Methodological standards and functional correlates of depth in vivo electrophysiological recordings in control rodents. A TASK1-WG3 report of the AES/ILAE Translational Task Force of the ILAE

Summary

This paper is a result of work of Translational Task Force of the ILAE Neurobiology Commission. The aim is to provide acceptable standards and interpretation of results of electrophysiological recordings in vivo in control rodents.

The recording of EEG from the scalp is an important component in the investigation of patients with epilepsy. These recordings are used in the diagnosis, classification and management of epilepsy (including localization as part of investigation for epilepsy surgery). From a translational perspective it is therefore important to define characteristics of the rodent EEG that reflect the human counterpart. However, there are some key differences between recording in humans and recording in rodents. In humans, extracranial (scalp) EEG is standardized so that all patients have similar electrode placement allowing consistency across recordings. This is useful particularly for diagnostic screening in order to ensure adequate coverage of the cortex. In rodents, recordings are usually intracranial and may be used for monitoring seizures and other epileptiform events, but they are more often used to record physiological electrical activity to address a specific hypothesis and therefore electrode placement is consistent within an experiment but may not be consistent across experiments or across laboratories. This is analogous to intracranial EEG studies in epilepsy patients undergoing evaluation for targeted surgical treatment to control seizures, where the goal is to delineate the seizure onset zone¹. In these cases, electrodes are placed either at the surface of the brain (grids, strips) or directly into the brain parenchyma (depth electrodes). Electrode placement is tailored on a per-patient basis, and may also be influenced by practice variations between epilepsy surgery centers. In the last decade there has been an explosion of analytical tools to examine in vivo physiology, and these tools are being applied to the study of epilepsy physiology in people as well as animals. It should be noted that while scalp or extracranial EEG, intracranial EEG, and depth local field potential recordings are all ostensibly field potential recordings, these terms are specific to each recording type and should not be used interchangeably. Specifically, the term “local field potential” is not appropriate for recordings made from clinical macro-electrodes and should be reserved for micro-electrodes that are more often utilized in research settings.

This manuscript is a product of the TASK1-WG3 working group of the Joint American Epilepsy Society (AES) and International League Against Epilepsy (ILAE) Translational Task Force of the ILAE intended to provide an experts’ opinion on best practices and terminologies for the use of depth electrophysiological recordings from the brain of rodents used as experimental controls in animal studies; while rats and mice are most commonly used as model organisms to study neurological and psychiatric conditions, guinea pig and hamsters are occasionally used for these applications too. Specifically, we wish to outline the tools and the analytical methods to provide the research community with an understanding of how a particular study can be performed and how the data might be analyzed. We also provide some recommendations that may help in the design and reporting of studies utilizing such recordings.

Types of Electrodes

Consistent with the diversity of questions that can be answered with in vivo depth electrode recordings, the choice of electrode used for such recordings should be tailored to the question. Scalp electrodes or needle electrodes in the scalp tend to be used only in anesthetized animals for questions regarding presence or absence of seizures and epileptiform activity. Most electrophysiological recordings in rodents are done with chronically-implanted electrodes in freely-moving animals using either wired or more recently wireless techniques.

Screw electrodes and surface flexible microelectrode arrays

While conventional scalp electrodes have been used in larger rodents with some success, small stainless steel screw electrodes placed epidurally in burr holes drilled into the skull are most often used for rats and mice. These electrodes only record from the surface of the brain and are therefore most useful for monitoring for seizure or other epileptiform activity in cortical brain regions⁸.

Single and double wire depth electrodes

Single wires are the simplest form of depth electrodes used for in vivo recording. 75–500 μm-thick nichrome, tungsten or stainless steel wires are often used. The shank of the electrode is isolated with a variety of materials, such as polyimide, glass, epoxylite, etc. These electrodes are typically sharpened by electrochemical methods with a taper angle ranging from 5° to 30°. Also suited are carbon fiber electrodes which see more multiunit activity than wire electrodes and can be covered with detectors for chemical compounds for example ATP, ACh etc. Metal wires are commercially available in large rolls. Stereotrodes can also be fashioned out of a twisted pair of single wires. These offer the ability to use local (bipolar) or common (referential) references. Wires are implanted stereotaxically under general anesthesia and may be housed in guide cannulas for implantation into deeper brain structures in order to prevent bending. Dental cement is usually used to permanently affix the implant to the skull for chronic recording. Another option is to use portable manipulator attached to the skull. In this case, the electrode tip can be advanced in the brain and approach firing neurons, enabling excellent signal to noise ratio for single unit recordings. Available are also electrodes for detection of ions, pH and oxygen as well as glucose. Amperometric or voltametric devices permit detection of different neurotransmitters such as dopamine, norepinephrine and serotonin.

