Stress Testing of the Gut–Brain Axis for Precision Medicine in Functional Dyspepsia

1. Vagal–Microbial Integration

1.1. Preclinical Evidence: Vagal Manipulation Abolishes Microbiome-Dependent Phenotypes

Ten independent rodent studies that use subdiaphragmatic vagotomy (SDV) as the main perturbation show the same pattern: with the vagus intact, microbiome-targeted interventions produce the expected behavioural, neurochemical, or inflammatory effects; once it is severed, those effects disappear or are sharply attenuated. The evidence falls into three lines.

The single-strain probiotic studies are the cleanest test. Bravo et al.¹² showed that Lactobacillus rhamnosus JB-1 increased hippocampal GABA-receptor expression and produced anxiolytic behaviour in BALB/c mice — effects entirely abolished by SDV. Sgritta et al.¹⁸ replicated this dependency in an autism-relevant context: Lactobacillus reuteri rescued social behaviour deficits in Shank3B knockout mice only when the vagus was intact. Wang et al. (2020)¹⁴ demonstrated SDV blocked anhedonia phenotypes induced by Lactobacillus intestinalis and L. reuteri in antibiotic-treated mice.

Faecal microbiota transplantation (FMT) extends the same dependency to community-level interventions. Pu et al.¹⁶ showed FMT from depression-associated donors (Chrna7 knockouts) induced depression-like behaviour in recipients — an effect blocked by SDV. Siopi et al.¹⁷ demonstrated SDV abolished FMT-transferred behavioural and neurogenic phenotypes from chronically stressed donors.

Inflammatory and metabolic challenges round out the picture. Zhang J et al.¹³ showed SDV prevented lipopolysaccharide (LPS)-induced Lactobacillus depletion and depression-like behaviour. Wang et al. (2021)¹⁵ (Ephx2 knockouts), Yang Y et al.¹⁹ (Chrna7 knockouts), and Yue C et al.²⁰ (CFA chronic inflammatory pain) all demonstrated SDV blocked microbiome-dependent neurobehavioural phenotypes. Zhang Y et al.²¹ extended this to organ-specific manipulation: hepatic-branch vagotomy in murine cirrhosis altered Bacteroidetes:Firmicutes ratio and reduced neuroinflammatory markers. Across these ten studies (Table 1), microbiome-mediated phenotypes do not emerge when vagal signalling is disrupted in the relevant arc.

One boundary case is worth noting. Bercik et al.²⁴ showed antimicrobial-induced behavioural changes in adult mice were not abolished by vagotomy or sympathectomy, indicating that not all microbiome–behaviour effects are vagally mediated. The abolition seen above applies to interventions where the vagal arc is the main afferent route; antimicrobial effects seem to work through other pathways.

1.2. Human Observational Evidence: Vagal Tone and Microbiome Composition Co-Vary

Two human cohort studies, one large and cross-sectional, the other a smaller community sample, show that vagal tone and microbiome composition co-vary at the population level. Tsubokawa et al.²⁵ studied 950 Japanese adults and found higher heart rate variability (SDNN) correlated with greater microbial diversity (Shannon index) and enrichment of Lachnospiraceae and Bifidobacterium after multivariable adjustment. Ravenda et al.²⁶ studied 75 Italian community participants and demonstrated that lower vagal tone was associated with altered β-diversity (PERMANOVA P = 0.006), Prevotella enrichment, and Faecalibacterium prausnitzii depletion. The two cohorts differ in size, geography, and methods, yet point the same way: parasympathetic activity tracks anti-inflammatory commensal abundance.

1.3. Human Interventional Evidence: tVNS Shifts Microbiome Composition

Liu et al.²² published the first randomised controlled trial of transcutaneous vagus nerve stimulation in a gastrointestinal disorder with pre-specified microbiome outcomes — irritable bowel syndrome with constipation (IBS-C), n = 40. Active taVNS (25 Hz, 30 min daily, 4 weeks) produced significant increases in Bifidobacterium abundance (P = 0.038) and faecal butyrate (P = 0.011) compared with sham. The change in HF-HRV during the intervention period correlated with symptom improvement (r = −0.54, 95% CI −0.73 to −0.27), providing the first human evidence of a quantitative link between vagal engagement and clinical response in a tVNS–microbiome trial of disorders of gut–brain interaction.

