News on Population Genetics—Human Genome Research Shows Race Is Real

The human-genetics websites are abuzz over two papers in the July 6 issue of Science.  The papers discuss rare variants in the human genome, which highlight differences between big old localized populations.

Steve Hsu at infoproc:

Deep sequencing of the human genome, which reveals rare variants (here, defined as those found in fewer than 0.5 percent of the population), shows that there is actually more variation between groups than within groups. (So what you may have been taught in school is not true ─ sorry, that's how science works sometimes.) The figure below, from this July 6 Science article, shows that over 50 percent of rare genetic variants are found in African populations (which have greater genetic diversity) but not in European populations. About 41 percent of all rare variants are found only in Europeans and not in Africans, and only 9 percent of the variants are common to both groups.

Razib Khan at Gene Expression:

I suspect this is going to be a big deal for some time. For humans we are coming to toward the end of the SNP-age and entering into the whole-genome-age. That means that the emphasis on common variation at the genomic level is going to give way somewhat to rarer, more particular, variation. One of the major takeaways is that a lot of this variation is going to be population specific . . . If I read this right we may be entering into a golden age of demographic history reconstruction, as rare variants and whole-genome catalogs of a huge number of humans are going to allow us to generate a very fine-grained map of human population diversity.

One thing's for sure:  the continued survival of Lewontin's Fallacy ─ I still hear it all the time ─ is ever more inexplicable.

VanBUG: Andrew G Clark, Professor of Population Genetics, Cornell University

[My preamble to this talk is that I was fortunate enough to have had the opportunity to speak with Dr. Clark before the talk along with a group of students from the Bioinformatics Training Program.  Although asked to speak today on the subject of the 1000 genomes work that he's done, I was able to pose several questions to him, including "If you weren't talking about 1000 Genomes, what would would you have been speaking about instead?"  I have to admit, I had a very interesting tour of the chemistry of drosophila mating, parental specific gene expression in progeny and even some chicken expression.  Rarely has 45 minutes of science gone by so quickly.  Without further ado (and with great respect to Rodrigo Goya, who is speaking far too briefly - and at a ridiculous speed - on RNA-seq and alternative splicing  in cancer before Dr. Clark takes the stage), here are my notes. ]

Human population genomics with large sample size and full genome sequences

Talking about two projects – one sequencing a large number of genomes (1000 Genomes project), the other sequencing a very large number of samples in only 2 genes (Rare Variant studies).

The ability to predict phenotype from genotype is still small – where is the heritability?  Using simple snps is insufficient to figure out disease and heritibility.  Perhaps it’s rare variation that is responsible.  That launched the 1000 Genome project.

1000 Genome was looking to find stuff down to 1% of population.   (In accessible regions)

See Nature for pilot project publication of the 1000 Genomes project.. This included several trios (Parents and child).  Found more than 15M snps across the human genome.  Biggest impact, however, has been the impact on informatics – How do you deal with that large volume of snps?  Snp calling, alignment, codification, etc…

Much of the standard file formats, etc came from the 1000 Genomes groups working on that data. Biggest issue is (of course) to avoid mapping to the wrong reference!  “High quality mismatches” ->  Many false positives that failed to validate: misalignments of reads.  Read length improvements helped keep this down, as did using the insertions found in other 1000 Genome project subjects.

Tuning of snp callling made a big difference.  Process with validations made a significant impact.  However, for rare snps, it’s still hard to call snps.

Novel SNPs tend to be population specific.  Eg. Yoruban vs. European have different patterns of SNPs.  There is a core of common SNPs, but each has it’s own distribution of the rare or population specific SNPs.

“Imputation” using haplotype information (phasing) was a key item for making sense of the different sources of the data.

Great graph on fequency spectrum.  (Number of variants – log vs allele frequency (0.01 – 1)) Gives a lying out flat hockey stick.  Lots of very rare frequency snps, decreasing towards 1, but a spike at 1.

>100kb from each gene there is reduced variation (eg, Transcription start site.)

Some discussion of recombination hotspots, which were much better mapped by using the 1000 genome project data.

Another application: de novo mutation.  Identify where there are variations in the offspring where they are not found in either present.   Roughly about 1000 mutations per gamete.  ~3×10^-8 substitution per generation.

1000 Genomes project is now expanding to 2500 samples.  Trying to distribute across 25 population groups, with 100 individuals per group.

Well, what do we expect to discover from ultra-deep sampling?

