Email Icon Facebook icon Twitter Icon GooglePlus Icon Contact

User Top Menu

Schizophrenia Genetics 2015—Part 2, From Discovery to Understanding

10 Aug 2015

In SRF's five-part 2015 schizophrenia genetics update, reporter Michele Solis surveys leaders in the field about milestones, challenges, and current research.

See Part 1, Renaissance; Part 3, Rare Allure; Part 4, Rethinking Diagnoses; and Part 5, Plan of Action.

Download a PDF of the entire series.


August 11, 2015. Last year's prized genomewide association study (GWAS) of schizophrenia identified 128 single nucleotide polymorphisms, or SNPs, that were found more often in people with schizophrenia than in controls (see SRF related news report). These identify regions of the genome that harbor common DNA changes that increase risk for the disorder, and still more are expected to emerge in a larger GWAS currently being conducted by the Psychiatric Genomics Consortium (PGC).

But scientists are not waiting for the complete gene catalog before exploring what the findings might mean. Drug companies are already drawing up plans for finding drug targets; analysts are looking for themes among the implicated genes; and researchers are turning to postmortem brains or neurons derived from stem cells for functional insights.

"The results are seeding all sorts of biomedical research here at Cardiff, ranging from brain imaging to animals to cellular studies," said Michael O'Donovan of Cardiff University in Wales, who chairs the PGC's schizophrenia group. "For the first time, people are in a position to probe the biological consequences of genetic clues that you can pretty much guarantee are related to psychiatric disorders."

But if getting GWAS off the ground was strenuous, interpreting their findings will be an even longer haul. Most of the hits lie in relatively uncharted non-coding "regulatory" regions that control gene expression, leaving researchers in the dark about which genes are influenced. The tiny effect sizes of these SNPs may also make it hard to identify their effects in biological assays. And even when a true risk variant is pinpointed within the GWAS-implicated heap, understanding its function and dysregulation in schizophrenia is hampered by the inaccessibility of brain tissue.

"The genetics of schizophrenia aren't particularly different from the genetics of other complex human diseases," said Mark Daly of the Broad Institute in Cambridge, Massachusetts. "What sets schizophrenia apart is lack of access to the tissue that matters."

To get around this, researchers will draw from multiple types of data: snapshots of gene expression during brain development, as put together from postmortem brain mapping; human cell models of neurons derived from stem cells that can, with new genome editing techniques, systematically test the effects of risk variants on cell function; network analysis to generate ideas about the biology that is derailed in schizophrenia.

Even with all these data, making final inferences will be tricky.

"The genetics is difficult, but it's going to work. Making lines of cells and differentiating them is going to be hard, but that's going to work. But what we really have to figure out is how to get from cells to people," said Steven Hyman, director of the Stanley Center for Psychiatric Research at the Broad Institute in Cambridge, Massachusetts.

Finding culprits among nominees

Though GWAS findings narrow in on the sectors of the three billion base pairs of the human genome that contain risk variants, they don't actually pinpoint these variants. That's because the 128 genomewide significant SNPs flag stretches of DNA that tend to stick together during recombination—that is, they are in "linkage disequilibrium" with each other. These tight stretches may contain many genes, which constitute nominees for schizophrenia's risk factors.

Getting to the actual culprits will require fine mapping of these regions to find the causal variant driving the genomewide-significant signal. Doing this, however, has been harder than expected. Of thousands of associations found by GWAS for other diseases, only a handful have been tracked to a causal variant.

This may be because common variants in a locus have such subtle effects that they're hard to recognize. Another possibility is that rare variants in a locus may be driving GWAS signals, a situation referred to as a synthetic association (Dickson et al., 2010). However, some researchers argue this does not explain the bulk of the schizophrenia GWAS results (Wray et al., 2011).

"Synthetic association has not proved a fruitful resolution to fine mapping efforts," Daly said, noting that more comprehensive scans of less common alleles find that they don't account for GWAS signals as well as common ones do. This doesn't rule out the possibility that independent rare variants in the same genes implicated by GWAS will turn up, however.

New statistical methods are making headway in fine-mapping the thorniest region, the major histocompatibility complex (MHC) region on chromosome 6. This region contains 200 genes, yet has resisted analysis because it is replete with linkage disequilibrium. But recent conditional analyses of the region that take into account correlations between different SNPs, or different haplotype structures, have narrowed in on several independent signals (for details, see SRF related conference report).

