RESEARCH DOSSIER
Brain–Behavior Relationships
Abstract. This fundamental research monograph investigates how organized neural tissues, circuits, pathways, and large-scale networks contribute to specific behavioral capacities and performances. It defines brain–behavior relationships through distinct claims about association, necessity, sufficiency, causal perturbation, and mechanistic implementation. Historical evidence from neurological cases and sensory physiology is examined alongside experimental reinforcement research, neural plasticity, computational models, modern functional imaging, lesion–symptom mapping, and network neuroscience. The analysis compares neural contributions to perception, attention, executive control, learning, memory, valuation, and action. Critical attention is given to anatomical coverage, behavioral construct validity, head-motion artifacts, model identifiability, individual differences, clinical translation, and replication of brain-wide associations. The synthesis concludes that valid neural explanations depend on precise behavioral outcomes and convergent methods that discriminate causal hypotheses; it identifies priorities for developmental, perturbational, longitudinal, and population-based research.
Keywords: brain–behavior relationships; behavioral neuroscience; neuropsychology; cognitive neuroscience; brain lesions; causal inference; functional neuroimaging; neural networks; neuroplasticity; neural representation; executive control; reinforcement learning; memory systems; brain-wide association studies; replication
The Scientific Problem of Brain–Behavior Relationships
Brain–behavior relationships concern how organized neural structures and physiological processes contribute to the capacities, performances, experiences, and adaptive actions of organisms. The explanatory object is a relation between an explicitly defined behavioral phenotype and a specified neural property under describable conditions. A movement, remembered episode, choice, spoken word, and change of attention each require distinct behavioral operations before the neural evidence can be interpreted. The field therefore joins behavioral measurement to anatomy, electrophysiology, causal intervention, and computational analysis. A relation may describe population covariance, an intervention effect, a necessary contribution to performance, or an implementing mechanism; these claims differ in the evidence they demand. The empirical program developed by Krakauer et al. (2017) gives behavior the role of defining the scientific phenomenon that the neural explanation must explain.
The principal research questions ask which neural organizations support particular behavioral functions, how localized components cooperate through long-range pathways, how changing brain states changes performance, and how developmental and experiential histories alter these relations. Additional questions concern whether measurements generalize beyond a selected task, whether the same neural signal can participate in several psychological operations, and whether different neural systems can sustain function after injury. The field consequently requires an account of functional specificity that accommodates overlapping representations and adaptive reorganization. The distinction between a neural correlate and a neural requirement is especially important: a measurable activation can accompany task execution without establishing that the activated region is indispensable. The neuropsychological analysis of Rorden et al. (2007) explains why disruption studies complement activation measurements.
This monograph is an integrative critical research synthesis organized around historical demonstrations, mechanistic questions, the principal experimental and observational methods, convergent evidence from major behavioral domains, and the limits of contemporary population prediction. Sources include original experimental reports, historical clinical investigations, major theoretical papers, and methodological studies. It is not a protocol-driven systematic review; no exhaustive search, screening flow, or pooled effect estimate is claimed. The scope is broader than any single neural circuit and narrower than a general theory of cognition: the concern throughout is how evidence licenses an inference from measurable neural organization to particular behavioral operations. A distinction between anatomical location and explanatory function is maintained when interpreting the classical and modern literature.
Localization, Association, and the Historical Formation of the Field
Nineteenth-century clinical neurology established that patterns of injury could reveal functional specialization within the cerebrum. The reports on speech impairment associated with left frontal lesions provided a decisive form of evidence for a nonuniform functional organization; later language syndromes emphasized temporal regions and the connections between specialized processes. Historical cases should be read as studies of particular patients with imperfect anatomical and behavioral measurements, rather than as complete maps of a single faculty. In the contemporary reinterpretation by Boes et al. (2015), consequences of focal injury can be situated within networks connected to damaged tissue. The historical development therefore encompasses localization of vulnerable nodes and the long-standing problem of disconnection between otherwise functioning neural components.
The case of Phineas Gage illustrates both the explanatory power and the interpretive difficulty of single-person evidence. Gage survived a penetrating frontal injury in 1848, and John Harlow subsequently described changes in conduct that drew attention to the relation between frontal systems and behavioral regulation. Retrospective narratives have sometimes made the case appear much more behaviorally definite than its surviving documentation warrants. Anatomical reconstruction, missing preinjury standardized measures, possible recovery, and the circumstances of later reports all affect interpretation. Modern analyses of historical cases show that damage can interrupt white-matter connections beyond the visible cortical injury Boes et al. (2015). The proper conclusion is that frontal injury may transform social and executive behavior, with the precise underlying processes requiring independent investigation.
Localization and distributed organization describe complementary properties of neural systems. Specialized receptor fields, cortical areas, and subcortical circuits impose constraints on what information can be transformed, while communication among them permits flexible behavior. In their network account of attention, language, and memory, Mesulam (1990) treated complex cognition as arising from interconnected processing systems. Such accounts revise any picture in which one psychological noun corresponds permanently to one anatomically isolated site. The historical lesson extends to interpretation: a precise lesion–symptom association supports a functional hypothesis, but lesion distribution, vascular territories, diaschisis, and compensatory activity influence which deficits are actually expressed. Contemporary brain–behavior science inherits the localization problem as a problem of multiscale causal organization.
Classical Clinical Evidence: Speech, Memory, and Hemispheric Disconnection
Clinical observations of acquired language impairment remain fundamental because injury permits examination of what functions disappear, survive, or reorganize. Broca's nineteenth-century work identified an association between damage in the left frontal lobe and severely impaired articulated speech, but contemporary language models distribute phonological, semantic, syntactic, and motor operations across interconnected systems. The patient phenotype is indispensable: difficulties with speech production, auditory comprehension, naming, repetition, and reading cannot be combined into one undifferentiated language score without loss of explanatory specificity. A lesion can also disrupt distant sites by disconnecting pathways, making the apparent cortical center only part of the causal story. Network methods offer a way to formalize and test these competing interpretations Boes et al. (2015).
The classic study by Scoville & Milner (1957) compared memory function after medial temporal resections and reported persistent difficulty forming recent memories associated with sufficiently extensive bilateral damage. Its enduring contribution is the separation of memory formation from other cognitive abilities and the demonstration that human memory has distinguishable components. Yet surgical resections changed several adjacent structures, the patient series was clinically selected, and later research was needed to distinguish episodic recollection, semantic acquisition, working memory, and different forms of skill learning. The medial temporal system includes interconnected hippocampal and adjacent cortical structures Squire & Zola-Morgan (1991); assigning an entire memory system to one anatomical label misstates the evidence. Strong inference combines lesion extent, detailed behavioral dissociations, and independent experiments across species.