There are >3000 mutations in dystrophin.  (Ascertained cases of muscular dystrophy. – Flanagan et al, 2009, Human Mutation)

If you think of any gene, you can expect to find every gene mutated at every point across every population… eventually.  [Actually, I do see this in most genes, but not all... some are hyper conserved, if I've interpreted it correctly.]

Major problem, tho: sequencing error.  If you’re sampling billions of base pairs, with 1/100,000 error rate, you’ll still find bad base calls!

Alex Coventry: There are only 6 types of heterozygotes (CG, CT, GT, AC, AG, AT)… ancient technology, not getting into it – was developed for sanger.

Studied HHEX and KCNJ11 genes, sequenced in 13,715 people. Validated by Barcoding and 454 sequencing.

Using the model from Alex’s work, you could use a posterior probabilty of each SNP.  Helped in validating.  When dealing with rare variants, there isn’t a lot of information.

The punchline: “There are a lot of rare SNPs out there!”

Some data shown (site frequency) as sample data increases.  The vast majority of what you get in the long run is the rare SNPs.

Human rare variation is “in excess” of what you’d expect from classical theory.  So why are there so many variants?

Historical population was small, but underwent a recent population explosion in the last 2000 years. This allows for a rapid diversity to be generated as each new generation has new variants, and no dramatic culls to force this rare variation to consolidate.

How many excess rare variants would you expect from the population explosion?  (Guttenkunst et al, 2009, PLOS Genetics)  Population has expanded 100x in about 100 generations.  Thus, we see the core set, which were present in the population before the explosion, followed by the rapid diversification explosion of rare snps.

You can do age inferrence, then, with the frequency of SNPs.  older snps must be present across more of the population.  Very few SNVs are older than 100 generations.  If you fit the population model back to the expected SNV frequency in100 generations ago, the current data fits very well.

When fitting to effective sample size of humans, you can see that we’re WAY out of equilibrium from what the common snps would suggest.  [I'm somewhat lost on this, actually.  Ne (parent) vs n (offspring).  I think the point is that we've not yet seen consolidation (coalescence?) of SNPs.]

“Theory of Multiple Mergers”  Essentially, we have a lot of branches that haven’t had the chance to blend – each node on the variation tree has a lot of unique traits (SNPs) independent of the ancestors.  (The bulk of the weight of the branch lengths is in the many many leaves at the tips of the trees.)

[If that didn't make sense, it's my fault - the talk is very clear, but I don't have the population genetics vocabulary to explain this on the fly.]

What proportion of SNPs found in each new full genome sequence do we expect to be novel? (For each human.)  “It’s a fairly large number.”  It’s about 5-7%, Outliers from ]3-17%.  [I see about the same for my database,  which is neat to confirm.]  Can fit this to models: constant population size would give a low fraction (0.1%), with explosive model (1.4%) over very large sample sizes.

Rare variants are enriched for non-synonymous and premature terminations (Marth et al , submitted) [Cool - not surprising, and very confounding if you don't take population frequency into account in your variant discovery.]

What does this mean in complex diseases?  Many of our diseases are going to be caused by rare variants, rather than common variants.  Analogy of jets that have 4x redundancy, versus humans with 2x redundancy at the genome level.

Conclusions:

• Human population has exploded, but it has a huge effect on rare variations.
• Huge samples must be sequenced to detect and test effects
• Will impact out studies of diseases, as we have to come to terms with the effects of the rare variations.

Out of Africa: Startling New Genetics of Human Origins

Western Pygmies

I love population genetics for its ability to peer back into human history through the medium of DNA’s ATCGs.

One of the stars of this discipline is Sarah Tishkoff, a standout in African genetics, someone who will readily haul a centrifuge into the bush in Cameroon.

Tishkoff of the University of Pennsylvania is lead author on a paper published online July 26 in Cell that details  whole-genome sequencing of five individuals each from three extant hunter-gatherer groups—the Pygmies of Cameroon as well as the Hadza and the Sandawe of Tanzania. The results reveal millions of newly discovered genetic variants—differences in single genetic letters, the ATCGs—and indicate that early modern humans may have interbred long ago in Africa with another species of hominid (although the fossil record does not provide much support for the latter finding).

Tishkoff answered a few questions for us about this paper, co-authored with Joseph Lachance and 11 other researchers. An edited version of the interview appears below:

Please describe the research that led to the paper that was published today:

We’re the first ones to look at these diverse groups of hunter-gathers in Africa who descend from some of the most ancestral lineages in the world. They’re  interesting because they have very unique and distinct lifestyles There are few  populations that maintain this active hunter-gatherer lifestyle.

