Neural prostheses are devices that interface with the central or peripheral nervous system. They target at the capture, modulation or elicitation of neural activity, in most cases to record the information flow within a neural pathway for its online or later decoding, or to mimic or replace neural functionality that has been compromised or lost. While in theory any information carrying modality of the neuron could be tapped into (e.g., concentrations of neurotransmitters in the synaptic cleft, of energy carriers such as ATP or glucose, or of ions or oxygen; optical properties; ionic currents; membrane potential), most devices sample or alter the membrane potential (or a proportional quantity - thereto). Since the discovery and description of ‘body electricity’ by Luigi Galvani and Alessandro Volta in their studies on voltage-induced muscle contraction (Volta, 1793), metal electrodes have been used for establishing a bidirectional communication link between electrically excitable cells - and stimulation or recording apparatuses. From an engineering point of view, this has always been the least challenging and technologically least demanding approach. It just requires bringing a locally deinsulated conductor into close vicinity of a neuron to capture or induce fluctuations in its surrounding electrical field. Given the multitude of conductive materials to choose from and the ever growing number of technological possibilities for their structuring and processing into any desired shape and arrangement, historically, the interface of choice for neuroprosthetics has become the electrode array despite some of its fundamental conceptional shortcomings - . It has evolved into other clinically relevant readout derivatives such as electroencephalography (EEG), electromyography (EMG), electrocorticography (ECoG), and electroretinography (ERG). And it is still competitive with technologies making use of different recording or stimulation principles such as magnetic resonance imaging (MRI), magnetoencephalography (MEG), positron emission tomography (PET) or transcranial magnetic stimulation (TMS) (Clark, 1998; Lapitska et al., 2009). -
This chapter will review recent design trends for microelectrode arrays (MEAs) with an emphasis on flexible polymer devices, which may be exploited for neuroprosthetics. A brief synopsis on the history of in vitro and in vivo interfacing concepts with electrogenic cells introduces the chapter, followed by a discussion on their performance and limitations, to then take a look at latest strategies to overcome these limitations by resorting to new concepts, materials, and fabrication and modification processes. The chapter will conclude with the presentation and discussion of an innovative, versatile, and easy fabrication route for turning microchannel scaffolds into all-polymer neuroprosthetic electrode arrays. This strategy bears the potential of implementing a variety of secondary functionalities such as microfluidics, drug release schemes and optical stimulation paradigms, which may be operated in parallel to electrical recording and stimulation.
1.1. In vitro microelectrode arrays
To better understand the events at the cell-electrode interface, a variety of MEAs for the in vitro study of electrogenic cells have been developed over the past 40 years (Pine, 2006). Because they do not penetrate the cell membrane, they are considered ‘noninvasive’. They sample local fluctuations of the electrical field generated by the membrane potential. Thus, any change in membrane potential due to a local and selective flux of specific ions (mostly Na+ and K+) across the cell membrane will lead to a capacitively mediated shift of charges in a nearby conductor (Butt et al., 2003) or at the gate of a field effect transistors (FET) (Fromherz et al., 1991; Fromherz, 2006; Poghossian et al., 2009; Lambacher et al., 2011). While FETs are restricted to the sampling of these events, metals or semiconductors can also be used for actively modifying them. By charging the electrodes (or more generally speaking, the interface), such capacitive shift can thus be imposed onto the cell membrane - , which leads to a shift in its membrane potential and may result in the opening of voltage-gated ion channels (Bear et al., 2007). Such arrays are used to find physiologically ‘meaningful’ electrical communication parameters, to study the influence of electrode topography or (bio-)chemical functionalization on cellular events, and to characterize the physiologically induced changes of such interface over time. In most cases, they are compatible with a majority of microscopy techniques for simultaneous morphology studies or the optical screening of membrane potential-associated variables (e.g., by potential sensitive dyes or by imaging intrinsic changes of the optical properties of a cell, e.g., its refractive index) or activity-associated events (e.g., by calcium imaging). They are furthermore accessible to manipulation techniques such as drug delivery in pharmacological and toxicity assays (Gross et al., 1997; Johnstone et al., 2010), mechanical or laser microdissection in regeneration studies, or optical tweezers for biomechanical manipulation and force spectroscopy (Difato et al., 2011). While an in vitro system may not truthfully reproduce the conditions of an in vivo environment and the events therein, MEAs have nevertheless been widely adopted for screening studies. They have become a tool and test bed for better understanding the design requirements (e.g., material properties, coatings) of neural probes. Table 1 lists pioneering works and currently active groups or companies that have developed or commercialized key MEA technologies. Light-addressable devices were omitted (Bucher et al., 2001; Stein et al., 2004; Starovoytov et al., 2005; Suzurikawa et al., 2006). The terminology ‘passive devices’ refers to substrates with microelectrodes, tracks and connection pads that need to be connected to external amplification and signal processing hardware. ‘Active devices’ carry some of these electronics on-chip (Hierlemann et al., 2011). In ‘hybrid’ devices, MEAs and signal conditioning electronics are produced separately but packaged together into a standalone device. They can furthermore include other types of electrochemical sensors on-chip (e.g., for temperature, oxygen, pH, impedance,...).
1.2. From in vitro to in vivo
From a conceptual point of view, the readout and stimulation physics of in vitro electrode systems are identical to electrode-based in vivo probes. Also, the needs for amplification, filtering, analog-to-digital (A/D) signal conversion and signal (post-) processing electronics are almost the same. This holds for signal readout circuitry on the one hand, and for signal generators and digital-to-analog (D/A) converters for modulating neural activity with electrical stimuli on the other hand. It does not matter whether they are placed in direct vicinity of the electrodes as in recent ‘active’ MEA designs, or classically connected as modular hardware at the end of the electrode tracks of ‘passive’ MEAs. A comprehensive review by Jochum et al. surveys neural ampliﬁers with an emphasis on integrated circuit designs (Jochum et al., 2009).