The auricular branch targeted by transcutaneous stimulation is anatomically distinct from the subdiaphragmatic vagal branches manipulated in the rodent vagotomy paradigms summarised in Section 1.1; whether transcutaneous stimulation effectively engages downstream gastrointestinal and immune effectors in humans remains an open empirical question. The framework introduced in Section 3 addresses this question through a within-subject index of vagal arc engagement.

2. The Translational Gap

The mechanistic case for the vagal–microbial axis is strong (Section 1), yet clinical neuromodulation has not taken up microbiome readout, which is largely absent from vagal neuromodulation trials in DGBI. A recent systematic review by Veldman et al.²³ identified seven randomised controlled trials of non-invasive vagus nerve stimulation in gastrointestinal disorders (alongside earlier pilot work of chronic implanted vagal stimulation in Crohn's disease²⁷). Microbiome assessment was not a pre-specified outcome in any of them; the only RCT in a gastrointestinal disorder to include it — the Liu trial in IBS-C²² — was published subsequently. The seven reviewed trials measured symptomatic, autonomic, or inflammatory endpoints in isolation, despite the conceptual complementarity of all four readouts and the now-routine availability of 16S rRNA sequencing.

This omission has consequences. Without paired microbiome and autonomic data, a positive trial can be read only at the level of the proximate physiological response, and whether the clinical benefit works through, despite, or independently of microbial change cannot be answered. Recent preclinical work — Pan et al.²⁸ using multi-omics profiling, and Zhang X et al.²⁹ in a chronic unpredictable mild stress rat model — demonstrates tVNS-induced microbial composition shifts toward anti-inflammatory profiles with concurrent normalisation of inflammatory cytokines and behavioural outcomes, supporting biological plausibility. Recent translational reviews have similarly highlighted this gap in neurodegenerative disease contexts^30,31.

The barrier is conceptual, not methodological. Stool microbiome sequencing is cheap relative to trial budgets; consensus methods exist for microbiome analytical pipelines³² and stool sample handling^33,34, and tVNS parameter standardisation is established³⁵; and disorders of gut–brain interaction are exactly the indication where the four readouts are most tightly coupled. What is missing is a framework that makes microbiome assessment the natural complement to autonomic and symptomatic measurement.

3. A Framework for Reactivity Phenotyping

3.1. The Clinical Problem

Three patients with Rome IV postprandial distress syndrome come through clinic in turn. Their symptom diaries look alike, endoscopy is normal, and all three test Helicobacter pylori negative. One improves on a proton pump inhibitor, the second on a tricyclic antidepressant, the third on neither. Baseline measurements — fasting microbiome composition, resting heart rate variability, serum inflammatory markers, validated questionnaires — give partial signals, but none of them reliably predicts which patient will respond to what. In functional dyspepsia, that choice is still made largely by trial and error.

The framework operationalises the demand-reveals-reactivity principle through bidirectional perturbation. An ascending challenge — a standardised 300 kcal mixed-macronutrient liquid meal, chosen to evoke a controlled postprandial response in functional dyspepsia — engages the gut-to-brain limb through enteroendocrine, immune, and visceral-afferent pathways. A descending challenge — transcutaneous auricular vagus nerve stimulation delivered as a 30-min peri-prandial block (T = 0 to T = 30 min) at perception-based titrated amplitude (typically perception threshold + 0.5 mA, 25 Hz, 250 μs, left cymba conchae) with consensus-standard parameters³⁵ — engages the brain-to-gut limb through cholinergic and anti-inflammatory effectors. The combination, delivered across two visits in counterbalanced sham/active order with biomarker observation continuing through T = 180 min postprandial, is intended to reveal how the coupled system responds to demand and to what extent the vagus can regulate that response.

3.2. Three Indices of Reactivity and Modulation

Each index is designed to answer a distinct clinical question (Box 1). Like Rome IV, KDIGO staging, and RECIST response criteria, the architecture is constructed a priori from theory and then tested empirically. No one of the three indices can be fully derived from the others; together they describe a reactivity surface that no existing assessment captures.

