A feasibility study of multi-site,intracellular recordings from mammalian neurons by extracellular gold mushroom-shaped microelectrodes

The development of multi-electrode array platforms for large scale recording of neurons is at the forefront of neuro-engineering research efforts. Recently we demonstrated, at the proof-of-concept level, a breakthrough neuron-microelectrode interface in which cultured Aplysia neurons tightly engulf gold mushroom-shaped microelectrodes (gMμEs). While maintaining their extracellular position, the gMμEs record synaptic- and action-potentials with characteristic features of intracellular recordings. Here we examined the feasibility of using gMμEs for intracellular recordings from mammalian neurons. To that end we experimentally examined the innate size limits of cultured rat hippocampal neurons to engulf gMμEs and measured the width of the “extracellular” cleft formed between the neurons and the gold surface. Using the experimental results we next analyzed the expected range of gMμEs-neuron electrical coupling coefficients. We estimated that sufficient electrical coupling levels to record attenuated synaptic- and action-potentials can be reached using the gMμE-neuron configuration. The definition of the engulfment limits of the gMμEs caps diameter at ≤2–2.5 μm and the estimated electrical coupling coefficients from the simulations pave the way for rational development and application of the gMμE based concept for in-cell recordings from mammalian neurons.

In recent studies our laboratory has developed a different approach dubbed "IN-CELL recording" in which micrometer-sized, extracellular gold mushroom-shaped microelectrodes (gMμ Es) record attenuated synaptic and action potentials with the characteristic features of intracellular recordings 8,[25][26][27][28][29][30] . In these proof-of-concept studies we demonstrated that cultured Aplysia neurons tightly engulf gMμ Es to form a high seal resistance (R s ). This, together with increased conductance of the neuronal membrane that faces the electrode (the junctional membrane -jm), makes it possible to record action potentials and subthreshold synaptic potentials with qualities and biophysics similar to perforated patch recordings 31 . Ultrastructural studies have revealed that various cell types including NIH/3T3, CHO, PC-12, H9C2, HL-1 cell lines as well as primary cultured rat hippocampal neurons also engulf mushroom shaped electrodes [25][26][27]30,[32][33][34] , thus suggesting that the cell-biological mechanisms leading to gMμ E engulfment may be ubiquitous.
Historically the micrometer size and shape of the gMμ Es used in our laboratory was selected by mimicking the dimensions and shape of the post synaptic spine structures that extend from the dendrites of vertebrate neurons 26,35 . Nevertheless, from a practical neuroengineering point of view, larger diameter gMμ Es are expected to provide better electrical coupling.
Although no studies have attempted to optimize IN-CELL recordings by increasing the size of gMμ Es, several laboratories have explored the potential use of nanometric size mushroom shaped electrodes 32,34,36 . These studies are based on the reasoning that nanometric sized mushroom-shaped microelectrodes may be more suitable for interfacing with small (10-20 μ m diameter) mammalian neurons than the very large Aplysia neurons (50-80 μ m diameter) used in our proof-of-concept studies. These studies showed that approximately 500 nm mushroom shaped protrusions are engulfed by cultured HL-1 cells and rat hippocampal neurons. Nevertheless, in these studies the recorded field potentials lacked the features of IN-CELL recordings. Rather, they were characterized by being similar to classical extracellular recordings of biphasic or monophasic negative field potential with small amplitudes of 100-200 μ V 32, 34 .
The above background implies that to improve the electrical coupling between small mammalian neurons and extracellular gMμ Es for noninvasive long-term intracellular recordings and stimulation, larger rather than smaller gMμ Es should be considered, or other characteristics of the gMμ E should be improved.
So far, no study has attempted to quantitatively estimate the optimal gMμ Es size to obtain the maximal range of electrophysiological signaling amplitudes generated by mammalian neurons. Optimization of the neuron-gMμ E coupling coefficient depends on three classes of parameters: (a) the innate cell biological mechanisms that limit the gMμ E-cap diameter that can be totally engulfed by a mammalian neuron. (b) The cleft width formed between the neuron's membrane and the surface of the gMμ E. (c) The size-and material-dependent electrical parameters of the gMμ E. In the present study we examined these questions. Based on the experimental results we quantitatively estimated the expected levels of mammalian neurons-gMμ Es coupling coefficients. The biological examination revealed that the size limits of the gMμ Es cap that can be effectively engulfed by hippocampal neurons is 2-2.5 μ m. Beyond this diameter the neurons can adhere to the upper surface of the mushrooms cap but fail to engulf it. Computer simulations of the neuron-gMμ E configuration, which took into account the limited size of the mushroom cap and various structural and electrical parameters, then provided the range of electrical coupling coefficient that can be expected from the mammalian neuron-gMμ E configuration. The findings presented here constitute a bioengineering framework for the rational design, development and application of gMμ Es based platforms for IN-CELL recordings from mammalian neurons.

