Serial nasopharyngeal, fecal, and serum specimens of patients were collected at hospitals by Guangzhou Municipal Centers for Diseases Control and Prevention. When possible SARS was diagnosed in the waitress on January 2, 2004, serum, throat and rectal swabs were obtained from all 6 palm civets at the restaurant. It was reported that the animals were purchased from Xinyuan live animal wholesale market in Guangzhou. Serum samples from employees of the restaurant were obtained on January 4. Persons with positive results provided additional samples as needed. All specimens were stored at –80°C.

Serum samples were tested by enzyme-linked immunosorbent assay (ELISA), immunofluorescent antibody (IFA) test, and Western blot for specific immunoglobulin G (IgG) and IgM. Nasopharyngeal, throat, and rectal specimens were tested by reverse transcription–polymerase chain reaction for polyprotein (P) and nucleocapsid (N) genes of SARS-CoV. Gene sequences were determined directly from original samples. RNA was transcribed into cDNA (SuperScript, Invitrogen, Carlsbad, CA, USA) and subsequently used for PCR amplification. Complete spike (S) gene and whole genome sequencing of SARS-CoV virus was conducted by using 48 primer sets based on the sequence data of a SARS-CoV SZ3 isolate from palm civet (6) and an ABI 3730 Genetic Analyzer (Applied Biosystems, Foster City, CA, USA). Assembled genome sequences were compared with those of the first virus isolates of human (TOR2) and animal (SZ3) origin. Any nucleotide (nt) differences were double-checked and confirmed. Sequences from this study were deposited in GenBank (accession nos. AY572034–572038).

Samples from patients and animals were cultured in fetal rhesus kidney (FRhK-4) cells or Vero E6 cells for virus isolation as described (6,7). Cells with or without SARS-CoV virus infection were harvested and fixed in 2.5% glutaraldehyde (Electron Microscopy Sciences, Hatfield, PA, USA) for 4 h and post-fixed in 1% osmium tetroxide for 1 h. Cells were then transferred to a 1.5-mL tube and centrifuged at 1,000 rpm for 10 min. The supernatant was removed and a 2% agarose solution (55°C–60°C) was added to the cell pellet. After the agarose solidified, ≈1-mm cubes containing the cell pellet were cut and dehydrated in graded ethanol. The cubes were then embedded in epoxy resin. Ultrathin sections (70 nm) were prepared and stained with uranyl acetate and lead citrate. Sections were examined with a Philips (Eindhoven, the Netherlands) EM208S electron microscope.

Nucleotide and amino acid sequences were aligned by using MegAlign version 6.0 (DNASTAR, Madison, WI, USA). A neighbor-joining tree with bootstrap values was constructed to estimate phylogenetic relationships among sequences. Nucleotide positions were numbered based on the TOR2 SARS virus isolate (GenBank accession no. NC_004718) (8).

Epidemiologic investigations showed that 2 of the 4 recent SARS patients were linked with the restaurant serving palm civets as food. One patient was a 20-year-old waitress who became ill on December 26, 2003, with suspected SARS was diagnosed on January 2, 2004, and she was classified as a probable SARS patient by local health authorities on January 8, 2004 (4,5). She denied eating palm civet or being in close contact with them. However, palm civets were found in her work area and she often passed or stood a short distance from the animal cages. The other patient was a 40-year-old physician who ate at the restaurant on December 31, 2003, and first showed symptoms on January 7, 2004. His dining table was within 5 m of civet cages.

Serial serum samples from both patients were positive for IgG and IgM against SARS-CoV by ELISA, IFA, and Western blots. Positions 22907–23192 (286 bp) of the S gene were sequenced from a nasopharyngeal swab isolate from the waitress and from a fecal specimen from the physician (samples were obtained on January 5 and January 12, 2004, respectively). The 2 S gene sequence fragments were identical, but differed from all S gene sequences available on public databases. Attempts to isolate virus from these specimens by using Vero E6 cells were unsuccessful. Isolation of virus with the FRhK-4 cell line was not attempted because the volume of specimen from patients was limited.

The restaurant is in a 2-story building in downtown Guangzhou. Eight animal cages containing 6 palm civets (Paguma larvata) were stacked (2 cages per stack) at the front door of the restaurant. The cages were approximately 1 m from the sidewalk and 2 m from the first row of dining tables on the ground floor of the restaurant. Pedestrians walking in the street and customers dining on the ground floor could easily see the animals in the cages.

Both P and N genes of SARS-CoV were found by nested PCR in all throat and rectal swab specimens from 6 palm civets (Table 1) (9). Three complete genome sequences and 2 complete S gene sequences of SARS-CoV were found in rectal or throat swab specimens from 5 of the 6 palm civets (Table 1) (10). The 286-bp S gene sequences from isolates from the waitress and the physician were identical to 4 of 5 S gene sequences from palm civets from the restaurant, but differed from other sequences available from public databases (Table 2). SARS-CoV virus was isolated from FRhK-4 cells cultured with a rectal swab specimen of a palm civet, but not from Vero E6 cells. Cytopathic effects (CPE) of SARS-CoV virus on FRhK-4 cells were visible 4 days after culture with a fecal swab sample. Electron microscopy showed typical morphologic features of SARS-CoV virus in a thin section of the infected cell. A complete genome sequence of the SARS-CoV isolated from a palm civet was determined directly from the original sample and submitted to GenBank (accession no. AY572034).

