In this work, the fluid-solid heat transfer is studied by means of particle-resolved direct numerical simulations performed by using a Lagrangian VOF approach based on fictitious domain and penalty methods. The first non-isothermal test case studied in detail consists in a cold fluid flow across an array of hot particles at a given temperature. Particles are fixed and randomly arranged, and their volume fraction spans from 0.1 to 0.4 in order to cover the typical values encountered in fluid-particle dense flows. Three Reynolds numbers, with unity Prandtl number, are investigated. From the fully resolved fluid velocity and temperature fields, macroscopic quantities are obtained by means of spatial averages over the entire domain or over planes normal to the streamwise direction. These quantities are then used to investigate the gas-solid heat transfer at macroscopic scale with the goal to support two-fluid model development. At such macroscopic scale, the fluid energy equation makes appear a velocity-temperature correlation term which accounts for the microscopic heterogeneous distributions of both the fluid velocity and temperature due to the momentum and heat transfer with the solid particles. In this study, such a contribution is written in terms of a bulk-temperature tensor. In the presence of a unidirectional mean flow, this tensor reduces to the unique component representing the well-known bulk temperature, as classically defined. A connection between this contribution and the ratio of two Nusselt numbers based, respectively, on the fluid temperature and on the bulk temperature, is pointed out. These Nusselt numbers are computed and compared to the correlations available from the literature. Their ratio is also computed and compared to the literature. On the basis of such a ratio, a model for the fluctuating velocity-temperature contribution is proposed. In a second stage, particle-resolved numerical simulations were carried out for several configurations of non-isothermal liquid-solid fluidized beds with a limited number of particles (280, 520 and 1280) at a given temperature with the fluid flow injected at a uniform and constant cold temperature. Then, in this work, simulations results are mainly analyzed in terms of macroscopic variables. In particular, the measured fluid-particle heat transfer coefficient and fluid velocity- temperature spatial correlations are compared with the mathematical modeling approach derived for the fixed bed.