Research on surgically separated cerebral hemispheres brought another kind of intervention evidence: manipulating information flow between neural systems changes which information can guide an overt response. Sperry's account of hemisphere deconnection Sperry (1968) exemplifies the importance of response channel, stimulus presentation, and task demand in interpreting a behavioral deficit. A verbal report and a manual choice may depend on different routes through the injured or disconnected nervous system. These experiments also demonstrate why clinical findings must be interpreted within the actual testing arrangements rather than projected into global claims about personal identity or ordinary behavior. Taken together, aphasia, amnesia, and disconnection cases establish functional differentiation and interdependence while demonstrating that an observed psychological function depends on a configuration of accessible neural pathways.
Experimental Foundations: Sensory Coding, Reinforcement, and Plasticity
Electrophysiological recording in animals made it possible to connect controlled sensory input with the response properties of identifiable neurons. The visual cortex experiments of Hubel & Wiesel (1962) characterized receptive fields and orientation-dependent responses under carefully defined visual conditions. These findings established neural selectivity at a tractable level of analysis and supported later accounts of hierarchical and distributed visual processing. Their explanatory target was a particular response property in a prepared visual system, not a complete account of conscious perception. The transformation from firing pattern to visually guided behavior requires connections among sensory processing, attention, decision processes, motor planning, and feedback. That gap is an experimental research problem, and it prevents assuming that a single response-selective neuron encodes a whole perceptual experience.
The demonstration by Olds & Milner (1954) that electrical stimulation at particular brain sites could reinforce behavior provided a causal connection between neural perturbation and behavioral preference. Stimulation, however, can influence multiple cell types and fibers and can induce states whose psychological interpretations depend on the experimental contingencies. Decades later, recording work linked particular dopaminergic signals with changes in expected reward Schultz et al. (1997), and human functional imaging studied partially differentiated ventral and dorsal striatal contributions during instrumental learning O'Doherty et al. (2004). These lines of evidence illuminate relations among prediction, action selection, and reinforcement, while leaving room for distinctions among dopamine release, neuronal firing, subjective pleasure, salience, and effort. Neural signal and behavioral construct must be operationalized separately.
Experience can also alter the organization that supports subsequent behavior. The long-lasting synaptic potentiation observed after patterned perforant-path stimulation in rabbits Bliss & Lømo (1973) supplied an experimentally demonstrable form of persistent physiological change. At a larger organizational scale, Nudo et al. (1996) documented use-dependent changes in primary motor representations following skilled learning in adult monkeys. Plasticity is thus demonstrated at multiple scales, with different interventions, measurements, and timescales. Neither laboratory potentiation nor a shifting cortical map alone establishes that a specific human memory, skill, or recovery process has been fully explained. To connect physiological adaptation to psychological learning, investigators must demonstrate a task-specific behavioral trajectory, identify relevant changing neural mechanisms, and test whether the neural changes mediate performance.
Causal Inference: Correlation, Necessity, Sufficiency, and Mediation
The phrase brain–behavior relationship covers several logical claims. An observational correlation says that a neural measurement varies with a behavioral measure across selected units. A necessary contribution is supported when an adequately selective disruption impairs the target capacity under specified conditions. Sufficiency is stronger and depends on whether an intervention produces the specified component of behavior in a defined context. Mediation asks whether an intervention's behavioral consequences are transmitted through an intermediate process according to an identified causal model. These categories require attention to confounding, selection, task demand, compensatory strategies, and measurement error. Rorden et al. (2007) shows why lesion analysis addresses questions that activation maps alone cannot resolve, though lesions themselves rarely constitute perfectly isolated experimental manipulations.
Human lesions provide natural interventions, but their distribution is constrained by vascular anatomy, disease, trauma, and survival. A patient with a large lesion may have several deficits, while lesions in neighboring patients can overlap because of shared vascular territories. Even statistically robust lesion maps can therefore reflect correlated anatomical damage rather than an isolated functional node. Moore et al. (2024) develops the conceptual and practical conditions for mapping lesions to symptoms, including lesion coverage and appropriate behavioral endpoints. Disruption by transcranial stimulation offers a different intervention with its own uncertainty about spatial selectivity, dosage, and secondary network effects. Intracranial stimulation and animal optogenetic methods permit further interventions but typically address different populations and levels of function. Convergence improves causal interpretation when these differences are recognized.
A change in behavior following a neural manipulation is an empirical result; the mechanism proposed to explain that change is a further scientific claim. Researchers should specify the intermediate variables, temporal predictions, and competing causal paths before declaring mediation. Causal explanation also depends on performance ceilings, learning histories, bodily feedback, and environmental affordances: an intact circuit may appear irrelevant in a task that fails to recruit it. A neural subsystem can be involved without being uniquely necessary because parallel circuits or strategies support successful performance. The challenge is therefore to discover intervention-sensitive dependency under particular circumstances. A mechanistic account of explanation Craver (2007) makes component organization and interaction central to this task while leaving empirical identification to specific designs.
Anatomical Specialization Within Distributed Neural Systems
Anatomical specialization is the patterned distribution of computational constraints, connectivity, cytoarchitecture, and physiological response properties across the nervous system. Functional organization emerges as specialized populations exchange information through recurrent pathways. Mesulam (1990) described large-scale networks for several cognitive domains in terms of interacting components rather than a collection of independent centers. This perspective allows researchers to ask whether a behavioral impairment arises from local loss of processing, damage to a communication pathway, altered coordination, or an adaptive response to widespread disruption. The anatomical unit relevant to a task may be a layer, circuit, pathway, cortical field, or coordinated system; choosing among these units follows the hypothesized operation and the spatial scale accessible to the measurement.
Structural connectivity describes physical pathways such as white-matter tracts, whereas functional connectivity summarizes statistical dependence between signals and effective connectivity attempts to model directed influences within an explicit framework. A region can be functionally correlated with another without a direct axonal connection, and the same anatomical tract can participate in different transient task configurations. In functional MRI, correlation can reflect neural coupling, common inputs, physiological artifacts, or preprocessing decisions. Power et al. (2011) used graph analyses to identify communities in resting-state correlation patterns, making distributed organization measurable but not automatically establishing the information transferred along every edge. Connecting these maps to behavior requires additional hypotheses about task-dependent communication, perturbation, and developmental variability.
The methodological implication is to treat every neural map as an estimate with defined dependencies. Parcellation resolution, signal preprocessing, motion correction, regional averaging, and the time interval analyzed can change the observed structure. Individual variation is meaningful when stable after accounting for these decisions, and population templates remain useful when their purpose and generalization limits are explicit. Yeo et al. (2011) derived widely used seven- and seventeen-network partitions from resting-state functional data in a large sample. These partitions are organizational descriptions: they do not prescribe an invariant list of psychological faculties or imply that each network has one exclusive function. A strong brain–behavior account specifies how network properties constrain behavior in a particular task and why alternative network configurations would predict different outcomes.
