11 Oct 2016
by Michele Solis
Subtle yet widespread disruptions in gene expression mark the brain in schizophrenia, according to a study of over 500 brain samples that was published in Nature Neuroscience on September 26. From the CommonMind Consortium, the study sequenced RNA messages to provide the most comprehensive picture yet of gene expression in the brain in schizophrenia. The effort linked 20 percent of genetic risk signals previously identified for schizophrenia to changes in gene expression, pegged multiple suspect genes with roles in brain development, and detected 693 genes that were expressed slightly differently in schizophrenia, ranging from 1.03- to 1.33-fold changes compared to controls.
These modest changes underscore the polygenic nature of schizophrenia, said study leader Pamela Sklar of the Icahn School of Medicine at Mount Sinai in New York City.
“These results are fully in line with everything else we’ve observed about the disorder,” Sklar said. “I think this is just a correlate of the fact that there are many genes that are influencing the disease, and many of them are doing it through subtle effects on gene expression.”
The CommonMind Consortium is a private-public partnership between pharmaceutical companies and academics that came together to create a resource of gene expression in the brain, and last year they publicly released their first installment of data. Others, such as the Genotype-Tissue Expression (GTEx) project, the PsychENCODE Consortium, and the BrainSeq Consortium, are building similar resources.
These collaborations have been propelled by the realization that to understand the function of the genes implicated in psychiatric disorders, researchers need to examine their regulation in human brain tissue itself. While researchers have been trying to pinpoint brain gene expression important to schizophrenia for over 20 years, no one has been able to look at such a large cohort of brain samples before. The new study also upped the ante by employing RNA sequencing to capture all transcripts, including novel isoforms, which are not detectable by microarray techniques. The study also surveyed the entire genome rather than focusing on a single region (see SRF story).
“This is an enormous effort to catalog that kind of data, and I commend them for putting it out because it’s hard,” said Beth Thomas of the Scripps Research Institute in La Jolla, California, who was not involved in the study. Thomas studies gene expression in neurological diseases such as Huntington’s disease, but also collaborates with an Australian group that works on schizophrenia.
Many of the challenges that beset interpretation of previous studies remain in the new study, however. Likening the new study to a remake of an old movie, Thomas said, “Did they solve the problem of what genes are important in schizophrenia? Probably not really, but this is a step in the right direction. We have to keep chipping away at this work to get our answer.”
Control knobs and fish heads
First authors Menachem Fromer, Panos Roussos, and Solveig Sieberts started with postmortem brain samples from the dorsolateral prefrontal cortex of 258 people with schizophrenia and 279 controls. RNA sequencing revealed gene expression levels in each sample, which were then adjusted for confounding variables such as the interval between death and brain collection, RNA integrity, and age at death.
This seemed to successfully avoid artifactual gene expression changes, said Mark Vawter of the University of California, Irvine, who was not involved in the study. Vawter has identified gene expression in postmortem brain that can be attributed to such confounds (Vawter et al., 2006).
The researchers identified a remarkable 2+ million expression quantitative loci (eQTLs), which are single nucleotide polymorphisms (SNPs) associated with a gene’s expression level. For example, having no G at a particular SNP might be associated with low transcript levels of a particular gene, having one G with intermediate levels, and having two G's with high levels.
These eQTLs may act as control knobs of gene expression and could explain how genetic risk for schizophrenia exerts its effect. In fact, the study found that 20 percent of risk regions flagged by the latest genomewide association study (GWAS) for schizophrenia (see SRF story) contained an eQTL.
Eight of these eQTLs were associated with a single gene, including FURIN, which encodes a protein processor; TSNARE1 and SNAP91, both components of presynaptic terminals; CNTN4, a member of an extracellular matrix protein family; and CLCN3, a chloride channel.
Some of these genes had effects on brain development, as explored in a zebrafish model: Suppression of FURIN and overexpression of TSNARE and CNTN4 resulted in smaller head sizes. In a stem cell assay, FURIN suppression also disrupted cell migration. Such assays could help prioritize genes for follow-up.
“We're not saying these are actually models of schizophrenia,” Sklar said. “But they’re certainly perfectly reasonable models of brain development.”
The researchers also detected 693 genes that were differentially expressed between schizophrenia cases and controls. Disappointingly, the results did not reproduce effects seen in previous studies—in terms of both the gene involved and the magnitude of the change.
“I wouldn't say our study calls into question everything that came before,” Sklar said. “Generally, a first study will overestimate an effect size, and over time with larger samples, you approximate a more realistic effect size. But it certainly means that not all of the previously reported differentially expressed genes are necessarily going to pan out.”
The authors argued that their modest differences were not unexpected, given the very slight differences in allele frequencies between cases and controls, which can be as little as 1-2 percent.
Even the single gene targets of eQTLs did not show significant differences between cases and controls—a finding that may cause some to call into question their role in schizophrenia risk. But computational modeling suggested that about 28,000 samples would be required to detect such small changes. Alternatively, it may be that timing rather than level of transcription is what matters to risk—something not captured by the study.
However, Thomas suggested that pooling a large heterogeneous set of brain samples may have diluted any signal. For example, Thomas’ group found that differences between schizophrenia cases and controls were larger when they sorted brains according to duration of illness (Narayan et al., 2008).
“What needs to be done is to find a way of teasing out the heterogeneity,” she said. “I’m not sure adding more samples is the answer to getting a meaningful gene expression change.”
Vawter also suggested that integrating information about a person’s load of genetic risk, as captured by the polygenic risk score, might help. “As people’s genotypes approach the higher end of polygenic risk, we may see better differentiation of cases and controls,” he said.
Vawter is using CommonMind Consortium data to investigate the role of mitochondria in schizophrenia, but he said it is slow going because of the BAM file format in which the data comes.
“I would urge everyone to please publish in FASTQ file format so people can hit the ground running,” he said, noting that his postdoc has had to spend months converting the files and aligning them to the same version of the genome before beginning analysis. “The current format creates a lot of extra work just to save a little bit of storage space.”
In the meantime, the CommonMind Consortium continues to amass samples and is looking at other brain regions as well as other mechanisms of gene regulation such as epigenetic modifications and RNA editing. “Our goal is to learn as much as we can on the same samples, so we can integrate this for a more holistic picture,” Sklar said. “This is just the first installment.”