Providing data access, ensuring data protection

Readers of this post probably don’t need to be convinced that genetic association data have enormous potential for helping us to understand and treat complex diseases like type 2 diabetes. Significant associations between variants and diseases can suggest genes, or regions of the genome, that could be important for disease risk or progression—and this knowledge could help us identify new drug targets.

The Accelerating Medicines Partnership in Type 2 Diabetes (AMP T2D) is a pre-competitive partnership among the National Institutes of Health, industry and not-for-profit organizations, which is managed by the Foundation for the National Institutes of Health. Its mission is to make genetic association data accessible to the worldwide biomedical research community via the Type 2 Diabetes Knowledge Portal, in order to facilitate discovery of new targets for T2D treatment. But it can be a challenge to aggregate genetic data. The privacy of the individuals who contributed their health status and genomic sequences must always be protected, and there are many layers of regulation to ensure this. Restrictions at the institutional, regional, and national levels determine how data are handled and whether they can be transferred.

Until now, all of the results displayed in the Portal have been derived from data housed at the AMP T2D Data Coordinating Center (DCC) at the Broad Institute, where the Portal website resides. But some of the valuable data generated outside the U.S. cannot be transferred to the DCC. To address this issue, AMP T2D funded the development of a mechanism that enables researchers to interact with all of the data: federation. 

Federation means that data are housed at a site (a “federated node”) that meets their specific privacy requirements, but are made available for remote queries via the Portal. Results from such queries are served up alongside results from all of the datasets housed in the AMP T2D DCC. Researchers may browse and query data from any location without even needing to know where they reside.

A federated node has now been created at the European Bioinformatics Institute (EBI) and may be accessed via the T2D Knowledge Portal. Today, Portal tools and interfaces can query both data housed at the AMP T2D DCC at the Broad Institute and data at the EBI federated node. 

According to Paul Flicek, a Senior Scientist and Team Leader of Vertebrate Genomics at EMBL-EBI, “A key mission of EMBL-EBI is to make data available to the widest possible community. Seamlessly accessing stored in multiple locations via a single portal helps ensure that the data we store from many projects are maximally useful for additional research.”

The first dataset to be incorporated into the Portal via the EBI federated node is the Oxford BioBank exome chip analysis dataset, which contains association data for glycemic, lipid, and blood pressure traits from over 7,100 healthy subjects in Oxfordshire, U.K. The dataset is described on our Data page. Portal users can interact with this dataset in the same way (and with the same speed) as with other datasets. 

“Diabetes is a global problem, and it will take research and innovation on a global scale if we are to tackle it effectively,” says Mark McCarthy, Robert Turner Professor of Diabetic Medicine at University of Oxford. “The success of our research on the genetics of diabetes depends on access to data generated by groups around the world. The federated portal provides an additional set of tools that will allow us to jointly analyse those data sets wherever they happen to be based.” 

Federation represents both an important technical advance in handling and protecting data, and a significant step forward in democratizing and improving access to genetic association results. And because it is generally applicable to any kind of genetic association data, it has the potential to have an impact beyond T2D research, facilitating the study of other complex diseases and traits.

Insulin sensitivity comes into focus

Many different things can be seen in any landscape, depending on your focal point.
Image by Nicooo76 via Pixabay.

When photographing a landscape, different photographers choose different perspectives. Some capture a wide-angle view, while others focus on particular details.

It’s no different for researchers who use genome-wide association studies (GWAS) to investigate the genetic landscape of type 2 diabetes (T2D). A common perspective is to study the wide range of variants that are significantly associated with the presence of T2D in patients. But it can also be very informative to concentrate on individual traits related to the physiology of T2D. In a new paper in Diabetes, co-first authors Geoffrey Walford, Stefan Gustafsson, Denis Rybin, and fellow members of the Meta-Analyses of Glucose and Insulin-related traits Consortium (MAGIC) took this focused perspective to discover associations of genetic variants with insulin sensitivity.

Along with reduced insulin levels, the loss of insulin sensitivity (often termed insulin resistance) is a major hallmark of T2D. When muscle, liver, and fat cells become less able to respond to insulin, blood glucose levels rise. Since this can contribute to development of T2D and exacerbate its symptoms, knowing which genetic variants are associated with sensitivity to insulin could be informative for understanding pathways that contribute to T2D risk.