Autonomic readout enters all three indices through a sign-aligned multi-modal composite rather than any single physiological signal. Heart rate variability — the most accessible non-invasive correlate of cardiac vagal activity³⁶ — is shaped by respiration, baroreflex tone, posture, and attentional state, so reading vagal engagement off a single HRV value is unreliable. The framework therefore combines three complementary readouts: paced-breathing high-frequency HRV (recorded during a 0.25 Hz cued protocol that fixes the respiration confound), electrodermal activity as a marker of sympathetic outflow, and pupillometric reactivity as a central autonomic readout whose taVNS response has been reported in locus coeruleus–linked paradigms. Each enters as an absolute change from baseline, with a sign fixed by the literature: HF-HRV usually rises under vagal engagement and enters positively; electrodermal activity usually falls and enters negatively; pupil reactivity, whose direction under taVNS is less settled, enters positively under the locus coeruleus–activation hypothesis that Phase 1 tests. Combining the three reduces reliance on any one signal and lets the parasympathetic, sympathetic, and central readouts triangulate.

The Gut Reactivity Index (GRI) measures the size of the system's response to nutritional challenge under sham stimulation. It combines symptom burden as the area under the curve of the Numeric Rating Scale (NRS), magnitude of autonomic engagement, and acute inflammatory amplification over the full 0–180 min postprandial window — covering both the during-stim phase (0–30 min) and the post-stim cascade (30–180 min) — with each component z-standardised within the cohort. A high GRI flags a system that responds disproportionately to demand. The index plays two roles: a descriptive summary of overall reactivity reported at Phase 1, and the input to the within-subject contrast that defines the Vagal–Gut Modulation Index. GRI on its own does not direct therapy.

The Vagal Reserve Index (VRI) is a prerequisite check on the auriculo-vagal reflex arc, read from the multi-modal autonomic response to a 15-minute tonic taVNS segment given before the meal on the active visit. For each subject, VRI is the sign-aligned mean of three z-standardised absolute changes — in paced HF-HRV, electrodermal activity, and pupil diameter — between the stimulation segment and the rest period just before it. A low VRI marks a patient whose autonomic response to stimulation is blunted, which we read as a candidate marker of weak auriculo-vagal arc engagement — much like confirming that a target organ is reachable before predicting a drug response.

The Vagal–Gut Modulation Index (VGMI) is the within-subject change in the symptom–inflammatory response composite (GRIresp) between sham and active visits, measured under counterbalanced visit order. The autonomic-magnitude term is left out of this contrast because active stimulation is expected to raise vagal engagement, which would otherwise pull the difference against the symptom and inflammatory attenuation it is meant to detect. Each visit delivers a 30-min peri-prandial stimulation block (T = 0 to T = 30 min, sham or active) with biomarker observation continuing through T = 180 min, so that VGMI tests whether a defined peri-prandial taVNS dose modifies the 180-min cascade response to the same nutritional challenge. At the individual level, VGMI is a direct subtraction of within-subject indices; at the cohort level, hypothesis testing uses a mixed-effects model with subject as random effect and condition as fixed effect, exploiting within-subject pairing to suppress between-subject noise. A positive VGMI is interpreted as evidence that peri-prandial taVNS attenuates the response to the same challenge that produced the GRI; a near-zero VGMI indicates that the system's reactivity is, on this acute timescale and at the chosen dose, vagally unmodifiable. Because acute vagal accessibility and longer-term neuroplastic adaptation are distinct constructs, VGMI is interpreted as estimating an upper bound on acute modulability rather than as a deterministic gate on chronic response; sustained tVNS efficacy may exceed the acute prediction in patients with low VGMI. This distinction is reflected in the analytic role assigned to VGMI in Phase 2 as a continuous moderator rather than as an inclusion criterion (Section 4.2).

These three indices answer three distinct questions: how reactive is the system, how intact is the vagal arc, and how acutely modifiable is the response by vagal activation. Together they characterise whether and how strongly the system responds — but not why. Two patients with identical GRI and VGMI may require opposite first-line therapies, and existing measurements cannot tell them apart.

3.3. The Neural–Peripheral Dominance Index and the microbial Substrate Index

Take two patients with high GRI and high VGMI. Both respond acutely to vagal stimulation, yet the source of their reactivity differs. In one it is rooted in trait neural sensitisation — long-standing visceral hypersensitivity and affective comorbidity. In the other, peripheral processes dominate: acute postprandial inflammation and peripheral motor dysfunction. The first is a candidate for sustained vagal neuromodulation. The second responds acutely to taVNS but is liable to relapse without concurrent gut-directed therapy, especially against an unfavourable microbial substrate.