Multi-array depth electrodes

While single wire electrodes can be useful for some questions, such as 24hr monitoring for the presence or absence of seizure or epileptiform activity, most often multi-array electrodes (MAEs) are used as they allow for greater spatial resolution of signals as well as recording of multiunit (occasionally single unit) activity. MAEs can be used in vivo or in vitro. Implantable in vivo MAEs fall into two major categories: microwire and silicone-based probes. Microwire MAEs often use electrodes consisting of four (tetrodes) or eight (octrodes) single wires twisted together. Tetrodes and octrodes are often used for recording local field potentials (LFPs) as well as multiunit activity. Under certain conditions², the multiunit recordings can be post-processed to identify one or more single units (e.g.- putative action potentials from single neurons). The wires used in this instance are often smaller than stereotrode or single wire electrodes, 25 μm or 12.5 μm -thick nichrome or platinum-iridium wires. Multiple smaller wires allow for better discrimination of putative action potential waveforms from individual neurons through triangulation. Silicon probes are often more expensive than microwire array electrodes, but have a large number of recording sites in a relatively small volume. These recording sites are customizable and precisely distributed along the probe, allowing for the determination of spatial relationships of these signals unlike microwire electrodes. Probes may have single contacts for LFP and multiunit activity recordings or may have tetrode contacts that allow for discrimination of single unit waveforms in addition to LFPs. Due to the current flow in extracellular space, which is directly responsible for the field potential generation, a generated dipole produced in structures with elongated neuronal elements (i.e pyramidal cells in neocortex or hippocampus) will produce different polarities of extracellular signals depending on the exact electrode localization relative to the dipole. An example: synchronous hyperpolarization of large number of cortical pyramidal neurons generates depth-positive but surface-negative waves of LFP.³ Therefore, the exact location of electrodes needs to be reported.

Multi-electrode arrays with linear arrangement of recording sites 50–100 microns apart can be utilized for laminar analysis of field potential profiles. Current source density analysis of laminar profiles obtained with multi-electrode arrays is used to study local network arrangement by localizing the depth location of the generators of field potentials. This type of analysis is largely utilized in cortical structures characterized by a highly laminar organization of intrinsic and extrinsic connectivity that generates sizeable and spatially circumscribed extracellular current sinks and sources.

Glass electrodes for extracellular single unit recordings

Extracellular single unit recording of putative action potentials, and extracellular staining of recorded neurons can be done with the use of glass micropipettes. The pipettes are pulled from glass capillaries (usually 1.5 mm outside diameter [OD]) and filled with high concentration electrolyte (usually NaCl, 0.5–1 M); if extracellular staining is envisaged, 1–1.5% of dye (neurobiotin, biocytin etc. or viruses) is added. The tips of pipettes have to be broken to external diameter about 0.5–2.5 μm (larger diameters might be required for viruses). This type of pipette can be rapidly advanced into the brain tissue until the target structure reached (i.e. thalamus, hippocampus, deep cortical layers, etc). The maximal step of pipette advancement should be set to a few microns (typically 4–10 μm). In rodents, cortical neurons from supragranular layers do not fire many action potentials. Therefore, typically the pipette is rapidly advanced to layer 4 or deeper and then, it is further advanced with 4–10 μm steps, until neuronal firing is detected. In order to approach the neuron more closely, the step of pipette advancement should be set to 1–4 μm. To ensure that the recording is from a single neuron, the amplitude of the extracellularly recorded action potential should be in the range of 1–5 mV. If the amplitude is lower, there is a chance that the recording signal arises from several closely located neurons; if the amplitude is higher, it may be due to the pipette exerting mechanical pressure on the recorded neuron, which will increase spontaneous firing frequency and possibly injuring the investigated neuron. On a rare occasion, when the pipette is either very close to the neuron or touches the neuronal membrane, the glass electrodes can allow juxtacelular or quasi-intracellular measurements that reflect intracellular dynamics of neuronal activities with evident barrages of synaptic events and positive, like in intracellular recordings, action potentials. The extracellular dye injection is done at the end of recording session. Positive direct current pulses (200–500 ms, 50% duty cycle) should be delivered to the recording pipette containing the dye, and the amplitude adjusted to evoke strong neuronal firing. Typical amplitude of current pulses is 2–10 nA. Two to five (up to 15) minutes are typically needed to provide good neuronal staining. The extracellular dye is taken up by one or more neurons in close proximity to the tip of the electrode.

