes from the thalamorecipient layer (or layer 4) to the cortical surface, and second, to compare the response quality of graphene-surface recordings to that of penetrating layer 4 recordings.
Pure-tone pips (50 ms, 5-ms cos2 ramps) with 80 different frequencies and eight sound-pressure levels from 0 to 70 dB were delivered by an in-ear speaker. Following a craniotomy above the primary auditory cortex (AI) (Fig. 2a), the graphene array or penetrating electrodes were placed on the cortical surface or advanced to layer 4, respectively (Fig. 2b). We obtained local field potentials (LFPs) in response to different sounds and constructed caudal-rostral frequency maps (or tonotopic maps). The characteristic frequency (CF) was determined using the tip of a V-shaped tuning curve placed at each graphene/penetrating-electrode site, which represents the LFP at the lowest sound intensity. The CF distribution over the AI showed the expected frequency organization.
a Graphene array applied to the cortical surface or layer 4 of the primary auditory cortex. b Tonotopic cortical maps generated in response to various frequencies. The dashed circle represents the putative primary auditory cortex (AI) in graphene-surface (bi) and penetrating layer-4 (bii) recordings. c Receptive fields (RFs), tuning curves, and LFPs are shown in response to 4.1, 8.2, and 16.4 kHz frequencies (ci). Cortical surface and layer-4 maps differ significantly in RF size, tuning-curve size, firing rate, spike amplitude, and BW20 but not in cortical threshold (cii). Data are presented as the mean ± standard error (SEM). **p < 0.01; n.s.p > 0.05.
The overall organization of the surface maps was similar to that of the maps assessed in layer 4. However, differences in RF size, tuning-curve size, sound-evoked firing rate at the CF, bandwidth (at 20 dB above the hearing threshold; BW20), and spike amplitude except cortical-response threshold were observed between surface and penetrating recordings [Fig. 2c; all surface measures were normalized to layer 4 values. RF size: 387 ± 47% (n = 12, F = 23.05, p < 0.0001); tuning curve size: 139.9 ± 3.9% (n = 68, F = 48.43, p < 0.0001); firing rate: 143.6 ± 5.0% (n = 128, F = 18.33, p < 0.0001); spike amplitude: 78.7 ± 1.9% (n = 511, F = 120.0, p < 0.0001); BW20: 132.8 ± 3.3% (n = 68, F = 60.27, p < 0.0001); and cortical threshold: 97.1 ± 0.6% (n = 68, F = 3.17, p = 0.077)]. Moreover, no frequency dependence was observed for the altered responses across the tonotopic regions (Fig. S3).
The differences between surface and layer-4 recordings in the somatosensory cortex of the rats were similar to those in their primary AI. We also obtained recordings from superficial layer 1/2 with penetrating electrodes in the somatosensory cortex. Somatosensory surface maps with graphene electrodes were compared with cortical layer 1/2 and 4 maps assessed with penetrating electrodes. LFPs for each sensory stimulus to the whisker, forepaw, forelimb, hind paw, hindlimb, trunk, etc., were located in the somatosensory cortex. Somatosensory maps were constructed using the LFP amplitude to produce a rodent homunculus (Fig. 3a).
a Location of neural recordings across cortical layers (left). Cortical surface maps determined using graphene electrodes were compared to cortical layer 1/2, and four maps were produced using penetrating electrodes (middle). Color-coded somatosensory maps were constructed using the response amplitudes and overlapped with the homunculus (right). b Cortical responses and maps were recorded following the stimulation of each body part. Notably, the response amplitude and area stimulated by various somatosensory stimuli gradually decreased and expanded from layer 4 to layer 1/2 and to the cortical surface, respectively. c Response amplitude and area across cortical layers were quantified by the size and location of LFPs. The area was normalized to the area recorded using graphene electrodes because of the different geographical locations of each somatosensory part. The response amplitude and area gradually decreased and expanded, respectively, as the electrodes were moved from cortical layer 4 to superficial layer 1/2 and to the cortical surface. Cortical-surface recordings obtained using graphene arrays show weaker responses and less stimulus selectivity than those obtained with penetrating electrodes. Data are presented as the mean ± standard error (SEM). **p < 0.01; n.s.p > 0.05.
We did not detect differences in the topographic organization of stimulus selectivity between the recording methodologies. However, the cortical-surface recordings showed weaker responses and a more expanded RF area than those obtained with penetrating recordings in the superficial layer 1/2. These differences exceed the penetrating recordings in layer 4. [In Fig. 3b, c, all the measures were normalized to layer-4 values. Spike amplitude: layer 4 (n = 153) vs. layer 1/2 (n = 215), 68.0 ± 0.97%; layer 1/2 vs. surface (n = 1098), 14.4 ± 0.17%; layer 4 vs. surface; F = 1265.5; p < 0.0001 for all comparisons. RF size: layer 4 vs. layer 1/2, 250.7 ± 44.0%; layer 1/2 vs. surface, 363.5 ± 66.8%; layer 4 vs. surface; F = 1265.5, n = 6, p < 0.0001 for all comparisons.] The decreased selectivity from layer 4 to the superficial cortex was likely due to the distinct hierarchical structures of neural circuits from layer 4 to the surface, rather than the different recording materials, as responses in penetrating layer 1/2 recordings became weaker and less selective when compared with those in penetrating layer 4 recordings.