Yet, in vivo, new challenges arise. The electrode array encounters a rather different and more complex environment when compared to in vitro scenarios. Any array cannot longer be considered ‘noninvasive’, even if it will not penetrate the cell membrane. This is because a MEA has to be brought first into its recording position, which involves the opening and partial removal of the skull and of any protective encapsulation of the brain. Deep brain implants need to be furthermore inserted into the soft, yet densely packed brain tissue, thereby, in the best case, just displacing, and in the worst case, even destroying neurons, connections and glia cells along the penetration path. This may lead to a partial destruction of the tissue architecture, which the brain needs to compensate for. Apart from the insertion damage, a rigid neural implant may get repositioned upon a sudden movement of the head due to the inertial forces acting on the quasi-floating brain. Furthermore, the immune system may become activated and attack the implanted portion of a device. The implant materials are thereby exposed to a variety of compounds not found in in vitro systems that they may chemically react with. In bad scenarios, the reaction products may be toxic, thereby triggering a temporary if not chronic immune response - . Any change of the device material may furthermore be categorized as ‘degradation’, which could compromise device stability and performance over time. The latter may also be simply lowered by device encapsulation in tissue scars, thereby electrically insulating the recording and stimulation sites. This scarring is considered as one of the most common reasons for device failure. Stimulation electrodes carry additional risks of tissue damage by their electrochemical erosion or overheating (Dowling, 2008; Marin & Fernandez, 2010). In summary, implant materials should be chosen that are sufficiently biostable against alterations or degradation by the physiological environment. They should furthermore not trigger any immune response or alter cell physiology in an uncontrolled and undesirable fashion - . In other words, a neuroprosthetic recording device should behave as if it were not present, and a stimulation device should in addition induce a neural response as similar to neural signaling mechanisms as possible.
Another critical issue is the actual interconnection of the devices to the outside world. Signals are commonly transferred to extracorporeal signal conditioning and processing electronics by cables. Not only are the connection points between cable and device a source of failure due to the detachment of the cable ends to the device by chemical degradation or mechanical forces. Once passing through the skin or skull, the pass-through hole has to be well sealed and stabilized to not let contaminants pass and become a site of chronic infection, and to not let the cables move and thereby exert mechanical stress onto the surrounding tissue. Recent trends therefore target at the transmission of signals through the skull by telemetric technologies, which, however, pose new challenges with respect to miniaturization, circuit protection against humidity, and energy transfer for powering the telemetric electronics. And it does not provide a solution for stabilizing the internal circuit-to-MEA connection, which may still be exposed to movement-related stress.
1.3. Electrode arrays for in vivo electrophysiology
For getting a general overview on in vivo neuroprosthetic - implants and brain-machine interfaces including the diverse types of microelectrode arrays, their specific applications and main vendors, the reader shall be referred to recent overview articles and reviews (Hetke & Anderson, 2002; Rutten, 2002; Navarro et al., 2005; Cheung, 2007; Winter et al., 2007; Dowling, 2008; Hajjhassan et al., 2008; Lebedev et al., 2008; Wise et al., 2008; Stieglitz et al., 2009; Graimann et al., 2010; Mussa-Ivaldi et al., 2010; Rothschild, 2010; Hassler et al., 2011). This paragraph will just summarize some of the basic design concepts, general fabrication approaches and the limitations they impose on the performance of chronically implanted devices.
1.3.1. Design and fabrication aspects
In the 80’s of the last century, the boom in microfabrication technologies opened the door for designing elaborate multi-microelectrode arrays with spatially distributed recording or stimulation sites. However, the choice of carrier, conductor and insulator materials depends on (and is thereby limited by) the often harsh fabrication and processing conditions. Given that materials in tissue-contact also need to fulfill the condition of being biocompatible (that is foremost, not being cytotoxic) and biostable (that is, not become degraded by the physiological environment), the number of suitable materials is quite low. Almost all of them are considerably more rigid than soft tissue, and the devices made from them tend to have sharp edges. Anyone ever having experienced a splinter in the thumb will remember how painful - it is each time the splinter moves only the tiniest bit. The simple reason is: the splinter is rigid and edgy whereas the tissue of the thumb is not. Thus, despite the tremendous research investments into diverse neuroprosthetic technologies, intracortical probes still lack functional stability during chronic use due to the large discrepancy between their biomechanical and chemical properties and those of the tissue environment (Marin & Fernandez, 2010).