In vivo intracellular whole-cell patch clamp recordings

If extracellular unit recordings provide information on timing of action potentials only (neuronal output), intracellular recordings (patch-clamp or sharp) provide additional information on spike timing and on synaptic input activities of the recorded neuron. Patch clamp techniques can be utilized in vivo in anesthetized, head-restrained or, less commonly, freely-moving animals. The patch electrode is generally made from thin borosilicate glass tubes that have been pulled to a resistance of 7–8 MΩ when filled with artificial intracellular solution surrounding an AgCl wire. This technique is often called blind patch, as visualization of the neuron to be recorded from is impossible in intact specimens and confirmation of neuronal contact is made indirectly through monitoring the change in electrode resistance as the electrode is advanced through the neuropil⁴. During long-lasting recordings, the artificial intracellular solution used to fill up pipettes diffuses into recorded neurons, which usually alters neuronal intrinsic excitability. These techniques typically offer much lower yield in terms of cell numbers than microwire or silicone probe recordings, but allow researchers to use either voltage- or current clamp techniques to measure neuronal currents/potentials intracellularly, as well as allowing for reconstruction of neurons being recorded from, therefore allowing for a greater level of confidence in determining which neuron subtype has been recorded from. Patch-clamp recordings are also possible in freely behaving animals. To do this, the animal has to be briefly anesthetized in order to patch a neuron, then the patch pipette is glued to the skull, the anesthesia is removed and the animal is released⁵. Miniature, head-mountable, mechanically stabilized recording devices have also been utilized for in vivo blind patch in freely moving animals⁶. The practical use of this technique is very limited because the success rate to obtain stable recordings is low, particularly over time intervals that would be needed for recording during complex behaviors, and the technical difficulty is very high. Visually guided patch-clamp recordings are also possible from superficial layer neurons in mice if they are combined with two-photon microscopy.

In vivo intracellular recordings with sharp electrodes

Sharp electrodes can be successfully used to perform intracellular recordings in anesthetized animals or non-anesthetized, head-restrained animals⁷. These electrodes are typically filled with 2–3 M potassium acetate or less often 0.5 M potassium methylsulfate. When pulled, the tips of recording pipettes are usually very thin (<<0.1 μm). In order to obtain acceptable electrodes with sharp tips after filling with electrolyte, they have to be either broken or beveled to a typical 30–120 MΩ resistance. Sharp electrodes are used to obtain voltage recordings in current clamp mode. However, single electrode voltage clamp is also possible using switched voltage clamp or an iterative optimization of current injected into a cell. Typically, the electrode is advanced into brain tissue with rapid steps of 3–5 μm. Penetration of a cell (neuron or glia) is expected when a voltage jumps to a value −60 / −70 mV. Immediately after impalement, some negative current (usually not exceeding 1 nA) needs to be injected into recorded neuron and after a period of stabilization (1 min or a few minutes) the injected hyperpolarizing current can be removed.

The main advantage of sharp electrodes over patch-clamp electrodes is that the same sharp electrode can be used to impale consecutively several neurons while patch electrode has to be changed after every attempt to make a patch, successful or not successful. Given that the trunks of sharp electrodes are significantly thinner than the trunks of patch electrodes, the overall damage to investigated tissue caused by sharp electrodes is significantly smaller than the one caused by patch electrodes. However, there is a bias to recordings from large elements while blind patching allows also recordings from mossy fiber terminals in cerebellum or hippocampus.

Practical considerations

• □Membrane potential vs. trans membrane potential. When measured against some reference, multiple brain regions reveal the presence of some form of field potential. The same field potential is present both inside and outside neurons. During normal brain oscillations the amplitude of local field potentials is around several microvolts and the biggest normal field potential events, sleep slow waves, typically do not exceed 1 mV. Therefore, the intracellular recordings are only slightly contaminated by the extracellular field. Amplifiers enabling intracellular recordings, which subtract extracellular field, are not available commercially. A more practical approach is to consider simultaneous extra- and intracellular DC recordings with two similar electrodes separated by only tens of microns and referenced to the same remote reference. To obtain information on transmembrane potential the extracellular signal has to be subtracted from the signal obtained with intracellular pipette.
• □Electrode impedance. The impedance of the electrodes themselves is an important consideration, particularly for single unit recordings. For this reason, electroplating is often used prior to implantation to reduce the impedance of microwire electrodes to between 100 and 400 kΩ. Electroplating, usually using gold solution, increases the signal-to-noise ratio and allows for recordings of very small amplitude biological signals such as those from single action potentials, which are generally 60–200 μV. Care should be taken while electroplating, as excessive gold plating results in connection between two wires in a tetrode or octrode. It is noteworthy that electroplating is not possible in human recordings due to ethical constraints.
• □Extracellular spikes. Extracellular recording of spikes provides information on timing of action potentials, which is the basis of communication between neurons.
• □
  Intracellular activities. Intracellular recordings provide information on membrane potential of a cell at any given moment of time, which is the basis of field potential (LFP/EEG) generation.

  Recommendation: The type of recording electrode should be driven by the experimental question. The detail of which electrode (electrode material, type of electrode and type of recording), its manufacture and details of its implantation (specific implantation coordinates and method for affixing to the skull, whether the electrode is moveable or not) should be reported in manuscripts.

Data acquisition

Due to the small amplitude of the signals, pre-amplification is necessary. This is usually accomplished using homemade unity gain preamplifiers, although commercially available preamplifiers also exist. The preamplifier is directly connected to the implant affixed to the head of the rodent. These preamplifiers amplify the signal prior to transmission through a cable attached to the acquisition device. Although custom-built in-house cables are often used, several commercially available cables exist for transmitting these biological signals to any of several commercially available acquisition devices that will amplify and digitize the signal. Rarely digitization adds additional noise to the true electrophysiological signal. Digitization requires quantization of the analog signal at some integer value (or quantizing step). This approximation process adds quantizing noise to the original signal. Minimization of this noise can be accomplished by decreasing the size of the quantizing step relative to the signal being sampled. In practice, pre-amplification of the signal to scale it to the extent that it fills a reasonable portion of the dynamic range of the digitizer. Digitizer devices are available with varying numbers of channels and quantizing step ranges. Commutators are extremely useful when recording from freely moving animals non-wirelessly. These units permit unrestrained movement of the animal in any environment while minimizing torque on the cable and implanted electrode assembly.

