ABSTRACT
SARS-CoV-2, a positive single-stranded RNA virus, caused the COVID-19 pandemic. Although its sense-mRNA architecture was reported, its anti-sense strand was unexplored. Here, we deeply sequenced both strands of RNA and found SARS-CoV-2 transcription is strongly biased to form the sense strand. During negative strand synthesis, apart from canonical sub-genomic ORFs, numerous non-canonical fusion transcripts are formed, driven by 3-15 nt sequence homology scattered along the genome but more prone to be inhibited by SARS-CoV-2 RNA polymerase inhibitor Remdesivir. The drug also represses more of the negative than the positive strand synthesis as supported by a mathematic simulation model and experimental quantifications. Overall, this study opens new sights into SARS-CoV-2 biogenesis and may facilitate the anti-viral drug design.