The need for designing more flexible electrode arrays was already addressed in the 60’s of the last century (Rutledge & Duncan, 1966). Various strategies have been suggested since then to overcome some of the above mentioned limitations, particularly in the context of designing cochlear, retinal and deep brain implants. Today, the most commonly used flexible carrier and track insulation materials are polyimides (PIs), poly(p-xylylene) (PPX, and in particular poly(chloro-p-xylylene) (Parylene®-C)), poly(dimethylsiloxane) (PDMS), poly(tetrafluoroethylene) (PTFE), and occasionally less flexible liquid crystal polymers (LCPs) or photoresists (e.g., SU-8) (Navarro et al., 2005; Cheung, 2007; Myllymaa et al., 2009; Hassler et al., 2011). Noble metals - such as Pt, Ir, W and Au are sputter- or vapor-deposited, often requiring other metals (e.g., Ti, Cr) as adhesion promoters. They are structured into electrodes, tracks and connection pads by etching through a sacrificial mask of photoresist, glass or metals (Patrick et al., 2006; Wang et al., 2007; Mercanzini et al., 2008; Rodger et al., 2008). With few exceptions, each device needs to be fabricated in a clean room environment. Only recently, new strategies have been proposed to microstructure conductors under normal laboratory conditions that allow their transfer onto or embedding into polymer carriers that do not withstand high temperatures or vacuum. Hu et al. first sintered metal nanopowders within micropatterned, temperature-resistant multilevel microchannel quartz molds, which were then embedded into PDMS, parylene or polyimide. This allowed the generation of high-aspect (10:1) 3D conductors with, depending on the chosen mold generation process (ion milling, laser milling, deep reactive-ion etching (DRIE)), a wide variety of arbitrary shapes. A volume reduction between 5 % and 25% during sintering led to gaps between the conductor and the mold, which were filled with PDMS, thereby automatically insulating the tracks (Hu et al., 2006). A different approach was chosen by Dupas-Bruzek and coworkers, which was based on the UV (248 nm) laser-assisted activation of PDMS followed by the electroless deposition of Pt onto the laser-treated and thereby chemically modified surface areas (Dupas-Bruzek, Drean, et al., 2009; Dupas-Bruzek, Robbe, et al., 2009). Henle et al. created a PDMS-Pt-PDMS sandwich by placing a 12.5 µm thin Pt film directly onto a partially cured PDMS carrier substrate, structuring it by an IR (1064 nm) laser, manually discarding excess material and spin-coating a second PDMS layer on top (Henle et al., 2011). All of these fabrication approaches have in common that pads and electrodes get insulated during device fabrication and need to be reexposed in a post-processing step (e.g., by laser deinsulation or etching). Furthermore, tracks, pads and electrodes are always made from metals. Often, their long-term performance is limited due to the delamination of insulation layers or material fatigue over time. And finally, even such flexible neural implants still contain parts that are either more rigid than the surrounding tissue or lack the arbitrary deformability to follow its shape.
We therefore decided to deviate from common microelectrode array fabrication paradigms by resorting to a microchannel replication strategy with conductive polymers (CPs) completely replacing metals. This does not only allow the implementation of one and the same electrode layout in different types of insulating polymer backbones (e.g., of different shore hardness). It also gives more freedom in the choice of conductive materials that have biomechanical properties more similar to the embedding substrate. Fundamental proof-of-principle results have been published recently (Blau et al., 2011). The concept is sketched out in Fig. 1. It exemplarily illustrates the master fabrication and the replica-molding routes for in vitro polyMEAs with or without spike electrodes on the one hand, and copies of the master on the other hand. The initial master with bi-level microstructures for electrodes, conductor tracks and contact pads can be made out of two SU-8 layers following standard photolithography recipes - . A silicon wafer (m1) is spin-coated with SU-8 (m2), which is then soft-baked and illuminated through a photo mask (m3) to create the features of all three elements, the electrodes, tracks and pads. After a post-exposure bake (m4), a second SU-8 layer is spun on top of the first and soft-baked (m5), then exposed through the carefully aligned second photo mask with electrode and pad features only (m6). Thereafter follows a standard post-exposure bake (m7) and the development of both layers (m8). The individual layer thicknesses can be controlled by the spin coating parameters to define separate heights for buried track channels and for scaffold-penetrating electrodes and pad wells, respectively. Although the one-time fabrication of the molding master may still require a clean room infrastructure - , devices can be fabricated in a normal laboratory environment. In addition, a master can be replicated at negligible cost by the very same replica-molding procedure that is used to manufacture individual devices.
The bi-level master (1, grey, orange) (e.g., made of a high aspect ratio photoresist) is usually coated by an anti-adhesive film - before being filled with a curable polymer gel (2, sky blue) (e.g., PDMS). Alternatively, the master (or a sturdier copy thereof) can be used for hot-embossing into a thermomoldable polymer. A plane or micro-/nanostructured (zoom in) sacrificial carrier foil (green) levels the polymer with the highest protrusions of the master (f1). This ensures the complete penetration of the polymer by the elevated structural features of the master. After the curing of the thin polymer microchannel scaffold, such carrier helps in scaffold removal from the master without running the risk of tearing it apart. It furthermore allows the purposeful imprinting of a topofunctional texture into the PDMS surface and/or the electrodes of a polyMEA (e.g., for steering neural differentiation or guiding neurites). After flip over (f2), the carrier also temporarily supports the scaffold during the subsequent processing steps for generating a polyMEA with plane electrodes (route p). To create the electrodes, tracks and pads, the microchannels and cavities of the scaffold are filled with the CP (f3p). If the CP wets the channel-separating plateaus in this step, it can be rather easily scraped off after partial evaporation of the solvent. After curing, the scaffold is backside-insulated by a second polymer layer (f4p). If a foil or panel is chosen as the insulator and a film-forming CP as the conductor, the channels stay hollow and can serve for microfluidic or optical add-on applications. In this case, the sizes of the electrodes are less well defined. On the top-side of the polyMEA, they will have ring shapes with a through-hole diameter and wall thickness equaling the film thickness of the CP rather than being planar. However, a neuron could partially get into contact with the CP-coated channel sidewalls upon entering the channel or sending any of its processes into the channel. Thus, theoretically, the entire channel surface will serve as an electrode. Alternatively, all cavities are filled with the same or a similar curable polymer gel from which the scaffold was made of. Depending on the stickiness of the temporary carrier foil, the disk-like CP depositions on the carrier at the electrode sites and contact pads are more likely to be transferred to the polymer rather than sticking to the carrier. This is because a dried CP film usually has a nano-structured surface that gets physically entrapped by the polymer gel insulator. This issue is discussed in the next paragraph in more detail. After removing the carrier - , the polyMEA with planar electrodes (f5p) is ready for use. No other deinsulation step is required. If spike electrodes shall be generated instead (route s), the scaffold is transferred onto a carrier with cone-like indentations (red), which can be generated by e.g., anisotropic etching of Si (Jansen et al., 1996; Williams & Muller, 1996) or multidirectional UV lithography in photoresist (Yong-Kyu et al., 2006). After aligning those with the electrode through-holes of the scaffold, the CP can be filled into the cavities and cured (f3s). Rajaraman et al. recently presented a similar approach for pyramid-shaped metal electrodes (Rajaraman et al., 2011). As before, a backside insulation layer is applied (f4s) to give a polyMEA with spike electrodes after its peeling from the microstructured carrier. The spikes can either be used as such or, if a small recording/stimulation site is desired, be insulated by a thin coat of PDMS and then be clipped at their tops (e.g., by a razor-blade through a mask with holes) to expose the CP.