Results
Structural interfacing between neurons and protruding gold mushroom-shaped microelectrodes. To define the largest gMμ E size that can be engulfed by mammalian neurons we cultured dissociated 17 day old rat embryonic (E17) hippocampal neurons 37 on matrices of gold mushroom-shaped protruding micro-structures (gMμ P) of different sizes. The E17 culturing procedure yields a culture enriched by neuronal cells with only a few glial cells and thus can be used to examine the ultrastructural relationships between neurons and gMμ Ps ( Supplementary Fig. 1). The neurons were cultured on three matrices made of small, medium and large gMμ Ps with cap diameters of 1.5-2, 3-3.5 and 4-5 μ m and stalk diameters of 1, 2 and 3 μ m, respectively (Fig. 1). Because the thickness of the cell body cytoplasm of cultured rat hippocampal neurons is in the range of 1-2 μ m and the fact that mechanical deformation of the nuclear envelop may alter gene expression 38,39 , we kept the height of the gMμ P at 1.5-2 μ m (1-1.3 stalk and 0.5-0.7 μ m cap heights, respectively). The center to center spacing between the micro-protrusions was adjusted to generate spaces of 8 μ m between the perimeters of the mushroom caps. This spacing was selected to increase the probability for thin sections prepared for transmission electron microscopic (TEM) observation to run through the gMμ P but still avoid inhibition of the electrode engulfment as a result of overly densely spaced microstructures 40,41 . The entire surface of the matrices was functionalized by PDL and laminin.
Scientific RepoRts | 5:14100 | DOi: 10.1038/srep14100 Scanning electron microscopy (SEM) of the cultures revealed that independent of the gMμ P cap diameter, cell bodies and neurites adhere to the flat substrate in between the microprotrusions and to the caps or stalks of the gMμ Ps (Fig. 2).
To study the effects of gMμ P size on the extent of their engulfment by the neurons, we characterized the neuron-gMμ P interfaces by measuring the thickness (width) of the cleft formed between the neuronal plasma membrane and the gMμ P surfaces (Figs. 3 and 4 and Supplementary Figs. 2-4). Since we were interested in examining how the size of the gMμ P caps affects its active engulfment by the neurons, only cell bodies and large neurites that formed at least a single discernible physical contact (0-10 nm cleft) with the protruding structure were included in the quantitative analysis (Figs. 3 and 4). Measurements were made at intervals of 50 nm, perpendicular to the surface of the electrodes. The data were clustered to represent different areas of the gMμ P: (a) the upper part of the mushroom cap that faces the junctional membrane of the neurons, (b) the mushroom stalk, (c) the substrate surface that corresponds to the diameter of the mushroom cap, and (d) the lower part of the mushroom cap that faces the substrate . When the cleft thickness between the gold surface and the cells exceeded 300 nm it was not included in the calculations of the average cleft size. The fraction (in percent) of gMμ P surface with a cleft thickness smaller than 300 nm served as the "engulfment-level" parameter ( Fig. 4).
Examination of the tight contact formed between the neuron membrane and the upper surface of the gMμ P cap revealed an identical ultrastructure independent of the cap diameter (Figs. 3 and 4 and Supplementary Figs. 2-4). The tight contacts appearing along stretches of 0.2-1 μ m were interposed by short clefts of 5-10 μ m. The fact that independent of the cap diameter, a 0-10 nm narrow cleft is formed between the cell's plasma membrane and the upper surface of the gMμ P suggests that the rough surface  of the cap ( Fig. 1) facilitates membrane adhesion to it and that this characteristic adhesion pattern is independent of mechanical forces associated with the engulfment of the gMμ P stalk or the flat substrate around it (but see discussion in Santoro et al., 2014 33 ). It is important to note that quantitative assessment of the extracellular cleft formed between the plasma membrane of living cells and an artificial substrate such as the gMμ Ps may reflect unknown levels of geometric artifacts (shrinking or expansion) induced by the chemical fixative used, and/or the dehydration and embedding processes 42,43 . A number of studies have attempted to quantify the extent of such artifacts, which has been estimated in 3D brain tissues and long-term cultures of hippocampal slices 43,44 . These studies indicated that the processing of tissues for TEM imaging induces tissue "shrinking " in the range of 5-17% which in absolute terms in this study was equivalent to ± 1-4 nm 44 . It is conceivable to assume that the extent of shrinkage artifacts induced by fixative perfusion, dehydration and/or embedding of a ~10 μ m thick monolayer of neurons is less than that of in vivo brain tissues and cultured brain slices. Although the TEM images prepared in our study did not show any signs of expansion or shrinking we estimated the possible quantitative consequences of such artifacts on the simulated neurons-gMμ E electrical coupling coefficients. For instance, for measurements representing 10% shrinkage of the cleft width, the calculated coupling coefficients should be corrected by reducing the estimated coupling coefficient by 5.5-9% for a neuron-gMμ E (with a cap diameter of 1.5-2.5 μ m) with homogeneous cleft widths in the range of 10-100 nm (as detailed later).
