Salience Network: Insula, Attention, and Brain-State Switching
Author: Ukrainian Psychological Hub · Published: October 12, 2026 · Editorial Policy
The salience network (SN) is a distributed functional brain system associated with detecting biologically and behaviorally relevant events, coordinating attention, and integrating information about the body, environment, and current goals. Its widely described cortical hubs are the anterior insula and anterior cingulate or anterior midcingulate cortex. An unexpected sound, an important error, a painful signal, or a change in bodily needs can alter what the brain must prioritize. Functional connectivity studies suggest that salience-related regions interact with networks supporting internally directed cognition and task-focused control (Seeley et al., 2007; Menon and Uddin, 2010).
The term “brain-state switching” refers in this literature to changes in patterns of network participation, particularly between internal thought and externally directed demands. A prominent hypothesis assigns the anterior insula a coordinating role, although the evidence supports context-dependent, distributed interactions rather than a universally necessary switch operated by one cortical region. The meaning of “salience network” also varies among atlases, and its apparent functions depend on the task, participant, and measurement (Seeley, 2019; Uddin et al., 2023).
What Is the Salience Network?
A functional brain network is a set of regions whose measured activity exhibits systematic relationships. Functional connectivity commonly refers to statistical dependence between regional signals, whereas structural connectivity refers to anatomical pathways. The SN is identified through patterns involving anterior insular and cingulate territories and, depending on the research method, their broader cortical and subcortical partners. A network label is an operational description: it does not establish a rigid anatomical organ or a unique psychological faculty.
Salience is the significance of an event in its context. A brightly colored sign may stand out from its background through sensory contrast; a barely audible notification can be more relevant because of its learned meaning. Surprise, novelty, task relevance, motivational value, threat, and bodily discomfort can each influence priority, yet represent distinct explanatory variables. Some salient events recruit attention, while others may be acknowledged and ignored in the service of a more important goal. The SN participates in coordination of such priorities but is not the exclusive processor of sensory, motivational, or emotional importance (Uddin, 2015).
A practical definition is therefore: the salience network is a reproducible, distributed functional configuration centered on anterior insular and cingulate regions, studied for its contribution to detecting relevant demands and coordinating neural resources. This definition distinguishes the observed network pattern from specific, more speculative causal claims about how switching is implemented.
How the Network Was Discovered
Resting-state functional MRI made it possible to examine correlations among slow fluctuations in blood-oxygen-level-dependent (BOLD) signals when participants were not performing a prescribed task. Seeley and colleagues demonstrated dissociable intrinsic connectivity systems associated with salience processing and executive control in a landmark 2007 paper. The anterior insula and cingulate were prominent hubs of the salience-related organization, while lateral frontal and parietal regions characterized a partly distinct control system (Seeley et al., 2007).
In 2008, Sridharan, Levitin, and Menon investigated the relative timing and directional statistical relationships of activity among anterior insular regions, a central executive network, and the default mode network. They interpreted their findings as evidence that right frontoinsular regions helped initiate changes in other networks during salient events. That work gave the switching hypothesis much of its prominence. Directional models fitted to hemodynamic signals, however, do not have the causal specificity of interventions that directly manipulate identified neuronal populations (Sridharan et al., 2008).
Menon's triple-network framework subsequently connected the SN, default mode network, and central executive network into an influential account of cognition and psychopathology. Later research developed a wider range of interpretations involving homeostasis, interoception, distributed control, and subcortical arousal. Network naming has also become an explicit subject of debate. The field's history matters because popular presentations sometimes promote an original theoretical model to the status of a conclusively identified biological switch (Menon, 2011; Uddin et al., 2023).
Core Anatomy: Anterior Insula and Cingulate Cortex
The insula lies beneath the opercular cortex in the lateral sulcus and includes territories with differing structural connectivity and task-response profiles. Anterior regions integrate signals relevant to affect, bodily regulation, task demands, and control. Researchers often describe dorsal anterior insular areas as more associated with cognitive and attentional tasks, while ventral territories participate prominently in affective and autonomic functions. These tendencies overlap; the insula does not divide neatly into a single thought area and a single emotion area (Menon and Uddin, 2010).
