A person says, "My heart suddenly feels wrong."
That sentence can contain several different research questions.
What changed in the heart? What signal reached the nervous system? How strongly was it perceived? What did the person predict the sensation meant? Did the sensation trigger anxiety, or did anxiety change the sensation? Would the same person respond differently to a respiratory or gastrointestinal perturbation?
Sahib Khalsa's research programme is built around making those questions experimentally separable.
His work sits at the intersection of interoception, psychiatry, organ-specific physiology and experimental medicine. Rather than treating "body awareness" as one global capacity, the programme uses controlled perturbations and multimodal measurement to ask how internal signals are generated, sensed, interpreted and altered in clinical conditions.
Current role and programme
As of 23 September 2026, UCLA's Brain Research Institute lists Sahib Khalsa, MD, PhD as an Associate Professor in Residence in the Department of Psychiatry and Biobehavioral Sciences at the David Geffen School of Medicine and Director of Anxiety Disorders Research.
The current Khalsa Lab at UCLA describes research on heart-brain and gut-brain interactions using experimental medicine, pharmacological modulation, brain imaging, EEG and computational modeling. Its listed programme includes cardiovascular, respiratory and gastrointestinal interoception, anxiety disorders, anorexia nervosa and related conditions.
Khalsa previously held a major clinical-research role at the Laureate Institute for Brain Research and remains associated with that history in older literature. Current identity should therefore be taken from UCLA rather than inferred from legacy affiliations in publications.
A field-level roadmap
Khalsa was lead author of the 2018 consensus-style review "Interoception and Mental Health: A Roadmap," developed with a large multidisciplinary group associated with the 2016 Interoception Summit.
The paper helped articulate several enduring problems in the field.
Interoception is multimodal. Measurement approaches are inconsistent. Clinical studies often use different definitions. Some tasks depend strongly on signal intensity or knowledge. Mechanistic translation requires methods that can connect bodily perturbation, subjective report and neural response.
The roadmap's value is not that it solved those problems.
It made them harder to ignore.
For a Knowledge Library, the paper is especially useful because it resists the idea that one heartbeat task can stand in for the entire interoceptive system.
Experimental perturbation changes the question
Many interoception measures ask what people can perceive from naturally occurring bodily signals.
Khalsa's programme often takes a different route: deliberately perturb the bodily system and measure the resulting perception and neural response.
Cardiovascular work has used pharmacological probes such as isoproterenol to alter cardiac and respiratory sensations in a controlled setting. Respiratory paradigms can change inspiratory resistance. Gastrointestinal work can use mechanosensory perturbation, including newer ingestible or body-surface approaches.
The methodological advantage is important.
If the stimulus is manipulated, researchers can estimate response functions rather than relying only on whatever internal variation happened to occur during a resting task.
But perturbation is not purity.
A pharmacological challenge changes multiple physiological variables. Participants can form expectations. Clinical groups differ in medication, prior experience and symptom meaning. Experimental control narrows uncertainty; it does not eliminate interpretation.
Interoception as organ-specific science
Khalsa's work is valuable for the library because it makes modality specificity difficult to ignore.
The heart, lungs and gastrointestinal tract have different receptors, timescales, signal properties and behavioral relevance.
A person can be sensitive to respiratory resistance without showing the same pattern in cardiac perception. Gastrointestinal mechanosensation involves different physiology again.
This makes "interoceptive ability" a risky shorthand.
A more defensible model is a profile across modalities, tasks and dimensions.
The 2026 study on aging coauthored by Khalsa illustrates the direction. It examined healthy adults aged 18 to 79 across cardiac and respiratory modalities and combined performance-related and self-report dimensions. Whatever the final field consensus, that design is more informative than assuming one task captures the interior of the body.
Anxiety is both bodily and interpretive
Anxiety makes interoception clinically compelling because many anxiety symptoms are bodily: racing heart, shortness of breath, chest tightness, dizziness, nausea, trembling.
But the same sensation can be interpreted in radically different ways.
A strong heartbeat during exercise may be neutral. The same intensity during an unexpected pause at night may be read as danger. Prediction, context and previous panic can change the experience.
Khalsa's programme studies how bodily perturbations interact with anxiety-related neural and perceptual processes.
That does not mean anxiety is "caused by poor interoception."
Clinical symptoms can arise through multiple pathways, and interoceptive differences are neither necessary nor sufficient for a diagnosis.
The strongest use of interoception in psychiatry is mechanistic: identify sub-processes that may explain some symptom patterns in some people.
Eating disorders extend the programme beyond the heart
The lab's work on gut-brain signaling and anorexia nervosa addresses a domain where internal sensation is central to behavior yet difficult to measure cleanly.
Fullness, gastric distension, nausea, appetite and satiety involve mechanical, chemical, hormonal, learned and social processes.
