Research Article

The Genome Sequence of the SARS-Associated Coronavirus

See allHide authors and affiliations

Science  30 May 2003:
Vol. 300, Issue 5624, pp. 1399-1404
DOI: 10.1126/science.1085953


We sequenced the 29,751-base genome of the severe acute respiratory syndrome (SARS)–associated coronavirus known as the Tor2 isolate. The genome sequence reveals that this coronavirus is only moderately related to other known coronaviruses, including two human coronaviruses, HCoV-OC43 and HCoV-229E. Phylogenetic analysis of the predicted viral proteins indicates that the virus does not closely resemble any of the three previously known groups of coronaviruses. The genome sequence will aid in the diagnosis of SARS virus infection in humans and potential animal hosts (using polymerase chain reaction and immunological tests), in the development of antivirals (including neutralizing antibodies), and in the identification of putative epitopes for vaccine development.

An outbreak of atypical pneumonia, referred to as severe acute respiratory syndrome (SARS) and first identified in Guangdong Province, China, has spread to several countries. The severity of this disease is such that the mortality rate appears to be ∼3 to 6%, although a recent report suggests this rate can be as high as 43 to 55% in people older than 60 years (1). A number of laboratories worldwide have undertaken the identification of the causative agent (2, 3). The National Microbiology Laboratory in Canada obtained the Tor2 isolate from a patient in Toronto and succeeded in growing a coronavirus-like agent in African green monkey kidney (Vero E6) cells. This coronavirus was named publicly by the World Health Organization and member laboratories as the “SARS virus” (WHO press release, 16 April 2003) after tests of causation according to Koch's postulates, including monkey inoculation (4). This virus, which we refer to as SARS-HCoV, was purified, and its RNA genome was extracted and sent to the British Columbia Centre for Disease Control in Vancouver for genome sequencing by the BCCA Genome Sciences Centre.

The coronaviruses are members of a family of enveloped viruses that replicate in the cytoplasm of animal host cells (5). They are distinguished by the presence of a single-stranded plus-sense RNA genome about 30 kb in length that has a 5′ cap structure and 3′ polyadenylation tract. Upon infection of an appropriate host cell, the 5′-most open reading frame (ORF) of the viral genome is translated into a large polyprotein that is cleaved by viral-encoded proteases to release several nonstructural proteins, including an RNA-dependent RNA polymerase (Rep) and an adenosine triphosphatase (ATPase) helicase (Hel). These proteins, in turn, are responsible for replicating the viral genome as well as generating nested transcripts that are used in the synthesis of the viral proteins. The mechanism by which these subgenomic mRNAs are made is not fully understood. However, recent evidence indicates that transcription-regulating sequences (TRSs) at the 5′ end of each gene represent signals that regulate the discontinuous transcription of subgenomic mRNAs. The TRSs include a partially conserved core sequence (CS) that in some coronaviruses is 5′-CUAAAC-3′. Two major models have been proposed to explain the discontinuous transcription in coronaviruses and arterioviruses (6, 7). The discovery of transcriptionally active, subgenomic-size minus strands containing the antileader sequence and of transcription intermediates active in the synthesis of mRNAs (811) favors the model of discontinuous transcription during the minus strand synthesis (7).

The viral membrane proteins, including the major proteins S (Spike) and M (membrane), are inserted into the endoplasmic reticulum (ER) Golgi intermediate compartment while full-length replicated RNA plus strands assemble with the N (nucleocapsid) protein. This RNA-protein complex then associates with the M protein embedded in the membranes of the ER, and virus particles form as the nucleocapsid complex buds into the lumen of the ER. The virus then migrates through the Golgi complex and eventually exits the cell, likely by exocytosis (5). The site of viral attachment to the host cell resides within the S protein.

