Enhanced Geothermal Systems (EGS) allow for worldwide geothermal electricity production. They target deep (3-5 km), fractured rock reservoirs whose permeability is artificially increased through hydraulic stimulations (fluid injections). The injections modify the effective stresses, leading to new fracture creation and the reactivation of pre-existing ones, along with permeability changes. Today, the main issues with this technique are i) understanding and predicting the final fluid transport capacity of the reservoir and ii) avoiding the seismicity associated with the stimulation treatment. This thesis aims at understanding the two-way coupling between mechanical rock deformation and fluid transport in EGS reservoirs by reproducing them in the laboratory. Complex experimental protocols were developed and applied to various lithologies typically found in EGS. The experimental observations, coupled with analytical and numerical models, and with natural observations, allowed for the unravelling of some fundamental processes at play during EGS stimulations. A first part focused on how mechanical deformation influences fluid flow in i) anisotropic rocks and ii) fractures with customized roughness. In anisotropic rocks, the foliation orientation towards the stress field controls its mechanical and hydraulic transport properties. With ongoing deformation, anisotropic micromechanical models accurately predict the onset of damage, ultimate strength, porosity evolution, and fracture structure towards foliation orientation. Permeability is controlled by foliation orientation, fracture structure, and the applied stress. This shows that the full permeability tensor needs to be used with care on failed anisotropic rocks. In rock fractures with customized roughness, increasing the normal load decreases fracture transmissivity and thus fluid transport capacity. The transmissivity decrease is controlled by the contact geometry, as predicted by numerical models. Further, reversible shear loading and irreversible shear displacements (up to 1 mm offset) have little effect on transmissivity. Transmissivity evolution with shear displacement can be predicted by simple models only at low normal stress, where wear is not too prominent. These results question the concept of hydro-shear stimulations with small fault displacements. A second part targeted the effect of fluids across the earthquake cycle (nucleation and propagation) to approach induced seismicity. During nucleation, slight changes in fluid pressure drastically change the temporal evolution of earthquake precursors (slip and seismicity). Under all fluid pressure conditions, a semi empirical scaling relationship links the total moment (energy) released before the mainshock to its magnitude, in compatibility with several natural and anthropogenic earthquakes. This observation could help with real-time estimation of an impending earthquake's magnitude. During propagation, fluid pressure has a strong control on the thermal evolution of frictionally-heated asperity contacts. At low pressures, fluids vaporize and allow for contact melting, reducing shear resistance and leading to large magnitude quakes. At pressures close to the liquid-supercritical transition, water is an efficient thermal buffer, to the point that contact melting is impeded, leading to smaller quakes with little dynamic weakening. Consequently, injection strategies in EGS could be improved through careful extrapolation of the results.