The clinical question is not whether the system responds, but along which axis the response is propagating — and against what microbial substrate.

The Neural–Peripheral Dominance Index (NPDI) casts this axis as a contrast between trait neural vulnerability and acute peripheral reactivity. The neural sub-score Z_N combines three trait anchors of central neural loading: visceral sensitisation (Visceral Sensitivity Index, VSI) and limbic–affective loading (Patient Health Questionnaire 9-item, PHQ-9, and Generalised Anxiety Disorder 7-item, GAD-7)³⁷. These are clinically accessible proxies for central vulnerability, not direct neural measurements; external anchoring through neuroimaging and vagal-evoked potentials is a validation priority (Section 5). The peripheral sub-score Z_P is computed from acute Δ measures across the standardised challenge under sham conditions: change in serum interleukin-6 (IL-6), tumour necrosis factor-α (TNF-α), and soluble CD14 (sCD14). In the Extended formulation, Z_P additionally incorporates a gastric motility marker. NPDI is defined as the difference of the standardised sub-scores, NPDI = z(Z_N) − z(Z_P); each sub-score is z-standardised before subtraction so that the contrast is balanced across sub-scores of unequal component count. Direction: positive NPDI = predominant trait neural loading relative to acute peripheral reactivity; negative NPDI = predominant peripheral reactivity.

This is a deliberate departure from earlier reactivity-only formulations. The autonomic engagement signal already sits in GRI, VRI, and VGMI; assigning trait neural anchors to Z_N gives NPDI a separate role: it captures the long-standing central context against which acute peripheral reactivity plays out, rather than re-summarising the same Δ inputs. The contrast is deliberate — chronic central susceptibility against acute peripheral reactivity — and follows the idea that symptoms arise from the interaction between predisposition and provocation, not from processes on a single timescale. Throughout, 'neural' and 'peripheral' name predominant loading profiles defined by their component measures, not direct anatomical or causal sites. Anchoring these labels to neuroimaging or transabdominal vagal-evoked potentials is a multi-centre validation priority (Section 5).

The Substrate Index (SI) captures the trait-level microbial context in which acute reactivity occurs. It is computed at V0 — the baseline visit, separate from the challenge visits — as an isometric log-ratio balance between pre-specified anti-inflammatory commensal and pro-inflammatory taxa (Box 1). Because microbiome composition shifts from day to day, a single-sample SI is best read as a baseline coordinate rather than a strict trait, and its test–retest reliability is reported as a Phase 1 feasibility outcome. SI is reported alongside NPDI, and the clinical implications come from the joint (NPDI, SI) coordinate rather than NPDI alone: two patients with the same NPDI but different SI map to mechanistically distinct profiles. Keeping SI as a parallel coordinate, rather than folding it into NPDI, keeps dynamic acute reactivity and stable microbial substrate dimensionally separate.

Together the framework yields five coordinates: three reactivity indices (GRI, VRI, VGMI), one dominance contrast (NPDI), and one substrate coordinate (SI). Although the indices share some component variables, contrast and mean constructions are intended to extract orthogonal information from overlapping inputs. Non-redundancy is operationalised as a pre-registered hypothesis that the partial correlations of NPDI with each of GRI, VRI, VGMI and SI — controlling for the Δ inflammatory composite that NPDI and GRI share by construction — fall below 0.50, with parallel principal component analysis reported as a sensitivity check. A value at or above this threshold triggers pre-specified post-hoc review of the formula.

The framework makes a directly testable joint prediction: matching treatment to the (NPDI, SI) coordinate by mechanism outperforms mismatched pairing, with VGMI moderating the response so that acute vagal modulability scales the size of any taVNS effect. NPDI-positive patients are expected to respond to vagal neuromodulation, with effect size scaled by VGMI; NPDI-negative patients with low SI are expected to need motility- and symptom-directed therapy; and NPDI-negative patients with high SI are expected to need microbiome-directed therapy, as primary or adjunct treatment (Fig. 3). Phase 2 tests this joint prediction through a stratified factorial design (Section 4).

