RESEARCH DOSSIER
Brain Stimulation and Perturbational Causal Inference
Abstract. Brain stimulation provides experimental interventions for examining how neural circuits contribute to psychological capacities, but a treatment effect does not automatically identify a unique local mechanism. This critical integrative monograph investigates the causal logic, scientific history, physiological mechanisms, measurement methods and evidentiary limits of magnetic, transcranial electrical, intracranial and optogenetic perturbation. It analyzes chronometric TMS, state dependence, electric-field exposure, neural recruitment, combined TMS–EEG, sensory contamination, effective connectivity, the perturbational complexity index and behavioral construct validity. Original experimental foundations are considered alongside methods reviews, quantitative synthesis and expert safety guidance. The main conclusion is that strong perturbational inference joins a precisely defined intervention to verified target engagement, controlled network and peripheral effects, operationally valid behavior, and independently replicated mechanisms. Clinical translation and extension to artificial cognitive systems are assessed as distinct downstream problems with their own evidentiary conditions.
Keywords: brain stimulation; perturbational causal inference; transcranial magnetic stimulation; TMS–EEG; transcranial direct current stimulation; optogenetics; effective connectivity; perturbational complexity index; cortical excitability; causal intervention; sham control; electric-field modeling; cognitive neuroscience; behavioral neuroscience; neural networks
The Research Object: Perturbation, Neural Change, and Psychological Outcome
Brain stimulation and perturbational causal inference investigate what follows when an experimenter deliberately changes the physical state of neural tissue and observes a defined consequence. A perturbation is an externally controlled intervention with a specified delivery method, target, time, intensity, and comparison condition. The primary outcomes may be evoked neural activity, altered network propagation, response accuracy, reaction time, a motor command, a perceptual report, or a clinically meaningful change. Each outcome supports a different level of inference. The research object is therefore a mapping from intervention to measurable response under experimentally described conditions, joined to an explicit model of the process that connects them. Foundational syntheses of human noninvasive stimulation emphasize this experimental role Polanía et al. (2018).
The central scientific question asks which neural components contribute causally to an identified psychological operation and by what mechanism that contribution occurs. Other questions concern timing, anatomical specificity, necessity versus sufficiency, recruitment of connected tissue, plastic aftereffects, and variation across participants or brain states. A physical perturbation can change behavior through the intended neural pathway, through a remote circuit, through sensory experiences associated with stimulation, or through expectancy and task adaptation. Consequently, the observed contrast between active and comparison conditions requires both a valid behavioral measure and a defensible account of intervention delivery. The methodological integration proposed in Siebner et al. (2009) makes simultaneous network measurement an essential complement to behavioral testing.
This monograph is a critical integrative synthesis of original experiments, mechanistic research, methodological investigations, consensus guidance, and quantitative evidence reviews. It traces the development from intracranial stimulation and noninvasive magnetic stimulation to combined TMS–EEG, electric-field modeling, optogenetic circuit control, and perturbation-based measures of effective connectivity. The review does not claim exhaustive systematic searching, pooled effect estimates, newly collected data, or clinical treatment recommendations. Its analytical task is to distinguish experimentally identified causal effects from broader claims about localization, computation, personal experience, and clinical benefit. The same standards apply across healthy-volunteer research, animal preparations, and observations involving patients, although the ethical and inferential constraints differ.
Intervention Logic and the Meaning of a Causal Effect
An intervention establishes a causal contrast when conditions are made comparable apart from a well-characterized manipulation and its downstream consequences. In potential-outcome language, the experimental estimand concerns the difference between an outcome under a defined stimulation regime and that outcome under a comparison regime for a specified population and task. Random assignment can support identification of an average treatment effect, yet it does not by itself identify the neural mediator. An active pulse and a sham pulse can differ in tactile sensation, auditory input, muscle response, and perceived credibility. These differences are possible causal paths rather than administrative details. An intervention whose physical treatment is incompletely specified also has an ambiguous causal estimand.
Necessity, sufficiency, and modulation require separate experimental claims. A decrease in task performance after disruption suggests that the affected neural system participates in successful performance under those testing conditions, provided alternative pathways are controlled. Successful induction of a movement or report can show that activating a circuit is sufficient to produce an effect in that preparation; it does not show that the same activation normally generates that effect. An enhancement may arise through disinhibition, compensation, or changed decision criteria rather than through amplification of a uniquely necessary representation. The virtual-lesion research program Pascual-Leone et al. (2000) provided a productive language for causal interference, but the metaphor should be interpreted through concrete electrophysiological and behavioral tests.
The unit of intervention is often a field or a pattern of recruited elements rather than a single cell, anatomical parcel, or psychological faculty. The unit of observation can be an individual trial, a participant average, a localized signal, or a distributed state trajectory. These units must be aligned by preregistered contrasts, timing assumptions, and appropriately modeled dependence. Repeated trials establish a response distribution; they do not manufacture additional independent participants. The resulting analysis should report effect uncertainty and meaningful alternative explanations, including off-target and sensory mechanisms. Human stimulation research illustrates the value of distinguishing changes in observable performance from hypotheses about the underlying cognitive operation Hallett (2007).
Historical Origins: Electrical Reward and Magnetic Access to Cortex
Direct electrical stimulation became a fundamental strategy in experimental behavioral neuroscience because it created a controllable route from imposed current to observable action. The work of Olds & Milner (1954) reported positive reinforcement associated with electrical stimulation of septal and other rat brain regions. Its historical significance lies in demonstrating that stimulation can change the reinforcement contingencies sustaining behavior. Interpretation remains dependent on electrode locations, tissue recruited by current, motivational state, and the operant procedure. The experiment does not reduce reinforcement to a single center. It established a powerful perturbational question: which neural manipulations alter an organism's choice to perform a response when alternative explanations are experimentally constrained.
A major development in human research came when Barker et al. (1985) demonstrated noninvasive magnetic stimulation of the human motor cortex. A rapidly changing magnetic field induces electric currents within conductive tissue, permitting brief cortical perturbation through the intact skull. Early applications had unusually clear physiological endpoints: stimulation over motor structures could elicit muscle responses measurable as motor-evoked potentials. The technique created a bridge between basic neurophysiology and time-sensitive tests of brain–behavior relations. Subsequent cognitive studies extended the logic from motor readouts to visual, perceptual, language, and executive tasks, where outcomes are typically more complex and less anatomically direct.
The historical trajectory also changed the concept of localization. Whereas a lesion records consequences of enduring damage, a magnetic pulse alters activity transiently and can be placed at different time points within a task. Repetitive protocols produce aftereffects that may outlast stimulation, permitting comparisons of baseline and altered states. These advantages introduced new assumptions about what was physically perturbed and how behavioral differences should be interpreted. The overview by Pascual-Leone et al. (2000) framed virtual lesions, chronometry, and functional connectivity as distinct investigative uses. Their common foundation is experimental intervention, but each requires different controls and inferential safeguards.
