AI-Guided Stability Tuning of a Heterodimeric Linker for Programmable Protein Tube Architectures

The ability to rationally modulate the morphology of artificial higher-order protein assemblies remains limited, particularly for geometrically constrained architectures such as tubes. Here, we show that simple tuning of the interaction stability within a heterodimeric coiled-coil linker, guided by deep-learning-based predictions from ThermoMPNN, provides an effective strategy for programming the assembly behaviour of two-component protein tubes. By engineering a systematic stability gradient within the M3L2/p66α heterodimeric linker, we generated single-amino-acid variants that exhibit predictable shifts in the temperature window for tube formation and produce tubes with distinct diameters. Notably, the least stable variant uniquely accesses multilayered tube-in-tube architectures. Time-resolved electron microscopy reveals that these hierarchical structures arise through a sequential process involving the initial formation of thin tubes, subsequent wall thickening, and eventual development of nested tubes, highlighting the role of transient linker rearrangements near the thermal transition. Together, these findings establish coiled-coil stability as a minimal and tunable design parameter that governs higher-order morphology and offer a minimal and programmable framework for constructing rationally designed protein tube architectures.