Perception and the Relation Between Neural Codes and Experience
Visual processing demonstrates the transition from highly specific neural responses to flexible behavioral interpretation. Orientation tuning and receptive-field organization documented in Hubel & Wiesel (1962) establish that cortical neurons respond systematically to features of visual input. Yet recognition, search, object constancy, and visually guided action require integration across neuronal populations and across time. The meaningful behavioral variable may be discriminability, response time, error structure, adaptation, or generalization to a new viewing condition. Each measure connects to the neural system differently. A neuron selective for a feature can contribute to recognition without being sufficient for recognizing a complete object; population responses can preserve stimulus information unavailable from a single recording.
Multivoxel pattern approaches moved analysis beyond the amount of activation in a small area. The study by Haxby et al. (2001) showed that object-related information in ventral temporal cortex could be represented by distributed, overlapping response patterns. This finding encouraged a representational approach in which identity and category structure are assessed through patterns across many measurements. A classifier that discriminates two stimulus conditions demonstrates information in the measurement under a specified experimental distribution. It does not by itself prove that the organism uses that information for the task, or that the same pattern would generalize across changes in attention, familiarity, task instruction, or scanner. Linking representation to behavior calls for tests of perceptual performance and selective intervention.
Perceptual experience also depends on endogenous state. Expectation, attention, action plans, and multisensory feedback alter the processing of the same physical stimulus. Such dependencies create an opportunity to distinguish feedforward encoding from recurrent modulation and decision-related activity. If a cortical pattern appears after the perceptual choice rather than before it, the observed relation may partially reflect the decision or its report. If task difficulty increases both neural signal and response time, a common cause can explain their association. A rigorous experiment therefore separates stimulus evidence, instructions, motor output, and subjective report whenever feasible. The goal is a model of transformations that predicts discriminations and errors under intervention rather than a static list of locations activated by perceptual tasks.
Attention and Cognitive Control as Network-Dependent Performance
Attention comprises distinguishable operations involving alertness, spatial selection, orienting, competition among stimuli, and sustained task priorities. Posner & Petersen (1990) articulated a network-oriented model that helped organize experimentally separable attentional functions. Behavioral measures such as cueing benefits, costs of invalid cues, omitted targets, and changes in response speed provide tractable dependent variables. Neural correlations become meaningful when tied to these patterns and to a theory of the operation being measured. Attention should not be treated as one homogeneous process because neural manipulations can change readiness, selection, or response strategy differently. The same behavioral score can arise from multiple underlying processes, and dissociations help decide between them.
An influential prefrontal theory Miller & Cohen (2001) proposes that maintained goal representations bias information processing in other parts of the brain. This account explains why frontal systems are implicated in flexible behavior across many tasks without requiring a distinct anatomical module for each task. The multiple-demand framework Duncan (2010) emphasizes a distributed frontoparietal system recruited by diverse cognitively demanding activities. These proposals converge on coordinated control while differing in the proposed representational architecture and the breadth of functions to be included. Behavioral demands and network organization must be measured in the same experiment if the neural interpretation is to distinguish task set, working memory, monitoring, response inhibition, and effort.
A patient who fails a complex executive test may struggle because of disrupted rule maintenance, speed, language comprehension, motivation, or motor execution. Likewise, prefrontal activation during a demanding task can reflect several of these components. Lesion and imaging results are most informative when test batteries contain separable measures and when process models predict how responses change across conditions. Developmental and clinical comparisons introduce further requirements: identical test scores can be achieved with different strategies, and measured signals can change with age or medication. Neural accounts of cognitive control should integrate convergent behavioral dissociations, anatomical findings, longitudinal changes, and interventions that selectively vary the demands of maintaining goals or resolving interference.
Learning and Memory Across Neural Subsystems
Memory research demonstrates that neural damage can selectively compromise particular forms of learning while leaving others partly available. The bilateral medial temporal cases analyzed in Scoville & Milner (1957) contributed an anatomical foundation for studying new declarative learning. Squire & Zola-Morgan (1991) subsequently organized evidence for a medial temporal memory system connecting the hippocampus and neighboring cortices with distributed long-term representations. Contemporary interpretation distinguishes encoding, consolidation, retrieval, recognition, familiarity, and different memory contents. A behavioral memory score combines these operations unless the task design separates them. A lesion-associated deficit therefore motivates, but does not complete, a theory of which operation has been impaired and how neural mechanisms support it.
The relation between synaptic plasticity and memory is supported through multiple experimental programs, including the demonstration of persistent synaptic responses following patterned stimulation in Bliss & Lømo (1973). That observation shows how activity can produce enduring physiological modification, a plausible ingredient of experience-dependent learning. The mapping from synaptic change to human autobiographical memory is nevertheless many-to-many: a learning episode recruits distributed circuits, and synaptic modifications support numerous forms of adaptation. Strong linking evidence would connect behaviorally measured acquisition to relevant physiological dynamics and then test how experimentally manipulating those dynamics alters acquisition or retention. Such a program also requires attention to time, since short-term state changes, longer-term stabilization, and remote retrieval occur under different constraints.
Memory-related brain–behavior associations can be distorted by the conditions of testing. Repeated exposure creates practice effects, memory tasks often depend on attention and language, and emotional salience changes both encoding and reporting. Injury or disease may affect several components at once, while preserved performance can involve compensatory strategies. Comparisons among recognition, free recall, priming, procedural learning, and delayed retrieval are therefore more explanatory than a single global measure. The conceptual outcome is a system of conditional dependencies: distinct neural ensembles and pathways make identifiable contributions to learning under particular tasks, while distributed representations, adaptation, and interactions with other systems permit flexible behavior.
Reward, Value, Action Selection, and Motor Organization
Behavioral regulation by consequences supplied an early causal bridge between neural stimulation and action. The self-stimulation findings in Olds & Milner (1954) showed that activation of selected brain sites could modify subsequent response probability under reinforcement contingencies. Modern learning models distinguish prediction of reward, evaluation of actions, selection policy, and experienced outcome. Electrophysiological investigations in Schultz et al. (1997) identified firing patterns associated with differences between predicted and received rewards under particular learning conditions. These responses are important observations about neural coding, while broader claims about motivation or enjoyment require additional behavioral evidence. Neural activation must be situated within what the organism learned, what options were presented, and which actions could be performed.
Human instrumental-learning research has investigated differentiated roles of striatal regions, including the experimental functional-imaging results of O'Doherty et al. (2004). The study linked task-defined computational variables to activity in ventral and dorsal striatal territories, supporting a functional hypothesis about distinct components of reinforcement learning. Its inference rests on the assumed model, the measured blood-oxygen signal, and the particular task conditions; it cannot establish that actor and critic processes are anatomically exclusive. Model-based analysis gains explanatory strength when competing learning rules make separable behavioral predictions, neural signals are estimated without circular selection, and manipulations alter reward contingency or action availability in ways anticipated by the model.
Motor control makes neural–behavioral coupling especially clear because muscular actions, trajectories, errors, and adaptation can be recorded continuously. Skilled movement depends on spinal circuitry, sensory feedback, cerebellar and basal ganglia contributions, and cortical organization. Nudo et al. (1996) provided direct evidence of training-related changes in motor cortical maps in monkeys, making experience-dependent representation an experimental variable. Translating these findings to human rehabilitation requires further controls for spontaneous recovery, changing task difficulty, and repeated practice. Motor performance is also embodied: biomechanics and feedback alter the relation between activity in a neural circuit and the resulting movement. A complete account follows neural changes through system dynamics to observable corrections and learned skill.