Transcranial Magnetic Stimulation: Physical and Physiological Mechanisms
Transcranial magnetic stimulation, or TMS, passes a short intense current through a coil to create a changing magnetic field and an induced electric field in the head. Neural consequences depend on the resulting field in tissue, neuronal orientation, pulse waveform, state-dependent membrane properties, and recruitment of axons and local circuits. Coil placement is a proxy for the distributed electric field, not a guarantee that a single cortical point is uniquely manipulated. Motor cortex physiology provides accessible calibration through motor-evoked potentials, but that calibration is indirect for nonmotor targets. The physiological account in Hallett (2007) establishes why intensity and geometry must accompany any claim about function.
Different coil shapes and pulse configurations affect focality and depth, while skull shape, cerebrospinal fluid, gyral anatomy, and orientation shape the effective exposure. A seemingly identical scalp procedure can produce different intracerebral fields across participants. A behavioral null effect may therefore reflect insufficient target engagement, an inappropriate pulse time, or a task insensitive to the affected computation. Conversely, a positive contrast may be generated through recruitment of adjacent tissue or connected regions. These mechanisms motivate individualized targeting and direct checks of target engagement. Combining TMS with imaging and electrophysiology can reveal whether the intended circuit was affected and which remote systems responded Siebner et al. (2009).
The distinction between direct physical activation and the emergent functional response is consequential. A pulse may excite axonal elements, change balance among inhibitory and excitatory processes, or induce a cascade of polysynaptic activity. The later behavioral effect reflects the current state of the network and the task demands imposed at that moment. Terms such as facilitation and inhibition are useful for operationally defined effects but cannot be assigned universally on the basis of pulse polarity or frequency alone. Studies of interneuron recruitment and motor plasticity document pronounced physiological heterogeneity Hamada et al. (2013). An adequately specified TMS experiment connects a modeled field, a testable physiological change, and the behavioral hypothesis under examination.
Single-Pulse, Repetitive, and Theta-Burst Protocols
Single-pulse TMS permits millisecond-resolved intervention at selected events during a psychological task. Paired pulses can probe interaction among facilitatory and inhibitory circuits, while repetitive TMS changes the stimulation history experienced by a cortical network. Theta-burst stimulation organizes brief high-frequency bursts into patterned trains intended to induce aftereffects. These protocol classes represent different experimental treatments; findings from one cannot automatically be transferred to another. Their relevance to causal inference depends on a mechanistic account of timing, pulse intensity, and observed physiological response. The original theta-burst study Huang et al. (2005) demonstrated patterned effects on human motor cortical excitability under the tested parameters.
Protocol labels such as continuous or intermittent theta-burst summarize temporal structure, not a guaranteed universal direction of response. The degree and sign of aftereffects can depend on recruitment of cortical interneurons, baseline excitability, previous movement, sleep, medication, and person-specific anatomy. An aftereffect measured in motor-evoked potentials is also not identical to an aftereffect in memory, attention, or mood. The experiment by Hamada et al. (2013) examined variation in response to repetitive stimulation and related it to interneuron recruitment. Such results motivate explicit manipulation checks, separate reporting of responder variation, and caution when using a standardized dose as a substitute for measured target engagement.
Offline stimulation is usually delivered before behavioral assessment, enabling a relatively uninterrupted task and extending the period during which altered network organization may be studied. Online stimulation occurs during performance and can be timed to a cue, stimulus, response preparation, or feedback event. The approaches answer related but different questions: one tests the consequences of an induced state that persists, while the other tests disruption or facilitation at a specified moment. Carryover, learning, and practice must be considered in both. Methodological guidance on combined perturbation and imaging Siebner et al. (2009) underscores the need to distinguish acute field effects from adaptation that unfolds after the train has ended.
Chronometric Intervention in Perception and Action
Chronometric stimulation asks when a neural contribution becomes consequential for an identifiable behavioral operation. Pulses delivered at differing latencies relative to a stimulus or movement can generate a time-resolved intervention function. The logic depends on temporal alignment between the pulse, the hypothesized computation, and the measurement interval. If a pulse selectively worsens performance at one latency, that window may identify a vulnerable phase of processing, although stimulation effects can extend across multiple neuronal and behavioral events. The human cognitive program described by Pascual-Leone et al. (2000) introduced chronometry as a principal advantage of magnetic perturbation compared with sustained lesions and slow hemodynamic observation.
Experimental interpretation is strengthened by a task with separable processing stages and a model of performance that allows competing predictions. A visual discrimination task, for example, can record accuracy, response latency, and confidence, while manipulating when stimulation occurs relative to the visual input. A reduction in accuracy at a selected latency may reflect disruption of perceptual encoding, evidence accumulation, memory retention, or decision formation, depending on task structure. A temporal effect does not uniquely identify the computational stage without independent behavioral constraints. The broader noninvasive-stimulation synthesis Polanía et al. (2018) illustrates why psychological outcomes must be modeled at an appropriate level instead of equating each behavioral difference with a localized faculty.
Timing precision also exposes shortcomings in trial design. Stimulus-locked responses vary with arousal, anticipation, and previous trials; response-locked pulses may interact with ongoing motor preparation. Balanced pulse timing and jittered controls are therefore valuable, as are enough trials to estimate interactions between temporal window and stimulation target. The contrast of primary interest should be prespecified and considered alongside a multiple-comparisons plan. Pulse-induced auditory or somatosensory cues can themselves provide information about temporal structure and affect performance. A chronometric inference becomes persuasive when a behavioral timing interaction converges with independent target-engagement evidence, rather than when a single nominally significant window is selected from many tests.
Brain State, Prior Activity, and Context Dependence
Neural response to stimulation depends on the brain state that precedes the pulse. Background membrane excitability, spontaneous oscillatory phase, sensory adaptation, attentional allocation, medication exposure, and recently performed actions can alter which elements are recruited and how they respond. The state-dependency analysis by Silvanto et al. (2008) challenged simple accounts in which stimulation necessarily adds or removes a constant amount of neural activity. A pulse can have different behavioral effects under different adaptation conditions because it encounters different population response distributions. An intervention therefore has to be characterized jointly with the initial state in which it operates.
State dependence presents an experimental opportunity as well as a source of confounding. Investigators can deliberately manipulate prior stimulation, perceptual adaptation, or task difficulty and test whether the causal effect changes in a predicted direction. A properly crossed design distinguishes the main effect of state manipulation from the interaction between state and pulse. An interaction can constrain models of neural selectivity or competitive processing when the corresponding behavioral operations are independently defined. Nonetheless, state measures such as scalp EEG amplitude provide imperfect proxies for cellular conditions. The mechanistic diversity described in Hamada et al. (2013) is a reason to treat physiological moderators as testable variables.