Figure 3. Joint (NPDI, SI) coordinate and mechanism-concordant therapy assignment. Two-dimensional phenotype map combining the Neural–Peripheral Dominance Index (NPDI; x-axis) and the microbial Substrate Index (SI; y-axis). Positive NPDI indicates predominant trait neural loading (visceral sensitisation, affective comorbidity); negative NPDI indicates predominant acute peripheral reactivity (postprandial inflammatory and motor loading). Positive SI indicates greater microbial substrate burden (relative depletion of anti-inflammatory commensals); negative SI indicates a more favourable substrate. The three mechanism-concordant predictions tested in Phase 2 are: (i) neural-dominant (NPDI+) regardless of SI — vagal neuromodulation as primary therapy, with treatment effect scaled by VGMI; (ii) peripheral-reactive with low substrate burden (NPDI−, SI−) — motility- and symptom-directed therapy; (iii) peripheral-reactive with high substrate burden (NPDI−, SI+) — microbiome-directed therapy (low-FODMAP + soluble fibre) as primary or adjunct intervention. Quadrant boundaries are cohort-referenced zero-crossings of the standardised indices; their stability against published clinical cut-offs is evaluated as a sensitivity analysis, and concordance between categorical and continuous NPDI analyses is pre-registered as a confirmatory criterion. The therapeutic implications shown are pre-registered predictions for prospective Phase 2 testing, not validated treatment assignments. Abbreviations: NPDI, Neural–Peripheral Dominance Index; SI, Substrate Index; taVNS, transcutaneous auricular vagus nerve stimulation; VGMI, Vagal–Gut Modulation Index.

Figure 3. Joint (NPDI, SI) coordinate and mechanism-concordant therapy assignment. Two-dimensional phenotype map combining the Neural–Peripheral Dominance Index (NPDI; x-axis) and the microbial Substrate Index (SI; y-axis). Positive NPDI indicates predominant trait neural loading (visceral sensitisation, affective comorbidity); negative NPDI indicates predominant acute peripheral reactivity (postprandial inflammatory and motor loading). Positive SI indicates greater microbial substrate burden (relative depletion of anti-inflammatory commensals); negative SI indicates a more favourable substrate. The three mechanism-concordant predictions tested in Phase 2 are: (i) neural-dominant (NPDI+) regardless of SI — vagal neuromodulation as primary therapy, with treatment effect scaled by VGMI; (ii) peripheral-reactive with low substrate burden (NPDI−, SI−) — motility- and symptom-directed therapy; (iii) peripheral-reactive with high substrate burden (NPDI−, SI+) — microbiome-directed therapy (low-FODMAP + soluble fibre) as primary or adjunct intervention. Quadrant boundaries are cohort-referenced zero-crossings of the standardised indices; their stability against published clinical cut-offs is evaluated as a sensitivity analysis, and concordance between categorical and continuous NPDI analyses is pre-registered as a confirmatory criterion. The therapeutic implications shown are pre-registered predictions for prospective Phase 2 testing, not validated treatment assignments. Abbreviations: NPDI, Neural–Peripheral Dominance Index; SI, Substrate Index; taVNS, transcutaneous auricular vagus nerve stimulation; VGMI, Vagal–Gut Modulation Index.

The (NPDI, SI) coordinate sits as a mechanistic layer beneath Rome IV, not a replacement for it. Rome IV sorts patients by symptom pattern, which is reliable for epidemiology and regulatory use but is not designed to separate underlying mechanisms. The pre-registered Phase 1 hypothesis is that, within Rome IV PDS, the variance in NPDI explained by PDS-defining items alone falls below R² = 0.25, and that the joint (NPDI, SI) coordinate predicts response to mechanism-specific therapy. Extension to epigastric pain syndrome and other DGBI is planned for later phases. If the joint coordinate fails to pick out mechanistically distinct treatment-response profiles in Phase 2, the central premise of the framework is in doubt — because predictive value under perturbation is exactly what sets this apart from static stratification.

3.4. Design Principles

Three commitments are fixed before data collection.

First, tiered portability. The framework is specified at two tiers, so its validity does not depend on how well-resourced a centre is. Tier 1 (Core) uses measurements computable without specialised neurophysiology or motility infrastructure: standard ECG patch for paced-breathing HRV, an electrodermal sensor, and a desktop pupillometer. Pupillometry is research-grade equipment — inexpensive and portable, but not yet routine in secondary-care gastroenterology; Core deployment is therefore feasible in research settings and in centres with established autonomic-testing capacity. Tier 2 (Extended) additionally incorporates gastric motility (electrogastrography or ¹³C-octanoic acid breath test) in the peripheral sub-score and serum cortisol time-course as an HPA-axis (hypothalamic–pituitary–adrenal axis) anchor. Both tiers are pre-registered as comparable through a common base.

