If the mountain of unpublished genome-wide association studies on BioRxiv are any indication, Alzheimer’s scientists have a nice new set of risk loci to explore. Three large AD GWAS unearthed 13 new loci by expanding upon the previous GWAS made possible by the International Genomics of Alzheimer’s Project (IGAP). One of the new studies, led by a cadre of IGAP researchers, added more cases and controls to the previous sample. The other two were co-led by Stephan Ripke of Massachusetts General Hospital in Boston, Ole Andreassen of the University of Oslo, and Danielle Posthuma of VU University in Amsterdam; and Riccardo Marioni of the University of Edinburgh and Peter Visscher of the University of Queensland in Australia. Both grew their sample size by tapping into more than a quarter-million genetic samples in the UK Biobank. In an approach called GWAX, parental diagnosis, aka “AD-by-proxy,” was taken as a stand-in for AD cases. Together, the new hits strengthen the case for inflammation and lipid metabolism, but now also point to APP processing and tau-binding proteins as key orchestrators of LOAD.

  • Three Alzheimer’s GWAS posted on preprint server await formal publication.
  • Combined, they have been read nearly 3,000 times already.
  • They revealed 13 novel risk loci, including ADAM10, ADAMTS1/4, ACE, and APH1B.

“These are exciting manuscripts that move the field a step forward,” commented Jose Bras of University College London. “It is very positive that all were initially submitted to BioRxiv, making them open to everyone without publication delays. I hope we continue to see this for large-scale genetic studies.” The three manuscripts were posted January 15, February 20, and April 4, 2018; between them, their full pdfs have been viewed 2,800 times and tweeted widely while they undergo peer review.

"The current set of GWAS-related studies indicates that there are still significant loci to find for Alzheimer’s disease, and likely for other neurodegenerative conditions, as sample sizes increase,” commented Mark Cookson, National Institute on Aging, Bethesda, Maryland.

The largest published AD GWAS to date identified 19 AD risk loci (Jul 2013 conference news and Oct 2013 news). That landmark study—itself a meta-analysis of four individual GWAS including 17,000 AD cases and 38,000 controls—was the product of a collaborative effort by IGAP. The loci it found were estimated to account for about a third of the overall genetic burden of AD. To unveil more variants, especially rare ones with larger effect sizes, geneticists have since used whole-genome and -exome sequencing (Aug 2017 news). Another approach—exemplified by the “studies in waiting” on  BioRxiv —is to keep beefing up GWAS sample size.

The IGAP2 GWAS on BioRxiv was co-led by eight authors, including Jean-Charles Lambert of Institut Pasteur de Lille, France, and Margaret Pericak-Vance of the University of Miami, Florida. Co-first authors Brian Kunkle of U Miami and Benjamin Grenier-Boley at Pasteur acquired more samples from AD cases and controls, then pooled them with their previous GWAS. The resulting analysis included 21,982 cases and 41,944 controls, topping their previous sample by 29 and 13 percent, respectively. They investigated nearly 10 million common variants and more than 2 million rare ones for association with AD risk—up more than 60 percent from before. This discovery phase preceded replication in two additional cohorts, and the meta-analysis of all three cohorts reached a grand total of 89,769 samples.

The scientists confirmed association with AD in 18 out of 19 previously identified loci, as well as two loci found in other studies—the rare R47H variant in TREM2 and a variant in the ECDH3 locus. Four new loci came up: IQCK, ADAMTS1, ACE, and ADAM10. The AD risk variants landed in intergenic regions, so the researchers tried to narrow down the list of nearby genes for ones that might be responsible. They considered expression quantitative trait loci (eQTL), which are changes in gene expression associated with a given variant, in AD-relevant tissues and cell types, and they looked for differential expression of genes in postmortem brain tissue from AD patients versus controls. Ultimately, they presented this list of risk loci and candidate genes:

What about those? The metalloproteinase ADAM10, aka α-secretase, promotes non-amyloidogenic processing of APP and sheds the extracellular domain of TREM2. ADAMTS1, a different class of metalloproteinase, is more highly expressed in people with Down’s syndrome who have AD, and implicated in neuroprotection and -inflammation (Kuno et al., 1997; Wilcock, 2012; Gurses et al., 2016). ACE (angiotensin converting enzyme) expression in AD brain tissue is linked to Aβ load and AD severity, and ACE variants have been associated with AD risk in Wadi Ara, an Israeli Arab community with high rates of AD (Miners et al., 2009; Meng et al., 2006). PSMC5 is a proteasome subunit and ATPase that regulates expression of the major histocompatibility complex (Inostroza-Nieves et al., 2012), and CD79B is a receptor on B lymphocytes. Little is known about the function of IQCK or DEF8, but GPRC5B is a regulator of neurogenesis and obesity-mediated inflammation (Kurabayashi et al., 2013; Kim et al., 2012). 

The researchers also looked for biological pathways encompassing the risk loci. When including only common variants in the analysis, they found four functional clusters: APP metabolism/Aβ formation, tau protein binding, lipid metabolism, and immune response. This was the case even after removing ApoE from the analysis. Rare variants were not significantly over-represented in any biological pathway, however, lipid and Aβ pathways were nominally enriched with them. Notably, the pathways for common and for rare variants correlated, which suggests rare variants indeed influence AD-relevant biology.