The coronaviruses include a large number of viruses that infect different animal species. The predominant diseases associated with these viruses are respiratory and enteric infections, although hepatic and neurological diseases also occur. Human coronaviruses identified in the 1960s (including the prototype viruses HCoV-OC43 and HCoV-229E) are responsible for up to 30% of respiratory infections (12). Coronaviruses are divided into three serotypes: groups 1, 2, and 3 (13). Phylogenetic analysis of coronavirus sequences also identifies three main classes of these viruses, corresponding to each of the three serotypes. Group 2 coronaviruses contain a gene encoding hemagglutinin esterase (HE) that is homologous to that of influenza C virus. It is presumed that the precursor of the group 2 coronaviruses acquired HE as a result of a recombination event within a doubly infected host cell. We note that the Tor2 genome sequence appears to lack an HE gene.

Purification of viral particles and RNA, and DNA sequencing. Virus isolation was performed on a bronchoalveolar lavage specimen of a fatal SARS case belonging to the original case cluster from Toronto, Canada. Viral particles from this Tor2 isolate were purified, and the genetic material (RNA) was extracted (14) from the Tor2 isolate (15). The RNA was converted to cDNA by means of a combined random-priming and oligo(dT) priming strategy (14). Size-selected cDNA products were cloned, and single sequence reads were generated from each end of the insert from randomly chosen clones. Sequences were assembled and the assembly was edited to produce a draft sequence of the viral genome on 12 April 2003 (14). Rapid amplification of cDNA ends [RACE (14)] was performed to capture the 5′ end of the viral genome. The SARS genomic sequence has been deposited into GenBank (accession number AY274119.3). The final sequence we produced (also available as Release 3; is essentially identical to that released independently by the U.S. Centers for Disease Control (CDC) (16). We report additional bases in the Tor2 sequence that correspond to the 3′ (encoded) polyadenylation tail. Eight base differences between the two sequences could represent sequencing errors, polymerase chain reaction (PCR) artifacts, or mutable sites in the genome. The differences we detect between our sequence and that of the CDC are summarized in Table 1.

Table 1.

Nucleotide base differences between the Tor2 sequence and the Urbani sequence [(16),]. Boldface indicates a base difference resulting in an amino acid change (32); X indicates a nonconservative amino acid substitution.

PositionView inline Tor2 Urbani Frame Protein
Base Amino acid Base Amino acid
7,919 C A T V 1 Replicase 1A
16,622 C A T A 3 Replicase 1B
19,064 A E G E 3 Replicase 1B
19,183 T V C A 3 Replicase 1B
23,220 G A T S X 3 S (Spike) glycoprotein
24,872 T L C L 3 S (Spike) glycoprotein
25,298 A R G G X 2 ORF 3
26,857 T S C P X 1 M protein
  • View inline* GenBank AY274119.3.

  • Non–protein-coding features of the Tor2 SARS-CoV genome sequence. At the 5′ end of the genome, we detected a putative 5′ leader sequence with similarity to the conserved coronavirus core leader sequence, 5′-CUAAAC-3′ (6, 7). Putative TRS sequences were determined through manual alignment of sequences upstream of potential initiating methionine codons (see below) with the region of the coronavirus genome sequence containing the leader sequence (Table 2). Candidate TRS sequences were scored as strong, weak, or absent on the basis of inspection of the alignments.

    Table 2.

    Nucleotide position, associated ORF, and putative transcription regulatory sequences (see text for details). Numbers in parentheses within the alignment indicate distance to the putative initiating codon. The conserved core sequence is indicated in boldface in the putative leader sequence. Contiguous sequences identical to region of the leader sequence containing the core sequence are underlined. No putative TRSs were detected for ORFs 4, 13, and 14, although ORF 13 could share the TRS associated with the N protein.

    Base ORF TRS sequence
    21,479 S (Spike) CAACUAAACGAACAUG
    26,104 Envelope UGAGUACGAACUUAUG
    27,001 ORF 7 AACUAUAAAUU (62) AUG

    The 3′ untranslated region (3′UTR) sequence contains a 32–base pair region corresponding to the conserved s2m motif (17). The s2m motif is believed to be a universal feature of astroviruses that has also been identified in avian infectious bronchitis virus (avian IBV) and the ERV-2 equine rhinovirus. The high degree of conservation between the s2m motifs in these different viruses and their evolutionary distance suggests that the avian IBV and ERV-2 have acquired the s2m motif through separate horizontal RNA transfer events (17). The inferred distance of the SARS coronavirus to IBV from our phylogenetic analysis (Fig. 1) would also suggest that the SARS coronavirus has obtained its s2m motif through a horizontal transfer event.

