john hawks weblog

Tomislav Maricic and colleagues from Svante Pääbo's group have reported finding a regulatory change in the gene FOXP2 that may be of relevance to the evolution of human speech [1]. To telegraph the conclusion, the paper does not demonstrate that Neandertals or Denisovans were different from humans in speech or language-relevant phenotypes.

Most important, a substantial number of living people share the ancestral genotype inferred for Neandertals and Denisovans for the site considered in the study. It is a genetic change within living people that may have been important, but it is an instance where human variation includes the Neandertal genotype.

I'm going to let the paper's mini-review do the work of describing the background to the study:

Among humans, sequence variation around exon 7 shows an excess of derived nucleotide variants at high frequencies and of rare nucleotide variants, indicating that the region has been affected by a selective sweep (Enard et al. 2002; Zhang et al. 2002; Yu et al. 2009). It has been estimated that this happened within the last 200,000 years (Enard et al. 2002) or 55,000 years (Coop et al. 2008). Because it was initially assumed that at least one of the two amino acid substitutions were the cause of the sweep, it was expected that at least one of them would not be present in Neandertals, who shared a common ancestor with modern humans 370–450,000 years ago (Green et al. 2010). However, both nucleotide substitutions were found in two Neandertals from Spain (Krause et al. 2007) as well as in Neandertals from Croatia (Green et al. 2010), and in Denisovans, an extinct Asian hominin group related to Neandertals (Reich et al. 2010). Furthermore, it was found that linkage disequilibrium extends across exon 7 in present-day humans, which is not expected if one of the two amino acid changes in exon 7 was the target of selection (Ptak et al. 2009). Hence, although at least one of the two amino acid changes is very likely evolutionarily relevant given the functional data and the conservation of FOXP2, they are not likely to be the cause for the selective sweep. Assuming that a sweep did occur, it must therefore be caused by some other variant in the region, possibly affecting the regulation or splicing of FOXP2.

It was a big story that humans had a recent sweep in this gene, eliminating most of the variation, and that humans are different from other primates in the coding sequence. But the apparent timing of the sweep did not make sense in combination with the observation that Neandertals share the human coding sequence.

One resolution of these observations is the hypothesis that the human version of FOXP2 simply came from Neandertals. I wrote about a short paper by Graham Coop and colleagues in 2008 that went along similar lines ("FOXP2 is really recent, it really did introgress (if it's not contamination)"). Coop and colleagues substantiated the hypothesis of a recent selective sweep, but at the same time they did acknowledge that selection on some other linked locus might account for the evidence.

Maricic and colleagues have found another linked genetic change that could account for the sweep. In their scenario, the sweep was only the most recent of possibly several changes under selection to this gene. This most recent one involved a regulatory change within exon 7 of the gene that did not affect the coding sequence at all.

The sequence analysis carried out by Maricic and colleagues is very straightforward. They simply resequenced the gene region from Neandertal specimens to get a list of sites where Neandertals and Denisovans do not carry a derived human variant, and then resequenced the gene in 50 humans to see how many of the derived human mutations are high-frequency. The one they identify is both high-frequency and affects a candidate regulatory site. The site is a binding site for the transcription factor POU3F2. The rest of the paper documents their attempts to demonstrate an effect of this site on gene regulation in tissue culture. They conclude:

The transcription factor POU3F2 is expressed exclusively in the central nervous system (Schreiber et al. 1993), more specifically in postmitotic neurons and glia (Hagino-Yamagishi et al. 1997). Within the central nervous system, FOXP2 is expressed in postmitotic neurons (Ferland et al. 2003). Thus, it is reasonable to assume that POU3F2 regulates expression of FOXP2 in neurons. It is furthermore interesting that position 114076877 is located at the point in intron 8 of the FOXP2 gene where the pattern of allele frequencies among humans indicates that a functional change occurred that could be responsible for a positive selective sweep affecting the FOXP2 gene during the last 50,000 years (Coop et al. 2008). It is noteworthy that this is the only nucleotide variant in that region where the majority of present-day people carry a derived variant that is not present in Neandertals and Denisovans. Thus, it is possible that this change was positively selected recently during the evolution of fully modern humans.

However, the ancestral allele shared by Neandertals and Denisovans is also fairly common in some human populations today. As Maricic and colleagues conclude, the obvious thing to do is look at homozygote carriers of the allele to see if they're different from noncarriers:

The ancestral allele occurs at frequencies of ∼10% in some African populations (supplementary table S6, Supplementary Material online). Therefore, individuals homozygous for the ancestral allele can be expected to occur at a frequency of approximately 1% in the population. In such individuals, the phenotype of the ancestral allele should be observable even if is recessive to the derived allele. Further work will explore the phenotypes of such homozygous carriers of the ancestral allele and the consequences of the substitution at position 114076877 on FOXP2 transcription in model systems.

This all seems logical. We may not be able to say that Neandertals were just like us in FOXP2 -- but that's because we're not all alike. They're just like some of us.