Lesion–Symptom Mapping and the Problem of Anatomical Coverage
Lesion-based research analyzes the relation between an anatomical disruption and a specified loss or change in performance. Individual cases are useful for generating mechanistic hypotheses, while group approaches can estimate whether particular locations or pathways are statistically associated with impairment. The practical guide by Moore et al. (2024) describes the logic and limitations of lesion mapping in modern neuropsychology. Because damage is neither randomly allocated nor evenly distributed throughout the brain, anatomical coverage determines which regional claims can be estimated. A null result in poorly sampled tissue is uninformative about necessity. Lesion size, vascular dependence, covarying affected structures, and preexisting differences between patients can generate misleading associations. High-quality studies report these properties and examine the sensitivity of their maps to analytic assumptions.
Outcome measurement presents a parallel set of problems. A language score can combine articulation, comprehension, naming, and general processing speed; a memory score may depend on attention and response method. Mapping an aggregate phenotype can obscure which underlying operation depends on the damaged region. Rorden et al. (2007) advocates improved strategies for identifying lesion–symptom relations, including attention to statistical reliability and lesion coverage. Modern analyses may adjust for lesion volume, compare multivariate models, and test whether deficits are predicted by disconnection measures. Each approach imposes its own assumptions; controlling for lesion volume, for example, can change the causal quantity estimated if volume is part of the damage process. Robust conclusions compare explicitly declared estimands across reasonable alternatives.
Clinical injury data also pose temporal questions. Acute swelling, reduced blood flow, and transient network dysfunction may contribute to symptoms that later resolve. Rehabilitation, compensatory learning, and changes in environmental support can alter a patient's observable capacity. A single lesion scan and examination cannot determine whether the affected tissue supports the initial acquisition of a behavior, its routine execution, or long-term recovery. Longitudinal designs with repeated sensitive behavioral measures are preferable for those distinctions. The approach of Boes et al. (2015) adds functional-network connectivity associated with lesion sites, suggesting a path from dispersed anatomical injuries to shared symptoms. Such analyses remain dependent on connectivity reference data and should be combined with direct observations of each patient's network and recovery when available.
Electrophysiology, Hemodynamic Imaging, and Their Measurement Models
Neural recording methods sample different physical correlates of activity. Extracellular spikes reflect firing by sampled neurons; local field potentials combine nearby electrical processes; scalp electroencephalography captures spatially mixed electrical signals; and functional MRI estimates a vascular response related indirectly to neural activity. Measurement resolution, noise, selection of sampled tissue, and population accessibility differ across these approaches. In the physiological study of Hubel & Wiesel (1962), spike responses were directly related to controlled stimulus manipulations. Contemporary behavioral neuroscience often requires a similar alignment between the measurement model and the behavioral hypothesis, even when signals are spatially extensive or measured from humans. The label activity by itself hides distinctions that matter for mechanistic explanation.
Task functional MRI offers a statistical contrast between measured hemodynamic signals under specified conditions. When a region responds more strongly during a decision task, that evidence can support a forward inference about task sensitivity, provided design and analysis are sound. It does not establish a one-to-one mapping between a region and a mental process. Poldrack (2006) analyzed reverse inference as a probabilistic question dependent on how selectively the region responds across candidate processes. A region recruited during anxiety, for example, cannot identify the presence of anxiety without discriminating evidence against alternative explanations. Multivariate decoding permits quantitative predictions but similarly depends on training distributions, independent validation, and the psychological interpretation of the labels.
Resting-state connectivity methods analyze covariation in signals when no specific experimental task is imposed. Yeo et al. (2011) reported a reproducible large-scale partition of the human cortex, and Power et al. (2011) described functional communities using graph representations. Both studies contribute major organizational maps, but neither directly demonstrates what a participant was thinking during every correlated fluctuation. Motion can produce systematic spatial patterns in measured correlations, a concern documented by Satterthwaite et al. (2012). Respiration, cardiac physiology, scanner variation, and the analytic pipeline can also influence the estimate. Task and resting data are strongest when their limitations are checked independently and when their predictions about behavior survive alternative preprocessing and sampling choices.
Perturbation, Stimulation, and the Interpretation of Neural Interventions
Controlled stimulation changes a neural system and permits direct tests of the behavioral consequences. Historical electrical self-stimulation experiments Olds & Milner (1954) linked experimental activation to response probability under reinforcement conditions. Modern approaches include transcranial magnetic stimulation, electrical stimulation of cortical or subcortical tissue, pharmacological perturbation, and cell-type-sensitive methods in animals. They differ in selectivity, ethical constraints, timing, and the classes of mechanism that can be addressed. A stimulation effect provides a counterfactual comparison when the control condition is well chosen. It does not necessarily isolate a single cell population or psychological faculty: current spread, afferent pathways, sensory sensations associated with stimulation, and network-level propagation may all contribute.
Interpretability requires defining the intervention target before observing the behavioral effect. A reduction in response speed after stimulation can reflect altered motor execution, attention, arousal, or strategy, and the conditions of a control intervention determine which explanations are excluded. Dose–response comparisons, temporal targeting relative to task events, blinding, and independent behavioral measures strengthen causal analysis. Cross-species work can supply more selective interventions but introduces translation problems because sensory ecology, learned behavior, and circuitry vary across organisms. Studies of training-dependent motor representations Nudo et al. (1996) underscore the importance of measuring performance and neural adaptation together. A causal account of a behavior must demonstrate that the targeted change affects the operation named by the hypothesis.
Stimulation studies are particularly useful when compared with lesions and passive recordings. Activation recorded during performance establishes engagement; disruption that selectively impairs that performance suggests functional dependence; stimulation that changes a process at an appropriate phase can distinguish competing mechanisms. But a lack of impairment may reflect incomplete targeting or compensation rather than irrelevance, and stimulation-induced behavior need not mimic the natural computation. Network propagation adds additional ambiguity: an intervention at one site can produce changes in distant connected tissue. A realistic causal synthesis consequently treats the intervention as affecting an organized system and makes explicit predictions about both proximal physiological changes and downstream behavioral consequences.
Computational Models and Neural Representations of Behavior
Computational models translate descriptions of behavior into operations with parameters that can be estimated and tested. A reinforcement-learning model might distinguish an expected value, a prediction error, an action-selection temperature, and a learning rate. Each parameter receives empirical meaning through task contingencies and measured choices. Linking that model to neural evidence is informative when trial-by-trial predictions account for variance beyond simpler alternatives and when behavioral fit generalizes to new conditions. Schultz et al. (1997) provided a seminal connection between reward-prediction computations and neuronal responses, and O'Doherty et al. (2004) related computational components of instrumental learning to human imaging signals. Neither correspondence, by itself, proves that a formal algorithm is uniquely realized by a particular brain structure.
