SARS-CoV-2 variant analysis

A total of 7,948 high coverage sequences from human isolates were downloaded from GISAID on 04/20/2020, representing 6,797 unique DNA sequences. These were annotated as ‘sequences with < 1% Ns and < 0.05% unique amino acid mutations (not seen in other sequences in the database) and no insertion/deletion unless verified by the submitter’. The unique sequences were aligned to the reference genome with MAFFT experimental version 7.463 using options ‘–auto –addfragments’^24,25. The misaligned 3.2 kb fragment EPI_ISL_426413 was identified as Hepatitis B virus isolates JRC-HB01 by MEGABLAST against the NCBI nr database and was removed from the alignment. Further, a total of 40 inserts in the alignment that introduced gaps in the reference sequence were deleted as likely sequencing errors. The 11 largest inserts ranged from 99 to 155 bases, and the next largest were 3 base long. The genome submitter confirmed that those large inserts were likely assembly artefacts that will be corrected. Of the smaller inserts, a three base ‘TTT’ in-frame insertion was observed at nsp6 position 298 in 17 sequences, and is presumably real, although it is in a repetitive region of 8 consecutive Ts and therefore at the limit of accurate PCR amplification of mono-thymidine repeats²⁶. Gaps in aligned sequences were likewise treated as likely sequencing errors and replaced with the unknown base to keep translation in frame with respect to the reference sequence, taking into account the ribosomal frameshift in the nsp12 coding sequence. This process resulted in just 95 remaining premature terminations in over 160 K proteins; 43 of these were found consistently after amino acid 125 of nsp10 and likely represent true truncations. The others were distributed across the sequences of 11 protein families and are likely erroneous. The 27 protein alignments thus obtained, one for each of the SARS-CoV-2 proteins selected for this study, were transformed into a n by m matrix, where n is the length of the reference protein and m the length of the alignment, and processed column-wise. At each position in the alignment, the variant count was obtained by counting unique amino acids that differed from the reference; variability was quantified using the Shannon entropy, calculated as SE=-∑i=1nPi×log2Pi where n is the number of amino-acid types and P_i is the fraction of amino acid type i at that position^27,28. Codes for unknown amino acid (X) and translation termination (*) were ignored in all calculations.