If you don't remember your password, you can reset it by entering your email address and clicking the Reset Password button. You will then receive an email that contains a secure link for resetting your password
If the address matches a valid account an email will be sent to __email__ with instructions for resetting your password
1 These authors have equally contributed to this work.
Julio C. Hernandez-Pavon
Correspondence
Corresponding author. Legs + Walking Lab, Shirley Ryan AbilityLab (Formerly, The Rehabilitation Institute of Chicago), 355 E Erie St, 60611, Chicago, IL, USA.
Footnotes
1 These authors have equally contributed to this work.
Affiliations
Department of Physical Medicine and Rehabilitation, Feinberg School of Medicine, Northwestern University, Chicago, IL, USALegs + Walking Lab, Shirley Ryan AbilityLab, Chicago, IL, USACenter for Brain Stimulation, Shirley Ryan AbilityLab, Chicago, IL, USA
Neuroimaging Center (NIC), Focus Program Translational Neuroscience (FTN), Johannes Gutenberg University Medical Center, GermanyLeibniz Institute for Resilience Research (LIR), Mainz, Germany
Center for Mind/Brain Sciences - CIMeC, University of Trento, Rovereto, TN, ItalyDepartment of Neurology & Stroke, University of Tübingen, Tübingen, Germany
Department of Technical Physics, University of Eastern Finland, Kuopio, FinlandDepartment of Clinical Neurophysiology, Kuopio University Hospital, Kuopio, Finland
Department of Neuroscience and Biomedical Engineering, Aalto University, Espoo, FinlandBioMag Laboratory, HUS Medical Imaging Center, Helsinki University Hospital, Helsinki University and Aalto University School of Science, Helsinki, Finland
Department of Neuroscience and Biomedical Engineering, Aalto University, Espoo, FinlandBioMag Laboratory, HUS Medical Imaging Center, Helsinki University Hospital, Helsinki University and Aalto University School of Science, Helsinki, Finland
Department of Neuroscience and Biomedical Engineering, Aalto University, Espoo, FinlandBioMag Laboratory, HUS Medical Imaging Center, Helsinki University Hospital, Helsinki University and Aalto University School of Science, Helsinki, Finland
Department of Clinical and Movement Neurosciences, UCL Queen Square Institute of Neurology, University College London, London, United KingdomDepartment of Medical Sciences and Public Health, University of Cagliari, Cagliari, Italy
University of Adelaide, Adelaide, AustraliaSouth Australian Health and Medical Research Institute, Adelaide, AustraliaMonash University, Melbourne, Australia
Berenson-Allen Center for Noninvasive Brain Stimulation, Department of Neurology, Beth Israel Deaconess Medical Center, Harvard Medical School, Boston, MA, USA
Danish Research Centre for Magnetic Resonance, Centre for Functional and Diagnostic Imaging and Research, Copenhagen University Hospital - Amager and Hvidovre, Copenhagen, DenmarkDepartment of Neurology, Copenhagen University Hospital Bispebjerg and Frederiksberg, Copenhagen, DenmarkDepartment of Clinical Medicine, Faculty of Health and Medical Sciences, University of Copenhagen, Copenhagen, Denmark
Department of Neurology & Stroke, University of Tübingen, Tübingen, GermanyTemerty Centre for Therapeutic Brain Intervention, Centre for Addiction and Mental Health, Toronto, CanadaHertie Institute for Clinical Brain Research, University of Tübingen, Tübingen, GermanyDepartment of Psychiatry, University of Toronto, Toronto, Canada
Department of Neurology & Stroke, University of Tübingen, Tübingen, GermanyHertie Institute for Clinical Brain Research, University of Tübingen, Tübingen, Germany
Department of Neuroscience and Biomedical Engineering, Aalto University, Espoo, FinlandBioMag Laboratory, HUS Medical Imaging Center, Helsinki University Hospital, Helsinki University and Aalto University School of Science, Helsinki, Finland
TMS−EEG is a powerful technique for basic research and clinical applications.
•
The methodological combination of TMS−EEG is challenging.
•
The lack of standardization may affect reproducibility and limit the comparability of results produced across groups.
•
This article covers all aspects that should be considered in TMS−EEG experiments.
•
We provide methodological recommendations for effective TMS−EEG recordings and analysis.
Abstract
Transcranial magnetic stimulation (TMS) evokes neuronal activity in the targeted cortex and connected brain regions. The evoked brain response can be measured with electroencephalography (EEG). TMS combined with simultaneous EEG (TMS−EEG) is widely used for studying cortical reactivity and connectivity at high spatiotemporal resolution. Methodologically, the combination of TMS with EEG is challenging, and there are many open questions in the field. Different TMS−EEG equipment and approaches for data collection and analysis are used. The lack of standardization may affect reproducibility and limit the comparability of results produced in different research laboratories. In addition, there is controversy about the extent to which auditory and somatosensory inputs contribute to transcranially evoked EEG. This review provides a guide for researchers who wish to use TMS−EEG to study the reactivity of the human cortex. A worldwide panel of experts working on TMS−EEG covered all aspects that should be considered in TMS−EEG experiments, providing methodological recommendations (when possible) for effective TMS−EEG recordings and analysis. The panel identified and discussed the challenges of the technique, particularly regarding recording procedures, artifact correction, analysis, and interpretation of the transcranial evoked potentials (TEPs). Therefore, this work offers an extensive overview of TMS−EEG methodology and thus may promote standardization of experimental and computational procedures across groups.
]. However, the technique was not yet ready for broader use as the recorded cortical response was obscured by the TMS-induced electromagnetic artifact. A few years had to pass before the electromagnetic artifact problem was partially solved. In 1996, the first successful TMS−EEG study (published by Ilmoniemi et al. [
]) demonstrated the feasibility of the combination to record cortical excitability and connectivity. After these first successful recordings, the interest in using EEG to measure brain activation elicited by TMS has steadily increased. Consequently, this has opened new possibilities in basic and clinical research as noted in a recent review [
], multiple approaches to recording and analyzing the TMS−EEG data have been developed, and there is still no consensus on how to standardize the procedures for TMS−EEG preparation, data acquisition, and analysis. This article aims to review the state of the art in the field and provide, when possible, recommendations for successful TMS−EEG studies to eventually improve the reproducibility of experimental and analysis procedures across laboratories. We aim to share our expertise with the community, based on published data and personal experience. We have gathered several leading TMS–EEG experts, hoping to promote clarification of concepts, improvement of our practices, guidance for newcomers, and identification and addressing of open questions in the field.
1.1 Electrophysiological aspects of TMS–EEG
1.1.1 TMS
TMS excites axons in the brain via inductive electromagnetic stimulation. A strong, very brief, magnetic field is delivered to the brain via a transducing coil. The changing magnetic field induces a time-varying electric field (E-field) in the cortex. Depending on the orientation of the E-field with respect to the geometry of the cortex and cortical neurons, the E-field leads to a depolarization of axons in the stimulated brain area. Depending on the level of depolarization, action potentials may be triggered [
]. Trans-synaptic activation of neurons on which the excited axons impinge will induce postsynaptic currents in the dendritic arbor of cortical pyramidal neurons at the target site. Postsynaptic potentials are subject to summation spatially and/or temporally. If the summation is large enough and involves a sufficiently large area of the cortex, the postsynaptic currents will result in a measurable EEG signal. At the same time, the spread of activation along pyramidal neurons causes a secondary excitation or inhibition of connected subcortical structures and cortical brain regions. The temporospatial summation of postsynaptic currents in the dendritic arbor of pyramidal or other cells in connected cortical areas may also cause a measurable EEG signal, contributing to the transcranially evoked EEG response.