We propose that the different engulfment levels of small, medium and large gMμ Ps reflect the limited surface area and volume of the neuronal cell bodies and neurites. Unlike many cell types, including various cell lines, primary cardiomyocytes, 2-3 day old cultured mammalian primary neurons and the Aplysia neurons that have been used to study the structural and functional interfacing of cells with nano and micro-protruding structures [21][22][23][25][26][27]32,33,45 , the genetic blueprint of many CNS neurons allocate an order of magnitude larger fraction of the plasma membrane to the dendrites and axonal compartments and the rest to the relatively small cell body 46 . As a result, neuron cell bodies with limited cytoskeletal machinery, very limited cytoplasmic volume, a relative large nucleus and a small membrane surface area can only follow and engulf small gMμ Ps, but are unable to adapt their shape and dimensions to enwrap medium and large gMμ Ps.
In summary, the above observations demonstrate that gMμ Ps with cap diameters larger than 2-2.5 μ m are incompatible with the innate cell biological mechanisms that underlie the processes of gMμ P engulfment by hippocampal neurons. This innate neuronal property limits the maximal gMμ E cap dimensions that can be applied and thus define the physical limits of the maximal electrical coupling coefficient that can be expected from the gMμ E-neurons hybrid configurations. Based on the above experimental observations and using computer simulation we next analyzed the electrical coupling coefficients that can be expected to form between cultured hippocampal neurons and engulfed gMμ Es.
Estimate of the electrical coupling coefficient between gold mushroom-shaped microelectrodes and cultured neurons. The electrical coupling coefficient (CC) is defined as the ratio between the maximal voltage amplitude of a signal recorded by the device (electrode-amplifier system) and the voltage amplitude of the signal generated across the plasma membrane of a neuron. In the simulations presented below we took the shape, dimensions and electrical properties of the gold electrodes into account as well as the level of the electrode engulfment, the frequency of the electrical signals generated by the neurons, and the passive junctional membrane properties of the neuron.
The simulations were conducted using the SPICE simulation system (Tanner EDA v.15) of analog electrical circuits. The basic configurations of the equivalent electrical circuits used are shown in Fig. 5. The circuits depict the passive membrane properties of the neuron, the electrode, the amplifier and the cleft formed between the neuron and the electrode (Fig. 5). The neurons are grown in a conducting culture medium which is grounded by an Ag/AgCl electrode. In the model the neuron's surface area is subdivided into a non-junctional membrane (njm, red) that faces the grounded culture medium, and a junctional membrane (jm, blue) that faces the electrode. Each of these membrane compartments is represented by a resistor and capacitor in parallel R njm , C njm , R jm and C jm respectively. The cleft formed between the neuron and the electrode is represented by a resistor (the seal resistance-R s ). The electrode is represented by a resistor and capacitor (R e , C e respectively, see supplementary material for additional details). In the model the electrical signals generated by the neurons were simulated by voltage pulses fed into the cytosol (green) which is located in the analog circuit between the njm and jm (Fig. 5). Three neuronal signals that correspond to action potentials, synaptic potentials and slow membrane oscillations were used. For the simulation of the electrical coupling of action potentials we used a sine wave of 1,000 Hz. Synaptic potentials were simulated by a 100 Hz sine wave and slow membrane oscillations by 10 Hz sine waves.

The effects of shape and size of the gMμE on the coupling coefficients. We began by asking
to what extent the size of the gMμ E affects the neuron-electrode electrical CC. The simulated gMμ E was constructed of a 1 μ m high cylindrical stalk and a 0.5 μ m high mushroom cap, shaped like a half ellipsoid transected in the plane of the long diameter (Figs. 1 and 5). To better evaluate the importance of the detailed geometry of the gMμ E, we considered two modes of increasing the dimensions of the mushroom shaped microelectrode (Fig. 5b,c). In Model-A the diameter of the cylindrical stalk was maintained constant at 0.75 μ m (this diameter was selected as it can be fabricated using conventional lithography) while the diameter of the mushrooms cap was increased from 1.5 to 5 μ m. In Model-B the diameters of the cylindrical stalk and ellipse-shaped cap were increased concomitantly, but the cap's diameter was always kept 1 μ m larger than the stalk. This method of increasing the size mimicked the method used to fabricate the gMμ Ps in the biological experiments (Figs. 1-3). It should be noted that whereas the morphometric studies presented in the first part of the manuscript showed that hippocampal neurons cannot  Fig. 1). (d-g) Schematic drawings and analog electrical circuits of a gMμ E ((e), yellow) totally (d,e) and partially (f,g) engulfed by a neuron (green). Non-junctional membranenjm (red), junctional membrane -jm (blue), electrode -e (yellow), arrows in (f) indicate the surface of gMμ E exposed to the culture medium-R ef . engulf gMμ P with cap diameter > 2-2.5 μ m, we included gMμ Es with cap diameters ranging from 1.5 to 5 μ m in the simulation to illustrate the significant impact of the gMμ E dimensions on the expected CC.
