Novel integrative genomic tool for interrogating lithium response in bipolar disorder

Abstract

We developed a novel integrative genomic tool called GRANITE (Genetic Regulatory Analysis of Networks Investigational Tool Environment) that can effectively analyze large complex data sets to generate interactive networks. GRANITE is an open-source tool and invaluable resource for a variety of genomic fields. Although our analysis is confined to static expression data, GRANITE has the capability of evaluating time-course data and generating interactive networks that may shed light on acute versus chronic treatment, as well as evaluating dose response and providing insight into mechanisms that underlie therapeutic versus sub-therapeutic doses or toxic doses. As a proof-of-concept study, we investigated lithium (Li) response in bipolar disorder (BD). BD is a severe mood disorder marked by cycles of mania and depression. Li is one of the most commonly prescribed and decidedly effective treatments for many patients (responders), although its mode of action is not yet fully understood, nor is it effective in every patient (non-responders). In an in vitro study, we compared vehicle versus chronic Li treatment in patient-derived lymphoblastoid cells (LCLs) (derived from either responders or non-responders) using both microRNA (miRNA) and messenger RNA gene expression profiling. We present both Li responder and non-responder network visualizations created by our GRANITE analysis in BD. We identified by network visualization that the Let-7 family is consistently downregulated by Li in both groups where this miRNA family has been implicated in neurodegeneration, cell survival and synaptic development. We discuss the potential of this analysis for investigating treatment response and even providing clinicians with a tool for predicting treatment response in their patients, as well as for providing the industry with a tool for identifying network nodes as targets for novel drug discovery.

Figure 1

Bipolar disorder lymphoblastoid cell expression data in vehicle versus chronic lithium treatment. Affymetrix arrays (miRNA chips 3.0 for a and b; U133-Plus 2.0 Human gene profiling messenger RNA (mRNA) arrays for c and d) are shown depicting microRNAs (miRNAs) or mRNAs differentially expressed between lithium vs vehicle in responders (a and c) and non-responders (b and d) and then subject to hierarchical clustering. Significantly regulated miRNAs (determined by meeting these criteria ±1.5-fold regulation and unadjusted P-value <0.05) are shown for both groups (a and b). Note that in the responder group (a) when R31F-lithium (orange) is placed in a separate group, it then clusters with responder vehicle group rather than responder lithium group. Also, in the non-responder group (b) when 18M_veh (purple) is placed in the outlier group, it clusters with the non-responder lithium group. Also note the increased heterogeneity in the non-responder group where the vehicle group shows two sets of clusters. Significantly regulated mRNAs (determined by meeting these criteria ±1.2-fold regulation and unadjusted P-value <0.05) are shown for both groups (c and d). Note that in the responder group (c) when R39F-lithium and 45M_lithium (red) are placed in a separate group, they cluster with responder vehicle treatment. Also, in the non-responder group (d) 10F_Veh (gray) shows abnormal hybridization and is placed in a new group. This greatly enhances clustering in non-responder groups (lithium vs vehicle). There are fewer genes that pass the same significance/fold filter in the non-responder group (d) compared with the responder group (c). These results illustrate a greater heterogeneity in the non-responder group than the responder group, and this may be indicative of an intrinsic lithium influence on transcriptional targets in the responder group.

Figure 2

MicroRNA (miRNA)–messenger RNA (mRNA) network visualization. Depicted are three miRNA–mRNA networks generated by GRANITE (Genetic Regulatory Analysis of Networks Investigational Tool Environment) using an mRNA filter (>1.2-fold regulation), where target scan is used to predict miRNAs that are targeted by lithium-regulated mRNAs taken from the arrays. Each dot represents either a miRNA (in blue) or mRNA (in yellow or red) node. The blue core represents high edge density miRNA nodes and the outer mantel clusters represent relatively lower-density nodes, with the red ones representing the lowest density. The responder (a) network shows the greatest gulf between the blue core and the mantle of visualization due to the high degree of miRNAs contained at the core, with many having over 500 connections. The non-responder (b) network has a less-pronounced core that may be due to more sample heterogeneity. There is a cluster of miRNAs in the Let-7 family as shown. The common (c) network was generated to identify common lithium networks that are conserved between responders and non-responders.

Figure 3

Genes controlled by Let-7. Depicted is an illustration of genes that have microRNA (miRNA)-binding sites to be regulated by Let-7 in the non-responder (NR) GRANITE network.