Neutrality and selection on gene expression

There is a good case to be made that distinguishing neutrality from selection is now the central problem of molecular evolutionary biology. I don't intend to make the case, but I do want to discuss the problem. It arises for me because of a recent discussion of human-chimpanzee differences in gene expression, in a Science paper by Philipp Khaitovich and collaborators.

Reading this paper and some of its references has made me realize that one of the key aspects of the problem is that evolutionary biologists and molecular biologists often don't speak the same language. Sometimes the two groups may use exactly the same terms to mean different things --- yet, because they are mostly words borrowed from English, the meanings are similar enough to cause immense confusion.

These are my notes on gene expression differences between humans and chimpanzees. My focus is in pointing out things that might confuse, and attempting to determine the importance of the work to the project of uncovering the events and processes of human evolution. In that spirit, this is not a critique in any way, although I do include some critical comments; they are indications of the way this work differs from other kinds of evolutionary biology.

So, what do I mean when I say the meanings of words appear to be different? Consider the title of the paper:

Parallel Patterns of Evolution in the Genomes and Transcriptomes of Humans and Chimpanzees

This seems quite clear to an evolutionary biologist: humans and chimpanzees both evolved in a common direction from an ancestor that was different from both of them. At least, that's the usual meaning of "parallel evolution".

But in fact, the abstract makes clear that something very different is meant by the term "parallel" here:

The determination of the chimpanzee genome sequence provides a means to study both structural and functional aspects of the evolution of the human genome. Here we compare humans and chimpanzees with respect to differences in expression levels and protein-coding sequences for genes active in brain, heart, liver, kidney, and testis. We find that the patterns of differences in gene expression and gene sequences are markedly similar. In particular, there is a gradation of selective constraints among the tissues so that the brain shows the least differences between the species whereas liver shows the most. Furthermore, expression levels as well as amino acid sequences of genes active in more tissues have diverged less between the species than have genes active in fewer tissues. In general, these patterns are consistent with a model of neutral evolution with negative selection. However, for X-chromosomal genes expressed in testis, patterns suggestive of positive selection on sequence changes as well as expression changes are seen. Furthermore, although genes expressed in the brain have changed less than have genes expressed in other tissues, in agreement with previous work we find that genes active in brain have accumulated more changes on the human than on the chimpanzee lineage.

Now I'm confused. There is no evidence for parallel evolution here; rather, the expression profiles in the two species would seem to be parsimoniously explained by homology. In other words, there is nothing to suggest that the common ancestor of humans and chimpanzees had different expression levels in those tissues. Scanning the article, the word "parallel" is later used as a synonym for "similar"; and "the parallelism between sequence evolution and expression evolution" just means that both genetic divergence and expression differences depend on the tissue where genes are expressed. So "parallel patterns of evolution" is not about common pathways of selection, but instead a sort of tissue-dependent rate heterogeneity.

If that were the only thing confusing me, I wouldn't bother writing about it. But reading this stuff, I'm having constant Inigo Montoya moments: "You keep using that word. I do not think it means what you think it means."

Here's a passage from the conclusion:

In summary, we find that the patterns of evolutionary change in gene expression are largely compatible with a neutral model, in which different levels of constraints acting in different tissues add up for single genes (Khaitovich et al. 2005: 1853).

Of course, usually a neutral model is one in which there aren't any constraints...since the source of these constraints is selection.

But here, the "neutral model" is applied only to the changes in gene expression that did happen, not the changes that didn't. In a way, this distinction is analogous to Ohta's usage of "nearly-neutral" evolution: Lots of slightly deleterious changes in gene expression may have been precluded by selection, and only the truly neutral ones actually happened.

That's fair enough as it goes. And Khaitovich et al. (2004) make clear that the neutral model for gene expression is intended as a null hypothesis. It gives clear predictions for the differences we ought to expect, and in fact the differences that we observe do fit those predictions. So the burden is on an alternative explanation to explain the data better than drift. Until such an alternative surfaces, we are perfectly justified to say that humans and chimpanzees are different in gene expression mainly because of neutral evolution.

