Abstract
Context. Very massive stars (VMSs), with masses in excess of 100 M⊙, are known in the Galaxy and the Large Magellanic Cloud (LMC). They are mostly characterised by their strong stellar winds compared to normal massive stars. Their mass-loss rates have been calibrated at the metallicity of the LMC. No constraints exist at other metallicities.Aims. We aim to study the spectroscopic appearance of VMSs and their effect on the integrated light of starbursts at low metallicity.Methods. In the absence of empirical constraints, we adopted two frameworks for the mass-loss rates of VMSs: in one case, we assumed no metallicity dependence; in the other case, we assumed a linear scaling with metallicity. Under these assumptions, we computed evolutionary models for masses 150, 200, 250, and 300 M⊙ at Z = 0.2, 0.1 and 0.01 Z⊙. We computed the associated synthetic spectra at selected points along the evolutionary tracks. Finally, we built population synthesis models including VMSs based on our new VMS models.Results. We find that the evolution of VMSs depends critically on the assumptions regarding mass-loss rates. In case of no metallicity dependence, VMSs remain hot for all their lifetimes. Conversely, when mass-loss rates are reduced because of lower metallicity, VMSs follow a classical evolution towards the red part of the Hertzsprung-Russell diagram. VMSs display He II 1640 emission in most phases of their evolution, except when they become too cool. This line is present in the integrated light of population synthesis models down to 0.1 Z⊙ whatever the star formation history, and is also sometimes seen at Z = 0.01 Z⊙. He II 1640 is weaker in models that include a metallicity scaling of the mass-loss rates. The optical spectra of starbursts, especially the Wolf-Rayet bumps, sometimes display VMS signatures when these stars are present. At low metallicities, adding VMSs to population synthesis models produces more ionising photons down to ~45 eV. At higher energies, the ionising flux depends on age, metallicity, assumptions regarding VMS mass-loss rates, and on the very short phases at the end of VMS evolution. He II ionising fluxes large enough to produce some amount of nebular He II 4686 emission can be produced under specific circumstances. Our models are able to reproduce qualitatively and sometimes also quantitatively the UV spectra of star-forming regions. However, we are not able to clearly identify which mass-loss framework is favoured.Conclusions. Very massive stars can be identified down to 0.1 Z⊙, and potentially to 0.01 Z⊙ depending on the mass-loss rates’ metallicity scaling, through their He II 1640 emission. Their detailed evolution at these low metallicities, especially their mass-loss rates, can be constrained when more UV spectra of star-forming regions at low metallicities are available.