The only thing I would add is that the number of humans covered by the study is still quite small. The paper examined only 50 individuals from the HGDP set; additionally they considered the 1000 Genomes data. It is interesting that the Neandertal-Denisovan ancestral allele at this site is not present in several of the samples outside Africa in the 1000 Genomes data, but it is present in two of the American samples, and in all the African samples. So although the region looks like it was positively selected at some point during the last 100,000 years or so, we still can't yet say that the ancestral allele carried by Neandertals was disadvantageous within later populations.

Larger samples would settle that question. In the meantime, this study does point the way toward a wider analysis of differences in gene regulation among archaic human genomes.

References

I was really pleased to see the new paper by Erik Axelsson and colleagues [1] on the pattern of recent selection on domesticated dogs. As we began working on recent selection in humans, we expected that domesticated animals might exhibit similar patterns genome-wide. They are among the organisms most similar to humans in demography and ecological change: Domesticated animals have all undergone rapid shifts in diet, predator ecology and social dynamics after domestication, at the same time that they have experienced rapid increases in population size. That is a recipe for rapid adaptive evolution.

As in humans, the paper shows that dogs were selected strongly for a new agricultural diet. Just as in humans who descend from early agriculturalists, dogs have extensive duplication of the amylase gene. Humans express amylase in saliva, but as explained in the paper dogs only produce amylase in the pancreas, where it digests starches intestinally. Where this paper gets really exciting is when the authors began to investigate the entire metabolic pathway underlying starch digestion. The amylase gene AMY2B underwent duplications similar to those in humans, and not found in wolves. Two other genes that interact in starch digestion and glucose uptake did not undergo duplication but do show near-fixed haplotypes in dogs that are absent or very rare in wolves, and the paper shows using both biochemistry and phylogenetic comparison with herbivores and omnivores that the dog versions of these genes increase enzymatic activity on starches and glucose uptake.

In conclusion, we have presented evidence that dog domestication was accompanied by selection at three genes with key roles in starch digestion: AMY2B, MGAM and SGLT1. Our results show that adaptations that allowed the early ancestors of modern dogs to thrive on a diet rich in starch, relative to the carnivorous diet of wolves, constituted a crucial step in early dog domestication. This may suggest that a change of ecological niche could have been the driving force behind the domestication process, and that scavenging in waste dumps near the increasingly common human settlements during the dawn of the agricultural revolution may have constituted this new niche6. In light of previous results describing the timing and location of dog domestication, our findings may suggest that the development of agriculture catalysed the domestication of dogs.

So for those of you wondering why we feed dogs kibble instead of raw beef, here's the reason.

After finding candidate regions for selection across the genome, the authors ran a gene ontology analysis to see whether functional gene loci in these regions fall into any consistent categories. Along with the metabolic and digestive genes, they found

The most conspicuous cluster (11 terms) relates to the term ‘nervous system development’. The eight genes belonging to this category (Supplementary Tables 7 and 8) include MBP, VWC2, SMO, TLX3, CYFIP1 and SH3GL2, of which several affect developmental signalling and synaptic strength and plasticity. We surveyed published literature and identified 11 additional CDR genes with central nervous system function (Supplementary Table 9), adding to a total of 19 CDRs that contain brain genes. These findings support the hypothesis that selection for altered behaviour was important during dog domestication and that mutations affecting developmental genes may underlie these changes7.

That is a similar story to humans. We don't know what such genes might do, and unraveling what difference these genes may have made to behavior will take a lot of additional understanding of developmental biology. Much easier to work out what is going on when you can examine the biochemistry in vitro as with starch enzymes.

The paper also makes clear why finding evidence of selection can be a difficult empirical problem at the moment:

Uniquely placed sequence reads from pooled DNA representing 12 wolves of worldwide distribution and 60 dogs from 14 diverse breeds (Supplementary Table 1) covered 91.6% and 94.6%, respectively, of the 2,385 megabases (Mb) of autosomal sequence in the CanFam 2.0 genome assembly11. The aligned coverage depth was 29.8× for all dog pools combined and 6.2× for the single wolf pool (Supplementary Table 1 and Supplementary Fig. 1). We identified 3,786,655 putative single nucleotide polymorphisms (SNPs) in the combined dog and wolf data, 1,770,909 (46.8%) of which were only segregating in the dog pools, whereas 140,818 (3.7%) were private to wolves (Supplementary Table 2). Similarly we detected 506,148 short indels and 26,619 copy-number variations (CNVs) (Supplementary Files 1 and 2). We were able to experimentally validate 113 out of 114 tested SNPs (Supplementary Table 3 and Supplementary Discussion, section 1).

If that sounds confusing, that's because it is confusing. Right now whole-genome sequencing is not yet routine, and whole-exome sequencing is not routine for creatures other than people. So maximizing the available data means working with partial genomes at varying levels of coverage, often accumulated for other purposes by other research groups using different sequencing platforms. Verifying sequence differences is not trivial. Generating a sample of gene sequences from many individuals is challenging, particularly as different individuals may be covered or not for different parts of their genomes.