This is the most extensive study in Africa using high-coverage deeply detailed sequence data. We focused on three groups because they’re  anthropologically interesting. They’re thought to be descended from groups that are ancestral to all modern humans. We wanted to understand the genetic basis of adaptation to their local environment  including, for instance the short stature trait in Pygmies.

So what did you find?

We discovered 13 million variants and, of those variants, greater than 3 million are completely novel, meaning that they have not been reported in any database. The current public database has 40 million variants. So we found 3 million novel variants by simply sequencing 15 individuals. That increases by about 8 percent all known human genetic variation. It also demonstrates that we’re missing a lot of really important variation that’s out there, particularly in Africa, which is the  homeland of modern humans and a place where there’s been a lot of time for differentiation to have occurred in very diverse environments. What this means is that there’s s probably a lot of regional or population-specific variation out there that has not been that well characterized, some of which is functionally very important.

What about natural selection?

Natural selection seems to be operating more on the non-coding genome [the regulatory portion that does not contain genes] than the coding region. A lot of people are doing exome sequencing [looking only at genes]. I think they’re missing a lot of important variation.

In our study, we looked at what regions of these groups’ genomes were uniquely differentiated to their local environments. There wasn’t a huge amount of overlap between the groups—or between them and other non-hunter-gatherer groups from Africa.  Due to natural selection, we found there were distinctive adaptations for  immunity, taste and smell.

In the Pygmies, we discovered genes involved with thermal regulation, immunity and stature, all likely to be adaptive to a tropical environment. We pinpointed genes related to pituitary and thyroid function, the latter perhaps an adaptation to a low-iodine environment.

In the Sandawe, we found a variant for  melanin, a gene involved in skin color.  The Sandawe are among the most fair-skinned groups in Africa. When I went to work with them, they said, ‘We’re like brothers and sisters because you look like us.’ This is not because of any European admixture; they look like the San [a hunter-gatherer group from southern Africa]. When I said: ‘Where do you come from, they pointed to a mountain in the distance. When I said ‘Can you take me there?” we went but there was no road. We went through the bush and they showed me cave paintings. Having lived in South Africa, I’ve seen the cave paintings of the San.

What about interbreeding with other human species?

A number of studies have shown a low amount of interbreeding between early modern humans outside of Africa and archaic species outside of Africa including Neandertals and, in Asia, with the species they call Denisova.They’ve never found any evidence of Neandertal DNA in Africa. The problem is that you just don’t get good preservation of fossils in Africa.  So what we did was collaborate with Josh Akey and Ben Vernot at the University of Washington and used a statistic they developed to recognize regions of genome that appear to be of archaic origin.

The first thing we did is to test this statistic by applying it to non-Africans and we found a very strong enrichment for Neandertal DNA in those genomes. But we didn’t see that in the Africans. They had no Neandertal DNA. When we applied the statistic to Africans, though, we still saw a lot of evidence for interbreeding from a hominid who diverged from a common ancestor that we shared about 1.2 million years ago, about the time that Neandertals split off as well. This suggests that there could have been a sister species in Africa. What it was nobody knows. But it seems to show that modern humans have been interbreeding and it’s not unique to non-African species.

Why are African genetics so exciting?

Africa was the site of origin of all modern humans and if you want to learn about when, where and how we evolved, you want to look at this continent. It has a long history of population subdivision and adaptation of those populations to very distinct environments and a broad range of phenotypes, ranging from the short stature of the Pygmies to the  very tall stature of the pastoralists in the east. It also has very different disease exposure and very different disease prevalence throughout.

What’s next?

We want to expand our genome-wide analysis to other populations, and we want to do so with larger sample sizes. We’re going to continue to try to correlate genetic variants with different phenotypic traits. We’d love to do functional studies of these genes to see, for instance, how they are regulating pituitary development. Is there some totally novel mechanism involved. We’re going to look at the Pygmies and other groups with a systems approach. You can’t look at height, as an example, by itself. You have to look at it in relation to metabolism and immunity and see how everything interacts.

population genetics

This is a little bit different than most posts here. I have a paper out today inMolecular Ecology Resources:  “mmod: an R library for the calculation of population differentiation statistics” (doi: 10.1111/j.1755-0998.2012.03174.x). Looking around the web, there aren’t many simple expositions of just what a “differentiation statistic” might be, and why the “modern measures of differentiation” my little R package can calculate might improve on the more traditional ones. So,  I thought I’d have a go here. 

