Every time one of the cells in your body divides, it has to double its quota of DNA so that each daughter cell gets a complete set. DNA is a replicator—a molecule that can be accurately duplicated, admittedly with some help from proteins. DNA has been doing this for billions of years, well before there were humans, before animals existed, and probably before the first cells evolved.
But what came before DNA? Probably RNA, a related molecule. Certain types of RNA can store genetic information, just like DNA. And much like proteins, they can fold into three-dimensional shapes to speed up chemical reactions, among other functions—these are called ribozymes.
The dominant theory is that an “RNA world” preceded the origin of life. It’s possible that the Earth’s first true replicators were RNA molecules that could fold up to speed up their own replication. They copied themselves. They did so imperfectly, creating daughter molecules with slightly different sequences. Some of them copied themselves more efficiently, and left more descendants than their peers. Gradually, the entire population evolved towards ever more efficient replication.
But there’s a problem with this story. The RNA molecule we’re talking about would have been long and folded into a complex ribozyme. But the ribozymes that scientists can make today are simple, and made from very short pieces of RNA. You can imagine a simple molecule gradually growing and evolving into a more complex one, but that idea has problems too. Mathematical models predict that this burgeoning replicator would be unable to copy itself accurately enough, and would start accumulating errors. After a while, it would face an “error catastrophe”, where the build-up of mistakes crippled it.
But what if there wasn’t just one RNA replicator copying itself? What if, instead, there was a whole network of them? This idea was originally floated in 1971 by Nobel-winning chemist Manfred Eigen. “He came to the conclusion that an individual replicator couldn’t persist for very long, and came up with the idea of a hypercycle,” says Niles Lehman from Portland State University. That is, molecule A helps B to copy itself. B helps C, C helps D and so on, eventually looping back to A.
Eigen predicted the existence of hypercycles using mathematics. Now, Lehman has created something similar in a test tube. It’s a contrived set-up, and it doesn’t confirm that such networks were genuinely involved in the origin of life, but it shows that they can form, and that they become more complex over time. As James Attwater and Philipp Holliger from the University of Cambridge write in an accompanying piece, the study makes “a persuasive case for the benefits of cooperation even at this nascent stage of life. The first genes may not have been so selfish, after all.”
Back in 2001, the Human Genome Project gave us a nigh-complete readout of our DNA. Somehow, those As, Gs, Cs, and Ts contained the full instructions for making one of us, but they were hardly a simple blueprint or recipe book. The genome was there, but we had little idea about how it was used, controlled or organised, much less how it led to a living, breathing human.
That gap has just got a little smaller. A massive international project called ENCODE – the Encyclopedia Of DNA Elements – has moved us from “Here’s the genome” towards “Here’s what the genome does”. Over the last 10 years, an international team of 442 scientists have assailed 147 different types of cells with 24 types of experiments. Their goal: catalogue every letter (nucleotide) within the genome that does something. The results are published today in 30 papers across three different journals, and more.
For years, we’ve known that only 1.5 percent of the genome actually contains instructions for making proteins, the molecular workhorses of our cells. But ENCODE has shown that the rest of the genome – the non-coding majority – is still rife with “functional elements”. That is, it’s doing something.
It contains docking sites where proteins can stick and switch genes on or off. Or it is read and ‘transcribed’ into molecules of RNA. Or it controls whether nearby genes are transcribed (promoters; more than 70,000 of these). Or it influences the activity of other genes, sometimes across great distances (enhancers; more than 400,000 of these). Or it affects how DNA is folded and packaged. Something.
According to ENCODE’s analysis, 80 percent of the genome has a “biochemical function”. More on exactly what this means later, but the key point is: It’s not “junk”. Scientists have long recognised that some non-coding DNA has a function, and more and more solid examples have come to light [edited for clarity – Ed]. But, many maintained that much of these sequences were, indeed, junk. ENCODE says otherwise. “Almost every nucleotide is associated with a function of some sort or another, and we now know where they are, what binds to them, what their associations are, and more,” says Tom Gingeras, one of the study’s many senior scientists.
