The N-Terminus of Sophora tonkinensis Cytochrome P450s Evolves Neutrally yet Encodes Rich Functional Information: A Protein Language Model Analysis

Abstract The N-terminal membrane anchor of cytochrome P450s is essential for function yet exceptionally variable—a paradox resistant to alignment-based analysis. Using an alignment-free protein language model (ESM2) pipeline on 345 Sophora tonkinensis P450s, we show the N-terminal 50 residues evolve under pervasive neutral relaxation (mean dispersion 0.62), punctuated by a single constrained island: a PxxG structural hinge (P 21 , G 24 conserved at 98%/99%). Despite lacking sequence-level constraint, embedding-space decomposition resolves this scaffold into two separable channels. DIVA (unsupervised) captures a topological template encoding membrane-anchoring architecture, validated by physicochemical correlations recapitulating ER signal-anchor insertion and confirmed by DeepLoc- 2.1. PIVOT (supervised ablation) captures a family-associated signal whose peak position distinguishes CYP families (ε 2 = 0.44, permutation p < 0.0001) despite label-free training. The channels are largely separable: PIVOT peaks lie downstream of DIVA peaks in 84.4% of proteins, with mutually exclusive signal-loss sets. The P450 N-terminus thus integrates membrane-topological and family-identity information within a single relaxed region. Neutral evolution is reframed not as absence of information, but as the condition enabling multiplexed functional encoding—a mechanistic rationale for the functional importance of this variable anchor.