Multivariate Analyses of Codon Usage of SARS-CoV-2 and other betacoronaviruses

Haogao Gu; Daniel Chu; Malik Peiris; Leo L.M. Poon

A novel coronavirus outbreak took place in Wuhan, Hubei province, China in December 31 2019 1 . This novel coronavirus (SARS-CoV-2) causes pneumonia in patients 2 and it has 32 rapidly spread to other provinces in China and other countries 3 . This novel coronavirus 33 outbreak had raised global concern but current knowledge on the origin and transmission 34 route of the pathogen is still limited. The SARS-CoV-2 belongs to the genus Betacoronavirus, 35 which also includes two highly virulent human coronaviruses, SARS-CoV and MERS-CoV. 36 Apart from human, many animal species, such as bat, rat, camel, swine and hedgehog, can be 37 infected by different types of coronaviruses. Further sequence analyses of this novel and 38 other betacoronaviruses might provide additional information to better understand the 39 evolution of SARS-CoV-2. 40 Preferential codon usage is commonly seen in different organisms, and it has been evident 41 that the uneven codon usage is not neutral but related to gene expression or other selection 42 pressures 4-6 . There are two levels of codon usage biases, one is at amino acid level and the 43 other is at synonymous codon level. The amino acid composition of proteins can be an 44 important factor that explaining certain sequence traits. For example integral membrane 45 proteins that are enriched in hydrophobic amino acids can create significant codon usage 46 bias 7 . Amino acid composition sometime can also introduce confounding effects when one 47 only focuses on studying the variations of synonymous codon usage. The use of global 48 correspondence analysis (CA) and its derivatives within-group correspondence analysis 49 (WCA) and between-group correspondence analysis (BCA) to analyze codon usages can 50 overcome the above problem. In fact, WCA becomes "model of choice" for analyzing 51 synonymous codon usage in recent years, as it is more robust than other traditional methods 52 (e.g. CA with relative codon frequency or CA with RSCU values) 7,8 . This analytic approach, 53 however, has not been used in studying viral sequences. As the natural history of the SARS- 54 CoV-2 remains largely unknown, an in-depth codon usage analysis of this newly emerging 55 virus might provide some novel insights. 56 In this study, we used both CA and WCA to analyses codon usage patterns of a vast number 57 of betacoronavirus sequences. We found SARS-CoV-2 and bat SARSr-CoV have similar 58 amino acid usage. However, our analyses suggested that the spike and member genes of 59 SARS-CoV-2 have rather distinct synonymous codon usage patterns. Table S1 . 86 The codon count for every gene sequence input for the correspondence analysis was All the correspondence analyses in this study were performed individually for each gene, to 103 achieve better resolution on gene specific codon usage pattern. Gravy score provides an easy way to estimate the hydropathy character of a protein 15 . It was 117 used in this study as a proxy to identify proteins that are likely to be membrane-bound 118 proteins. The GRAVY score was calculated in a linear form on codon frequencies as: The overall amino acid and codon usage of the dataset are plotted in an ascending order 140 ( Figure 1 ). We observed that leucine and valine were the two most frequently used amino 141 acids in the four studied genes, while tryptophan, histidine and methionine were the three 142 least used ones. We also found that codons ending with cytosine or guanine were generally 143 le