During the polyMEA fabrication steps sketched out in Fig. 1, the CP will always get into contact with the transfer carrier foil (green) at the pad and electrode openings in the scaffold. Depending on the surface structure and chemistry of the foil, film-forming CPs like PEDOT:PSS from aqueous dispersion might stick more or less well to it (Fig. 2, a1 & a2). Upon peeling the foil from hollow channels, the sub-µm-thick CP membrane spanning the electrode through-hole may get partially or completely torn off and/or stay stuck to the carrier foil (Fig. 2, a2, left channel). If, instead, CP-coated channels are backfilled with a polymer gel like PDMS, the gel will get mechanically entrapped in the CP film due to its roughness (on the nm scale) thereby acting as a mechanical support for the membrane after curing (Fig. 2, a2, right channel).
Both scenarios give useful devices for different application needs. A neuron may just grow its processes into a CP-coated channel without electrode membrane (Fig. 2, a3, left channel). Thus, the recorded signal can be easily attributed to that very neuron. Alternatively, if the well diameter were reduced to a few µm, a planar-patch-like recording array could be created. As recently reported by Klemic et al. (Klemic et al., 2005), Chen and Folch (Chen & Folch, 2006), Tonomura et al. (Tonomura et al., 2010), and Hofmann et al. (Hofmann et al., 2011), neurons tend to seal such microapertures to result in high signal-to-noise (S/N) ratios and selectivity. A back-filled channel with an intact electrode membrane (Fig. 2, a3, right channel) or a gPDMS composite (Fig. 2, b1-3) will act like a classical electrode instead. Even if the CP membrane were missing, the electrode would be defined by the nm-thick ring-shaped CP coating at the end of a channel. It is still an open question whether the impedance of these ring electrodes is sufficiently low to capture or induce neural signals.
1.3.2. PDMS as a soft and flexible substrate material
The platinum-catalyzed addition-crosslinking of vinyl-endblocked silicone polymers and silicone polymers with SiH functionality gives medical grade polydimethylsiloxane (PDMS) (e.g., Dow Corning, Sylgard 184; Wacker, Elastosil RT 601), a rubberlike polymer with particular properties (Colas & Curtis, 2004). Not only can a wide variety of different hardness and tear resistances be chosen from. It is furthermore dimensionally stable and will not shrink or expand upon curing. If not chemically activated (e.g., by plasma oxygenation or other types of surface functionalization (Donzel et al., 2001)), it reversibly adsorbs to almost any smooth surface but will not react with it. It demonstrates outstanding bioperformance and has a favorable toxicological profile in diverse medical contexts (Briquet et al., 1996). It is therefore FDA-approved and part of common implants that are in direct tissue contact (e.g., breast implants, contact lenses, tubing in heart surgery, catheters) (Colas, 2001; Curtis & Colas, 2004). It can furthermore be used as a molding material to cast itself - or various other materials such as plasters, concrete, wax, low-melt metal alloys (tin, pewter) or resins (urethane, epoxy or polyester) for master replication purposes (Smoothon, 2008). After depositing a conductive seed layer, it can also be electroplated to give LIGA-type master replicas (Jung et al., 2008).
1.3.3. Flexible, polymer-based electrode materials
The desire of incorporating electronics into bendable and stretchable lightweight consumer devices (e.g., rollable displays, garment, conformal solar panels) has driven the research and development of unconventional elastomeric conductors (Rogers et al., 2010) including inks based on single-walled carbon nanotubes (Sekitani et al., 2009) or silver nanoparticles (Ahn et al., 2009). Despite their lower electrical conductivity (Kahol et al., 2005), conductive polymers promise to become a low-cost alternative to metals for their flexibility and easier processability (Inganäs, 2010). Currently, PEDOT:PSS is one of the CPs with the highest conductivity, is transparent, forms bendable films, is chemically inert and non-cytotoxic. It has therefore found its way into the biosciences as a selective sensing layer in biosensors (Janata & Josowicz, 2003; Lange et al., 2008; Rozlosnik, 2009) and more recently as a feature enhancer of metal microelectrodes (Guimard et al., 2007; Widge et al., 2007). Besides the two conductivity modes (electron-hole and ionic) of CPs, their biophysical properties and process-dependent micro- and nano-topographies seem to enhance their bio-acceptance and tissue integration (Ateh et al., 2006; Guimard et al., 2007; Owens & Malliaras, 2010).
