Multipath TCP (MPTCP) is regarded as a promising solution to aggregate the bandwidth of multiple paths to generate throughput benefits in heterogeneous networks. However, transmitting data over multiple paths simultaneously often leads to out-of-order issues due to the asymmetry of heterogeneous paths. In this paper, we propose a novel packet scheduling mechanism named Adaptive Delay-Aligned Scheduling (ADAS) for multipath transmission in heterogeneous wireless networks. ADAS utilizes the wisdom of the last-hop connected to the receiver to solve the out-of-order problem and improve the overall throughput at the same time. Specifically, ADAS equips a virtual link loop on the last-hop to buffer the out-of-order packets within time threshold and further schedules and sends them to the receiver as sequentially as possible, which looks as if all packets take the same time to travel across the network. In this way, the delay-aligned scheduling is achieved and the out-of-order problem can be effectively addressed. Besides, an adaptive weighting algorithm is proposed to dynamically adjust the time threshold to avoid over scheduling and improve the overall throughput. Extensive experiments demonstrate that ADAS outperforms state-of-the-art mechanisms. Besides, a lower out-of-order rate of 5.73% and a higher overall throughput of 6.76 Mbps can be achieved through combining ADAS with the current scheduling mechanisms.