And what’s in the remaining 20 percent? Possibly not junk either, according to Ewan Birney, the project’s Lead Analysis Coordinator and self-described “cat-herder-in-chief”. He explains that ENCODE only (!) looked at 147 types of cells, and the human body has a few thousand. A given part of the genome might control a gene in one cell type, but not others. If every cell is included, functions may emerge for the phantom proportion. “It’s likely that 80 percent will go to 100 percent,” says Birney. “We don’t really have any large chunks of redundant DNA. This metaphor of junk isn’t that useful.”
That the genome is complex will come as no surprise to scientists, but ENCODE does two fresh things: it catalogues the DNA elements for scientists to pore over; and it reveals just how many there are. “The genome is no longer an empty vastness – it is densely packed with peaks and wiggles of biochemical activity,” says Shyam Prabhakar from the Genome Institute of Singapore. “There are nuggets for everyone here. No matter which piece of the genome we happen to be studying in any particular project, we will benefit from looking up the corresponding ENCODE tracks.”
There are many implications, from redefining what a “gene” is, to providing new clues about diseases, to piecing together how the genome works in three dimensions. “It has fundamentally changed my view of our genome. It’s like a jungle in there. It’s full of things doing stuff,” says Birney. “You look at it and go: “What is going on? Does one really need to make all these pieces of RNA? It feels verdant with activity but one struggles to find the logic for it.
Think of the human genome as a city. The basic layout, tallest buildings and most famous sights are visible from a distance. That’s where we got to in 2001. Now, we’ve zoomed in. We can see the players that make the city tick: the cleaners and security guards who maintain the buildings, the sewers and power lines connecting distant parts, the police and politicians who oversee the rest. That’s where we are now: a comprehensive 3-D portrait of a dynamic, changing entity, rather than a static, 2-D map.
And just as London is not New York, different types of cells rely on different DNA elements. For example, of the roughly 3 million locations where proteins stick to DNA, just 3,700 are commonly used in every cell examined. Liver cells, skin cells, neurons, embryonic stem cells… all of them use different suites of switches to control their lives. Again, we knew this would be so. Again, it’s the scale and the comprehensiveness that matter.
“This is an important milestone,” says George Church, a geneticist at the Harvard Medical School. His only gripe is that ENCODE’s cells lines came from different people, so it’s hard to say if differences between cells are consistent differences, or simply reflect the genetics of their owners. Birney explains that in other studies, the differences between cells were greater than the differences between people, but Church still wants to see ENCODE’s analyses repeated with several types of cell from a small group of people, healthy and diseased. That should be possible since “the cost of some of these [tests] has dropped a million-fold,” he says.
The next phase is to find out how these players interact with one another. What does the 80 percent do (if, genuinely, anything)? If it does something, does it do something important? Does it change something tangible, like a part of our body, or our risk of disease? If it changes, does evolution care?
[Update 07/09 23:00 Indeed, to many scientists, these are the questions that matter, and ones that ENCODE has dodged through a liberal definition of “functional”. That, say the critics, critically weakens its claims of having found a genome rife with activity. Most of the ENCODE’s “functional elements” are little more than sequences being transcribed to RNA, with little heed to their physiological or evolutionary importance. These include repetitive remains of genetic parasites that have copied themselves ad infinitum, the corpses of dead and once-useful genes, and more.
To include all such sequences within the bracket of “functional” sets a very low bar. Michael Eisen from the Howard Hughes Medical Institute said that ENCODE’s definition as a “meaningless measure of functional significance” and Leonid Kruglyak from Princeton University noted that it’s “barely more interesting” than saying that a sequence gets copied (which all of them are). To put it more simply: our genomic city’s got lots of new players in it, but they may largely be bums.
