This is the accepted version of the article: Masvidal-Codina E., Illa X., Dasilva M., Calia A.B., Dragojevic T., Vidal-Rosas E.E., Prats-Alfonso E., Martínez-Aguilar J., De la Cruz J.M., Garcia-Cortadella R., Godignon P., Rius G., Camassa A., Del Corro E., Bousquet J., Hébert C., Durduran T., Villa . High-resolution mapping of infraslow cortical brain activity enabled by graphene microtransistors. Nature Materials, (2019). 18. : 280 - . 10.1038/s41563-018-0249-4. Available at: https://dx.doi.org/10.1038/s41563-018-0249-4 High-resolution mapping of infraslow cortical brain activity enabled by graphene microtransistors Eduard Masvidal-Codina1, Xavi Illa2,1, Miguel Dasilva3, Andrea Bonaccini Calia4, Tanja Dragojević5, Ernesto E. Vidal-Rosas5, Elisabet Prats-Alfonso2,1, Javier Martínez-Aguilar1,2, Jose M. De la Cruz Sanchez4, Ramon Garcia-Cortadella4, Philippe Godignon1, Gemma Rius1, Alessandra Camassa3, Elena Del Corro4, Jessica Bousquet4, Clement Hébert4, Turgut Durduran5,6, Rosa Villa1,2, Maria V Sanchez-Vives3,6, Jose A Garrido4,6* and Anton Guimerà-Brunet1,2* 1 Institut de Microelectrònica de Barcelona, IMB-CNM (CSIC), Esfera UAB, Bellaterra, Spain 2 Centro de Investigación Biomédica en Red en Bioingeniería, Biomateriales y Nanomedicina (CIBER-BBN), Madrid, Spain 3 Institut d’Investigacions Biomèdiques August Pi i Sunyer (IDIBAPS), Barcelona, Spain 4 Catalan Institute of Nanoscience and Nanotechnology (ICN2), CSIC and The Barcelona Institute of Science and Technology (BIST), Campus UAB, Bellaterra, Barcelona, Spain 5 ICFO-Institut de Ciéncies Fotòniques, The Barcelona Institute of Science and Technology, Castelldefels, Barcelona, Spain 6 Institució Catalana de Recerca i Estudis Avançats (ICREA), Barcelona, Spain * Correspondence: anton.guimera@imb-cnm.csic.es; joseantonio.garrido@icn2.cat Keywords: cortical spreading depression, DC-ECoG, graphene, infraslow. 1 Abstract 2 Recording infraslow brain signals (< 0.1 Hz) with microelectrodes is severely hampered by current 3 microelectrode materials, primarily due to limitations resulting from voltage drift and high electrode 4 impedance. Hence, most recording systems include high-pass filters that solve saturation issues but 5 come in hand with loss of physiological and pathological information. In this work, we use flexible 6 epicortical and intracortical arrays of graphene solution-gated field-effect transistors (gSGFETs) to 7 map cortical spreading depression in rats and demonstrate that gSGFETs are able to record, with 8 high-fidelity, infraslow signals together with signals in the typical local field potential bandwidth. 9 The wide recording bandwidth results from the direct field-effect coupling of the active transistor, in 10 contrast to standard passive electrodes, as well as from the electrochemical inertness of graphene. 11 Taking advantage of such functionality, we envision broad applications of gSGFET technology for 12 monitoring infraslow brain activity both in research and in the clinic. 2 13 Recently, there has been a particular resurgence of interest in fluctuations of brain activity occurring 14 at < 0.1 Hz, commonly referred to as very slow, ultraslow or infraslow activity (ISA)1. ISA is 15 suggested to have a unique neurophysiological basis2, and to be indicative of brain states (e.g. sleep, 16 anesthesia, coma, wakefulness)2-4. ISA is also correlated with resting-state networks in functional 17 magnetic resonance imaging5 and may significantly contribute to the high variability observed in the 18 time course of physiological signals6,7. Interestingly, cortical spreading depression (CSD)8,9, a slowly 19 propagating wave of near-complete depolarization of neurons and astrocytes followed by a period of 20 electrical activity suppression, occurs at infralow frequencies. CSD is often triggered in individuals 21 suffering stroke or brain injury as well as migraines and recent research has shown that CSD plays a 22 significant role in brain pathophysiology10-12. For this reason, monitoring electrophysiological signals 23 below 0.1 Hz can be very valuable for clinical diagnosis, prognosis and therapy in neurocritical 24 care13-15. 25 Non-invasive techniques such as electroencephalography (EEG) and magnetoencephalography 26 (MEG) have been used to study ISA16,17. However, their limited spatial resolution, and averaged 27 signal impose serious limitations, e.g. scalp EEG alone is not sufficient for CSD detection14,18. 28 Hence, invasive electrophysiological techniques are the most widely used to record infraslow 29 brainwaves. The proper recording of ISA requires the use of direct-coupled amplifiers and extremely 30 stable and low-impedance invasive electrodes. Traditionally, liquid-filled glass micropipettes are 31 used, which allow only one or few-point measurements19 and therefore impose serious mapping 32 limitations. For higher spatial resolution and mapping, non-polarizable silver/silver chloride 33 (Ag/AgCl) electrodes could be used, which prevent charge accumulation at the interface and 34 therefore voltage drift. However, due to the toxicity of silver, the use of such electrodes for human or 35 chronic animal in vivo monitoring is not an option20. This has fostered the search for alternative 36 microelectrode materials with low impedance and drift, although none has yet been found capable of 37 offering performance comparable to Ag/AgCl electrodes21. Current ISA recordings in humans are 38 performed with platinum electrodes, which challenge CSD detection due to artefacts and transients13. 39 Moreover, miniaturization of electrode size to achieve higher spatial resolution may cause intrinsic 40 high-pass filtering of ISA due to the associated electrode impedance increase22,23. Other invasive 41 optical techniques, such as calcium imaging are also used to monitor ISA, but still nowadays they 42 present serious challenges in resolving high-frequency activity for a large number of neurons24,25 and 43 their intrinsic need of indicators limits the translation to clinical use. Consequently, there is a 44 pressing need for a technique that allows measuring large-scale, high-spatiotemporal resolution 45 electrophysiological signals including infralow frequencies in a potentially fully implantable, 46 nontoxic, clinical-scale system. (Table S1). 47 As an alternative to the commonly used microelectrode technology, recording electrophysiological 48 signals with field-effect transistors (FETs) offers several advantages: they are less sensitive to 49 environmental noise thanks to their intrinsic voltage-to-current amplification, and they can be easily 50 multiplexed26. Nonetheless, the difficulties in combining high gate capacitance and carrier mobility 51 silicon FETs with flexible materials has historically hampered their use for in vivo recordings27. 52 Graphene solution-gated field-effect transistors (gSGFETs) have been proposed to potentially 53 overcome most previous drawbacks28. Graphene flexibility allows gSGFETs to be embedded in ultra- 3 54 soft and flexible substrates without loss of performance29, while its wide electrochemical window and 55 biocompatibility allows direct contact with biological fluids and tissues and ensures a safe operation 56 in in vivo conditions30. In addition, the two-dimensional nature of graphene provides the highest 57 surface-to-volume ratio possible, making graphene very sensitive to charges at its surface. Taking 58 advantage of the above-mentioned properties, in previous works, we demonstrated that gSGFETs are 59 able to record local field potentials31,32. 