Biologists often want to be able to measure the degree to which a population is divided into smaller sub-populations. This can be an important thing to quantify, because sub-populations within highly structured populations are, to some extent, genetically distinct from other sub-populations and therefore have their own evolutionary histories (and perhaps futures).

To illustrate this point I’ve run some simulations. Imagine if we had 5 subpopulations, each with a thousand individuals. In each population we will follow the fate of a locus with two alleles, R and r that have no effect on survival or reproduction and start with frequencies 0.8 and 0.2 respectively (these numbers motivated by this post). In the absence of gene flow between these populations (Panel 1) the frequency of ther allele bounces around due to genetetic drift (evolutionary change, after all, is inevitable). Crucially though, changes in one population can’t effect other populations so we end up with substantial among-population differences in allele frequency. In the next two panels, in each generation a proportion of each population’s individuals (0.001 and 0.01 respectively) are drawn from the other populations in the simulation. Now that the populations are sharing genes the lines that represent their allele frequencies pull together  (that is, the among-population variation is reduced).




 

One way to quantify the among-population variation displayed in these simulations is to look at the number of heterozygotes you expect to observe across the entire population. The final values for P(r) in the first simulation were {0.33, 0.47. 0.88. 0.10. 0.33} with a mean frequency of 0.42 (so the frequency of the Rallele would be 0.58). Knowing our Hardy Weinberg, if we had one big population with two alleles, one being at a frequency of 0.42 we’d expect to get 2pq = 2 * 0.42 * 0.58 = 0.40 heterozygotes. We can call that number H_T for expected total heterozygosity. But thats not what we’d actually see in this case. The sub-populations that make up this larger population have their own allele frequencies, when we calculate the expected proportion of heterozygotes for each of these populations by themselves we end up with {0.44, 0.49, 0.21, 0.18, 0.44} for a within-population expected heterozygosity (H_S) of 0.35*. This lack of heterozygotes within sub-populations compared with the total population expectation will always arise when genetic drift makes sub-populations distinct from each other.
Masatoshi Nei  used this pattern to propose a statistic to quantify population divergence called G_ST, which he defined like this:

 G_{ST =} (H_T - H_S ) / H_T

Nei’s motivaton with G_ST was to generalise Sewall Wright‘s F_ST **, which was defined for diploid organisms and two-allele systems, so that it could be used for any genetic data. But there’s a problem with this formulation. Because H_T  is always larger than H_S  and can’t be greater than one, the maximum possible value of
G_ST  is 1-H_S. This dependency on the within-population genetic diversity means comparisons between studies, and even between loci in one study, are difficult (since H_S will likely be different in each case). This is particularly worryingly for highly polymorphic makers like microsatellites, which can give values of H_S as high as 0.9, severely constraining the possible values of G_ST.


Although the problem of
G_ST‘s dependence on H_S has been known for a while, it’s taken some time for new statistics that get around this problem to be developed. Philip Hedrick (doi: 10.1554/05-076.1) along with Patrick Meirmans (doi: 10.1111/j.1755-0998.2010.02927.x) introduced G”_ST  - a version of G_ST that is corrected for the observed value of H_S as well as the number of sub-populations being considered. Meirmans used a similar trick to define φ’_ST  (doi: 10.1111/j.0014-3820.2006.tb01874.x), another F_ST analogue that partitions genetic distances into within- and between-population components. Most recently, Lou Joust introduced an entirely separate statistic, D, that  directly measures allelic divergence (doi 10.1111/j.1365-294X.2008.03887.x).


The statistical programming language R is becoming increasingly popular among biologists. Although there is a strong suite of tools for performing population genetic analyses in R, code to calculate these “new” measures of population divergence have not been available. My package, mmod, fills this gap.  I won’t give too many details of the package here, as that’s detailed in the paper and the package is will documented. Briefly, mmod has functions to calculate the three statistics described above (and Nei’s
G_ST ), as well as pairwise versions of each statistic for every population in a datastet. It also allows users to perform bootstrap and jacknife re-sampling of datasets, the results of which are returned as user-accessable objects which can be examined with any R function (there is also a helper function to easily apply differentiation statistics to bootstrap sample and summarise the results) . The library is on CRAN, so installation is as easy as typing “install.pacakge(“mmod”)”, the source code is up on github. If want to use the package I’d suggest reading the vignette (“mmod-demo”) before you dive in.