Second, compositional rigour. Microbiome relative abundances are inherently compositional and require log-ratio treatment for valid statistical inference³². The Substrate Index is computed as an isometric log-ratio balance with Bayesian multiplicative replacement of zero counts³⁸. The numerator and denominator taxa sets (Box 1) are locked prior to Phase 1 data collection. All composite-index inputs are first winsorised at ±3 SD on raw values, then z-standardised within the cohort before averaging — an order pre-registered to ensure replicability.

Third, reactivity over baseline, with trait anchoring where required. The framework is principally perturbation-based: GRI, VRI, VGMI, and the peripheral sub-score of NPDI enter as Δ values across the standardised challenge. Trait anchors (VSI, PHQ-9, GAD-7 in the neural sub-score of NPDI; SI as the substrate coordinate) are incorporated where transient challenge dynamics alone are insufficient to identify stable mechanistic loading. This division — acute reactivity in GRI/VRI/VGMI and Z_P, trait loading in Z_N and SI — is deliberate, preserving the framework's commitment to measuring how the system responds while acknowledging that mechanism-discriminating contrasts require both response and predisposition information.

The five-coordinate architecture (GRI, VRI, VGMI, NPDI, SI) is deliberately over-complete at the proof-of-concept stage, so that it can be falsified along several dimensions at once; later phases are expected to drop the dimensions that do not contribute, leaving a clinically minimal version. Pre-registration limits analytical freedom and backs the claim that the framework is predictive, not just descriptive.

4. A Two-Phase Validation Strategy

4.1. Phase 1—Proof-of-Concept

Phase 1 asks whether the framework can be computed reliably in real patients and whether its pre-registered structural predictions hold within a single Rome IV subtype. A proof-of-concept cohort of patients with Rome IV postprandial distress syndrome and demographically matched healthy controls (n = 20 per group) completes a baseline visit followed by two experimental visits in counterbalanced randomised order. Each experimental visit follows the same timeline: pre-meal acclimatisation, 15-min tonic stimulation segment (sham or active depending on visit assignment) yielding the VRI input, 20-min physiological washout, 5-min paced-breathing reference for autonomic baseline, then a standardised 300 kcal mixed-macronutrient meal at T = 0. Stimulation is delivered as a 30-min peri-prandial block from T = 0 to T = 30 min (matching the visit's sham/active assignment), with biomarker observation continuing through T = 180 min. Stimulation amplitude is individualised by perception-based titration at each visit and recorded as a covariate. The sham visit yields GRI(sham) and the input measures to NPDI; the active visit yields GRI(active) and, by within-subject difference, VGMI. SI is computed from a baseline stool sample at the baseline visit. Preclinical evidence underpinning the framework was selected against contemporary methodological reporting standards³⁹ to ensure mechanistic claims rest on adequately powered, bias-controlled models.

Phase 1 reports six pre-registered hypotheses through descriptive statistics with confidence intervals (not formal hypothesis tests, which require Phase 2 power) covering distributional spread of NPDI within Rome IV PDS and balanced representation of NPDI signs (H1), non-redundancy of NPDI against the other reactivity indices and SI (H2), within-subtype orthogonality against Rome IV PDS items (H3), external convergent validity (H4), feasibility of predictive validity testing (H5; formal predictive testing is reserved for Phase 2), and substrate–reactivity interaction (H6). Sample size justification, hypothesis thresholds, blinding and stimulation parameter specifications, the reliability criterion for repeat measurement of GRI under sham conditions, the multi-modal autonomic acquisition protocol, adverse-event reporting standards, and go/no-go criteria for progression to Phase 2 are pre-registered in the Supplementary Protocol prior to recruitment (OSF registration DOI 10.17605/OSF.IO/46X9F), with adjudication by an independent data monitoring committee.