This debate is unlikely to quieten any time soon, although some of the heaviest critics of ENCODE’s “junk” DNA conclusions have still praised its nature as a genomic parts list. For example, T. Ryan Gregory from Guelph University contrasts their discussions on junk DNA to a classic paper from 1972, and concludes that they are “far less sophisticated than what was found in the literature decades ago.” But he also says that ENCODE provides “the most detailed overview of genome elements we’ve ever seen and will surely lead to a flood of interesting research for many years to come.” And Michael White from the Washington University in St. Louis said that the project had achieved “an impressive level of consistency and quality for such a large consortium.” He added, “Whatever else you might want to say about the idea of ENCODE, you cannot say that ENCODE was poorly executed.” ]
Where will it lead us? It’s easy to get carried away, and ENCODE’s scientists seem wary of the hype-and-backlash cycle that befell the Human Genome Project. Much was promised at its unveiling, by both the media and the scientists involved, including medical breakthroughs and a clearer understanding of our humanity. The ENCODE team is being more cautious. “This idea that it will lead to new treatments for cancer or provide answers that were previously unknown is at least partially true,” says Gingeras, “but the degree to which it will successfully address those issues is unknown.
“We are the most complex things we know about. It’s not surprising that the manual is huge,” says Birney. “I think it’s going to take this century to fill in all the details. That full reconciliation is going to be this century’s science.”
Find out more about ENCODE:
So, that 80 percent figure… Let’s build up to it.
We know that 1.5 percent of the genome codes for proteins. That much is clearly functional and we’ve known that for a while. ENCODE also looked for places in the genome where proteins stick to DNA – sites where, most likely, the proteins are switching a gene on or off. They found 4 million such switches, which together account for 8.5 percent of the genome.* (Birney: “You can’t move for switches.”) That’s already higher than anyone was expecting, and it sets a pretty conservative lower bound for the part of the genome that definitively does something.
In fact, because ENCODE hasn’t looked at every possible type of cell or every possible protein that sticks to DNA, this figure is almost certainly too low. Birney’s estimate is that it’s out by half. This means that the total proportion of the genome that either creates a protein or sticks to one, is around 20 percent.
To get from 20 to 80 percent, we include all the other elements that ENCODE looked for – not just the sequences that have proteins latched onto them, but those that affects how DNA is packaged and those that are transcribed at all. Birney says, “[That figure] best coveys the difference between a genome made mostly of dead wood and one that is alive with activity.” [Update 5/9/12 23:00: For Birney’s own, very measured, take on this, check out his post. ]
That 80 percent covers many classes of sequence that were thought to be essentially functionless. These include introns – the parts of a gene that are cut out at the RNA stage, and don’t contribute to a protein’s manufacture. “The idea that introns are definitely deadweight isn’t true,” says Birney. The same could be said for our many repetitive sequences: small chunks of DNA that have the ability to copy themselves, and are found in large, recurring chains. These are typically viewed as parasites, which duplicate themselves at the expense of the rest of the genome. Or are they?
The youngest of these sequences – those that have copied themselves only recently in our history – still pose a problem for ENCODE. But many of the older ones, the genomic veterans, fall within the “functional” category. Some contain sequences where proteins can bind, and influence the activity of nearby genes. Perhaps their spread across the genome represents not the invasion of a parasite, but a way of spreading control. “These parasites can be subverted sometimes,” says Birney.
He expects that many skeptics will argue about the 80 percent figure, and the definition of “functional”. But he says, “No matter how you cut it, we’ve got to get used to the fact that there’s a lot more going on with the genome than we knew.”
[Update 07/09 23:00 Birney was right about the scepticism. Gregory says, “80 percent is the figure only if your definition is so loose as to be all but meaningless.” Larry Moran from the University of Toronto adds, “Functional” simply means a little bit of DNA that’s been identified in an assay of some sort or another. That’s a remarkably silly definition of function and if you’re using it to discount junk DNA it’s downright disingenuous.”
This is the main criticism of ENCODE thus far, repeated across many blogs and touched on in the opening section of this post. There are other concerns. For example, White notes that many DNA-binding proteins recognise short sequences that crop up all over the genome just by chance. The upshot is that you’d expect many of the elements that ENCODE identified if you just wrote out a random string of As, Gs, Cs, and Ts. “I’ve spent the summer testing a lot of random DNA,” he tweeted. “It’s not hard to make it do something biochemically interesting.”