To simplify, we began with simulations assuming that the neurons adhere to the entire surface of the gMμ E and to the flat substrate along an area defined by the projection of the mushroom cap onto the substrate, which corresponds to a 0.5 μ m ring around the base of the stalk (Fig. 5d). The width of the cleft between the electrode surface and the neuronal plasma membrane (d j ) was set to be homogeneous and equaled 10, 25 or 100 nm. The cleft resistivity, which corresponds to the cleft thickness, was calcu- (where ρ is the culture medium resistivity which equals 100 Ω cm). The simulations integrated seven parameters related to the dimensions of the gMμ Es and influence the CC. These were the surface areas of the jm and the electrode (e), the corresponding values of R jm , C jm ; R s ; R e and C e (See Materials and Methods and Supplementary Material). The simulations were conducted for mushroom-shaped electrodes with cap diameters of 1.5-5 μ m (Fig. 5a), and for junctional membranes with resistivities of 8 or 80 Ω cm 2 (see Materials and Methods) and a capacitance of 1 μ F/cm 2 . Figure 6a,b illustrate the CC as a function of the mushroom cap diameters for the two modes of gMμ E geometric growth. The simulations tested three different cleft thicknesses (d j ) of 10, 25 and 100 nm, using specific R jm of 80 Ω cm 2 and frequencies mimicking action potentials, synaptic potentials and endogenous membrane oscillations. In both gMμ E model geometries, the CC declined as the cleft thickness increased (Fig. 6a,b from left to right). Comparison of the range of CC values obtained by the two gMμ E models revealed that the detailed mushroom geometry affected the value of the expected CC (compare the CC values in Fig. 6a,b). Recall that the two models differed solely with regard to the diameters of the gMμ E stalk: in gMμ E Model-A the stalk diameter was maintained constant (0.75 μ m), whereas in Model-B the stalk diameter increased with increasing cap diameter to maintain a constant relationship of a cap diameter exceeding the stalk diameter by 1 μ m (see Supplementary Material). In general the comparison of the CC levels in models A and B (Fig. 6) revealed that: (a) in both models the larger the diameter of the Figure 6. The electrical coupling coefficient between neurons and gold mushroom shaped microelectrodes (gMμEs) as a function of the mushroom cap diameters. Two modes of size change were considered (Fig. 5b,c): in (a) Model A, the stalk diameter was kept constant while the cap diameter was increased; in (b) Model B, the diameter of the mushrooms cap and stalk increased, keeping the gMμ E cap diameter 1 μ m larger than the stalk. The simulations were conducted assuming a homogeneous membrane-gMμ E cleft thickness (d j ) of 10, 25 or 100 nm, for three impulse frequencies depicting membrane oscillations (10 Hz), synaptic potentials (100 Hz, both oscillations and synaptic potentials are depicted by a black curve) and action potentials (1 kHz, red) and for junctional membrane resistivity (R jm ) of 80 Ω cm 2 . All parameters related to the dimensional changes of the simulated gMμ Es (R jm , C jm ; R s ; R e , C e ) were integrated in the simulations using a specific membrane capacitance of 1 μ F/cm 2 . Since the first part of this study established that cell biological mechanisms limit the engulfment of gMμ Ps to a maximal cap diameter of 2-2.5 μ m, we focused the next paragraph on a description of the estimated CC values for gMμ E with the maximal cap diameters that can be effectively engulfed.
These simulations imply that the maximal CC that can be expected using favorable theoretical conditions of a totally engulfed gMμ E configuration, a cap diameter of 1.5-2.5 μ m, an R s produced by a homogeneous, very narrow 10 nm cleft and a R jm of 80 Ω cm 2 is at best 2.47% for the subthreshold potentials (mimicked by low frequencies of 1 and 100 Hz) and close to 2.2% for action potentials (mimicked by the 1 kHz frequency) (Fig. 6). These CC levels are sufficient for recordings of attenuated action potentials with amplitudes in the range of those reported by intracellular passive nanostructures 16,19,22,23 . Nevertheless, the CC levels of the slow frequencies are insufficient for recordings of synaptic potentials and membrane oscillations with source amplitudes in the range of 1-5 mV. With CC values of 2.47%, synaptic potentials of 1-5 mV will be attenuated to the noise levels of the system and below it.
As it is reasonable to assume that the innate cell biological mechanisms that limit hippocampal neurons from engulfing larger gMμ E then 2-2.5 μ m cannot be altered, and that a better seal resistance than that formed by a cleft of 10 nm cannot be achieved, we next examined the prospects of improving the CC by reducing the junctional membrane resistance.
The expected impact of reduced junctional membrane resistance on the electrical coupling coefficient. It should be noted that in the proof-of-concept experimental studies using the Aplysia neurons-gMμ E hybrid configuration our laboratory reported on the recording of action potentials with amplitudes reaching approximately 20 mV and synaptic potentials of ~2 mV. That is in these experiments the recorded signals were attenuated to approximately 25% of the input potentials 28,29 . To account for these high CC levels it was necessary to assume that R jm should have a lower value than that directly derived from the input resistance of a neuron and the fractional area that serves as the junctional membrane 28,29 . To account for the high CC levels obtained in the Hai et al. experiments 28,29 the value of R jm had to be reduced by a factor of 1,000 from 100 GΩ to 100 MΩ . It was suggested that the lower R jm may have been generated by the membrane curvature (formed around the gMμ E) which in turn increased the density of the ionic channels within the curved patch of the junctional membrane 47,48 .