The practical consequence is that protocol effects should be expressed conditionally. Claims about stimulation facilitating cognition require the relevant task, population, state, and outcome; they do not describe universal properties of the apparatus. Training, fatigue, expectancy, circadian conditions, and participant beliefs may influence the observed contrast even when a stimulation device is calibrated correctly. Repeated-measures designs can reduce stable between-person variance but introduce carryover and order effects. Randomized orders and sufficiently separated sessions help, provided the presumed washout is justified. State-dependent analyses should be reported with uncertainty and with avoidance of retrospective subgroup explanations that overfit noisy data Polanía et al. (2018).
Transcranial Direct Current Stimulation and Excitability
Transcranial direct current stimulation, or tDCS, applies weak direct current through electrodes placed on the scalp. The resulting electric field can modulate neural membrane polarization and excitability, with consequences shaped by anatomy, montage, intensity, orientation, and current distribution. In a foundational human motor study, Nitsche & Paulus (2000) observed stimulation-dependent changes in motor-evoked potentials and explored how current direction, duration, and intensity influenced aftereffects. This work established a physiological phenomenon under controlled conditions. It did not establish that the same polarities necessarily improve or impair every cognitive process, since motor excitability and psychological performance are distinct measurement domains.
Electrical stimulation is sometimes described using anodal excitation and cathodal inhibition. These expressions summarize certain early motor experiments but become unreliable as blanket descriptions of complex cortical networks. Neurons differ in orientation, morphology, current sensitivity, and connectivity, and ongoing activity changes the functional consequence of small polarization shifts. The magnitude of the electric field in target tissue may be only a fraction of that suggested by the scalp current because it is shaped by intervening tissue. Anatomical modeling by Opitz et al. (2015) shows that electrode placement and individual anatomy meaningfully alter intracerebral electric-field distribution. An experiment with the same surface electrode arrangement is therefore not necessarily equivalent across individuals.
Interpretation of cognitive tDCS outcomes requires stronger evidence than a plausible physiological mechanism alone. In their quantitative review of single-session studies of healthy adults, Horvath et al. (2015) reported no significant effects among the analyzed cognitive outcomes, while noting variation in tasks and study protocols. That finding challenges sweeping enhancement claims within the scope of that review rather than ruling out every effect in all settings. Positive isolated reports must be evaluated against adequate blinding, independent replication, effect-size uncertainty, and the chosen behavioral construct. A precise research program distinguishes evidence that current reaches tissue, evidence of neural modulation, and evidence that such modulation changes a particular cognitive ability in a reliable way.
Electric-Field Modeling, Exposure, and Individualized Dose
An electrode montage or coil position specifies how stimulation is applied externally, whereas the intervention experienced by neural tissue is the intracerebral field. The mapping depends on geometry and conductivities of scalp, skull, cerebrospinal fluid, gray matter, and white matter. Realistic head models can estimate spatial electric-field distributions and compare proposed stimulation arrangements, although model predictions inherit uncertainty from segmentation, material parameters, and assumptions about tissue orientation. The study by Opitz et al. (2015) demonstrates why head anatomy and electrode placement deserve explicit consideration in tDCS experiments. Model-informed exposure constitutes a substantial improvement over treating current delivered at the surface as a person-independent cortical dose.
Physiological validation matters because computational predictions alone do not prove engagement of a hypothesized neuronal population. In cross-species studies of transcranial electric stimulation, Vöröslakos et al. (2018) combined measurements in animals and human-related tissue preparations to investigate field attenuation and threshold conditions. Their results highlighted the distance between externally delivered current and direct neural effects and motivated alternative pulse configurations. Generalization from rodents, cadaver preparations, and particular electrophysiological readouts to living human cognition remains a separate question. An experimentally valuable model therefore links dose prediction to measured physiological change in the same target system.
Individualized dosing can use structural images, motor thresholds, navigation, electrode geometry, impedance measures, and physiological response functions. Each component answers a different question. Motor threshold calibrates a motor pathway, not the complete field distribution for prefrontal targets; structural targeting improves reproducibility but does not prove successful modulation. Dose–response experiments should span scientifically and ethically permissible ranges and evaluate whether increasing estimated target exposure changes the hypothesized neural response and behavioral endpoint. Nonmonotonicity and interindividual differences remain plausible. The research synthesis by Polanía et al. (2018) emphasizes a coherent combination of biophysics, neurophysiology, and cognitive theory rather than one-dimensional comparisons of current amplitude.
Waveform, Alternating Current, and the Problem of Entrainment
Electrical stimulation protocols differ in temporal structure, including direct current, alternating current, short pulses, and patterned waveforms. Frequency-specific alternating stimulation may be designed to influence oscillatory timing, while broadband or pulsed interventions probe other physiological sensitivities. The relevant neural mechanism cannot be inferred solely from matching an applied frequency to a recorded EEG rhythm. Scalp EEG measures aggregate electrical activity, and several distinct neuronal processes can contribute to similar spectral features. A causal claim about oscillatory entrainment requires showing that stimulation changes neural timing in a manner distinguishable from measurement artifact and from peripheral sensory effects. The field-level considerations developed by Opitz et al. (2015) remain relevant across electrical waveforms.
The term entrainment is strongest when the phase or frequency of neural activity changes systematically with the external drive and when evidence rules out purely additive electrical contamination. Aftereffects lasting beyond stimulation raise additional questions about learning, plasticity, and the persistence of altered dynamics. These should be studied with stimulation-off measurements and suitable sham controls. Disentangling a direct frequency-specific neural process from an indirect behavioral state change may require electrophysiology, task manipulations, and computational models. The experimental work of Vöröslakos et al. (2018) demonstrates why biological responsiveness and the magnitude of intracerebral fields must be evaluated before an externally applied waveform is interpreted as a sufficient neural intervention.
Interpretation also depends on the outcome metric. Power in one frequency band can change through movement, vigilance, eye activity, or altered network coupling without uniquely identifying a psychological state. Behavioral enhancement should be demonstrated through independently validated tasks with an appropriate control for test familiarity and expectancy. Claims should specify stimulation parameters and electrode locations in enough detail for reproducibility. A method can be valuable for studying resonance and network susceptibility even when its clinical or cognitive improvement remains uncertain. This distinction allows a productive scientific program that evaluates parameter-sensitive mechanisms without relying on broad claims of universal cognitive optimization Horvath et al. (2015).
Optogenetics and Cell-Type-Specific Perturbation
Optogenetics provides a different form of causal access by expressing light-sensitive proteins in selected neuronal populations and using light to alter their activity. The foundational report by Boyden et al. (2005) demonstrated millisecond-scale optical control of neuronal spiking in mammalian neurons using channelrhodopsin. Genetic targeting can restrict the intervention to a defined class of cells more precisely than scalp stimulation usually can. Nevertheless, genetic markers are operational tools, and the recruited population can still contain functional heterogeneity. Light intensity, scattering, expression variability, synaptic propagation, and developmental compensation all shape what the experiment actually changes. Cellular selectivity is therefore a claim to be validated, not an automatic property of the technology.
