High 3He/4He ratios in some basalts have generally been interpreted as originating in an incompletely degassed lower-mantle source1, 2, 3, 4, 5, 6, 7, 8, 9. This helium source may have been isolated at the core-mantle boundary region since Earth's accretion4, 5, 6. Alternatively, it may have taken part in whole-mantle convection and crust production over the age of the Earth7, 8, 9; if so, it is now either a primitive refugium at the core-mantle boundary8 or is distributed throughout the lower mantle7, 9. Here we constrain the problem using lavas from Baffin Island, West Greenland, the Ontong Java Plateau, Isla Gorgona and Fernandina (Galapagos). Olivine phenocryst compositions show that these lavas originated from a peridotite source that was about 20 per cent higher in nickel content than in the modern mid-ocean-ridge basalt source. Where data are available, these lavas also have high 3He/4He. We propose that a less-degassed nickel-rich source formed by core-mantle interaction during the crystallization of a melt-rich layer or basal magma ocean5, 6, and that this source continues to be sampled by mantle plumes. The spatial distribution of this source may be constrained by nickel partitioning experiments at the pressures of the core-mantle boundary.