Ok, while I understand that this sort of article might be interesting, honestly it's only scratching the surface of the current state of the art (and the idea of TE's acting as fodder for evolution isn't exactly new).
Short reply -- Get Martin Nowak's book Evolutionary Dynamics
Long reply -- Even Nowak's work is just the beginning (disclaimer: my current Ph.D. thesis work is building off of Nowak's start). Let's start with this notion that TE's are vital for evolution. Real answer: sort of. You see, there's an awful lot of species that don't have an overabundance of "junk" DNA, yet are still able to adapt and evolve. The reason that TE's at least seem to be more important in higher order plants and animals is because of how such organisms organize their genomes.
Ok, before I get carried away on a long rant, here's the heart of the matter for the hacker audience:
Proteins fold into domains. These are compact structures with a rough upper limit to their size that typically can carry out one biochemical function. In E. coli, most genes code for proteins with only one domain. So, E. coli can evolve by grabbing new genes, turning genes on or off, or in some circumstances, evolve new functionality from existing domains through random mutation. (In reality, random mutation takes a long time to produce anything useful, and the creation of novel domains appears to no occur. There's a theory growing in prominence that all of the protein domain folds that exist today were present when life began, and may even represent independent life originating events...but I'm getting off track.)
In humans, most of our proteins are multi-domain. Not all of these domains, however, are catalytic. That is, in a human protein with 5 domains, maybe only one actually has biochemical activity. The function of the rest is to modulate that activity or localize the protein to one part of the cell or another. Also, the mechanisms controlling when genes get turned on and off in humans is much more complex than in bacteria. Therefore, evolution in higher order plants and animals is much more likely to occur through a "shuffling" of domains and regulatory elements. Because TE's are good at "shuffling" DNA, it's not surprising that having a healthy dose of this sort of "junk" DNA is advantageous. Of course, that's not all...there's also neutral evolution and pseudo-genes and epigenetic inheritance, etc. Biology is really on the cusp of exploding (oh, and I'm writing a book about that too ;-).
tl;dr -- The genetics of higher order plants and animals is not unlike a program which relies heavily on many libraries. If you swap out one XML parser for a better XML parser, you'll get better performance (and more customers!). Transposable elements function (sort of) like biology's equivalent of a linker, and can help organisms swap in and out libraries/protein domains.
"There's a theory growing in prominence that all of the protein domain folds that exist today were present when life began, and may even represent independent life originating events"
Who in the world is advocating for that theory? It sounds silly, on its face...one has only to spend an afternoon browsing the SCOP or PFAM databases to see that there's been a huge amount of recent evolution at the fold level, and that the domains that we know about have appeared over a very long time.
Gah! Ok, apologies. I have to learn to stop over-oversimplifying!
What I mean there was this: Mathematical modeling of the evolution of protein folds points toward the impossibility of convergently evolving folds (but not convergently evolving functionality). This implies that two proteins with the same fold can reasonably be assumed to have evolved from a common ancestor protein, even if that conclusion cannot be reached from sequence information.
Unfolded proteins (at least, above a certain length) cause problems for living things. Thus, the ability of evolution to freely explore fold-space is constrained. There is some interesting work going on in this space looking at the possibility networks of interrelated folds that don't pass through unfolded intermediates, but I think it's too soon to say, for certain, that these networks are sufficient to generate truly "novel" folds.
As for the SCOP classification system, my personal view is that it tends to be on the restrictive side. Of course, that's the point of SCOP: to robustly categorize folds. As for PFAM, it's been a while, but last I looked they still don't consider any 3D structural information in their classifications. I guess what I'm trying to say here is that, whether or not "novel" protein folds are actively appearing depends on your definition of "novel". If I mutate a residue in the middle of a helix that breaks the helix in two, is that a novel fold? If I then insert a few more amino acids and turn that one helix into a helix-turn-helix, is that a novel fold?
The theory I alluded to is not my own, but I can admire the logic behind it. The idea is that it is possible to group many folds through sequence and other (i.e. threading) means into derived folds. However, even when you do this, you don't arrive at a rootless tree. Instead, you find that there are somewhere (depending on who you ask) between 800 and 1300 "roots" to the fold family tree.
Presumably, these roots represent novel abiogenesis events. At the very least, these root folds must have existed before "modern" biology (i.e. the sort that cannot tolerate unfolded states) began. Whether they all have a common ancestor or not is very much up for debate, but so far I don't know that we have conclusive evidence one way or the other.