A second principal cortical hub lies in dorsal anterior cingulate and anterior midcingulate territories. These regions contribute to performance monitoring, effort, action selection, conflict-related demands, and coordination of physiological responses. The precise borders of dorsal anterior cingulate, anterior midcingulate, and cingulo-opercular areas depend on the anatomical and functional conventions used. A cingulate response during an error does not automatically establish an isolated error detector, because effort, arousal, motor adjustment, and uncertainty can also change during the task.
The system has connections extending into frontal operculum, lateral frontal regions, ventral striatum, amygdala, thalamus, hypothalamus, and certain brainstem nuclei. Seeley's account treats these relationships as important to the system's regulation of homeostatic demands. Some studies reserve the term salience network for a narrower cortical set; others include more extensive subcortical partners. Readers should always inspect the map or parcellation rather than assume every study contains identical nodes (Seeley, 2019).
Importantly, a region is not identical to a network. The insula participates in several distinct anatomical and functional interactions and is also studied for interoception, pain, and subjective states. The N16 EN-30 article on insular anatomy will own the dedicated region-level overview when published; the present canonical owns the salience-network mechanism.
How Salience Influences Attention and Behavior
Anterior insular and cingulate activations recur across studies of unusual sensory events, performance errors, bodily discomfort, response inhibition, unexpected outcomes, and demanding transitions. The commonality suggests an involvement in identifying circumstances that may require processing resources or behavioral priorities to be adjusted. Yet shared activation can result from several factors that covary across tasks. Experiments need to distinguish novelty, expected value, physiological arousal, cognitive effort, decision uncertainty, and instructions about what to attend to (Uddin, 2015).
Consider someone reading while a phone vibrates. Sensory pathways register the signal. Memory and learned value establish whether the interruption merits attention. Body-state information may contribute to urgency, while goal-directed control determines whether to respond now or resume reading. The SN can participate in coordinating these influences. The example demonstrates how a network can have a broadly relevant function without performing every computational step or generating a complete conscious experience.
The network is therefore often assigned three overlapping roles: detecting relevant changes, integrating internal and external demands, and coordinating flexible access to control. These roles need not be expressed together in every trial. A surprising sound can be noticeable without requiring an executive task switch; an anticipated transition between two working-memory operations may require control without strong sensory surprise. Well-designed experiments test such dissociations.
The Triple-Network Model and Brain-State Switching
The default mode network (DMN) includes medial prefrontal, posterior cingulate, precuneus, and related regions associated with autobiographical, semantic, social, and internally generated cognition. The frontoparietal control or central executive network includes lateral prefrontal and parietal systems involved in flexibly implementing task goals and working-memory demands. The salience network has been proposed to influence which configurations are appropriate as priorities change (Menon, 2011).
In an auditory oddball experiment, unexpected tones may recruit frontoinsular regions while portions of the DMN show reduced task-related BOLD response and control regions become more engaged. Such observations helped produce the switching hypothesis. They are not evidence for a universal rule that the SN literally turns the DMN off and executive networks on. Internal cognition can be goal directed; the DMN sometimes cooperates with frontoparietal systems during planning and memory, and several networks can operate simultaneously. See our companion Default Mode Network article for its independent mechanisms.
A further distinction is between changing cognition and changing arousal. Shifting from recollecting a conversation to answering a question differs from waking suddenly, becoming vigilant, or moving between sleep-related states. A 2026 review by Li and colleagues focuses on sensory-induced brain-wide arousal changes and stresses subcortical sensory and state-regulatory systems. This evidence complements the triple-network model while making it inappropriate to attribute all arousal transitions to one cortical hub (Li et al., 2026).
Does the Right Anterior Insula Cause Network Switching?