In the simulations depicted in Fig. 6 we used an R jm value of 80 Ω cm 2 which corresponds to 1 GΩ for a gMμ E with a cap diameter and stalk diameters of 1.75 and 0.75 μ m respectively. To explore the range of R jm values that would enable the recording of subthreshold synaptic potentials and membrane oscillations from cultured hippocampal neurons we next examined (Fig. 7) the relationships between CC and R jm assuming a fully engulfed gMμ E configuration, a cleft thickness of 25 nm (a more realistic cleft than 10 nm) and an electrode cap diameter of 1.5 and 2.5 μ m (Model B). The simulation in Fig. 7 showed that for R jm of 100 MΩ (R jm resistivity of 8 Ω cm 2 ) the CC of the frequencies that mimicked synaptic potentials and membrane oscillations was still below 10% and thus could hardly couple a 1 mV high synaptic potential. Nevertheless, a small additional decrease in R jm from 100 MΩ to 50-80 MΩ was sufficient to increase the CC to the 10% level and thus to allow "in-cell" recordings of the electrophysiological signaling repertoire from 1 mV and above (inserts in Fig. 7). It is conceivable that such a decrease in R jm may be induced by membrane curvature as discussed by Hai et al. 28 , by chemical functionalization of the gMμ E with nano-pore forming molecules 49,50 , by electroporation 16,19,22,23 or by molecules that lead to recruitment of ionic channels 12,28 . The expected neuron-gMμE coupling coefficients of partially engulfed gMμE. One of the expected benefits of the use of a gMμ E is the improved source separation of the electrophysiological signaling with respect to classical large surface planar electrodes. Nevertheless, the transmission electron images described in the first part of this manuscript revealed that a single gMμ P may be contacted or partially engulfed by a number of neuronal elements (neurites or cell bodies Figs. 2,3c and 4a). Thus, we next estimated the CC formed between neurons or neurites and gMμ Es as a function of the "engulfment level" (the percentage of the electrode surface area directly in contact with the neuron) and thereby estimated the expected amplitudes that can be recorded by gMμ Es that are contacted by a number of neurons.
For this simulation we used an analog electrical circuit that depicted the fraction of the surface area to which a cell membrane was adhered by two parallel RC circuits (Fig. 5f,g). One circuit represented the fraction of the gMμ E in contact with the neuron's junctional membrane and the other represented the circuit in direct contact with the grounded culture medium. For the simulation the values of R jm , C jm , and R s were calculated to correspond to the fraction of the surface area that was in contact with the electrode (see Supplementary Material). The calculations of the CC as a function of the contacted surface area were made for gMμ E cap diameters of 1.5 and 2.5 μ m and stalk diameters of 0.5 and 1.5 μ m respectively, for different R jm resistivities at different frequencies. Since gold resistivity is orders of magnitude smaller than the resistances formed between the gold surface and the culture solution (~2.35 μ Ω cm) 51 , we simulated the occupied and free surface areas of the electrodes in an abstract manner without attempting to simulate the location of the contact between the neuron and the electrode.
The simulation revealed that the CC declined rapidly as a function of the electrode surface exposed to the bathing solution (Fig. 8). In fact, the model predicted that a junctional membrane resistivity value of 80 Ω cm 2 and contact area of ≤ 50% between a neuron and an electrode with a cleft of 25 nm does not enable recordings of any signals, since the estimated CC is as low as 0.001-0.002% (Fig. 8a). Under identical conditions, if the junctional membrane resistance was lowered to ≤ 8 Ω cm 2 , attenuated action potentials, but not synaptic potentials and membrane oscillation, could be detected (Fig. 8b).
This implies that in spite of its small dimension, a single gMμ E can record spike activity generated by a number of neurons that form a direct contact or partially engulf a gMμ E.
The prospects of using the neuron-gMμE configuration for Recording the basic electrophysiological repertoire generated by cultured neuronal network. Given that the noise level of MEA platforms is in the range of 20-40 μ V, CC values of approximately 10% for synaptic potentials and membrane oscillations (with amplitudes as small as 1 mV) and a CC of 0.5-1% for action potentials (with amplitudes in the range of 70-100 mV) may be sufficient to enable on-line acquisition of the basic electrophysiological signaling repertoire of cultured mammalian neurons.
We assume that whereas the innate cell biological limits of gMμ E engulfment and gMμ E-plasma membrane cleft width cannot be reduced, other parameters that influence the CC level could nevertheless be improved to reach CC levels of ≥ 10%. Specifically, the gMμ E engulfment levels, and junctional membrane conductance could be improved by the use of: (a) Engulfment promoting peptides as discussed by Hai et al. 28,29 ; (b) Pore forming molecules localized at the gMμ E caps 12,49,50 ; (c) Electroporation as described by number of groups 16,19,22,23 . Although the physical properties of the junctional membrane are dominant in defining the CC, lowering the gMμ E impedance could also contribute and improve the CC. This could be achieved by applying nanometric layers of electro-active materials such as conducting polymers, carbon nanotubes, graphene and hybrid organic-inorganic nanomaterials on the electrode surface 52 . Finally, the stray capacitance of the MEA system could be improved and would also improve the CC level. In summary, it is conceivable that the use of the extracellular gMμ E-neuron configuration could be used for recordings of the entire electrophysiological repertoire in the range of 1-1000 Hz with amplitudes above 1 mV.