Sampling frequency in analog to digital conversion is an important consideration⁷. The Nyquist frequency dictates the minimum sampling rate needed to analyze a signal of a certain frequency, which is the double of the sinusoidal process. That is to say, in order to get an accurate measure of an oscillation (7–12 Hz), one should use a sampling frequency of at least 24 Hz. This is, of course, a minimum. High frequency components of an analog signal often introduce an error in digitalization process. In order to reduce this error an anti-alias filter, a low-pass filter is recommended. Because the brain waves are not purely sinusoidal, the Nyquist frequency should be dictated by the highest expected slope of the recorded process. Practically speaking, sampling rates of at least 3 times the upper limit of the frequency contents of interest should be used. The typical sampling rates are 3–10 times the maximum frequency of interest. As the sampling frequency increases, the file size will also increase, thus posing technical challenges for investigators interested in 24-hour monitoring of multiple animals in parallel. For LFPs, a minimum sampling frequency of 400 Hz is recommended. High gamma range oscillations, typically in the 80–200 Hz range but potentially up to 500 Hz, overlapping with the range of multiunit firing, may be detected depending on sampling rate. Activity in this range has received much recent attention in both animal and human epilepsy studies, where high frequency oscillations (HFOs) have been recorded with micro- and standard clinical electrodes from neocortex and hippocampus. The action potential duration of cortical fast-spiking cells (a type of interneuron) is 0.5 ms with a 0.15–0.2 ms rising time and 0.3–0.4 ms decay time. In order to accurately describe the waveform of the action potential rising time with a minimum of 3 samples, a 0.05 ms sampling period is needed. Therefore, for recordings intended for postprocessing to identify single units, a sampling frequency of at least 20 kHz should be used, to enable identification of individual putative action potentials from high-impedance microelectrodes. Higher sampling frequencies are highly preferred, with many commercially available systems allowing for sampling at 33 to 40 kHz. These higher sampling frequencies increase the number of data points collected and this in turn allows for superior spike sorting; however, as a trade-off data file sizes are larger, necessitating increased storage space. Practical advice: No high frequency devices should be present at the recording set up because they can interfere with recording system.

Filtering of the signal is also an important consideration. As with sampling frequency, the low and high pass filters should be determined on an experiment-to-experiment basis depending on the question to be answered. The frequency-response curves should be carefully assessed in each case. Typically, finite impulse response (FIR) filters are preferred for isolating high frequency activity, while the more computationally tractable infinite impulse response (IIR) filters suffice for low frequencies. Multitaper and wavelet analysis methods may also be considered, depending on the desired temporal and frequency resolution⁶. In general, for LFP extraction from most brain regions, low pass filtering of 200Hz should be sufficient to answer many questions in control animals. High pass filtering of 1Hz is often recommended due to artifacts introduced by slow cable movements (see below). To obtain correct information of the shape of slow waves or to record infraslow oscillation DC amplifiers and non-polarizing electrodes must be used. For single unit experiments, multiunit activity may be filtered from 500 Hz to 9 kHz, although the choice of corner frequencies varies substantially between studies.

Extracellular voltage potentials are always recorded as a difference between two potentials, whether those potentials are close in proximity (generally referred to as bipolar recording) or whether they are spatially disparate (referred to as monopolar or referential recording), selecting a reference is important. When recording LFPs in rodents, it is often best to select a common reference, generally an electrode placed into the space above the cerebellar parenchyma for this purpose. For single unit recordings, a local reference is best.

Practical considerations: Noise and artifact

Electrical noise from environmental electromagnetic fields can contaminate biological signal recordings. To minimize this, Faraday cages are often used for recordings from both freely moving and anesthetized animals. This can be accomplished using grounded sheets of wire mesh or by purchasing a commercially available faraday cage. Proper attention to grounding can also minimize electrical noise, e.g. ensuring ground paths are well-insulated and avoiding ground loops if multiple pieces of equipment are active. Other noise reduction methods include unplugging nearby electrical equipment, recording in locations with no large equipment nearby, such as refrigerators, and using radiofrequency (RF) shielded rooms. Acquisition systems with notch filters and line noise adaptation can reduce 50/60Hz noise as well. Muscle artifact due to sniffing or chewing can be a problem when freely moving animals are used. These artifacts are generally removed during post-processing; however, as removal is unlikely to be complete, visual review to select artifact-free segments for analysis may be required.

Recommendations: The results from electrophysiological recording can be dramatically biased by incorrect recording setups. Therefore, detail on the settings used for recording should be reported in manuscripts (specific pre-amplifiers and commutators used, sampling frequency, any filtering or post-processing used to remove artifact). Sampling frequency of three or more times the maximum frequency of interest should be used.