TMS is based on electromagnetic induction, described by Faraday's law. A TMS pulse is initiated by flowing an intense current (∼5 kA) through the TMS coil windings. This current produces a time-varying magnetic field that penetrates the scalp and skull unimpeded, inducing an E-field. The brain is a conductor; therefore, eddy currents (i.e., currents that circulate in closed loops and in opposite directions than the currents in the TMS coil) are induced in the brain that can depolarize neurons, producing neuronal firing. TMS is thought to activate cortical neurons that have axonal bends or other geometrical inhomogeneities or endings in the induced E-field, as the E-field along neurites changes most rapidly at these locations [
Chapter 1 background physics for magnetic stimulation. Supplements to clinical neurophysiology transcranial magnetic stimulation and transcranial direct current stimulation, proceedings of the 2nd international transcranial magnetic stimulation (TMS) and transcranial direct current stimulation (tDCS).
]. The strength of the magnetic pulse is in the order of 1–3 T, with a rise time of about 50–100 μs. Because of the short pulse duration, the temporal resolution of TMS is sub-milliseconds, which allows for real-time modulation of the brain. The spatial extent of the cortical area stimulated by TMS depends on the coil geometry, stimulus intensity, target area, and, therefore, coil-to-cortex distance [
]. As magnetic fields attenuate rapidly with distance and as the induced E-field approaches zero at the center of the head, TMS stimulates superficial cortical layers more strongly than deeper layers. However, the induced neuronal activity depends also on other aspects (like the position and orientation of neuronal structures and membrane characteristics). In summary, besides stimulating the target area and surrounding tissues, TMS indirectly activates synaptically interconnected sites, a feature exploited in brain connectivity studies [
]. When the stimulation intensity (SI) is adequate, locally evoked action potentials may propagate along anatomical connections across cortical layers within the same cortical column and to other cortical and subcortical regions (e.g., Ref. [
Fig. 1Chain of events triggered by the TMS pulse. (1–2) A current pulse flows through the TMS coil (max I ∼ 5 kA) and produces a brief (∼100 μs) but strong magnetic field (max B ∼ 1–3 T). (3) The changing magnetic field induces an E-field (∼50–100 V/m) in the brain which in turn (4) produces a flow of electric current in the tissue (∼0.1 mA/mm2) (5) The flow of current (i.e., ions). produces local membrane depolarization (>∼10 mV). (6) Voltage-gated ion channels are opened and (7) action potentials are generated in axons where depolarization reaches the firing threshold. (8) Neurotransmitters are released in the synaptic cleft. (9) Postsynaptic currents are generated, which lead to postsynaptic excitatory (and inhibitory) potentials that in turn lead to action potential generation if the firing threshold is exceeded. This transsynaptic activation represents the activation of networks. The potential differences (E-fields), resulting from postsynaptic currents, drive volume currents inside the head and the scalp [
]. (10) The TMS-induced activation can be recorded with EEG. Note that the EEG signal can be described with a linear model, Y = B + A + N (see Section 6.1). Figure created with BioRender.com.
The brain activity evoked by TMS can be recorded with different neuroimaging techniques such as EEG, functional magnetic resonance imaging (fMRI), near-infrared spectroscopy and positron emission tomography (for a review see Refs. [
]). However, the most successful and thus commonly used combination has been with EEG because it is a widespread method, is less expensive than other neuroimaging techniques, and is technically the least complicated to be combined online with TMS.
1.1.2 EEG
Despite developments in measurement technology, the basic principles of EEG remain unchanged from Berger's time [
]. EEG, with its millisecond temporal resolution and a spatial resolution of centimeters, is widely used for non-invasively studying the electrophysiological dynamics of the brain [
]. EEG measures electrical potential differences between pairs of electrodes placed on the scalp. The recorded signal is a linear mixture of source-current amplitudes, and the signals in neighboring electrodes commonly correlate [
]. Action potentials have a short duration compared to postsynaptic potentials; for this reason, action potentials do not overlap as much in time and synchronize much less than postsynaptic potentials. Furthermore, due to their symmetric current distribution, the E-field generated by action potentials decays faster with distance than that of postsynaptic currents [
]. Postsynaptic potentials are primarily confined to the dendrites and cell bodies. When a sufficient number of neurons – several thousand or more – with similar overall orientation produce synchronous postsynaptic currents, the resulting E-field and volume currents summate, making it possible to record the cortical EEG response at the scalp level.
1.1.3 TMS–EEG
The combination of TMS with EEG has been relevant for addressing fundamental neuroscientific questions in new ways. In particular, the two techniques complement each other, in that causal information provided by TMS overcomes the correlational nature of EEG data, whereas the ability to record from the whole scalp provides a global picture of the brain activity generated by the E-field. One of the main advantages of using TMS–EEG is that outcome measures, derived from EEG responses to TMS (i.e., evoked potentials or brain oscillations) can be used as a neurophysiological marker of excitability or connectivity for any brain area, including the regions where TMS does not generate a proxy of cortical/cortico-spinal excitability, such as motor evoked potentials (MEPs) or phosphenes [
]. Although TMS−EEG data can be analyzed in the time and frequency domains, so far, most studies have focused on the former, the so-called TMS-evoked potentials (TEPs).
1.1.4 TEPs and TMS-triggered oscillations
TEPs are brain potentials time-locked to the TMS pulse [
]. To study TEPs, the signal is averaged across trials. The initial TMS-evoked response is presumably produced by the activation of neurons concentrated in the targeted area followed by the activation of axonally interconnected areas [
The TEPs consist of positive (P) and negative (N) deflections that reflect a spatio-temporal superposition of excitatory and inhibitory postsynaptic potentials, like the so-called event-related potentials (ERPs) [
]. Although the neurophysiological underpinnings of TEPs remain to be completely elucidated, they are considered a genuine, reproducible measure of cortical reactivity [
]. TMS of the primary motor cortex (M1) evokes several peaks, described at approximately 15 (N15), 30 (P30), 45 (N45), 60 (P60), 100 (N100), and 180 (P180) milliseconds [
]. However, recently it has been shown that later peaks (>∼80 ms) such as N100 and P180 may be contaminated by sensory-evoked responses (see Sections 3.5, 4.2.3, and 4.2.4), while very early peaks, such as the N15, can be contaminated by cranial muscle responses (see Section 4.2.2).
TEPs are detectable up to 400−500 ms around the stimulation area as well as in distant inter-connected brain areas [
]. Accordingly, for some TEP components, the maximal amplitude is recorded by the electrodes close to the stimulation site, while others may be more prominent over distant electrodes, e.g., over the contralateral hemisphere [
Modulation of electroencephalographic responses to transcranial magnetic stimulation: evidence for changes in cortical excitability related to movement.
TMS effects on brain activity can be further investigated in the frequency domain. When a cortical area is perturbed by TMS, the neuronal response as measured by EEG tends to oscillate at a specific natural frequency [
]. Part of this response may be explained by the phase alignment of ongoing local brain oscillations through the effect of the TMS pulse on the targeted cortex [
]. Therefore, TMS–EEG can be used to manipulate and investigate brain rhythms by measuring the impact of a TMS pulse on EEG and associated behavioral effects [
David O., Kiebel S.J., Harrison L.M., Mattout J., Kilner J.M., Friston K.J. Dynamic causal modeling of evoked responses in EEG and MEG. Neuroimage 2006;30(4)1255–1272.