Circuit perturbation is particularly informative when anatomical, electrophysiological, and behavioral evidence can be connected. In mice, Kravitz et al. (2010) manipulated genetically distinguished striatal projection pathways and observed directionally different changes in motor behavior in experimental models relevant to parkinsonian movement. The study supports causal roles for defined pathway populations under its testing conditions. It does not show that every behavioral function is implemented by an isolated channel whose activation has a fixed meaning across organisms or states. Connections with local interneurons, downstream nuclei, and feedback loops remain essential parts of the causal system, and effects depend on the preparation and baseline pathology.
The interpretive advantage of optogenetics is its potential to compare carefully selected neural populations under timing-specific interventions. The main constraints include invasiveness, dependence on transgenic or viral delivery methods, possible photoelectric and thermal artifacts, and limited direct generalization to intact human cognition. A well-designed experiment compares expression and light conditions, establishes physiological engagement, and tests behavioral hypotheses alongside negative controls. Repeated stimulation may alter plasticity, arousal, or strategy, creating effects beyond the targeted instant. These design demands preserve the distinction between a precisely administered local manipulation and a fully explained behavioral mechanism Boyden et al. (2005).
Intracranial Electrical Stimulation and the Reach of Local Intervention
Intracranial stimulation directly delivers current through electrodes implanted or placed in close proximity to neural tissue. It can provide access to deep structures that are difficult to reach with external magnetic or electrical methods. Experimental uses range from local excitability tests to evoked experiences and alterations of motor, sensory, and cognitive behavior. The physical current spreads according to electrode geometry and local tissue properties; it can activate passing fibers and remote circuits. An electrode coordinate is therefore a useful spatial descriptor but does not uniquely identify the neurons responsible for an evoked outcome. Historical reward experiments Olds & Milner (1954) illustrate how measurable behavior can be influenced through such interventions.
Neural stimulation at one site can change the effective input to multiple downstream areas, depending on anatomy and task state. To identify a causal pathway, investigators may compare electrodes, amplitudes, stimulus patterns, recorded neural responses, and behavioral consequences. Such experiments become more informative when stimulation of distinct nodes produces differentiable predictions under one explicit mechanistic model. A direct physiological response can confirm recruitment without showing that the same recruitment accounts for a later change in behavior. Conversely, an unexpected effect may expose a pathway that structural anatomy alone did not predict. Network-oriented accounts of intervention Siebner et al. (2009) explain why remote effects are not merely experimental noise.
Invasive human recordings and stimulation often arise in clinical contexts, which create both opportunities and restrictions. Participants may have seizures, altered medications, localized pathology, or electrodes positioned for medical rather than basic-research purposes. The sampling of accessible brain tissue is consequently selective. Clinical safety requirements determine permissible stimulation amplitudes, observation periods, and procedures. Cross-person comparisons must account for unequal placement and heterogeneous histories. Experimental reports should describe the clinical population and separate observations about the stimulated patient from claims about typical human psychology. Informed authorization, medical oversight, and research governance belong to the causal validity of the study because they shape which interventions and outcomes can be observed.
Network Propagation, Effective Connectivity, and Causal Organization
Effective connectivity concerns directed influences that one neural process exerts on another under an explicit model or experimental manipulation. Structural connectivity describes anatomical pathways, whereas functional connectivity commonly describes statistical dependence between measured signals. Perturbation can help discriminate among these descriptions by asking whether stimulation at one node produces time-ordered responses elsewhere. The combined TMS–EEG experiments of Ilmoniemi et al. (1997) provided early demonstrations that a focal magnetic pulse could evoke an electrical response that propagated across cortical areas. The finding helped establish the methodological idea of measuring interregional consequences of an experimentally timed input.
Propagation does not automatically prove a direct monosynaptic link. Responses recorded at distant electrodes may reflect several anatomical routes, recurrent interactions, common sources, or volume-conduction and inverse-model limitations. Timing, anatomical constraints, multisite recording, and comparisons across targets can narrow the hypotheses. In healthy humans, simultaneous TMS and other neuroimaging methods have been used to study distributed effects of stimulation and to characterize changes in circuit responsiveness Siebner et al. (2009). A network statement is therefore strongest when it specifies the nodes, the intervention and response variables, the temporal window, and the limits of the recording model.
The distinction between local excitability and propagation is particularly important for psychological explanation. A system may respond strongly near the pulse but fail to transmit organized activity through a larger network, or it may transmit a response without producing a measurable change in a particular behavior. These outcomes concern different aspects of neural organization. In sleep research, Massimini et al. (2005) used TMS and high-density EEG to compare evoked cortical propagation across conscious wakefulness and nonrapid-eye-movement sleep. The experiment established state-dependent differences in perturbation responses, while broader theoretical interpretation of those differences requires independent behavioral and physiological evidence.
TMS–EEG: Measuring the Brain's Response to a Controlled Pulse
Combined transcranial magnetic stimulation and electroencephalography, or TMS–EEG, records the electrical response that follows a controlled magnetic pulse. The method can address questions about cortical reactivity, inhibition, propagation, and state-dependent network dynamics, with temporal resolution suited to early and late evoked components. The original development reported by Ilmoniemi et al. (1997) showed how EEG could reveal responses in connected cortical territories after stimulation. Subsequent methodological synthesis Tremblay et al. (2019) assembled evidence on acquisition, physiology, analysis, and potential clinical applications. The core measurement is a spatiotemporal signal conditioned on stimulation parameters, recording setup, and analysis decisions.
A TMS-evoked potential contains multiple overlapping processes. Direct activation of stimulated tissue can coexist with network transmission, auditory responses to coil discharge, scalp nerve activation, muscle contractions, eye movements, and electrical artifacts. An EEG component defined by peak latency or electrode location is an operational feature rather than an independently identified receptor, neurotransmitter system, or psychological state. Its value emerges from convergent manipulations, physiological benchmarks, and reproducibility. The consensus analysis by Tremblay et al. (2019) describes opportunities and complications for relating TMS–EEG features to cortical excitability or clinical differences. Rigorous interpretation requires the researcher to specify what the measurement can rule in and what remains ambiguous.
Good studies combine precise pulse triggering, suitable amplifier behavior, field and coil documentation, sensory controls, artifact-aware preprocessing, and independent comparison measures. Averaging across many trials can improve signal-to-noise ratio but may conceal changing state or mix different response classes. Source reconstruction adds hypotheses about neural generators and depends on a model of head conductivity and recorded sensors. Inference from the resulting waveform to effective connectivity must therefore be supported by time-resolved causal contrasts rather than assumed from a visually plausible scalp map. The methodological principle is to preserve the integrity of the stimulation–response relationship at every stage of acquisition and processing Siebner et al. (2009).
Sensory Contamination, Sham Conditions, and Signal Separation
An active TMS pulse creates a physical stimulus that participants can hear and feel. Coil clicking, skin and muscle sensations, and involuntary movement may affect EEG signals, subjective expectations, or task performance independently of the intended cortical action. A sham condition should reproduce the relevant nonspecific experiences while minimizing the intracerebral field, yet exact matching can be technically difficult. Because the physical stimulation and sensory co-stimulation are linked, simple subtraction of active and sham average waveforms relies on an assumption that sensory and direct cortical effects combine in a sufficiently separable way. Methodological work on optimized sham and a pharmacological challenge explored this issue empirically Gordon et al. (2023).