Types of recordings

A 24-hour video-EEG monitoring is the optimal choice for seizure detection in rodents, but this may pose considerable regulatory and technical challenges. Recording can be achieved using either tethered or wireless telemetry equipment. Tethered systems require the use of commutators to allow animals to move around their cage. Wireless systems have been developed with multi-channel capability and battery life of hours to days, eliminating the need for wired recording. These systems tend to be more expensive and increase surgical complexity by requiring subcutaneous or intraperitoneal implantation of battery packs. Telemetry systems generally have fewer channels, making multisite monitoring more difficult.

Recording in anesthetized animals is sometimes used to evaluate oscillatory activity as anesthetized animal recordings minimize noise and artifact. However, care should be used when choosing an anesthesia, as anesthetics can alter brain function. Typically, urethane, barbiturates, ketamine or combinations thereof are used for anesthesia alongside local anesthetics applied to the scalp. When choosing an anesthetic, duration of recording and whether or not the animal will be sacrificed after the recording should be considered. Urethane anesthesia is very long-acting and may maintain physiological brain activity better than shorter-acting anesthetics, such as ketamine that typically lasts less than an hour, or barbiturates lasting several hours. However, urethane anesthesia can generally be used only in terminal experiments, as it is carcinogenic and not approved for survival use by most Institutional Animal Care and Use Committees. Care should be taken to achieve a level of anesthesia sufficient to maintain suppression of withdrawal reflex to a tail or paw pinch. Breathing rate and body temperature should be monitored and maintained within normal limits for the duration of the recording as well. Caution should be taken for the use of gas anesthesia in epilepsy research. The main advantage of gas anesthesia is precise and nearly instantaneous control of its depth. When delivered, halothane, isoflurane and similar anesthetics prevent the development of seizure activities. However, after long surgeries with these anesthetics, rebound paroxysmal discharges can be often observed.

Recording of LFPs and multi-unit activity can also be done in animals that are either pellet chasing in an open field or during tasks that assess cognition. The aim of these types of recordings is to link electrophysiological parameters to cognitive performance. Some of the implications of these types of recordings are discussed in the sections below.

Tools for analysis

Visual analysis of biological signals is often used as a first-pass measure for eliminating periods with prominent noise or artifact, and is the gold standard for detecting relevant EEG features such as seizures^{9; 10}. However, for high-throughput processing, MATLAB (Mathworks, Natick, MA) or R (Free Software Foundation, Boston, MA) programs can be used for this purpose, keeping in mind that the sensitivity and specificity of automated methods may be limited due to variability of physiological/pathological processes. Signal processing on the LFP and single unit firing activity is also frequently achieved using these programs. Although some signal processing software packages are available for MATLAB and R, many laboratories also use code written in-house in order to address specific questions for which there are no standard tools.

In order to explore data in the frequency domain, Fourier transform or wavelet analyses are typically used to create power spectrograms across a range of frequencies (Fig. 1, a). The precise method selected generally depends on the desired time and frequency resolution. The sampling frequency and low pass filter settings determine the highest frequency that can be represented in the spectrum while the total duration of signal to be analyzed and the high pass filter settings determine the lowest frequency that can be represented in the spectrum. Short-time Fourier analysis can be used to monitor changes in frequency bands over time, and these changes can be related to behavior when the recordings are from awake, behaving animals. It is also of interest to examine LFP recorded from multiple electrodes simultaneously and to determine whether the signals are synchronous. Although LFP is known to be an imperfect index of unit activity, it is sometimes useful to examine areas sharing synchronized oscillatory activity. This is typically done using methods such as cross-correlation or coherence analyses¹¹. Because most types of oscillatory activities have rhythmicity, but are not purely sinusoidal, phase synchrony analysis might give deeper biological insights than regularly used analyses of time synchrony¹².

Fig. 1. Tools for analysis of electrophysiological signals.

Frequency information from local field potential recordings (a, top panel) can be extracted using a Fourier transform and plotted over time in a spectrogram (a, bottom panel). This allows for visualization of periods of theta (black circles in a) or other frequency oscillations of interest. Local field potentials recorded from each of four tetrode wires, shown as different color signals in (b), can be post-processed and filtered from ~500–9000 Hz to allow for detection of putative action potential waveforms in the signal (b; example marked with vertical line). Timestamps of putative action potentials are extracted and waveforms are “clustered” so that the activity from single neurons can be resolved (colors in c) based on unique waveform shapes on each tetrode wire.

Intracellular activities are usually analyzed with approaches similar to analysis of EEG/LFP signals. In addition, averaging is commonly used as analytical approach. The reference point for time-average analysis is usually either a particular time point of simultaneously recorded EEG/LFP signal (i.e. maximum of depth negative component of paroxysmal wave), or an intracellularly recorded event (i.e. time of half-amplitude of transition from hyperpolarized to depolarized state)¹³.