4.2. Phase 2—Factorial Validation

Phase 2 tests the framework's central clinical claim — that the joint (NPDI, SI) coordinate predicts mechanism-specific treatment response — in an independent multi-centre cohort of about 120 patients with Rome IV PDS. The design is a 2 × 2 factorial: participants are first phenotyped using the Phase 1 perturbation protocol, then stratified by NPDI sign and randomised within stratum to either home-based taVNS³⁵ or a structured low-FODMAP diet supplemented with soluble fibre over twelve weeks. SI enters the primary model as a continuous covariate and VGMI as a continuous moderator of the treatment × NPDI interaction, so that acute vagal modulability scales the expected magnitude of taVNS-driven effect (Section 3.3).

The dietary arm targets the peripheral and substrate axes through complementary mechanisms: low-FODMAP restriction reduces fermentable substrate available to gas-producing taxa⁴⁰, and soluble fibre supports butyrate-producing commensal abundance⁴¹. The combination is selected over single-strain probiotic comparators evaluated in adjacent DGBI⁴² on grounds of mechanism alignment with the peripheral and substrate coordinates. The pre-specified directional prediction is that mechanism-concordant treatment cells outperform discordant cells on a composite primary outcome combining the PAGI-SYM and the FD-specific Nepean Dyspepsia Index, with convergent improvement on both instruments above their established minimal clinically important differences reported as a co-primary responder analysis.

Treatment dosing parameters, sample-size justification, randomisation and blinding procedures, primary and secondary statistical models, durability and responder endpoints, continuous-predictor sensitivity analyses, and pre-registered out-of-sample validation are detailed in the Supplementary Protocol. Subsequent empirical calibration through regularised regression may refine relative component weighting; this refinement complements the a priori Phase 1 architecture rather than replacing it, with concordance between the a priori and empirically calibrated formulations required for full claim of phenotype-guided stratification.

5. Research Priorities

5.1. Microbiome Embedding in tVNS Trials

The most immediately feasible priority is to embed microbiome assessment as a secondary outcome in ongoing and planned tVNS trials. The marginal cost of 16S rRNA stool sequencing (US$50–150 per sample) is small compared with the trial budget³³, and biological plausibility is supported both by the Liu trial demonstrating Bifidobacterium and butyrate shift over four weeks²² and by recent preclinical multi-omics work demonstrating tVNS-induced anti-inflammatory microbial shifts in chronic stress models²⁸. Minimum reporting standards developed jointly between the bioelectronic medicine and microbiome communities, modelled on existing tVNS methodology consensus³⁵, would convert the current ad-hoc landscape into systematic evidence.

5.2. Cross-Disorder Extension

Beyond Phase 1 and Phase 2, three cross-disorder extensions are tractable: IBS-D, where motility–microbiome interaction provides a distinct test of the (NPDI−, SI+) coordinate⁴⁰; IBD–DGBI overlap presentations, where the vagal anti-inflammatory mechanism has direct therapeutic relevance and validated inflammatory markers anchor the peripheral sub-score independently of subjective symptoms^43,44; and major depressive disorder with prominent gastrointestinal comorbidity, where the framework's bidirectional logic intersects with an established clinical population⁴⁵. Cross-disorder application tests whether the joint (NPDI, SI) coordinate captures a partially transdiagnostic dimension or whether mechanism-specific phenotypes are more disorder-bound than the framework's logic anticipates.

5.3. Neurobiological Validation Priorities

The framework's 'neural' and 'peripheral' labels rest on operational definitions; their mapping to anatomically distinct subsystems is empirically open. Four strategies would close the gap between these operational labels and a direct mechanistic readout. First, functional neuroimaging during the nutritional challenge — with regions of interest including the insular cortex, the nucleus tractus solitarius, and the limbic regions co-activated in visceral perception — would test whether NPDI-positive patients show differential central engagement relative to NPDI-negative patients. Second, transabdominal or cervical vagal-evoked potentials would provide a direct electrophysiological readout of vagal arc engagement, complementing the indirect multi-modal autonomic composite. Third, time-resolved measurement of autonomic–central coupling (for example, brain–HRV coherence during challenge) would test whether the trait psychometric anchors in Z_N track moment-to-moment central modulation as predicted. Fourth, integration of stimulation-amplitude dose-response with central engagement readouts is now tractable: a recent study using 4-min taVNS in sleep-deprived healthy participants demonstrated that higher self-titrated stimulation amplitudes (range 12–26 mA at the tragus) were associated with smaller decrements in vigilance and mood under conditions of high physiological pressure⁴⁶, consistent with dose-dependent vagal fibre recruitment in that setting. Because this evidence comes from a different electrode site, a much higher amplitude range, and a non-gut task, whether the same relationship operates in a postprandial gut-brain paradigm and links to NPDI stratum in disorders of gut-brain interaction is an open Phase 3 priority. These four priorities define a neurobiological validation track that would convert the framework's mechanistic claims from operational to anatomical.