Signal cleaning can remove large artifacts but can also alter the neural components under study. Independent component analysis, interpolation around the pulse, masking, and source reconstruction involve analyst choices that influence waveforms and derived indices. A credible measurement report explains when data were discarded, which parameters were used, what signal was recovered, and how results depend on reasonable alternative choices. Experimental evidence on cortical and sensory responses indicates that a stereotyped later waveform may occur after several cortical stimulation sites and sensory control conditions Gordon et al. (2023). Such findings make site specificity and direct physiological validation central requirements for strong claims about early and late TMS–EEG components.
Sham validity should be assessed for each scientific estimand. In a behavioral experiment, perceived credibility and acoustic distraction may be decisive; in TMS–EEG, the morphology of stimulus-evoked sensory potentials and amplifier recovery may be decisive. Active control at another cortical site can introduce a real brain effect rather than serve as a neutral placebo. Conversely, scalp stimulation controls can induce neural sensory responses that must be analyzed rather than ignored. Transparent reporting of all control differences enables later replication and quantitative evidence synthesis. The research objective is a credible counterfactual contrast whose remaining differences plausibly isolate the intervention mechanism of interest Tremblay et al. (2019).
Perturbational Complexity and the Scientific Study of Consciousness
Perturbational approaches to consciousness attempt to measure how widely and diversely a brain responds when an investigator delivers a controlled input. Rather than relying only on spontaneous neural activity or a person's verbal report, the approach estimates properties of the evoked spatiotemporal pattern. In their wakefulness and sleep study, Massimini et al. (2005) observed distinct patterns of response propagation after TMS, including more restricted subsequent responses during nonrapid-eye-movement sleep. This work provided an empirical foundation for investigating the complexity of integrated cortical responses across states. The interpretation concerns organization of evoked dynamics, which must be connected cautiously to psychological and clinical states.
The perturbational complexity index, or PCI, was introduced by Casali et al. (2013) to quantify the complexity of cortical responses evoked by TMS and recorded with EEG. The method identifies significant spatiotemporal activation patterns and uses a compression-based measure of their algorithmic complexity. Its theoretical rationale links differentiation and integration of causal interactions to the capacity for conscious experience. The authors tested the measure across healthy wakefulness, sleep, anesthesia, and clinical conditions involving impaired responsiveness. This is a theory-guided measurement program that combines a physical intervention, a statistical processing pipeline, and an interpretive model; its numerical outcome is not a direct recording of subjective experience.
Clinical applications require particular attention to target populations and diagnostic reference standards. An unresponsive patient may retain capacities that cannot be expressed behaviorally, while a complex evoked response may arise under conditions differing from those in initial validation studies. PCI thresholds and classification rules must be assessed for prospective performance, artifacts, protocol dependence, and population transportability. The recovery observations of Rosanova et al. (2012) connected changes in cortical effective connectivity with clinical recovery of consciousness, strengthening interest in longitudinal perturbation measurements. The general scientific challenge is to relate an objective neural index to carefully established clinical and experiential constructs without collapsing them into one another.
Causal Claims About Metacognition and Higher Cognition
Stimulation studies of metacognition examine whether perturbing a specific neural system changes judgments about one's own performance. The behavioral distinction is important: first-order task accuracy may remain stable while confidence discrimination, visibility judgments, or calibration changes. In an influential prefrontal theta-burst experiment, Rounis et al. (2010) reported changes in metacognitive visual awareness and analyzed them using signal-detection methods. The finding generated a causal hypothesis about prefrontal contribution to metacognitive sensitivity. Such a hypothesis is more precise than saying that a stimulation site is the location of consciousness, because it names an operational measure and an experimental contrast.
Identifying a metacognitive effect requires appropriate separation of sensory evidence, discrimination performance, decision criterion, and confidence scale use. Without that separation, lower visibility reports could reflect a conservative response strategy, reduced attention, or changes in task engagement. In a direct replication and extension of the Rounis paradigm, Bor et al. (2017) reported no metacognitive impairment after prefrontal or parietal theta-burst stimulation; the investigators included active-control conditions and examined the impact of exclusion and signal-detection decisions. Their result directly limits generalization from the positive original experiment Rounis et al. (2010). The divergent outcomes should motivate prospective independent replication, reliable target-engagement checks, and transparent treatment of unstable metacognitive estimates. The broader framework in Polanía et al. (2018) treats brain stimulation as one component of convergent causal evidence.
The same inferential architecture applies to executive control, language, memory, and social cognition. A pulse may alter errors or latencies in a complex task without pinpointing a single psychological computation. Process models can distinguish evidence strength, inhibitory thresholds, attention shifts, and response conflict, thereby generating predictions that differ across stimulation conditions. Experiments then become tests of these alternatives rather than demonstrations that a named cortical area possesses a named faculty. The virtual-lesion and chronometric approaches summarized by Pascual-Leone et al. (2000) are useful precisely when their temporal and anatomical parameters are linked to discriminating behavioral predictions.
Construct Validity, Behavioral Assays, and Mechanistic Precision
A causal effect of stimulation on an outcome variable is only as psychologically informative as the variable's measurement model. Accuracy on a task can reflect perception, memory, motivation, motor control, speed–accuracy tradeoffs, or response bias. A questionnaire can reflect current affect, demand characteristics, item interpretation, and stable disposition. Selecting a familiar task does not resolve these ambiguities. Perturbational research should define the psychological construct before selecting the neural intervention, then identify which observable features would support or contradict proposed mechanisms. The review by Polanía et al. (2018) demonstrates how stimulation studies can be integrated with carefully chosen behavioral paradigms.
Evidence becomes more discriminating when multiple indicators of the same construct converge and when theoretically adjacent constructs display predicted differences. For example, confidence discrimination can be tested alongside perceptual accuracy and response criterion, while response inhibition can be separated from general response slowing. Within-person task contrasts may isolate particular operations, provided practice effects and unequal difficulty are considered. Brain stimulation then functions as an intervention on a mechanistic hypothesis, not as a general-purpose validation device for a questionnaire or an anatomical atlas. The metacognitive analysis presented in Rounis et al. (2010) illustrates the importance of separating different stages of a judgment.
An experimentally observed behavioral effect can also arise through changes in strategy over repeated trials. Participants may learn to compensate for disrupted performance, use auditory cues produced by stimulation, or shift their effort after recognizing the active condition. Reaction-time distributions and trial-level models can disclose patterns lost in averaged scores. Task sequencing, response mapping, training, and measurement intervals should consequently be documented. Convergent physiological measures can strengthen evidence for target engagement, but they do not cure a behaviorally invalid construct. The scientific payoff of perturbation depends on the clarity of the psychological explanation that is actually being tested Hallett (2007).