The right anterior insula became a major candidate coordinator because some fMRI studies reported its involvement preceding or statistically predicting changes in default-mode and executive-system signals. Menon and Uddin's theoretical synthesis suggested that anterior-insula responsiveness to important events might facilitate access to attention and working-memory resources. Molnar-Szakacs and Uddin later proposed a gatekeeper account based on effective connectivity, functional flexibility, and anatomical connections (Menon and Uddin, 2010; Molnar-Szakacs and Uddin, 2022).
A gatekeeper model is an explanatory hypothesis, not a literal neuron-based toggle. Fitting temporal or directional statistical models to fMRI data requires assumptions about how neural activity generates vascular BOLD changes. A relatively early hemodynamic response in one region cannot, by itself, prove that its neurons directly command another network. Other pathways, including cingulate, frontal, thalamic, and brainstem systems, may initiate, mediate, or reciprocally respond to the same transition.
This makes the research question precise: under which experimental conditions is activity in a given insular population necessary or sufficient to alter a measurable network state or behavior? Lesion and perturbation studies, direct electrophysiology, and longitudinal within-person analyses contribute different constraints. Reliable cross-species translation requires identifying homologous circuits and comparable behavioral functions, not merely using the same network name.
Interoception, Homeostasis, and Salience
Interoception encompasses the sensing, representation, and interpretation of information about internal physiological condition, including breathing, cardiovascular activity, digestive signals, temperature, and other bodily processes. Internal information travels through multiple peripheral and central pathways, including brainstem, thalamic, and insular systems. The anterior insula has consequently become important to theories linking bodily state, feeling, motivation, and cognitive priorities (Critchley and Harrison, 2013).
Changes in thirst, breathing during exertion, nausea, or discomfort may demand that a person adjust behavior. A network sensitive to such demands could also react to relevant external signals because both influence adaptive allocation of resources. Seeley's homeostatic account explicitly relates the SN to perceiving and responding to demands on the organism, offering a biological rationale for its diverse experimental responses (Seeley, 2019).
Physiological perception, subjective intensity, objective discrimination accuracy, and confidence in a bodily judgment are different measurements. An anterior-insula BOLD signal cannot independently establish that someone is interoceptively accurate, anxious, or experiencing a certain emotion. Research combining autonomic physiology, behavior, subjective reports, and neural recordings is necessary to evaluate such claims.
Salience, Novelty, Prediction Error, and Reward
Novelty describes a difference from prior experience; surprise describes a discrepancy from expectation and can be modeled in multiple ways; prediction errors compare actual and anticipated outcomes. Motivational value concerns the consequences of an event for an organism's goals or needs. Each may increase the priority of processing, and laboratory paradigms sometimes confound them. Consequently, a change in SN activation following a rare reward or unexpected omission cannot by itself identify the encoded computation.
Attention has multiple interacting forms. Dorsal attention systems, ventral attention territories, thalamocortical pathways, and executive-control networks participate in different forms of selection, orienting, vigilance, and maintained task set. Some SN-related signals may indicate when priorities should change, while partner systems participate in implementing sensory and behavioral selection. Nomenclature varies enough that a researcher must name the operational network definition (Uddin et al., 2023).
Reward involves learning, reinforcement, valuation, motivation, and action selection distributed across striatal, cortical, and midbrain circuitry. Dopamine contributes to several aspects of these processes, with important differences among neuronal populations and behavioral tasks. The salience network overlaps functionally with motivational processing but is not interchangeable with the reward system. Our Brain Reward System article retains the canonical treatment of those neural mechanisms.
How Researchers Measure the Salience Network
Resting-state functional MRI estimates correlations between relatively slow BOLD time series in regions of interest. This can identify reproducible statistical networks and differences in their measured organization. BOLD is a vascular signal indirectly related to neural activity; it is not a direct electrode recording of individual neurons or a structural image of axonal pathways. Head motion, breathing, heart rate, sleepiness, preprocessing, scanner parameters, and the definition of regions can change measured connectivity (Power et al., 2011; Yeo et al., 2011).