But we should recognize that it is a very particular interpretation of "neutrality": one that gives a substantial role to selection. This may help explain another bit of confusion; why Khaitovich et al. (2004) include this:

In fact, even at the level of morphology, it has been argued that many features are not adaptive, but instead result from physical constraints or historical accidents (Gould and Lewontin 1979) (Khaitovich et al. 2004:e132).

This is a citation to "The spandrels of San Marco", but one that is misapplied. Gould and Lewontin's (1979) arguments about the limits of adaptation concern the proper definition of adaptations and the logic for inferring past selection. In their paper, they consider ways that features of organisms may actually reflect selection on related features or structures, elements of architectural necessity, or genetic drift. But Gould and Lewontin did not give any examples of characters that evolved by drift, nor even mention the role of drift directly on morphological characters. Their most interesting argument is that for polygenic traits the level of selection per gene may be so slight as to allow many nonoptimal alleles to be fixed. The "physical constraints or historical accidents" parts of the paper may accord with the spirit of the gene expression work, but they are not about neutrality or genetic drift -- they are about interpreting the proper role of selection.

The key is that many things from other fields may look alike, or seem to be analogous, but that doesn't mean that they are alike or analogous. That's why it is important to use terms precisely, because sometimes they don't mean what you think they mean.

Finding selective constraints

Delving into the papers, it is possible to find some facts about gene expression in humans and chimpanzees that may help clarify the role of expression in the evolution of human characteristics. But the statistical test of neutrality makes clear that the goals of the molecular biologist and the evolutionary biologist are often very different from each other. Both may be interested in evidence of constraints, and these are not too problematic to identify:

Our analyses show that each tissue is associated with a certain level of evolutionary constraints acting on the genes expressed in it -- for instance, brain imposes more constraints than liver. These constraints add up across tissues so that genes expressed in many tissues are subject to more constraints than are genes expressed in few tissues. The signatures of these constraints are seen both at the level of DNA sequence differences and at the level of expression differences (Khaitovich et al. 2005:1851).

No problem: the evolutionary change in both coding sequences and gene expression depends on what kind of tissue the genes are expressed in. The pattern of selection on genes acting in the brain has kept them more similar between chimpanzees and humans than genes that are active in the liver. This is true to an even greater extent for genes expressed in many tissues. These genes are much more similar between humans and chimpanzees than are genes expressed only in the liver.

Notice, I said "pattern of selection" instead of "level of evolutionary constraints". I think this distinction makes a bit of difference when considering the passage immediately following the last one:

We have recently suggested that the evolution of gene expression patterns largely conforms to the predictions of a neutral model of evolution (23), i.e., that most expression differences observed within and between species are selectively neutral or nearly neutral. Because most evolutionary changes in nucleotide sequences conform to a neutral theory (24), the parallelism between sequence evolution and expression evolution observed here supports the notion that most evolutionary changes in gene expression are similarly selectively neutral or nearly neutral (23) (ibid., references in original).

This doesn't follow. They have found that the pattern of selection is similar in humans and chimpanzees. They conclude that changes separating humans and chimpanzees are therefore not the result of selection. Again, I'm confused.

Citation number 23 from the passage above is a previous paper by Khaitovich and colleagues (2004). In it, the authors demonstrate that much evolution in gene expression does fit the expectations of neutral evolution. The tests they applied are listed in the abstract to the paper:

(1) expression differences between species accumulate approximately linearly with time; (2) gene expression variation among individuals within a species correlates positively with expression divergence between species; (3) rates of expression divergence between species do not differ significantly between intact genes and expressed pseudogenes; (4) expression differences between brain regions within a species have accumulated approximately linearly with time since these regions emerged during evolution. These results suggest that the majority of expression differences observed between species are selectively neutral or nearly neutral and likely to be of little or no functional significance (Khaitovich et al. 2004:e132).

Khaitovich et al. (2004) show all of these propositions to be approximately true among hominoids.

