Sometimes, gene duplication occurs within a species: part of the chromosome may be duplicated, causing one, a few, or many genes to have more copies of themselves within the species. The descendants of the duplicates and the original are homologous are they are descended from a common ancestor. This type of homology is called paralogy: a homology due to a duplication event (para == in parallel).

In another case, the genes can be homologous due to speciation: a new species (A1) diverges from the original (A0), carrying highly similar genetic loads. The gene for, say, brown eyes in A1 and the gene for brown eyes in A0 are also homologous: derived from the gene of hemoglobin in A0. This time, the homology is called orthology: it is not due to in-species duplication, but due to speciation itself (ortho == exact).  The definitions of orthologs and paralogs were given by Walter Fitch in a seminal paper published in 1970.

One of the first protein structures to be solved was that of hemoglobin, the oxygen carrying protein complex in our blood. Scientists noticed that hemoglobin in jawed mammals has three different protein chains: alpha, beta and gamma. Their amino acid sequences were very similar, suggesting that the genes encoding for hemoglobin are highly similar, suggesting homology. Since all jawed mammals have hemoglobin, and they all had alpha, beta and gamma chains, the conclusion was that the duplication of the original genes happened in the common ancestor of  jawed mammals, before they split up into different species. Hence, the alpha, beta and gamma chains in hemoglobin are paralogous: homologous due to duplication preceding speciation. However, gamma-hemoglobin was shown to have a different function than beta or alpha (more on that in a bit).  The conclusion from this observation was the Ortholog Conjecture and it can be stated as follows: paralogs (reminder: homologs due to duplication) diverge in function more than orthologs (homologs due to speciation). A model was proposed for this observation: when genes duplicate within a species’ genome, there is less selective pressure on one copy to perform the same function. Thus, it can accumulate mutations and eventually adopt a different function. The ortholog conjecture states that paralogs mostly differ in function, whereas orthologs mostly do not. The ortholog conjecture is a very powerful statement because, if we have two proteins known to be orthologs, we can infer that they have the same function, whereas paralogs may not (if they had enough time to diverge). The ortholog conjecture is therefore a fundamental tenet in molecular phylogenetics, and is also a tool used to predict the function of proteins. If two homologous proteins are found out to be orthologs, then it is assumed they have the same (or highly similar) functionality.

A crack in the ortholog conjecture was formed in study published late 2009 in a paper published by Romain A. Studer and Marc Robinson-Rechavi. I blogged then about their study:

Romain A. Studer and Marc Robinson-Rechavi challenge common wisdom by publishing a study that says: “it ain’t necessarily so”. They look at three alternative models of molecular function evolution: (i) subfunctionalization after duplication; (ii) neofunctionalization after duplication; and (iii) the ‘alternative model’ of equal change after duplication or speciation. Subfunctionalization holds that after duplication, each of the two copies of the gene performs only a subset of the functions of the ancestral single copy. Neofunctionalization holds that one of the two genes possesses a new, selectively beneficial function that was absent in the population before the duplication. The ‘alternative model’ states that the gain of new function is not preferential to paralogs and that orthologs may gain new functions at the same rate that paralogs do.

Studer and Robinson-Rechavi claim that few studies have been made to study the scope of any of these proposed models. They then lay out study designs for doing so, challenging other evolutionary biologists (and themselves?) to conduct these studies and examine whether the common wisdom that orthologs maintain function while paralogs gain function. What I like about this paper is that it not only makes a strong case for challenging conventional wisdom, it also lays out a series of possible routes of study to be taken up by others.

Now two studies have widened this crack to a rather large crevasse. The first is a study by scientists in Indiana University. In a way, this new publication is a response to Studer & Robinson-Rechavi’s call to arms on points (i) and (ii). The IU scientists (the Radivojac lab and the Hahn lab at the School of Informatics at Indiana University, Bloomington, IN) examined hundreds of pairs of orthologous and paralogous genes from the mouse and human genomes. They then examined whether paralogs had a higher functional similarity, or rather orthologs.  What they found certainly defied the ortholog conjecture:

The relationship between functional similarity and sequence identity for human-mouse orthologs (red) and all paralogs (blue). (A) Biological pathway (B) molecular function. From PLoS Comput Biol 7(6): e1002073 under CC licence.