While graphite and, in particular, carbon black are usually not categorized as polymers, they share some of their properties with respect to their extended carbon backbone. For their high electrical conductivity, biological inertness, low price and easy handling, they are excellent filler materials for creating flexible, voluminous conductor tracks or coatings with silicones or polyurethanes as the matrix (Calixto et al., 2007; Huang et al., 2011). As with any conductive filler (e.g., antimony- or indium-doped tin oxide, silver (Ahn et al., 2009; Gong & Wen, 2009; Larmagnac et al., 2010; Zhang et al., 2011), carbon, or any form of their nanoparticle derivatives (Sekitani et al., 2009; Pavesi et al., 2011)), a conductive polymer can be generated when the percolation threshold - of the filler has been surpassed (Kirkpatrick, 1973; Milliken and Company, 1997). Rather than being a thin film conductor - such as PEDOT:PSS or most of the vapor-deposited metal electrodes, the composite fills the entire volume of the microchannel, thus being less prone to dramatic changes in its conductivity upon deformation - . As shown for other flexible wire composites (Ahn et al., 2009), it is hypothesized that the main reason for the stability in conductivity of such flexible wires is the disperse distribution of graphite flakes in the PDMS. Upon shifting these flakes against each other, the current will just follow an alternative or even newly established conductive pathway between different flakes being or getting into contact with each other. Thus, by filling channels of extended dimensions with gPDMS, the shortcoming of its lower conductivity compared to metals can be partially compensated by its bulk-like distribution.
1.3.4. Electrode functionalization and post-processing strategies
The postprocessing of devices serves two main goals:
the improvement of the electrical characteristics of the electrodes (mainly with respect to the decrease of their electrical impedances and/or the increase of their reversible charge delivery capacity (CDC) - , and
the enhancement of their biocompatibility for their better tissue integration.
The electrical impedance is a more general concept of electrical resistance; it describes the frequency-dependent resistance of an electrical conductor. At ‘0 Hz’, the alternating current (AC) impedance of an electrode is identical to its direct current (DC) resistance. Over the physiologically relevant frequency range between 0.5 - 100 Hz (relevant for slow oscillations as in local field potentials (LFPs)) and 1 - 5 kHz (for the capture of individual action potentials (APs) of neurons - ), the impedance of conductors can decrease by 2-3 orders of magnitude. In general, it can be stated that the smaller the geometrical electrode area (e.g., πr2 for disk-like electrodes with radius r), the higher the impedance. The impedance of a small-diameter electrode can be lowered by increasing its real (or effective) surface area, e.g., by adding a microtopography to the electrode (e.g., by increasing its roughness). Depending on the structure of that surface topography (e.g., fractal or columnar) and the type and electrical properties of the material that creates it, the capacitance - C of the electrode may be affected as well. Decreasing the impedance will increase the signal-to-noise ratio and decrease the required voltage U to charge the electrode-electrolyte interface. Increasing the capacitance of the electrode gives room for more charges Q to accumulate on the electrode surface at a given voltage (Q = C·U), thereby increasing the number of separated ionic charges at the electrode-electrolyte-tissue interface. Consequently, the increase in the locally generated electrical field may increase the likelihood of sufficiently shifting the membrane potential over its depolarization threshold in stimulation experiments - . The effect of decreasing electrode impedances by depositing non-planar metal layers (e.g., porous Pt-black, columnar TiN or fine-grained IrOx) has been exploited for a long time. A new trend is the deposition of carbon nanotubes (CNTs), which have favorable electrical and biophysical properties. They not only improve the impedance and the charge transfer capacity by more than one order of magnitude, but provide nano-textured anchoring points that cells can make use of (Keefer et al., 2008; Park et al., 2009; Malarkey & Parpura, 2010). Recent studies show the same effect for CPs. Ludwig et al. reported on an about 10-fold decrease in impedance at 1 kHz for electrochemically deposited PEDOT coatings on Ø 15 µm Au recording electrodes, which reduced noise levels by about half (Ludwig et al., 2011). In earlier studies, the same effect was demonstrated for polypyrrole (PPy) (Cui et al., 2001). Kotov et al. and Beattie et al. recently summarized the potential of nanomaterials for neural interfaces (Beattie et al., 2009; Kotov et al., 2009).
Many works have addressed the issue of providing signaling cues on electrode and substrate surfaces to mask the non-biological properties of a material, and to alleviate the acute and chronic disturbances imposed by a neuroprosthetic device onto its surrounding biological environment (Leach et al., 2010). By adsorbing polycations onto ready-made CP layers (Collazos-Castro et al., 2010) or by entrapping or intermingling cell adhesion and differentiation promoting proteins or their fragments (e.g., neural cell adhesion molecules (NCAMs), nerve growth factor (NGF), laminin and fibronectin) into the CP during its electrochemical or self-assembled monolayer (SAM) formation, it could be shown that brain-derived cells attached preferentially to the CP (Cui et al., 2001; Widge et al., 2007; Asplund et al., 2010; Green et al., 2010; Thompson et al., 2010). Recent studies indicate that, apart from its overall geometry and surface chemistry, the nano and micro surface texture of a substrate may have considerable impact on cell adhesion, differentiation, cell morphology and gene expression as well (Wilkinson, 2004; Barr et al., 2010). Depending on the type of starting material, the deposition parameters (temperature, pH, U, Q, time) and the choice of solvents and auxiliary components (such as surfactants), the form, texture and order of CPs can be fine-tuned (George et al., 2005; Yang et al., 2005; Abidian et al., 2010). And since Wong et al. observed that the shape and growth of (endothelial) cells could be noninvasively controlled by just switching the oxidation state of fibronectin-coated PPy (Wong et al., 1994), and that current flow through PPy would promote protein synthesis and neurite outgrowth (Schmidt et al., 1997), later works exploited this combination of conductivity and particular geometries of CPs for the programmable control of e.g., neurite extension, protein adsorption and cell adhesion, or for the spatially defined release of ions, antibiotics, anti-inflammatories, neurotransmitters and other signaling factors (Abidian et al., 2010; Ravichandran et al., 2010; Sirivisoot et al., 2011; Svennersten et al., 2011). While the roughening of electrodes and the incorporation of biofunctional cues in all cases provide mechanical and biochemical anchoring points for cells, CP coatings are lightweight and usually less brittle than metal deposits. Furthermore, unlike metals, both CNTs and CPs are chemically accessible to covalent pre- or post-modification with bioactive molecules (Ravichandran et al., 2010).