What does this mean? The motivation for the neutral theory was the discovery of abundant electrophoretic variation in natural populations. This appeared to contradict the established principles of genetics, which had been based on the assumption that most natural populations are monomorphic for a "wild-type" allele, and that polymorphism is rare. In contrast, electrophoretic studies found that polymorphism is ubiquitous. Under the assumption that this polymorphism is not reflected at the level of morphology or behavior, the neutral hypothesis proposed that the polymorphism has no phenotypic importance, and therefore no possible correlation with fitness.

Applied to gene expression, we might make a prediction: that the variability in gene expression (i.e. polymorphism) is great, and that many changes in gene expression have no fitness effect. The actual extent of gene expression differences in natural populations (of humans and chimpanzees) is not obvious in these studies -- for the most part, differences within and between these species is concealed within cluster diagrams, which are based on distance formulae.

But if we look at the diagram presented in Khaitovich et al. (2005:1851), a conclusion emerges:

Figure 1 from Khaitovich et al. (2005:1851)

The study concludes that genes expressed in testis were likely under positive selection because of the large between-species divergence compared to the small within-species divergence.

But notice that only in this case does the ratio of within-species to between-species divergence look anything like that for gene sequences! For the other tissues, the between-species divergence is incredibly short compared to the amount of divergence between individuals within species. In contrast, most gene sequences show around a tenfold greater branch length between humans and chimpanzees than within either species.

This problem is discussed by Khaitovich et al. (2004):

The fact that the overall accumulation of expression differences conforms to a selectively neutral model does not mean, of course, that all expression differences between species are selectively neutral. As for nucleotide changes, some changes in gene expression will have had phenotypic consequences and some of these will have become fixed due to positive selection. To identify such gene expression differences, we propose to use the ratio of divergence between species to diversity within species, akin to the tests suggested for quantitative genetic traits (Charlesworth 1984; Lynch and Hill 1986; Turelli et al. 1988) and in agreement with recent suggestions by Rifkin et al. (2003) or Hsieh et al. (2003). However, to do this it is necessary for each gene considered to distinguish the gene expression diversity caused by genetic differences between individuals from the diversity caused by environmental factors. This is crucial since the environmental component is likely to be much larger than the genetic component. For example, under strict neutrality and no environmental influence, we expect a divergence to diversity ratio that is equal to the ratio of time of divergence of the species to the average time to the common ancestors of the individuals sampled within a species. This would be about 1:10 for humans and chimpanzees (Chen and Li 2001; Lander et al. 2001). However, the observed ratio is approximately 1:3, suggesting that the environmental component is on the order of three times bigger than the genetic component. Studies of gene expression differences among individuals with different genetic relatedness will eventually allow an estimation of the genetic component of expression variation.

Khaitovich et al. (2004) attempted to use pseudogene expression as a neutral test of the ratio of between-species divergence to within-species diversity. But this is clearly an error, since the expression of these pseudogenes must depend on other factors (e.g. promoters and inhibitors, which are coded by active genes). It makes no difference that the expressed pseudogenes themselves do not affect fitness; whatever controls their expression may well affect many other (non-neutral) things, including the expression of active genes. Thus, pseudogene regulation might well be expected to be very much like the expression of active genes -- if indeed the pseudogenes' expression were controlled by unique-to-pseudogene promoters, then we might not expect them to still be expressed at all.

This improper use of pseudogene comparisons may really have led them astray on the interpretation of testis-specific expression changes. If the ratio of between-species to within-species divergence is closer to 10:1, then it might mean that the environmental influence on testis-expressed genes is lower.

In any event, it would appear that the extent of gene expression differences within species is very great indeed. The similarity in between-to-within species divergence ratios among tissue types shows that the pattern of selection among tissue types may be similar. But the relatively low between-species divergence shows that far fewer gene expression changes become fixed within species than occur within species. Some of this difference may be the consequence of environmental difference --- but it should be noted that environments vary between species as well as within them. From the great environmental differences between humans and chimpanzees, we might well expect between-species gene expression differences to be disproportionately great, not disproportionately minor. So a better explanation would appear to be strong selective constraints on most within-species variants.

Of course, many of the differences between humans and chimpanzees in gene expression may have no fitness consequences. If this number of neutral changes is very large, in fact, it will make it very difficult to find evidence for adaptive changes in gene expression.

