Skip to main content
Premium Trial:

Request an Annual Quote

Integrative Network Approach Used to Profile Drug Response in Bipolar Disorder


NEW YORK (GenomeWeb) – A National Institute of Mental Health-led team has demonstrated the feasibility of using an open-access computational pipeline to visualize genomic datasets in interactive networks corresponding with drug response in individuals with bipolar disorder or other conditions.

In a proof-of-principle study appearing online in Translational Psychiatry last week, the researchers used this approach — known as the "Genetic Regulatory Analysis of Networks Investigational Tool Environment," or GRANITE — to narrow in on microRNA and messenger RNA patterns that differ in samples from individuals with bipolar disorder who respond to lithium treatment and those who don't.

"The novelty of this approach is to be able to combine this genomic data together to make larger response networks for lithium response in bipolar patients," co-corresponding author Joshua Hunsberger told GenomeWeb. "We used GRANITE to build these rather large response networks that combined miRNA and mRNA data."

Hunsberger was based at NIMH when the study was done. He is currently a research associate in regenerative medicine at Wake Forest University.

Although the current comparison was performed at a single time point following lithium treatment, he and his co-authors noted that GRANITE can also be used to evaluate expression data that's been collected during a time course experiment to track the consequences of acute and chronic treatment.

Hunsberger noted that the tool appears to have promise for tackling a wide range of research and clinical questions, too.

For example, those involved in the study suggested that the GRANITE pipeline may prove useful for discovering new drug candidates, delving into the roots of drug response or of a disease itself, or distinguishing therapeutic doses of a compound from those that are toxic or ineffective.

In the clinic, meanwhile, the team hopes to eventually see such approaches used to not only help in prescribing appropriate treatments, but also for predicting patient outcomes given a particular therapy and/or drug dose.

To that end, the researchers are considering ways of integrating SNP information into the pipeline, Hunsberger said, since such variants should be identifiable from straightforward tests on DNA found in an individual's blood sample.

"We'd like [clinicians] to be able to genotype their patients and then, from that genotype, be able to say, 'Okay, we think there's a 70 percent chance that you're going to respond well to lithium,' for instance," he explained.

The current iteration of GRANITE was developed with the help of funding from the National Institutes of Health — including NIMH, NIAID, and NIAID's Bioinformatics and Computational Biosciences Branch.

A key software component called Gravel was developed under the GRANITE project to produce networks from gene expression and miRNA expression input data.

For their current analysis, Hunsberger and his colleagues applied the GRANITE-Gravel tool for establishing molecular networks related to lithium response using Affymetrix array-based miRNA and mRNA profiles generated for cell lines produced from blood samples for eight individuals with bipolar disorder who responded to lithium treatment and eight non-responders.

In the case of the non-responders, the team included participants who had been taking lithium for at least two years and used blood tests to confirm that individuals had been taking the prescribed dose of the drug, Hunsberger noted.

From the miRNA and mRNA networks obtained before and after lithium treatment, the researchers identified an apparent role for Let-7 family miRNAs in lithium response.

The group generated several networks using information at more than 1,000 mRNAs and nearly 300 miRNAs with a version of the GRANITE pipeline that was outfitted with TargetScan software for predicting miRNA-mRNA interactions.

In cells from both the responder and non-responder individuals, the study's authors saw a dip in the expression of Let-7 family miRNAs after lithium treatment. But this decline was more widespread in LCLs from the responder group.

"The Let-7 family could represent a novel target for [lithium] response," they wrote, "although further investigation is warranted."

Similarly, the networks highlighted genes that were differentially expressed in the responders and non-responders — patterns they verified using quantitative reverse transcription PCR.

From these results, they concluded that "[t]he network analysis algorithms and visualizations implemented in GRANITE may be instrumental in biomarker identification that potentially could aid in predicting [lithium] responsiveness in patients, as well as providing insights in other similarly complex phenotypes."

Hunsberger noted that the GRANITE pipeline is modular, meaning it's possible to swap in different algorithms for assessing miRNA targets and other data types depending on an investigator's preference and the nature of the project at hand. It also currently contains open-source libraries, such as the interactive visualization program Gephi, and can be run without commercial software provided users have a Java virtual machine installed.

Likewise, GRANITE is expected to be compatible not only with array-based miRNA and RNA expression data, but also with expression profiles produced by RNA sequencing.

Researchers can use it to assess data for any tissue type of interest, Hunsberger noted — from LCLs such as those considered in the proof-of-principle bipolar disorder analysis to tissues developed via an induced pluripotent stem cell intermediate to more easily obtained blood samples.

The analysis may take anywhere from a few minutes to a few days, depending on the type of analysis being done and the level of network resolution that users are after, co-author Rajdeep Singh, an information systems and global solutions scientist with Lockheed Martin, told GenomeWeb in an email message.

Co-authors on the study spent several months finding ways to decrease the time needed to run the tool so that the initial analysis is ready to go just a couple minutes after obtaining the appropriate expression data.

In addition to further validating findings from the current study, the researchers hope to see GRANITE paired with SNP profiles as a means of establishing predictive models for drug response in bipolar disorder and other conditions.

"The future goal would be to validate this tool and show that some of the networks we've developed can, in fact, help to predict patients' lithium response by expanding [the networks] and adding in the SNP genotyping," Hunsberger said. "But that would, of course, require a much larger sample size."

If sufficient resources are available, he noted, it would be beneficial to establish databases where users could upload their data to build up integrated network data generated through various disease and drug response experiments.

"If you generate a large database and allow people to use the tool and develop some of these networks, I think that would be a great way to use crowd-sourcing to really move the field forward," Hunsberger argued. "There's a lot of data out there that folks have and if they're willing to share that, then we have a novel tool to integrate all of that data together."

In the meantime, the group is eager to hear about other researchers' experience with GRANITE, he noted. "We're excited to hear folks' thoughts and get feedback on what they think of this tool and how it can be potentially improved."