Models of cognitive control similarly predict how goals influence subsequent transformations. The integrative proposal by Miller & Cohen (2001) connects maintained prefrontal activity with biased processing elsewhere. The framework in Duncan (2010) emphasizes flexible multiple-demand recruitment across task domains. These models can be compared using experimental manipulations that separately change goal maintenance, conflict, novelty, and memory load. The relevant question is whether the competing proposals yield different neural and behavioral predictions under controlled conditions. Large numbers of regressors or elaborate architecture diagrams do not guarantee identification; different model parameters can produce nearly indistinguishable responses, and correlated model variables can make neural associations unstable.
Representational analyses raise a further distinction between decodability and utilization. A statistical classifier may recover stimulus category, response direction, or reported emotion from a distributed signal. That observation establishes information available to the classifier in its sampling distribution, not necessarily information used causally by the organism. Haxby et al. (2001) demonstrated distributed representations for objects, helping expand the idea of a neural code beyond an isolated feature detector. Claims about representation become stronger when patterns predict behavior out of sample, transfer to new stimuli, change with interventions in a theoretically anticipated way, and are interpreted with explicit alternative models. The significance of neural coding depends ultimately on how the coded distinction participates in the behavior being explained.
Convergence Across Methods and the Hierarchy of Evidence
Evidence for a brain–behavior relationship gains credibility when independent techniques make compatible predictions about the same clearly measured function. Anatomy identifies plausible pathways, electrophysiology constrains signal dynamics, perturbation tests dependence, neuropsychology reveals selective impairments, and behavioral experiments test functional consequences. Their agreement can eliminate explanations that fit any single method alone. For example, a cognitive-control account becomes stronger when tasks distinguish goal maintenance from response conflict, distributed neural activity tracks those operations, and targeted perturbations selectively change the predicted component. This is convergence across claims and designs, rather than an accumulation of pictures of the brain. The mechanistic framework of Craver (2007) helps organize proposed components and their operations.
The hierarchy of evidence follows the inference target. Randomized perturbation is especially useful for a causal effect under a well-defined intervention, whereas longitudinal observation may be best suited for describing development and large representative samples for estimating population associations. Lesion studies supply distinctive necessity evidence when measurement and anatomical assumptions are sufficiently satisfied. A large cross-sectional correlation remains relevant to prediction without establishing an intervention effect. Marek et al. (2022) showed the scale of sampling needed for stable discovery of many population-wide neuroimaging associations, while Kang et al. (2024) demonstrated that design choices and the distinction between within-person and between-person variation materially affect replicability. The value of a study therefore depends on the question it can answer.
Contradictory results should be analyzed by determining whether the underlying tasks, populations, measurements, and estimands genuinely coincide. A lesion may produce impairment in a demanding task but negligible change in everyday activity supported by redundancy or adaptation. A brain–behavior association may replicate among older participants but disappear among younger people with less variation in the measured characteristic. A classifier can predict within the original scanning center yet fail elsewhere because of site effects. Such differences can reveal the boundary conditions of the phenomenon. The appropriate synthesis preserves the distinction between an observed result, a statistical generalization, and a proposed mechanistic explanation, and specifies what further evidence would separate the alternatives.
Neurodevelopment, Learning Histories, and Changing Neural Dependencies
Brain–behavior relationships change as organisms develop and acquire experience. The same behavioral response can depend on different strategies at different ages, and the corresponding neural measurements can change as anatomy, connectivity, and task familiarity develop. Developmental studies should distinguish a difference in capacity from a difference in the strategy used to express that capacity. Age-related patterns of activation may reflect efficiency, task difficulty, compensatory recruitment, or changes in vascular physiology. Likewise, the maturation of a network is an anatomical or physiological trajectory whose behavioral importance must be established through linked measurements. Yeo et al. (2011) provides a population-level description of large-scale network organization, but developmental explanations require age-sensitive evidence and repeated observations beyond the adult template.
Experience-dependent cortical organization is supported by the animal motor-learning results of Nudo et al. (1996), while physiological plasticity mechanisms were experimentally investigated by Bliss & Lømo (1973). These studies concern different species, structures, and endpoints; they should not be conflated into a single universal effect of training. Learning a skilled movement may alter the distribution of cortical representations while simultaneously engaging changes in other circuits, muscles, attention, and feedback use. A mechanistic study should examine whether behavioral change precedes, coincides with, or follows physiological reorganization, and whether the relationship is preserved when practice amount and task difficulty are controlled. Repeated measures help to distinguish stable individual differences from the within-organism changes produced by acquisition.
Recovery after neurological injury adds a different developmental timescale. Spontaneous improvement, functional substitution, task adaptation, and environmental accommodation can all change outcome measures without restoring the exact preinjury implementation. A postinjury increase in activation may represent successful compensation, inefficient effort, disinhibition, or an epiphenomenal response. These possibilities require longitudinal trajectories and interventions that distinguish behavioral gains from changes in neural signal. The lesion-network approach of Boes et al. (2015) underscores that a focal injury can disturb functionally connected regions, while treatment can potentially alter communication beyond the injury. The scientific question is which observed neural changes account for the recovery of a specified behavioral operation and under what conditions that relation generalizes.
Individual Differences, Sampling, and Cross-Population Generalization
Much human neuroimaging research seeks correlations between brain-derived variables and differences in cognition, personality, psychopathology, or performance. These studies typically estimate between-person relationships, and their conclusions depend on the range of the measured trait, demographic composition, reliability of each variable, and confounding by acquisition procedures. A small brain–behavior correlation can be genuine but poorly reproducible in a limited sample. Marek et al. (2022) systematically investigated the replicability of brain-wide association studies and showed that stable population estimates often require samples numbering in the thousands. Its message concerns the statistical power and reproducibility of discovery across many comparisons, rather than the invalidity of all small, well-controlled experimental studies.
Sampling strategy affects the estimand. A cohort selected to represent a narrow age group will contain less age variance than a lifespan sample, while clinical case–control sampling changes both the behavioral distribution and the likely biological correlates. Kang et al. (2024) analyzed 63 longitudinal and cross-sectional MRI studies involving 77,695 scans and demonstrated that design modifications can substantially influence standardized effects and replicability. The paper also showed that longitudinal sampling is not uniformly superior: within-person and between-person associations can differ. Consequently, a cross-sectional coefficient should not be interpreted as the effect expected when one person changes over time. Reports should specify whether the claim concerns ranking individuals, tracking individual change, or estimating the response to an intervention.
Generalization also depends on measurement equivalence and the environment. A cognitive task developed in one language or educational context may recruit different strategies elsewhere, even when the summary accuracy is similar. Scanner hardware, preprocessing, recruitment site, and motion distributions can introduce patterns that are predictive of group membership but irrelevant to the proposed psychological mechanism. Satterthwaite et al. (2012) demonstrated systematic head-motion effects on functional connectivity measurement and their potential relevance for developmental comparisons. Robust research uses independent validation samples, sensitivity analyses, transparent exclusion rules, and explicit reporting of the population to which the result applies. Individual differences become meaningful brain–behavior explanations when both sides of the association have interpretable constructs and credible transport beyond the original dataset.
