1 Jul 2013
2 July 2013. On 25 April 2013, the last day of the International Congress on Schizophrenia Research in Orlando, Florida, four scientists from the Lieber Institute for Brain Development in Baltimore, Maryland, described their application, BrainCloud, a tool for analyzing gene expression and DNA methylation patterns across the lifespan and as a function of genetic variation. The symposium, titled “BrainCloud: An Integrative Approach to Genomic Brain Research,” was chaired by Joel Kleinman of the National Institute of Mental Health in Bethesda, Maryland.
Kleinman explained that BrainCloud was created to generate a better understanding of the genetic variation that confers risk for mental illness. In an effort to go beyond mere statistical associations, the application allows researchers to look at the effect of genetic variants on molecular phenotypes such as gene expression and epigenetics in hopes of better understanding the putative associations of single nucleotide polymorphisms (SNPs) with schizophrenia.
BrainCloud is a freely available, stand-alone application that comes in two flavors: 1) microarray gene expression data obtained from 269 subjects (see SRF related news story) and 2) DNA methylation data from 108 subjects (see SRF related news story). Both datasets use tissue obtained from the dorsolateral prefrontal cortex (DLPFC) of control subjects, eliminating many of the confounding factors, such as medication, that are present in subjects with mental illness. Every subject is well characterized, with genotype, medical history, histopathology, and tissue quality information available. In addition, with an age range that spans the majority of the human lifespan, BrainCloud is ideal for examining age-related changes in gene expression and DNA methylation.
The first speaker, Andrew Jaffe, discussed the methods used to identify global patterns of transcription across the lifespan using BrainCloud, focusing on different approaches to pattern finding. The original publication of the dataset (Colantuoni et al., 2011) used principal component analysis to find patterns of fetal gene expression that reversed in early postnatal life. However, because each principal component must be independent of all other principal components, principal component analysis has limited ability to detect subtle gene expression changes, Jaffe said.
He then described a second approach he has used to find richer global age-related patterns of gene expression. First, he used surrogate variable analysis to renormalize the original data, allowing for greater removal of variability (both experimental and biological in nature). Next, he employed a Bayesian approach termed Coordinated Gene Activity in Pattern Sets (CoGAPS) that reduced the statistical constraints of pattern finding to identify distinct temporal patterns of gene expression. Jaffe found a much more complex age-related trajectory of gene expression than originally reported: several distinct waves that peaked during fetal development, one during the first year of life, another during early adolescence, one that plateaued in early adulthood, and finally, a wave that rose during aging. The renormalized data, Jaffe noted, are available on the new, Web-based version of BrainCloud.
Barbara Lipska compared the transcription and DNA methylation BrainCloud datasets, reviewing the major findings of the two BrainCloud publications to date (see SRF related news story; SRF news story). Expression data are available for roughly 30,000 genes, while methylation data are available for about 14,500 genes. Lipska reported that both the mRNA expression and DNA methylation plots have similar clusters between fetal and postnatal life, with the largest changes located during fetal development. “Fetal samples stand out as being very different from the rest of the cohort,” she noted. In addition, gene splicing is developmentally regulated, in some genes by DNA methylation, and differs among genotypic variants. Both mRNA expression and DNA methylation are strongly predicted by genotypic variants for a large number of genes.
Thomas Hyde was next up at the podium, where Kleinman noted the former would be “showing us the future of BrainCloud.” The next generation of BrainCloud, said Hyde, will have many more new datasets. In its next iteration, BrainCloud will include data from a much larger cohort of subjects spanning the major mental illnesses: 398 controls, 212 subjects with schizophrenia, 96 with bipolar disorder, 165 with major depression, and 18 with autism. In addition to cDNA microarrays, SNP chip and DNA methylation microarrays will be added, along with tissue from the hippocampus. Finally, a new look at the transcriptome will be provided using RNA-Seq, a high-throughput sequencing technique that can quantify RNA content as well as sequence variations. As an example, Hyde showed how RNA-Seq maps have been built from adult and fetal samples of Axin1, a negative regulator of the Wnt signaling pathway that is important in embryonic development. He used these maps to find new splicing events that were then validated using PCR.
As he concluded his talk, Hyde reflected on the nearly two decades of work that has gone into the collection of the samples included in BrainCloud, all obtained at the same site and assessed by the same diagnostician. “We hope that this unified collection can be a resource for all of you for your research,” he said.
In a departure from the rest of the symposium, the final speaker, Kristin Bigos, showed how BrainCloud data can be used to explore clinical research questions, focusing on their application to clinical pharmacology. For example, by examining the developmental trajectory of the genes encoding cytochrome P450 (CYP450) enzymes that are known to be involved in drug metabolism in the liver, Bigos demonstrated a role for these genes in brain development and adult brain function.
Bigos also showed that BrainCloud data can be used to “explore how differences in brain expression related to genetic variation can be exploited as novel potential drug pathways.” CACNA1C is a calcium channel gene that is associated with risk for schizophrenia and bipolar disorder (see SRF related news story; SRF news story). Using BrainCloud, Bigos found that the risk genotype is associated with increased mRNA expression in the DLPFC, suggesting a potential treatment target. This finding provided a rationale for the ongoing clinical trial using a CACNA1C antagonist to treat schizophrenia and bipolar disorder. Bigos also used BrainCloud data to examine the relationship between mRNA expression and risk variants in GRM7, a metabotropic glutamate receptor that is associated with schizophrenia and bipolar disorder.—Allison A. Curley.