60 In this work, we investigate the potential of graphene microtransistors to record infraslow brain 61 activity by performing in vivo recordings where we use, gSGFETs for both epicortical and 62 intracortical mapping of cortical spreading depression. We found that graphene microtransistors are 63 excellent devices for recording infraslow signals, performing similarly to solution-filled glass 64 micropipettes while additionally offering the possibility of performing spatially-resolved mapping. 65 Importantly, gSGFETs do not compromise the acquisition of signals in the conventional local field 66 potential bandwidth, therefore allowing recording in a wide frequency bandwidth. Furthermore, we 67 also demonstrate that gSGFET technology can be used in combination with optical techniques, such 68 as laser speckle contrast imaging, to obtain 2-D maps of neurovascular coupling. 69 Fabrication and characterization of gSGFET arrays 70 A gSGFET is a device in which graphene is used as channel material, contacted by two metal leads 71 (source and drain terminals), in direct contact with an electrolyte solution or conductive biological 72 tissue where a reference electrode is placed and used as gate terminal (Fig. 1a). We fabricated 12 µm- 73 thick flexible probes containing arrays of gSGFETs in both epicortical and intracortical designs (Fig. 74 1b) using the process previously reported32. The arrays were placed in zero insertion force connectors 75 for interfacing with recording electronics (Fig. 1c). Characterization consisted in measuring the 76 transfer curve, drain current (𝐼𝑑𝑠 ) vs gate-source voltage (𝑉𝑔𝑠 ), of all gSGFETs in each array with a 77 fixed drain-source voltage (𝑉𝑑𝑠 ). The small dispersion of the charge neutrality point obtained 78 (CNP=243.6 ± 6.1 mV), which is defined as the 𝑉𝑔𝑠 voltage associated to the minimum current value 79 of the transfer curve, indicates the homogeneity of the transistors (Fig. 1d). Importantly, since 𝑉𝑔𝑠 and 80 𝑉𝑑𝑠 are shared, a small CNP dispersion allows near-optimal recording performance for all gSGFETs 81 in the same array. The leakage current (𝐼𝑔𝑠 ) for all gSGFETs in the array was also measured, being in 82 the nA range throughout the voltage sweep (Fig. 1e), demonstrating the good insulation of the 83 passivation layer and the negligible reactivity of the graphene. Furthermore, we measured the 84 frequency response of the transconductance (gm), which indicates the efficiency of the signal 85 coupling (𝜕𝐼𝑑𝑠 ⁄𝜕𝑉𝑔𝑠 ). The negative gm for Vgs values lower than the CNP results in an inversion 86 (180º phase) of the signals measured at such bias while for Vgs values higher than the CNP the signal 87 phase is preserved. In both cases, we obtained constant gm values in a wide bandwidth (Fig. 1f-g). 4 88 89 Fig. 1 | Flexible graphene solution-gated field-effect transistor array technology and characterization. a, 90 Schematic of a graphene transistor polarized in common gate mode. b, Optical microscope images of the 91 active area of a 4×4 gSGFET array and a 15 channel intracortical array. c, Photograph of the neural probe after 92 peeling from the wafer and being introduced into a zero insertion force connector. d-g, Steady-state and 93 frequency response characterization of a 100x50-μm2 gSGFET array in 10 mM phosphate buffered saline 94 (PBS) with a drain-source voltage bias (Vds) of 50 mV. d, gSGFET transfer curves (blue lines), drain-source 95 current (Ids) vs gate-source voltage (Vgs), together with the mean (dark blue) and standard deviation (blue 96 shade). Boxplot inset shows charge neutrality point dispersion (center line, median; box limits, upper and 97 lower quartiles). e, Leakage current (Igs) of all gSGFETs in the array throughout the voltage sweep. f, Transfer 98 curve (blue squares and line) and its first derivative (transconductance (gm), black line) of a gSGFET. g, 99 Frequency response of the transconductance at two different points of the transfer curve (f): Vgs lower than the 100 CNP (green), where gm is negative resulting in a signal inversion (180º phase); and Vgs higher than the CNP 101 (orange), where gm is positive and thus results in no inversion (0º phase) . Independently of the branch of the 102 transfer curve where a gSGFET is polarized, the module of gm is similar to the steady-state value for a wide 103 bandwidth (≈ 0 - 1 kHz). 104 5 105 106 Fig. 2 | Infraslow, local field potential, and wide-band in vivo gSGFET recordings of cortical spreading 107 depression (CSD). a, Schematic of the gSGFET recording setup and signal post processing methodology. The 108 custom electronic circuit is used to perform the in vivo characterization (transfer curve) and record the 109 transistor current in the low-pass-filtered (LPF) and the band-pass-filtered (BPF) bands. From the combination 110 of both signals and taking into account the current-to-voltage conversion, the calibrated wide-band signal 111 (𝑉𝑠𝑖𝑔 ) is obtained. b, Schematic of a rat skull depicting the craniotomy (grey area), the location of the gSGFET 112 array and micropipette as also the frontal craniotomy where 5mM KCl was applied to induce CSDs. c, 113 Electrophysiological recordings obtained with a gSGFET epicortical array during the induction of four CSD 114 events (blue shade). From top to bottom: LPF signal, BPF and voltage-converted wide-band signal. d, 115 Voltage-converted wide-band signal of a CSD event recorded by a gSGFET and spectrogram showing the 116 characteristic silencing of activity. e, Comparison of a CSD signal recorded by a graphene transistor and a 117 solution-filled glass micropipette with a Ag/AgCl wire demonstrating the excellent similarity in shape, 118 magnitude and time span. 119 6 120 In vivo wide-band recordings with gSGFETs 121 Cortical spreading depression10,12,19 was chosen to illustrate the capabilities of graphene transistors to 122 record electrophysiological signals in a wide bandwidth. Experimentally, two craniotomies were 123 performed over the left hemisphere of isoflurane-anaesthetized Wistar rats: a larger craniotomy over 124 the primary somatosensory cortex, where the epicortical probe was placed, and a smaller one in the 125 frontal cortex, where 5 mM KCl was applied locally to induce CSD (Fig. 2b). A custom electronic 126 circuit allowed us to simultaneously record at two frequency bands: low-pass filtered band (LPF, ≈0- 127 0.16 Hz) and band-pass filtered band (BPF, 0.16 Hz-10 kHz) with different gains (104, and 106 128 respectively) to avoid amplifier saturation due to the high-amplitude CSD signal. In a first set of 129 experiments, we recorded the LPF and BPF signals with an epicortical gSGFET array during the 130 induction of CSD events (Fig. 2c). The graphene transistors were polarized in the hole conduction 131 regime, i.e. Vgs < CNP (negative gm)resulting in an inversion of the recorded LPF and BPF current 132 signals with respect to the voltage signal occurring at the gate. The LPF signal shows the very slow 133 CSD event whereas the BPF signal corresponds to the local field potential, revealing the silencing of 134 activity characteristic of cortical spreading depression. It is important to note that the high frequency 135 content of the steep depolarization seen in the BPF signal at the beginning of each CSD event is 136 generally the unique information of the CSD seen in AC-coupled recordings. After calibrationthe 137 wide-band electrophysiological signal can be obtained (see Fig. 2 a, c and Data Analysis section in 138 Methods). The calibration procedure eliminates both the variations associated with the different 139 current levels and the transconductance differences at the bias point between the transistors (Fig. S1). 