Clinical Neuropsychology, Symptoms, and the Limits of Localization
Clinical neuropsychology makes functional dependence visible through impairment, but a clinical syndrome is seldom the output of one isolated neural component. Language difficulties, amnesia, disturbances of spatial attention, motor deficits, and changes in social behavior may each involve overlapping networks and multiple cognitive operations. Moore et al. (2024) recommends integrating anatomical lesion data with suitable psychological measures while considering lesion distribution and methodological bias. Neuropsychological profiles are therefore most informative when symptom dimensions, preserved capacities, premorbid characteristics, and associated neurological changes are characterized separately. A diagnosis names a clinical grouping; it does not by itself establish that all individuals sharing that grouping possess the same neural mechanism.
Symptoms following a focal lesion may sometimes map to the functional connectivity of the lesion site more consistently than to a common damaged location. Boes et al. (2015) supplied a method for investigating such network localization across lesions producing related neurological outcomes. This expands the possibilities for mechanistic hypotheses and clinical interpretation, but it also introduces assumptions about reference connectivity and variability across individuals. Clinical extrapolation should test whether a derived network predicts independent patients and whether selective manipulation of a network node changes the target symptom. Any proposed treatment additionally requires evidence of safety, effectiveness, and benefit for an appropriately characterized patient population; anatomical plausibility alone does not provide such evidence.
Neural differences reported in psychiatric samples are especially vulnerable to causal ambiguity. A measured difference may precede symptoms, reflect illness duration, follow medication exposure, vary with sleep or stress, or arise from selection and measurement conditions. The population-replicability analysis in Marek et al. (2022) and the design-focused evidence in Kang et al. (2024) bear directly on the interpretation of small imaging studies of psychological traits. A region-level association is an invitation to refine a functional model, not an individual diagnostic rule. Research translation should identify whether an observed signal improves prediction beyond available behavioral and clinical data, whether it is reliable for a specific person, and whether an actionable intervention changes outcomes in controlled studies.
Neural Signatures, Prediction, and the Boundary Between Decoding and Explanation
Multivariate neural signatures attempt to predict outcomes from distributed patterns rather than single-area activity. In four studies totaling 114 participants, Wager et al. (2013) developed and tested a functional-imaging signature related to experimentally induced physical pain. The work illustrated how independent conditions and multivariate classification could strengthen prediction of a well-defined experimental target. Its outcome must nevertheless be interpreted with respect to the stimuli, tasks, selected samples, and validation conditions studied. A successful prediction of heat-evoked pain intensity does not establish a universal biological test for every painful condition, nor does it prove that the predictive pattern is the minimum circuitry necessary to generate pain. Predictive performance and mechanistic necessity remain separate research claims.
Any decoder can exploit correlations that are stable within its development sample, including those caused by instructions, response strategy, physiological artifacts, or a confounded training set. Poldrack (2006) explains the limits of inferring a mental process from a regional signal when the signal is shared among processes. Multivariate approaches can improve discrimination if the model is trained and tested independently, the feature-selection procedure is nested within validation, and the label has construct validity. External validation should involve new individuals and, where relevant, new scanners, task configurations, and populations. Calibrated performance for group averages cannot be assumed to deliver accurate individual-level conclusions, especially when base rates and measurement quality change.
Mechanistic explanation demands a more complete chain than prediction. Investigators must show how the measured signals arise, how they represent or transform information relevant to the behavior, and how intervention on relevant components alters the behavior under competing model predictions. The coding findings of Haxby et al. (2001) and the reinforcement-learning analyses in O'Doherty et al. (2004) provide methodological precedents for connecting multivariate and computational constructs with performance. Both also reveal a general problem: several neural organizations can yield similar behavioral outputs, and the same measured neural property may support several functions. A reliable decoder is valuable as a scientific instrument; claims of causal interpretation require further converging tests.
Evidence Quality, Replication, and Responsible Interpretation
Replicability is a property of an explicitly specified result under a defined replication design. It includes independent sampling, consistent operationalization, suitable measurement reliability, and an analytic procedure sufficiently transparent to reproduce the claim. Marek et al. (2022) documented a fundamental sample-size problem for many brain-wide associations, and Kang et al. (2024) demonstrated that sampling variability and longitudinal structure can alter that problem substantially. These findings motivate greater statistical precision and careful specification of the research target. They do not imply that every neural experiment requires a population cohort of thousands: tightly controlled within-participant interventions and high-dimensional population correlations address different questions, signal-to-noise conditions, and effect structures.
Methodological quality begins with reliable behavioral constructs. Performance scores should be evaluated for test–retest consistency, ceiling or floor effects, practice effects, and the possibility that respondents use different strategies. Neural measures require transparent preprocessing, artifact evaluation, motion monitoring, quality thresholds, and sensitivity to analytical decisions. Multiple comparisons, flexible model selection, selective outcome reporting, and circular feature choice inflate apparent evidence. The specific motion artifacts described by Satterthwaite et al. (2012) show that a nuisance variable can imitate biologically meaningful variation. A scientifically useful report therefore includes enough information to determine how the primary effect was defined and whether reasonable alternate choices change it.
Research interpretation must preserve uncertainty about the causal mechanism where evidence does not discriminate alternatives. Functional imaging generally supports statistical task associations; lesion and perturbation designs can bear more directly on dependence but contain their own sources of bias; computational models explain more when they yield uniquely testable predictions. The interdisciplinary argument of Krakauer et al. (2017) gives precise behavioral analysis a foundational role in neuroscience, because a neural mechanism is intelligible only relative to a behavior whose form and conditions are well defined. Critical synthesis is achieved by making these inferential boundaries explicit and by identifying the next decisive experiment rather than by describing neural maps as self-evident explanations.
Competing Explanatory Frameworks and Unresolved Mechanistic Disputes
Contemporary disputes over brain–behavior relationships often arise from different definitions of the explanatory target. A localization account may ask which tissue is required for a function, a network account may ask how the operation is coordinated across regions, and a computational account may ask what transformation explains performance. These questions generate different empirical predictions even when the same data are used. Mesulam (1990) articulated a distributed neurocognitive architecture, while Miller & Cohen (2001) developed a model of goal-dependent prefrontal bias and Duncan (2010) described recruitment of a broadly useful multiple-demand system. Their claims need comparison at the level of measurable task operations and intervention-sensitive predictions, not merely resemblance between anatomical diagrams.