Task-based fMRI compares conditions such as rare versus frequent tones, easy versus demanding trials, or internal versus external attention. A contrast can show that a region's BOLD response differs reliably, but psychological specificity requires independent control of novelty, arousal, effort, emotion, and actions. Group averaging can blur nearby network boundaries or mask substantial individual variation. A single participant's activation map cannot automatically classify subjective experience.
Dynamic functional connectivity examines whether estimated interactions change over time. Effective-connectivity analyses introduce directional statistical models. Both can generate mechanistic hypotheses, but estimates depend on sampling frequency, hemodynamic assumptions, noise, preprocessing, and model selection. The inferential leap from BOLD correlation or apparent lag to biological causality is especially large when vascular responses differ across regions.
EEG detects scalp electrical potentials related to synchronized neural populations, and MEG measures associated magnetic fields. They offer excellent temporal resolution compared with typical fMRI but have source-localization limitations, especially for deeply situated insular territories. Intracranial recording, lesions, neurostimulation, and carefully targeted animal interventions can add causal information, although their populations and experimental constraints differ. See Brain Imaging Methods for a comparative treatment.
Evidence for Switching: Converging Observations and Causal Tests
The initial network segregation experiments established that salience-related and executive-related patterns can be detected with resting functional connectivity, while subsequent task studies tied frontoinsular responses to changes in other networks. Sridharan and colleagues' 2008 research supplied evidence consistent with directional coupling, within the limits of hemodynamic measurement and model assumptions. These findings support a scientifically useful switching hypothesis; they do not make every transition anatomically identical (Seeley et al., 2007; Sridharan et al., 2008).
In traumatic brain injury research, Bonnelle and colleagues reported an association between salience-network pathway integrity and the regulation of default-network responses during inhibitory control. Structural damage supplies constraints beyond ordinary activation studies, but traumatic injury affects multiple connected pathways and the resulting findings must be interpreted within observational and clinical heterogeneity (Bonnelle et al., 2012).
A 2023 rat study reported that manipulating anterior insular neurons using optogenetic methods altered a default-mode-like network. Such work directly intervenes in neural circuits under defined conditions and can therefore support causal claims in the studied animal preparation. Translation to humans requires separate evidence regarding cell types, network homology, behavioral state, and the corresponding human intervention; a finding in rats cannot establish a universal human cortical switch (Nature Communications, 2023).
The combined picture is strongest for distributed, context-sensitive coordination and weaker for the claim that a particular right insular node always initiates transitions. Competing hypotheses emphasize reciprocal cortico-cortical exchange, brainstem and thalamic input, homeostatic regulation, task-dependent control demands, and individual functional architecture.
Salience Versus Cingulo-Opercular and Ventral Attention Networks
Different atlases divide the cerebral cortex into differing numbers of communities. Networks labeled salience, cingulo-opercular, and ventral attention can overlap anatomically, yet may be separated or combined depending on spatial resolution, input data, and community-detection method. Some schemes distinguish cingulate and opercular territories on the basis of sustained control, others emphasize their coupling with insula and subcortical systems. The seven-network atlas and finer network partitions reveal useful broad structure without requiring one universally fixed vocabulary (Yeo et al., 2011; Power et al., 2011).
An expert workgroup reviewed controversies in network nomenclature in 2023 and emphasized reporting network scale, atlas, interindividual variability, dynamics, subcortical involvement, and multimodal evidence. Those recommendations are essential when comparing apparently contradictory results. One article's “salience network connectivity” may refer to regions that another paper assigns to a distinct system. Inferences about stronger or weaker salience processing require looking beyond the label (Uddin et al., 2023).
Precision functional mapping samples a person's brain repeatedly to identify their individual system boundaries. It can improve anatomical specificity, but it does not convert measured network organization directly into a unique psychological interpretation. The relation between network maps and meaningful behavior must still be tested.
