Multistable attractor dynamics as a systems level hypothesis for treatment refractoriness in rheumatic disease

Abstract Treatment refractoriness in chronic rheumatic disease — formalised for rheumatoid arthritis (RA) by the EULAR 2021 difficult-to-treat (D2T) criteria and affecting ~12% of patients — is not addressable by any single-pathway model. We ask whether a multi-axis dynamical-systems framework can generate, as a testable hypothesis, the attractor-like persistence of refractory disease. We present the 3-Axis Integrative Framework (3-AIF), a six-variable ordinary differential equation (ODE) system coupling mucosal tolerance (Axis 1), the energy-gated Nerve–Adipose–Immune (NAM) danger-sensing unit (Axis 2), and the integrated stress response (Axis 3) through Hill-function cross-talk and an mTORC1–AMPK metabolic gate. Under a biologically motivated parameter set, the system is multistable, with three co-existing stable attractors—healthy, refractory-disease, and severe-collapse—separated by two saddle points, and shows fold-catastrophe hysteresis, mirroring the clinical spectrum from partial response to complete refractoriness. Local sensitivity analysis ranks three Axis-2 parameters (parasympathetic resolution, danger auto-amplification, and environmental forcing) as dominant, consistent with emerging vagus nerve-targeted neuroimmune modulation in RA. Single-axis ablation confirms that cross-axis coupling is required for the fold transitions. A cross-disease transcriptomic consistency analysis across six public GEO datasets, using an expression- and variance-matched permutation null, finds the 3-AIF axis genes significantly more dysregulated than matched background genes in the two RA cohorts ( p = 0.0016 and p < 0.0001) but not in the smaller non-RA cohorts. Because these datasets lack documented D2T patients, we frame the results as molecular plausibility rather than validation, and note the need for longitudinal multi-omics data in formally classified D2T cohorts.