Studying selection requires a fairly large sample of genomes. This paper establishes evidence of selection on a few things in which domesticated dogs are mostly the same, and all are different from wolves. In other words, these are "complete sweeps" or "near-complete sweeps", in which a new genetic variant has become mostly fixed within the domesticated dog sample. A larger sample of dogs would be able to test selection with a broader range of strength and initial date, including "partial sweeps" and selection on standing variation that may have already existed in ancestral wolves before being subject to selection in domesticated dogs. So this paper opens a new area of inquiry on the causes of domestication without ruling out that we will discover much, much more about the history of selection in dogs.

One really cool possibility is that we will uncover convergent or parallel patterns of selection in dogs with different geographic origins. Already we know that body size and pigmentation have been subject to selection in different dog breeds, and that single genes transferred across breeds have been important parts of that process. There are a few cases in humans where the extensive geographic dispersal of a single adaptive variant can explain the present distribution of a trait. But in many more cases, different human groups have attained traits by parallel selection on different genetic variants. Because humans control the breeding of dogs and traded dogs across long distances in historic times, we may find that dogs are much less affected by parallelism and much more by long-distance gene flow than humans. But we won't know until we put that hypothesis to the test.

References

Last week's issue of Science included a perspective piece by my UW colleagues Dominique Brossard and Dietram A. Scheufele, from Life Science Communication [1]. They focus on the impact of technology and internet communication on the public understanding of science.

People find information online today very differently from the way people used to find information, whether from the traditional printed press or in libraries. Information in broad, authoritative works such as encyclopedias, textbooks or indexes involved highly selective editing by humans, moderated by expert opinion. A reader looking in any printed encyclopedia would be likely to see the same basic facts and be directed to the same essential references.

Now, computer algorithms do much of this job by tracking what people choose to look at after they have searched for a topic or keyword. This changes the process of information discovery, and as Brossard and Scheufele discuss, may introduce feedbacks into the process with unpredictable effects:

[T]here are often clear discrepancies between what people search for online, which specific areas are suggested to them by search engines, and what people ultimately find. As a result, someone's initial question about a scientific topic, the search results offered by a search engine, and the algorithms that a search provider uses to tailor retrieved content to a search may all be linked in a self-reinforcing informational spiral in which search queries and the resulting Web traffic drive algorithms and vice versa (7). This raises an interesting paradox when it comes to relatively new scientific topics, such as nanotechnology, that are still unfamiliar to many people: Is the World Wide Web opening up a new world of easily accessible scientific information to lay audiences with just a few clicks? Or are we moving toward an online science communication environment in which knowledge gain and opinion formation are increasingly shaped by how search engines present results, direct traffic, and ultimately narrow our informational choices?

I encounter this problem here with my weblog. It is very difficult to design an effective presentation strategy for topic-specific searches on a website. It is also hard to maintain internal search capacity on a site the size of this one, with content that comprises both original text and bibliographic references. As you can tell by the fact that I frequently deactivate internal searching altogether, this has been a pain for me to develop and maintain.

The more newsworthy part of this essay is a reference to the effects of online comments after articles about science and technology topics. Brossard and Scheufele refer to a recent conference that covered this topic, and the results of a study in which subjects were exposed to the same story but with different types of comment sections:

Disturbingly, readers' interpretations of potential risks associated with the technology described in the news article differed significantly depending only on the tone of the manipulated reader comments posted with the story. Exposure to uncivil comments (which included name calling and other non–content-specific expressions of incivility) polarized the views among proponents and opponents of the technology with respect to its potential risks. In other words, just the tone of the comments following balanced science stories in Web 2.0 environments can significantly alter how audiences think about the technology itself.

Anyone who reads comments sections following news articles surely will have noticed the rotten wealth of trolls and other idiots who inhabit such forums. I thought about Brossard and Scheufele's piece again today when I read a post by Dan Conover at Xark: "Why I shut down comments". The post reflects on how blog communities have changed since the early days of blogging in 2005. This timeframe has coincided with the growth of social media of other types, such as Facebook and Twitter, which have given many people a closed community for sharing comments and perspectives with like-minded folks. Conover observes that the trolls and spam are more persistent, causing a rapid degradation of the value of comment sections of many blogs.

This isn't of course universal. Many blogs continue to have rich and varied comment sections with their posts, and some (like mine) never had any comments at all. What I find more interesting is this passage:

I believed then, as I believe now, that the ability to comment and share across horizontal, informal networks is the killer app for the 21st century.

Which sounds nice.

Unfortunately, newspaper and other traditional-media websites, for all their hand-wringing concerns about libel and civility circa 2005, are typically the worst offenders when it comes to building quality comment cultures. We've taught users bad habits and turned comment sections into troll ghettos.

Comments on professional news websites are almost always useless, misguided, or malevolent. Combine this with Brossard and Scheufele's claim that the tone of comment sections affects readers' comprehension of science and technology stories, and I propose a hypothesis: Professional news websites may be the worst way to communicate science, because their comment policies undercut science comprehension.

References