# Interacting pathology, regulation and learning: a bounded evidence map

17 September 2026. Primary-source companion for the Program Audit Studio. Research interpretation, not a treatment recommendation or a systematic review. Prepared as a research companion; distinct from attributed course maps.

## Answer to the question

**Yes: there are established research methods for studying interacting physiological systems, and explicit mathematical models that retain an ongoing inflammatory input while tracking regulation, pain and cardiovascular consequences.** They are useful starting points for a richer account than a choice between continuing bodily disease and a learned alarm. However, this search did not identify one validated model combining pathogen dynamics, learning, subjective experience and mind-body treatment response across chronic illnesses. The selected papers illustrate different pieces and different levels of maturity; their compatibility does not validate a combined theory. [Network methods](https://doi.org/10.1371/journal.pone.0142143), [coupled physiological model](https://doi.org/10.1113/JP280883), [extended inflammatory model](https://doi.org/10.1038/s41540-024-00473-y).

An intervention may alter distress or one regulated response while other disease processes remain active. Conversely, nervous-system involvement does not identify a learned cause, an error, or a process accessible to a particular exercise. The important research questions are **which process changes, by what intervention, with what other processes still present, and with what consequences for symptoms, function, disease activity and burden**. The two RA trials below make different, bounded contributions to that distinction. [MBSR trial](https://doi.org/10.1002/art.23010), [implanted-stimulation trial](https://doi.org/10.1038/s41591-025-04114-7).

## Seven sources and their exact jobs

### 1. Network physiology: measured coordination, not a universal disease mechanism

**Bartsch et al. (2015), “Network Physiology: How Organ Systems Dynamically Interact.”** DOI **10.1371/journal.pone.0142143**. [Primary full text](https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0142143).

- **Population/time:** 36 healthy adults, 18 women and 18 men, ages 20–40; average 7.8-hour sleep recordings. Signals were standardized to one-second resolution and analyzed in moving windows.
- **Established here:** coordination among EEG bands, cardiac, respiratory, ocular and motor signals changes with sleep stage. This is a method for measuring context-dependent network organization.
- **Nonlinearity:** the paper motivates nonlinear feedback, but its reported time-delay-stability links derive from lagged cross-correlations and temporal stability. They do not identify a unique nonlinear causal law, directional mechanism or treatment target.
- **Limit:** no infection, chronic-illness cohort, learning intervention or efficacy comparison. A physiological network diagram should not become a claim that all illness is one dysregulated network.
- **Reading depth:** primary Data, Methods/TDS, network construction and selected results; supplements and code not rerun.

### 2. Coupled inflammation, temperature, pain and circulation

**Dobreva et al. (2021), “A physiological model of the inflammatory-thermal-pain-cardiovascular interactions during an endotoxin challenge.”** DOI **10.1113/JP280883**. [Published abstract](https://pubmed.ncbi.nlm.nih.gov/33450068/); [author preprint](https://arxiv.org/pdf/1908.07611).

- **Population/time:** published model calibrated to 20 healthy young men after endotoxin challenge and checked against a separate human dataset; acute, hours-long measurements.
- **Established here:** an explicit coupled model reproduces inflammatory and physiological trajectories. Its maintained pathological input and combined-treatment examples are simulations, not treatment trials in infected patients.
- **Nonlinearity:** specified nonlinear activation/inhibition and regulatory relationships are fitted within a proposed structure. Fit and independent short-challenge validation do not establish every assumed coupling or sustained-infection prediction.
- **Limit:** no learned symptom response or psychological intervention is modeled. Endotoxin exposure is not a replicating infection.
- **Reading depth:** published abstract; preprint methods, equations and limitations. The preprint is not interchangeable with the final version: intervention terminology and calibration/validation wording differ. Consequently, its detailed parameters should not be imported into our display as verified final-publication specifications.

### 3. A newer inflammation model explicitly tests prolonged input

**Relouw et al. (2024), “Mathematical model of the inflammatory response to acute and prolonged lipopolysaccharide exposure in humans.”** DOI **10.1038/s41540-024-00473-y**. [Primary article](https://www.nature.com/articles/s41540-024-00473-y); [authoritative full-text XML](https://www.ebi.ac.uk/europepmc/webservices/rest/PMC11621538/fullTextXML).

