With the designed six pairs of primers, all six overlapping S-gene fragments were amplified by nested RT-PCR from 29 RNA samples. However, most RNA samples initially included in the study, though positive for SARS-CoV with BNI primers, failed to simultaneously generate all six overlapping S-gene fragments and were excluded from further sequence analysis. Disintegration of the virus and low viral load in the raw samples likely accounted for these failures.

One hundred and thirteen sequence variations distributed in nine variant sites were identified in analyzed sequences that were compared to the reference BJ01 strain of SARS-CoV. BJ01 is an isolate from a tissue-culture propagated sample (16) and is used as reference strain in other studies (9,10). With the exception of one site (position 21702), other variant sites have not, to our knowledge, been documented in humans. Seven of nine variant sites were nonsynonymous. Figure 2 shows the identified variant sites compared to the reference sequence.

We identified novel variant sites and the coexistence of sequences with and without S-gene substitutions in SARS-CoV. Theoretically, a replicating RNA virus expresses a range of genetic and phenotypic variants and has the potential to generate novel virions, which may be selected in response to environmental pressures. RNA viruses generally tolerate high levels of mutagenesis because of their limited genetic complexity (17). Mutations have the potential to be pathogenic (e.g., giving the virus immunity to neutralizing antibodies, cytotoxic T cells, or antiviral drugs [18–20]). The dynamics of error copying and sequence decomposition are time-dependent. In HIV infection, for example, one adaptive substitution in the env gene occurred every 3.3 months or 25 viral generations, averaging across patients (21).

In our study, a higher variation frequency in the S gene was identified for SARS-CoV compared to previous reports (7–10). This difference may be due to a broader sample collection covering a longer timespan of infection. In addition, since virus isolates were not passaged in culture, the whole mutant repertoire is more likely to be detected, since no reverse mutation occurs in cell culture. Our observation most likely reflected the real situation in vivo. Variations were unlikely to result from Taq polymerase errors, since we repeated the experiments for all variations from preparing independent RNA and RT-PCR products and used Platinum Pfx DNA polymerase, which has a high fidelity, to confirm the results in some cases. We could not exclude the possibility that some variations were from defective genomes. However, the fact that the variations remained detectable in the sequences from two or three specimens of the same patient, obtained at different times, suggested that these variations might be active and extensible in vivo.

Sequences with and without substitutions (referred to BJ01) were simultaneously detected in the sequences from seven samples, which suggests the existence of SARS-CoV quasispecies. Furthermore, S-gene sequences from different samples collected at different times from the same patient showed similar, but not exactly identical, variation profiles in four participants (patients 4, 5, 6, and 19 in Table 1); this implies that a dynamic mutational process may exist in vivo. Table 2 summarizes the variations occurring in 29 analyzed S-gene sequences from 20 individual SARS patients.

One nonsynonymous change observed at position A1023G is within the heptad repeat (HR) domains, which is thought to be important for virus entry, and previous study on MHV showed that it would have some effect on virus infection (22). At this stage, we cannot rule out the possibility that this change affects the biological outcome of the virus, but further experiments need to be addressed in the near future.

We observed the coexistence of the S-gene sequences with and without substitutions and time-dependent variation profile in some patients. These observations suggest the possible existence of SARS-CoV quasispecies in an acute infection. In this study, however, the limitation of clinical sample collection and difficulty in directly amplifying full-length S gene from raw clinical samples restricted further extensive study for dynamic mutant distributions of the virus. In addition, the sequencing clone number was conditioned by the scale of the project, and this may have led to some minor variant sequences escaping analysis. Another factor possibly affecting the stability of the viral genome is the administration of the antiviral drug ribavirin. That ribavirin enhances mutagensis of RNA viruses has been addressed (23). Therefore, the artificial effect of ribavirin on the SARS-CoV mutant spectrum remains to be clarified.

The genetic variation of SARS-CoV remains limited in relation to many other RNA viruses such as HIV-1, HCV, and MHV. The probable reason is that SARS-CoV only causes an acute, self-limited infection, which may prevent persistent long-term mutant development in vivo as occurs in chronic RNA viral infections. Notably, some modules in the S protein remain conserved, e.g., the fusion-important HR domains. Although some variations may predict changes of protein functional features, no obvious correlation exists between mutation and clinical disease manifestation from the limited data reported here. Instead, the variation profile was closely correlated with epidemiography; e.g., patients 3–8 were infected in one hospital.

In conclusion, we report here some new variant sites in the S gene of coronavirus and possible existence of SARS-CoV quasispecies in some patients, though in limited numbers. This knowledge furthers our understanding of this emerging virus.