Randomization, Blinding, Sham, and Active Comparators
Randomized sham-controlled stimulation studies seek a counterfactual comparison between active and nonactive or alternative intervention conditions. Successful randomization balances expected background characteristics, but blinding may be difficult when active stimulation produces distinctive sensations. Subject and assessor masking should therefore be evaluated rather than assumed from the label double-blind. A trial can report whether participants guessed their allocation, whether the two procedures differed in discomfort, and how expectancy or prior experience was measured. Objective neurophysiological endpoints can reduce some subjective reporting problems while creating their own sensory and recording confounds Gordon et al. (2023).
Crossover designs are attractive for transient interventions because each participant can serve as a comparator across conditions. Such designs depend on adequate washout and control of order, learning, and sustained aftereffects. Parallel-group designs reduce carryover but often require larger samples to achieve comparable precision. Active-control stimulation at another site may help test anatomical selectivity but can influence behavior through its own network effects. Sham controls and active controls therefore answer different questions. A design with multiple comparison conditions can characterize both expectancy and functional specificity if prespecified primary contrasts and adequate statistical power are maintained Siebner et al. (2009).
Safety, tolerability, and blinding are intertwined. Intensity and coil placement influence both field delivery and somatosensory experience; changing them solely for indistinguishability may alter the causal intervention. A good research protocol specifies how the sham apparatus approximates sensory cues and which differences remain. Statistical plans should account for repeated measures, treatment-by-time interactions, clustering, and heterogeneity rather than treating every recorded trial as an independent randomized subject. Reporting null and adverse findings is part of the same evidentiary standard. Expert TMS guidance Rossi et al. (2021) provides an important safety framework for deciding which experimental contrasts are ethically feasible.
Target Engagement, Dose–Response, and Mediation Analysis
A strong mechanistic study establishes three linked elements: the stimulation procedure changes a predicted physiological variable, that change predicts a specified behavioral consequence, and competing paths are experimentally constrained. The intermediate variable is a candidate mediator, not a mechanism established simply by correlating two post-intervention measurements. Formal mediation inference requires assumptions about mediator–outcome confounding, timing, measurement reliability, and potential treatment-induced confounders. Randomizing the stimulation exposure does not automatically randomize the mediator. Combined TMS and neuroimaging methods Siebner et al. (2009) offer direct opportunities to test target engagement while leaving these additional assumptions open to scrutiny.
A dose–response curve adds information when the dose is expressed as a credible measure of neural exposure and when tested intensities cover a theoretically meaningful range. External voltage, coil power, electrode current, and modeled tissue field represent different variables. For tDCS, Opitz et al. (2015) shows why scalp configuration and participant anatomy influence the field experienced by cortex. For repetitive TMS, Hamada et al. (2013) illustrates variation in recruited cortical circuits and aftereffects. A monotonic dose–behavior relation would support some mechanisms, whereas nonmonotonic or null relations could reflect saturation, homeostasis, different recruitment thresholds, or insensitive behavior. These possibilities should be defined prospectively where feasible.
One route toward stronger mediation is triangulation: independently manipulate the proposed mediator when ethically and technically possible, measure its temporal precedence, and test whether stimulation effects disappear or change when a mechanistically specific pathway is blocked or altered. Cell-type-specific optical interventions can support such comparisons in animal circuits, while human studies often rely on multiple imperfect measures. A well-characterized effect of direct and indirect basal-ganglia pathway stimulation Kravitz et al. (2010) exemplifies the value of linking controlled circuit access to behavior, though it remains specific to the mouse preparations and task circumstances examined. Mechanistic conclusions must travel only as far as their interventions and measurements permit.
Heterogeneity, Replication, and the Limits of Average Effects
Individual responses to brain stimulation are heterogeneous because skull and cortical anatomy, neural recruitment, learning history, vigilance, baseline excitability, and medication may differ. Statistical heterogeneity can also reflect imprecise measurement or protocol variation rather than distinct biological types. Estimating an average treatment effect is useful, but it does not establish that all participants benefit or that a sample-derived responder category is stable. In human motor studies, Hamada et al. (2013) investigated variability in plastic responses associated with recruited interneuron circuits. The finding motivates repeated physiological measurements and cautious interpretation of correlations between baseline markers and outcome changes.
Replication is particularly important when effects are small, stimulation protocols are flexible, and behavioral tasks offer many alternative endpoints. The quantitative review by Horvath et al. (2015) found little reliable evidence for single-session tDCS effects on several tested cognitive outcomes in healthy adults, illustrating why selective publication of successful contrasts can distort impressions of efficacy. Differences between a failed replication and a genuine moderator must be analyzed through adequate power, measurement equivalence, and documented differences in intervention delivery. A robust evidence synthesis considers positive, null, and contradictory studies with attention to study design rather than attributing every discrepancy to participant individuality.
Recommended practices include preregistered primary outcomes, sufficient independent participants, within-person reliability assessments, reported effect sizes and confidence intervals, and sharing of stimulation and preprocessing parameters. Multisite replication can test transportability across equipment, staff, and populations, while individual-level stimulation models can assess whether predicted field delivery explains some variance. An exploratory finding may guide later experiments, but should be identified as exploratory when initially reported. The broader methods literature Polanía et al. (2018) encourages integrating neurophysiology and behavior; an intervention finding achieves cumulative scientific value when its contrast is stable enough for independent teams to reproduce.
Clinical Translation, Safety, and Evidence Boundaries
Noninvasive stimulation methods are also studied in psychiatry, neurology, rehabilitation, and pain, but therapeutic efficacy raises questions beyond laboratory causal mapping. A laboratory contrast may establish an acute physiological change without demonstrating that repeated treatment improves a patient-centered endpoint. Clinical trials require appropriate diagnosis, outcome validity, comparator procedures, follow-up periods, adverse-event reporting, and analysis of benefits against risks. The comprehensive expert recommendations in Rossi et al. (2021) address safety, training, ethical requirements, and regulatory issues for TMS in healthy and patient populations. Such guidance provides the safety framework within which scientifically informative perturbations may be conducted.
The use of TMS in clinical neurophysiology illustrates the distinction between a mechanism probe and an intervention with a demonstrated health benefit. Measures of cortical reactivity or connectivity may be useful for detecting a biological state or predicting outcomes, but they require separate validation as biomarkers before they can guide medical decisions. The review by Tremblay et al. (2019) surveyed prospective TMS–EEG clinical applications, methodological dependencies, and measurement challenges. In studies of disorders of consciousness, Rosanova et al. (2012) associated changes in cortical propagation with recovery, while Casali et al. (2013) developed a perturbation-based index requiring clinical validation across patients and conditions.
Ethical research design includes screening for medical risks, appropriately trained personnel, monitoring of discomfort and adverse events, responsible recruitment, and the special protections applicable to vulnerable populations. The safety profile differs among magnetic, electrical, implanted, and optogenetic methods; evidence from a rodent preparation cannot establish safety for human interventions. Research reports should state whether a protocol was experimental, clinically approved, or used under a specified investigational authorization. Conclusions about treatment benefit should follow trials and professional guidance, not extrapolation from basic circuit responses. The present synthesis supplies no individual prescription or stimulation protocol; clinical application remains subject to qualified supervision and the applicable regulations Rossi et al. (2021).