To be clear, though, I am guilty of oversimplification in the line you've picked out.
Nice post. (For the record, the simplification is understandable...I got my PhD in this stuff, so I'm not exactly the target audience.)
I agree that convergent evolution of protein folds seems unlikely to have happened frequently, but I'm not willing to dismiss it as impossible over the course of evolutionary time-frames. More importantly, I'm not willing to extend that idea to the conclusion that all protein folds must have been extant at the beginning of evolutionary history. There are simply too many ways for new protein folds to be produced -- and not necessarily by stepwise point mutations through stable intermediates.
You're right that PFAM isn't a structural classification system per se; they do what you describe a bit later, and build massive sequence profiles to detect homologies. But at the sequence identity levels used to build PFAM, you can safely assume that any sequence within a family that has a solved structure will probably share that structure's overall fold. The details will be different, but the fold will be conserved. That's really the point of PFAM (and why it has a symbiotic relationship with SCOP).
But here's the problem with drawing too many structural conclusions from sequence analysis: with state-of-the art algorithms, we can detect structural homologies out to about 25-30% sequence identity. Beyond that, we just don't know how to call two structures "similar" or "dissimilar", without having the structures themselves.
Point being, you can't really say that those 800-1300 "roots" of the evolutionary tree are independent starting points. All you can conclude is that our tools aren't good enough (or there isn't enough data) to trace back the evolutionary tree to the point where those "roots" may have converged.
For the record, I'm not trying to be pendantic or argumentative. This is one of those few fun debates that makes the field interesting. ;-)
Thanks. I'm always game for a good debate. As I mentioned, my thesis-in-progress (T-minus 3 months...fingers crossed!) is in a related field (evolutionary dynamics/computational biochemistry), but I used to be more involved in protein structure/bioinformatics back in the day (worked, for a time, in the lab that maintains ecogene.org).
The last debate I had regarding convergent evolution of protein folds got rather heated. This is one of those classic problems that can't be approached without some amount of hand-waving, and depending on which way you wave, you can arrive at different results. In some respects it boils down to Levinthal's paradox, except with evolutionary moves in place of topological ones. The one big unknown that you would need to find before you could make any sort of educated approach at the issue is what fraction of all possible protein chains of a certain size have stable, fast-folding structural minima. If that number is high, then short hops from one to the other could very likely result in convergence.
As for the SCOP/PFAM part of the story, the 25-30% "twilight zone" for sequences yielding related structures has a counterpart with structures that are topologically related but with low (or I've even seen cases of essentially no) sequence identity. That is, if you look at a group in PFAM, then take all of the members with structures in that group, gather all of the corresponding SCOP groups, add the members from the SCOP groups, then for each of those look for related PFAMs or other sequences...essentially, what you're doing is a structure informed PSI-BLAST (sort of what ESPRESSO is to T-COFFEE).
Now, if you attempt to fill the gaps from lack of structures by running each of your sequences through something like Skolnics TASSER threading algorithm, and using the best predicted folds to grow your group, this is how you can use distant sequence and structure homologies to construct "master" fold groups. This is, roughly speaking, how the 800-1300 number is arrived at.
Admittedly, the more structures we solve and genomes we sequence, the better this sort of technique will get. In many ways, this is biology's analog to cosmology and the big bang: we can look further and further out into space, but at some point, we can know what happened any earlier than some small time after the big bang without recreating those conditions. Likewise, we can sequence and solve more structures, but I don't doubt that at some point we will hit a wall and need to start attempting to recreate abiogenesis.
I haven't looked into it, but I wonder what would happen if you plotted the proportion of intergenic dna in the genome and species reproduction rate. Thoughts?
Not much. For unicellular organisms, size of genome correlates relatively well with time to divide (which, if you only consist of one cell, is the same as reproducing). This is one of the reasons that bacteria and virii have such compact genomes.
Once you get past the single-celled stage, though, the correlation essentially disappears. Reproduction of a multicellular organism is a highly coordinated team effort involving multiple cells/tissues. For these organisms, "intergenic" DNA will correlate more closely with complexity of regulation (but not close enough to be horribly interesting).
It's fairly well-established that non-coding DNA regions are important for biological function. From regulatory elements that work via DNA and RNA bending, to the catalytic RNA molecules that are essential for genome maintenance, the portions of the genome that don't actually code for genes are full of interesting stuff. So the title is a bit sensational, really.