]. Since this topic is out of the scope of this paper and has been widely discussed elsewhere, we refer the reader to previous literature (e.g., Refs. [
]). However, researchers should carefully distinguish between TMS-evoked responses (i.e., signals that are phase-locked and thus survive averaging of single trials) and TMS-induced responses (i.e., signals that are not phase-locked and thus cancel out during averaging; e.g., Ref. [
]). The latter requires the calculation of time-frequency representations (TFR) at the single-trial level with subsequent averaging to preserve the oscillatory activity that is related to but not phase-locked to the TMS pulse. Notably, this measure, which can also involve certain baseline normalization operations and is sometimes referred to as TMS-related spectral perturbation (TRSP), reveals a mixture of phase-locked and non-phase-locked responses that are difficult to disentangle [
Throughout this paper, we will mostly refer to TEPs when describing EEG responses to TMS, but the same considerations apply to TMS evoked and TMS-induced oscillatory activity, except where otherwise stated.
2. TMS–EEG instrumentation
This section aims to provide a comprehensive overview of the equipment currently available to acquire TMS–EEG data and to discuss how different settings/parameters affect the quality of the recordings. To do so, we reviewed published evidence, reported practices, and experiences documented by different laboratories.
The instrumentation to acquire TMS−EEG data typically includes a) TMS device and coils, b) TMS-compatible EEG amplifier, and c) TMS-compatible electrodes. The integration of a neuronavigation system is highly recommended to keep the TMS coil on the desired target with the same orientation and angulation throughout the session and across visits in the case of longitudinal measurements [
]. In addition, the use of a neuronavigation system is mandatory in studies involving patients with structural brain lesions, since stimulation of severely damaged areas does not elicit any EEG response [
]. In the following sections, we describe each component.
2.1 TMS stimulators
Currently, there are several TMS stimulators available on the market. When performing TMS−EEG studies, the following properties can be useful.
1.
Option to control the recharge delay: a change in the potential of the coil during the capacitor recharging can cause electrical artifacts in the EEG recording. Since the recharging typically occurs in a time window overlapping with the relevant signal, it is crucial to set the time of recharge outside the temporal window of interest (i.e., the recharge delay should not overlap with the relevant post-TMS signal). To meet this requirement, most of the stimulators currently available on the market (for instance, some versions of MagVenture, Nexstim, Magstim, and Deymed stimulators) include a recharge delay option that allows one to choose the recharge time (see Section 4 for more details on this artifact).
2.
Generation of different pulse waveforms: the most used are monophasic and biphasic waveforms, although available stimulators can generate other waveforms, such as half-sine and trapezoidal.
3.
Some stimulators can change the induced current direction in the coil: this may be relevant to studying the effect of the induced E-field direction on brain activity.
4.
Compatibility with different TMS coil sizes/shapes. For example, this can be helpful to perform multi-site TMS−EEG studies where 2 or more coils are placed on the head.
5.
Cooling system to run long protocols: to improve the signal-to-noise ratio (SNR) of TMS–EEG data, it is generally recommended to average a sufficient number of trials. During stimulation, the TMS coil heats up at a rate that depends on stimulation intensity (SI) and may need to stop working upon reaching a specific temperature because of safety issues. Liquid- or air-cooled coils reduce coil heating.
6.
Triggering signal communication between TMS stimulator, EEG, and neuronavigation: the communication between hardware is crucial, i.e., controlling properties of the stimulator (like SI, inter stimulus interval/randomization) via an external device or, e.g., a navigation system.
2.2 TMS coils
Currently, there are many different types of TMS coils [
Safety and recommendations for TMS use in healthy subjects and patient populations, with updates on training, ethical and regulatory issues: expert Guidelines.
]. Overall, the coil choice depends on the TMS protocol to be performed. Their shape, size, and winding determine the induced E-field and, therefore, the focality and depth of penetration, which impact the brain volume stimulated [
], but so far, there is no systematic study of the effect of the TMS coils on TMS-related EEG responses. In addition, the type of coil may also determine to what extent cranial muscles near the area of interest will be stimulated, affecting the EEG recordings. Therefore, one should be aware of scalp, facial, and neck muscle activations; for instance, the double-cone coil may trigger strong muscle twitches affecting the EEG recordings, e.g., Ref. [
], allowing electronically controlled stimulation of multiple brain areas at different times and intensities, (for an example of TMS–EEG and mTMS see Ref. [
In this section, we will briefly outline our current knowledge about the two most common pulse shapes, in TMS–EEG recordings, i.e., monophasic and biphasic waveforms [
Monophasic and biphasic pulses are defined by the amplitude ratio of the first and second phases of the E-field waveform. Monophasic pulses are shorter (usually around 100 μs) and consist of a steep initial current flow in the coil, which is responsible for neuronal depolarization. A switch or a diode in the stimulator prevents the coil current from flowing in the reverse direction (Fig. 2). Nevertheless, when the coil current (and the consequent magnetic field) returns to zero, an induced current in the brain in the opposite direction is always present. However, this current in the opposite direction only ends the depolarization phase, it will not trigger any action potentials; therefore, the biologically relevant current is monodirectional [
Fig. 2Comparison of monophasic and biphasic pulses. The monophasic pulse (left panel) consists of a steep initial current flow, whereas the biphasic pulse (right panel) consists of two half-cycles of opposite polarity (see text for a detailed description). The figure shows the time course of monophasic and biphasic magnetic pulse with magnetic field strength B (solid line) and its rate of change dB/dt (dashed line), which correlates with induced electric field strength. Reproduced with permission from Funke [
Transcranial magnetic stimulation of rodents: repetitive transcranial magnetic stimulation—a noninvasive way to induce neural plasticity in vivo and in vitro.
in: Manahan-Vaughan D. Handbook of behavioral neuroscience. Elsevier,
2018: 365-387
Biphasic pulses are longer (up to several hundreds of μs) and usually consist of at least two half-cycles of opposite polarity but similar amplitude (thus, with an amplitude ratio close to 1), and a shape that is slightly variable across stimulators. In contrast to monophasic pulses, each coil current phase can effectively stimulate the cortex [
Non-invasive electrical and magnetic stimulation of the brain, spinal cord, roots and peripheral nerves: basic principles and procedures for routine clinical and research application. An updated report from an I.F.C.N. Committee.
]. In other words, for a monophasic pulse, the first phase is more relevant for exciting cortical neurons, whereas, for a biphasic pulse, the second phase is more effective. Because of this difference, the monophasic pulse is preferred when investigating the effects of current direction, which are less pronounced with the biphasic pulse [
]. For example, the optimal current direction of monophasic pulses in the brain tissue for M1 stimulation is posterior to anterior and lateral to medial [
]. To produce this current (by a current changing in the opposite direction in the coil) with Magstim devices, the handle of the coil should point backward for monophasic pulses and forward for biphasic pulses [
]. For MagVenture (MagPro), the optimal current for M1 stimulation with default settings is generated with the handle pointing forward for monophasic pulses and backward for biphasic pulses. The difference between Magstim and MagVenture stimulators is determined by the current direction in the coil, which goes from the handle towards the end of the coil for MagVenture and vice-versa for Magstim (see Fig. 3).
Fig. 3Example of induced current direction by two different stimulators. In Magstim stimulators (left figure) the current in the coil flows from the top to the handle as indicated by the curved arrows. The current induced in the brain flows in the opposite direction and is therefore defined as posterior to anterior as depicted by the straight arrow. In other stimulators, such as MagVenture (right figure), the opposite is true. The current in the coil flows from the handle to the top as shown by the curved arrows and therefore the induced current in the brain flows from the front to the back, i.e., it is an anterior to-posterior current.