Causal Identification Failures and Competing Explanations
Perturbation creates an opportunity for causal inference, yet several routes from treatment to outcome can violate an investigator's preferred interpretation. An active pulse may drive peripheral afferents, change task arousal, evoke discomfort, influence expectancy, or recruit connected tissue outside the intended target. If any such pathway changes the measured behavior, the total treatment effect remains real while the claimed local neural mechanism is underidentified. TMS–EEG exemplifies the problem because evoked electrical signals can combine cortical response, sensory response, and recording artifact. Optimized sham studies Gordon et al. (2023) show why experimentally separating these sources is necessary for credible physiological interpretation.
Another failure is equating temporal or spatial coincidence with causal exclusivity. Stimulation at a cortical coordinate can alter several processes, and a deficit may be expressed only when the task prevents compensatory recruitment. A response measured at a distant site may reflect network propagation rather than a directly stimulated anatomical connection. A motor-evoked potential can change even when the cognitive task of interest remains stable. The imaging consensus Siebner et al. (2009) emphasizes the distinction between immediate stimulation effects and distributed network adaptation. Strong claims specify the causal contrast and the evidence that excludes plausible rival paths, with residual ambiguities made visible rather than hidden by anatomical terminology.
Statistical problems compound mechanistic ambiguity. Selecting the best-performing coil target after examining outcomes, searching many latency windows, using flexible artifact rejection, or defining responder subgroups retrospectively can capitalize on noise. Small correlated samples and repeated trial counts may create misleadingly narrow standard errors if subject-level variation is ignored. Null findings can also be inconclusive when target engagement is unknown. A transparent causal report should distinguish intervention efficacy, physiological engagement, construct measurement, and pathway mediation and state which were directly tested. The synthesis by Horvath et al. (2015) demonstrates how aggregate evidence can challenge compelling but weakly replicated behavioral claims.
Multiscale Integration and Future Research Programs
Future perturbational research can become more explanatory by linking physical stimulation fields, cellular recruitment, circuit propagation, computational operations, and observable behavior within one explicit inferential design. Each scale contributes an independently measurable variable. Models can predict which intervention should change which downstream response under a given brain state, and experiments can distinguish among competing patterns. Optogenetics provides access to selected neural populations Boyden et al. (2005), while TMS–EEG tracks time-resolved human network responses Ilmoniemi et al. (1997). Cross-scale integration should preserve the differences between species, recording technologies, outcome validity, and ethically permissible interventions.
Adaptive targeting and closed-loop stimulation introduce additional causal questions because the delivered treatment depends on previously measured neural signals. Such systems are dynamic treatment regimes rather than fixed exposures. The choice of trigger, state estimator, intervention timing, and feedback variable can influence the resulting behavior. Valid evaluation calls for randomized or otherwise identifiable comparisons of policies, transparent treatment histories, and explicit handling of time-varying confounding. The state-dependence framework in Silvanto et al. (2008) gives a conceptual foundation for examining why a brain's response changes with its condition, while field modeling Opitz et al. (2015) can help characterize the actual exposure across candidate targets.
An Artificial Era research horizon extends methodological comparison to artificial cognitive systems, where controlled input perturbations, component ablations, and causal tracing can probe engineered information processing. Comparisons are informative when the units and intervention assumptions are specified: disrupting a software module and stimulating a biological network differ in their physical substrates, reversibility, access, and internal state. A common methodological question is whether an imposed intervention produces the predicted change in a defined functional output. This is a proposed interdisciplinary extension, not an empirical finding that biological and artificial systems share identical psychological mechanisms. The research program advances through explicit operational definitions and tests that distinguish alternative causal models.
Conclusion
Brain stimulation advances causal neuroscience by placing carefully defined interventions between hypotheses about neural organization and measurable physiological or behavioral outcomes. Historical electrical stimulation, noninvasive magnetic pulses, direct current modulation, genetically targeted optical control, and combined TMS–EEG methods each offer a distinctive range of access. Their evidentiary strength depends on how accurately the intervention is characterized, whether its neural effect is independently observed, and whether the measured task validly represents the psychological operation being investigated. The development from Barker et al. (1985) to network-based TMS–EEG and optogenetic studies Boyden et al. (2005) has substantially expanded the available experimental contrasts.
The reviewed evidence supports several stable conclusions. Perturbation can establish effects that correlational imaging alone cannot identify; stimulation acts within distributed and state-dependent systems; individualized field exposure matters; sensory and recording artifacts can compromise mechanism attribution; and behavioral outcomes must be evaluated independently of neural target engagement. Work on cortical propagation Massimini et al. (2005), perturbational complexity Casali et al. (2013), and cognitive interventions Rounis et al. (2010) demonstrates both the conceptual reach and the inferential demands of the approach. Disagreement and null results are part of the cumulative evidence, especially when claims extend from physiological modulation to generalized cognitive enhancement Horvath et al. (2015).
The central research priority is a transparent chain from intervention definition to identified causal contrast, physiological mechanism, behavioral construct, replication, and warranted generalization. Future experiments should combine mechanistic controls, model-informed dosage, robust sham procedures, longitudinal observation, and adequately powered cross-population tests while honoring the safety constraints of each technology Rossi et al. (2021). The resulting account of brain–behavior relationships treats causation as a scientific property to be established through discriminating evidence. The strongest outcome of perturbational research is an increasingly precise explanation of how organized neural systems contribute to particular psychological functions under specified conditions.
