With the progress of autonomous vehicles (AVs), the vehicular digital twin has emerged as an effective solution for enhancing the safety and reliability of AVs. One fundamental challenge in vehicular digital twin networks lies in the effective data synchronization between physical vehicles (PVs) and digital twins (DTs) under resource constraints. To tackle this challenge, this paper proposes a novel synchronous paradigm incorporating sampling, communication, and prediction components to enhance synchronization performance under transmission cost constraints. Particularly, we employ a cutting-edge concept, age of information, as the performance evaluation criterion and optimize synchronization performance by formulating a resource-constrained average AoI minimization problem. We transform it into a constrained Markov decision process and introduce a constrained deep reinforcement learning-based solution, namely the sampling, communication, and prediction co-design algorithm. Simulation outcomes reveal that our approach delivers enhanced synchronization performance with lower transmission costs than the baseline. As far as we know, this is the first paper to establish a unified framework for the co-design of sampling, communication, and prediction in vehicular digital twin networks.