But before we explain the results, a word about function. The function of a protein has several aspects which are context-dependent; two important ones are the molecular function of the protein, and the biological process in which it participates. For example, the molecular function of all hemoglobins  is noted as  oxygen binding and oxygen transport. However, they are different in the processes, or pathways, in which they participate: gamma-hemoglobin participates in the transport of oxygen in the fetus. The complex which contains gamma-hemoglobin has a higher affinity to oxygen, and thus able to extract oxygen in the placenta from the maternal oxygenated hemoglobin and transport it to the fetus.

Now we can explain the figure above. Graph (A) above shows the functional similarity for the biological pathway aspect and how it is affected by the sequence identities of the hundreds of orthologs (red) and paralogs (blue) examined between human and mouse. Graph (B) shows the functional similarity of the molecular function aspect.

The X-axis is the sequence identity percentage between any pair of sequences: the higher the percent identity, the less divergent are the sequences, the more inclined we should be to think that the pair of proteins performs the same function since they diverged less. The Y-axis shows the fraction of functional similarity. Looking at graph (B) above, we see that paralogs which are 100% identical, have (almost always) the same function . But sequences of orthlogous proteins between human and mouse have only about 65% functional similarity, on average. What does that mean? In the database they looked at, each gene has a set of words associated with it, describing what it does. The IU scientists found that only about 65% of the keywords in orthologous sequence pairs overlapped, on average. Whereas for paralogs 100% overlapped. And those are for sequences which are identical! This means that even if we find identical protein sequences in human and in mouse, it does not mean that they have the same molecular function. On the other hand, paralogs, will generally have more similar functions. So the ortholog conjecture has been stood on its head here: paralogs are the ones that would generally have the same function, whereas orthologs diverge more in function. This holds true for up to about 50% sequence identity, when the picture seems to reverse itself. Graph (A) depicts the differences in the biological pathway aspect. Here, the differences are even more striking. The paralogs which are 90-100% identical between human and mouse participate in almost exactly the same pathways in both organisms. But orthologous proteins which are 90-100% identical the functional similarity is much lower:  only about 65%.

So what does this all mean?

First, it means that, at least between human and mouse, paralogs are better predictors of function than orthologs. And why would that be? To answer this question, let’s look closer at the graphs above. Note that while for paralogs the functional similarity decreases rapidly with sequence similarity, for orthologs the functional similarity remains roughly the same no matter how similar or different the orthologs are to each other, and even when they are 100% identical their functions vary to some extent! The reason: the experimental study of function in two human and in mouse  takes place in different contexts. The species-specific context is what causes the differences in annotation, and in the overall function. Also, all the orthologs in the study are of the same age, dating back to the human-mouse lineage split 75 million years ago. The paralogs predate that split, and may be of different ages: the split may predate the human / mouse split by 10 million years, 100 million years, or 1 billion years. Thus orthologs, regardless of their actual sequence similarity, have the same age, and paralogs do not. But why should proteins of the same age share the same level of (not so high) functional similarity? The authors of the study reply:

While there is no direct role for “time” in evolution that is not tied to mutation, we suggest that what time represents here is the evolution of the cellular context: the sum of the evolutionary changes over all of the directly and indirectly interacting molecules. If this context evolves at a steady rate (i.e. the average amount of functional change among all of the interacting molecules remains relatively constant), then protein function will appear to evolve at a steady rate, a rate largely disconnected from the level of an individual protein’s sequence divergence. — PLoS Comput Biol, Vol. 7, No. 6.

The strongest evidence they find for this hypothesis, is that even proteins with 100% are annotated differently. To wit:

For example, Liao and Zhang [50] found that >20% of genes that are essential for viability in humans are not essential in mouse. It is unlikely that changes to the proteins themselves have made them essential or not, but rather that their context in cellular and organismal networks has evolved. –ibid.

The proteins may not have changed substantially, but their environment changed, giving them a different role. Think about changing jobs after moving to a new place where there is no employer providing your exact old job you were used to. You may have been an embedded systems programmer, but now you are a website programmer. So context goes a long way to explain changes in ortholog function.