140 In each CSD event a small positive shift of 1-2 mV generally precedes the depression, immediately 141 after which a steep negative change (≈ -20 mV) can be observed, which slowly recovers during the 142 next minute or so. The CSD-associated silencing of high-frequency activity and its progressive 143 recovery is shown in the voltage wave and spectrogram of Fig. 2d. In order to confirm the fidelity of 144 the CSD recordings of the gSGFET technology, simultaneous recordings with a solution-filled glass 145 micropipette with a Ag/AgCl wire were conducted. The infraslow deflection associated with CSD as 146 measured by gSGFETs has a very similar shape, magnitude and temporal duration than the signal 147 recorded by a micropipette (see Fig.2e and Fig.S2: cross-correlation = 0.85 ± 0.1 for the recording of 148 two CSD events). 149 ISA recording capabilities with gSGFETs and microelectrodes 150 A second set of experiments was designed to compare the performance of gSGFETs with 151 microelectrodes in in-vivo direct-coupled recordings. CSD was induced and simultaneously recorded 152 with an gSGFET epicortical array located more posterior to a neural probe containing groups of 153 triodes of 50 µm diameter gold microelectrodes 200 µm apart in which one microelectrode of each 154 triode was modified by deposition of platinum black to lower its impedance (Fig. S3). Data shown in 155 Fig. 3 corresponds to a representative experiment of n=3 independent subjects. Fig. 3a shows that 156 gold and platinum black recordings exhibit very large and diverse baseline offsets as well as 157 oscillations and drifts (-7.9 ± 3.3 mV/h, n=10 and -3.6 ± 1.6 mV/h, n=6), while the gSGFET signals 158 are very stable (1.1 ± 1.0 mV/h, n=15). Importantly, gSGFETs record significantly higher amplitudes 159 for the CSD events (-13.3 ± 1.8 mV) in comparison with gold (-4.7 ± 1.6 mV) and platinum black (- 7 160 3.0 ± 0.7 mV) microelectrodes. Figure 3b highlights one of the intrinsic limitations of microelectrode 161 technology for the measurement of ISA: polarization-induced drift. 162 163 Fig. 3 | Comparison of DC-coupled gSGFET and microelectrode recordings of cortical spreading 164 depression. a-c, Representative data of one of a total of three independent experiments. a, Direct-coupled 165 recordings of 100 x 50 µm2 gSGFET transistors and gold and platinum black 50 µm diameter microelectrodes. 166 The vertical dashed lines show the time when KCl (5 mM) was applied to induce a CSD. b, DC-offset 167 removed recordings of a representative channel of each type. Black lines illustrate the mean drift: dotted and 168 dashed correspond to gold and platinum black microelectrodes, respectively, and the dash-dotted line 169 corresponds to gSGFETs. c, DC-offset removed recordings of a representative channel of each type and the 170 same signal filtered at 0.002 Hz to remove oscillations and drift; the gSGFET signal does not require any 171 filtering and is therefore not distorted. 172 The drift of the baseline potential superimposed over the huge voltage offsets is problematic as it can 173 lead to saturation of the amplifiers used to record the signal. More importantly, baseline oscillations 174 in the infralow frequencies, will potentially hamper the determination of the exact characteristics of 175 CSD, such as amplitude or waveform, as the required high-pass filter used to remove such effects 8 176 will alter the signal shape (see Fig.3c and Fig. S4). Another intrinsic limitation of microelectrode 177 technology is based on the relation between the microelectrode impedance and the input impedance 178 of the recording equipment (𝑍𝑒′ and 𝑍𝑎′ , respectively)21,23. The recorded signal (𝑉𝑖𝑛 ) is determined by 179 the voltage divider formed by both impedances: 𝑉𝑠𝑖𝑔 (𝑓) 𝑍𝑎′ (𝑓) 𝑉𝑖𝑛 (𝑓) = 𝐼(𝑓)𝑍𝑎′ (𝑓) = ′ (1) 𝑍𝑎 (𝑓) + 𝑍𝑒′ (𝑓) 180 Eq. (1) implies that when 𝑍𝑎′ is not substantially larger than 𝑍𝑒′ , the recorded signals will be 181 attenuated and shifted with respect to 𝑉𝑠𝑖𝑔 22. It is important to highlight that the 𝑍𝑎′ >>𝑍𝑒′ requirement 182 to achieve a voltage gain equal to 1 could be compromised, especially at very-low frequencies, when 183 the electrode area is scaled down, due to the inverse relation between electrode impedance and its 184 area, leading to high-pass filtering of the recorded signals. By measuring the impedance of both 185 electrode types and modelling the preamplifier impedance with the values reported by the 186 manufacturer, we obtained the voltage gain (𝑉𝑖𝑛 /𝑉𝑠𝑖𝑔 ) of the equivalent circuit formed by the 187 recording electrode and the amplifier, see Fig. 4a-b.Fig. 4c shows a representative CSD recorded by a 188 gSGFET and gold and platinum black microelectrodes and the mean amplitude of the first peak for 189 each type. For the 50 µm diameter gold microelectrodes, an attenuation lower than 50% is expected 190 from Fig. 4b, which is in agreement with the experimental results. For the platinum black electrodes 191 we tentatively attribute the higher than predicted attenuation to in vivo electrochemical processes that 192 impact the electrode response at very low frequencies33. We assign the superior performance of 193 gSGFETs to the following main reasons. First, graphene exhibits an excellent DC stability, as 194 demonstrated by low in vivo drift. We attribute this to the low density of states of pristine graphene 195 near the Fermi level, which decreases the overall electronic overlap with redox species34, and to the 196 low density of extrinsic electron transfer sites, i.e. defects and edges, all contributing to the excellent 197 electrochemical inertness of CVD graphene24,35,36. The low leakage current measured (Fig. 1e) also 198 supports the electrochemical inertness. 9 199 200 Fig. 4 | Microelectrode and gSGFET recording modes: considerations for infraslow recordings. a, Cross- 201 sectional view and superimposed electric equivalent circuit models of a recording electrode and a gSGFET. 202 For an electrode, the electrode-electrolyte interface, is modelled simply as a capacitor and a resistor in parallel 203 (Re-e, Ce-e). Vin, the voltage at the input of the amplifier is determined by the voltage divider formed by 𝑍𝑒′ and 204 𝑍𝑎′ , the effective electrode and amplifier impedance, respectively. Rs represents the electrolyte resistance. In 205 the case of a gSGFET, Vsig modulates the graphene channel resistance (Rds) by field-effect through the gate 206 capacitance (Cg-e), which results in current variations (ids) proportional to the transconductance value at the 207 bias point, plus the voltage signal (which is mostly negligible for small amplitude electrophysiological 208 signals), as seen in the small signal model. b, Mean and standard deviation of the impedance module 209 (experimental data) of nine 50 µm diameter gold (blue) and six platinum black (grey) microelectrodes together 210 with 𝑍𝑎′ and calculated voltage gain (𝑉𝑖𝑛 /𝑉𝑠𝑖𝑔 ) for each microelectrode type. c, Recordings of a CSD event for 211 each type of microelectrodes and a gSGFET. Horizontal lines represent the mean value of CSD amplitude. 