- **Population/time:** digitized experiments in healthy young men; calibration bolus groups of 10 and 15; validation across additional doses and a 10-person, three-hour infusion. Infection scenarios extend to 12–36 hours.
- **Established here:** specified cytokine/mRNA dynamics and temperature/heart-rate outputs predict held-out challenge conditions, with recognized errors including excessive TNF and delayed IL-6 recovery.
- **Nonlinearity:** nonlinear regulatory structure has empirical dose/time constraints. Clinical infection input curves were selected through trial and error to yield plausible ranges; these were not independently measured infection trajectories. No treatment was simulated.
- **Limit:** this is stronger evidence for bounded inflammatory dynamics than for clinical infection prediction. It does not include cognition, symptom learning or mind-body interventions. mRNA persistence is biochemical memory, not a measured belief.
- **Reading depth:** full article body, calibration, validation, infection simulation and limitations; no numerical reproduction or supplement audit.

### 4. Inflammation can be studied as an input to learning systems

**Queirazza et al. (2024), “Mild exogenous inflammation blunts neural signatures of bounded evidence accumulation and reward prediction error processing in healthy male participants.”** DOI **10.1016/j.bbi.2024.03.044**. [Author-hosted published paper](https://eprints.gla.ac.uk/323723/2/323723.pdf).

- **Population/time:** 19 healthy men in a randomized, placebo-controlled crossover vaccination study; learning task and fMRI 3–4 hours after injection. Active infection was excluded.
- **Established here:** inflammatory manipulation changed selected model-associated neural responses during decisions and reward feedback. **Task performance and fitted behavioral-model parameters did not show significant condition differences.**
- **Nonlinearity:** a sequential-sampling/reinforcement-learning model was fitted and compared with alternatives. This is not a demonstrated bidirectional nonlinear inflammation–learning feedback loop.
- **Limit:** small male sample, acute challenge, and task hypotheses absent from preregistration. It does not show that a person’s beliefs perpetuate inflammation, that poor recovery means faulty learning, or that retraining reverses active disease.
- **Reading depth:** published PDF abstract, methods, model/performance results and discussion, especially sections 3.4–3.5 and limitations; no raw-data rerun.

### 5. Disease-specific challenge evidence keeps tissue processes in view

**Appelman et al. (2024), “Muscle abnormalities worsen after post-exertional malaise in long COVID.”** DOI **10.1038/s41467-023-44432-3**. [Primary full text](https://www.nature.com/articles/s41467-023-44432-3).

- **Population/time:** 25 long-COVID participants with post-exertional malaise and 21 recovered controls; maximal exercise, blood/muscle measurements before and one day after, symptoms followed for up to seven days.
- **Established here:** illness-associated metabolic/muscle abnormalities and post-exercise physiological changes accompany this symptom pattern. Assay sample sizes vary. The study separates slow exercise-capacity features from acute worsening.
- **Nonlinearity:** this design did not estimate a nonlinear dynamical coupling law or test the hypothesized longer-term downward feedback cycle.
- **Limit:** neither active replicating infection nor a learned-alarm mechanism is established. The measured viral-protein signal did not distinguish groups. It does not test psychological treatment, generalize to all long COVID/ME/CFS, or supply an individual safe-exertion rule.
- **Reading depth:** primary cohort/methods, selected muscle/metabolic results, viral-protein comparison and discussion; no independent assay or supplement reproduction.

### 6. Psychological benefit can be tested during continuing inflammatory illness

**Pradhan et al. (2007), “Effect of Mindfulness-Based Stress Reduction in rheumatoid arthritis patients.”** DOI **10.1002/art.23010**. [PubMed abstract](https://pubmed.ncbi.nlm.nih.gov/17907231/); [publisher](https://acrjournals.onlinelibrary.wiley.com/doi/10.1002/art.23010).

- **Population/time:** 63 adults with RA, randomized to an eight-week MBSR course plus four months of reinforcement or waitlist. Usual rheumatological care continued; baseline remission was an exclusion.
- **Established here:** no significant between-group outcomes at two months; distress and well-being improved at six months. Disease activity did not show an intervention benefit.
- **Nonlinearity:** not a network or mechanistic-coupling test. Randomization tests the package against this comparator, not a specific mediator or feedback equation.
- **Limit:** small pilot and waitlist comparator; no proof that mindfulness changes RA pathology or a universal inference of no inflammatory effect. This supports a bounded example of adjunctive benefit without requiring the disease to have ended.
- **Reading depth:** primary abstract and publisher methods extract; complete result tables, harms ascertainment and supplements not independently reviewed in this pass.