1.3.5. Performance of polyMEA devices
The electrical, mechanical and optical characteristics as well as the recording performance of in vitro polyMEAs and epidural in vivo probes derived therefrom have been presented in a recent proof-of-concept study (Blau et al., 2011). The main characteristics are summarized below. polyMEAs for in vitro applications were designed to fit the pin-layout of a commercial 60-channel amplifier system (Multi Channel Systems). Their overall dimensions of 4.9 x 4.9 cm2 matched commercial MEAs. Their minimal thickness could not stay below the height of the bilayer master features, which were between 200 and 500 µm. To reach the standard height of 1 mm of commercial MEAs, either temporary spacers or a permanently fused PDMS backside insulation coat were used. Device flexibility depended on device thickness. With increasing thickness above 200 µm, the PDMS still stayed flexible but slowly lost its surface conformability. A polyMEA with gPDMS electrodes, tracks and pads of approximate thickness of 300 µm is depicted in Fig. 3a. Electrode diameters of the 8 x 8 (- 4 corner electrodes) electrode matrix were nominally 80 µm. Electrode spacing was 400 µm with the exception of an 800 µm cross-shaped gap in the center of the electrode matrix. Device transparency of a polyMEA with PEDOT:PSS conductive films and PDMS-flooded cavities is demonstrated in Fig. 3b& c. At the edge of a cortico-hippocampal cell carpet, individual neurons and connections can be identified on top of electrodes and buried connection tracks alike. Between 1 Hz and 5 kHz, polyMEAs with PEDOT:PSS electrodes (Ø below 100 µm) had almost flat impedances of about 1.2 MΩ on average (Fig. 3d, blue dotted line). In contrast, gPDMS composite electrodes of same dimensions had significantly lower impedances of about 35 kΩ at 1 kHz with a logarithmic increase to about 15 MΩ at 1 Hz (Fig. 3d, black dashed line), which is typical for plain metal electrodes (Fig. 3d, orange line for Ø 30 µm TiN on Ti or on ITO electrodes). Despite their rather large electrode diameters and high impedances, PEDOT:PSS electrodes were able to capture action potentials and local field fluctuations alike. And despite of the larger noise floor between ~ 20 µV (with gPDMS counter electrode or Pt wire GND) and 50 - 80 µV (with internal PEDOT:PSS counter electrode), signal-to-noise ratios (S/N) were typically 5 for neural recordings and up to 100 for cardiomyocyte recordings. This was confirmed with cortico-hippocampal co-cultures (Fig. 3f), retinal whole mounts and acute cardiac tissue preparations (both not shown). Local field responses to visual stimuli in epicortical recordings from the visual cortex of mice were clearly visible after 59x averaging (Fig. 3g).
Judging from the low phase of the impedance at low frequencies, the electrodes of the polyMEA devices show mostly resistive behavior. As discussed above, stimulation electrodes are usually designed to have a highly capacitive character (Cogan, 2008). Theoretically, the polymer electrodes therefore may need to be altered in postprocessing steps to increase their capacitance. However, impedance spectra were taken in a non-physiological scenario. Several groups have already demonstrated that PEDOT-decorated electrodes are good and sufficiently stable performers in neural stimulation experiments with a CSC of 2.3 - 15 mC/cm2, which is already close to the 25 mC/cm2 reported for IrOx stimulation electrodes (Merrill et al., 2005; Cui & Zhou, 2007; Cogan, 2008; Wilks et al., 2009; Boretius et al., 2011).
We tested 80 µm PEDOT:PSS and gPDMS electrodes on dense carpets of cortical and hippocampal neurons from rats in cultured neural networks for their suitability as voltage-controlled stimulation electrodes without success - . It may very well be that polymer electrodes of same or different diameters will nevertheless perform sufficiently well as stimulation electrodes when resorting to charge-controlled stimulation protocols or different stimulation waveforms. It is furthermore likely that, apart from the dimensional and electrical properties of an electrode and the medium composition, the distance and the capacitive and dielectric properties of the lipid double layer of the cell membrane have a larger than thought impact on the CSC - . If true, the stimulation performance of an electrode should depend on the type of cellular environment it is tested in.
1.3.6. Interconnection technologies
Connecting a microelectrode array to any kind of electronics is a critical issue. The mechanical clamping of contact pads requires a mechanism that is difficult to miniaturize and which might simply detach. Classical (wire) bonding introduces materials with limited flexibility that, due to stress and/or the contact chemistry between the different materials (including the humidity absorbed by a packaging compound), become the location of corrosion and break. By design, polyMEA arrays may alleviate the bonding issue. When the microchannel tracks are filled with elastomeric gPDMS (or alternatively with hydrogel conductors (Guiseppi-Elie, 2010)), they may be bent and twisted to a large degree. This property, together with the liberty to shape channels with increasing volumes downstream of the electrodes, allows for integrating ribbon cable-type wiring to external electronics as part of the electrode array. Thus, the point of connection may be placed wherever the mechanical stress onto the connector may be least. In one experimental approach, the gPDMS connection pads at the end of a supracortical prototype array were simply slipped into a standard dual row connector during their temporary depression (Fig. 4). Upon pressure release, the rubbery gPDMS was wedged by the pins, thus ensuring proper seating and electrical connectivity. And because the thickness of the back insulation layer of an in vivo polyMEA can be applied non-uniformly - (e.g., thin at the recording site, thick at the connector site), the contact pressure can be tuned to fit different connector types. The concept resembles that of zebra strip connectors. After insertion, probe and connector can be fixated by sewing and be encapsulated by PDMS or epoxies.