Monophasic and biphasic pulses present unique advantages and disadvantages; the choice will therefore depend on the research question. Previous studies [
] have shown that biphasic waveforms are more effective, i.e., require lower magnetic fields to stimulate the cortex (e.g., lower resting motor threshold) and, therefore, may be preferred for TMS−EEG experiments, given that the severity of many TMS-related artifacts increases with the SI (e.g., muscle artifacts) [
]. Following the stimulation of a dummy head, two independent studies reported that monophasic pulses induced a larger artifact compared to biphasic pulses, but the EEG signal returned to the baseline levels within 5 ms after the pulse delivery regardless of the waveform type. It is worth noting that, while these results indicate that the artifact duration does not depend on the waveform, the 5 ms interval hinges on the EEG equipment and recording parameters (Section 2.4). Furthermore, while the duration of the initial artifact was found to be similar, this does not rule out effects on later artifacts (some of the authors of this paper have indeed reported that the monophasic waveform causes an offset that slows the return of the EEG signal to the baseline).
The effect of the TMS pulse waveform on brain activity has been recently investigated by Casula and colleagues [
] using TEPs. The authors found that TEPs between 50 and 200 ms were characterized by a larger amplitude when evoked by monophasic compared to biphasic pulses [
]. However, the effect of pulse shape on the TEPs has not been systematically investigated and more studies are needed.
2.4 EEG amplifiers
The first methodological challenge associated with recording EEG during TMS is the strong E-field generated by the magnetic pulse, which can saturate the recording amplifiers for several seconds. To overcome this problem, a sample-and-hold circuit was introduced to control the recording apparatus and lock the EEG signal [
], thereby avoiding saturation of the recording amplifiers and allowing one to record the response generated by the stimulation after the hold period. In more recent years, a different generation of amplifiers has gained popularity and has replaced the sample-and-hold circuit approach. These amplifiers have been designed to work in high time-varying magnetic fields, thus avoiding saturation, and have in principle the advantage of allowing the EEG to be acquired continuously. However, the stimulus artifact covers a small amount of signal, possibly including the initial response of the directly stimulated cortical target, that cannot be recovered with current preprocessing methods (see Section 4). For an overview of different TMS-compatible EEG systems, see Supplementary Materials (Table S1 and questionnaires).
Despite the lack of systematic investigations, we know that some recording parameters are more effective than others in limiting the impact of the initial electrical artifact, which is a high-amplitude and high-frequency signal. As shown in Fig. 8 in Freche et al. [
]), and as recommended by many manufacturers (Table S1 Supplementary Materials), an adequate sampling rate must be selected, together with a corresponding low-pass cutoff. The lower the low-pass is, the longer the ripples created by the interaction of the filter with the TMS pulse artifact last. If sampled at very high rates, the pulse artifact lasts only as long as the actual TMS pulse and also reflects the pulse shape. With lower sampling rates (and thus lower anti-aliasing low-pass filters), filter ripples increase in amplitude and duration, and longer pulse artifacts arise. For example, with the same amplifiers and experimental setting, Veniero et al. [
] reported a 10 ms artifact with a sampling rate of 1 kHz.
As reported in Table S1 (Supplementary Materials), all TMS-compatible EEG amplifiers can record data with a high sampling rate. It is worth mentioning that some companies report that with a sampling rate of ∼20 kHz, the artifact duration is below 2 ms or even below 1 ms when sampling at 80 kHz (in line with Freche et al., [
To avoid further rippling, additional low-pass filters must be avoided where possible or carefully chosen. While low-pass filters reduce the pulse artifact amplitude, they increase its duration. Since the EEG signal covered by the pulse artifact cannot be recovered and is later removed, its amplitude and clipping can be ignored, and one should aim to reduce its duration as much as possible. For similar reasons, DC amplifiers are to be preferred over AC amplifiers, since high-pass filters also interact with the pulse artifact and introduce artificial trends/drift in the signal around the TMS pulse (for a detailed discussion on high-pass filters effects, see [
]). Of note, high-pass filters can also tamper with later artifacts and TEPs. For DC amplifiers, either no high-pass filters or a very low one (e.g., 0.016 Hz, i.e., 10 s time constant) should be used to prevent/reduce such trends.
For a list of available TMS-compatible EEG systems see Supplementary Materials, where we report the results from a questionnaire, we have asked several manufacturers to fill out with general information about each system.
2.5 EEG electrodes
In standard EEG, four types of electrodes can be used: passive, active, dry, and sponge. However, conventional EEG electrodes cannot be used with TMS [
] because the magnetic pulse induces eddy currents (i.e., currents that circulate in closed loops) and causes electrode heating. These issues can be reduced using sintered Ag/AgCl pellet or C-ring electrodes (i.e., ring electrodes with a slit to prevent current induction in a closed ring), which have been used in most TMS–EEG studies. A disadvantage of pellet electrodes is the considerable amount of preparation time required to reduce the impedances to acceptable values (5 kΩ or less). The so-called Multitrodes (EasyCap) are C-ring electrodes in which the Ag/AgCl coating is located on the inner instead of the lower surface of the C-ring. Since the contact surface is larger and more easily accessible, many authors of this paper have reported that impedances can be lowered more quickly. C-electrodes are usually preferred because they reduce eddy currents induced by TMS, which may contribute to the decay artifacts (see Section 4.1.3).
2.5.1 Active vs. passive electrodes
Active electrodes (AEs) have been introduced in electrophysiology only in recent years. Compared to traditional passive electrodes (PEs), which act as simple recording sites, AEs entail preamplification of the signal directly at the electrode stage. When recording standard EEG, this feature provides several advantages, such as the reduction of electrical line noise and the recording of a better signal at higher electrode impedance levels. In addition, the ease of montage and the fast preparation of AE recordings result in shorter experimental sessions and a reduction of discomfort for participants.
Recently, a few studies have used AE with new active amplifiers to record EEG during TMS [
] and revealed no significant difference in amplitude or scalp topography. However, some AE users have observed an increase in the decay artifact (see Section 4.1.3) duration that should be further investigated. Moreover, while AEs reduce the preparation time, their larger thickness increases the coil-to-cortex distance and requires higher TMS intensity, which might impair the EEG signal quality and lower the spatial specificity of the stimulation. This also unfavorably affects the activation threshold and should be acknowledged when reporting and comparing threshold values between studies [
]. Overall, while AEs seem a useful addition to the TMS−EEG field, more studies are needed to assess their performance in different experimental settings.
2.5.2 How many electrodes do we need to record acceptable EEG responses?
A common question in the field is how many electrodes should be used. The original International 10–20 system was devised with the intention that each electrode would inform about brain activity in the underlying cerebral structure [
]. The electrode potentials were usually measured with respect to the same reference electrode, resulting in controversial discussions about the proper reference electrode location. Currently, as we understand the sensitivity patterns of the EEG signals, we do not need to worry about the reference electrode “problem”. Referencing is a linear data transformation therefore the data can be re-referenced offline. Unless the reference position is particularly prone to local artifacts (from movement, sweating, TMS, etc.), a later re-referencing to the common average (or any other preferred linear recombination) allows recovering the reference signal, so that the referencing during recording is arbitrary.