### 7. A neural intervention can change outcomes in active autoimmune disease

**Tesser et al. (online 2025; issue 2026), “Vagus nerve-mediated neuroimmune modulation for rheumatoid arthritis: a pivotal randomized controlled trial.”** DOI **10.1038/s41591-025-04114-7**. [Primary paper](https://www.nature.com/articles/s41591-025-04114-7); [full-text XML](https://www.ebi.ac.uk/europepmc/webservices/rest/PMC12823386/fullTextXML).

- **Population/time:** 242 patients with active RA and inadequate response/intolerance to biological/targeted drugs; implanted stimulation versus sham for three months, then open-label follow-up to 12 months. Conventional DMARD care continued.
- **Established here:** three-month ACR20 responses were 35.2% versus 24.2%; adjusted difference 11.8 percentage points. The endpoint includes joint examination and patient-reported measures. Serious related events were perioperative (1.6% of 243 enrolled).
- **Nonlinearity:** randomized evidence for the specified device intervention, not validation of a nonlinear network model or every proposed mediator.
- **Limit:** a device dose cannot be equated with breathing, meditation, felt safety or retraining. Longer follow-up lacked a sham comparator and allowed treatment augmentation; sponsor designed/analyzed the trial with investigators.
- **Reading depth:** primary abstract plus full-text results, outcomes, methods, safety and sponsor role; supplements not independently audited.

## What the Commons can responsibly take from this

This is a **research architecture proposal**, not a new combined mechanism established by the seven papers. It should be displayed separately from a source-faithful reconstruction of what a course teaches.

1. Preserve continuing inputs: pathogen or antigen exposure where established, tissue damage, medication, environment and metabolic constraints. A trigger should not disappear from the map just because a regulatory loop is added.
2. Separate regulatory dynamics from learning. An inflammatory feedback, a baroreflex, learned avoidance and a propositional belief are different processes and require different measurements.
3. Give arrows a scope and an evidentiary job: measured temporal association, experimental manipulation, fitted model assumption, tested prediction, or clinical treatment effect. “Nonlinear” is not a stronger evidence grade.
4. Keep intervention targets and outcome layers separate. Changing an input, changing response gain, changing appraisal, and changing daily participation can have different implications even when the symptom report looks similar.
5. Ask which predictions discriminate models. An ongoing-input model, a learned-maintenance model and a mixed model should make different time-course or intervention predictions. Adding more arrows does not improve a model unless the extra structure helps explain or predict observations.

The study design needed for a particular claim might combine repeated physiological and experiential measurements with a defined intervention and comparator, prospective harms, and prediction in new data. Testing a nonlinear response would require enough variation in dose/state/time and explicit comparison with simpler alternatives. A significant statistical interaction, an attractive network figure, or a symptom improvement alone is not that demonstration.

**Confidence:** high that these quantitative research traditions exist and are more specific than a binary account; moderate in this deliberately small selection as an orientation; low that the papers license person-level choice of a mind-body intervention during active infection. No universal chronic-illness model or treatment-selection rule was established in this review.

## Corpus integration and search record

This fits existing `NAVIGATION.md` orientation and `papers/between-symptom-and-disease/MEASUREMENT-ROADMAP.md`: ongoing pathology, response dynamics, learning and treatment can be simultaneous research variables. `FIELD-RATIONALE.md` already contains the distinct RA trial boundaries. This memo adds concrete modeling methods and their validation limits; it does not replace those documents or alter the claim ledger.

Live searches on 17 September 2026 covered network physiology/sleep, inflammation models with sustained input, inflammation and computational learning, RA mindfulness, and neuroimmune trial evidence. Primary publishers, author repositories, PubMed and Europe PMC supplied the sources. No acute histamine or immune-engram paper was used as a supporting example. This was purposive selection, not an exhaustive or preregistered search; omitted fields and contrary results remain a limitation.

Publisher/PMC web rendering failed for parts of two articles. The official Europe PMC full-text API supplied their article bodies. The Dobreva final full text remained unavailable in this pass; its author preprint was explicitly distinguished. Local working text extractions are for source checking only and are not proposed publication assets.