Interestingly, about a month after the IU paper was published, another paper from the Robinson-Rechavi lab was published, which also talks about homologs between human and mouse. In this study Gharib and Robinson-Rechavi reviewed previous literature listing several types of functional divergence of orthologs between human and mouse. They had some additional findings. For example, about 11% of the orthologous genes were alternatively spliced, meaning that the end products, proteins, were different between human and mouse.  They also listed specific phenotypic effects: genes which are linked to diseases in humans, but mutations in their mouse orthologs have no effects on mice. They cite studies that found that over 20% of genes which are essential in human are non-essential in mice (an essential gene is just that: if the organism does not have it, or it is mutated, the effects are fatal, and the organism does not develop past very early stages).  Their literature review concluded that 10-20% of ortholog pairs between human and mouse cannot be used for functional transfer. The IU study implies a higher percentage. Both studies conclude that a common practice in molecular evolution studies, the use of orthologs to infer function, should be seriously looked at.

(Full disclosure: Dr. Radivojac & I are collaborators, although our collaboration is unrelated to this study).

Nehrt, N., Clark, W., Radivojac, P., & Hahn, M. (2011). Testing the Ortholog Conjecture with Comparative Functional Genomic Data from Mammals PLoS Computational Biology, 7 (6) DOI: 10.1371/journal.pcbi.1002073

Gharib, W., & Robinson-Rechavi, M. (2011). When orthologs diverge between human and mouse Briefings in Bioinformatics DOI: 10.1093/bib/bbr031

Fitch, W. (1970). Distinguishing Homologous from Analogous Proteins Systematic Zoology, 19 (2) DOI: 10.2307/2412448

Why am I harping on this again?  Because the term “low homology” managed to sneak itself, of all places, into the title of a paper published in Bioinformatics.  Ouch. Bioinformaticians should know better.

Just for kicks, I decided to look at how many papers were published this year (January 1 through today)  using the misuse of terms in their title or abstract. Here are the results:

• “high homology”  134
• “low homology”: 13 (well, that’s low)
• “highly homologous”: 140
• “distant homologs”: 7
• “close homologs”: 7
• “percent homology”: 1

I could not find others such as “weak homologs”, “strong homologs”. Small mercies. Well, there is some work to do still in removing bad habits.

But while I gripe, someone is actually doing something about the whole terminology muddle. Specifically, Marc Robinson-Rechavi and his group in The University of Lausanne have created an ontology for describing the “relation between biological objects which resemble or are related to each other sufficiently to warrant a comparison“.

An ontology is a formal representation of concepts and the relationships between them.  It is usually hierarchical, with the terms going from the general to the specific. You may be familiar with the Gene Ontology as  standard representation of the different function of genes.

Example of the Biological Process ontology in the Gene Ontology

Marc’s group is creating an ontology for describing biological similarities in a hierarchical fashion, going from the general to the specific. At the top they have “similarity”. The four terms under that are “homology”, “homoplasy”, “functional equivalence” and “homocracy”.

Homocracy is a term suggested in 2003 by Claus Nielsen and Pedro Martinez for describing organs/structures which are organised through the expression of identical patterning genes. The rationale being that many homologous organs may be homocratic, but some homocratic organs may not be homologous.  Homoplasy means similarity due to convergent evolution, but not due to common ancestry. Fins on a tuna and a dolphin are homoplasic, but not homologous. However, the  fore fins on a dolphin are homologous to our arms, being descended from the forelimbs of the common ancestor of humans and dolphins.

The deepest annotated branch is homology, and going into the whole thing here would be long and arduous. But it is very well-crafted ontology. You can play around with the HOM ontology to see more of the terms, and also see their annotations at the OBO foundry.

Top terms of the HOM ontology. You can explore more on http://keg.cs.uvic.ca/ncbo/flexviz/FlexoViz.html#

Now, if someone could sort the terminology muddle between the different dialects of the English language…

Peter (watching Cricket on British TV): What the hell is he talking about?
Englishman: Oh, it’s Cricket. Marvelous game, really. You see, the bowler hurls the ball toward the batter who tries to play away a fine leg. He endeavors to score by dashing between the creases, provided the wicket keeper hasn’t whipped his bails off, of course.
Peter: Anybody get that?
Cleveland: The only British idiom I know is that “fag” means “cigarette.”
Peter: Well, someone tell this “cigarette” to shut up.

Source TV Guide courtesy Fox

Of course nothing will change, since I am too late in the game, no one listens to me anyway and I do not see the six readers of this blog rallying to eradicate microhomology.

No I am not bitter. Mild and bitter perhaps, but only after 5 o’clock.

(Thanks to F.B.  for the inspiration).