    Fig. 1.

    Phylogenetic analysis of SARS proteins. Unrooted phylogenetic trees were generated by clustalw 1.74 (33) using the BLOSUM comparison matrix and a bootstrap analysis of 1000 iterations. Numbers indicate bootstrap replicates supporting each node. Phylogenetic trees were drawn with the Phylip Drawtree program 3.6a3 (34). Branch lengths indicate the number of substitutions per residue. GenBank accession numbers for protein sequences are as follows: (A) Replicase 1A: BCoV (bovine coronavirus), AAL40396; HCoV-229E (human coronavirus), NP_073555; MHV (mouse hepatitis virus), NP_045298; IBV (avian infectious bronchitis virus), CAC39113; TGEV (porcine transmissible gastroenteritis virus), NP_058423. (B) Membrane glycoprotein: PHEV (porcine hemagglutinating encephalomyelitis virus), AAL80035; BCoV, NP_150082; IBV 1, AAF35863; IBV 2, AAK83027; MHV, AAF36439; TGEV, NP_058427; HCoV-OC43, AAA45462; FCoV (feline coronavirus), BAC01160. (C) Nucleocapsid: MHV, P18446; BCoV, NP_150083; IBV 1, AAK27162; IBV 2, NP_040838; FCoV, CAA74230; PTGV (porcine transmissible gastroenteritis virus), AAM97563; HCoV-229E, NP_073556; HCoV-OC43, P33469; PHEV, AAL80036; TCV (turkey coronavirus), AAF23873. (D) S (Spike) protein: BCoV, AAL40400; MHV, P11225; HCoV-OC43, S44241; HCoV-229E, AAK32191; PHEV, AAL80031; PRCoV (porcine respiratory coronavirus), AAA46905; PEDV (porcine epidemic diarrhea virus), CAA80971; CCoV (canine coronavirus), S41453; FIPV (feline infectious peritonitis virus), BAA06805; IBV, AAO34396.

    Predicted protein coding features of the Tor2 SARS-CoV genome sequence. ORFs were determined initially through sequence similarity to known coronavirus proteins. This approach identified replicases 1a and 1b, the S protein, the small envelope (E) protein, the M protein, and the N protein. ORFs that did not match database sequences were identified if they were larger than 40 amino acids, unless a strong match to the TRS consensus was found close to and upstream of the potential initiating methionine residue. We note that Rota et al. (16) did not identify potential proteins of less than 50 amino acids. We attempted to identify putative TRSs upstream of all ORFs, both known and predicted (Tables 2 and 3). However, TRSs are not required for transcription of all coronavirus genes, because internal initiation from larger RNA transcripts is also able to facilitate translation (18, 19). Certain ORFs overlap (ORFs 10 and 11, by 12 amino acids; Fig. 2), and some are contained entirely within another ORF or ORFs (ORF 4 and ORFs 13 and 14; Fig. 2). The biological relevance of these ORF predictions remains to be established, but in the cases of ORFs 10 and 11, we detect strong matches to the TRS consensus in close proximity to their respective initiating methionine codons (Table 2). Construction of unrooted phylogenetic trees using the set of known proteins and representatives of the three known coronaviral groups reveals that the proteins encoded by the SARS virus do not readily cluster more closely with any one group (Fig. 1). Hence, we propose that this isolate be considered the first representative of “group 4” coronaviruses.

    Fig. 2.

    Map of the predicted ORFs and s2m motif in the Tor2 SARS virus genome sequence.

    Table 3.

    Features of the Tor2 genome sequence.