I’m keen to hear about bugs or feature requests from users, just email them to david.winter@gmail.com

---

Reference:


Winter, D.J. (in press). MMOD: an R library for the calculation of population differentiation statisticsMolecular Ecology Resources : dx.doi.org/10.1111/j.1755-0998.2012.03174.x

* mmod actually uses nearly unbiased estimators for these parameters, to deal with the way small population samples can mis-represent the actual allele frequencies in populations.

** I don’t want to write an entire history of F-statisitcs here, because it’s a big and murky topic, but I did want to make the point that the formulation I gave for G_ST  is often presented as “Wright’s F_ST ” in genetics courses. Wright was certainly aware that his statistic was related to the proportion of heterozygotes you expect to get in a populaiton, but, when he introduced F-statistics in general, and F_ST  in particular, he was really dealing with correlation among gametes at various levels of population structure. Unfortunately, there are now many many definitions of F_ST  floating around, and it’s probably pointless to argue about a “right one”. If you use my package I encourage you to be explicit about, and cite, the particular statistic that you are using. For each of the the F_ST  analogues that the package calculates the in-line help contains the correct reference. 

Genomics Enables Scientists to Study Genetic Variability in Human Populations

Thinking about population genetics often brings to mind visions of animals in the wild being swept along by the tide of natural catastrophes, soil depletion, or predation. However, over the past ten years the field of population genetics has undergone major renovations because of recent advances in gene sequencing and screening technologies. These technological innovations have allowed scientists to tackle bigger and broader questions related to population trends, and to study genetic variation on a much broader scale than ever before possible with older methods, such as test crosses, random sampling, and field work. Today, discoveries can be facilitated by the ever-expanding field of genomics, which is the use of large databases for the purpose of studying genetic variation on a large scale across many different organisms.

What is genomics?

Figure 1: Genome size is the total number of base pairs in an organism.  While the number of genes in an organism's DNA (red bars) varies from species to species (numbers at right), it is not always proportional to genome size (blue bars).  Note how many genes a fruit fly can squeeze out of its relatively small genome.

Until recently, the term genome was used to describe the complete set of chromosomes that made up a given species. Today, scientists use the termgenome to refer to the complete set of DNA sequences derived from each chromosome of a given species. Genomics is a relatively new and ever-expanding field dedicated to the study of defining genomes in this more specific way.

The direct analysis of the genome of an organism, or the genomes of a group of organisms, is now possible through advances in the efficiency of DNA sequencing and large-scale genetic screening. These new high-throughput methods allow researchers to collect vast amounts of information about genetic variation in very short periods of time.

Genomics has also shown that the size of a genome (i.e. the number of nucleotide pairs it contains) is not necessarily proportional to the number of genes contained within it. Some organisms, like the fruit fly, fit a considerable number of genes into a relatively small genome, whereas humans and mice possess many extra "unused" nucleotide pairs that do not encode genes (Figure 1).

See how human genomes compare to others

How many genes does it take to build a human being?

Although early reports suggested that human chromosomes might contain as many as 100,000 different genes, we now know that the 24 different human chromosomes altogether contain 20,000-25,000 different genes. However, it is likely that many of those genes are not absolutely required.

How can we study human genetic variation?

Humans can also be the focus of population genetics studies, as they too have been subject to the forces of change over long periods of time. Recently, the DNA sequence of the entire human genome was deciphered in a massive effort called The Human Genome Project (HGP). This project sequenced the DNA of each human chromosome from end to end, determined the DNA sequence of every human gene, and mapped the precise location of every human gene to a particular region of a human chromosome.

With this information in hand, scientists now have a baseline definition of every human gene. With this baseline, they are beginning to study how the DNA sequences of human genes can vary among individuals and populations. In fact, scientists can currently study the variability of those genes (i.e. all allelic forms) in different populations around the globe. Early results from these studies indicate that humans are identical over the vast majority of their genome. The apparently striking phenotypic variation among human beings around the world can be accounted for by only an exceptionally small number of genetic differences. Genes that code for skin color, facial features, or body size represent a small fraction of the DNA that comprises the total human genome.

Variation in the human genome: SNPs

More on human populations

• See a model of recent human evolution
• What are some human allelic variations?
• What is the Human Genome Project (HGP)?
• The HGP: The contributions of Francis Collins and Craig Venter

After the completion of the HGP in 2003, researchers began to pinpoint locations within the genome that varied among individuals. These scientists discovered that the most common type of DNA sequence variation found in the human genome is the single nucleotide polymorphism (SNP, pronounced "snip"). There are approximately 10 million SNPs in the human genome.