Another way to fine-map GWAS loci looks to the genomes of non-European ethnicities, which have not so far contributed much to genetic studies. For example, Sub-Saharan Africans have the "oldest" genomes, with smaller haplotype blocks. Finding a GWAS-fingered SNP within one of these smaller blocks would constrain the risk variant to a smaller region. The Stanley Center has begun a series of collaborations to collect DNA from diverse populations around the world, including Africa, Japan, and Mexico.

These approaches won't deliver an unequivocal causal variant for each and every locus, however, which means researchers will have to take indirect paths toward understanding the GWAS results. One option is to do targeted sequencing of the genes influenced by these variants to provide a sense of naturally occurring variation that would likely include risk factors for human diseases.

But much of the trouble in interpreting GWAS stems from an incomplete grasp of the human genome and its component parts. Most of the GWAS clues for schizophrenia lie outside of genes, in non-coding DNA that is thought to control how, when, and where different genes are turned on or off. This casts schizophrenia as a disorder of perturbed gene expression rather than one of broken proteins.

The genome's "control panel" is just starting to come to light through projects such as the Encyclopedia of DNA Elements (ENCODE) project, which identified regulatory elements that control gene expression such as promoters that provide landing pads for transcription machinery (see SRF related news report). More recently, the Roadmap Epigenome Mapping Consortium identified the DNA modifications in many cell types that flagged enhancers, short stretches of DNA that can increase gene transcription, even from a distance. A more brain-focused initiative called PsychENCODE began in 2013 to examine the epigenomic landscape of neurons taken from postmortem brain samples, including those from people with schizophrenia. Together, these efforts will create a picture of how transcription unfolds in different cell types at different times, which will help researchers pick out any disease-related irregularities.

Mapping the brain's molecular landscape

With risk for schizophrenia embodied in the ups and downs of gene expression, the brain's profile of transcripts will be a critical touchstone for interpreting GWAS results.

"If the change is not in the amino acid sequence, it has to be in the regulation of gene expression by whatever mechanism—epigenetic, microRNA, non-coding RNA, promoters, enhancers—pick your mechanism; it doesn't matter " said Daniel Weinberger of the Lieber Institute for Brain Development in Baltimore, Maryland. "It's in the transcript—it has to be in the transcript."

If schizophrenia has its roots in early brain development, as is generally accepted, then the critical, risk-related perturbations to gene expression might only be glimpsed in a subset of neurons in fetal brain tissue. Databases of gene expression in the human brain are being generated across the lifespan for different brain regions (e.g., BrainSpan; BrainCloud), composed of data from either microarrays that probe for specific transcripts, or from RNA sequencing, which captures all transcripts, many new.

These data can be mined to find SNPs associated with expression, known as eQTLs (expression quantitative trait loci). Finding overlaps between human brain-specific eQTLs, regulatory elements such as enhancers or promoters, and schizophrenia-associated SNPs can help prioritize which genetic lead to follow up. This strategy showed researchers that a GWAS-implicated SNP in an intron of CACNA1C, a calcium channel subunit, flagged a nearby enhancer. The enhancer interacted with a promoter some distance away, and the risk variant resulted in decreased expression of CACNA1C in vitro (Roussos et al., 2014).

Transcriptome profiling in postmortem brains may help find genes hiding in GWAS signals, too. The Lieber Institute Pharma RNA-seq Consortium, a collaboration between the Lieber Institute and multiple pharmaceutical companies to sequence the transcriptomes of 1,600 brains, recently reported that genomewide-significant SNPs in a region of chromosome 10 all regulated expression of one gene nearby (for details, see SRF related conference report).

Eventually, postmortem brain studies may also implicate specific circuits or cell types involved. Studies of gene expression networks in autism have pointed to perturbations in excitatory neurons in the deep layers of cortex during mid-fetal development (Willsey et al., 2013). Getting ever more granular, a technology called drop-sequencing will allow high-throughput single-cell transcription profiling, which could lead scientists precisely to the altered circuits in mental disorders (Macosko et al., 2015; Klein et al., 2015).

Stem cells and networks

With hundreds of variants of interest expected for schizophrenia, the field also needs a high-throughput scheme to analyze the biological effects of each variant. To do this, stem cells—not knockout animals—are in the front lines. Currently, the Stanley Center is constructing libraries of well-characterized embryonic stem cell and induced pluripotent stem cell (iPSC) lines that carry individual, or combinations of, risk alleles introduced through genome editing technology.

