The train of discovery powered by genome-wide association studies (GWAS) is losing steam, so researchers are turning to a different experimental fuel to drive progress on the genetics of Alzheimer's and Parkinson's diseases. The sense of slowing GWAS returns and accelerating alternatives were recurring themes at joint Keystone meetings called Alzheimer's Disease—From Fundamental Insights to Light at the End of the Translational Tunnel and Parkinson’s Disease: Genetics, Mechanisms and Therapeutics, Symposia. Both were held 2-7 March in Keystone, Colorado.
Andrew Singleton of the National Institute on Aging, Bethesda, Maryland, set the tone with his estimation that one-third of PD risk is genetically heritable, yet only about 10 percent of the responsible genes are identified (see Keller et al., 2012). GWAS have unearthed common variants that pose low risk, Singleton said, but fall short when it comes to discovering rare variants that pose higher risks. Furthermore, for each locus that pops up in a GWAS, researchers have to parse the one among dozens of genes nearby that is affected by the variation.
Reinforcing Singleton, Thomas Gasser of the University of Tübingen, Germany, said that as GWAS continue to grow larger, researchers reap smaller rewards. Gasser presented unpublished findings from a recent GWAS doozy, which included nearly 14,000 PD patients. The study yielded 28 PD-associated loci, six of which were new, he said. “Larger sample sizes yield more hits, but at some point it starts to level off,” Gasser noted, pointing to a plateauing graph of GWAS hits versus sample size. “It may not make much sense to keep going higher.” Furthermore, all 28 loci Gasser identified pose only a combined 3.3-fold elevated risk for PD—a level he called “sobering.”
AD geneticists may have plucked all the low-hanging fruit, as well. GWAS by Gerard Schellenberg, University of Pennsylvania, Philadelphia, and colleagues, which together comprised more than 70,000 people, turned up 22 AD risk variants of which 11 were new (see Oct 2013 news story). However, Schellenberg said their GWAS has not yet outlived its usefulness; it can be combined with other techniques that increase power to uncover elusive genotypes. "We rely on imputing to determine genotypes," he told Alzforum. Geneticists use imputation, which is based on the knowledge that certain sequences are usually co-inherited, to fill in missing sequence data. Researchers have incorporated data from the 1,000 Genome Project for GWAS imputation, but this dataset contains sequencing information from 14 different ethnic backgrounds, meaning any one population is represented only by a few hundred sequences or less (see Nov 2012 news story). "Once we have 25,000 genomes sequenced, we can better impute, and that will help us find rare variants," said Schellenberg.
Nevertheless, Richard Mayeux, Columbia University, New York, noted that the field is now leaning more on whole-genome and whole-exome sequencing. For example, last year exome sequencing identified variants in the phospholipase D3 gene that double the risk for AD (see Dec 2013 news story). Mayeux and Schellenberg are on the steering committee for the Alzheimer's Disease Sequencing Project (see Dec 2013 news story). It will sequence the genomes of 566 people in 111 families plus exomes in 11,000 other individuals, including 500 cases and the same number of controls, plus an additional 1,000 people from affected families. Mayeux said that using this data, his group has found six loci that may explain why the age at onset ranges so widely, from 45 to 75 years, in people who carry the G206A presenilin 1 mutation.
Other researchers have begun combining genetics with other measures to hunt risk alleles. In a short talk, Philip De Jager, Brigham and Women's Hospital, Boston, outlined an epigenomics approach to identifying risk factors for AD. De Jager, together with David Bennett at Rush University, Chicago, has correlated DNA methylation and micro RNA expression in the dorsolateral prefrontal cortex with the presence of amyloid plaques in the brain. They determined DNA modification, postmortem, at more than 400,000 CpG-rich, methylation-susceptible sequences in people from the Religious Orders Study and the Memory and Aging Project being conducted at Rush. Of 740 participants, almost half had been diagnosed with AD, a quarter had mild cognitive impairment, and the others were cognitively normal when they died. In people with plaque pathology, 117 regions were differentially methylated and 13 nearby genes were differentially expressed, De Jaeger said. A replication study using samples from a Mayo Clinic cohort confirmed the association for seven of those genes: Bin 1, ACACB, AP3M2, DDB1, DDOST, DLL1, and Slc17a7. A Bin1 single-nucleotide polymorphism (SNP) had previously come up in GWAS studies, but De Jager said that was unrelated to methylation changes. Interestingly, these methylation and expression changes occurred in asymptomatic individuals, suggesting an early event in disease.