10 212 The second reason why graphene microtransistors can record infraslow signals is related to their 213 working mechanism, which is significantly different from that of electrodes. In gSGFETs, voltage 214 oscillations near the active graphene channel modulate the current flow along it (see schematic and 215 small-signal model in Fig. 4a). Eq. 2 shows the relation between the recorded current (Ids−rec ) and 216 the signal (Vsig ): 𝐼𝑑𝑠−𝑟𝑒𝑐 (𝑉𝑔𝑠 , 𝑉𝑠𝑖𝑔 ) = 𝐼𝑑𝑠 (𝑉𝑔𝑠 ) + 𝑖𝑑𝑠 (𝑉𝑔𝑠 , 𝑉𝑠𝑖𝑔 ) = 𝐼𝑑𝑠 (𝑉𝑔𝑠 ) + 𝑔𝑚 (𝑉𝑔𝑠 + 𝑉𝑠𝑖𝑔 ) 𝑉𝑠𝑖𝑔 , (2) 217 where Ids is the current at the bias point Vgs and ids the current variation induced by the gate signal. 218 This equation is valid and frequency-independent as long as 𝑔𝑚 is also frequency-independent. In 219 this work (Fig. 1g), we provide evidence that the transconductance of gSGFETs remains constant in a 220 wide bandwidthand and that this behaviour is preserved with further downscaling of gSGFETs 221 (Fig.S5). 222 Mapping cortical spreading depression with gSGFETs 223 As an example of the potential of gSGFET technology, we mapped the propagation of CSD events 224 using a 4x4 epicortical gSGFET array and compared the signals with what is observed in 225 conventional high-pass filtered recordings (Fig. 5a-b). The recording of the CSD event with the 226 gSGFET array reveals that while the onset of the negative shift is similar for all gSGFETs, there is 227 much more variety in the subsequent recovery, with some transistors exhibiting a second negative 228 shift with higher amplitude than the first one. This effect can also be observed in the last two frames 229 (corresponding to 80 s and 90 s) of the spatial maps of gSGFET recordings (Fig. 5b) where recovered 230 and still depressed brain areas coexist. Importantly, this information is lost in conventional 231 microelectrode recordings, where only the CSD onset is observed due to the high pass filter in the 232 recording electronics. We found that the mean duration of CSD events is 47 ± 8 s and a speed of 233 propagation of 8 ± 1 mm/min (n=10 CSDs collected from two different subjects). 234 Under physiological conditions, there is a neurovascular response, vasodilatation and increased 235 regional cerebral blood flow (rCBF) due to spreading depolarization that causes spreading 236 hyperemia10. However, most studies on CSD neurovascular coupling have been performed with 237 mapping techniques for the rCBF while electrical activity is measured only at two sites with glass 238 micropippetes5. Here, taking advantage of the gSGFET technology, we designed an experiment in 239 which we could simultaneously map both variables. Fig. 5c provides further evidence of the 240 spreading depolarization and hyperemia neurovascular coupling. We used a non-contact, wide-field 241 technique, laser speckle contrast imaging (LSCI)37, that allows imaging variations of rCBF38. 242 Experimentally, a craniotomy was performed in a Wistar rat and a continuous-wave temperature 243 controlled laser diode and a camera were mounted to image a wide area inside in which an epicortical 244 16-channel gSGFET array was placed. After 5mM KCl administration, CSD was induced, which was 245 followed by an increase in rCBF that slowly returned (4-5minutes) to basal values (Fig. 5c). 246 We also performed in vivo experiments with intracortical probes consisting of a linear array of 15 247 gSGFETs spanning the entire depth of a rat cortex (Fig. 6a). From both the ordered recording and the 248 spatiotemporal voltage map (Fig. 6b), it can be seen how CSD occurs in the whole cortex depth. A 11 249 transition from a superficial long depolarization to a shorter one preceded and followed by a 250 hyperpolarization in the deeper layers can be clearly observed. 251 Outlook 252 In this work we show that gSGFETs can record neural signals in a wide electrophysiological 253 bandwidth, from infralow (<0.1 Hz) frequencies to the typical local field potential bandwidth, similar 254 to solution-filled glass micropipettes but with the capability of overcoming their spatial sampling 255 limitations. Importantly, this capability does not depend on a given transistor size but is preserved 256 among a wide range of device sizes, which brings freedom when designing an array for a given 257 application. There are two main reasons that explain this unique recording capability: the direct DC- 258 coupling of transistors, in contrast to standard passive electrodes; and the excellent electrochemical 259 stability of graphene. Making use of these features, gSGFET technology opens the possibility of 260 mapping infraslow oscillations with high fidelity and spatial resolution (epicortically and 261 intracortically). This can lead to a better understanding of the brain regions where ISA is initiated, its 262 propagation to other areas and a clarification of the interplay of different cellular types, which are 263 still poorly understood1,2,39. Additionally, the wide recording bandwidth of gSGFETs can help in 264 determining the relation between ISA and higher frequency signals17,40 and contribute to a better 265 understanding of the genesis of local field potentials41 and of cortical wave propagation features42,43. 266 Since 2014, work exploiting both the transparency and electrical conductivity of graphene has 267 allowed simultaneous local field potential recordings using graphene microelectrodes, and imaging or 268 optical stimulation at the same position, which has profound implications in neuroscience24,44. Our 269 work demonstrates that graphene transistors can be used together with imaging techniques, such as 270 LSCI to map infraslow electrophysiological signals and regional cerebral blood flow. This 271 combination of techniques holds great potential and can contribute to a better understanding of 272 neurovascular coupling phenomena. 273 12 274 275 Fig. 5 | Mapping cortical spreading depression with graphene transistors. a, Infralow frequency signals 276 recorded by a 4x4, 400 µm grid spacing, gSGFET array (black lines) during the occurrence of a CSD event as 277 illustrated in the top left schematic. The contour plot shows the time delays of the onset of CSD with respect to 278 the mean time illustrating the spatiotemporal course of the CSD. b, Interpolated spatial voltage maps showing 279 the propagation of the same CSD event as measured by the gSGFET array. a,b High pass filtered recordings at 280 0.1Hz (red lines in a and bottom spatial voltage maps in b) are included to illustrate the loss of signal 281 information in conventional microelectrode recordings. c, Schematic of a rat skull depicting the laser speckle 282 contrast imaging field-of-view and the position of the gSGFET array. Electrical recordings and optical 283 imaging were performed directly on the cortical surface. Time evolution of the upper right and lower left 284 graphene microtransistors as well as the regional cerebral blood flow (rCBF) measured at the same position 285 showing their co-occurrence. Colour maps represent the spatial value of the extracellular voltage as measured 13 286 by the gSGFET array (top) and the rCBF (bottom) at a given set of times after the induction of a CSD event. 