The experimental results and simulations conducted in this study now make it possible and justifiable to proceed with the fabrication of gMμ Es-MEA with maximal dimensions that can be engulfed by cultured rat hippocampal neurons, and then experimentally validate that extracellular gMμ Es can record the basic electrophysiological repertoire of cultured mammalian neurons. Successful in vitro application of gMμ Es based MEA will pave the way to applying the technology for in vivo use. Aside from common technological issues that affect the use of all in vivo MEA-platforms, specific concerns related to the in vivo use of gMμ Es based MEA will have to be addressed. These include: (a) the mechanical stability of gMμ Es to withstand sheer forces during MEA-platform insertion into the brain tissue is not known and may have to be adjusted; (b) whereas under in vitro conditions the neurons come into initial Scientific RepoRts | 5:14100 | DOi: 10.1038/srep14100 contact with the gMμ Es through gravity, under in vivo conditions the initial neuron-gMμ E contact will need to depend on other mechanisms. For example, by attracting the neurons towards the gMμ Es by molecular signaling, or by chemical functionalization of the gMμ Es surface with molecules that stabilize neuron-electrode adhesion once random contacts are formed; (c) one crucial problem in the in vivo use of gMμ Es-MEA is the expected competition between glia and neurons for the engulfment of gMμ Es. Under in vitro conditions the problem can be dealt with by using protocols to prepare neuronal cultures with very low glia densities. Although a solution to the issue of foreign body encapsulation by glia under in vivo conditions has been extensively investigated by a large number of laboratories, an effective solution is still not available.

Materials and Methods
Fabrication of gold mushroom shaped micro protrusion matrices. Gold mushroom-shaped micro-protrusion matrices (gMμ P) were prepared on 200 μ m glass wafers (AF45 Schott Glass) by means of photolithography and electroplating techniques. Briefly, the wafers were coated with an Au layer (60 nm of thermal evaporation) on top of a 10 nm Ti adhesion layer (e-gun evaporation), spin-coated with Shipley photoresist S1813 (4,000 RPM) hard baked for 10 min at 120 °C. Next, the photoresist layer was exposed to UV using a photomask with 1 μ m holes with a pitch of 8 μ m, 2 μ m holes with a pitch of 10 μ m, and 3 μ m holes with a pitch of 12 μ m determining the stem diameter of the gMμ P. (Karl Suss MicroTec MA6 mask aligner, UV 365 nm W = 26 mW/cm 2 , exposure time: 2.7 s). Development was done using AZ726 for 35 s after which the developing agent was washed from the samples in de-ionized water. gMμ Ps were then electroplated using Neutronex gold plating solution at a current density of 0.2 A/cm 2 for 45 min. The photoresist layer was then stripped using acetone and IPA. The gMμ P matrices were attached to the bottom of plastic culture dishes using SylGard (Dow Corning) and dried for 48 h at 60 °C.  Note that the CC of a neuron that contact 50% of a gMμ E, with R jm of 80 Ω cm 2 (a), is too small to enable recording of any electrophysiological signals (dashed lines). On the other hand, the CC of a neuron that contacts 50% of a gMμ E, with R jm of 8 Ω cm 2 (b), may generate small action potentials with amplitude in the range of 100-200 μ V (dashed lines).

Surface functionalization.
Scientific RepoRts | 5:14100 | DOi: 10.1038/srep14100 Cell culture. Rat hippocampal neurons were obtained from 17 d old embryos, as described by Kaech and Banker 37 . Briefly, a pregnant WT (Sprague Dawley) female rat was deeply anesthetized with isoflurane, the embryos removed and decapitated. The hippocampus was removed and treated with papain (Sigma-Aldrich) for 45 min, and serially triturated. Cell density at plating was 250,000-500,000 cells/ml. Cells were seeded in attachment/seeding medium [Neurobasal medium, 5% FBS, 2% B27, 1% GlutaMAX (all from Life technologies), 1% Penicillin-Streptomycin Amphotericin B Solution (Biological Industries)] on gMμ P matrices (Fig. 1). 24 h after culturing, the seeding medium was replaced with serum-free maintenance/feeding medium (Neurobasal medium, 2% B27, 1% GlutaMAX, 1% Penicillin-Streptomycin Amphotericin B Solution). At 3 days in vitro (DIV) 2.5 μ M ara-c (Sigma-Aldrich) was added to prevent glial cell proliferation. Half of the maintenance medium was replaced every 3-5 days by astroglial conditioned medium (consisting of 1/2 astroglial conditioned medium and 1/2 feeding medium). Hippocampal cultured cells were kept at 37 °C in a humidified atmosphere of 5% CO 2 . Cultures were kept up to 7-21 DIV. All procedures and experiments were approved and performed in accordance with approved guidance by the Committee for Animal Experimentation at the Institute of Life Sciences of the Hebrew University of Jerusalem.