But insulin sensitivity is difficult to measure. Earlier GWAS have used simple estimates of insulin sensitivity, such as fasting levels of insulin, and have discovered a handful of genetic variants that influence insulin sensitivity. The “gold standard” test, the euglycemic clamp, involves giving patients continuous infusions of insulin and glucose and monitoring their blood glucose every few minutes. It’s expensive and time-consuming—not a test that is practical to perform on the tens of thousands of subjects that are commonly used in GWAS.

The authors wondered whether they could instead use an index that combines several measurements, each relatively easy to make. It’s an index with a long name: the modified Stumvoll Insulin Sensitivity Index (ISI). Developed by Stumvoll and colleagues in 2001, this index can be derived in a variety of ways. The authors chose the ISI requiring just three measurements: fasting insulin levels; glucose levels two hours after a glucose load; and insulin levels two hours after a glucose load. This ISI is as good as or better than other estimates of insulin sensitivity and correlates well with the euglycemic clamp.

So the researchers looked for variants associated with the Stumvoll ISI in nearly 17,000 participants in the discovery phase of the work. They added another 13,300 in the replication phase, adding up to about 30,000 in the combined meta-analysis. Since obesity, measured by body mass index (BMI), can affect insulin sensitivity, the authors added BMI to some of their statistical models.

First, the authors found associations between the ISI and other variants already known to affect simple measures of insulin sensitivity. This provided reassurance that the ISI was properly detecting genetic influences on insulin sensitivity. After discovery, replication, and meta-analysis, two novel genetic variants were associated with ISI at genome-wide significance (P-value < 5.0 ×10^-8) in a model that tested the effect of the variant, age, sex, and the interaction between the variant and BMI: variant rs12454712, near the gene BCL2, and variant rs10506418, near the gene FAM19A2.

How might these variants affect insulin sensitivity? There’s a lot more work to be done before that question can be answered. Additional studies will need to clarify whether these variants, which are near BCL2 and FAM19A2, affect these or other genes, and then how these variants actually cause changes in insulin sensitivity. 

There are some clues already in the published literature. The variant rs12454712 near BCL2 has previously been found to be associated with T2D, supporting the hypothesis that this region of the genome contributes to T2D risk through reducing insulin sensitivity. And the gene itself (BCL2) has already been implicated in glycemic metabolism: inhibiting bcl2 improves glucose tolerance in a mouse model, while a drug that inhibits the protein product of the gene (BCL2) increases blood glucose levels in certain chronic lymphocytic leukemia patients. So there’s even more reason to suspect that the rs12454712 variant might affect insulin sensitivity via BCL2.

There is as yet no evidence linking the protein FAM19A2 function to glucose metabolism, so the jury is out on whether the variant rs10506418 affects FAM19A2 or some other nearby gene. 

By focusing on a detail of the T2D-related genetic landscape, this study has teased out two variants that may give us clues about the physiology of insulin sensitivity and the development of T2D. And that’s a valuable addition to our overall picture of T2D genetics!

Better summaries of variant information convey the most important information at a glance

We’ve made significant improvements to the information we display on the Variant pages of the T2D Knowledge Portal. The summary at the top of each Variant page (view an example) now shows the reference nucleotide and the variant nucleotide at that position. Transcripts covering the variant are listed, along with several important details for each transcript: the change caused by the variant in the encoded protein sequence (if applicable); the Sequence Ontology term describing the consequence of the variation (for example, “missense variant”); and the expected effect of the variant on protein function, as predicted by the PolyPhen and Sift algorithms.

Summary section of the Variant page

Just below the summary on the Variant page, we’ve also improved the graphic showing the association of the variant with T2D and related traits. We’ve re-named this section “associations at a glance” because it immediately shows the most important information about these associations. 

At-a-glance section of the Variant page. Click the image to view a larger version.

The boxes in this graphic represent the associations of this variant with T2D (at the top) and with other traits (below, in an expandable section). Under the hood, the software is now pulling up information more quickly so that the display is more responsive. We’ve also made it more pleasant to look at, tidying up the shape of the boxes and the alignment of the information they contain.

But beyond the style improvements, we’ve added a lot of substance. Where available, each association now includes the odds ratio (for dichotomous traits) or the effect size (for continuous traits) and the direction of effect. Positive effects are shown in blue, and negative effects in purple. 

We’ve also added the sample size, in black text in the bottom left corner of the box, for each data set. This indicates the total number of individuals involved in the study. And if available, the frequency and count of the variant in the data set are shown in red and blue text at the bottom middle and bottom right corner of the box, respectively. The count indicates the number of haplotypes in the set that contain the variant, while the frequency indicates the occurrence of the variant allele in the sampled population.