Each electrode derivation measures the difference between two scalp potentials, informing us about one dimension of the source current distribution in the brain. This dimension, described by the sensitivity pattern or lead field of the derivation, depends on the placement of the electrodes as well as the details of the conductivity distribution of the head. When the number of electrodes is increased after the first few dozen, the marginal benefit of each new recording channel diminishes quickly because nearby electrodes sense nearly the same potential [
]. It has been found that the rank of the data obtained with a large electrode set is typically 30–50, meaning that with optimal placement on the scalp, 30–50 electrodes would be enough to gather the spatial information that is available to EEG [
]. Because the electrode placement is usually not optimized, about 60 electrodes (in practice often 64) is sufficient to obtain almost all signal components available from scalp recordings [
However, a couple of advantages are offered by a larger number of electrodes. First, if an electrode channel becomes noisy or non-functional in a 256-channel system, virtually no spatial dimension is lost, since the redundant channels can provide the lost information. Second, if one can assume that the noise in neighboring electrode channels is statistically independent (as it is if the noise is mainly from the electrode contact and the amplifiers), the overall SNR is increased; in effect, signals from neighboring channels will be effectively averaged in the course of data analysis. Thus, because the source-level SNR is in principle approximately proportional to the square root of the number of channels with uncorrelated noise, increasing the number of electrodes from 64 to 256 could double the source-level SNR [
]. In fact, some data-cleaning methods, such as the source-estimate-utilizing noise-discarding algorithm (SOUND) algorithm (see Section 6.3.3 for details), utilize cross-validation between channels to detect the channel-specific noise. For these methods, “oversampling” the EEG spatially is beneficial when estimating the noise distribution. However, SNR improvements can be obtained also by improving electrode contacts and by lowering the noise level in amplifiers. Third, artifacts due to the activation of cranial muscles could be more accurately pinpointed with additional electrodes. Since TMS activates muscles only under the coil, in some experiments it would suffice to add just a few muscle-activity-detecting electrodes over the TMS target area. The extra electrodes would enable one to measure and model the spatial pattern of the electrical activity of the muscle so that the artifact could be removed from the rest of the data.
2.6 Neuronavigation
Neuronavigation has become increasingly important in TMS research, as it increases stimulation accuracy and efficacy [
]. With navigated TMS (nTMS), the coil position and orientation can be monitored in real-time, ensuring appropriate stimulation of the target area throughout the experimental session [
]. This reduces possible inter-trial variability in the TMS–EEG recordings due to coil movement and increases accuracy by reducing the risk of stimulating a slightly different area [
]. As neuronavigation systems can store information on the coil position and orientation, they also ensure comparable targeting across multiple sessions and result reproducibility [
Optically tracked neuronavigation increases the stability of hand-held focal coil positioning: evidence from "transcranial" magnetic stimulation-induced electrical field measurements.
]. Some systems can mark EEG trials when displacements from the target occur.
Advanced neuronavigation systems compute the induced E-field in the brain, which enables precise anatomical stimulus targeting; the strength of the E-field serves also as a stimulation intensity that is independent of coil or stimulator type (see Section 3.2). Using nTMS to align the direction of induced current relative to the underlying gyral pattern is furthermore expected to increase TMS effectiveness. The strength of stimulation is enhanced when the current is perpendicular to the target gyrus relative to when it is parallel (for modeling see Ref. [
]) (see also Section 3.4). Existing nTMS systems estimate the induced E-field using spherical conductor models to take into account the local curvature of the skull [
]. While TMS–EEG studies may benefit from using neuronavigation systems based on realistic head models, such models have not yet been implemented online due to the computational cost [
]. The idea is that automatic positioning allows us to target many cortical areas in a reasonable amount of time with high precision. A caveat of robot-navigated TMS–EEG, however, can be increased levels of line noise in the EEG data from the electronics of the robot, which may require a spacer-mediated gap between the TMS coil and EEG cap and/or additional grounding measures.
3. General aspects of TMS–EEG
3.1 Number of trials (Signal-to-noise ratio)
One of the most common questions people in the EEG and TMS−EEG community ask is “How many trials do I need to acquire in my experiments to obtain meaningful TEPs or oscillations?”. Although these are simple questions, they do not have a simple answer.
The number of trials depends on the meaningful signal in relation to the noise content, i.e., the SNR. The SNR depends on the square root of the number of trials [
], provided that the meaningful signal and noise remain similar from trial to trial. In more detail, let S be the size of the signal, N the size of the noise on a single trial, and T the number of trials. The SNR on a single trial is defined as S/N (the signal divided by the noise). The total SNR of averaged responses, such as TEPs, is then equal to (S/N) ∗ sqrt(T) (the single-trial SNR multiplied by the square root of the number of trials). The closer the meaningful signal level gets to the level of noise, the more trials are required. However, if the meaningful signal is below the noise level in a single trial, even more trials are required. The required number of trials also depends on the set quality criterion, i.e., the required SNR. Suppose the required SNR is known and the single-trial signal level and noise levels are known. In that case, the required number of trials can be calculated. As noted above, the increase in SNR is not linear; therefore, doubling the number of trials does not double the SNR. For instance, to double the SNR from 100 trials, one needs to measure 400 trials. This means that, after a certain point, increasing the SNR further would lead to very lengthy experiments without significant benefit. The power law of SNR has additional positive implications. When a sufficient number of trials have been recorded, one should not be too concerned to reject contaminated epochs, as this will have only a minor impact on the potential maximal SNR. For instance, after recording 300 trials, one can reject 30 trials and decrease the theoretical maximal SNR by only 5%.
When TEPs are the signals of interest, a good starting point to set the number of trials could be looking at studies that have investigated test-retest reliability and reproducibility [
]. Many of these studies suggest that around 100 clean trials (note: clean refers to the number of trials after exclusion of artifactual epochs) are sufficient to have reliable TEPs. However, most studies have been performed on motor areas and therefore, this conclusion might not apply to other areas. Additionally, weak cortical responses tend to require more trials than strong cortical responses. For example, it has been reported that the reliability of the TEP peaks is dependent on the investigated component, and the concordance between trials plateaus after 60 trials, while the smallest detectable difference continues to improve with added trials [
] suggested that the number of trials needed for a high SNR range between 150 and 300, depending on the intensity of stimulation (as an empirical rule, the higher the intensity, the lower the number of trials). While this is a good approach, care should be taken since the strength of the cortical response varies from one location to another [
], and increasing the SI may also have an impact on TMS-induced activation of cranial muscle, voltage decay, and sensory evoked potentials. Therefore, different target regions might require a different number of stimuli. For instance, stimulation of frontal areas is more prone to artifacts than motor areas and a larger number of trials may be required since there is a higher likelihood of rejecting bad trials due to artifacts (e.g., eye-blinks and muscle contractions). However, following good practice during TMS−EEG preparation and recordings might help to decrease noise and get better SNR (see Section 5) with a reasonable number of trials.
The number of trials should also be chosen considering the type of outcome measure we are interested in. Therefore, we recommend referring to the relevant EEG literature to define the number of trials. As an example, indexes related to the frequency domain, such as pre-stimulus phase estimation are known to depend strongly on the number of trials (for a review see [
], who confirmed that if the measure of interest is the phase of the EEG signal immediately preceding the TMS pulse, the phase-estimation algorithm depends strongly on SNR.
3.2 TMS threshold determination
There are several ways to determine the TMS SI or threshold, which depend on the outcome measure of choice and a somewhat arbitrarily defined criterion. Thresholds can be determined by measuring motor responses, phosphene perception, in principle also the amplitude of TEPs, or estimated by simulations of the induced E-field.
Motor responses: The most common way to determine the SI is to measure the motor threshold (MT) in a resting muscle. This is done by first mapping the M1 cortical representation for the target muscle and then finding the optimal position and coil orientation, for that muscle, thereby maximizing the E-field at the cortical representation area (“hotspot”) of the muscle. The MT is measured by directing the E-field to the hotspot and is typically defined as the minimum TMS intensity able to evoke MEPs of at least 50 μV peak-to-peak in the contralateral muscle of interest (to the stimulated hemisphere) in 5 out of 10 consecutive trials e.g., [
Non-invasive electrical and magnetic stimulation of the brain, spinal cord and roots: basic principles and procedures for routine clinical application. Report of an IFCN committee.
]. Of note, due to TMS-induced E-field spreading and overlapping cortical representations, MEPs are also elicited in muscles adjacent to the one examined [
Longer transcranial magnetic stimulation intertrial interval increases size, reduces variability, and improves the reliability of motor evoked potentials.
]. If the SI for TMS−EEG measurements is based on the MT, one should consider using the same ISI and jitter for threshold estimation and TMS−EEG protocols.