Immunohistochemistry and fluorescent microscopy. Cultured hippocampal cells were immunolabeled as previously described 30 (Supplementary Fig. 1). Briefly, samples were fixed by 4% paraformaldehyde (Sigma-Aldrich) in Hank's Balanced Salt Solution (HBSS, Biological Industries) for 30 min, washed with HBSS before membrane permeabilisation with 0.1% TritonX-100 (BDH Chemicals) in HBSS for 30 min. After washes with Tween 0.1% (J.T. Baker) in HBSS, cells were incubated for 1 h in blocking solution [BS, 2% chicken albumin (Sigma-Aldrich) in Tween 0.1%]. Then samples were incubated with primary antibodies in 1% BS overnight at 4 °C. Neurons were labeled for neuron-specific intermediate filaments with mouse anti neurofilament antibodies, and glial cells were labeled for glial fibrillary acidic protein (GFAP) with primary anti-GFAP rabbit monoclonal antibodies. The next day the samples were washed repeatedly with 0.1% Tween and incubated with secondary antibodies in 1% BS for 1 h: goat anti-mouse secondary antibodies conjugated to Cy2 (Jackson ImmunoResearch Laboratories, Inc), and goat anti-rabbit secondary antibodies conjugated to Cy3 (Life technology). Cells were counterstained with the nuclear marker DAPI (Sigma-Aldrich) for 1 h, at room temperature. Samples were washed with HBSS, and stored at 4 °C in anti-fade n-propylgallate (Sigma-Aldrich) solution in 50% glycerol until imaging. Confocal imaging of the immunolabeled cultures was done using the D-Eclipse C1 imaging system (Nikon) mounted on an Eclipse TE-2000 microscope (Nikon). Images were collected and processed using EZ-C1 software (Nikon). Scanning was done in sequential mode: red was excited with a 543 nm He-Ne laser and collected with 605 ± 75 band pass filter, green was excited with a 488 nm Argon laser and collected with a 515 ± 30 band pass filter, blue was excited with a 405 nm diode and collected with a 450 ± 35 band pass filter. Images were prepared using the open-source image analysis program ImageJ (NIH, USA) and Photoshop CS6.
Electron microscopy. For TEM analysis, cells cultured on gMμ P matrices were fixed, dehydrated and embedded in Agar 100 within the culturing dish as previously described 53 . Briefly, the neurons were fixed by 3% glutaraldehyde in a 0.1 M cacodylate buffer with a pH 7.4 for 1 h, at RT. The cells were then washed in a 0.1 M cold cacodylate buffer (pH = 7.4) (Agar Scientific, Stansted, UK). Post fixation was done with 1% osmium tetroxide (Next Chimica, Centurion, South Africa) and 0.6% K 3 Fe(CN) 6 for 1 h, at RT. The cells were then washed in a 0.1 M cold cacodylate buffer (pH = 7.4) (Agar Scientific, Stansted, UK). Dehydration was carried out through a series of increasing concentrations of ethanol solutions, and finally the neurons were embedded in Agar 100 (Agar Scientific). Then the glass and Ti layer were etched using 39% hydrofluoric acid (for approx. 0.5 h). The Au layer was etched by a diluted Au etcher (I 2 /KI/H 2 O), leaving the gold-spine structures intact. Thereafter, the agar block, including the cells, was re-embedded in Agar 100 in a flat mold. This doubly embedded preparation was then thin-sectioned.
Measurements of cleft width from TEM images were done digitally using the image analysis program ImageJ and Photoshop CS6 (Fig. 4, Supplementary Fig. 2-4). The cleft width was measured every 50 nm perpendicularly to the gMμ P surface. The significance of the differences between average cleft width values of the different areas was analyzed using student T tests and one-way ANOVAs.
For SEM analysis, cells were cultured for 6-7 d on gMμ P matrices, were fixed and dehydrated as described above. Prior to the critical dry process the cultures were dehydrated by fresh 100% EtOH for 30 min. Critical point drying was conducted in liquid CO 2 in a SAMDRI-PVT-3D apparatus (Tousimis, USA). Once dried the samples were sputtered with gold in an SPI-Module TM Sputter Coater Module (SPI Supplies, USA). Images were taken with an Extra High Resolution Scanning Electron Microscopy MagellanTM 400L using an accelerating voltage of 5 kV.
Computer simulation. Computer simulations were done using SPICE (Tanner EDA v.15), and the passive analog electrical circuit depicting a gMμ E interfaced with a neuron as shown in Fig. 6 8,28,29 . Calculations and graphs presentations were made using MATLAB (20014A). The main purpose of the simulations was to quantitatively characterize the relationships between the dimensions and shape of gMμ E and the CC levels between the electrodes and cultured rat hippocampal neurons. The simulated gMμ Es were constructed of ellipsoid-shaped caps with a constant height of 0.5 μ m, and an ellipsoid Scientific RepoRts | 5:14100 | DOi: 10.1038/srep14100 cap diameter ranging from 1.5 to 5 μ m (Fig. 5a). The cylindrical stalks of the gMμ E were always constructed of constant 1 μ m heights. Two modes of gMμ E models were considered (Fig. 5a,b): in Model A the stalk diameter was kept constant at 0.75 μ m while the diameter of the cap increased (Fig. 5a). In Model-B (Fig. 5b) the stalk diameter increased from 0.5 to 4 μ m while the ellipsoidal cap increased from 1.5 to 5 μ m. The detailed calculations of the gMμ E surface are given in the Supplementary Material (Supplementary paragraph 2.1 on gMμ E surface area calculations).