287 Representative data of one of a total of two independent experiments. 288 289 Fig. 6 | Depth profile of the infralow-frequency voltage variations induced by cortical spreading 290 depression in a rat cortex. a, Layout of the fabricated 15-channel graphene intracortical probe and 291 ordered local field potential recordings. Infralow-frequency recordings (black lines) during the 292 occurrence of a CSD event. Dashed lines, have been interpolated from nearby transistors. Depth 293 position is indicated by the layer number and corpus callosum (CC). b, Colour maps of the temporal 294 course of the infraslow changes during a CSD event across the depth of a rat cortex. a-b, Same signal 295 high-pass filtered at 0.1 Hz (red lines) and their spatio-temporal colour map are included to illustrate 296 the loss of information in conventional microelectrode recordings. 297 In the particular case of CSD, where no non-invasive electrophysiological technique has been 298 demonstrated capable of its monitoring, the adoption of invasive DC-coupled electrode recordings 299 has been proposed to provide further diagnostic information and easy and direct detection of CSDs 13. 300 gSGFET technology emerges as a potential preclinical as well as clinically relevant tool to help 301 determine the relation of CSDs to neural disorders such as migraine, malignant stroke, subarachnoid 302 and intracranial haemorrhage, and traumatic brain injury. If the challenges of translating gSGFET 303 technology to the clinics, such as chronic and safe operation and human compatibility are surpassed, 304 gSGFETs could be applied in neurointensive care monitoring12,14or for CSD intraoperative 305 monitoring since there is evidence that CSD can occur during neurosurgical procedures 45. 306 Importantly,in contrast to electrodes where a signal is needed to measure electrode impedance, the 307 possibility to measure the characteristic transfer curve of a gSGFET in vivo at any time, allows 308 assessing the stability as well as the signal coupling magnitude (transconductance) during an implant 14 309 lifetime, therefore easing the evaluation of its chronic performance. In summary, our work 310 demonstrates that gSGFET arrays are ideal candidates to fill the gap of a large-scale, high- 311 spatiotemporal recording technology that covers a wide electrophysiological bandwidth in a 312 potentially fully implantable, nontoxic, clinical-scale device. By measuring the full bandwidth of 313 brain activity with high spatiotemporal resolution we will be able to improve our understanding of 314 brain function in health and disease status, and develop better diagnostic and therapeutic procedures 315 for those affected. 316 References 317 1 Hughes, S. W., Lőrincz, M. L., Parri, H. R. & Crunelli, V. in Progress in Brain Research Vol. 318 193 (eds Eus J. W. Van Someren et al.) 145-162 (Elsevier, 2011). 319 2 Mitra, A. et al. Spontaneous Infra-slow Brain Activity Has Unique Spatiotemporal Dynamics 320 and Laminar Structure. Neuron, doi:https://doi.org/10.1016/j.neuron.2018.03.015. 321 3 Lecci, S. et al. Coordinated infraslow neural and cardiac oscillations mark fragility and 322 offline periods in mammalian sleep. Science Advances 3, e1602026 (2017). 323 4 Mitra, A., Snyder, A. Z., Tagliazucchi, E., Laufs, H. & Raichle, M. E. Propagated infra-slow 324 intrinsic brain activity reorganizes across wake and slow wave sleep. Elife 4 (2015). 325 5 Hiltunen, T. et al. Infra-slow EEG fluctuations are correlated with resting-state network 326 dynamics in fMRI. Journal of Neuroscience 34, 356-362 (2014). 327 6 Leopold, D. A., Murayama, Y. & Logothetis, N. K. Very slow activity fluctuations in monkey 328 visual cortex: implications for functional brain imaging. Cerebral cortex 13, 422-433 (2003). 329 7 Kelly, A. C., Uddin, L. Q., Biswal, B. B., Castellanos, F. X. & Milham, M. P. Competition 330 between functional brain networks mediates behavioral variability. Neuroimage 39, 527-537 331 (2008). 332 8 Chung, D. Y. & Ayata, C. in Primer on Cerebrovascular Diseases (Second Edition) (eds 333 José Biller et al.) 149-153 (Academic Press, 2017). 334 9 Hartings, J. A. et al. The continuum of spreading depolarizations in acute cortical lesion 335 development: examining Leão’s legacy. Journal of Cerebral Blood Flow & Metabolism 37, 336 1571-1594 (2017). 337 10 Dreier, J. P. The role of spreading depression, spreading depolarization and spreading 338 ischemia in neurological disease. Nat Med 17, 439-447 (2011). 339 11 Dreier, Jens P. & Reiffurth, C. The Stroke-Migraine Depolarization Continuum. Neuron 86, 340 902-922, doi:https://doi.org/10.1016/j.neuron.2015.04.004 (2015). 341 12 Lauritzen, M. et al. Clinical relevance of cortical spreading depression in neurological 342 disorders: migraine, malignant stroke, subarachnoid and intracranial hemorrhage, and 343 traumatic brain injury. Journal of Cerebral Blood Flow & Metabolism 31, 17-35 (2011). 344 13 Hartings, J. A. et al. Direct current electrocorticography for clinical neuromonitoring of 345 spreading depolarizations. Journal of Cerebral Blood Flow & Metabolism 37, 1857-1870 346 (2017). 15 347 14 Dreier, J. P. et al. Recording, analysis, and interpretation of spreading depolarizations in 348 neurointensive care: review and recommendations of the COSBID research group. Journal of 349 Cerebral Blood Flow & Metabolism, 0271678X16654496 (2016). 350 15 Kovac, S., Speckmann, E.-J. & Gorji, A. Uncensored EEG: The role of DC potentials in 351 neurobiology of the brain. Progress in Neurobiology, 352 doi:https://doi.org/10.1016/j.pneurobio.2018.02.001 (2018). 353 16 Vanhatalo, S., Voipio, J. & Kaila, K. Full-band EEG (FbEEG): an emerging standard in 354 electroencephalography. Clinical Neurophysiology 116, 1-8, 355 doi:http://dx.doi.org/10.1016/j.clinph.2004.09.015 (2005). 356 17 Vanhatalo, S. et al. Infraslow oscillations modulate excitability and interictal epileptic activity 357 in the human cortex during sleep. Proceedings of the National Academy of Sciences of the 358 United States of America 101, 5053-5057 (2004). 359 18 Hofmeijer, J. et al. Detecting Cortical Spreading Depolarization with Full Band Scalp 360 Electroencephalography: An Illusion? Frontiers in Neurology 9, 361 doi:10.3389/fneur.2018.00017 (2018). 362 19 Ayata, C. & Lauritzen, M. Spreading depression, spreading depolarizations, and the cerebral 363 vasculature. Physiological reviews 95, 953-993 (2015). 364 20 Stensaas, S. S. & Stensaas, L. J. Histopathological evaluation of materials implanted in the 365 cerebral cortex. Acta Neuropathologica 41, 145-155, doi:10.1007/bf00689766 (1978). 366 21 Li, C. et al. Evaluation of microelectrode materials for direct-current electrocorticography. 367 Journal of neural engineering 13, 016008 (2015). 368 22 Nelson, M. J., Pouget, P., Nilsen, E. A., Patten, C. D. & Schall, J. D. Review of signal 369 distortion through metal microelectrode recording circuits and filters. Journal of neuroscience 370 methods 169, 141-157 (2008). 371 23 Stacey, W. C. et al. Potential for unreliable interpretation of EEG recorded with 372 microelectrodes. Epilepsia 54, 1391-1401 (2013). 