IgG antibodies against SARS-CoV were detected in 2 (5.1%) of 39 employees of the restaurant. This was higher than that observed in the control groups (1%–3%) (11). One employee tested positive for IgM against SARS-CoV in serum samples obtained on January 4 and January 13. Results became negative by January 17, 2004, with no illness or fever in the previous 2 months. This employee worked as a head waitress and often helped customers select palm civets from animal cages. A cook in the restaurant also tested positive for IgG antibody to SARS-CoV.

Comparison of 5 complete S gene sequences (3,768 nt) from palm civets at the restaurant, 22 S gene sequences from SARS patients in the early 2003 epidemic, and 2 viruses isolated from palm civets in 2003 showed 60 nt polymorphisms. Only 5 signature nt variations (SNVs) were observed in the 5 complete S gene sequences from palm civets determined in this study, indicating that SARS-CoV sequences from civets at the restaurant were not different from those of the original animal SARS source. We also observed that 21 SNVs could be used to distinguish viruses with high pathogenicity and infectivity from those with low pathogenicity and infectivity relative to clinical presentation and transmission events (Table 2).

Three of 5 complete S gene sequences from palm civets at the restaurant did not contain any of the 21 SNVs. The remaining 2 isolates (Civet014 and Civet020) had only 1 or 2 SNVs. In contrast, 11 of 22 SARS-CoV strains isolated from humans in Canada, Germany, and Vietnam had all 21 SNV mutations (Table 2). It should be noted that the first human SARS-CoV isolated, GD01, had 17 of 21 SNVs (Table 2). This virus caused severe infections in humans, but did not spread from Guangdong Province (12). Similar SNV patterns were observed in other isolates from patients at the beginning of the 2003 epidemic (13). Four isolates (ZS-A, SZ-B, SZ-C, and HGZ8L1-A) had 18 of 21 SNVs and were obtained from patients with contact histories traceable to some of the earliest independent cases, but with no further transmission recorded (13). Virus GZ02 had 17 SNVs. Another group of 2 early isolates, HSZ-B and HSZ-C, had 19 SNVs, in addition to an 82-nt deletion (13). Virus GZ60, which was isolated from nasopharyngeal aspirates of a healthcare worker at Guangdong Chest Hospital on February 18, 2003, had 18 SNVs (10). Guan et al. named this virus SARS-CoV subcluster A1, together with GD01 and GZ43 (10). None of the sequences of these early-phase isolates have been observed in the middle or later phase of the epidemic, suggesting these isolates had low or mild infectivity (13).

When deduced amino acid sequences were analyzed, 15 signature amino acid variations (SAAVs) were observed that could distinguish between viruses with low or high pathogenicity and infectivity. Three of the 5 recent SARS-CoV isolates from palm civets had no SAAVs, while viruses isolated from outbreaks in various countries had all 15 SAAVs (7,10,14,15). The isolates from the early phase of the 2003 epidemic (GD01, ZS-A, SZ-B, SZ-C, HGZ8L1-A, HSZ-B, HSZ-C, and GZ60) had 12 or 13 SAAVs (6,7,13). SARS-CoV SZ3 and SZ16 had 7 SAAVs mutations (6). The S protein sequence predicted for the first SARS case of 2003–2004 had only 3 SAAVs (13).

When the complete genome sequences of SARS-CoV determined in specimens from palm civets at the restaurant (n = 3), animal markets (n = 2), and patients (n = 23) were compared, the 29-nt deletion (positions 27869–27897) was absent in all isolates from palm civets at the restaurant and at the market, but was present in 22 of 23 patient isolates. The only human isolate (GZ01) without the 29-nt deletion was from a patient in the 2003 epidemic. In addition to S gene sequences, another 42 SNVs were identified, of which 33 were located on the gene encoding P protein (open reading frame [ORF] ab), and on 9 other genes for uncharacterized proteins: ORF 3 (5 SNVs), membrane protein (2 SNVs), and N protein (2 SNVs). However, when complete genome sequences of SARS-CoV from palm civets at the restaurant were compared with those of isolates from palm civets from the market, only 37 SNVs were identified and located on genes encoding P protein (20 SNVs), S protein (11 SNVs), ORF 3a (3 SNVs), M protein (1 SNV), and N protein (2 SNVs). All nucleotide changes were observed in virus sequences of palm civets from the market, but not in virus sequences of animal isolates from the restaurant.

Analysis of the S gene of SARS-CoV showed that viral isolates of animal origin clustered into 2 distinct groups. Group A is represented by SZ3 and SZ16, which were isolated from palm civets in 2003. Group B is represented by viruses found in palm civets at the restaurant (Figure) (6). Analysis of complete genome sequences showed the same relationships (data not shown). These most recent SARS patients were therefore infected by SARS-CoV that is most closely related to virus isolates from palm civets at the restaurant (Figure) (6).