Next, we examined whether graphene-based cortical surface stimulation could induce sensory map reorganization. Previous studies have shown that theta burst stimulation (TBS), a physiologically relevant high-frequency stimulation, is effective in inducing neuroplasticity and has therapeutic applications24,25,26. The cortical-surface and layer 4 LFPs in response to sound stimuli were simultaneously monitored following the TBSs (Fig. 4a). Five successive TBSs spaced 10 min apart induced significant enhancements in RF size, tuning curve size, firing rate, spike amplitude, spike duration, and BW20 in both cortical-surface and layer-4 recordings without any observable frequency dependence of the altered responses (Fig. S4). [Fig. 4b; values were normalized to pre-TBS values. RF size: surface, 1558 ± 263% (n = 13, F = 23.96, p < 0.0001); tuning-curve size: layer 4, 148.7 ± 9.8% (n = 44, F = 13.37, p = 0.0004) and surface, 204.3 ± 5.3% (n = 448, F = 126.0, p < 0.0001); firing rate: layer 4, 204.2 ± 13.8% (n = 44, F = 15.72, p = 0.0002) and surface, 198.5 ± 7.9% (n = 446, F = 91.14, p < 0.0001); spike amplitude: layer 4, 134.3 ± 4.9% (n = 103, F = 51.36, p < 0.0001) and surface, 222.8 ± 19.9% (n = 61, F = 47.56, p < 0.0001); BW20: layer 4, 160.2 ± 21.7% (n = 14, F = 3.59, p = 0.06) and surface, 154.1 ± 5.7% (n = 38, F = 27.39, p < 0.0001)].
a Illustration of cortical stimulation at both the cortical surface and layer 4 using a graphene array. b Surface and layer-4 LFPs induced by auditory stimuli were simultaneously monitored before and after TBS at the cortical surface (bi). Cortical surface stimulation increased the RF size, tuning curve size, firing rate, spike amplitude, spike duration, and bandwidth (bii). c Using auditory brain slices, neural responses at the cortical surface were monitored using a stimulating electrode at cortical layer 4. TBSs were applied at two different locations: one stimulus (TBS1) at the cortical surface and the other (TBS2) at cortical layer 4. Although both TBS1 and TBS2 induced neural enhancement (bottom), the enhancement with TBS1 was larger than that with TBS2 (upper right). Data are presented as the mean ± standard error (SEM). *p < 0.05; **p < 0.01.
Furthermore, we examined whether Hebbian synaptic plasticity was involved in the observed map reorganization. In sectioned brains, neural responses were monitored at the cortical surface following electrical stimulation at cortical layer 4. Once stable responses were detected, after 20 min, four TBSs were applied at two different locations: one stimulus (TBS1) at the cortical surface and the other (TBS2) at cortical layer 4. These stimuli mimicked cortical-surface stimulation and sensory-driven input, respectively (Fig. 4c). Both TBS1 and TBS2 induced neural enhancement [TBS1, 121.7 ± 6.7% (n = 6, t = 3.84, p = 0.012); TBS2, 111.5 ± 1.9% (n = 6, t = 9.17, p = 0.0003)]. Interestingly, TBS1 produced a significantly larger enhancement (F = 5.14, p = 0.046). This outcome indicated that, compared with deep stimulation, cortical-surface stimulation rendered cortical responses more susceptible to changes.
Cortical-surface stimulation-enhanced responses were similarly observed in the somatosensory cortex (Fig. 5). The LFP responses to somatosensory stimuli were monitored before and after the TBS. For example, forepaw stimulation-induced LFP deflections at several graphene-recording spots (Fig. 5a). The response amplitude resulting from forepaw stimulation was significantly enhanced up to 60 min following successive TBSs, while the response amplitude without the TBS remained constant. [Fig. 5b, before TBS vs. after TBS, n = 19, (20 min) 155.8 ± 17.1%, (40 min) 164.8 ± 15.1%, (60 min) 202.1 ± 20.5%, paired t test, p < 0.05, and with TBS vs. without TBS, Fisher’s PLSD, F > 23.27, p < 0.0013 in each comparison.] It is notable that the RF size was not studied because of the limited size of our graphene array. These results indicated that cortical surface stimulation induces remodeling of sensory maps along the columnar network.
a In the somatosensory cortex, local field potential responses to sensory stimuli were monitored before and after TBS. The amplitude of neural responses was significantly enhanced when forepaw stimulation was paired with TBS, whereas response amplitudes without TBS remained uniform over time. b Response amplitude resulting from forepaw stimulation was significantly enhanced up to 60 min following successive TBSs, whereas response amplitudes without TBS remained constant. Data are presented as the mean ± standard error (SEM); **p < 0.01.