    Feature Start—EndView inline No. of amino acids No. of bases Frame Candidate TRSView inline Rota et al. ORFView inline
    ORF 1a 265-13,398 4,382 13,149 +1 N/AView inline 1a
    ORF 1b 13,398-21,485 2,628 7,887 +3 N/A 1b
    S protein 21,492-25,259 1,255 3,768 +3 Strong S
    ORF 3 25,268-26,092 274 825 +2 Strong X1
    ORF 4 25,689-26,153 154 465 +3 AbsentView inline X2
    E protein 26,117-26,347 76 231 +2 Weak E
    M protein 26,398-27,063 221 666 +1 Strong M
    ORF 7 27,074-27,265 63 192 +2 Weak X3
    ORF 8 27,273-27,641 122 369 +3 Strong X4
    ORF 9 27,638-27,772 44 135 +2 Weak N/R
    ORF 10 27,779-27,898 39 120 +2 Strong N/R
    ORF 11 27,864-28,118 84 255 +3 Weak X5
    N protein 28,120-29,388 422 1,269 +1 Strong N
    ORF 13 28,130-28,426 98 297 +2 AbsentView inline N/R
    ORF 14 28,583-28,795 70 213 +2 AbsentView inline N/R
    s2m motif 29,590-29,621 N/A 30 N/A N/A N/R
  • View inline* End coordinates include the stop codon, except for ORF 1a and s2m. The right coordinate of ORF 1a is the end position of the ribosome slippage site (UUUAAC). The most likely frameshift site, based on alignment with other replicase proteins, is 13,392.

  • View inline See text for details.

  • View inline Corresponding ORFs from Rota et al. (16). N/R indicates the feature was not reported.

  • View inline§ These ORFs overlap substantially or completely with others and may share TRSs.

  • View inline N/A, not applicable.

  • The coding potential of the 29,751-base genome is depicted in Fig. 2. Recognizable ORFs include the replicase 1a and 1b translation products, the S glycoprotein, the E protein, the M protein, and the N protein. We have, in addition, conducted a preliminary analysis of the nine novel ORFs in an attempt to ascribe to them a possible functional role. These analyses are summarized below.

    The replicase 1a ORF (base pairs 265 to 13,398) and replicase 1b ORF (base pairs 13,398 to 21,485) occupy 21.2 kb of the SARS virus genome (Fig. 2). Conserved in both length and amino acid sequence to other coronavirus replicase proteins, the genes encode a number of proteins that are produced by proteolytic cleavage of a large polyprotein (20). As seen in other coronaviruses and as anticipated, a frame shift interrupts the protein-coding region and separates the 1a and 1b reading frames.

    The Spike (S) glycoprotein (Fig. 2; base pairs 21,492 to 25,259) encodes a surface projection glycoprotein precursor predicted to be 1255 amino acids in length. Mutations in the gene encoding the Spike protein have previously been correlated with altered pathogenesis and virulence in other coronaviruses (5). In some coronaviruses, the mature Spike protein is inserted in the viral envelope, with most of the protein exposed on the surface of the viral particles. It is believed that three molecules of the Spike protein form the characteristic peplomers or corona-like structures of this virus family. Our analysis of the Spike glycoprotein with SignalP (21) reveals a high probability of a signal peptide (probability 0.996) with cleavage between residues 13 and 14. TMHMM (22) reveals a strong transmembrane domain near the C-terminal end. Together these data predict a type I membrane protein with the N terminus and the majority of the protein (residues 14 to 1195) on the outside of the cell surface or virus particle, in agreement with other coronavirus Spike protein data. Supporting this conclusion, it has recently been shown that for HCoV-229E virions, residues 417 to 546 are required for binding to the cellular receptor, aminopeptidase N (23). However, it is known that various coronaviruses use different receptors, and hence it is likely that different receptor binding sites are also used.

    ORF 3 (Fig. 2; base pairs 25,268 to 26,092) encodes a predicted protein of 274 amino acids that lacks significant BLAST (24), FASTA (25), or PFAM (26) similarities to any known protein. Analysis of the N-terminal 70 amino acids with SignalP provides weak evidence for the existence of a signal peptide and a cleavage site (probability 0.540). Both TMpred (27) and TMHMM predict the existence of three transmembrane regions spanning approximately residues 34 to 56, 77 to 99, and 103 to 125. The most likely model from these analyses is that the C terminus and a large 149–amino acid N-terminal domain would be located inside the viral or cellular membrane. The C-terminal (interior) region of the protein may encode a protein domain with ATP-binding properties (ProDom ID PD037277).