Using a similar approach, De Jager correlated plaque pathology with micro RNA expression. Postmortem tissue samples from 711 people revealed that 25 of 309 miRNAs tested were expressed differently in people with amyloid pathology than in those without. Analysis of target genes for those miRNAs suggests that they may be up- or downregulated, as well. Combining this with the methylation data, the researchers are building a network of methylation sites and miRNAs that drive expression changes associated with pathology. De Jaeger said that miR-1260 emerged as a master regulator. Keystone attendees considered this approach attractive for other pathologies, such as Lewy bodies or TDP-43 inclusions.
What about known risk variants? Scientists at the meeting agreed that they need to better understand how those lead to disease. Investigators presented clues obtained by weaving in data from protein interaction networks and expression analysis. “Genetics doesn’t need to happen in a vacuum anymore,” said Singleton. “Integrating data is key to proving association and understanding function.”
In some cases, the integration starts with digging deeper into the data already available. “GWAS filter out all but the most robust hits,” Gasser told Alzforum, explaining that among the genes that don’t make the grade could lie true associations that act within the guise of a larger network. Sifting through 249 near-misses from the GWAS conducted by the International Parkinson’s Disease Genomics Consortium (IPDGC), Gasser found that half of them were related to the immune system. This allowed his team to generate an “inflammatory subtype” based on 100 SNPs and use that to stratify PD patients. The scientists subsequently screened the blood of 300 PD patients from the GWAS. They found higher levels of the key inflammatory cytokine IL-6 in patients harboring the inflammatory genetic fingerprint than in those who did not, indicating that the genetic association translated into a phenotype. While the findings are preliminary, Gasser said such experiments have the potential to group patients into different disease categories that could lead to more personalized treatment. “We’re not going to look at PD patients as one big pool anymore,” Gasser told Alzforum. His lab is currently stratifying patients based on a genetic subtype of mitochondrial dysfunction, as well.
Besides tapping networks to link GWAS hits to functional changes, researchers can learn from focusing on individual hits. For example, GTP cyclohydrolase 1 (GCH1), a gene involved in the production of dopamine and linked to dopamine-responsive dystonia (DRD), popped up as a risk factor in the IPDGC GWAS. As with a majority of those hits, the GCH1 variant fell into a non-coding region. However, comparing exome-sequencing data from PD patients and controls, researchers led by Niccolo Mencacci at University College London unearthed 10 rare variants within GCH1’s coding region. The new finding strengthens the idea that defects in dopamine production are a cause of PD, rather than just a symptom, in some people, Gasser said.
With GWAS hits in non-coding regions and rare variants in the coding region, the genetic architecture of GCH1 represents the norm for PD risk factors, Gasser said. Variants in non-coding regions likely control expression levels of nearby genes, but how they modulate expression or even which genes they affect can’t be determined by GWAS data alone.
“There are probably 20 to 30 loci from PD GWAS studies, each one lying near 20 to 30 genes,” commented Mark Cookson of the NIA. “So we have to ask: What’s our best shot?” Answering that question will require smart priorities, Cookson said. Schellenberg agreed this will be challenging. Of the AD GWAS hits to date, 90 percent are in non-coding regions, and it will be difficult to predict which genes they control. “We know that only 20 percent of regulatory elements affect the closest gene,” he said. On top of that, while the mean distance between regulatory element and target gene is about 120 kilobases, it can be 1.4 megabases or farther, said Schellenberg. Layering on more complexity, he reminded the audience that multiple enhancers can affect a given gene and a single enhancer can affect multiple genes, not always to the same extent. "The challenge of translating GWAS data is to identify which gene is relevant and determine whether risk variants increase or decrease expression," he said.
Taking a stab at this, Vincent Plagnol of University College London studied changes in gene expression associated with 16 established PD risk loci identified by GWAS. Using microarray analysis of brain samples from healthy controls, Plagnol searched for expression changes of genes thought to be under control of the GWAS SNPs. Of the 16 non-coding hits, only three correlated with nearby expression quantitative trait loci (eQTLs), which are genomic regions that regulate gene activity. In other words, three of 16 GWAS hits —RAB7L1, SPPL2B, and GPNMB—affected expression of nearby genes. However, Plagnol cautioned that his study was underpowered to find eQTLs that may affect expression of genes farther away. Larger sample sizes and RNA sequencing methods, rather than eQTL screening, would be necessary to uncover such variants (see Sept 2013 news story).