Post-processing of extracellular recordings to identify single units can be accomplished by first extracting multiunit activity from the raw traces by filtering in the desired range (typically >500Hz). The signal is then thresholded (usually >3× the RMS of the background) and putative action potential waveform spikes are then isolated from the signal (Fig. 1, b). Tetrode or other multichannel electrodes are used so that waveforms from single neurons are recorded on multiple electrodes, allowing confirmation of single neuron identification based on the waveform “signature” recorded on multiple electrodes with commercially available software (Fig. 1, c). Such software include packages like SpikeSort 3D (Neuralynx, Bozeman, MT), Offline Sorter (Plexon, Dallas, TX), Spike2 (CED, Cambridge, UK). Spike sorting is also possible with open source software such as Open Ephys. Candidate waveforms are generally discriminated based on maximum and minimum spike voltage, peak to trough amplitude of the spike, peak and trough time, shape of the spike, spike duration and principle component analysis of each spike. Wavelets (with or without Lillifors tests), independent component analysis and factor analysis are also commonly used for feature extraction.

Once the signals from individual neurons have been clustered and timestamps of each spike are extracted, firing frequency, inter-spike interval, and spike amplitudes and half-widths can be estimated. In addition, behaviorally regulated firing of single neurons can be examined. For example, spatially regulated firing of a subpopulation of pyramidal cells in the hippocampus called “place cells” is generally assessed using positional rate distribution maps. Such rate maps produce a graphical representation of where a neuron fires in space, usually as a normalized heat map of putative action potentials per unit area in an open arena. The total number of spikes per unit area (pixel) is divided by the amount of time the animal spent in this area. Putative pyramidal neurons are considered place cells if they have firing fields with at least nine contiguous pixels and firing rates above the mean for the whole session.

Recommendations: The type of recording and the tools for analysis are dependent upon the experimental question. However, biases and errors can occur if the recording and analysis are not appropriate and therefore detail of the methods (including specific signal processing algorithm and any new code used) should be reported in manuscripts.

Interpreting the data: Local field potentials and single units in information processing and cognition

In the mammalian brain, inputs into neural networks are processed in a way that that these networks collectively encode and retrieve information, including spatial representations and associative memories in order to guide decision making and behavioral response. The assessment of such networks, obviously, depends on the spatial sampling of the electrodes, which is not necessarily similar from one study to another, and even from one individual animal to another.

Neural networks can be investigated at a variety of scales from the level of the synapse, to the level of single neuron firing characteristics, moving up to the level of the networks of neurons producing oscillatory activity in the brain. Several mechanisms for the relationship between neuronal firing and information encoding have been proposed: rate coding, population coding and temporal coding. Although these will be described separately below, many of these processes are occurring simultaneously within the same network and should not be conceptualized as independent mechanisms with independent contributions to information processing and cognition.

Neocortical and thalamocortical oscillations

Throughout life cortical and thalamic neurons display and often generate different types of oscillations with frequencies from 1 cycle per minutes up to 200 Hz (Fig. 2). The types of oscillations depend on the brain states. Table 1 provides information on types of neocortical and thalamocortical oscillation, associated behavioral state and on their origin.

Fig. 2. Oscillations in the thalamocortical system.

a. Frequency band of oscillations recorded in the thalamocortical system. b. Above, a segment of local field potential recorded during slow-wave sleep in cat’s associative cortex. Below, Fast Fourier Transformation of the signal shown above. (a, from ref³⁵, b, unpublished observation).

Table 1.

Normal hippocampal, neocortical and thalamocortical oscillations.

Oscillation	Behavioral state	Known origin
Infra-slow (period seconds to minutes)	All states, including natural states and anesthesia.	Unknown, possibly metabolic.
Slow 0.1–1 Hz and Delta (1–4Hz)	Slow wave sleep, several forms of anesthesia (e.g., ketamine-xylazine, Urethane).	Intracortical, but thalamus actively contribute to synchronization
Theta rhythm (4–12 Hz)#	Exploration, REM sleep.	Septo-hippocampal network (unclear whether theta activity recorded in neocortex also originates in septo-hippocampal network). Theta rhythm in rodents is more pronounced that in other species.
Alpha rhythm (8–10 Hz)	In primates and carnivores dominates occipital lobe during wakefulness with eyes closed.	Unknown, but in vitro data suggest that thalamus has sufficient machinery to generate continuous alpha rhythm.
Spindles (10–16 Hz) Transient waxing and waning oscillations usually lasting 0.5–2 sec. Currently spindles are subdivided on slow (10–13 Hz) and fast (13–16 Hz)	Predominantly recorded during stage 2 sleep, can be recorded during deep sleep.	Spindles have been described as being generated in thalamus. However, several properties of slow spindles raise questions about involvement of other brain regions.
Sigma rhythm (10–15 Hz)	All states of vigilance, but stronger during slow-wave sleep.	Unknown. Sometimes the terms sleep spindles and sigma rhythm are used to describe the same phenomenon.
Mu rhythm (7–14 Hz) and associated beta rhythm (20 Hz)	Quiet wakefulness, over somatosensory cortex. Stops when the movement is present.	Unknown. Some suggest that this rhythm is generated by somatosensory thalamic nuclei.
Beta rhythm (16–29 Hz)	Waking and drowsiness; may also be accentuated by GABAergic medications e.g. benzodiazepines.	Likely cortical.
Gamma rhythm. Currently subdivided into low gamma (30–80 Hz) and high gamma (90–120 Hz)	Waking or modulated by sleep slow waves.	Likely local cortical networks, but can be synchronized with thalamic activities.
Neocortical ripples (140–200 Hz)	All states of vigilance, but different power: wake < REM < SWS < anesthesia < seizure onset zone.	Intracortical (for neocortical ripples), depend on gap-junctions.

