Marco Island, Florida (where the Advances in Genome Biology and Technology conference is being held) is certainly the place to be this week.

Greg Baute’s optimistic predictions about the year 2010 in sequencing may prove more accurate than my own pessimism yet.

But now I know the people who will be the first to find out the answers to these questions. Today (yesterday by the time this is scheduled to publish) Pacific Biosystem’s announced where the first ten of their new sequencing machines will be going:


View Pacific Biosystems Sequencers in a larger map (click the markers to see the names of the institutions receiving the sequencers)

You’ll notice the two urban areas that are receiving two are the San Francisco Bay Area, and St. Louis.

I don’t know what else to say, I just thought this would be fun data to map.

I do wish Cornell had been included in the first round as I’d really like more sequence data from diverse varieties of corn and PacBio sequencers should have an easier time with the highly repetitive maize genome than second generation sequencers.

They tell me patience is a virtue…

*Pacific Biosystems is the first company to produce “third generation” sequencing machines. The PacBio sequencer reveals the sequences of longer stretches of DNA all at once. Imagine assembling a sequenced genome as putting together a puzzle. The fewer, bigger, pieces, the easier the puzzle. It’s the difference between the 500 and 1000 piece versions of the same puzzle. My understanding is there are other companies developing third generation sequencing technology, but they are certainly keeping a much lower profile. Pacific Biosystem’s public relations has had most biologists I know, including myself, waiting breathlessly for the release of these instruments since my first year of grad school.

As long as SOLiD sequencing can keep giving Illumina a run for its money, the price of sequencing is going to keep dropping, and the R&D departments of both companies will be working round the clock to keep the improvements coming (SOLiD is already promising upgrades that will triple the amount of sequence generated per run, while cutting the cost of each run by half (6x reduction in cost/GB of sequence)… by the end of this year.)

People talk about Moore’s law as saying computers double in speed every 18 months.* I’ve heard estimates that the cost of sequencing is dropping as much at 90% per year. To get a sense of how impressive that is, if those rates of growth keep up, after nine years, computers would be 64 times as fast, and sequencing costs per base pair would be less than one-billionth  (that is 1/1,000,000,000) what they are today.

*This isn’t what was actually stated as Moore’s law, and even if it was, the speed of computer processors stopped growing some time ago, with the extra transistors (the thing actually predicted to increase by Moore’s law), getting devoted to additional cores instead. So todays computers can’t do a single task faster than last years computers, but they can do more things quickly at once (depending on the task programmers can down get around this by breaking individual tasks down into smaller pieces that can each be worked on by a different processor core at the same time). Read more on wikipedia.

PolITgenomics has a great, detailed post up about the new HiSeq 2000 sequencing machines from Illumina. Illumina sequencers (still often called Solexa, at least by me) generate lots and lots of short DNA sequences all at once. Assembling a genome out of those sequences can be hard** but the latest generation of Illumina sequencers, including the HiSeq 2000, generates sequences almost twice as long as those used to assemble the Panda genome which was published late last year (100 bp long vs ~50). The biggest difference between the HiSeq 2000 and existing Illumina sequencers is the increase in sequence data generated per machine per day.

The Genome Analyzer IIx, which was previously the top of the line model from Illumina, generates 25 gigabases of sequence per run and takes fives days to run, or 5 gigabases (a little over 1.5 times the amount of sequence in the human reference genome) of dna sequences per day, although, keep in mind, that’s all in the form of short hundred base pair fragments, that then has to either be pieced together like a giant jigsaw puzzle, or matched onto an existing genome to reveal unique mutations present in the newly sequenced individual or species***. Either way, it’s already a huge amount of data.

The new machine generates sequences that are the same length but makes them a little faster (4 days) and generates a LOT more of them (100 gigabases of sequence).  A single machine is generating an average of 25 gigabases of sequence per DAY. Illumina had to redesign parts of their analysis software because the new instrument generates sequence faster than it could be saved to a hard drive!

And to top it all off, according to Omics! Omics! the Beijing Genomics Institute has already ordered 128 of the new sequencers. That’s enough sequencing machinery to generate 3200 gigabases (1000 times the size of the human genome) of sequence data per day. In 2009 the Joint Genome Institute which is part of the Department of Energy and a sequencing behomoth in its own right generated just over 1000 gigabases of high quality base calls. 0nce these machines are up and running the Beijing Genomics Institute will be producing twelve times the total 2009 sequence output of JGI every four days (on average they’ll produce much sequence as JGI did in 2009 every eight hours!).****   I don’t know what they intend to do with all that sequencing power, but I expect that it’ll be something incredible… actually they’ll probably end up doing lots of different things.

Where the number 30,000 comes from:

In a previous post I calculated a top of the market sanger sequencer from 2005 (just before the first next generation sequencing techologies came on the market) with 96 capillary tubes which could be run eight times per day would generated 750 kilobases of sequence per day. The new machine from Illumina generated 25 gigabases per day (ot 25,000,000 kilobases to keep my units consistent). 25,000,000/750 = 33,333 (rounded to the closest whole machine)

*This overstates the improvement as Sanger sequencers still generate longer sequences that are easier to piece together, but this is mostly only an issue with sequencing entirely new genomes, and a lot of the research that’s being done today is in organisms where there’s already a reference genome to map sequence data onto (for example in humans) which makes the sorted lengths of sequences generated by Illumina sequencing a much smaller issue.

