Despite its overwhelming clinical importance for understanding and mitigating the COVID-19 pandemic MESHD COVID-19 pandemic MESHD
, the protein-coding gene content of the SARS-CoV-2 genome remains unresolved, with the function and even protein-coding status of many hypothetical proteins unknown and often conflicting among different annotations, thus hindering efforts for systematic dissection of its biology and the impact of recent mutations. Comparative genomics is a powerful approach for distinguishing protein-coding versus non-coding functional elements, based on their characteristic patterns of change, which we previously used to annotate protein-coding genes in human, fly, and other species. Here, we use comparative genomics to provide a high-confidence set of SARS-CoV-2 protein-coding genes, to characterize their protein-level and nucleotide-level evolutionary constraint, and to interpret the functional implications for SARS-CoV-2 mutations acquired during the current pandemic. We select 44 complete Sarbecovirus genomes at evolutionary distances well-suited for protein-coding and non-coding element identification, create whole-genome alignments spanning all named and putative genes, and quantify their protein-coding evolutionary signatures using PhyloCSF and their overlapping constraint using FRESCo. We find strong protein-coding signatures for all named genes and for hypothetical ORFs 3a, 6, 7a, 7b, and 8, indicating protein-coding roles, and provide strong evidence of protein-coding status for a recently-proposed alternate-frame novel ORF within 3a. By contrast, ORF10 PROTEIN
shows no protein-coding signatures but shows unusually-high nucleotide-level constraint, indicating it has important but non-coding functions, and ORF14 and SARS-CoV-1 ORF3b PROTEIN
, which overlap other genes, lack evolutionary signatures expected for dual-coding regions, indicating they do not produce functional proteins. ORF9b PROTEIN
has ambiguous protein-coding signatures, preventing us from resolving its protein-coding status. ORF8 PROTEIN
shows extremely fast nucleotide-level evolution, lacks a known function, and was deactivated in SARS-CoV-1, but shows clear signatures indicating protein-coding function worthy of further investigation given its rapid evolution and potential role in replication. SARS-CoV-2 mutations are preferentially excluded from evolutionarily-constrained amino acid residues and synonymously-constrained nucleotides, indicating purifying constraint acting at both coding and non-coding levels. In contrast, we find a conserved region in the nucleocapsid that is enriched for recent mutations, which could indicate a selective signal, and find that several spike-protein PROTEIN
mutations previously identified as candidates for increased transmission and several mutations in isolates found to generate higher viral load in-vitro disrupt otherwise-perfectly-conserved amino-acids, consistent with adaptations for human-to-human transmission.