The Human Genome Project: how twenty centres shared one genome

The Human Genome Project is often described as one of the great scientific achievements of the twentieth century. We tend to remember the dates — the first draft announced in 2000, completion announced in 2003 — and less often the rather practical question underneath them: how did researchers decide who would sequence which part of the human genome?

The answer says something about how large scientific collaborations actually work. The public Human Genome Project was not a set of laboratories independently sequencing the same DNA. It was a coordinated international effort in which institutions agreed to divide the work, share their results, and follow common standards. And the division was anything but equal.

A genome too large for one laboratory

When the project formally began in October 1990, sequencing roughly three billion DNA base pairs was a technical challenge on a scale biomedical research hadn't really attempted before. Sequencing machines were slow, and much of the technology the project would eventually rely on didn't exist yet.

The public consortium — known formally as the International Human Genome Sequencing Consortium — eventually brought together 20 universities and research centres across six countries: the United States, the United Kingdom, Japan, France, Germany and China. More than 2,000 researchers^cite took part in total, across institutions, funding agencies and levels of seniority.

Five centres did most of the work, and became known informally as the G5: the Wellcome Trust Sanger Institute in Cambridge, the Whitehead Institute/MIT Center for Genome Research, Washington University in St Louis, Baylor College of Medicine in Houston, and the US Department of Energy's Joint Genome Institute. Smaller institutions contributed specific chromosome regions, physical mapping, or helped finish particularly difficult stretches of sequence.

The logic was straightforward, even if the outcome wasn't even: rather than have everyone work on everything, the consortium assigned chromosomes and chromosome regions to centres according to their capacity, their existing expertise, and their willingness to take responsibility for seeing a piece through to completion.

Who had which chromosome

Responsibility for a chromosome wasn't always a single, permanent assignment. A chromosome typically had a lead centre, sometimes other laboratories responsible for specific regions, and additional groups brought in to help close gaps. By the time the sequence was declared essentially finished in 2003, the lead responsibilities looked roughly like this:

A couple of entries are worth a second look. Chromosome 19 — the most gene-dense of all, despite being one of the smaller chromosomes — went to Stanford, where its density made it unusually slow to finish cleanly. Chromosome 21 went to RIKEN as part of a dedicated sub-consortium, reflecting its early priority as the chromosome linked to Down syndrome; chromosome 22, assigned to Sanger, became the first human chromosome to be finished, in 1999.

The seven chromosomes without one named lead weren't neglected — they were sequenced through genuinely shared effort across multiple centres at different points in the project. That's a messier story than a clean one-to-one table, but it's the more honest one.

Why Sanger sequenced almost a third of the genome

The Wellcome Trust Sanger Institute ended up contributing something close to a third of the entire public sequence. That wasn't the result of some international agreement assigning Britain a fixed share. It was largely practical: Sanger had substantial Wellcome Trust funding, a large sequencing infrastructure, and — under John Sulston and later Allan Bradley — the ability to scale up production faster than most other centres.

The consortium allocated work according to what centres could realistically deliver, not according to a notion of fairness. Equal shares would have made little sense between laboratories with very different numbers of machines, staff and funding. Sanger's outsized contribution is a reminder that equal participation and equal production aren't the same thing — and that one centre doing more doesn't make another centre's work less important. Some regions were simply harder to sequence than others, and some groups were better placed to do the slow, unglamorous work of closing gaps and fixing errors.

Cooperation, under competitive pressure

The consortium wasn't working in a vacuum. In 1998, Craig Venter announced that his company, Celera Genomics, would attempt to sequence the human genome independently, using a different strategy and promising to go faster. The announcement intensified pressure on the public project considerably.

But one of the defining choices of the public effort had already been made two years earlier. Under the Bermuda Principles, agreed in 1996, the heads of the major sequencing centres committed to submitting genome-sequence assemblies above a certain size to a public database within 24 hours of generating them — not at publication, immediately. A laboratory finishing part of chromosome 6 was producing information that a group working on chromosome 11, or a researcher anywhere else in the world, could use the next day.

The public consortium and Celera ultimately published their results side by side, in Nature and Science respectively, in February 2001. The project reached its declared completion milestone in April 2003 — accounting for 92% of the genome, with under 400 gaps — two years ahead of its original schedule. It would take until 2022, and a different consortium using long-read technology, to fill in what remained.

The final sprint owed as much to a computer science graduate student as to any sequencing centre. In the spring of 2000, with the consortium needing a way to stitch together millions of overlapping fragments into something resembling chromosomes, Jim Kent at UC Santa Cruz spent about a month writing an assembly program called GigAssembler, running it on a hastily built cluster of PC workstations his colleague David Haussler had to get emergency funding to buy. Kent's first full assembly came together on 22 June 2000 — days before the White House announcement, and, as it happened, just ahead of Celera's own. (More on that race, and the computing it took to win it, in a later Note.)

What the division of labour tells us

The most interesting thing about the Human Genome Project, to me, is that its success didn't require every participant to contribute equally. It required something more useful: an agreement about who was responsible for what, shared technical standards, and a commitment that the resulting information would belong to everyone, not to whoever got there first.

Writing twenty-five years after the project launched, its own leaders — Eric Green, James Watson and Francis Collins — pointed to the same thing: success depended, they wrote, on researchers being willing to set aside individual achievement for the sake of the collective result.

Sanger's large share doesn't make the work done in Japan, France, Germany, China or elsewhere in the US any less significant. A chromosome that needed five centres to finish wasn't a failure of planning — some regions are genuinely harder than others. The genome belonged to no single laboratory, and that was rather the point.

A note on sources: the chromosome table reflects lead-centre responsibilities recorded around the 2003 "finished sequence" milestone, drawn from UCSC's human genome chromosome reports. Draft-stage assignments earlier in the project sometimes differed, and the exact division of the seven chromosomes listed as shared is worth checking against the original 2001 Nature paper and its 2004 finishing companion before citing a specific centre. General project facts (consortium size, Bermuda Principles, completion figures) are drawn from NHGRI's own account at genome.gov and from Green, Watson and Collins, "Twenty-five years of big biology", Nature 526, 29–31 (2015). The Jim Kent/GigAssembler account draws on UC Santa Cruz's own news archive and contemporary reporting in Dr Dobb's Journal; sources differ on whether the cluster ran 50 or 100 machines, so that detail is left out here.