Although the MT is measured from M1, it is commonly used to set the SI in non-motor areas as it is simple, fast (depending on the exact MT determination method), it can be reliably determined with a number of pulses as low as 17 [
]. The limitation of this approach lies in the assumption that sensitivity to TMS for non-motor areas is similar or correlated to that of M1. This does not seem to be the case, for example, see Stewart et al. [
]. Unique cytoarchitectonic features could affect how a brain region reacts to TMS. Also, simple anatomical features such as variations in scalp-to-cortex and, therefore, coil-to-cortex distance have to be taken into account; this is automatically done in navigation systems where the cortical E-field is computed (see below). In TMS, the magnetic field decreases with the square-distance; therefore, the farther the coil-to-cortex distance, the weaker the magnetic field and the induced E-field in the cortex. As the coil-to-cortex distance varies between brain areas/targets, it is challenging to know which percentage of MT should be used for other areas, and practices on how to adjust the TMS intensity vary substantially between research laboratories (for a simple metric to account for coil-cortex distance see [
Instead of recording motor responses with the EMG, some groups determine the TMS threshold by visually observing muscle twitches. However, this approach overestimates the MT and is not considered suitable for reproducible measurements [
] and standardizing methods across users. Visual observation of muscle twitches can be useful to ensure that the recorded MEPs mainly reflect the target muscle of interest.
Phosphene perception: In visual areas, the SI can be based on the phosphene threshold (PT). Phosphenes are illusory percepts, often described as visual flashes perceived immediately after the TMS-pulse, thought to occur from the direct activation of the visual cortex [
]. The PT is calculated similarly to MT, but rather than relying on objectively measurable responses (i.e., MEPs), it depends on the participants’ subjective report (they are asked to indicate the presence/absence of phosphenes). As the relevant parts of the visual cortex may be located deeper than the primary motor cortex, the PT is typically higher than the MT [
Induced E-field: Another way to determine the SI is to calculate the induced E-field at the target and select the TMS intensity that generates the desired E-field [
Within-subject effect of coil-to-cortex distance on cortical electric field threshold and motor evoked potentials in transcranial magnetic stimulation.
] and can be used for any cortical area. One limitation is that this technique requires the use of advanced neuronavigation and participants’ MRIs (see Section 2.6), which might not always be available. Furthermore, the online E-field calculation is only available in a few TMS/Navigation systems (for which the underlying algorithms for E-field estimation are not openly available). However, open-source software, which takes into account the subject-specific anatomy, for offline E-field modeling is available (e.g., www.simnibs.org [
]); and is now widely used in the field of transcranial electrical and magnetic stimulation. In contrast to accurate finite-element calculators, such as Simnibs, commercial online E-field estimators are based on computational simplifications. For instance, one such neuronavigation system is based on computing the E-field inside a sphere, fitted to the local subject-specific geometry. The computational differences between different systems can lead to discrepancies in the E-field estimations across different studies. Thus, online E-field monitoring might be most useful to normalize the TMS dose within a cohort and to ensure test-retest reliability within a subject.
Finally, the relationship between the TMS-induced E-field and the activation of the target site has to be further investigated. Factors influencing neuronal excitability such as axonal geometry may affect the required E-field in a way that is difficult to predict based on a priori information, i.e., we do not know the intensity and orientation of the E-field that should be applied to effectively stimulate the cortex. Previous studies have shown that when stimulating the visual cortex: a) with E-field intensities below 50 V/m, post-stimulation activity is indistinguishable from baseline EEG activity (i.e., no TEPs could be elicited); b) TEP amplitudes progressively increase with the intensity of the induced E-field; c) at 120 V/m there is a substantial activation of the target area [
]. Importantly, E-field estimates do not consider the possible effects of other factors such as the TMS pulse waveform and duration or the spatial extent of the E-field with a certain intensity, which may contribute to the temporal and spatial summation of the induced activations and thus to the ability of a TMS pulse to evoke action potentials in cortical neurons.
TEP amplitude: The SI can also be determined by searching stimulus parameters that maximize TEP amplitudes. In analogy with the motor hotspot search, the position, orientation, and intensity of the TMS can be adjusted to optimize the impact of the stimulation on the underlying neuronal circuits while minimizing artifacts at the same time. This approach relies on the visual inspection of the data in real-time during the recording (rt-TEP software, [
]). At first, visualization of single-trial data allows to immediately assess the presence of evoked muscle activity or other TMS-related artifacts; if the cortical target is not too close to cranial muscles, small adjustments of coil orientation and/or position are often enough to reduce the impact of these artifacts on the EEG signal [
]. Subsequently, the effectiveness of the stimulation can be evaluated by measuring peak-to-peak amplitude of average TEPs (re-referenced to the average reference) obtained after a limited number of pulses (e.g., 20-trial average) in the first 50 ms after TMS in the channels closest to the stimulation site. Specifically, EEG responses to TMS are expected to show larger amplitude a) in the channels close to the stimulation site compared to distant channels, b) at early latencies compared to late latencies, and c) in the channels of the stimulated hemisphere compared to the contralateral ones. Based on these TEP features, the peak-to-peak amplitude of the largest component measured in the first 50 ms in the channel closest to the stimulation site represents a readout of the impact of TMS on the cortex. The reliability can be further enhanced by combining multiple EEG channels into linear combinations that enhance the sensitivity of the readout to the region of interest.
The peak-to-peak amplitude of the early and local EEG response to TMS after averaging 20 trials correlates with the signal-to-noise ratio of a full session in which 80–100 trials are averaged and depend on the amplitude and variability of spontaneous EEG (see Supplementary results in Ref. [
]). Although it is not possible to set an absolute value for the ideal peak-to-peak amplitude, in principle it could be possible to estimate a reasonable endpoint based on the number of trials to be collected and on the amplitude of ongoing EEG activity.
This approach implies that the effects of TMS parameters (intensity, site, orientation) are assessed in real-time and adjusted (if needed) to minimize muscle artifacts and maximize the strength of the initial cortical activation; thus, it may imply a deviation from precise targeting requirements (e.g., while stimulating over cortical sites associated with a certain assumed function or dysfunction), for improving data quality. In conclusion, relying on a real-time EEG readout during the experiment provides immediate control over undesired artifacts. This approach is most effective while stimulating cortical structures close to the midline where cranial muscle activation can be reduced by small adjustment of TMS parameters and becomes more challenging when more lateral cortical areas are targeted [
]. Several studies have described the input-output characteristics of TMS−EEG responses, i.e., how they change as a function of the SI, and they mostly indicate a linear relationship at typical SIs, at least on M1 and prefrontal cortex (e.g. Refs. [
] for non-linear intensity–amplitude relationship in visual areas). In other cases, SI may be defined through known behavioral effects from the literature, hence ensuring suprathreshold SI. For instance, in a recent series of TMS−EEG experiments on Frontal Eye Fields (FEF)-control over posterior brain signals, Veniero et al. [
] used a fixed SI of 65% of maximum stimulator output (MSO), which was defined based on prior studies revealing that exactly this intensity effectively activates FEF and its projections as inferred from behavioral TMS effects on visual attention tasks [
], FEF-TMS at this suprathreshold SI (relative to behavioral effects) led to changes in intrinsic brain oscillations at occipital sites, i.e., in remote connected areas. Besides suprathreshold SI, there is also evidence that subthreshold SI (with respect to MT) can be sufficient to induce TMS−EEG responses, albeit likely confined to the local level. It has been shown that stimulation of the left and right M1 and prefrontal cortices at 60% MT is sufficient to evoke measurable brain activity [
]. In M1, this E-field strength can induce visible TMS−EEG peaks, but these SIs (commonly less than 50% of MT) may not be enough to activate the whole motor network [
The question about the SI necessary to activate transcallosal and other long-range pathways, as detected with TMS−EEG, is still open and will also depend on the population under investigation (e.g., brain responses of patients with major depression are altered when compared to healthy volunteers (e.g., Ref. [
]. However, in TEPs these effects have not been studied as extensively as in MEPs, as the impact of only a few coil orientations and locations has been tested. Different coil orientations influence TEP polarities [
]. In some participants, varying the coil location near the hotspot slightly influences TEP amplitudes, whereas, in others, it also affects the TEP waveform [
], where the magnitude of the entrained alpha oscillations was at its maximum when the coil was oriented to induce currents perpendicular to the target gyrus [
3.5 How to deal with the EEG responses caused by co-stimulation of peripheral structures of the nervous system
TMS typically causes somatosensory and auditory sensations because it might not only activate cortical neurons, but also nerves innervating the face, jaw, and neck muscles. Even when no muscles are activated, the pulse causes scalp sensations due to the excitation of afferent nerves (e.g., trigeminal nerve) or to mechanical stimulation of the skin by coil vibrations (e.g., a tapping sensation). In addition, a clicking sound is produced by the coil wires when the pulse is discharged and can activate auditory pathways through air and bone conduction. These sensory inputs may lead to peripherally evoked EEG responses which contaminate transcranially evoked EEG responses that result from direct cortical activation. The peripherally evoked potentials may not only contaminate transcranially evoked EEG responses but may also modulate them through neurophysiological interactions.
