There are several examples of de-novo gene formation. See work from Dave Begun's lab @ UC Davis. Levine et al. (PNAS) is one example from his lab. There are also various chimeric gene structures that originate from duplications followed by deletions and the like (http://www.genetics.org/content/181/1/313.abstract). These two types of "new genes" are relatively uncommon compared to "plain old" tandem duplications in genomes. These processes are all still relevant, as they are the outcomes of a mutational process that gives rise to them. Natural populations segregate large numbers of copy-number variants for genes, and there is an extensive (albeit somewhat confusing) literature on the topic in humans and some more limited work in Drosophila species.

Kevin, I am skeptical of the claims of whole gene exonization. I suspect that some genes may pseudogenize or become inactive (through frameshift etc) in all but one lineage which can give the appearance of de novo formation from ncDNA. However, some recruitment of intronic or UTR sequences (like Sdic) into coding regions is possible, but this doesn't produce a wholly new gene. The examples I gave from the hox and integrin families are not addressed by Begun and others.

When your are asking, how does gene duplication explain major innovations in biochemical function, then one is faced with the question of how divergent is the new paralog copy. A very nice review of the different fates of these copies is e.g. Innan & Kondrashov, Nature Review Genetics, 2010 or Connant and Wolfe, Nature Review Genetics, 2008. There are several examples of the rise of a new gene functions in one copy (see e.g. Hasselmann et al. 2008, Nature, the case of the sex determining gene in honey bee), which come together with modifications in protein domains and gene structure. If you talking about "new type" of genes, one should carefully adress the question, if the underlying gene function is somehow new when compared to its ancestral copy.

The best example - among many others - are the different globin genes which are specifically regulated during development.

Another way to get new genes is the fusion between 2 simper genes, according to the exon shuffling model. It is supposed that the sequences coding for the different modules of the protein correspond to one or more exons. However, some introns may have been lost, after the fusion. Maybe you will find our paper on the origin of WFIKKN in PLOSone usefull.

There are at least three possibilities for the emergence of completely new protein structures observed in a phylogeny.

However a study that compares them head-to-head in term of their evolutionary contribution is extremely difficult to assess. It'd be great to see it though.

1. The most addressed is that a gene duplicates and loses its selective pressure, and before it's lost as a pseudogene, it enters a phase of adaptive evolution, which forces it into unexplored portions of the sequence/funcional space. This possibility, however, has only a few recognized examples (like antifreeze proteins in cold resistant fishes).

2. The second one, which has gathered attentions over the last years is de novo from non coding regions. There are numerous examples for this, including protein and non protein coding genes in yeasts, mice, humans, arabidopsis, ants, drosophila and fishes.

3. The third one has been recognized in viruses for a long time, it's called overprinting of alternative reading frames, and implies the exaptation of a previously unused reading frame, different from the one that bears the current selective pressure. The extent to which this happens in large eukaryotes is still unknown, although the author of de novo paper on yeast says it's a high proportion of what seems de novo.

For introductory references i'd suggest the following literature:

http://www.annualreviews.org/doi/abs/10.1146/annurev-ecolsys-110411-160513

http://www.nature.com/nrg/journal/v12/n10/abs/nrg3053.html

http://onlinelibrary.wiley.com/doi/10.1002/9780470015902.a0024601/abstract

http://www.ncbi.nlm.nih.gov/pubmed/20651121

More elaborate studies, similar to the ones highlighted above by Begun and Jones:

http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3401362/

http://www.ncbi.nlm.nih.gov/pubmed/18493065

http://www.ncbi.nlm.nih.gov/pubmed/21948395

http://www.cell.com/trends/genetics/abstract/S0168-9525%2807%2900299-5

http://www.biomedcentral.com/1471-2164/14/117

www.ncbi.nlm.nih.gov/pubmed/19733073

http://www.ncbi.nlm.nih.gov/pubmed/19064677

www.ncbi.nlm.nih.gov/pubmed/19716618

On overprinting:

http://www.ncbi.nlm.nih.gov/pubmed/22821011

http://www.pnas.org/content/89/20/9489.full.pdf

From my point of view, the evolutionary significance of gene duplication is to increase the likelihood of their expression and thus to strengthen the reliability of signs , in coding which they participate. On the other hand, repeatedly duplicated genes are probably serve as a "storage" of mutations that are harmful for the current environmental conditions, but giving the evolutionary advantage of any other conditions. Thus, when environmental conditions change quickly organisms can " find" the most useful for these conditions mutations from accumulated set and therefore quickly adapt to them.