Does a long interspecies branch length indicate positive selection? If we could control for environmental effects, it might. But this would be a very course test of selection --- it would require that the number of positively selected (i.e. adaptive) changes be very great -- perhaps more than the number of neutral changes. Here, we are looking at the expression levels of all genes (or at least a substantial proportion of them) within a tissue. Even if the tissue underwent substantial changes during human evolution, it is probable that these changes involved a relatively small proportion of all the genes that are active in a tissue.

How do gene expression differences relate?

At this point, it's worth thinking about what we aim to explain. I am interested mainly in three things: how selection caused the evolution of humans from an ape ancestor; whether the absolute number of adaptive changes in the human genome was high enough to substantially affect nucleotide diversity; and the extent to which human anatomical and behavioral evolution may have depended upon regulatory changes versus protein structural changes.

As far as the number of selected changes in expression, the proportion of selected changes is not that informative. If the total number of changes (selected and neutral) changes is very large -- as it appears to be -- then a very small proportion of expression changes would still be a substantial number of changes. And it still remains possible that a relatively high proportion of differences between species were selected, even if there is a high level of within-species diversity, because of the unknown role of environment. And the same gene might easily have been under selection for its expression multiple times, or different times in different tissues. So there is far to go to figure out the role of adaptive change in expression on overall gene diversity.

The extensive diversity in gene expression within species is very interesting from the perspective of regulatory vs. protein evolution. It may be that there is more extensive diversity among people in gene expression than in protein structure, and the fact that chimpanzees show a similar pattern may suggest that it is generally true. The role of selection depends on how heritable such variation in expression actually is, but considering the extensive variation and its continuous nature, it may be that gene regulation forms a much more facile substrate for adaptive evolution than does amino acid sequence change.

One question is whether gene expression differences say very much about what we are interested in explaining about human evolution. For example, there is this passage from Khaitovich et al. (2004), concerning expression differences in different regions of the brain:

Our data show that tissues that diverged recently have very similar gene expression profiles irrespective of the differences in function. For instance, the transcriptome of Brodmann's area 44 in the left hemisphere (Broca's area) is very similar to that of the prefrontal cortex in both humans and chimpanzees, although it is known to be involved in speech processing in humans while it must have another function in chimpanzees (Kandel et al. 2000). This is what we would expect if the time since divergence rather than the extent of functional differences determined the magnitude of transcriptome change. Thus, although a number of expression differences between brain regions surely correspond to functional differences, our findings suggest that a sizeable proportion of the differences are functionally neutral.

Of course, they are not saying that there are no differences in gene expression between humans and chimpanzees in Broca's area; they are merely saying that the scale of differences in this area does not differ from other areas of the brain. So there clearly could be differences in gene expression that have functional importance in this region.

But what if there weren't? A real possibility is that the important differences between humans and chimpanzees lie in the circuitry of this region, and not in the function of the neurons themselves. Indeed, the expression profiles of the neurons might be entirely identical for all we know, and the key differences might lie in the embryology of the developing neural circuits. These embryological differences themselves would be the product of differences in gene expression, but only at a particular stage of ontogeny.

Clearly we need more than a one-dimensional account of expression differences. The evolutionary differences between humans and chimpanzees are determined by gene interactions that have a time component. What's worse, these depend on the interactions among developing (and differentiating) tissues, so that the in vivo differences in expression may not be easily modeled with in vitro methods.

All this is to say we still have a lot to learn about gene expression in human evolution. Also, it is clear that different kinds of biologists need to read more of each other's work. The lack of familiarity with the use of common words really has the potential to lead to confusion. Fortunately, that sounds the same in all kinds of biology.


Khaitovich P, Hellmann I, Enard W, Nowick K, Leinweber M, Franz H, Weiss G, Lachmann M, Pääbo S. 2005. Parallel patterns of evolution in the genomes and transcriptomes of humans and chimpanzees. Science 309:1850-1854. Full text online

Khaitovich P et al. 2004. A neutral model of transcriptome evolution. PLoS Biol 2:e132. Full text (free)