These cell lines can be made into neurons, and researchers plan to systematically compare neurons with a risk allele to those without it, but otherwise genetically identical. Because synapses are thought to play a role in schizophrenia risk, these neurons can also be coaxed to form brain "organoids" that recapitulate some aspects of neural organization, including synaptic connections between neurons (see SRF related news report).

"Undoubtedly, we will find cellular phenotypes, but the hard, hard problem in all of this is knowing what these cellular phenotypes mean for humans," Hyman said.

The emphasis on human cell models not only comes from recognition that regulatory, transcription-controlling elements are poorly conserved across species, but also from past disappointments in which insights based on rodent models did not apply to humans.

This "translational valley of death" puts a premium on human data when selecting potential drug targets, said Nick Brandon of AstraZeneca in Cambridge, Massachusetts, which is collaborating with the Lieber Institute to develop human stem cells for drug discovery. He sees human cell models taking a leading role in probing compounds for drug responses or efficacy, leaving animal models for testing drug safety and specific mechanisms of action.

"We absolutely still need animal models," he said. "But we just have to take the animals for what they are and not overinterpret them. We need to focus on what they are telling you about particular mechanistic aspects of your target or your compounds."

Another strategy to probe function relies upon bioinformatics, which offers tools to search for common themes among the genes nominated by GWAS. Called pathway or network analysis, this approach asks whether these genes are enriched within specific biological processes, or pathways.

Although the field has not yet settled on standard methodologies, the approach could highlight how variants of small effect flag problems in critical biological processes, which as a whole may be better therapeutic targets than any individual risk variant.

A pathway analysis of the PGC's earlier round of hits highlighted histone methylation, which controls gene expression, synapses, and immune signaling (see SRF related news report). These are very general processes that impinge upon all of development, however, leaving any therapeutic insights unclear. Adding more genomewide-significant hits to the analysis might lead to more specific components of these pathways, or schizophrenia's heterogeneity could stymie that hope.

"It's my own personal prejudice that there are lots of ways the brain can go wrong to lead to the disorders we currently call psychosis. If true, then it's quite likely there will be a lot of fairly general pathways," said Cardiff's O'Donovan.

Where do the old candidates stand?

As the new, GWAS-sanctioned genes capture everyone's attention, what to make of those old, pre-GWAS era candidates? Genes such as neuregulin (NRG1), dysbindin, and catechol-O-methyltransferase (COMT) were identified by linkage in smaller family studies or by candidate gene studies that explored a gene based on ideas about its role in schizophrenia biology. Except for a hit near the DRD2 gene, which encodes the D2 subtype of the dopamine receptor, the primary target of antipsychotic drugs, these old candidates don't turn up in the PGC's GWAS. Though future GWAS iterations may eventually bring to light some of these old favorites, opinions vary on how to reconcile them with the new data.

"The way to reconcile them is that they don't reconcile," said Thomas Lehner of the National Institute of Mental Health (NIMH) in Bethesda, Maryland, noting the exception of the MHC region, which was immediately recognized in early linkage studies. A recent survey shares this sentiment (Farrell et al., 2015).

Others say that the GWAS design only rules things in rather than out. "What we can say is that the data and the types of analyses we've done don't particularly favor any of the old candidates," O'Donovan said. "But that's a very different statement from saying the old candidates aren't involved."

Citing NRG1 as an example, O'Donovan suggested the old candidates may reside in complicated haplotypes that SNP arrays used in GWAS do not capture well.

Another idea is that the old candidates may be context dependent, so that a given gene doesn't produce a signal unless it combines with a particular environment or genomic background—effects of which would be lost in the giant samples needed for GWAS. The PGC's latest GWAS did not find evidence for gene-gene interactions (called epistasis) between pairs of risk SNPs, but this does not rule out the existence of interactions among three or more genes.

"We're going to find that the reason some of these candidate genes didn't stay significant was because they were critically dependent on other factors that varied tremendously across different samples," Weinberger said.

Brandon, who has pursued many of these previous candidates, feels they still have a lot of mileage left in them. "We would be foolish to throw these away," he said, adding that AstraZeneca is keeping them in mind as the new data emerge. "We are certainly trying to be inclusive of these prior findings because you never know when things interact."

Our next story in the Schizophrenia Genetics 2015 series—"Rare Allure"—will detail the hunt for another category of possible contributors to liability for schizophrenia: the rare and elusive genetic glitches that could have robust effects on risk.—Michele Solis.

See Part 3, Rare Allure.