    ORF 4 (Fig. 2; base pairs 25,689 to 26,153) encodes a predicted protein of 154 amino acids. This ORF overlaps entirely with ORF 3 and the E protein. Our analysis failed to locate a potential TRS sequence at the 5′ end of this putative ORF. However, it is possible that this protein is expressed from the ORF 3 mRNA using an internal ribosomal entry site. BLAST analyses fail to identify matching sequences. Analysis with TMpred weakly predicts a single transmembrane helix.

    The gene encoding the small envelope (E) protein (Fig. 2; base pairs 26,117 to 26,347) yields a predicted protein of 76 amino acids. BLAST and FASTA comparisons indicate that the predicted protein exhibits significant matches to multiple envelope (alternatively known as small membrane) proteins from several coronaviruses. PFAM analysis of the protein reveals that the predicted protein is a member of the well-characterized NS3_EnvE protein family (26). InterProScan (28, 29) analysis reveals that the protein is a component of the viral envelope, and conserved sequences are also found in other viruses, including gastroenteritis virus and murine hepatitis virus. SignalP analysis predicts the presence of a transmembrane anchor (probability 0.939). TMpred analysis of the predicted protein reveals a similar transmembrane domain at positions 17 to 34, consistent with the known association of this protein with the viral envelope. TMHMM predicts a type II membrane protein with most of the hydrophilic domain (46 residues) and the C terminus located on the surface of the viral particle. In some coronaviruses such as porcine transmissible gastroenteritis virus (TGEV), the E protein is essential for virus replication (30). In contrast, in mouse hepatitis virus (MHV), although deletion of the gene encoding the E protein reduces virus replication by more than four orders of magnitude, the virus still can replicate (31).

    The gene encoding the membrane (M) glycoprotein (Fig. 2; base pairs 26,398 to 27,063) yields a predicted protein of 221 amino acids. BLAST and FASTA analyses of the protein reveal significant matches to a large number of coronaviral matrix glycoproteins. The association of the Spike glycoprotein (S) with the matrix glycoprotein (M) is an essential step in the formation of the viral envelope and in the accumulation of both proteins at the site of virus assembly (5). Analysis of the amino acid sequence with SignalP predicts a signal sequence (probability 0.932) that is not likely cleaved. TMHMM and TMpred analyses indicate the presence of three transmembrane helices, located at approximately residues 15 to 37, 50 to 72, and 77 to 99, with the 121–amino acid hydrophilic domain on the inside of the virus particle, where it is believed to interact with the nucleocapsid. PFAM analysis reveals a match to PFAM domain PF01635 and alignments to 85 other sequences in the PFAM database bearing this domain, which is indicative of the coronavirus matrix glycoprotein.

    ORF 7 (Fig. 2; base pairs 27,074 to 27,265) encodes a predicted protein of 63 amino acids. BLAST and FASTA searches yield no significant matches indicative of function. TMHMM and SignalP predict no transmembrane region; however, TMpred analysis predicts a likely transmembrane helix located between residues 3 and 22, with the N terminus located outside the viral particle. Similarly, ORF 8 (Fig. 2; base pairs 27,273 to 27,641), encoding a predicted protein of 122 amino acids, has no significant BLAST or FASTA matches to known proteins. Analysis of this sequence with SignalP indicates a cleaved signal sequence (probability 0.995) with the predicted cleavage site located between residues 15 and 16. TMpred and TMHMM analyses also predict a transmembrane helix located approximately at residues 99 to 117. Together these data indicate that ORF 8 is likely to be a type I membrane protein, with the major hydrophilic domain of the protein (residues 16 to 98) and the N terminus oriented inside the lumen of the ER/Golgi or on the surface of the cell membrane or virus particle, depending on the membrane localization of the protein.