Because long, and in particular, high-impedance wires may act like antennas which tend to pick up noise from the environment, they are usually avoided. Instead, high-to-low impedance conversion electronics are placed as closely to the electrodes as possible (as in ‘active’ MEA devices). However, today, the rigidity of any type of conversion electronics would still compromise device flexibility. Therefore, proper shielding (like in coaxial cables) remains the only alternative. Also in this case, PEDOT:PSS or gPDMS may substitute graphite-based conductive lacquers to create a mechanically more flexible, tightly device-wrapping shielding. When graphite is mixed into non-cured PDMS, the viscosity of the paste, once it has reached the desired conductivity, can be decreased temporarily with solvents such as iso-propanol. The external surface - of a device can thus be painted with such slurry gPDMS mix, which is then cured at elevated temperatures (80 °C – 150 °C). During the curing, the solvent will evaporate, resulting in a homogeneous conductive coating. It can be insulated by an upper coat of non-conductive PDMS. If a spot of the conductive gPDMS is kept deinsulated, it can serve as a counter electrode as part of the gPDMS shield. And, if necessary, the impedance of the gPDMS spot can be further tuned by electrochemical deposition of other conductors.
1.3.8. Other observations
For some not-yet understood and investigated reasons, PDMS seems to provide a more favorable surface for cell and tissue adhesion than other common culture substrates. In two acute slice experiments (retinal whole mounts) it was observed that, after an initial weigh-down by a nylon-stocking ensheathed platinum U-wire for enhancing the tissue electrode contact, the weight could be removed after 30 minutes without compromising the signal quality. While the PDMS and the electrodes had been coated with poly-D-lysine and laminin in a standard procedure for enhancing cell adhesion on MEAs, this type of stickiness had never been observed in our lab with insulation layers made of silicon nitride, silicon dioxide (glass) or polymer (photo-) resists such as SU-8 or polyimide. A similar observation has been reported by Guo et al. (Guo et al., 2010). There, however, the improved contact to the tissue surface was mainly attributed to the geometries of the conical well encasing of the electrodes. At this point, it is not yet clear to what degree the short O2 plasma hydrophilization - of the PDMS (for the better wetting with the adhesion mediators) contributed to its enhanced tissue interaction by not only altering the surface chemistry but also its morphology (Cvelbar et al., 2003). A process-related micro- or nano-texturing of the PDMS surface may have played a role as well (Barr et al., 2010). However, the most likely cause may be the tendency of silicones to absorb lipids (e.g., from the cell membrane) resulting in partial cell or debris fusion with the PDMS surface and its dimensional swelling (Mchenry et al., 1970; Colas & Curtis, 2004).
Preliminary results indicate that PEDOT:PSS can be embedded into a polymer matrix made of polyvinyl alcohol (PVA), glycerin and a di- or tricarboxylic acid as a crosslinker to render the PVA insoluble. However, this composite of high transparency and largely uncompromised conductivity will slightly swell in an aqueous environment. While a change in device geometries within the body is generally undesirable, a slight swelling of a polymer and/or its (hydrogel) electrodes may actually be favorable to enhance the electrode-tissue contact after device insertion. The water-uptake of PDMS itself is very low (below 1%). However, this might be just sufficient to stabilize device position within the tissue.
1.3.9. Open issues
As mentioned before, PDMS is very hydrophobic. Therefore, it is not wettable by aqueous or polar dispersions of organic conductors. Oxygen plasma treatment will render the PDMS surface hydrophilic by creating and exposing hydroxyl, carboxyl and peroxide groups on its surface, though. Depending on the storage conditions, this hydrophilicity is temporally more or less stable (Donzel et al., 2001). Under ambient conditions, it will usually degrade rapidly after the first 30 minutes. It can be anticipated that with shrinking channel feature sizes, the presented method of filling these channels (by coating the entire scaffold backside with the CP dispersion and then scraping it from the plateaus after the partial evaporation of the solvent) may not necessarily work well anymore. However, by playing with the two extremes of wettability, a channel-only plasma treatment may solve the problem. By temporarily covering the polyMEA with two adhesive sheets on both sides during plasma exposure, only the channel walls will be hydrophilized. After sheet removal and upon spreading the dispersion onto the scaffold, it will self-distribute within the channels only. For each channel geometry, finding the proper plasma parameters (power, frequency, pressure, temperature, time) might not be trivial, though.
While the softness and flexibility of all-polymer MEAs is one of their main assets, they have one major drawback: a MEA will not be easy to insert into dense tissue. A removable insertion device may alleviate this problem, though. During device fabrication, stiffer insertion and guidance aids could be embedded into the polymer microchannel scaffold such as anti-stick-coated polymer or glass fibers, which would then be withdrawn once the polyMEA is in its final position.
The used type of PDMS already breaks after 100% tear. Softer and more stretchable polyurethanes or silicones are available that would render the devices even more flexible and tear-resistant. However, most of them have still to be tested for their biocompatibility. And some of them are only milky translucent, which make them less suitable for concurrent cell imaging studies.
1.3.10. Optional strategies and future directions
The flexibility of PDMS can be exploited for fabricating spherically bent neural probes that may become useful as retinal implants or for electroretinogram recordings as described e.g., by Rodger et al. for parylene-based platinum electrode arrays (Rodger et al., 2008). While the current master production gives microstructures on a plane, a copy thereof may take on any shape. It only requires the placing of a PDMS scaffold onto the inverted shape of the desired topography during master reproduction in epoxy. The bending of the PDMS slab will certainly distort some of the microchannel features. But as long as the curvatures are not too sharp, electrode geometries will not be compromised. The concept is sketched out in Fig. 5.
