Transcribing electric guitar solo in real-world performance is challenging because of the interference of background accompaniments, the strong coupling between music pitch and playing technique, and the limited resource of data annotation. To address these issues, we first propose a new guitar solo dataset for this task. Then, we propose a transcription model which learns an output space jointly constructed with notes, playing techniques, and two sets of meta-class labels named note states and technique groups, such that the model can harness the layered relationship among different note-level event and playing technique classes. The proposed model outperforms the state-of-the-art guitar solo transcription and note transcription models.