“The immune system and lipid metabolism continue to come up as pathways strongly associated with AD, but now there’s data suggesting the involvement of APP metabolism and tau binding proteins,” commented Bras. “The former is particularly interesting since it links early and late-onset AD to the same mechanism—a finding that recent analyses from DIAN and ADNI have also shown.”

UK Biobank Yields Many Proxy ‘Cases’
Two other GWAS took a different approach to boost sample size. They designated as cases “AD-by-proxy,” aka people with one or both parents diagnosed with AD. This allowed them to tap into the massive UK Biobank data set, which includes parental AD diagnosis of its participants. The validity of this genome-wide association-by-proxy approach, dubbed “GWAX,” was established using the UK Biobank cohort for 12 diseases, including AD (Liu et al., 2017). 

The larger of the two GWAX, led by Ripke, Andreasson, and Posthuma, posted on BioRxiv on February 20. It unfolded in three stages. First came meta-analysis of three cohorts of clinically diagnosed AD cases and controls—IGAP’s 2013 GWAS, the Psychiatric Genomic Consortium (PGZ-ALZ), and Alzheimer’s Disease Sequencing Project (ADSP)—totaling some 24,000 AD cases and 55,000 controls. Stage 2 was the UK Biobank GWAX, which included 74,793 AD-by-proxy cases and 328,320 controls. Stage 3 was the meta-analysis of stages 1 and 2, bringing the total sample size to more than 455,000 people. The scientists replicated their findings in DeCODE, an independent Icelandic cohort with more than 6,500 AD cases and 174,000 controls.

In all, first author Iris Jansen and colleagues found 29 AD risk loci. Sixteen had been reported in the 2013 IGAP study. Thirteen had not but four of those, including CD33, had popped up in other studies. This left nine new loci to explore. Most of these SNPs resided in non-coding regions with an open chromatin state, implying an activating effect on transcription. To pare down the list of potentially causal genes nearby, the researchers conducted functional analyses, including eQTL, to look for genes expressed in AD-relevant tissues, as well as chromatin interaction mapping. This uncovered multiple potential genes per SNP. The researchers drew attention to three interesting genes associated with the novel loci: ADAM10, which also came out of Kunkle et al., ADAMTS4, and APH1B. ADAMTS4 is yet another metalloproteinase implicated in AD pathogenesis; APH1B is a sub-unit of the γ-secretase enzyme complex that processes APP to generate Aβ.

The scientists identified lipid metabolism and APP processing as AD risk pathways. Genes that drove these pathway associations were expressed highly in immune tissues, including whole blood, spleen, and lung, as well as in microglia.

The other UK Biobank-based study, led by Marioni and Visscher, posted on BioRxiv on January 15. These researchers conducted an AD-by-proxy GWAX from the UK Biobank samples, then meta-analyzed it with the 2013 IGAP GWAS. Because this study did not use samples from the PGZ-ALZ, ADSP, or DeCODE, it was slightly smaller than Jansen et al.

The researchers found 24 AD risk loci, six of which were novel. One—ACE—later also popped up in Kunkle et al.’s expanded IGAP GWAS, while three others—ADAM10, ADAMTS4, and CLNK—also emerged from Jansen et al. The other two loci were CCDC6 and PLCG2. Interestingly, two of the genes nearest these loci—CLNK and PLCG2— interact in a signaling cascade implicated in immune function, and rare variants in PLCG2 are reportedly protective against AD (Aug 2017 news). 

John Hardy of University College London commented that by boosting the sample size of established AD GWAS, the two UK Biobank studies identified new provisional AD risk loci that await further confirmation. “Both studies use family history of dementia as part of their analysis. This is clearly imperfect, as the authors realize, but it does add power, and the fact that just ‘AD by proxy analyses’ find the established GWAS hits gives some justification for this approach,” he wrote.

Hardy flagged a potential problem, which he called the “audit” issue. “It is getting increasingly difficult to know whether individuals, in either the control or disease category, have been entered into the combined analyses more than once, and the audit trail is extremely difficult to follow,” he wrote. But despite these caveats, Hardy said the main findings were clear: “Microglial activation and lipid metabolism as well as APP metabolism, repeatedly show up as the incriminated pathways to disease.”

Bras raised a similar issue. “Since some of the data sets are shared as summary statistics, is it plausible that the same samples are included in multiple studies, which may become problematic, particularly for the analysis of lower-frequency variants,” he wrote. “Also, the overlap in findings isn’t complete, which may suggest that some of these hits are false positives.” He added that given the enormous size of the studies, it may not be feasible to truly replicate them.

“This expansion of analyses to a larger data set identifies more loci and confirms the enrichment of microglial-expressed genes involved in lipid metabolism/innate immunity. In addition, these papers identify variants linked to APP metabolism,” commented Alison Goate of the Mount Sinai School of Medicine in New York. She noted that because the studies used largely overlapping samples, it is no surprise that many of the new loci overlapped, or were part of the same biological pathways.—Jessica Shugart

Comments

  1. A general note on these three manuscripts is that it is very positive that they were initially submitted to BioRχiv, making them open to everyone without publication delays. I hope we continue to see this for large-scale genetic studies.