Gregory asks why, if ENCODE is right and our genome is full of functional elements, does an onion have around five times as much non-coding DNA as we do? Or why pufferfishes can get by with just a tenth as much? Birney says the onion test is silly. While many genomes have a tight grip upon their repetitive jumping DNA, many plants seem to have relaxed that control. Consequently, their genomes have bloated in size (bolstered by the occasional mass doubling). “It’s almost as if the genome throws in the towel and goes: Oh sod it, just replicate everywhere.” Conversely, the pufferfish has maintained an incredibly tight rein upon its jumping sequences. “Its genome management is pretty much perfect,” says Birney. Hence: the smaller genome.
But Gregory thinks that these answers are a dodge. “I would still like Birney to answer the question. How is it that humans “need” 100% of their non-coding DNA, but a pufferfish does fine with 1/10 as much [and] a salamander has at least 4 times as much?” [I think Birney is writing a post on this, so expect more updates as they happen, and this post to balloon to onion proportions].]
[Update 07/09/12 11:00: The ENCODE reactions have come thick and fast, and Brendan Maher has written the best summary of them. I’m not going to duplicate his sterling efforts. Head over to Nature’s blog for more.]
* (A cool aside: John Stamatoyannopoulos from the University of Washington mapped these protein-DNA contacts by looking for “footprints” where the presence of a protein shields the underlying DNA from a “DNase” enzyme that would otherwise slice through it. The resolution is incredible! Stamatoyannopoulos could “see” every nucleotide that’s touched by a protein – not just a footprint, but each of its toes too. Joe Ecker from the Salk Institute thinks we should be eventually able to “dynamically footprint a cellular response”. That is, expose a cell to something—maybe a hormone or a toxin—and check its footprints over time. You can cross-reference those sites to the ENCODE database, and reconstruct what’s going on in the cell just by “watching” the shadows of proteins as they descend and lift off.)
Find out more about ENCODE:
The simplistic view of a gene is that it’s a stretch of DNA that is transcribed to make a protein. But each gene can be transcribed in different ways, and the transcripts overlap with one another. They’re like choose-your-own-adventure books: you can read them in different orders, start and finish at different points, and leave out chunks altogether.
Fair enough: We can say that the “gene” starts at the start of the first transcript, and ends at the end of the final transcript. But ENCODE’s data complicates this definition. There are a lot of transcripts, probably more than anyone had realised, and some connect two previously unconnected genes. The boundaries for those genes widen, and the gaps between them shrink or disappear.
Gingeras says that this “intergenic” space has shrunk by a factor of four. “A region that was once called Gene X is now melded to Gene Y.” Imagine discovering that every book in the library has a secret appendix, that’s also the foreword of the book next to it.
These bleeding boundaries seem familiar. Bacteria have them: Their genes are cramped together in a miracle of effective organisation, packing in as much information as possible into a tiny genome. Viruses epitomise such genetic economy even better. I suggested that comparison to Gingeras. “Exactly!” he said. “Nature never relinquished that strategy.”
Bacteria and viruses can get away with smooshing their protein-encoding genes together. But not only do we have more proteins, but we also need a vast array of sequences to control when, where and how they are deployed. Those elements need space too. Ignore them, and it looks like we have a flabby genome with sequence to spare. Understand them, and our own brand of economical packaging becomes clear. (However, Birney adds, “In bacteria and viruses, it’s all elegant and efficient. At the moment, our genome just seems really, really messy. There’s this much higher density of stuff, but for me, emotionally it doesn’t have that elegance when we see in a bacterial genome.“)
Given these blurred boundaries, Gingeras thinks that it no longer makes sense to think of a gene as a specific point in the genome, or as its basic unit. Instead, that honour falls to the transcript, made of RNA rather than DNA. “The atom of the genome is the transcript,” says Gingeras. “They are the basic unit that’s affected by mutation and selection.” A “gene” then becomes a collection of transcripts, united by some common factor.
There’s something poetic about this. Our view of the genome has long been focused on DNA. It’s the thing the genome project was deciphering. It is converted into RNA, giving it a more fundamental flavour. But out of those two molecules, RNA arrived on the planet first. It was copying itself and evolving long before DNA came on the scene. “These studies are pointing us back in that direction,” says Gingeras. They recognise RNA’s role, not as simply an intermediary between DNA and proteins, but something more primary.