Clinical Research: Depression, Anxiety, and Biomarkers
Network analyses have identified associations involving salience-related regions in depression, anxiety, psychosis research, chronic pain, neurological injury, and other contexts. These are research-level observations across heterogeneous groups. Similar patterns can occur across conditions, and findings differ with task, medication, age, arousal, imaging protocol, symptom profile, and analytic choices. The triple-network model offers a framework for studying mechanisms, not a universal diagnostic classifier (Menon, 2011).
A consequential 2024 Nature study used repeatedly sampled individualized brain scans to investigate depression. Lynch and colleagues found cortical expansion of a frontostriatal salience-related network in multiple depression samples and described some symptom-related changes in specific connectivity measures. The work improved precision through repeated measurement and replication. Its network boundaries and connectivity patterns are not an established stand-alone clinical test for depression, and the findings do not justify diagnosing an individual from a single research scan (Lynch et al., 2024).
Clinical translation demands external validation, reliable acquisition, appropriate comparison groups, meaningful sensitivity and specificity, prospective prediction, and evidence that test use improves care. A brain network association is distinct from a symptom, diagnostic criterion, prognosis, or proven treatment target. Disorder-specific clinical claims therefore belong in their corresponding English Hub owners; for a relevant evidence review see Depression and the Brain.
Scientific Limitations, Alternative Models, and Neural Myths
Reverse inference is a central limitation. Because anterior insular and cingulate regions respond across varied tasks, their activation cannot uniquely demonstrate attention, fear, pain, uncertainty, or a feeling of importance. Functional connectivity measures dependence among signals and cannot independently distinguish direct anatomical communication from common inputs or indirect pathways. Temporal precedence in BOLD time series is not neuronal timing without additional assumptions and evidence.
Competing accounts emphasize salience detection, gatekeeping access to control, regulating physiological demands, broad task-set or effort adjustments, and changes in arousal. They can yield related patterns of activation while differing in predicted interventions and behavioral consequences. The field benefits from preregistered models, cross-method comparisons, individual mapping, and experimental contrasts that disentangle the different computations (Seeley, 2019; Molnar-Szakacs and Uddin, 2022).
It is a myth that the salience network is a single attention center or that the insula is the only site where emotions are felt. It is also inaccurate to say that SN activation always shuts down mind wandering, or that higher connectivity always indicates better functioning. There is no validated general-purpose salience-network health score that interprets a routine clinical scan as proof of a “broken” or “overstimulated” network. Statements that simple exercises repair this system require direct controlled evidence rather than an appealing brain diagram.
Practical Implications for Understanding Cognitive Flexibility
The most useful implication is a more coherent account of how priorities change. Sensory events, learned significance, bodily regulation, and current goals jointly influence processing. The same sound may deserve immediate attention in one setting and almost none in another. Large-scale network research encourages examining the coordination of those influences instead of expecting complex behavior to be located in one cortical area.
For psychologists and researchers, methodological discipline is central. Define salience explicitly, measure attention or behavior independently, record physiological confounds, specify the atlas used to label the network, and report which conclusions depend on statistical association versus intervention. When evaluating potential clinical applications, assess whether a neural measure makes decisions more accurate or improves outcomes beyond established behavioral and clinical assessments.
For the English Psychology Hub network, this article connects naturally to the existing Neural Circuits overview and companion default-mode, imaging, reward, and emotion-neuroscience pages. Planned reciprocal links to the dedicated insula and frontoparietal control network articles should be added only after those reserved pages are published.
Frequently Asked Questions
What is the salience network's main function? It is associated with identifying contextually relevant demands, integrating information about internal and external conditions, and helping coordinate changes in how other brain systems are engaged. Several proposed mechanisms remain under investigation (Seeley, 2019).
Is the salience network the same as the anterior insula? No. The anterior insula is a key hub in many definitions, while cingulate regions and sometimes subcortical partners are also included. The insula additionally participates in many functions beyond this particular network.
