This letter investigates the popularity-aware content caching problem in resource-constrained Cellular-Vehicle-to-Everything (C-V2X) scenarios. A federated Deep Q Network (DQN)-based content caching scheme for vehicle cluster is presented to maximize the fee of caching vehicles. We first cluster vehicles based on their mobility and preferences, and combine historical request frequency with preferences to derive the popularity of the content. Moreover, multiple cluster heads within the coverage area of the Central Server (CS) use DQN to train local caching models, and the CS aggregates local model parameters to obtain a global popular content caching model. Simulation results verify that the proposed scheme outperforms other benchmarks in terms of the cache hit rate and content delivery rate.