373 24 Kuzum, D. et al. Transparent and flexible low noise graphene electrodes for simultaneous 374 electrophysiology and neuroimaging. Nature communications 5, 5259 (2014). 375 25 Deneux, T. et al. Accurate spike estimation from noisy calcium signals for ultrafast three- 376 dimensional imaging of large neuronal populations in vivo. Nature Communications 7, 377 12190, doi:10.1038/ncomms12190 378 https://www.nature.com/articles/ncomms12190#supplementary-information (2016). 379 26 Fang, H. et al. Capacitively coupled arrays of multiplexed flexible silicon transistors for long- 380 term cardiac electrophysiology. Nature biomedical engineering 1, 0038 (2017). 381 27 Heremans, P. et al. Mechanical and electronic properties of thin‐film transistors on plastic, 382 and their integration in flexible electronic applications. Advanced Materials 28, 4266-4282 383 (2016). 384 28 Hess, L. H., Seifert, M. & Garrido, J. A. Graphene transistors for bioelectronics. Proceedings 385 of the IEEE 101, 1780-1792 (2013). 386 29 Kim, B. J. et al. High-Performance Flexible Graphene Field Effect Transistors with Ion Gel 387 Gate Dielectrics. Nano Letters 10, 3464-3466, doi:10.1021/nl101559n (2010). 16 388 30 Kostarelos, K., Vincent, M., Hebert, C. & Garrido, J. A. Graphene in the Design and 389 Engineering of Next‐Generation Neural Interfaces. Advanced Materials 29 (2017). 390 31 Benno, M. B. et al. Mapping brain activity with flexible graphene micro-transistors. 2D 391 Materials 4, 025040 (2017). 392 32 Hébert, C. et al. Flexible Graphene Solution‐Gated Field‐Effect Transistors: Efficient 393 Transducers for Micro‐Electrocorticography. Advanced Functional Materials 28, 1703976, 394 doi:doi:10.1002/adfm.201703976 (2018). 395 33 Shinwari, M. W. et al. Microfabricated Reference Electrodes and their Biosensing 396 Applications. Sensors 10, 1679 (2010). 397 34 Chen, S., Liu, Y. & Chen, J. Heterogeneous electron transfer at nanoscopic electrodes: 398 importance of electronic structures and electric double layers. Chemical Society Reviews 43, 399 5372-5386 (2014). 400 35 Brownson, D. A. & Banks, C. E. The electrochemistry of CVD graphene: progress and 401 prospects. Physical Chemistry Chemical Physics 14, 8264-8281 (2012). 402 36 Brownson, D. A. C., Munro, L. J., Kampouris, D. K. & Banks, C. E. Electrochemistry of 403 graphene: not such a beneficial electrode material? RSC Advances 1, 978, 404 doi:10.1039/c1ra00393c (2011). 405 37 Valdes, C. P. et al. Speckle contrast optical spectroscopy, a non-invasive, diffuse optical 406 method for measuring microvascular blood flow in tissue. Biomedical optics express 5, 2769- 407 2784 (2014). 408 38 Boas, D. A. & Dunn, A. K. Laser speckle contrast imaging in biomedical optics. Journal of 409 biomedical optics 15, 011109 (2010). 410 39 Shibata, M. & Suzuki, N. Exploring the role of microglia in cortical spreading depression in 411 neurological disease. Journal of Cerebral Blood Flow & Metabolism 37, 1182-1191, 412 doi:10.1177/0271678x17690537 (2017). 413 40 Mitra, A. & Raichle, M. E. How networks communicate: propagation patterns in spontaneous 414 brain activity. Phil. Trans. R. Soc. B 371, 20150546 (2016). 415 41 Herreras, O. Local field potentials: myths and misunderstandings. Frontiers in neural circuits 416 10, 101 (2016). 417 42 Massimini, M., Huber, R., Ferrarelli, F., Hill, S. & Tononi, G. The sleep slow oscillation as a 418 traveling wave. Journal of Neuroscience 24, 6862-6870 (2004). 419 43 Capone, C. et al. Slow Waves in Cortical Slices: How Spontaneous Activity is Shaped by 420 Laminar Structure. Cerebral Cortex, 1-17 (2017). 421 44 Park, D.-W. et al. Graphene-based carbon-layered electrode array technology for neural 422 imaging and optogenetic applications. Nature communications 5, ncomms6258 (2014). 423 45 Carlson, A. P. et al. Cortical spreading depression occurs during elective neurosurgical 424 procedures. Journal of neurosurgery 126, 266-273 (2017). 425 426 17 427 Acknowledgments 428 This work was funded by the European Union’s Horizon 2020 research and innovation programme 429 under Grant Agreement No. 696656 (Graphene Flagship) and No. 732032 (BrainCom). This work 430 has made use of the Spanish ICTS Network MICRONANOFABS partially supported by MINECO 431 and the ICTS 'NANBIOSIS', more specifically by the Micro-NanoTechnology Unit of the CIBER in 432 Bioengineering, Biomaterials & Nanomedicne (CIBER-BBN) at the IMB-CNM. E.M.C. worked in 433 this manuscript during the PhD in Electrical and Telecommunication Engineering at the Universitat 434 Autònoma de Barcelona. E. d. C. thanks to Spanish Ministerio de Economía y Competitividad for the 435 Juan de la Cierva postdoctoral grant IJCI-2015-25201. T. Du acknowledges support from Fundació 436 CELLEX Barcelona, Ministerio de Economía y Competitividad /FEDER (PHOTODEMENTIA, 437 DPI2015-64358-C2-1-R), the “Severo Ochoa” Programme for Centres of Excellence in R&D (SEV- 438 2015-0522) and the Obra social “la Caixa” Foundation (LlumMedBcn). 439 Author contributions 440 E.M.C. did most of the fabrication and characterization of the gSGFET arrays, contributed to the 441 design and performance of the in vivo experiments, analyzed the data and wrote the manuscript. X.I. 442 designed the neural probes and fabricated the microelectrode arrays. A.B.C. contributed to the 443 fabrication and characterization of the gSGFET arrays. M.D. performed the in vivo experiments. 444 P.G., C.H., J.B. and E.P.A. contributed to the growth of the CVD graphene. E.P.A., E.dC. and 445 J.M.dC.S. contributed to the transfer of graphene. E.P.A., E.dC.G. and G.R. contributed to the 446 characterization of CVD graphene. J.M.A. contributed to the fabrication of the custom electronic 447 instrumentation and development of a python-based user interface. A.C. contributed to the CSD 448 propagation analysis. R.G.C. contributed in the noise characterization and analysis of the devices. 449 T.Dr., E.V. and T.Du. contributed to the in vivo measurements and analysis of cerebral blood flow. 450 M.D., M.S.V., A.G.B, R.V. and J.A.G. participated in the design of the in vivo experiments and 451 thoroughly reviewed the manuscript. A.G.B. contributed in the design and fabrication of the custom 452 electronic instrumentation, development of a custom gSGFET python library and in the analysis of 453 the data. All authors read and reviewed the manuscript. 454 Competing interests 455 Patent application (nº P201831068) filled by CSIC, ICREA, CIBER, ICN2 and IDIBAPS; inventors: 456 A.G.B., E.M.C., X.I., R.V., M.S.V. and J.A.G.; concerning a graphene transistor system for 457 measuring electrophysiological signals (pending). 