^#The specific term “theta rhythm” in rodents is often used to describe a 4–12Hz frequency oscillation [therefore encompassed by both theta (4–7Hz) and alpha (8–12Hz) oscillations in humans and primates] of septo-hippocampal origin, which is often observed in control rodents during exploration and REM.

REM: rapid eye movement; SWS: slow wave sleep

Neuronal Oscillations in the Hippocampus

Oscillations in the hippocampus are extensively studied in the context of cognition in epilepsy and healthy control animals. LFP oscillations are mainly comprised of excitatory and inhibitory synaptic currents, calcium spikes and intrinsic currents and resonances from fluctuations in membrane potential through HCN and other channels¹⁴. Hippocampal LFPs are dominated by two main rhythms that reflect the collective activity of local populations of neurons: “theta rhythm” and gamma. In the rodent “theta rhythm”, a prominent rhythm between 4–12 Hz, encompassing what would be considered both theta and alpha oscillations in humans and primates, is seen particularly during exploratory movement and REM sleep in rodents¹⁵. Gamma rhythm, sometimes broken into low gamma from 30–90 Hz and high gamma from 90–120 Hz, usually occurs at the peak of theta¹⁶. Theta in the hippocampus is broadly broken down into type 1 and type 2: type 1 “atropine resistant” and type 2 “atropine sensitive”. Suppression of theta by inactivation of the medial septal inputs to the hippocampus induces deficits in working memory as well as profound spatial memory deficits.

Gamma frequency oscillations in the LFP also indicate information processing in rodents. It is believed that gamma rhythms provide a coordinating mechanism binding ensembles of neurons that encode information^{15; 16}. Gamma and theta are related in CA1 of the hippocampus, in that the phase of theta modulates the amplitude of gamma¹⁹. Inputs from the entorhinal cortex and from CA3 are coordinated at different gamma frequencies, each appearing at a different phase of theta²⁰. It is proposed that such coordination of theta-gamma oscillations plays a role in long-term plasticity in CA1 and specific aspects of information processing.

In contrast with theta rhythm, sharp wave ripples (SWRs) dominate the hippocampal LFP when the animal is immobile either during periods of quiet wakefulness or sleep²¹. The fact that SWRs occur only during non-REM, when theta is not present, lends further credence to the idea of competing mechanisms for the two oscillations leading to a SWR/theta dichotomy. Sharp wave ripples are very large amplitude deflections (sharp waves) followed by very high frequency (~200 Hz) oscillations (ripples) in the LFP. These large events represent hypersynchronous activation of populations of neurons in the hippocampus²⁰. Despite its seemingly hypersynchronous appearance, the temporal organization of single neuronal firing patterns within the SWR is sequentially organized such that it resembles reactivation of the neuronal firing patterns found during behavior in a compressed timescale. This observation led to the hypothesis that these events are involved in transfer of previous behavioral experiences from the hippocampus to the neocortex for long-term storage²³. This hypothesis is supported by evidence showing that disruption of SWRs interferes with memory. Patterns of neuronal activity in SWRs replay recently acquired and pre-existing information that may be used to influence decisions, plan actions and could underlie insight.

Taken together, all of these oscillations within the hippocampus are an indicator of information processing occurring within the hippocampus; coordination of theta and gamma oscillations between the hippocampus and other structures, such as the prefrontal cortex, indicates communication between structures. When oscillations in LFPs are locked either in phase or in amplitude, this is generally regarded as a period of information transfer from the one structure to the other. It is believed that this coherence in LFP oscillations of neuronal groups participating in the same function allows them to interact efficiently, because their communication windows are open at the same times. Evidence for such a communication scheme has been observed in multiple frequency bands, brain regions and cognitive processes²⁴.

Temporal Coding

As alluded to above, neuronal oscillations seem to be important indicators of temporal coding of neurons. Since a single synaptic input is not sufficient by itself to induce firing in a neuron, synaptic inputs need to involve large numbers of neurons firing at a high rate or to be precisely timed so that excitatory postsynaptic potentials produced at various locations in the dendritic tree can induce neuronal firing. It is proposed that neurons that fire synchronously participate in the same process and form, as an ensemble coding entity. Additionally, millisecond-range synchrony between neurons that occurs during neuronal oscillations plays a critical role in induction of synaptic plasticity, which is thought to underlie storage of information at both long and short timescales²⁵. Ostensibly, this is typically examined through phase-locking of single unit firing with specific frequencies in the LFP, however more broadly temporal coding can be conceptualized as any variation of neuronal firing that occurs in time.