This additional information can help you evaluate the significance of associations. The sample size and variant count determine the power of the data set to establish the association. The higher the power, the more accurate the estimate of the variant’s effect.

Finally, when a variant is associated with other traits in addition to T2D, those traits in the same category are labeled with the same color. For example, in the display above, proinsulin levels, fasting glucose, HOMA-B, and two-hour glucose—all glycemic phenotypes—are labeled in orange, while triglycerides, LDL cholesterol, and cholesterol—lipid phenotypes—are labeled in red. This lets you see easily when a variant is linked to multiple traits that could reflect a common process or pathway, possibly offering a clue to the mechanism by which it affects physiology.

So this improved graphic now gives you an idea, literally at a single glance, of how strongly a variant is associated with T2D, how significant that association is, and whether it is also associated with other traits. 

We made these improvements in response to suggestions from scientists who use the T2D Knowledge Portal. We hope to hear your feedback too!

Variant Finder results may be saved, shared, and bookmarked

You may have noticed that our Variant Finder tool has a cleaner look and clearer instructions. But did you know that you can also save your search parameters, to re-create your search later or share it with a colleague?

First, construct your search. Here’s an example:

Click the image to view a larger version

After you click “Submit search request” you’ll be taken to the results page:

Click the image to view a larger version

And here’s the URL of the results page for this example search:

http://www.type2diabetesgenetics.org/variantSearch/launchAVariantSearch/?%5B%7B%22phenotype%22%3A%22T2D%22%2C%22dataset%22%3A%22ExChip_82k_mdv2%22%2C%22prop%22%3A%22P_VALUE%22%2C%22value%22%3A%225e-15%22%2C%22comparator%22%3A%22%3C%22%7D%2C%7B%22phenotype%22%3A%22FG%22%2C%22dataset%22%3A%22GWAS_MAGIC_mdv2%22%2C%22prop%22%3A%22P_VALUE%22%2C%22value%22%3A%225e-8%22%2C%22comparator%22%3A%22%3C%22%7D%2C%7B%22phenotype%22%3A%22PI%22%2C%22dataset%22%3A%22GWAS_MAGIC_mdv2%22%2C%22prop%22%3A%22P_VALUE%22%2C%22value%22%3A%225e-10%22%2C%22comparator%22%3A%22%3C%22%7D%2C%7B%22prop%22%3A%22MOST_DEL_SCORE%22%2C%22value%22%3A%222%22%2C%22comparator%22%3A%22%3D%22%7D%5D

It isn’t pretty, but it encodes the search. You can bookmark it, save it, or email it and you’ll get back the same result next time you enter it in a browser.

There’s one small caveat here. On the results page, you can modify the results table by clicking on the + signs in the table header to see options for adding more data to the table. But if you do this, those changes will not be encoded in the URL (we plan to enable this in the future); only the original search is encoded.

Let us know how you like this feature and what other features might be useful to you. And check out our mini-tutorial on the Variant Finder to see full instructions on how to use this tool. 

New graphics and table summarize variant associations at a glance

Our variant information pages now contain two new sections that make it easy to see quickly whether a variant is associated with type 2 diabetes or related traits, and just how significant those associations are.

At the top of the Variant page for one particular variant, the section titled “Is (variant name) associated with disease?” opens to show the associations of that variant with T2D in all datasets that are currently available via the Portal (view an example). Click the link “expand associations for all traits” to see significant associations with other T2D-related traits.

Each box represents an association between this variant and a trait as detected in one data set, and the color of the box indicates the significance of the association. Dark green shows genome-wide significance (p-value < 5 x 10e-8); medium green shows locus-wide significance (p-value < 5 x 10e-4); and light green denotes nominal significance (p-value < 0.05). Associations that do not meet the threshold for significance are shown in a white box.

These new graphics make it easy to see quickly that the variant rs13266634 is strongly linked to T2D, fasting glucose levels, and proinsulin levels.

Just below this section, the “Association statistics across traits” table gives complete details about the associations between the variant and multiple traits. The same shades of green show the most significant associations.

In this table with more details about the associations of this variant, the consistent color scheme highlights significance levels.

Information shown in this table for the variant-trait associations may include p-value, direction of effect, odds ratio, minor allele frequency, and effect size. The table can be sorted by trait name. Where a variant-trait association was detected in more than one study, the most significant result is shown; plus signs allow you to expand the table and view results from additional studies. Some datasets can also be expanded to show associations in different ancestries or cohorts.

We’re still developing these new features, and your feedback could help us make them even better. Please explore them and let us know what you think!