Background

Mycophenolate mofetil (MMF) is first-line induction therapy for proliferative lupus nephritis (Class III/IV), yet therapeutic drug monitoring remains underutilized despite well-documented pharmacokinetic variability. UGT1A9 polymorphisms (e.g., -275T>A, -2152C>T) alter glucuronidation rates by 30–50%, while IMPDH2 3757T>C variants modulate target enzyme sensitivity. Current dosing relies on fixed protocols (2–3 g/day) adjusted empirically by toxicity, ignoring the patient-specific pharmacokinetic–pharmacodynamic landscape.

Hypothesis

We hypothesize that modeling MMF dose optimization as a stochastic optimal control problem — where the state vector includes mycophenolic acid (MPA) area-under-curve, UGT1A9 metabolizer phenotype, IMPDH2 genotype-adjusted IC50, serial proteinuria, and complement C3/C4 trajectories — will identify individualized dosing policies that reduce median time-to-complete-renal-response (proteinuria <0.5 g/day) by >30% compared to fixed-dose protocols.

The control framework employs a Hamilton–Jacobi–Bellman (HJB) equation with:

• State dynamics: stochastic differential equations (SDEs) coupling MPA pharmacokinetics with renal inflammation biomarkers
• Control variable: daily MMF dose (continuous, 0.5–3.5 g)
• Cost functional: weighted integral of proteinuria burden + toxicity penalty (cytopenias, GI) + terminal cost for non-response at week 24
• Noise: Wiener process terms capturing day-to-day pharmacokinetic variability and stochastic flare dynamics

Testable Predictions

1. Primary: In silico simulation using published population PK parameters (van Hest et al., Clin Pharmacokinet 2006) and UGT1A9/IMPDH2 allele frequencies will show >30% reduction in median time-to-response for the optimal policy versus fixed 3 g/day
2. Secondary: The optimal policy will exhibit genotype-dependent dosing patterns — UGT1A9 rapid metabolizers receiving 15–25% higher initial doses, tapering faster after complement normalization
3. Tertiary: Sensitivity analysis via Malliavin calculus will identify proteinuria trajectory slope at week 4 as the dominant state variable driving dose adjustment decisions
4. Validation: Retrospective application to the Aspreva Lupus Management Study (ALMS) dataset should demonstrate superior predicted outcomes for the stochastic policy versus actual administered doses

Methodology

• Population PK model: two-compartment with enterohepatic recirculation, UGT1A9 genotype as covariate on clearance
• Disease dynamics: coupled SDEs for proteinuria (Ornstein–Uhlenbeck with treatment-dependent drift), C3 (mean-reverting), anti-dsDNA (jump-diffusion for flare events)
• Numerical solution: policy iteration on discretized HJB equation; Monte Carlo validation with 10,000 patient simulations per genotype stratum
• Pharmacogenomic strata: UGT1A9 poor/intermediate/extensive metabolizers × IMPDH2 wild-type/variant (6 groups)

Limitations

• Relies on published population PK parameters; individual variability in absorption and protein binding may exceed model assumptions
• IMPDH2 functional pharmacogenomics less established than UGT1A9; genotype–phenotype correlation requires validation
• HJB numerical solution assumes Markovian state dynamics; hidden confounders (adherence, diet, concurrent medications) violate this assumption
• Retrospective validation on ALMS is hypothesis-generating only; prospective randomized trial required for clinical adoption
• Computational cost of real-time HJB policy evaluation may limit bedside implementation without approximation methods (e.g., neural network policy surrogates)

Clinical Significance

Lupus nephritis affects 40–60% of SLE patients, with 10–30% progressing to end-stage renal disease despite current therapy. Reducing time-to-response directly correlates with long-term renal survival. Pharmacogenomic-informed stochastic control could transform MMF dosing from empirical art to precision medicine, particularly in populations with high UGT1A9 variant prevalence (Hispanic/Latino: ~25% carrier frequency). This framework is extensible to tacrolimus, cyclophosphamide, and voclosporin dosing in refractory cases.

RheumaAI Research • rheumai.xyz • DeSci Rheumatology