Recently, a few articles have triggered an intense discussion in the TMS–EEG community, opening a debate about the extent to which EEG responses to TMS are caused by direct cortical stimulation or include potentials elicited by sensory input associated with TMS [
]. Therefore, more attention has been paid to the use of control and sham stimulation during TMS–EEG experimentation. In the following, we will discuss strategies that can be used to control for peripherally evoked EEG responses, the most suitable depending on the experimental design and aim of the study [
Several procedures have been proposed to deal with the auditory stimulation that accompanies the TMS pulse delivery. Some strategies assume the linear summation of the activity generated by TMS and the auditory activation. Here, TEPs are recorded without the presence of masking noise, and the auditory evoked potential is mathematically removed either with the use of independent component analysis (ICA)-based approaches or by recording an additional auditory sham session that will be subtracted from, or at least compared to, the contaminated TEPs. Another strategy consists of controlling for auditory stimulation by playing a continuous noise to mask the coil click, such as white noise, colored noise, or a noise adapted to the spectral characteristics of the click itself and tailored in real-time based on participants' perception [
] developed and shared a tool to easily implement the latter solution with any type of coil and stimulator and to manipulate the standard noises in both time and frequency domains. Crucially, the use of this tool and the generated customized noise has been demonstrated to be effective at lower volume intensities (quantified by sound pressure level measurements) compared to the standard noises. It should be noted though that noise-masking may introduce a change in functional resting-state brain connectivity similar to the effect induced by scanner noise during fMRI [
]. This change in “brain state” might alter the brain's responsiveness to TMS.
While there is reasonable evidence that air-conducted auditory evoked responses can be suppressed by masking noise, at least under certain experimental conditions [
]. Studies are warranted to systematically assess whether or how concurrent noise exposure shapes the TEPs. Instead of masking the coil click with additional noise, one may try to reduce the coil click as much as possible. Recently, a TMS coil with substantially reduced acoustic noise has been developed by attaching the windings to a surrounding damping casing separated by an air gap [
]. The acoustic noise of the coil click was reduced by 18–41 dB. However, this coil has not been tested in TMS−EEG experiments yet.
Complementing the efforts to mask or minimize auditory and somatosensory co-stimulation, several groups have used “realistic sham stimulation” to replicate the coil click and the sensation of a real magnetic stimulation without significantly stimulating the brain tissue [
]. In TMS–EEG experiments, one option that has been explored is complementing TMS with cutaneous electrical stimulation. The TMS coil is used to reproduce the clicking noise, whereas electrodes attached to the scalp [
]. Despite all efforts to develop a realistic multisensory sham stimulation, none of the reported procedures have been able to perfectly match the peripheral co-stimulation of real TMS (see, for instance, Refs. [
]). This is mainly because the somatosensory percept related to TMS and electrical stimulation are qualitatively different and can be distinguished by the subjects [
], that permits isolation of the effect of interest. If the study aims to evaluate the effects of an experimental manipulation (e.g., learning), a pre/post-test design offers the advantage of testing the same participant at different timepoints, i.e., before and after the intervention, with the same TMS parameters. Likewise, studies that aim at testing the task-dependent modulations of TEPs may include recordings with the same TMS parameters in different task conditions. If this is the case, the sensory stimulation will be the same across time points or conditions, and differences in EEG can be attributed to direct cortical stimulation, provided that the experimental manipulation does not change the processing of sensory input. This strategy has been used in several TMS–EEG studies (e.g., Refs. [
]). The “comparative strategy” assumes that the interventional protocol does not change the peripherally evoked EEG response elicited by TMS. Although this might not always be the case it should be controlled when needed for the research question and protocol. Participants might habituate or become sensitized to peripheral co-stimulation, introducing order effects on peripherally evoked EEG responses in TMS–EEG experiments. The intervention itself may directly modulate the peripherally evoked EEG responses or indirectly by changing the arousing or attentional effects of peripheral co-stimulation on the TEP.
A similar comparative strategy has been applied in studies aiming at characterizing excitability and connectivity of a brain area in different states or during a task. In this case, the experimental design should include conditions that can be compared to answer the research question. Not many studies have used TMS–EEG during a cognitive task, but in this case, having a control task while keeping the stimulation parameter constant would ensure equal sensory stimulation. As an example, Morishima et al. [
] traced FEF connectivity in a face discrimination task and compared it to the same measure obtained in a motion discrimination task (note that faces and moving dots were presented simultaneously). Another approach entails the use of TMS pulses delivered at different intervals from an event of interest (e.g., movement onset, visual stimulus). The comparison of TEPs evoked during different “tasks”, “task epochs”, or “states” can still be influenced by task-specific, epoch-specific or state-specific modulations of the central processing caused by peripheral co-stimulation (e.g., resulting in gating or attentional shifts).
In TMS studies without EEG, a control site is often used to control for unspecific effects and establish site-specificity. However, the stimulation of different sites may induce distinct scalp sensations and muscle activation [
]. Others have explored the possibility of applying TMS controls over the same site but changing coil orientation from a more effective orientation (E-field induced perpendicular to the target gyrus) to a less effective orientation (E-field parallel to the gyrus) [
]. This should keep peripheral activation similar across conditions (e.g., from sounds), although differences in somatosensation due to different muscle fibers being activated by the two coil orientations cannot be excluded.
Therefore, many approaches have been explored but no consensus has been reached yet on the best approach. It is important to consider EEG responses caused by co-stimulation of peripheral structures when designing a study and apply the solution that is most reasonable for the purpose of the study.