A second dispute concerns what should count as a neural representation. Experimental selectivity is one measurable property, decodability another, and causal use in guiding an action a third. The early recordings in Hubel & Wiesel (1962) and the distributed pattern analysis in Haxby et al. (2001) address distinct versions of the representation problem. Representation becomes a stronger mechanistic claim when the theorized content predicts errors, invariance, generalization, and responses to targeted changes in input or circuitry. Otherwise, a successful pattern classification may be compatible with alternative computational explanations. Similar interpretive questions arise when a prediction-error signal is detected: Schultz et al. (1997) supplies physiological evidence relevant to learning models, while the specific role of the signal in behavior requires independent manipulations.
A third dispute is methodological. Some investigators prioritize richly measured individuals and causal perturbations, while others analyze large samples to estimate associations among population differences. Both approaches can supply relevant evidence because the corresponding hypotheses concern different quantities. The reproducibility results in Marek et al. (2022) and Kang et al. (2024) show why large and appropriately designed cohorts are important for many between-person associations. They do not substitute for carefully executed causal experiments, just as a striking patient dissociation does not estimate prevalence or a general population effect. The productive resolution is explicit specification of the target estimand, the model, the sources of bias, and the conditions under which two forms of evidence should converge.
From Biological Behavior to the Artificial Era: Scope of Interdisciplinary Translation
The history of brain–behavior research supplies principles of evidence that are useful beyond any single substrate: behavioral capacities require operational definition, measured internal organization can support competing functional interpretations, and causal claims demand interventions capable of separating alternatives. In the Artificial Era, research on artificial cognition and human–artificial collaboration can apply these principles while specifying its own mechanisms and performance criteria. The relevance is methodological rather than anatomical. Neural tissue, computational hardware, social interaction, and trained artificial systems have different causal organizations; a particular brain correlate cannot be transferred as proof of a corresponding state in an artificial system. Krakauer et al. (2017) reinforces the methodological lesson that observed behavior must remain a primary research object.
Comparative inquiry can ask how humans and artificial systems maintain information across tasks, adapt to feedback, coordinate actions, or distribute problem solving between agents. Such questions benefit from distinguishing externally successful performance from the mechanism that generates it. Experimental perturbations, ablation of computational components, and controlled changes in the information available to participants can reveal dependencies under specified conditions. The logic resembles the distinction between neural engagement and necessity developed in Rorden et al. (2007), though an intervention on a digital architecture is not equivalent to damage in a living nervous system. Transferable scientific standards include explicit task specification, attention to generalization, independent validation, and causal testing when feasible.
The philosophical framework Artificial Era · From Homo to Artificial motivates broader comparative research into the organization of cognition and behavior. Within this monograph, statements about artificial systems are research questions and conceptual applications, not empirical conclusions about subjective experience. The decisive scientific issue is which behavioral and organizational variables permit reliable comparison and what form of evidence would distinguish functionally similar performances with different internal mechanisms. A fruitful interface with Psychology of Artificial, Postsubjective Psychology, and the philosophy of Artificial Sapiens requires this methodological clarity and sustained attention to the distinct causal properties of each investigated system.
Priorities for Future Research and Stronger Scientific Designs
An immediate research priority is to improve the alignment between the psychological operation being studied and the neural measurement intended to explain it. Investigators should isolate identifiable sources of behavioral variability, articulate competing process models, and collect measurements whose temporal and spatial resolution are appropriate to the predicted mechanism. The result is a more discriminating study than one that correlates undifferentiated task scores with a whole-brain map. Moore et al. (2024) illustrates how carefully specified lesion analysis can strengthen functional inference, while Poldrack (2006) shows how the specificity of an activation pattern constrains the interpretation of a psychological process. The design should announce which conclusion would follow from each possible empirical outcome.
A second priority is to investigate causal dynamics and recovery across timescales. Repeated behavioral testing, longitudinal neural sampling, and appropriately controlled interventions can test whether a candidate neural change precedes improvement, mediates it, or follows compensation. The training and reorganization evidence in Nudo et al. (1996) provides a precedent for studying adaptive representations, while long-lasting synaptic potentiation Bliss & Lømo (1973) supplies a physiologically tractable mechanism that can be placed into more specific learning hypotheses. Future work should measure when several routes can support one behavior, which route is selected under changing demands, and how environmental support alters observable impairment.
A third priority is reproducible generalization. Large heterogeneous cohorts are needed for some population associations; deeply characterized within-person studies and perturbations are needed for other causal questions. Marek et al. (2022) and Kang et al. (2024) establish that sample design, variance, and repeated observations shape expected replicability. Independent replication should preserve the underlying estimand, while transport studies deliberately test different instruments, populations, scanners, and tasks. Transparent data provenance, validated behavioral outcomes, preregistered confirmatory contrasts where appropriate, and systematic analysis of uncertainty will strengthen the field. Research progress will be judged by mechanisms that survive discriminating tests and predict meaningful behavior beyond the original experimental setting.
Conclusion
Brain–behavior relationships are structured dependencies between neural organization and explicitly characterized behavioral functions. The historical sequence from clinical lesions through sensory physiology, reinforcement, plasticity, and modern imaging establishes that specialized tissue and distributed communication both contribute to behavior. Foundational reports such as Scoville & Milner (1957), Hubel & Wiesel (1962), Olds & Milner (1954), and Sperry (1968) opened distinct experimental routes into memory, perception, motivation, and functional disconnection. Later models clarified that neural responses become explanatory when interpreted relative to behavioral operations, competing mechanisms, and the functional connections that permit performance. The resulting field is unified by a scientific problem: how to infer the contribution of particular neural organizations to what organisms do.
The strongest evidence comes from coordinated designs that distinguish statistical prediction, task sensitivity, causal necessity, and explanatory mechanism. Lesion mapping Moore et al. (2024), network analysis Power et al. (2011), and theoretically informed experimental interventions can support different parts of that inference. The limits of reverse inference Poldrack (2006) and the replicability requirements of population-wide studies Marek et al. (2022) constrain what any one method can justify. Behavioral constructs, anatomical sampling, measurement reliability, study design, individual history, and contextual conditions jointly determine the validity of the result. Scientific integration therefore requires explicit descriptions of these conditions, not a universal claim that one class of methods reveals the whole relationship.
The mature research program treats brain and behavior as parts of an organized causal inquiry in which function is defined through testable performance, neural observations constrain mechanism, and interventions test the dependencies that theories predict. The open problems of dynamic reorganization, cross-population transfer, distributed computation, and individualized causal explanation remain central. Their resolution will improve psychological science, clinical neuropsychology, and carefully bounded comparative research across biological and artificial systems. The enduring conclusion is that reliable brain–behavior knowledge depends on rigorous functional measurement and the convergence of independent evidence across levels of neural organization.