Sigh… people don’t seem to learn. It’s been almost 22 years (yikes!) since a distinguished group of scientists published a letter in Cell calling for a responsible use of the word “homology”. If you were born when that letter was published, then in the US you can already drink legally. And you may very well want to, by the time you finish reading this post.

As of today there are one hundred and sixty seven articles listed  in PubMed with the phrases “distant homology” or “remote homology” in either the title or the abstract.

Please: make it stop.

Homology is a qualitative term.  It means having a common evolutionary origin. Two genes / proteins / organs are either homologous, or they are not. They cannot be “somewhat homologous” or “partially homologous” or (a favorite among molecular and structural biologists) “distantly / remotely homologous”.

Homology is inferred from similarity.  Similarity is quantitative. If organs are sufficiently similar, like mammalian forelimbs, then they are considered to be homologous. They maybe more similar (like the hands of humans and chimpanzees), or less similar (like human hand and a bat wing). Nevertheless, once they pass a certain similarity threshold, homology is inferred. The same applies to sequences of proteins and nucleic acids.  Similarity can be measured. Different degrees of similarities can be compared and scaled.

If two protein sequences are aligned, and 40% of the amino acids in the alignment are identical, then the two sequences have a 40% identity. The do not have a 40% homology. They are  homologous, and the homology is inferred from the similarity.  We observe that the two sequences are similar, and then we conclude that they are homologous. We use the sequence similarity, as measured by percent identity, to trace a line of common descent for those proteins we deem homologous.

(As an aside I should say that the percentage of sequence identity, or %ID is not a very good measure for inferring homology, nor is it for measuring similarity. It is an easy one to use: but it is very coarse and prone to errors. There are many better measures out there, including statistical ones like e-values, p-values or information theoretic ones like bit scores. But I digress, and this is a matter for another post.)

But once we confuse observations with conclusions, things quickly become an impossible muddle.

Am I not not just picking nits here? I mean, surely when the term “distant homology” comes up in a paper or in conversation, we all know the meaning. Distant homology means having a common evolutionary origin,  but with a common ancestor that was around a long time ago. “Distant homology” is intuitive, brief yet understandable. it is less cumbersome than: “homologous, with a distant common ancestor, as concluded form a low yet statistically significant similarity” which is what we really should say if we properly separate observations from conclusions, as captain nitpick would have us do.

Allow me to answer with two examples.  First, I have read several papers discussing “structural homology”  in the context of protein structure. Those papers that discuss structural homology were actually using a verbal shortcut for  a homology inferred from structural similarity. That is, they inferred common descent from protein structural similarity. This kind of inference is highly contentious, and while not necessarily wrong, must be done with great care and proper caveats. However, once the researchers rolled up observations with conclusions by using the “structural homology” verbal shortcut, they absolved themselves from convincing the reader that structural similarity is indeed a good measure of homology, and jumped directly to the conclusion that there is indeed an homology here. The framework for inferring homology from sequence similarity is well worked out, but not so for structure, yet.   Therefore, even if we do use the verbal shortcut “distant homology”, we can only use it by virtue of having a certain measure of similarity well-established already, as in sequence based similarity. If it is not well established, and in using structural similarities, we fail to go through the proper scientific channels that consist of providing convincing observations prior to providing conclusions.

Second: even worse is the use of the term “functional homology”. This is a clear case of the word homology used as a drop-in synonym for similarity. The misnomer “functional homology”  is typically used in studies where proteins that are clearly not homologous perform similar functions. Why infer evolutionary descent when clearly that was not intended in the first place? Well, once you start confusing similarity with homology, observations with conclusions, and make them synonymous, this is what happens.

So don’t even start this confusion.  Separate observations from conclusions, and make the former support the latter. Homology is qualitative, similarity is quantitative.  Genes cannot be distantly homologous any more than a woman can be a little pregnant.

Now you can have that drink. Unless you are a little pregnant.

Gerald R. Reeck, Christoph de Haëna, David C. Teller, Russell F. Doolittle, Walter M. Fitch, Richard E. Dickerson, Pierre Chambon, Andrew D. McLachlan, Emanuel Margoliash, Thomas H. Jukes and Emile Zuckerkandl (1987). “Homology” in proteins and nucleic acids: A terminology muddle and a way out of it Cell, 50 (5) DOI: 10.1016/0092-8674(87)90322-9