The resistance and capacitance of the gMμ E (R e , C e ), the value of the seal resistance between the electrode and the neurons (R s ) and the dimensions of the junctional membrane surface area, its resistance and capacitance were dependent on the surface area of the gMμ E (cap, stalk, and flat ring shaped area underneath the gMμ E cap to which the membrane adhered) and the engulfment level of the electrode by a neuron. These parameters were calculated in the following manner: gMμE resistance and capacitance. The impedance of the gMμ E itself is negligible with respect to the impedance created at the interface between the ionic solution and the gMμ E. For this reason the values of R e and C e used here refer to the resistance and capacity of the electrode/ionic solution interface. To define R e and C e for gMμ E of different sizes and shapes we calculated the values of the electrode resistivity (Ω cm 2 ) and the specific electrode capacitance (F/cm 2 ). This was done by first measuring the impedance of freshly fabricated gMμ Es with a cap diameter of 1.75 μ m, a height of 0.5 μ m and a stalk diameter of 0.75 μ m using an HP 4284A Precision RLC meter, at 1 KHz, at room temperature, in 0.9% NaCl solution, between individual gMμ Es and a counter Ag/AgCl electrode. The average electrode resistance was 3.5 MΩ and the capacitance was 5.1 PF (Supplementary material paragraph 2.2). These measurements were used to calculate an average electrode surface resistivity value of 0.28 Ω cm 2 and an average specific capacitance of µ 65 F cm 2 . For each size and shape of the gMμ E, the electrode resistance and capacitance was calculated relative to the surface area of a gMμ E with a measured resistance value as described above. Formally, where R e,0 -electrode resistance for a gMμ E with the above mentioned dimensions, c e -electrode capacitance, , C e 0 -electrode specific capacitance, S 0 -Surface area for this gMμ E, S -Surface area as a function of the dimensions. The electrode capacitance was calculated in a similar manner.
Calculation of the seal resistance (R s ). The seal resistance between the neuronal membrane and the gMμ E was calculated by integration of infinitesimal resistors connected in series along the cleft path (between the electrode and the cell membrane) from the center top of the gMμ E cap to its stalk-base and along a 0.5 μ m (ring shape) flat substrate that surrounds the stalk base (see Supplementary material paragraph 2.3 on the calculation of the seal resistance). For the calculations we assumed that the current generated by the neurons flowed through homogeneously distributed ion channels along the entire junctional plasma membrane (see Supplementary Fig. 5). Note that in earlier studies we assumed for the calculation of R s that the current flowed from a single point source at the center top of the gMμ E cap. Since the direction of the current flow in the present model was from the mushroom cap along the stalk, through the small flat ring-shaped region around the stalk, each infinitesimal resistor only affects the fraction of current generated "above" it. The effective contribution of the "infinitesimal resistors" to the seal resistance is therefore given by the naive calculation of resistance, multiplied by a weight function defined as the ratio between the electrode area above it and the total area of the electrode.
Using the calculated surface area of the gMμ E, we next calculated the corresponding values of: (a) the junctional membrane resistance and capacitance, (b) the seal resistance assuming different cleft thickness and the specific resistivity of the culture medium (100 Ω cm). For the simulations we used the following parameters: (1) the non-junctional membrane resistance of 100-250 MΩ (R njm ) has been measured in a large number of publications 54,55 . The values of the junctional membrane resistance (R jm ) were derived by dividing the resistivity of the non-junctional membrane by the junctional membrane surface area. These values correspond to plasma membrane resistances of 100 MΩ , 1 GΩ , and 100 GΩ for a gMμ E with a cap diameter of 1.75 μ m and a stalk diameter of 0.75 μ m. Based on the earlier results in Hai et al. 28,29 R njm was estimated two different values of membrane resistivity that represent three estimated values of channel densities in the plasma membrane which curves around the electrodes. These were defined as 8 Ω cm 2 (corresponding to 100 MΩ for R jm that faces a gMμ E with a cap diameter of 1.75 μ m and a stalk diameter of 0.75 μ m) and 80 Ω cm 2 (corresponding to a 1 GΩ R jm that faces a gMμ E with a cap diameter of 1.75 μ m. (3) The estimated junctional membrane capacitance (C jm ) was calculated for any given contact surface area between the simulated cells and the simulated gMμ E and the universal value of the specific membrane capacitance (1 μ F/cm 2 ). (4) The voltage pulses that simulated action potentials, synaptic potentials and membrane oscillations were delivered to the simulated neurons between the junctional (jm) and non-junctional membranes (njm). (5) An amplifier input capacitance of 8 pF and a resistance of 100 GΩ were used in all simulations.