Find out more about ENCODE:
For the last decade, geneticists have run a seemingly endless stream of “genome-wide association studies” (GWAS), attempting to understand the genetic basis of disease. They have thrown up a long list of SNPs – variants at specific DNA letters—that correlate with the risk of different conditions.
The ENCODE team have mapped all of these to their data. They found that just 12 percent of the SNPs lie within protein-coding areas. They also showed that compared to random SNPs, the disease-associated ones are 60 percent more likely to lie within functional, non-coding regions, especially in promoters and enhancers. This suggests that many of these variants are controlling the activity of different genes, and provides many fresh leads for understanding how they affect our risk of disease. “It was one of those too good to be true moments,” says Birney. “Literally, I was in the room [when they got the result] and I went: Yes!”
Imagine a massive table. Down the left side are all the diseases that people have done GWAS studies for. Across the top are all the possible cell types and transcription factors (proteins that control how genes are activated) in the ENCODE study. Are there hotspots? Are there SNPs that correspond to both? Yes. Lots, and many of them are new.
Take Crohn’s disease, a type of bowel disorder. The team found five SNPs that increase the risk of Crohn’s, and that are recognised by a group of transcription factors called GATA2. “That wasn’t something that the Crohn’s disease biologists had on their radar,” says Birney. “Suddenly we’ve made an unbiased association between a disease and a piece of basic biology.” In other words, it’s a new lead to follow up on.
“We’re now working with lots of different disease biologists looking at their data sets,” says Birney. “In some sense, ENCODE is working form the genome out, while GWAS studies are working from disease in.” Where they meet, there is interest. So far, the team have identified 400 such hotspots that are worth looking into. Of these, between 50 and 100 were predictable. Some of the rest make intuitive sense. Others are head-scratchers.
Find out more about ENCODE:
Writing the genome out as a string of letters invites a common fallacy: that it’s a two-dimensional, linear entity. It’s anything but. DNA is wrapped around proteins called histones like beads on a string. These are then twisted, folded and looped in an intricate three-dimensional way. The upshot is that parts of the genome that look distant when you write the sequences out can actually be physical neighbours. And this means that some switches can affect the activity of far away genes
Job Dekker from the University of Massachusetts Medical School has now used ENCODE data to map these long-range interactions across just 1 percent of the genome in three different types of cell. He discovered more than 1,000 of them, where switches in one part of the genome were physically reaching over and controlling the activity of a distant gene. “I like to say that nothing in the genome makes sense, except in 3D,” says Dekker. “It’s really a teaser for the future of genome science,” Dekker says.
Gingeras agrees. He thinks that understanding these 3-D interactions will add another layer of complexity to modern genetics, and extending this work to the rest of the genome, and other cell types, is a “next clear logical step”.
Find out more about ENCODE:
ENCODE is vast. The results of this second phase have been published in 30 central papers in Nature, Genome Biology and Genome Research, along with a slew of secondary articles in Science, Cell and others. And all of it is freely available to the public.
The pages of printed journals are a poor repository for such a vast trove of data, so the ENCODE team have devised a new publishing model. In the ENCODE portal site, readers can pick one of 13 topics of interest, and follow them in special “threads” that link all the papers. Say you want to know about enhancer sequences. The enhancer thread pulls out all the relevant paragraphs from the 30 papers across the three journals. “Rather than people having to skim read all 30 papers, and working out which ones they want to read, we pull out that thread for you,” says Birney.
And yes, there’s an app for that.
Transparency is a big issue too. “With these really intensive science projects, there has to be a huge amount of trust that data analysts have done things correctly,” says Birney. But you don’t have to trust. At least half the ENCODE figures are interactive, and the data behind them can be downloaded. The team have also built a “Virtual Machine” – a downloadable package of the almost-raw data and all the code in the ENCODE analyses. Think of it as the most complete Methods section ever. With the virtual machine, “you can absolutely replay step by step what we did to get to the figure,” says Birney. “I think it should be the standard for the future.”
Find out more about ENCODE:
Compilation of other ENCODE coverage