Interoception offers one lens, not a total explanation.
The move toward gut mechanosensation is methodologically important because it broadens a field historically dominated by cardiac tasks.
It also increases complexity. Gastrointestinal physiology unfolds on slower timescales and can be shaped by motility, meal history, medication and gastrointestinal conditions.
A mechanistic programme must therefore remain organ-specific rather than merely transplanting concepts from heartbeat studies.
The search for biomarkers needs restraint
Clinical neuroscience often seeks biomarkers: measurable features that help identify mechanisms, predict outcomes or stratify treatment.
Interoceptive tasks and heartbeat-evoked neural signals are potential candidates.
Khalsa coauthored a 2026 comprehensive meta-analytic review of the heartbeat-evoked potential, a brain response time-locked to cardiac activity. The review is useful precisely because it treats the signal as a neurophysiological phenomenon requiring synthesis across methods and health conditions.
A measurable neural signal is not automatically a clinical biomarker.
For that status, researchers need reliability, specificity, prospective validation, clinically meaningful discrimination and evidence that the measure adds value beyond simpler information.
The word biomarker should come late, after validation, not early as a synonym for interesting physiology.
Experimental therapeutics is not ordinary wellness language
Khalsa's institutional profile describes neuroscience-informed clinical applications and experimental therapeutics.
In research, experimental therapeutics has a demanding meaning: perturb a candidate target, measure whether the target changed, and examine whether clinically relevant outcomes change as predicted.
That differs from calling an intervention "nervous-system based" because it involves breathing, floating or body attention.
Some of Khalsa's work has examined Reduced Environmental Stimulation Therapy, often called floatation-REST, in anxiety-related contexts.
Those studies should be read as specific interventions with specific populations, protocols and outcomes.
They do not validate every commercial float practice or generic claim that reducing stimulation regulates everyone.
Measurement is part of the science, not paperwork around it
Interoception research repeatedly encounters a temptation to measure what is convenient and then rename it as the construct.
Khalsa's programme pushes against that by combining perturbation, physiology, subjective report, neuroimaging, EEG and computational modeling.
No one layer replaces the others.
A physiological change shows that the body changed. A report shows what the participant experienced. Neural data characterize central response. A computational model proposes how observations may be generated.
The layers can constrain one another.
They should not be collapsed.
This principle is as important for applied Inner Technology as it is for laboratory research.
If a programme claims to increase interoceptive capacity, it needs to specify which layer should change.
Resting measures and perturbation measures answer different questions
A resting heartbeat task asks how a person relates to a signal that is occurring naturally at its current intensity. A perturbation experiment asks how perception and physiology change when the internal state is deliberately shifted.
These approaches should complement rather than replace each other. Resting measures may better approximate ordinary conditions but suffer from limited and variable signal strength. Perturbation can create a clearer dose-response relation but introduces an artificial state and may engage expectation or stress responses that differ from daily life.
Khalsa's programme is especially informative because it makes the perturbation explicit. That allows the research question to move from "Are you good at sensing your body?" toward "How does this person detect and interpret a controlled change in this organ system?"
The second question is narrower and scientifically stronger.
Computational modeling can separate hidden processes
Behavioral performance often bundles sensitivity, bias, prior expectation and response strategy into one score. Computational models can attempt to separate those latent components.
In interoception, such models may estimate how strongly a participant weights prior expectations versus incoming sensory evidence, how uncertainty changes across trials, or how learning updates beliefs about bodily state.
The benefit is explanatory precision. The danger is model reification. A fitted parameter is not a directly observed biological quantity simply because it has a psychologically meaningful name. Different models can fit the same behavior, and parameter recovery must be demonstrated.
Computational psychiatry is therefore most persuasive when model comparison, out-of-sample prediction and independent physiological measures constrain interpretation.
Clinical heterogeneity is the reason to pursue mechanisms
Diagnostic categories are broad. Two people with the same anxiety diagnosis can have different bodily triggers, beliefs, avoidance patterns and physiological profiles. Interoceptive neuroscience is attractive partly because it may identify dimensions that cut across diagnoses.
But transdiagnostic relevance does not mean universal mechanism. A dimension can appear across several disorders while mattering only for a subset of patients.
Mechanistic stratification is useful only if it improves prediction or treatment selection beyond existing clinical information. That evidential step remains harder than demonstrating a group difference.
Ethics enters as soon as perturbation becomes intervention
Deliberately altering heartbeat, breathing or gut sensation can be informative precisely because these signals are salient. That salience also creates ethical responsibilities. Participant selection, medical screening, adverse-event monitoring and clinical oversight are part of the method, not bureaucratic additions.
The same principle should govern any applied translation. A laboratory probe performed under controlled medical conditions cannot be copied casually into consumer practice simply because the underlying mechanism sounds interesting.