That said, transposable elements are really not the more interesting parts of the non-coding stuff -- we understand them well, and they don't do much of anything except disrupt genes during reproduction. To draw a rough analogy: if you have a chunk of code with lots of changes in whitespace, it's probably also under heavy development. You would be wrong to conclude that "whitespace is essential for software evolution", but you might well be able to quantify the software development progress by tracking only the whitespace.
>the portions of the genome that don't actually code for genes are full of interesting stuff.
Thats true, but if you estimate that 2% of the genome is genes, perhaps a further 1% is regulatory elements and catalytic RNA. That still leaves a heck of a lot of 'junk'.
9 comments
[ 2.8 ms ] story [ 28.3 ms ] threadShort reply -- Get Martin Nowak's book Evolutionary Dynamics
Long reply -- Even Nowak's work is just the beginning (disclaimer: my current Ph.D. thesis work is building off of Nowak's start). Let's start with this notion that TE's are vital for evolution. Real answer: sort of. You see, there's an awful lot of species that don't have an overabundance of "junk" DNA, yet are still able to adapt and evolve. The reason that TE's at least seem to be more important in higher order plants and animals is because of how such organisms organize their genomes.
Ok, before I get carried away on a long rant, here's the heart of the matter for the hacker audience:
Proteins fold into domains. These are compact structures with a rough upper limit to their size that typically can carry out one biochemical function. In E. coli, most genes code for proteins with only one domain. So, E. coli can evolve by grabbing new genes, turning genes on or off, or in some circumstances, evolve new functionality from existing domains through random mutation. (In reality, random mutation takes a long time to produce anything useful, and the creation of novel domains appears to no occur. There's a theory growing in prominence that all of the protein domain folds that exist today were present when life began, and may even represent independent life originating events...but I'm getting off track.)
In humans, most of our proteins are multi-domain. Not all of these domains, however, are catalytic. That is, in a human protein with 5 domains, maybe only one actually has biochemical activity. The function of the rest is to modulate that activity or localize the protein to one part of the cell or another. Also, the mechanisms controlling when genes get turned on and off in humans is much more complex than in bacteria. Therefore, evolution in higher order plants and animals is much more likely to occur through a "shuffling" of domains and regulatory elements. Because TE's are good at "shuffling" DNA, it's not surprising that having a healthy dose of this sort of "junk" DNA is advantageous. Of course, that's not all...there's also neutral evolution and pseudo-genes and epigenetic inheritance, etc. Biology is really on the cusp of exploding (oh, and I'm writing a book about that too ;-).
tl;dr -- The genetics of higher order plants and animals is not unlike a program which relies heavily on many libraries. If you swap out one XML parser for a better XML parser, you'll get better performance (and more customers!). Transposable elements function (sort of) like biology's equivalent of a linker, and can help organisms swap in and out libraries/protein domains.
Who in the world is advocating for that theory? It sounds silly, on its face...one has only to spend an afternoon browsing the SCOP or PFAM databases to see that there's been a huge amount of recent evolution at the fold level, and that the domains that we know about have appeared over a very long time.
What I mean there was this: Mathematical modeling of the evolution of protein folds points toward the impossibility of convergently evolving folds (but not convergently evolving functionality). This implies that two proteins with the same fold can reasonably be assumed to have evolved from a common ancestor protein, even if that conclusion cannot be reached from sequence information.
Unfolded proteins (at least, above a certain length) cause problems for living things. Thus, the ability of evolution to freely explore fold-space is constrained. There is some interesting work going on in this space looking at the possibility networks of interrelated folds that don't pass through unfolded intermediates, but I think it's too soon to say, for certain, that these networks are sufficient to generate truly "novel" folds.
As for the SCOP classification system, my personal view is that it tends to be on the restrictive side. Of course, that's the point of SCOP: to robustly categorize folds. As for PFAM, it's been a while, but last I looked they still don't consider any 3D structural information in their classifications. I guess what I'm trying to say here is that, whether or not "novel" protein folds are actively appearing depends on your definition of "novel". If I mutate a residue in the middle of a helix that breaks the helix in two, is that a novel fold? If I then insert a few more amino acids and turn that one helix into a helix-turn-helix, is that a novel fold?