458 Methods 459 Graphene growth and characterization 460 Graphene layers were grown by Chemical Vapor Deposition (CVD) using one of the following 461 procedures: a) A lamp-heated rapid thermal CVD equipment from Jipelec and 25 µm thick, 99.8 % 462 metal basis copper foil provided by AlfaAesar have been employed. Prior to graphene CVD growth, 463 copper foils were sequentially cleaned in acetic acid and acetone, and finally rinsed in isopropyl 464 alcohol (IPA). Sample dimensions were 6 x 5 cm2. Growth processing conditions consisted in 10 465 minutes at 750 °C, 200 sccm H2 plus 5 minutes at 800 °C, 25 sccm CH4 / 200 sccm H2. b) Chemical 466 vapour deposition was on a 4.5x7 cm2 copper foil (Alfa Aesar, annealed, Coated). Prior to the 467 growth, the copper foil was electropolished during 5 min at a fixed current density of 62 mA/cm-2 in 18 468 a solution containing H20 (1 L) + H3PO4 (0.5 L) + ethanol (0.5 L) + isopropanol (0.1 L) and urea (10 469 g). Then, the copper foil was loaded in a planar quartz tube (1600x60 mm) and heated by a three zone 470 oven. A first annealing step at 1015 ºC under a 400 sccm argon flow at 100 mbar during 1 h was 471 followed by a 15-min growth step at 12 mbar under a gas mixture of 1000 sccm argon, 200 sccm 472 hydrogen and 2 sccm of methane. The sample was then cooled down under a 400 sccm argon flow 473 by removing the quartz tube from the oven. For all samples, a complete Raman characterization was 474 performed using a Witec spectrograph (Fig.S6a-d). Raman maps of 30x30 m2 were registered with a 475 spatial resolution lower than 1 m2 (using a 50x objective). We used a 488 nm excitation wavelength 476 to minimize the copper substrate luminescence signal. The laser power was kept below 1.5 mW to 477 avoid sample heating and a 600 g/nm grating was used to provide a pixel to pixel spectral resolution 478 below 3 cm-1. 479 gSGFET array fabrication and characterization 480 Four-inch silicon wafers were used as a support to build the devices. First, a 10-µm-thick polyimide 481 layer (PI-2611, HD MicroSystems) was spin-coated to be used as substrate and hard-baked at 350ºC 482 to complete the imidation process. Graphene transistors were fabricated in a sandwich-like structure. 483 For that, a first layer of metal (Ti/Au, 10/100 nm) was evaporated and defined in a standard lift-off 484 process using the image reversal photoresist AZ5214E (Clariant GmbH, Germany). Then, single- 485 layer graphene was transferred by electrochemical delamination46. After removing the PMMA 486 protection layer, the graphene active areas were defined by means of an oxygen-based reactive ion 487 etching (RIE). A second metal layer (Ni/Au, 20/200nm) was evaporated and defined in a similar 488 standard lift-off process avoiding the use of ultrasounds in order to maintain graphene integrity. SU-8 489 (SU-8 2005, MicroChemCorp., USA) a permanent epoxy-based negative photoresist was used to 490 passivate the metal leads while defining the graphene channel and metal contacts. Finally, the 491 polyimide substrate was structured in a deep-RIE process using the thick AZ9260 positive 492 photoresist (Clariant GmbH, Germany) as an etching mask. Polyimide probes were directly peeled 493 off from the wafer and placed in a zero insertion force (ZIF) connector to be interfaced with our 494 custom electronic instrumentation. Current-voltage measurements of graphene transistors were 495 performed in common gate mode with a fixed drain-source voltage (𝑉𝑑𝑠 =50 mV) varying the gate- 496 source voltage (𝑉𝑔𝑠 ) vs. a Ag/AgCl reference electrode in 0.01 M PBS solution. Steady-state was 497 ensured by acquiring only after time derivative of 1 s of current is below 5e-7 A/s. The total leakage 498 current was measured for the whole array and corresponds to the sum of the individual leakage 499 currents of all transistors in the array. The frequency response of the transconductance was measured 500 by applying a sum of sinusoidal signals at the electrolyte solution through the reference electrode and 501 by measuring the modulation of the drain current. The acquired signals were splitted into two bands, 502 low frequencies (≈0-10 Hz) in which drain-source current was simultaneous acquired for all 503 transistors in an array, and higher frequencies (10 Hz-30 kHz) in which each transistor was recorded 504 individually (Fig. S7). Data reporting the root-mean-square gate voltage noise dependence on 505 transistor area is included in Supplementary Figure S8 for a better characterization of current 506 gSGFET technology. 507 Microelectrode array fabrication and characterization 508 The flexible microelectrode array was fabricated in polyimide in a very similar process. Here, a 509 Ti/Au (20/200 nm) metal layer was evaporated on a 10 µm-thick polyimide-covered four-inch silicon 510 wafer to define the metal tracks and the microelectrodes, while a second polyimide layer (2 µm thick) 511 was used as the passivation layer. Two subsequent etching steps were used to open, firstly, the 512 microelectrode active areas and, secondly, to structure the polyimide in order to define the probe 513 geometry which is the same as in Illa et. al.47. Platinum black was deposited in some electrodes (Fig. 514 S3a) by constant polarization amperometry. A voltage of -0.2V against a Ag/AgCl reference 19 515 electrode was applied during 15 s. Impedance spectra were measured against a Ag/AgCl reference 516 electrode using a Solartron SI 1260 equipment (Solarton analytical, UK) with 20 mV signal 517 amplitude (Fig. S3b). 518 Ethical approval and animal handling 519 All experimental procedures were conducted in accordance with the European Union guidelines on 520 protection of vertebrates used for experimentation (Directive 2010/63/EU of the European Parliament 521 and of the Council of 22 September 2010) and all experiments were approved by the ethics 522 committee of the Hospital Clinic de Barcelona. Rats were kept under standard conditions (room 523 temperature 23 ±1ºC, 12:12-h light-dark cycle, lights on at 08:00), with food (A04, Harlan, Spain) 524 and water available ad libitum. 525 In vivo recordings 526 Eleven adult male Wistar rats (225-375 g) were used in this study. Animals were deeply 527 anaesthetized with isoflurane (4% induction, 1-3% maintenance) and all pressure and incision points 528 were infiltrated with local anesthetic lidocaine. Once under the surgical plane of anesthesia, animals 529 were transferred to a stereotaxic frame with body temperature constantly monitored and maintained 530 at 37ºC by means of a thermal blanket. A craniotomy and durotomy were performed on the left or 531 right hemisphere in order to record with epicortical or intracortical arrays, respectively. Additionally, 532 a craniotomy and durotomy were performed over the prefrontal cortex to topically administer 5 mM 533 KCl to induce cortical spreading depression. The large craniotomy was centred at 43 mm anterio- 534 posterior (AP) and 42.5 mm medio-lateral (ML) and was 6 mm AP by 4.5 mm ML in size while the 535 smaller craniotomy, located at 50 mm AP and 42 ML, was 2.5 mm AP by 1.25 mm ML. A Ag/AgCl 536 electrode pellet was inserted in temporal muscle and used as reference both for recordings and for the 537 measurement of the transistors transfer curve. All recording probes, either gSGFETs or 538 microelectrodes, were placed directly on the cortical surface and kept in place by adherence to the 539 tissue (Fig. S9a-b). A custom electronic instrumentation was used (Fig. S10), which provides the 540 current-to-voltage conversion and the bias control for each channel. The instrumentation splits the 541 recorded signals into two bands with different gains: low-pass filtered (< 0.16 Hz, 104 gain) and 542 band-pass filtered (0.16 Hz < f < 160 kHz, 106 gain). In the experiments where only the gSGFET 543 array was measured the low-pass filtered signals and bias control was managed by a data acquisition 544 system (National Instruments USB-6353), while the band-pass filtered signals were directly acquired 545 by a commercial electrophysiological recording system consisting of a programmable gain amplifier 546 (Multichannel Systems, GmbH) and digitizer interface (CED 1401 and Spike2 software, Cambrigde 547 Electronic Design, UK). The LPF and BPF bands were sampled at 1 Hz and 5 kHz respectively. Prior 548 to the beginning of the recordings, the transfer curve of the gSGFET was measured in situ to 549 determine the optimum bias point, generally around -0.1 V of the CNP (Fig. S9c-d). 