Relationships between the precise timing of single neuron firing activity and the phase of theta oscillations in the LFP differ across cell types²⁶. Pyramidal cells preferentially fire at the trough of theta, and basket cells shortly after the peak. Place cells, and to some extent fast spiking interneurons, have a more subtle relationship to theta rhythm: these cells oscillate at a faster rate than the ongoing LFP theta, which produces action potentials that appear earlier in the theta phase on successive theta cycles as the rat crosses the cell’s place field²⁷. This process is called phase precession.

Rate coding

Rate coding is the process by which firing rates of single neurons encode information. Rate coding in hippocampal neurons is also thought to be crucial for cognition. The most obvious example of rate coding in the hippocampus is the hippocampal place cell, whose firing is modulated by the location of the rat in space²⁸. Firing rates of other neuronal populations are regulated by other aspects of the rodent’s environment, such as head direction cells, or are task-relevant such as those neurons that upregulate their firing during specific phases of a task. It is estimated that approximately 30 to 40% of hippocampal pyramidal cells have very well defined place fields, although likely all hippocampal pyramidal cells encode some spatial information with varying fidelity²⁹. Individual place cells encode specific areas in space with high specificity, fidelity and longevity. Together with the fact that hippocampal lesions abolish spatial memory in rodents, the place cell phenomenon leads to the hypothesis that place cell firing is important for spatial cognition.

Population coding

In population coding, information is encoded by the joint activation of an ensemble of neurons rather than individual neurons within this ensemble, as in rate coding. An example of population coding is the activity of primary motor cortex neurons in macaques trained to reach targets organized in a circle and visual discrimination primary visual cortex³⁰. While individual neurons preferentially increase their firing for multiple directions of movement, the vector sum of the activity of the ensemble of neurons uniquely predicted the direction of motion. Subsequent studies have also shown that population coding accurately predicts stimulus features and behaviors in multiple brain areas and species, including localization of a rat on the basis of place cell activity³¹. Population coding has a number of advantages, including the ability of a finite set of neurons to encode huge numbers of different stimuli simultaneously as individual neurons can have different, but overlapping input selectivity. Compared to rate coding alone, this can be done relatively quickly and is extremely sensitive to changes in inputs.

Relevance to human recordings

Animal model experiments are of critical importance in detailing the cellular processes underlying normal and pathological processes. This is facilitated by drawing parallels directly between animal model and clinical recordings. In some situations, investigations in epilepsy patients using newly available high resolution or microelectrode arrays can help to bridge the “bench to bedside” gap.

Currently, two types of microelectrode depth recording devices are FDA approved for human use: the “Utah” array³², which can be implanted onto the neocortical surface to a depth of 1.0 or 1.5 mm, and depth arrays with added microwires that are typically used to record deep structures such as the hippocampus, amygdala, and cingulate cortex. Increasingly, investigational devices capable of multiscale recording are available for use, for example the Ulbert laminar multielectrode³³ and the surface-recording “NeuroGrid”³⁴, that have been reported to detect single units. The data acquisition and analysis techniques and considerations described above translate well to these human recordings, although key differences in rodent vs human EEG patterns need to be kept in mind. The defined frequency ranges may differ between species, for example high gamma and hippocampal theta. Normal variant patterns are much better understood in human recordings than in rodents, as there is a wealth of clinical correlation of human EEG going back several decades. These topics are covered in depth in other sections.

Recommendations:

The value of animal models lies in the ability to conduct controlled experiments, obtain larger sample sizes of both seizures and subjects with less variability, record from sites not normally clinically accessible, and record with better spatial coverage and/or resolution than is available from clinical recordings. However, clear description of the methods and terminology used for in vivo recording allow for proper interpretation and replication of experiments.

Descriptions should include method and type of recording, brain location and/or cell type where the recording was performed, behavioral state of the animal, frequency band of interest and any pre- or post-processing done to the local field potential signal. Appropriate use of the methods outlined above facilitate the study of temporal, rate and population coding, which help probe systems-level mechanisms for cognition and behavior in the context of rodents with epilepsy.

Conclusions

Neuronal depth recording of local field potentials and single neuron waveforms is an extremely powerful tool to probe the neural networks. This information can be used to understand epilepsy and epileptogenesis, as well as aide in the understanding of cognition, both in physiological and pathophysiological circumstances. An ever-growing variety of tools are available for acquiring, analyzing and interpreting these data, including some that are commercially available and customizable.

Bullet point summary.

• ○Recommended types of electrodes and recording techniques used to answer specific experimental questions
• ○Recommendations to avoid biased interpretation of electrophysiological results
• ○Brief description of activities generated by corticothalamic system and hippocampal formation in control rodents
• ○Neuronal coding types and interpretation
• ○Relevance of results obtained in rodents to human electrophysiological recordings