5.4. Phase 3 Confirmatory Programme

Phase 3 is conceived as a multi-centre confirmatory programme integrating three priorities: external validation of the (NPDI, SI) stratification across additional sites with diverse cohorts; external anchoring of the framework's mechanistic labels through the neurobiological measurements outlined in Section 5.3; and sample-size recalibration to detect clinically meaningful but smaller treatment × NPDI interactions than Phase 2 is powered to identify. Mechanistic dissection of the cholinergic anti-inflammatory pathway and short-chain fatty acid signalling — through targeted antagonists of α7 nicotinic acetylcholine receptors⁹ or FFAR2/3 blockade⁴⁷ alongside tVNS — would assign causal weight to each candidate mediator, alongside formal causal mediation analysis⁴⁸. Integration with existing tVNS–microbiome efforts, including an obesity protocol combining tVNS with 16S rRNA profiling⁴⁹, would accelerate cross-indication evidence.

5.5. Limitations

Methodological. The autonomic composite, although designed to mitigate single-signal dependence, depends on equipment — pupillometry in particular — not yet routine in clinical gastroenterology; broader translation will require simplified implementations using only HRV and electrodermal channels. Sham control in tVNS is methodologically imperfect, since sub-threshold electrical stimulation is not biologically inert; blinding fidelity is reported as a feasibility outcome at both phases. Phase 2 lacks a no-treatment or waitlist arm, so absolute treatment effects relative to natural history are not directly estimable; the test of mechanism-concordance through the treatment × NPDI interaction is, however, robust to this limitation by design.

Design and scope. The cohort selectivity of proof-of-concept will limit early generalisability, and explicit external validation is required before clinical claims extend beyond research deployment. The dietary comparator combines low-FODMAP restriction and soluble fibre on mechanism-specificity grounds; the relative contribution of each component is a Phase 2 secondary analysis rather than an established prior. Stimulation dose-finding remains an active research question: the 30-min peri-prandial paradigm at perception-based titrated amplitude reflects current parameter consensus while permitting individualisation, but whether longer or repeated stimulation blocks would amplify acute modulability remains to be tested. Recent dose-response evidence in a healthy population subjected to sleep deprivation⁴⁶ reports that higher self-titrated amplitudes are associated with smaller vigilance and mood decrements during periods of greatest physiological pressure. This is hypothesis-generating only: it comes from a different electrode site (tragus), an amplitude range far above this protocol's titrated dose, and a non-gut paradigm, and it concerns delivered amplitude whereas the protocol delivers a fixed PT + 0.5 mA, so MCA reflects tolerance capacity rather than dose. MCA is therefore carried as an exploratory moderator of taVNS response in Phase 2, not a pre-registered confirmatory one (see Supplementary Protocol). Whether tVNS-induced microbiome effects are stimulation-frequency dependent — Li et al. demonstrated 20 Hz effects in a rodent chronic stress model⁵⁰ — remains unresolved in humans. Paediatric and elderly populations, cross-cultural microbiome calibration, and integration with pharmacological treatment strata define the scope of work that follows Phase 2.

Statistical and framework status. Power calculations for Phase 2 assume a moderate-to-large treatment × NPDI interaction (Cohen's f = 0.30); since clinical interaction effects are often smaller, Phase 2 is powered as a proof-of-signal for a sizeable interaction rather than a definitive test, and smaller but clinically meaningful effects would not be reliably detected at n = 120. A null result there would not by itself rule the framework out; definitive testing of smaller effects is deferred to an adequately powered Phase 3. The framework is not yet validated; what is offered here is a structured pathway to validation, with pre-specified criteria for both progression and abandonment. If the pathway is travelled and the criteria met, precision medicine in functional dyspepsia — and, in time, across the broader landscape of disorders of gut–brain interaction — moves from aspiration to instrument.