EQTL studies should serve as a prerequisite for moving forward with more costly functional studies, Plagnol said. “We shouldn’t base functional studies on GWAS hits alone,” he told Alzforum. Plagnol urged labs to readily share eQTL data.
This kind of work is crucial for understanding genome-wide associations, said Patrick Lewis of the University of Reading in England, who co-organized the Parkinson's disease meeting. It adds weight to various GWAS signals, and moves us closer to figuring out what they’re doing at the cellular level, he said.
The non-coding GWAS hits associated with LRRK2 and SNCA did not turn up in Plagnol’s eQTL screen. This is disappointing because those two genes are also known for rare Mendelian coding variants that cause PD. However, in a recent study, Plagnol did identify two exons in LRRK2 that appeared to be upregulated in people carrying a nearby non-coding variant not associated with PD (see Trabzuni et al., 2013). Why only two of LRRK2’s 51 exons were upregulated by the variation remains to be seen, Plagnol said.
“We are struggling to understand the genome-wide association at the LRRK2 locus,” said Lewis, who was a co-author on Plagnol’s LRRK2 mapping study. “It doesn’t seem to be an eQTL. It may be a splicing locus, but at the moment that really isn’t clear.”
Cookson commented that because the brain expresses low levels of LRRK2, correlating it to eQTLs could prove difficult. However, he noted that the expression analysis is a mathematically rigorous way to prioritize variants, and could bring researchers closer to understanding which ones play a role in disease.
For his part, Cookson took a different approach to clarify LRRK2’s murky role in PD. He presented recently published data from an unbiased screen of LRRK2 protein-interaction partners. It pinpointed Rab7L1, a previous GWAS hit and a gene Plagnol had tagged as an eQTL (see Feb 2014 news story). Cookson identified other proteins that interact with LRRK2 and Rab7L1: GAK, Hsc70, and BAG5. All play a role in the clearance of vesicles in the trans-Golgi network. The lab is now seeding protein arrays with LRRK2’s interaction partners to expand the network. The synergy between GWAS, eQTL, and protein-interaction studies in identifying Rab7L1 gave Cookson hope for progress. “Rab7L1 is a pretty good candidate in that locus. When we combine these multiple prioritization strategies, something will rise to the top,” Cookson said.—Jessica Shugart and Tom Fagan.
- Paper Alert: New Alzheimer’s Genes Published
- Genetics Project Update: Over 1,000 Genomes and Counting
- Phospholipase D3 Variants Double Sporadic AD Risk
- Alzheimer’s Whole-Genome Data Now Available From the NIH
- RNA Sequencing Helps Identify Functional Variants from GWAS
- LRRK2 Interactions Identify New Parkinson’s Genes, Implicate Autophagy
- Keller MF, Saad M, Bras J, Bettella F, Nicolaou N, Simón-Sánchez J, Mittag F, Büchel F, Sharma M, Gibbs JR, Schulte C, Moskvina V, Durr A, Holmans P, Kilarski LL, Guerreiro R, Hernandez DG, Brice A, Ylikotila P, Stefánsson H, Majamaa K, Morris HR, Williams N, Gasser T, Heutink P, Wood NW, Hardy J, Martinez M, Singleton AB, Nalls MA. Using genome-wide complex trait analysis to quantify 'missing heritability' in Parkinson's disease. Hum Mol Genet. 2012 Nov 15;21(22):4996-5009. Epub 2012 Aug 13 PubMed.
- Trabzuni D, Ryten M, Emmett W, Ramasamy A, Lackner KJ, Zeller T, Walker R, Smith C, Lewis PA, Mamais A, de Silva R, Vandrovcova J, , Hernandez D, Nalls MA, Sharma M, Garnier S, Lesage S, Simon-Sanchez J, Gasser T, Heutink P, Brice A, Singleton A, Cai H, Schadt E, Wood NW, Bandopadhyay R, Weale ME, Hardy J, Plagnol V. Fine-Mapping, Gene Expression and Splicing Analysis of the Disease Associated LRRK2 Locus. PLoS One. 2013;8(8):e70724. PubMed.
No Available Further Reading