SNPwatch: Large Studies Find SNPs with Small Effects on Height

SNPwatch gives you the latest news about research linking various traits and conditions to individual genetic variations. These studies are exciting because they offer a glimpse into how genetics may affect our bodies and health; but in most cases, more work is needed before this research can provide information of value to individuals. For that reason it is important to remember that like all information we provide, the studies we describe in SNPwatch are for research and educational purposes only. SNPwatch is not intended to be a substitute for professional medical advice; you should always seek the advice of your physician or other appropriate healthcare professional with any questions you may have regarding diagnosis, cure, treatment or prevention of any disease or other medical condition. tallshadow2.jpgThree articles published online this week in the journal Nature Genetics report a bumper crop of SNPs associated with human height. Unlike other physical characteristics such as obesity that are caused by a mix of genetic and environmental factors, 90% of normal variation in human height is due to DNA alone. But unraveling the details of the genetics has been difficult. Last year, two large studies (here and here) looked for SNPs associated with height. Only two convincing candidates were found, and each of them accounted for only about a half-centimeter difference in height. The three new studies (here, here, and here) were able to find more SNPs associated with height by using extremely large sample sizes — two of the studies had more than 30,000 participants and the third study looked at more than 15,000 people. Large sample sizes increase the chance that SNPs with small effects will be found. Like the previously found height SNPs, the new SNPs each individually account for only a fraction of a centimeter in height difference. The authors of all three papers note that statistical analyses of their data indicate that many more small effect SNPs are waiting to be found. Even bigger sample sizes may be necessary to detect these SNPs. Many of the SNPs associated with height are in genes with well-documented functions in processes such as cell division, cell-to-cell signaling, and gene regulation. Other SNPs, however, are in genes about which little is known. “There may be more than a hundred genes which affect our height, many of which will work in surprising or unpredicted ways,” said Mike Weedon, lead author of one of the papers. “The challenge now for us is to understand how they influence growth in the body. This could open up new avenues for treating a range of diseases,” said Weedon. Want to know how you stack up? 23andMe users can see their data for 46 distinct height-related SNPs in the Genome Explorer (now called Browse Raw Data) (in some cases we substitute a SNP from the three new studies with an equivalent one that is included on our chip). We’ve used the data in each paper to calculate the approximate effect in centimeters for each SNP. The effect shown is the increase in height (in cm) that each copy of the “tall” version of the SNP would give a person compared to someone who had two “short” versions. At this point, these findings apply only to people of European ancestry. picture-5.png Photo by Dave2003/istockphoto
  • What about how genome-wide association failed to find significant associations despite such a large sample? This was brought to my attention by Daniel MacArthur at

  • Well, it depends on what you mean by significant. Each of the associations the three papers report is *statistically* significant.

    But, as you’ve noticed, none of them is very big.

    As Daniel MacArthur noted, genome-wide association studies have trouble finding SNPs that have really small effects. This was exactly the problem height researchers were facing. Height seems to be influenced by a huge number of variants that each only contribute a tiny effect.

    As MacArthur notes, the solution to this problem is brute force — using really large numbers of subjects in the studies. And that’s exactly what the authors of the papers reported on here did. Two of the papers looked at more than 30,000 people!

  • dgmacarthur

    Firstly, it’s clear that brute force will only get you so far: even the largest of the three studies managed to find variants that together explain only 3.7% of the total variance in height! That means that the SNPs laid out in the table above, while fun to look at, are essentially meaningless to individual 23andMe customers. Their effects (half a centimetre this way or that) will be completely drowned out by the effects of the genetic and environmental variables explaining the remaining 96.3% of the variance.

    These SNPs capture such a small proportion of the total variance that they provide no real useful data that might help you explain, for instance, which side of the family your tall stature came from, or how tall your kids are likely to be.

    As for the brute force approach being the best way forward: researchers are likely to capture a little more of this variance by pulling together samples from existing studies and adding in new ones, but this approach will have rapidly diminishing returns.

    That’s because current chip-based approaches can only capture common variants associated with height (or other complex traits, or common diseases). As I explained in the post linked to by Andrew, there are entire classes of genetic variation – such as rare variants, or copy-number variation – that probably underlie a major chunk of the variation in these traits but are essentially completely invisible to existing chips. They simply won’t be captured by chip-based approaches, regardless of how large the sample sizes are. (For the same reasons, they also won’t be tagged by the SNPs on 23andMe’s existing platform.)

    That doesn’t completely negate the use of SNP chips by researchers or by personal genomics companies – they’re the best we have at the moment, unless you have a spare $350,000 to pay for whole-genome sequencing! But customers should be aware that these chips do (and always will) provide a seriously incomplete picture of the total genetic variation contributing to human traits and common disease risk.


  • Daniel,
    Thanks for raising some excellent points. It’s true that SNPs give you only a piece of the picture, and often a small piece at that. And increasing sample sizes will take you only so far. But you have to start somewhere!

    In the case of height, the first genetic association of any kind came out only a few months ago, with the publication of a SNP in the gene HMGA2. Though the SNPs mentioned in this post explain only a small percentage of variation in height population-wide, in particular cases they may account for a lot more than that. For example, two siblings may find that a substantial fraction of their height difference can be explained by these SNPs.

    It’s also important to make a distinction between using SNPs as research tools as opposed to probes of a particular person’s genetics. It’s true that researchers have a limited ability to discover associations involving CNVs, rare variants, deletions, etc. using chips (though CNVs and deletions are within reach to some extent). But once those associations are found through other means 23andMe may be able to detect some of them with our custom chip.