Sequencing the SARS Virus

In April 2003, we at the Genome Sciences Centre (GSC) publicly released the first complete sequence assembly of the coronavirus now believed to be the cause of Severe Acute Respiratory Syndrome (SARS). The GSC has been using Linux for all of its analysis, storage and network infrastructure since its inception in 1999. The sequence data from the SARS Project was stored, processed and publicly distributed from a number of Linux servers, from the capable and slim IBM x330 to the behemoth eight-way Xeon x440. Linux has provided a flexible infrastructure allowing us to automate nearly every process in our sequencing pipeline. With the support of the Linux community, by way of newsgroups, Web articles and HOWTOs, we have been able to leverage commodity and mid-range hardware in an incredibly cost-efficient manner.

Since the first documented incidence of SARS on November 16, 2002, the virus has been responsible for a total of 8,458 cases reported in China (92%), Canada (3%), Singapore (2%) and the United States (1%), as well as in more than 25 other countries. SARS mortality rate is roughly 5–10% and as high as 50% in people older than 60. As of June 24, 2003, SARS has claimed 807 lives and has had a profoundly negative impact on the economies of the affected regions—China alone stands to lose billions of dollars in revenue from tourism and taxation.

On March 27, 2003, the director of our Centre, Marco Marra, and our project leader, Caroline Astell, decided to sequence the SARS coronavirus. At 1AM on April 7, 2003, approximately 50ng of genetic material from the Tor2 isolate of the pathogen, derived from a patient in Toronto, Canada, arrived from the Level 4 National Microbiology Lab in Winnipeg, Canada. Five days later, on April 12, 2003, our 29,751 base assembly of the sequence of the Tor2 isolate (Tor2/SARS) of the coronavirus was posted to a Zope/Plone page on our Apache server for public access. A few days later, the sequence of the Urbani isolate was posted by the (Centers for Disease Control) CDC in Atlanta, Georgia.

Before the 1990s, technology to collect large amounts of sequence information rapidly did not exist. The Human Genome Project (HGP) began in 1991, and by 1999 only 15% of the sequence had been collected. However, thanks to new methods, which were developed during the 1990s, the HGP turned sharply toward completion. By mid-2000, 90% of the human sequence was available, and currently the genome sequence essentially is complete. Data from sequencing projects like HGP is stored and publicly accessible through NCBI's Genbank.

Figure 1. A panorama of our sequencing lab: 1) barcodes representing procedures, 2) the Tango liquid handling platform, 3) –112°F freezers, 4) power supplies for thermocyclers, 5) ABI 3730XL sequencers, 6) ABI 3700 sequencers, 7) x330 sequencer control cluster, 8) network/power connections and 9) vent ducts for sequencers.

During its first ten years of operation (1982–1992), Genbank collected just over 100MB of sequence in 80,000 records. During the next decade (1992–2002) Genbank's rate of growth skyrocketed, and the database grew to 29GB—ten times the size of the human genome—in 22 million records. Genbank receives on the order of 10,000 sequence records each day from sequencing labs across the world. One of these labs is the GSC, which on April 13, 2003, deposited the sequence of Tor2/SARS to Genbank. To see how Linux was involved in the process leading to the submission of sequence gi:29826276, we need to go back to the beginning.

In June 1999, the lab consisted of six beige-box computers and just as many people. The central file server (2xP3-400, 512MB of RAM, Red Hat 5.2 and 2.0.36 kernel) was serving out three RAID-0 18GB SCSI disks using a DPT IV card. Another 50GB of software RAID was exported by a second machine (P3-400). With three other Linux clients and a Microsoft Windows NT station, these machines were on the BC Cancer Agency (BCCA) network.

Figure 2. First-generation server hardware: 1) VA Linux VAR900 2xXeon-500 exporting 1TB, 2) Raidion.u2w RAID controllers, 3) 2x8x36GB SCSI disks and 4) VA Linux 2230s and 3x10x72GB SCSI disks.

The timing of our beginnings worked to our advantage. Like all research labs, we needed to share disks, distribute processes, compile software and store and munge data. In other words, all the things at which UNIX excels. Had we started 2–3 years earlier, adopting the fledgling Linux would have been difficult. It's likely that, instead of now relegating inexpensive old PCs to office or less-intensive network tasks, we would be trying to maximize return on our substantial investment of aging Sun servers. Fortunately, it turned out that it was possible to buy the relatively inexpensive PCs, install Linux and have a robust, flexible and incredibly cost-effective UNIX environment. Thanks to Linux, it was no longer necessary to spend an entire salary on a UNIX workstation.

It was a good time to choose Linux. The 2.0 kernel was rock solid; the NFS server was stabilizing, and a choice of full-featured desktop environments was available. We were able to download or compile the essential toolbox for bioinformatics analysis, such as the open-source workhorses of the HGP: BLAST (sequence comparison), Phred (base calling of traces produced by sequencers), Phrap (sequence assembly) and Consed (visualization of sequence assemblies), as well as various sequence and protein databases. Of course, Perl filled in any cracks. Our cost of entry into getting computational work done was low, and we could spend grant funds more efficiently to expand the lab (Figure 1).