A permeable G310 constriction limits K+ flux in the activated KCNQ2 pore and is disrupted by the encephalopathy variant G310S
KCNQ2 (Kv7.2) subunits contribute to the channels that carry the neuronal M-current, and variants in KCNQ2 cause epilepsies that range from self-limited neonatal seizures to developmental and epileptic encephalopathy (DEE). Structures of KCNQ2 bound to PIP2 and cannabidiol show a widened activation gate, yet it has remained unclear which point of the activated pore limits K+ flux and how DEE variants at that point act. We simulated the KCNQ2 pore domain (PDB 8J01) in a POPC bilayer under an applied field of about 565 mV, restraining only the terminal residues of S5 and S6 so that G310 and S314 were free. The activated pore conducted K+ at 4.3 pS (95% CI 2.5-6.9; 17 net filter crossings in 1120 ns over four replicas), while the closed-state pore (PDB 7CR3) conducted none in 560 ns. Conduction proceeded by knock-on through a filter that was doubly occupied in 84-92% of frames. Along S6 the narrowest position of the activated pore was not the canonical gate residue S314 but the glycine of the Pro-Ala-Gly motif, G310: activation widens S314 by 0.37 nm but G310 by only 0.07 nm, and restraining G310 to its closed-state width in one replica abolished gate crossings. The DEE variant G310S reduced conduction about six-fold (G310S/wild-type ratio 0.16, 95% CI 0.02-0.66; p = 0.003). Its serine hydroxyls project into the constriction and halve the water column in the G310 plane (8.7 to 4.7 waters), an effect reproduced in CHARMM36m (9.9 to 4.3). The first coordination shell of a K+ ion at the constriction was, however, conserved (6.3 vs 6.5 oxygens), because the serine hydroxyls replace the waters they displace. The hydroxyls did not retain the ion: contacts lasted tens of picoseconds, and ions were instead turned back at the serine ring, fewer reaching the gate plane (45 vs 79 excursions per microsecond) and fewer of those crossing it (5% vs 23%). In mixed tetramers carrying one or two variant subunits the effect on the water column was additive, each subunit removing about a quarter of it (f = 0.22 and 0.58 of the full effect). These results identify G310 as the flux-limiting constriction of the activated KCNQ2 pore in our simulations and point to a permeation defect as one component of G310S loss of function.