550 For gSGFETs comparison experiments with microelectrodes and the glass micropipettes with 551 Ag/AgCl wire(≈0.15 MΩ) a total of four subjects was used: two subjects were measured with 552 gSGFETs, microelectrodes and a micropipette, one with gSGFETs and microlectrodes (data from 553 Fig.3) and another one with gSGFETs and a micropipette (data from Fig. 2e, and Fig.S2). A custom 554 Simulink model was used to simultaneously measure graphene transistors through an adapted 555 g.HIamp biosignal amplifier(g.tec medical engineering GmbH, Austria) while microelectrodes and 556 the solution-filled glass micropipette were recorded using an g.USBamp (g.tec medical engineering 557 GmbH, Austria). The same reference electrode was used by both amplifiers and signals were sampled 558 at 4.8 kHz. 559 Laser speckle contrast imaging 20 560 For the measurement of the regional cerebral blood flow (rCBF), a laser speckle contrast imaging 561 (LSCI) system was used which consists of a continuous-wave temperature-controlled laser diode 562 (785 nm, Thorlabs, Germany) for homogenous full-field illumination and a charge-coupled device 563 camera (sc640-120fm, Basler, Germany), with an exposure time of 5 ms, which captures the diffused 564 light scattered from the imaging area. The speckle contrast was calculated for the predefined region 565 of interest (ROI) at each pixel in temporal domain over 100 frames, to ensure good signal-to-noise 566 ratio. The statistics of different noise sources37 was accounted for when calculating the speckle 567 contrast. Speckle contrast was then related to a rCBF index (BF) as reported in 38,48. Finally, the 568 relative blood flow (ΔrCBF) was calculated as: 𝐵𝐹 − 𝐵𝐹𝐵 𝛥𝑟𝐶𝐵𝐹 = ∗ 100 [%] 𝐵𝐹𝐵 569 where 𝐵𝐹𝐵 corresponds to the basal regional blood flow. Fig. S11 shows the area where LSCI was 570 measured in Fig. 5c. 571 Data Analysis 572 All data were analyzed using Python 2.7 packages (Matplotlib, Numpy and Neo) and the custom 573 library PhyREC. The conversion of the recorded current signals (LPF and BPF) to a voltage signal 574 was performed by summation of both signals and interpolation in the in vivo measured transfer curve 575 of the corresponding gSGFET. The transfer curve was always measured, at least, at the beginning 576 and end of every acute experiment, and generally some more transfer curves measurements were 577 performed along the duration of the experiment. Comparison of the evolution of the in vivo measured 578 transfer curves was systematically performed during data analysis (see Fig. S12a) to ensure that no 579 significant variations are present and to detect (if there are) any misbehaving transistor. Moreover, all 580 recordings presented in the manuscript have been calibrated with the nearest transfer curve measured 581 (following the procedure shown in Fig. S1) to ensure high fidelity in the voltage-converted signals. 582 For visualization purposes microelectrode recordings were filtered (band-stop, 48-52 Hz) and down 583 sampled at 300Hz. For the propagation analysis, the baseline of the signal was estimated as the mean 584 value of the signal until the positive deflection. We defined the onset of the CSD as the onset of the 585 negative shift and detected it using a threshold (Fig. S13a). We defined the WaveTime of each wave 586 as the mean time of the triggers detected in the 16 transistors and constructed a TimeLagMatrix 587 containing time lags for each channel computed with respect to the WaveTime (Fig. S13c). We 588 interpolated the known time lags with a thin-plate smoothing spline technique (Fig.5a). The velocity 589 of the propagation has been estimated computing the gradient of the TimeLagMatrix on the grid43. To 590 determine the direction of the waves, a vector starting at the point with higher negative delay (leader 591 of the propagation) and pointing to the one with the highest positive delay (follower of the 592 propagation) was transformed into polar coordinates to obtain the angle (Fig. S13b). For the 593 colormaps of Fig. 5b,c and Fig. 6b a bicubic interpolation was perfomed for visualization purposes. 594 Reference electrode 595 Voltages at drain and source terminals used to operate graphene transistors are referred to the 596 reference electrode. The reference electrode is generally grounded in anaesthetized subjects to ensure 597 stable recordings, since the subject is grounded at many points. However, the requirement of the 598 reference electrode to be grounded is not necessary; provided that the reference electrode is properly 599 positioned in a non-active location and does not have drifts and oscillations that interfere with the 600 recording, a proper operation of the graphene transistor is achieved. Importantly, gSGFETs are less 601 sensitive than microelectrode technology to the baseline drift associated with the reference electrode. 21 602 Commonly, baseline drift can lead to saturation of the amplifiers used for microelectrode DC- 603 coupled recordings, while the operation principle of graphene transistors does not lead to saturation. 604 The drift of the reference electrode shifts the biasing point which could lead to non-optimal 605 performance of gSGFETs. However, this can be easily solved by changing the transistor bias to the 606 new optimal value, which can be obtained from measuring an in vivo transfer curve. 607 608 Code Availability 609 Custom code developed for neurophysiological analysis of gSGFET signals is available at: 610 https://github.com/aguimera/PhyREC. 611 Data Availability 612 The experimental data that support the figures within this paper and other findings of this study can 613 be accessed by contacting the corresponding authors. Authors can make data available on request, 614 agreeing on data formats needed. 615 Methods References 616 37 Valdes, C. P. et al. Speckle contrast optical spectroscopy, a non-invasive, diffuse optical 617 method for measuring microvascular blood flow in tissue. Biomedical optics express 5, 2769- 618 2784 (2014). 619 38 Boas, D. A. & Dunn, A. K. Laser speckle contrast imaging in biomedical optics. Journal of 620 biomedical optics 15, 011109 (2010). 621 43 Capone, C. et al. Slow Waves in Cortical Slices: How Spontaneous Activity is Shaped by 622 Laminar Structure. Cerebral Cortex, 1-17 (2017). 623 46 de la Rosa, C. J. L. et al. Frame assisted H[sub 2]O electrolysis induced H[sub 2] bubbling 624 transfer of large area graphene grown by chemical vapor deposition on Cu. Applied Physics 625 Letters 102, 022101-022104 (2013). 626 47 Illa, X., Rebollo, B., Gabriel, G., Sánchez-Vives, M. V. & Villa, R. in SPIE 627 Microtechnologies. 951803-951803-951806 (International Society for Optics and Photonics). 628 48 Bandyopadhyay, R., Gittings, A., Suh, S., Dixon, P. & Durian, D. J. Speckle-visibility 629 spectroscopy: A tool to study time-varying dynamics. Review of scientific instruments 76, 630 093110 (2005). 631 22