Mechanistic intimacy raises the standard of care.
What the evidence does not show
Khalsa's research does not show that interoception is one global trait.
It does not show that heartbeat perception captures respiratory or gastrointestinal sensing.
It does not show that people with anxiety or eating disorders share one bodily-processing deficit.
It does not show that stronger interoceptive sensitivity is always beneficial.
It does not show that a heartbeat-evoked potential is a validated diagnostic biomarker for ordinary clinical use.
It does not show that pharmacological perturbation reproduces every naturally occurring panic or bodily state.
It does not show that floatation-REST or any other experimental intervention is universally effective.
It does not show that organ-specific mechanisms can replace social, psychological or developmental explanations of distress.
And it does not justify marketing a generic "interoception training" programme as clinically validated without direct trials.
A 2026 direction: lifespan and multimodality
The 2026 aging study by Lee, Nishiyama, Khalsa and Duarte is notable because it asks how interoceptive sensitivity and sensibility vary across adulthood while sampling more than one modality.
The study included 78 adults aged 18–79 and assessed cardiac and respiratory processes alongside self-reported interoceptive sensibility.
The sample is not large enough to settle lifespan questions.
Its importance is architectural: age is treated as a potential modifier, and interoception is measured as multidimensional and multimodal.
That is the direction a mature field needs.
General claims should emerge from designs that can actually test generality.
Current relevance to human-AI systems
Khalsa's work also matters indirectly for AI-based health and affective technologies.
Biosensing systems increasingly combine heart rate, respiration and other physiological signals with machine learning. The promise is personalization. The danger is inference inflation.
Organ-specific perturbation research shows why physiology requires context. The same measured change can have different meanings across people and situations. Clinical interpretation depends on measurement validity, baseline, comorbidity and the exact question being asked.
An AI model can improve prediction without making the latent construct simple.
High classification accuracy in one dataset does not turn a physiological signal into transparent access to subjective state.
Institute synthesis: internal sensing needs experimental humility
The Institute uses experimental humility as a synthesis principle here: the more intimate the claim about a person's internal state, the more carefully the measurement chain should be exposed.
This phrase is Institute language, not Khalsa terminology.
What was perturbed?
What was measured?
What was reported?
What was modeled?
What was inferred?
Those five questions prevent a useful biomarker candidate from becoming a story about the whole person.
Why this profile belongs independently
Khalsa's programme is a useful bridge between basic interoception science and clinical translation because it combines organ-specific perturbation with multiple measurement layers.
The profile belongs independently not because the researcher has authority, but because following the programme clarifies what rigorous translational interoception research requires: construct precision, controlled perturbation, multimodality, clinical boundaries and explicit measurement.
It also supplies a valuable counterweight to cultural uses of "interoception" that have become detached from physiology.
Editorial limitation note
This profile does not infer treatment efficacy from mechanism studies, does not treat candidate biomarkers as clinical tests, and does not generalize findings from one interoceptive modality or clinical sample to all people.
Research references
- UCLA Brain Research Institute. Sahib Khalsa, MD, PhD: current institutional profile. Accessed 23 September 2026. https://bri.ucla.edu/people/sahib-khalsa-m-d-ph-d/
- Khalsa Lab, UCLA. Our Research. Accessed 23 September 2026. https://khalsalab.semel.ucla.edu/home/research/
- Khalsa SS, Adolphs R, Cameron OG, et al.; Interoception Summit 2016 participants. Interoception and Mental Health: A Roadmap. Biological Psychiatry: Cognitive Neuroscience and Neuroimaging. 2018;3(6):501-513. DOI: 10.1016/j.bpsc.2017.12.004. https://pubmed.ncbi.nlm.nih.gov/29884281/
- Wilkinson EM, Jones HG, Kohler M, Khalsa S, Moseley GL, Berryman C. A Comprehensive Meta-Analytic Review of the Neurophysiological Properties of the Heartbeat Evoked Potential in Health and Disease. Biological Psychiatry: Cognitive Neuroscience and Neuroimaging. 2026. DOI: 10.1016/j.bpsc.2026.04.012. https://pubmed.ncbi.nlm.nih.gov/42092635/
- Lee K, Nishiyama R, Khalsa SS, Duarte A. How aging shapes interoception: A multimodal, multidimensional analysis. Journals of Gerontology: Series B. Published online 20 August 2026. DOI: 10.1093/geronb/gbag169. https://pubmed.ncbi.nlm.nih.gov/42625395/
- Desmedt O, Luminet O, Walentynowicz M, Corneille O. The new measures of interoceptive accuracy: A systematic review and assessment. Neuroscience & Biobehavioral Reviews. 2023;153:105388. DOI: 10.1016/j.neubiorev.2023.105388. https://pubmed.ncbi.nlm.nih.gov/37708919/