Does the SN switch the default mode network off? Some tasks are associated with increased salience-system responses and reduced activity in parts of the DMN, but there is no established universal on/off relation. These networks may interact reciprocally and can support different cognitive demands together.
How does salience differ from attention? Salience describes potential relevance or priority, while attention covers the selection, sustained processing, and control of information. Multiple attention systems contribute; a prominent event can be ignored when goals require it.
Why is the right anterior insula often mentioned? Its recruitment and some directional connectivity findings motivated influential models in which it helps coordinate transitions between network patterns. The strength and necessity of that proposed role depend on the task and methods (Sridharan et al., 2008).
Can a salience-network scan diagnose anxiety, depression, or ADHD? Current research findings do not justify treating an SN image or connectivity value as a general individual diagnostic test. Diagnoses rely on established clinical evidence and criteria.
Does increased SN activity mean someone is stressed? No. Many factors, including cognitive effort, novelty, bodily regulation, pain, and task structure, can affect measured activity. The psychological meaning depends on controlled contrasts and other measurements.
Can the salience network be trained to improve focus? Behavioral training may influence performance and neural measures, but a claim that an exercise repairs or specifically optimizes the salience network requires direct testing and meaningful outcomes.
Is the ventral attention network identical to the salience network? Terminology and spatial boundaries differ among atlas systems. Results should be interpreted with the specific regions and definitions used rather than treating labels as universal synonyms.
What is the evidence for brain-state switching? Functional-network studies, task-related changes, injury-associated findings, and targeted interventions in animal models jointly motivate the hypothesis, while each supports different levels of causal interpretation (Bonnelle et al., 2012).
Related Articles
References
Bonnelle, V., et al. (2012). Salience network integrity predicts default mode network function after traumatic brain injury. Proceedings of the National Academy of Sciences, 109, 4690–4695. DOI.
Critchley, H. D., and Harrison, N. A. (2013). Visceral influences on brain and behavior. Neuron, 77, 624–638. DOI.
Li, M., Tan, K., Liu, T., and Yuan, K. (2026). How salient sensory stimuli induce brain-wide state alterations. Annual Review of Neuroscience, 49, 143–166. DOI.
Lynch, C. J., et al. (2024). Frontostriatal salience network expansion in individuals in depression. Nature, 633, 624–633. DOI.
Menon, V. (2011). Large-scale brain networks and psychopathology: A unifying triple network model. Trends in Cognitive Sciences, 15, 483–506. DOI.
Menon, V., and Uddin, L. Q. (2010). Saliency, switching, attention and control: A network model of insula function. Brain Structure and Function, 214, 655–667. DOI.
Molnar-Szakacs, I., and Uddin, L. Q. (2022). Anterior insula as a gatekeeper of executive control. Neuroscience and Biobehavioral Reviews, 139, 104736. DOI.
Power, J. D., et al. (2011). Functional network organization of the human brain. Neuron, 72, 665–678. DOI.
Seeley, W. W. (2019). The salience network: A neural system for perceiving and responding to homeostatic demands. Journal of Neuroscience, 39, 9878–9882. DOI.
Seeley, W. W., et al. (2007). Dissociable intrinsic connectivity networks for salience processing and executive control. Journal of Neuroscience, 27, 2349–2356. DOI.
Sridharan, D., Levitin, D. J., and Menon, V. (2008). A critical role for the right fronto-insular cortex in switching between central-executive and default-mode networks. Proceedings of the National Academy of Sciences, 105, 12569–12574. DOI.
Uddin, L. Q. (2015). Salience processing and insular cortical function and dysfunction. Nature Reviews Neuroscience, 16, 55–61. DOI.
Uddin, L. Q., et al. (2023). Controversies and progress on standardization of large-scale brain network nomenclature. Network Neuroscience, 7, 864–905. DOI.
Yeo, B. T. T., et al. (2011). The organization of the human cerebral cortex estimated by intrinsic functional connectivity. Journal of Neurophysiology, 106, 1125–1165. DOI.