**In my post on the panda genome I compared it to trying to piece together a novel 500 times as long as War and Peace using phrases this short:

nglish ambassador’s? Today is Wednesday. I must put

adron of varicolored horsemen. Two of them rode side

***If you’re one of the people like me who is geeky enough to want to sequence their own genome, but a lot more money to throw around ($48,000-68,000 last I read about it) this is exactly the way your genome would be sequenced. The Illumina sequencing machines generate gigabases of short DNA sequences that are then mapped by computers onto the reference human genome that was produced at a cost of $2.7 billion dollars over 13 years (under budget and ahead of schedule believe it or not!). By mapping imperfectly matched sequences, this technique will determine which base pair sequences are different in your genome compared to the reference sequence (ie you have an A where the reference human genome has a G). It’s like putting a puzzle together on top of another, already completed, puzzle with almost exactly the same picture.

***To be fair, JGI is continuing to tool up as well. The amount of sequence they generated increased by a factor of eight between 2008 and 2009.

Sequencing the maize genome took four years and 30 million dollars. Today Virginia Tech announced the University of Minnesota and themselves had received a $908,000 grant to sequence the Turkey genome in two years. I don’t know how big or complex the turkey genome is, but the idea of sequencing a whole new species for less than a million dollars is still pretty cool.

h/t to 538 they’ve got more cool turkey statistics over there.

Wild strawberry (left) and domesticated strawberry (right)

Clearly one of the major traits early strawberry growers selected for was bigger fruits. Which makes sense since it takes about the same amount of time an effort to pick a strawberry either way, but if you’re picking the ones on the right you’ll have more pounds of fruit picked at the end of the day.

But even in this case, the similarity in form hides major changes at the genome left. Strawberries went through two whole genome duplications during domestication (looks like it’s more complicated than I made it sound see comments), so each of the cells in the strawberries on the right contain eight copies of each chromosome, while the strawberry on the left contains the more standard two copies of each chromosome.

On the other end of the spectrum is maize. Maize is like nothing else on this earth, and for the longest time, many botanists and agronomists were convinced it must have been domesticated from a wild species that had since gone extinct. In fact the wild ancestor, a species of teosinte, of maize is so closely related to corn they can still mate with each other, the key definition of a species.

Teosinte was initially disregarded as the ancestor of corn because the two plants look very different. I could show you this if I’d been able to find a public domain photo of teosinte (or taken a picture of some of the plants when I rotated in a lab that was growing them). As is, you’ll have to take my word for it. To my eye a corn plant (at least up until it flowers) looks more like sorghum than it does like teosinte. The ears of corn also bear little resemblance to the tiny row of seeds produced by the female flowers of teosinte. (For a comparison of the two, look at the picture on the left here.)

Anyway, long before the corn genome came out, the fact corn had been domesticated from teosinte had been established. Some really clever classical genetics** suggested that 5-6 major genetic changes were responsible for most of the obvious physical changes between teosinte and corn. The Doebley lab actually mapped at least two of those changes, teosinte branched 1 (TB1) and teosinte glum architecture (TGA1).

Teosinte branched1 was one of the first genes that got me really excited about maize genetics. In fact it was this specific paper, assigned to me by an awesome post doc who refused to let me work for him until I know what I was talking about. The teosinte branched1 gene is actually expressed more in domesticated corn than in corn’s teosinte ancestors the result of a mutation in front of the gene. The gene’s job is to turn off clusters of cells near the base of each leaf which otherwise would develop into whole new stalks. The result is obvious:

From left to right mutant teosinte branched1 corn (where the tb1 gene doesn't work properly), normal corn, and sorghum. Key point: notice how the teosinte branched1 mutant corn and sorghum plants both have many stalks, while most corn has a single big one. The photo of the tb1 plant comes from maizeGDB the other two are my own.

But domestication isn’t just about a few big physical changes, there are a bunch of traits that farmers and farming select for both intentionally and unintentionally. Everything from when a crop flowers to how it tastes is under constant selection. So while 5-6 major traits may explain a lot of the obvious changes, the story of corn’s domestication includes many more genes. The publication of the corn genome is making it easier to track down those other domestication genes. Ed Buckler’s group is already identifying many of them.

When a single form of a gene which creates some desirable trait and is selected for, it displaces all the other variations of that gene. A telltale sign of selection for a particular gene is that comparing the versions of the gene from different individuals shows fewer differences than expected. Ed Buckler’s group generated more than 32,000,000,000 bases (the equivalent of 13 whole maize genomes) from 27 maize lines around the world. They identified 148 regions that where very similar between all 27 lines, a glaring difference from the normally incredible genetic diversity of corn. Finding these new candidates for domestication genes (I’ll be fascinated to learn more about which genes are on that list) were made possible by three things:

• The corn genome sequencing project
• The current generation of massively parallel sequencing technologies that make generating 32 gigabases of sequence possible for less than millions of dollars
• The hard work of people like Michael Gore and Jer-ming Chia the two lead authors on this paper.

*The single biggest exception is probably seafood. Also venison, and other wild game, and well as a few forms of wild berries and nuts.

**Classical genetics is the kind you can do without a sequenced genome or most of the modern molecular biology tools that make life so much easier for biologists of my generation. In this case, by mating corn and teosinte and looking at the ratios of traits in the grandchildren

I’ve got several posts on the maize genome coming out scheduled to go up later today. Living on the west coast (not to mention having a circadian clock that seems convinced I should actually be living on Honolulu time) it’s the only way to get information up in time for morning readers in most of the US.

Anyway, hopefully some of what I’ve written makes sense (I’ll be running a lot of long computational jobs at work so I’ll have plenty of time to answer questions in the comment sections about all the stuff I’ve written that doubtless makes no sense at all). But to start us off this morning, how about a short (<4 minutes) video from Patrick Schnable one of the two lead authors on the maize genome paper. After four years of talking about the corn genome project as well as it’s challenges and benefits, one gets very good at it.*

See the video in it’s original context here. I’m assuming since ISU provides embedding code they’re ok with me showing it here.

*Fair disclosure, there are important reasons I may be biased in my evaluation.