    ORF 9 (Fig. 2; base pairs 27,638 to 27,772) encodes a predicted protein of 44 amino acids. FASTA analysis of this sequence reveals some weak similarities (37% identity over a 35–amino acid overlap) to Swiss-Prot accession Q9M883, annotated as a putative sterol-C5 desaturase. A similarly weak match to a hypothetical Clostridium perfringens protein (Swiss-Prot accession CPE2366) is also detected. The functional implications, if any, of these matches are unknown. TMpred predicts the existence of a single strong transmembrane helix, with little preference for alternate models in which the N terminus is located inside or outside the particle. Similarly, ORF 10 (Fig. 2; base pairs 27,779 to 27,898), encoding a predicted protein of 39 amino acids, exhibits no significant matches in BLAST and FASTA searches but is predicted to encode a transmembrane helix by TMpred, with the N terminus located within the viral particle. The region immediately upstream of ORF 10 exhibits a strong match to the TRS consensus (Table 2), providing support for the notion that a transcript initiates from this site. ORF 11 (Fig. 2; base pairs 27,864 to 28,118), encoding a predicted protein of 84 amino acids, exhibits only very short (9 or 10 residues) matches to a region of the human coronavirus S glycoprotein precursor (starting at residue 801). Analyses by SignalP and TMHMM predict a soluble protein. As was the case for ORF 10, a detectable alignment to the TRS consensus sequence was found (Table 2).

    The gene encoding the nucleocapsid protein (Fig. 2; base pairs 28,120 to 29,388) yields a predicted protein of 422 amino acids. This protein aligns well with nucleocapsid proteins from other representative coronaviruses, although a short lysine-rich region (KTFPPTEPKKDKKKKTDEAQ) (32) appears to be unique to SARS. This region is suggestive of a nuclear localization signal, and although it contains a hit to InterProDomain IPR001472 (bipartite nuclear localization signal), the function of this insertion remains unknown. It is possible that the SARS virus nucleocapsid protein has a novel nuclear function, which could play a role in pathogenesis. In addition, the basic nature of this peptide suggests that it may assist in RNA binding.

    ORF 13 (Fig. 2; base pairs 28,130 to 28,426) encodes a predicted protein of 98 amino acids. BLAST analysis fails to identify similar sequences, and no transmembrane helices are predicted. ORF 14 (Fig. 2; base pairs 28,583 to 28,795) encodes a predicted protein of 70 amino acids. BLAST analysis fails to identify similar sequences. TMpred weakly predicts a single transmembrane helix.

    Conclusions. We used genome sequencing to determine that the virus named by the WHO as causally associated with SARS is a novel coronavirus. This has been confirmed by the sequence of two independent isolates: the Tor2 isolate, reported here, and the Urbani isolate, reported by the CDC (16). Although morphologically a coronavirus (3), this SARS virus is not more closely related to any of the three known classes of coronavirus, and we propose that it defines a fourth class of coronavirus (group 4) and that it be referred to as SARS-CoV. Our sequence data do not support a recent interviral recombination event between the known coronavirus groups as the origin of this virus, but this may be due to the limited number of known coronavirus genome sequences. Apart from the s2m motif located in the 3′UTR, there is also no evidence of any exchange of genetic material between the SARS virus and non-Coronaviridae. These data are consistent with the hypothesis that an animal virus for which the normal host is currently unknown recently mutated and developed the ability to productively infect humans. There also remains the possibility that the SARS virus evolved from a previously harmless human coronavirus. However, preliminary evidence suggests that antibodies to this virus are absent in people not infected with SARS-CoV (3), which implies that a benign virus closely related to the Tor2 isolate is not resident in humans. Identification of the normal host of this coronavirus and comparison of the sequences of the ancestral and SARS forms will further elucidate the process by which this virus arose.

    The availability of the SARS virus genome sequence is important from a public health perspective. It will allow the rapid development of PCR-based assays for this virus that capitalize on novel sequence features, enabling discrimination between this and other circulating coronaviruses. Such assays will allow the diagnosis of SARS virus infection in humans and, critically, will consolidate the association of this virus with SARS. If the association is further borne out, SARS virus genome–based PCR assays may form an important part of a public health strategy to control the spread of this syndrome. In the longer term, this information will assist in the development of antiviral treatments, including neutralizing antibodies and development of a vaccine to treat this emerging and deadly disease.

    Supporting Online Material

    Materials and Methods


    References and Notes

    View Abstract

    Stay Connected to Science

    Navigate This Article