Joseph--you asked about the potential for de novo origin of genes. That is conceptually separate from what you describe with the hox genes (fairly standard example of gene duplication). The work from the Begun lab has been quite careful--there is no evidence that a gene exists in the syntenic region throughout the fly phylogeny. Further, population-level sequencing demonstrates the existence of the de novo gene in all samples.

In a way, your question "is gene duplication still relevant" is poorly-formed. If there are ancient gene duplications in the genome, it stands to reason that the process is still going on (e.g., dozens of papers from Manyuan Long's group). However, that is not the only mechanism by which new functions evolve.

Kevin, the origin of the 7 proto-genes of the hox cluster is not "fairly standard". There is no evidence that they are related by gene duplication, unlike the other genes in the cluster and copied on different chromosomes. As for "de novo", I still think it is dubious, largely based on my reading of one paper used in support of it: Recent de novo origin of human protein-coding genes (Knowles and Mclysaght, 2009): http://genome.cshlp.org/content/early/2009/08/31/gr.095026.109

It turns out that the authors have discovered three genes that suffered a frameshift and subsequent deactivation in the lineage that preceded the great apes. They were subsequently reactivated in the human lineage. They are NOT "new genes", just resurrected ones.

I think you should check out Ashley Byun's research on protein subcellular relocalization. She's identified gene duplications that have changes in the signal peptide that sends them to new organelles, where they adopt new functions. She had a 2007 review in TREE and has new data showing fairly strong evidence that gene duplications lead to diversification of function. More info at my blog here: https://lxmx.wordpress.com/2013/06/28/evolution-2013-greatest-hits/

Hi,

As Rafik said, there is some "de novo" gene origination going on that is beyond doubt: the cases where a new ORF is created by overprinting an ancestral frame in a different frame (creating an overlapping gene arrangement - maybe "dual coding" is more precise). Because the ancestral frame was already here, it's possible to do a good alignment with outgroups and be CERTAIN there was no other gene there before (in the other reading frame).

eg

articles by our group

http://www.ncbi.nlm.nih.gov/pubmed/22821011

http://jvi.asm.org/content/83/20/10719.long

http://www.ncbi.nlm.nih.gov/pubmed/23966842 (incredible case of gene nursery)

and

http://www.ncbi.nlm.nih.gov/pubmed/23847207 (rather detailed on the mechanism of origination)

Overprinting is very frequent in viruses but studies by Pavel Baranov and other bioinfo studies (see refs in "gene nursery" paper) suggest that they may be many overlapping genes in the human genome- hundreds wouldnt amaze me. http://www.ncbi.nlm.nih.gov/pubmed/22593554

So that's many cases of de novo protein origination...

There is a reason I didn't mention the Knowles paper.

David (Karlin) Two points about your interesting gene nursery paper. Has anyone gone the next step of the overprinted gene scenario, and found genes that are reading in only one frame, but clearly originated by overprinting i.e one gene > overprint > duplicate > two genes? Second, you state that "Since virus genes diverge very fast..." - that presumably refers to, for example, divergence in a new host, but not long term changes in a hallmark gene.

Hi Adrian,

- people have found genes in eukaryotes and elsewhere that appeared to have originated by frameshifted translation from duplicate genes- in fact there's been quite a few studies but they're never been followed up (eg rarely cited and each study reinvents the wheel and maybe the genes aren't proven or annotated the reference genomes in and none of this reaches the mainstream consciousness). As to the opposite, frameshift then duplication and loss of the overlap separately in both overlaps... I get asked the question often but could find no example, or very short. If both phenomenon exist they're probably relatively rare.

- when I say viruses diverge very fast, well Sacha Gorbalenya showed a slide where an enzyme (I think a protease) of the Picornaviridae family has accumulated more changes than the equivalent proteases of bacteria, archaea and eukaryotes put together - so yes they do diverge extremely far and explore the universe more than other organisms - I don't know about long term changes (nobody knows, do they, but I get your point! I'll try to phrase it differently, any suggestion?