A worldwide effort known as the HapMap Project is mapping SNPs and other genetic variants in human populations around the world. By mapping the distribution of SNPs among different human populations, researchers can begin to learn which types of variation are most common in certain regions of the world. This information will help explain human origins and disease risks as well as how they relate to environmental conditions, both past and present. To date, the HapMap project has identified over 3.1 million SNPs across the human genome that are common among individuals of African, Asian, and European ancestry.

The HapMap database has also helped foster a new type of research in personalized medicine called the genome-wide association study (GWAS). With these studies, the distribution of SNPs is determined in hundreds, or even thousands, of people with and without a particular disease. Comparisons between diseased and non-diseased groups of individuals help determine which SNPs co-occur with disease symptoms. With this information in hand, scientists can carry out statistical analyses to help predict whether a certain SNP is associated with a specific disease, with the hope of identifying individuals who may be at risk.

For example, in a recent study conducted in the United Kingdom, researchers genotyped 2,000 individuals who had one of seven common disorders. Next, those individuals were compared to 3,000 genotyped control individuals who did not have the common disorders. With these comparisons, the researchers identified new genetic markers associated with increased risk for disorders such as heart disease and diabetes. In the future this study will be expanded to include 36,000 more individuals, and it will focus on 14 more health-related disorders as well as individual responses to certain drugs. Using these types of studies, scientists can sample large numbers of people and make meaningful predictions regarding disease risk for individuals based on the presence or absence of genetic markers within their genome.

Genomics and biological discovery

Genomic data can support discovery in diverse areas of biology, including medicine, systematics, and conservation biology. Like many histories, the history of genomics is fraught with conflict, disagreement, and excitement. The personalities and ideas that have shaped genomics even included a race between publicly funded and corporate genome sequencing groups that resulted in a tie at the finish line. Several subspecialty areas of genomics are also expanding as the community of scientists within them grows. These relatively new areas of genomic investigation include: epigenomics (the study of DNA modification), transcriptomics (the study of cellular RNA content), and proteomics (the study of proteins that characterize a particular cell). 

Mobile elements across human populations

Total citations

• Data not available
  Web of Science
• Data not available
  CrossRef
• Data not available
  Scopus

Online attention

Altmetric score (what's this?)

• 0 readers on Mendeley
• 0 readers on CiteULike
• 0 readers on Connotea

1,281Page views

Explanation of terms and methodology

• Sources: Web of Science, CrossRef, Scopus, WebTrends, and Altmetric

  Frequency of updates: In most cases, our metrics data is updated hourly, In some cases, such as page views, the data is not available the first 48 hours after an article is published.

• Citations

  Single number count for article citations from each service's database (may vary by service). The citations counts are reliant on the availability of the individual APIs from Web of Science, CrossRef, and Scopus. These counts are updated daily once they become available. Once a citation count is available, the list of articles citing this one is accessible by clicking on the circle for that citation source.

• Page views

  A cumulative count of full-text article views that includes HTML views and PDF downloads. These data reflect only usage on the nature.com journal platform, as tracked by WebTrends tagging, and are only for articles published on or after 1 January 2012. NPG content is also available on other third party platforms and in repositories around the world; we are currently not including article usage data from any of these third-party services. The page views data is available 48 hours after online publication and is updated daily.

• News, blogs and Google+ posts

  The number of times an article has been cited by individual mainstream news sources, blog post, or member of Google+ along with a link to the original article or post. News articles, blog posts and Google+ posts do not always link to articles in a way that can be picked up by aggregators used by Altmetric, so the listed links are not necessarily a reflection of the entire scope of media, blog or Google+ interest. Further, the list of blogs and news sources covered is manually curated by Altmetric and thus is subject to their discretion for inclusion as a scientific blog or media source. The news, blog, and Google+ posts are provided by Altmetric and are updated hourly.

• Altmetric score

  Altmetric calculates a score based on the online attention an article receives. Each coloured thread in the circle represents a different type of online attention and the number in the centre is the Altmetric score. The score is calculated based on two main sources of online attention: social media and mainstream news media. Altmetric also tracks usage in online reference managers such as Mendeley and CiteULike, but these do not contribute to the score. Older articles will typically score higher because they have had more time to get noticed. To account for this, Altmetric has included the context data for articles of a "similar age" (published within 6 weeks of either side of the publication date of this article).