References
- Bliss, T. V. P., & Lømo, T. (1973). Long-lasting potentiation of synaptic transmission in the dentate area of the anaesthetized rabbit following stimulation of the perforant path. The Journal of Physiology, 232(2), 331–356. https://doi.org/10.1113/jphysiol.1973.sp010273
- Boes, A. D., Prasad, S., Liu, H., Liu, Q., Pascual-Leone, A., Caviness, V. S., Jr., & Fox, M. D. (2015). Network localization of neurological symptoms from focal brain lesions. Brain, 138(10), 3061–3075. https://doi.org/10.1093/brain/awv228
- Craver, C. F. (2007). Explaining the brain: Mechanisms and the mosaic unity of neuroscience. Oxford University Press. https://doi.org/10.1093/acprof:oso/9780199299317.001.0001
- Duncan, J. (2010). The multiple-demand (MD) system of the primate brain: Mental programs for intelligent behaviour. Trends in Cognitive Sciences, 14(4), 172–179. https://doi.org/10.1016/j.tics.2010.01.004
- Haxby, J. V., Gobbini, M. I., Furey, M. L., Ishai, A., Schouten, J. L., & Pietrini, P. (2001). Distributed and overlapping representations of faces and objects in ventral temporal cortex. Science, 293(5539), 2425–2430. https://doi.org/10.1126/science.1063736
- Hubel, D. H., & Wiesel, T. N. (1962). Receptive fields, binocular interaction and functional architecture in the cat's visual cortex. The Journal of Physiology, 160(1), 106–154. https://doi.org/10.1113/jphysiol.1962.sp006837
- Kang, K., Seidlitz, J., Bethlehem, R. A. I., Xiong, J., Jones, M. T., Mehta, K., Keller, A. S., Tao, R., Randolph, A., Larsen, B., Tervo-Clemmens, B., Feczko, E., Miranda Dominguez, O., Nelson, S. M., Lifespan Brain Chart Consortium, Schildcrout, J., Fair, D. A., Satterthwaite, T. D., Alexander-Bloch, A., & Vandekar, S. (2024). Study design features increase replicability in brain-wide association studies. Nature, 636, 719–727. https://doi.org/10.1038/s41586-024-08260-9
- Krakauer, J. W., Ghazanfar, A. A., Gomez-Marin, A., MacIver, M. A., & Poeppel, D. (2017). Neuroscience needs behavior: Correcting a reductionist bias. Neuron, 93(3), 480–490. https://doi.org/10.1016/j.neuron.2016.12.041
- Marek, S., Tervo-Clemmens, B., Calabro, F. J., Montez, D. F., Kay, B. P., Hatoum, A. S., Donohue, M. R., Foran, W., Miller, R. L., Hendrickson, T. J., Malone, S. M., Kandala, S., Feczko, E., Miranda-Dominguez, O., Graham, A. M., Earl, E. A., Perrone, A. J., Cordova, M., Doyle, O., . . . Dosenbach, N. U. F. (2022). Reproducible brain-wide association studies require thousands of individuals. Nature, 603, 654–660. https://doi.org/10.1038/s41586-022-04492-9
- Mesulam, M.-M. (1990). Large-scale neurocognitive networks and distributed processing for attention, language, and memory. Annals of Neurology, 28(5), 597–613. https://doi.org/10.1002/ana.410280502
- Miller, E. K., & Cohen, J. D. (2001). An integrative theory of prefrontal cortex function. Annual Review of Neuroscience, 24, 167–202. https://doi.org/10.1146/annurev.neuro.24.1.167
- Moore, M. J., Demeyere, N., Rorden, C., & Mattingley, J. B. (2024). Lesion mapping in neuropsychological research: A practical and conceptual guide. Cortex, 170, 38–52. https://doi.org/10.1016/j.cortex.2023.10.001
- Nudo, R. J., Milliken, G. W., Jenkins, W. M., & Merzenich, M. M. (1996). Use-dependent alterations of movement representations in primary motor cortex of adult squirrel monkeys. Journal of Neuroscience, 16(2), 785–807. https://doi.org/10.1523/JNEUROSCI.16-02-00785.1996
- O'Doherty, J., Dayan, P., Schultz, J., Deichmann, R., Friston, K., & Dolan, R. J. (2004). Dissociable roles of ventral and dorsal striatum in instrumental conditioning. Science, 304(5669), 452–454. https://doi.org/10.1126/science.1094285
- 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
- Poldrack, R. A. (2006). Can cognitive processes be inferred from neuroimaging data? Trends in Cognitive Sciences, 10(2), 59–63. https://doi.org/10.1016/j.tics.2005.12.004
- Posner, M. I., & Petersen, S. E. (1990). The attention system of the human brain. Annual Review of Neuroscience, 13, 25–42. https://doi.org/10.1146/annurev.ne.13.030190.000325
- Power, J. D., Cohen, A. L., Nelson, S. M., Wig, G. S., Barnes, K. A., Church, J. A., Vogel, A. C., Laumann, T. O., Miezin, F. M., Schlaggar, B. L., & Petersen, S. E. (2011). Functional network organization of the human brain. Neuron, 72(4), 665–678. https://doi.org/10.1016/j.neuron.2011.09.006
- Rorden, C., Karnath, H.-O., & Bonilha, L. (2007). Improving lesion-symptom mapping. Journal of Cognitive Neuroscience, 19(7), 1081–1088. https://doi.org/10.1162/jocn.2007.19.7.1081
- Satterthwaite, T. D., Wolf, D. H., Loughead, J., Ruparel, K., Elliott, M. A., Hakonarson, H., Gur, R. C., & Gur, R. E. (2012). Impact of in-scanner head motion on multiple measures of functional connectivity: Relevance for studies of neurodevelopment in youth. NeuroImage, 60(1), 623–632. https://doi.org/10.1016/j.neuroimage.2011.12.063
- Schultz, W., Dayan, P., & Montague, P. R. (1997). A neural substrate of prediction and reward. Science, 275(5306), 1593–1599. https://doi.org/10.1126/science.275.5306.1593
- Scoville, W. B., & Milner, B. (1957). Loss of recent memory after bilateral hippocampal lesions. Journal of Neurology, Neurosurgery & Psychiatry, 20(1), 11–21. https://doi.org/10.1136/jnnp.20.1.11
- Sperry, R. W. (1968). Hemisphere deconnection and unity in conscious awareness. American Psychologist, 23(10), 723–733. https://doi.org/10.1037/h0026839
- Squire, L. R., & Zola-Morgan, S. (1991). The medial temporal lobe memory system. Science, 253(5026), 1380–1386. https://doi.org/10.1126/science.1896849
- Wager, T. D., Atlas, L. Y., Lindquist, M. A., Roy, M., Woo, C.-W., & Kross, E. (2013). An fMRI-based neurologic signature of physical pain. New England Journal of Medicine, 368(15), 1388–1397. https://doi.org/10.1056/NEJMoa1204471
- Yeo, B. T. T., Krienen, F. M., Sepulcre, J., Sabuncu, M. R., Lashkari, D., Hollinshead, M., Roffman, J. L., Smoller, J. W., Zöllei, L., Polimeni, J. R., Fischl, B., Liu, H., & Buckner, R. L. (2011). The organization of the human cerebral cortex estimated by intrinsic functional connectivity. Journal of Neurophysiology, 106(3), 1125–1165. https://doi.org/10.1152/jn.00338.2011