David, frameshift mutations almost inevitably produce premature stop codons and a very short ORF. If a frameshift has occurred in a duplicate gene in the past, it must have been subsequently suppressed and compensated for by other mutations. See my paper on the frameshift responsible for the evolutionary divergence of the KPNA gene family: https://www.researchgate.net/publication/51105792_An_ancient_frame-shifting_event_in_the_highly_conserved_KPNA_gene_family_has_undergone_extensive_compensation_by_natural_selection_in_vertebrates?ev=prf_pub

The idea that a frameshift can produce an entirely new ORF of sufficient length, and one that encodes a functional protein is, I believe, a total nonsense. But please tell me if you think I am mistaken.

Article An ancient frame-shifting event in the highly conserved KPNA...

Thanks Joseph, I'll read it with interest but can you please send me a PDF, I don't have access to journals at the moment (you could also upload it as a PDF in ResearchGate for the future, that'll allow easy access).

Many viruses produce, through frameshift, "entirely new ORFs of a certain length (eg >100aa)", see my papers or that of Andrew Firth for instance. These proteins are completely indispensable to the viruses. And these viruses infect eukaryotic cells so they are subjected to the laws of translation of eukaryotes (though they control and change them) So I am quite unsure what your reasoning is based on. Perhaps the viruses are subject to very different pressures that allow any ORF that has a function, even minutely useful, to ...persist. But there's nothing that prevents a virus or eukaryote to generate a short ORF by frameshift and then lengthen this ORF progressively anyway. I do think in fact there are many cases like those in eukaryotes but they're not believed or noticed 'cause official genome annotators reject most things non-canonical...

David, it is very simple and logical reasoning. There are 3 stop codons in the 64 codon genetic code. That means that, in a randomized DNA sequence, we should expect there to be an in-frame stop site every 20 amino acids or so subsequent to translation. I am sorry, but I just do not believe your notion that viruses can produce "indispensable" proteins through a frameshift, generating an "entirely new ORF" - we had such a claim made in the now famous "nylonase" case, as proposed by Ohno 30 years ago. It has since been discarded: http://www.jbc.org/content/280/47/39644.long (free paper). Moreover, even if the virus did manage to generate an ORF of sufficient length, there is only an infinitely remote chance of having this randomly encoded peptide actually being functional. I suspect that you have probably have made a mistake in your interpretation of the data like the great Susumu Ohno did.

Hi Joseph,

thanks for the reference. I have made no calculation or interpretation. The fact is that viruses frequently encode long, overlapping genes, and you can ask any virologist at your institution to confirm this. The most famous one (and perhaps the most intriguing) is the overlap between the Reverse transcriptase and the Glycoprotein of the Hepatitis B virus, longer than 1800nt. This overlap and the proteins it encodes has been studied for >20 years by hundreds of virologists. For instance this paper http://www.ploscompbiol.org/article/info%3Adoi%2F10.1371%2Fjournal.pcbi.1003162 gives a list of 27 overlapping genes proven beyond doubt, with references to the experimental publications, and their function, when known (quick link to table 1:

http://www.ploscompbiol.org/article/fetchObject.action?uri=info:doi/10.1371/journal.pcbi.1003162.t001&representation=PNG_M)

The overlapping frames are accessed by a wide variety of mechanisms, eg leaky scanning, Internal Ribosomal Entry Site (IRES), ribosomal frameshift, subgenomic RNA, stop-codon readthrough, splicing etc.

Because both frames in the hepatitis B overlap encode indispensable proteins, ie the Reverse transcriptase and the Glycoprotein, it seems puzzling how they may have originated and you may exert your sagacity, mathematical skills, and biological intuition to try and understand that.

Regarding your calculations, since I'm sure they're right, then how overlapping frames can exist in viruses at all? Nobody knows, but I suppose some of the assumptions may not apply here; for instance that overlapping genes originate in a "randomized" sequence; the idea of "sufficient" length (overlapping genes may evolve very progressively from very short ORFs); and the hypothesis that a long overlapping frame, if it managed to originate despite your misgivings, would not be functional. In fact, a newly accessed ORF, even if not functional per se and no matter how short, would be expect to alter the quantity of the other overlapping frame produced (by diverting the ribosome, for instance, in the case of ribosomal frameshift, or of leaky scanning), and could thus provide a selective advantage leading to it being retained. This idea has been proposed among others by Niv Sabath and Andrew Firth; and we provide a rather striking potential example in the deltaretrovirus gene overlap, in the article above.