The theory I alluded to is not my own, but I can admire the logic behind it. The idea is that it is possible to group many folds through sequence and other (i.e. threading) means into derived folds. However, even when you do this, you don't arrive at a rootless tree. Instead, you find that there are somewhere (depending on who you ask) between 800 and 1300 "roots" to the fold family tree.
Presumably, these roots represent novel abiogenesis events. At the very least, these root folds must have existed before "modern" biology (i.e. the sort that cannot tolerate unfolded states) began. Whether they all have a common ancestor or not is very much up for debate, but so far I don't know that we have conclusive evidence one way or the other.
To be clear, though, I am guilty of oversimplification in the line you've picked out.
I agree that convergent evolution of protein folds seems unlikely to have happened frequently, but I'm not willing to dismiss it as impossible over the course of evolutionary time-frames. More importantly, I'm not willing to extend that idea to the conclusion that all protein folds must have been extant at the beginning of evolutionary history. There are simply too many ways for new protein folds to be produced -- and not necessarily by stepwise point mutations through stable intermediates.
You're right that PFAM isn't a structural classification system per se; they do what you describe a bit later, and build massive sequence profiles to detect homologies. But at the sequence identity levels used to build PFAM, you can safely assume that any sequence within a family that has a solved structure will probably share that structure's overall fold. The details will be different, but the fold will be conserved. That's really the point of PFAM (and why it has a symbiotic relationship with SCOP).
But here's the problem with drawing too many structural conclusions from sequence analysis: with state-of-the art algorithms, we can detect structural homologies out to about 25-30% sequence identity. Beyond that, we just don't know how to call two structures "similar" or "dissimilar", without having the structures themselves.
Point being, you can't really say that those 800-1300 "roots" of the evolutionary tree are independent starting points. All you can conclude is that our tools aren't good enough (or there isn't enough data) to trace back the evolutionary tree to the point where those "roots" may have converged.
For the record, I'm not trying to be pendantic or argumentative. This is one of those few fun debates that makes the field interesting. ;-)
The last debate I had regarding convergent evolution of protein folds got rather heated. This is one of those classic problems that can't be approached without some amount of hand-waving, and depending on which way you wave, you can arrive at different results. In some respects it boils down to Levinthal's paradox, except with evolutionary moves in place of topological ones. The one big unknown that you would need to find before you could make any sort of educated approach at the issue is what fraction of all possible protein chains of a certain size have stable, fast-folding structural minima. If that number is high, then short hops from one to the other could very likely result in convergence.
As for the SCOP/PFAM part of the story, the 25-30% "twilight zone" for sequences yielding related structures has a counterpart with structures that are topologically related but with low (or I've even seen cases of essentially no) sequence identity. That is, if you look at a group in PFAM, then take all of the members with structures in that group, gather all of the corresponding SCOP groups, add the members from the SCOP groups, then for each of those look for related PFAMs or other sequences...essentially, what you're doing is a structure informed PSI-BLAST (sort of what ESPRESSO is to T-COFFEE).
Now, if you attempt to fill the gaps from lack of structures by running each of your sequences through something like Skolnics TASSER threading algorithm, and using the best predicted folds to grow your group, this is how you can use distant sequence and structure homologies to construct "master" fold groups. This is, roughly speaking, how the 800-1300 number is arrived at.
Admittedly, the more structures we solve and genomes we sequence, the better this sort of technique will get. In many ways, this is biology's analog to cosmology and the big bang: we can look further and further out into space, but at some point, we can know what happened any earlier than some small time after the big bang without recreating those conditions. Likewise, we can sequence and solve more structures, but I don't doubt that at some point we will hit a wall and need to start attempting to recreate abiogenesis.
Interesting times!
Once you get past the single-celled stage, though, the correlation essentially disappears. Reproduction of a multicellular organism is a highly coordinated team effort involving multiple cells/tissues. For these organisms, "intergenic" DNA will correlate more closely with complexity of regulation (but not close enough to be horribly interesting).
That said, transposable elements are really not the more interesting parts of the non-coding stuff -- we understand them well, and they don't do much of anything except disrupt genes during reproduction. To draw a rough analogy: if you have a chunk of code with lots of changes in whitespace, it's probably also under heavy development. You would be wrong to conclude that "whitespace is essential for software evolution", but you might well be able to quantify the software development progress by tracking only the whitespace.
Thats true, but if you estimate that 2% of the genome is genes, perhaps a further 1% is regulatory elements and catalytic RNA. That still leaves a heck of a lot of 'junk'.