References
- Barker, A. T., Jalinous, R., & Freeston, I. L. (1985). Non-invasive magnetic stimulation of human motor cortex. The Lancet, 1(8437), 1106–1107. https://doi.org/10.1016/S0140-6736(85)92413-4
- Bor, D., Schwartzman, D. J., Barrett, A. B., & Seth, A. K. (2017). Theta-burst transcranial magnetic stimulation to the prefrontal or parietal cortex does not impair metacognitive visual awareness. PLOS ONE, 12(2), e0171793. https://doi.org/10.1371/journal.pone.0171793
- Boyden, E. S., Zhang, F., Bamberg, E., Nagel, G., & Deisseroth, K. (2005). Millisecond-timescale, genetically targeted optical control of neural activity. Nature Neuroscience, 8(9), 1263–1268. https://doi.org/10.1038/nn1525
- Casali, A. G., Gosseries, O., Rosanova, M., Boly, M., Sarasso, S., Casali, K. R., Casarotto, S., Bruno, M.-A., Laureys, S., Tononi, G., & Massimini, M. (2013). A theoretically based index of consciousness independent of sensory processing and behavior. Science Translational Medicine, 5(198), 198ra105. https://doi.org/10.1126/scitranslmed.3006294
- Gordon, P. C., Song, Y. F., Jovellar, D. B., Rostami, M., Belardinelli, P., & Ziemann, U. (2023). Untangling TMS-EEG responses caused by TMS versus sensory input using optimized sham control and GABAergic challenge. The Journal of Physiology, 601(10), 1981–1998. https://doi.org/10.1113/JP283986
- Hallett, M. (2007). Transcranial magnetic stimulation: A primer. Neuron, 55(2), 187–199. https://doi.org/10.1016/j.neuron.2007.06.026
- Hamada, M., Murase, N., Hasan, A., Balaratnam, M., & Rothwell, J. C. (2013). The role of interneuron networks in driving human motor cortical plasticity. Cerebral Cortex, 23(7), 1593–1605. https://doi.org/10.1093/cercor/bhs147
- Horvath, J. C., Forte, J. D., & Carter, O. (2015). Quantitative review finds no evidence of cognitive effects in healthy populations from single-session transcranial direct current stimulation (tDCS). Brain Stimulation, 8(3), 535–550. https://doi.org/10.1016/j.brs.2015.01.400
- Huang, Y.-Z., Edwards, M. J., Rounis, E., Bhatia, K. P., & Rothwell, J. C. (2005). Theta burst stimulation of the human motor cortex. Neuron, 45(2), 201–206. https://doi.org/10.1016/j.neuron.2004.12.033
- Ilmoniemi, R. J., Virtanen, J., Ruohonen, J., Karhu, J., Aronen, H. J., Näätänen, R., & Katila, T. (1997). Neuronal responses to magnetic stimulation reveal cortical reactivity and connectivity. NeuroReport, 8(16), 3537–3540. https://doi.org/10.1097/00001756-199711100-00024
- Kravitz, A. V., Freeze, B. S., Parker, P. R. L., Kay, K., Thwin, M. T., Deisseroth, K., & Kreitzer, A. C. (2010). Regulation of parkinsonian motor behaviours by optogenetic control of basal ganglia circuitry. Nature, 466(7306), 622–626. https://doi.org/10.1038/nature09159
- Massimini, M., Ferrarelli, F., Huber, R., Esser, S. K., Singh, H., & Tononi, G. (2005). Breakdown of cortical effective connectivity during sleep. Science, 309(5744), 2228–2232. https://doi.org/10.1126/science.1117256
- Nitsche, M. A., & Paulus, W. (2000). Excitability changes induced in the human motor cortex by weak transcranial direct current stimulation. The Journal of Physiology, 527(3), 633–639. https://doi.org/10.1111/j.1469-7793.2000.t01-1-00633.x
- Olds, J., & Milner, P. (1954). Positive reinforcement produced by electrical stimulation of septal area and other regions of rat brain. Journal of Comparative and Physiological Psychology, 47(6), 419–427. https://doi.org/10.1037/h0058775
- Opitz, A., Paulus, W., Will, S., Antunes, A., & Thielscher, A. (2015). Determinants of the electric field during transcranial direct current stimulation. NeuroImage, 109, 140–150. https://doi.org/10.1016/j.neuroimage.2015.01.033
- Pascual-Leone, A., Walsh, V., & Rothwell, J. (2000). Transcranial magnetic stimulation in cognitive neuroscience—Virtual lesion, chronometry, and functional connectivity. Current Opinion in Neurobiology, 10(2), 232–237. https://doi.org/10.1016/S0959-4388(00)00081-7
- Polanía, R., Nitsche, M. A., & Ruff, C. C. (2018). Studying and modifying brain function with non-invasive brain stimulation. Nature Neuroscience, 21(2), 174–187. https://doi.org/10.1038/s41593-017-0054-4
- Rosanova, M., Gosseries, O., Casarotto, S., Boly, M., Casali, A. G., Bruno, M.-A., Mariotti, M., Boveroux, P., Tononi, G., Laureys, S., & Massimini, M. (2012). Recovery of cortical effective connectivity and recovery of consciousness in vegetative patients. Brain, 135(4), 1308–1320. https://doi.org/10.1093/brain/awr340
- Rossi, S., Antal, A., Bestmann, S., Bikson, M., Brewer, C., Brockmöller, J., Carpenter, L. L., Cincotta, M., Chen, R., Daskalakis, J. D., Di Lazzaro, V., Fox, M. D., George, M. S., Gilbert, D., Kimiskidis, V. K., Koch, G., Ilmoniemi, R. J., Lefaucheur, J.-P., Leocani, L., Lisanby, S. H., . . . Hallett, M. (2021). Safety and recommendations for TMS use in healthy subjects and patient populations, with updates on training, ethical and regulatory issues: Expert guidelines. Clinical Neurophysiology, 132(1), 269–306. https://doi.org/10.1016/j.clinph.2020.10.003
- Rounis, E., Maniscalco, B., Rothwell, J. C., Passingham, R. E., & Lau, H. (2010). Theta-burst transcranial magnetic stimulation to the prefrontal cortex impairs metacognitive visual awareness. Cognitive Neuroscience, 1(3), 165–175. https://doi.org/10.1080/17588921003632529
- Siebner, H. R., Bergmann, T. O., Bestmann, S., Massimini, M., Johansen-Berg, H., Mochizuki, H., Bohning, D. E., Boorman, E. D., Groppa, S., Miniussi, C., Pascual-Leone, A., Huber, R., Taylor, P. C. J., Ilmoniemi, R. J., De Gennaro, L., Strafella, A. P., Kähkönen, S., Klöppel, S., Frisoni, G. B., George, M. S., . . . Rossini, P. M. (2009). Consensus paper: Combining transcranial stimulation with neuroimaging. Brain Stimulation, 2(2), 58–80. https://doi.org/10.1016/j.brs.2008.11.002
- Silvanto, J., Muggleton, N., & Walsh, V. (2008). State-dependency in brain stimulation studies of perception and cognition. Trends in Cognitive Sciences, 12(12), 447–454. https://doi.org/10.1016/j.tics.2008.09.004
- Tremblay, S., Rogasch, N. C., Premoli, I., Blumberger, D. M., Casarotto, S., Chen, R., Di Lazzaro, V., Farzan, F., Ferrarelli, F., Fitzgerald, P. B., Hui, J., Ilmoniemi, R. J., Kimiskidis, V. K., Kugiumtzis, D., Lioumis, P., Pascual-Leone, A., Pellicciari, M. C., Rajji, T., Thut, G., Zomorrodi, R., . . . Daskalakis, Z. J. (2019). Clinical utility and prospective of TMS–EEG. Clinical Neurophysiology, 130(5), 802–844. https://doi.org/10.1016/j.clinph.2019.01.001
- Vöröslakos, M., Takeuchi, Y., Brinyiczki, K., Zombori, T., Oliva, A., Fernández-Ruiz, A., Kozák, G., Kincses, Z. T., Iványi, B., Buzsáki, G., & Berényi, A. (2018). Direct effects of transcranial electric stimulation on brain circuits in rats and humans. Nature Communications, 9, 483. https://doi.org/10.1038/s41467-018-02928-3