Similar to classical soft-lithography, the microchannels could also serve for the assisted transfer of conductive patterns onto other carrier substrates (Fig. 6). When placed onto a (nano-porous) carrier (e.g., a (filter) membrane) (Fig. 6-1), the channels may be filled with any kind of colloidal conductor material (Fig. 6-2). Upon applying a vacuum (underneath the membrane), the solvent will evaporate thereby leaving a conductor pattern on the membrane (Fig. 6-3). By filling the electrode and pad cavities with a sacrificial paste (e.g., wax) (Fig. 6-4), removing the microchannel scaffold (Fig. 6-5), spin coating an insulation layer on the top (and bottom) side of the carrier (Fig. 6-6) and removing the sacrificial plugs from electrodes and pads in a final step (Fig. 6-7), a CP electrode array with design features similar to that reported by Guo et al. can be generated in the most straightforward and cost-efficient fashion (Guo et al., 2010).
PEDOT:PSS can be purchased as an inkjet-compatible formulation. Thus, the filling of the microchannels with the conductor could not only be further automated, but different thicknesses or blends be deposited in different regions.
Alternatively, after the local laser-assisted alteration of the PDMS channel surfaces and the autocatalytic deposition of a Pt priming layer (Dupas-Bruzek, Drean, et al., 2009), EDOT or other precursors could be polymerized electrochemically to give electroconductive electrodes and tracks. The microchannels and the PDMS itself could be furthermore exploited in controlled drug-release strategies (Fig. 7b) (Colas, 2001; Musick et al., 2009). Various neural implant design studies with included microfluidics have already been reported (e.g., (Metz et al., 2004; Suzuki et al., 2004; Seidl & Et al., 2010)) and their benefit been discussed recently (Musienko et al., 2009). Although the stable coupling of the microchannels to outside fluidics might be possible (e.g., by making use of multilayer bonding concepts (Zhang et al., 2010) or reversible mechanical, pressure- or vacuum-assisted interconnection strategies (Chen & Pan, 2011)), without doubt it will be even more challenging than the design of fail-proof electrical connectors as discussed above.
With or without taking advantage of microfluidic connectors, neural processes could grow into the microchannels (Fig. 7b, left) as demonstrated by various groups (Morin et al., 2005; Claverol-Tinture et al., 2007; Benmerah et al., 2009; Lacour et al., 2010). This would increase the likelihood of identifying the actual origin of the bioelectrical signals. Alternatively, the preloading of empty or CP-coated microchannels with slow-release (electroconductive) hydrogels carrying diverse drugs could be pursued (Fig. 7a), thereby steering neural differentiation, regeneration and activity with growth or signaling factors, alleviating probe insertion damage by antibiotics and anti-inflammatory drugs, or attenuating the formation of glial scars by mitotic inhibitors (Peppas et al., 2006; Guiseppi-Elie, 2010). In that case, the electrical conductivity of the gel should be sufficiently high to warrant the coupling between the neuron and the conductive PEDOT:PSS film covering the microchannel walls. Alternatively or in combination, the PDMS itself could be loaded with drugs that are either soluble in PDMS or stored in porous cavities in a local silicone co-formulation. Delivering organic drugs through polymeric microchannels bears the risk of undesirable dissolution and accumulation of the compounds within the polymer over time, though. PDMS is particular prone to absorb e.g., organic solvents (Lee et al., 2003). In consequence, the absorption and release kinetics would have to be tested for each substance. While this absorption behavior could be a disadvantage in acute studies, it may become advantageous for chronic drug delivery where a slow and constant release of a drug is desired. In combination with the local deposition of photoelectric polymers, light-mediated electrical stimulation sites could be created (Fig. 7c) (Antognazza et al., 2009; Facchetti, 2010; Ghezzi et al., 2011). On a similar line, optical fibers or waveguides could be embedded into these channels for the light stimulation of optogenetic probes (Fig. 7d) (Im et al., 2011). Neural activity from light-responsive neurons could thus be recorded from the PEDOT:PSS electrodes at the end of PEDOT:PSS-coated channels upon their optical stimulation through the very same channel.
Neuroprosthetic devices should mimic as best as possible the tissue they are placed into. The tissue would then accept them as its own or just ignore them. They should furthermore be chemically, mechanically and functionally time-invariant for uncompromised performance. Fabricating electrode arrays exclusively from soft polymeric materials may be one step into that direction. The innovative concept of filling bi- or multi-level microchannel electrode array scaffolds with polymer conductors opens several new routes for designing and fabricating neuroprosthetic devices not only on the laboratory bench, but also through existing replica mass production schemes (e.g., mold-injection, hot-embossing). In contrast to metal MEAs (Sadleir et al., 2005), polymer conductors may turn out to be compatible with computer tomography (CT) and magnetic resonance imaging (MRI), minimizing or avoiding image artifacts (Chen & Wiscombe, 1998; Flanders & Schwartz, 2008). Given the vast choice of insulation and conductor materials, device properties and performance can be tuned in a multitude of ways. While thin coatings of the microchannel walls with PEDOT:PSS lead to largely transparent devices suitable for combined electrophysiological and microscopy in vitro studies, currently, film stability with respect to bending and stretching is still limited. Nevertheless, depending on the fabrication approach, electrodes can be shaped as ring electrodes or classical area electrodes. Alternatively, channels can be filled entirely with rubber-like gPDMS or other conductive PDMS or PU composites, rendering the devices excellently stable to bending and stretching. If both transparency and flexibility are required, a hybrid strategy of combining PEDOT:PSS electrodes with gPDMS tracks and connector pads may be chosen. While the presented results refer to proof-of-concept studies with polyMEAs still having rather large electrode and track dimensions, there is no conceptual hurdle that prevents their further miniaturization. It can thus be foreseen that the presented polyMEA concept heralds a new generation of implantable neuroprosthetic electrode arrays.