Best,

David, I think you are confusing overlapping ORFs - which occur even in mitochondrial DNA - with novel ORFs accidentally produced by a frameshift mutation. You can't seriously claim a "de novo" origin of an alternative ORF that already exists but which may lie dormant and under-expressed! Overlapping genes, like alternative splicing, are just an efficient way of storing information using a minimum amount of sequence space. They cannot be used to explain the origin of new genes at the heart of my question. I do, however, think that domain accretion may be possible through a frameshift mutation that pushes the stop site further downstream and extends the ORF.

Hi Joseph,

Ok, I understand there's been a confusion. I called proteins originated by overprinting "de novo" by opposition as origination from a pre-existing protein coding sequence (proteins generated by overprinting have a sequence unrelated to any other, without statistically significant similarity to others, and some of them have 3D folds unlike any other; whereas proteins originated by duplication, fusion, or fission of pre-existing coding sequences will have at least part of their 3D fold structurally similar to others). I use the terminology used by others in the field (cf earlier reply by Rafik Neme) but agree that "originated by overprinting" is more precise.

Now comes the time to understand what *you* mean by "novel ORFs accidentally produced by a frameshift mutation" and "the origin of new genes at the heart of my question" -can you please detail very, very precisely (do you mean from non-coding ORFs? If you mean from a coding ORF in which insertion of a single nucleotide generates a downstream sequence that is different, this is very similar to overprinting in the sense than a cryptic ORF may be accessed in this way). But I probably havent' understood what you mean - can you please take the time to explain very precisely. Thanks. D

David, in the case of a gene duplication, the notion that a frameshift will produce an alternative ORF and functional gene product - unrelated to the original sequence - is highly implausible. Unfortunately, a paper by Scherer et al. (2006) was misinterpreted as meaning just that: http://www.sciencedirect.com/science/article/pii/S0888754306001807 (full text available). What the authors actually found were instances where a frameshift has occured in part of the sequence but was suppressed downstream by a mutation that restored the rest of the ORF. I have also found that this is a very common event in duplicate genes. So, the frameshift just created a short region of sequence divergence rather than an entirely novel ORF. The notion that non-coding sequences can be recruited as novel ORFs also suffers from the problem of premature stop codons, as well as the fact that it is highly unlikely that they would be functional even if accidentally transcribed and translated.

If I understand you properly, you suggest that an indel mutation can mean that a "cryptic" or "dormant" alternative ORF can be accessed because it puts the overlapping ORF now in-frame and the other one out-of-frame. I don't see anything wrong with that at all, but the frameshift mutation is really acting as a switch determining which of the two pre-existing ORFs is transcribed and translated. So the real question regards the origins of both ORFs rather than the fact that they overlap each other.

The original question asks about recent evolution still involving gene duplication. The example of the antifreeze protein which involved the large scale amplification of short repeats is very recent and cured several times independently during the recent ice ages a few tens of thousands of years ago under extreme selective pressure.

The recent big surprise that a large part of the genetic diversity between individuals is due to copy number variation shows how there is more than enough available basic material for gene duplication diversity generation.

The rapid growth and diversification of the interferon-alpha gene family implies that there has been recent (viral-based?) selective pressure at work. One non-functional factor that might be involved here would be the possibility of having differential control of related genes in response to different stimuli in different tissues. This has been seen (I don't think anyone mentioned this yet in this discussion) in the evolution of the eye lens in animals and birds, occurring at least twice maybe fie or six times independently; the eye-lens protein (crystallin) in the Chimney Swift and Humming-Bird families being a recruited lactose dehydrogenase in cell cytoplasm of other cells, or argino-succinate lyase in many other vertebrates. Differential tissue expression of gene variants is also seen between embryonic and adult tissues. Dual functionality (enzymatic function and serendipitous(?) structural function) as well as the fact that gene duplication would be advantageous in facilitating differential gene regulation in various tissues comes into play here. There would be selective pressure at work here for birds that are flying all day long in the sunshine and the fact that their lens crystallin is more UV resistant.

Also there is ongoing evolution under human selective pressure in animal breeding. Duplication of short repeats within coding sequences effect fine tuning of some gene functions encoded, as for example in the muzzle curvature of dogs or in polydactyly.

I hope these comments have addressed some aspects of the question that others have not covered.