3.6 Triggering of TMS based on EEG features “open- and closed-loop”
Resting TMS−EEG can provide valuable information about the general excitability state or connectivity of the cortex. However, the information obtained about the causal role of specific brain phenomena, such as cortical oscillations, is limited, because there is no obvious way to control these activities. Triggering TMS based on the current brain state can directly probe the role of different cortical functions. There has been some confusion regarding the terminology when it comes to brain-state-dependent vs. -independent and closed- vs. open-loop TMS (for a recent discussion see Ref. [
]). Triggering TMS in real-time, based on particular EEG features (e.g., oscillatory phase and amplitude of specific frequency bands), allows brain-state-dependent TMS as compared to brain state-independent TMS. The latter is when TMS is applied through some predefined sequence (e.g., with a certain ISI ± some jitter) and therefore disregarding the current brain state. Beyond brain-state-dependent stimulation, closed-loop operation requires that a particular parameter of a system is monitored continuously and that TMS parameters (control signals) are adjusted (e.g., intensity and timing of TMS) accordingly to achieve, maintain, or change the monitored parameter as desired (e.g., aiming at a particular kind of brain state). The prime example of a closed-loop is a thermostat that measures the temperature and modifies the flow of hot water to a radiator to reach and maintain a preset temperature value. However, if the control signal does not change the monitored parameter (e.g., if TMS does not change the monitored brain state), and if this change does not feed back to the stimulation parameters, the loop remains open [
]. All studies published so far, therefore, represent at best open-loop brain state-dependent TMS−EEG since TMS-related EEG artifacts and peripheral co-stimulation evoked/induced responses currently still prevent continuous EEG monitoring in real-time.
An open-loop real-time approach is represented by the TMS pulse to the brain delivered at a predefined brain state (e.g., phase), implying that the induced brain response (e.g., TEPs) does not influence the characteristics of the next TMS pulse. In essence, the state of the brain is used to guide the TMS, delivered based on a parameter decided a priori, allowing an improvement in testing the brain response in specific conditions. The other approach is defining a closed-loop, which implies controlling the brain state via TMS to reach and maintain the TMS-induced response within a predefined range. In this condition, the induced brain response provides feedback for adjusting the TMS parameters via a feedback loop [
In this context, EEG−TMS (i.e., TMS guided by EEG) can be used to characterize the physiology of endogenous oscillations, both in terms of phase-dependent excitability (e.g., which phase of the sensorimotor μ-rhythm corresponds to maximum corticospinal excitability) [
]. The promise of such EEG-triggered TMS protocols is not only that a stronger and more reliable plastic response may be achieved at the site of stimulation, but also that specific neural pathways may be modulated, when synchronizing the stimulation with EEG-derived brain connectivity states.
In terms of signal processing, whereas the pre-stimulus EEG period is unaffected by the TMS artifact, averaging cannot be used in the same way to remove random noise. Since each trial must be considered individually, signal quality issues (baseline fluctuations, eye blinks, periods of low amplitude in the oscillation of interest, etc.) are critical. Especially slow drifts caused by the previous TMS pulse when recording in DC mode can be problematic; this needs to be considered in the preparation and online signal processing pipeline.
When using oscillatory brain activity as a “state marker” to trigger TMS, the state effects will critically depend on the method used to capture the ongoing oscillatory activity [
]. Due to the limited spatial resolution of EEG, the oscillatory activity at the sensor level may reflect a mixture of activity from various cortical regions rather than being generated locally in the cortex targeted by TMS [
4. The artifact problem in TMS−EEG: non-physiological and physiological signals
The TMS pulse can induce different artifacts, which can be of non-physiological or physiological nature. These artifacts can be time-locked or non-time-locked to the TMS-pulse. Both have been described in several publications [
]. In this section, we review known EEG artifacts generated by TMS, clarify their nature and present possible solutions to deal with them.
4.1 Non-physiological artifacts
Non-physiological artifacts are induced by the TMS pulse, and their origin is electromagnetic or mechanical.
4.1.1 Pulse artifact or electromagnetic artifact
This is the largest artifact generated by the TMS pulse (Fig. 4). It is electromagnetic in nature and is produced by the electromotive force induced in the loops formed by EEG electrode leads. It can be up to several volts, masking the brain signals and saturating EEG amplifiers, limiting the use of simultaneous TMS–EEG.
Fig. 4TMS pulse artifact recorded using a sampling rate of 5 kHz and an anti-aliasing low-pass filter of 1 kHz (resulting in filter ripples or ‘ringing’). In addition, signal saturation can be observed for the first large negative deflection around 1 ms.
Solution: this artifact cannot be avoided; however, TMS-compatible EEG amplifiers have been developed, allowing one to handle this artifact (see Supplementary Materials for a list of TMS–compatible EEG systems). The best strategy we have is to reduce the pulse artifact duration to its minimum. As explained before, a sufficient dynamic range, adequate sampling frequency, and high-enough cut-off frequency for the anti-aliasing low-pass filters can reduce the artifact duration significantly.
4.1.2 TMS recharge artifacts
This artifact is produced when the capacitors, which store the electric charge required for TMS, are recharged. The recharge artifact can look like a spike, an abrupt signal jump, an exponential decay, or a waning high-frequency discharge, depending on the TMS device used (Fig. 5). This artifact can corrupt the EEG recordings and be mistakenly interpreted as a brain signal, particularly if low-pass filtering is applied or TFRs are calculated before inspecting the data.
Fig. 5Example of recharge artifact. A) When the recharge delay is not set by the experimenter, the Magstim Standard Rapid2 generates a recharge artifact that peaks at different latencies depending on the stimulation intensity. In this example, the artifact peaked at 30, 36, and 42 ms after the pulse delivery at an intensity of 50, 60, and 70% of MSO respectively. Note that the amplitude of the artifact does not change with the intensity. B) Recharge artifact caused by the MagVenture MagPro X100 when the recharge delay is set at 500 ms from the pulse delivery (Modified from https://www.fieldtriptoolbox.org/assets/img/tutorial/tms-eeg/art_recharge_2.png).
Solution: in newer TMS stimulators, the timing of the capacitor recharge can be manually adjusted; therefore, the recharge artifact can be delayed and set to occur outside of the time window of interest. When the stimulator does not allow us to adjust the delay, it is important to determine the exact onset of the recharging from the manufacturer or by performing phantom recordings [
] to facilitate the offline removal and interpolation of uncorrupted signals. It is important to note that, in some TMS systems, the recharge delay may vary depending on the SI, although there would be a consistent latency at a given SI [
Additionally, in some devices, brief (few ms) low amplitude spikes may be visible, which are not time-locked to the TMS pulse but reflect maintenance recharging of the capacitors while idling (this can be observed in some MagVenture stimulators). Custom modifications of the device allow to transiently prevent maintenance recharging for time windows of interest. Alternatively, moving median filters (width of a few ms) allows for post-hoc removal.
4.1.3 Decay artifact
Different authors have referred to this artifact as decay artifact, discharge artifact, or electrode polarization artifact [
Artifact correction and source analysis of early electroencephalographic responses evoked by transcranial magnetic stimulation over primary motor cortex.
]. In many cases, the electrode–skin interface can be polarized by electric currents between the electrolyte gel and the recording electrode. When an electrode is polarized, it might take hundreds of milliseconds after the TMS pulse for the charges to return to equilibrium. This typically leads to an exponentially decaying charge, the decaying current being proportional to the remaining polarization voltage [
]. Note that the artifact can consist of several different decaying components with different time constants.
Solution: polarization artifacts can be minimized by choosing non-polarizable electrode materials and electrolyte, as well as by low contact impedance. By ensuring the best possible conductance between the scalp and the electrode, one can shorten the time constant of the capacitive behavior of the electrode–skin connection, thus shortening the lifetime of the artifact. Low impedances (that can be further minimized by mini-punctures of the skin) have been shown to reduce the size of the pulse and decay artifacts [
]; this is relevant because skin potentials are slow shifts that can lead to an increase in low frequencies that affect the EEG recordings. Finally, minimizing the impedance of the skin–electrode interface decreases the thermal voltage noise [
01696-0&id=gr1.jpg){kind=link}
01696-0&id=gr2.jpg){kind=link}
01696-0&id=gr3.jpg){kind=link}
01696-0&id=gr4.jpg){kind=link}
01696-0&id=gr5.jpg){kind=link}