Having worked for many years on peptide display libraries, producing so-called 'never born' genes based on synthetic oligonucleotides one gets a feeling for the type of serendipity which is requires to get function, even simply one of getting say 10 interatomic interactions in order to find a ligand that will bind a fairly flat groove on the surface of an enzyme. Any pentamer or even heptamer can be rapidly isolated for a deep pocket interaction immediately. For targets without deep pockets, working with random oligos encoding, say, a 29mer peptide, it is possible to isolate regions containing such functional (affinity) features at a frequency of say 1 in a hundred experiments from libraries containing some 10e9 variants ( a minute fraction of the potential variants which could be produced within a short oligonucleotide (say 100bp); but far in excess of the sequence variety actually present in the human genome ) and where further refinement is obtained by random point mutation of the lead gene. This is essentially the mechanism that our immune system uses in generating new specific antibodies, whereby comparisons of using natural and synthetic oligos in the regions which are involved in the intramolecular binding show that there has already been a selection for natural regions which are better at contributing to binding affinity. Generation and selection thus functions using the strategy of recombining several short segments, selecting for a weak affinity (function) and then site-specific mutation and further selection.

These empirical findings thus support the observation that most 'novel' genes will still be initially generated as variants of existing genes or by using small functional domains (remembering Wally Gilbert's theory that exon structure favours recombination between short regions encoding functional 25 to 50 aa domains) and subsequent (slow or rapid; depending on selection pressure) incremental changes, compared to using some 'untried' random sequence of serendipitous DNA.

Gene duplication is the only empirical mechanism available for the birth of genes as of now. Yet another mechanism is de-novo gene birth, yet we have no proof for this hypothesis, see http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3401362/

It depends what you mean by empiricism, certainly I have seen no one doubting the origination of new protein-coding genes by overprinting in viruses... See all my papers and that of others above. So if it happens by overprinting why not just by using a preexisting ORF...

Thanks Tobias. Actually, the phenomenon that you mention has been observed several times in viruses! Some viral proteins that have demonstrably originated completely de novo (since they arose by overprinting a preexisting, ancient reading frame) and are known to have NEW FOLDS AND FUNCTIONS: p19 from the plant viruses tombusviruses, which has both a previously unknown fold and a new way of binding to short interfering RNAs [www.ncbi.nlm.nih.gov/pubmed/14697199]; and ORF9b of SARS coronavirus, which has a completely new (very weird) fold www.ncbi.nlm.nih.gov/pubmed/16843897

They're both discussed in several of my papers e.g www.ncbi.nlm.nih.gov/pubmed/22821011 and www.ncbi.nlm.nih.gov/pubmed/19640978.

We are trying to identify more of these proteins and to solve their 3D structure eg the dataset in www.ncbi.nlm.nih.gov/pubmed/23966842 ...

Regards, David

Hi Tobias, yes indeed I would say they are almost textbook definition of adaptative conflict! It is also spot-on that small viral genomes probably don't have mechanisms in place to duplicate, or they would get big fast; on the contrary, large viruses (herpes, pox, or the giant, recently discovered mimivirus or pandoravirus) have lots of duplicated genes.

In general I would say that a solution (at least partial) to the conflict is that there's often one of the overlapping genes (in general the ancestral one) that's constrained, whereas the one originated de novo by overlapping it is much less constrained (it's a "slave" of the other one). The redundancy of the genetic code seems to give quite some leeway. However, yes, as I say above and give references, two cases are known (tombusvirus p19 and Coronavirus ORF9b) where the de novo proteins are structured -their 3D structure has been solved!), and in both cases, amazingly they overlap two proteins that are ALSO structured and therefore very constrained (respectively the viral movement protein and the viral Nucleoprotein, whose structure is known). This must generate an enormous constraint (!) but how it's managed by the virus hasn't been studied.

We recently discovered a related phenomenon which may be much more common (the P/C gene overlap of Paramyxovirinae). P and C adopt the arrangement above where one gene is constrained and the other one variable except for a short region that encodes simultaneously two short, crucial functional motifs in the P and C frames. This obviously imposes a great constraint on that particular region that should be so deleterious it is probably compensated by an evolutionary advantage. Indeed, this phenomenon had been predicted earlier - if theres a high mutation rate and the motifs are short (which is the case), then encoding simultaneously two important motifs decreases the size of the "fragile" genetic region encoding them and is advantageous www.ncbi.nlm.nih.gov/pubmed/15638463.

Our article is in minor revision and should be published soon in PLoS One.

Notice that's very different from encoding simultaneously two highly constrained proteins - that's another feat entirely, and the how and the why are still mysterious!

I'd say that de novo proteins start by probably being mostly flexible and unconstrained and can evolve towards being structured (though not necessarily). That's not solving the mystery! :-)