    These are exciting manuscripts that move the field a step forward. The AD-by-proxy approach seems to have validity in the UK Biobank data set, showing high genetic correlation estimates with clinical AD. This opens the door to studies with sample sizes that otherwise would be years away from being possible to achieve by any one group.

    There are some aspects that should be kept in mind, however: Independent replication of new loci becomes a tricky issue when one uses over 400,000 samples, simply because there are no other data sets of similar size. Since some of the data sets are shared as summary statistics, is it plausible that the same samples are included in multiple studies. This may become problematic, particularly for the analysis of lower-frequency variants. Also, the overlap in findings isn’t complete, which may suggest that some of these hits are false positives.

    The immune system and lipid metabolism continue to come up as pathways strongly associated with AD, but now there’s data suggesting the involvement of APP metabolism and tau binding proteins. The former is particularly interesting, since it links early and late-onset AD to the same mechanism—a finding that recent analyses from DIAN and ADNI have also shown.

    These are remarkable studies and it is very exciting to see these novel approaches being applied to the research in dementia.

  2. Jansen et al. and Marioni et al. add new samples to the established Alzheimer GWAS, and by that means they increase the (n) of these studies and lead to the provisional identification of new loci for the disease. Both studies use family history of dementia as part of their analysis. This is clearly imperfect, as the authors realize, but it does add power, and the fact that just “AD by proxy analyses” find the established GWAS hits gives some justification for this approach.

    A problem with these analyses is what I would call the “audit” issue. It is getting increasingly difficult to know whether individuals, in either the control or disease category, have been entered into the combined analyses more than once, and the audit trail is extremely difficult to follow. Clearly, if some cohorts are being counted twice, this makes one less confident in the outcome. Because of the challenges in accessing full data on each sample, it is difficult to assess on an individual sample basis, or even on a cohort basis.  

    However, the main findings continue to be clear: Microglial activation and lipid metabolism as well as APP metabolism, repeatedly show up as the incriminated pathways to disease. Of course, both lipid metabolism and microglial activation are recognized to be very broad terms, and defining which aspects of these complex (and probably interrelated) pathways are at play is clearly an important task.  

    The next task, one presumes, is to put these two studies together and do yet another audited meta-analysis. This likely will push even more loci from the same pathways over the magical Bonferroni correction for statistical significance and declaration.

  3. The current set of GWAS-related studies indicates that there are still significant loci to find for Alzheimer’s disease, and likely for other neurodegenerative conditions, as sample sizes increase. By definition, these will be loci with smaller effect sizes than those previously characterized, usually changing risk of disease by a few percent, but they are important to document as they should lead us to a deeper understanding of the gene pathways underlying neurodegeneration.

    The current set of papers supports this contention in that not only are new replicated gene candidates such as ADAM10 nominated, but also a previously discussed enrichment of immune-related genes is confirmed in these larger studies. Taken together, these studies strongly suggest that there are potential therapeutic targets around inflammation that might be helpful in sporadic AD.

    The one caveat that I would identify to modify this strong conclusion—and this is a generic concern with GWAS—is that while these groups use sensible strategies to identify best candidate genes at each locus, in some ways this might give a false sense of precision. Due to linkage disequilibrium, GWAS identifies groups of SNPS across each locus. Therefore, the pathway-based analyses are provisional but promising.

  4. To read the paper on bioRxiv, click here

    View all comments by Lucia Huntington
  5. To read the paper on bioRxiv, click here

    View all comments by Lucia Huntington
  6. To read the paper on bioRxiv, click here

    View all comments by Lucia Huntington

Make a Comment

To make a comment you must login or register.

References

News Citations

  1. Pooled GWAS Reveals New Alzheimer’s Genes and Pathways
  2. Paper Alert: New Alzheimer’s Genes Published
  3. The Search for the Missing AD Heritability Turns Up New Rare Variants
  4. Searching for New AD Risk Variants? Move Beyond GWAS

Alzpedia Citations

  1. ADAM10
  2. TREM2

Paper Citations

  1. . Molecular cloning of a gene encoding a new type of metalloproteinase-disintegrin family protein with thrombospondin motifs as an inflammation associated gene. J Biol Chem. 1997 Jan 3;272(1):556-62. PubMed.
  2. . Neuroinflammation in the aging down syndrome brain; lessons from Alzheimer's disease. Curr Gerontol Geriatr Res. 2012;2012:170276. PubMed.
  3. . Pathophysiological Function of ADAMTS Enzymes on Molecular Mechanism of Alzheimer's Disease. Aging Dis. 2016 Aug;7(4):479-90. Epub 2016 Jan 11 PubMed.
  4. . Angiotensin-converting enzyme levels and activity in Alzheimer's disease: differences in brain and CSF ACE and association with ACE1 genotypes. Am J Transl Res. 2009;1(2):163-77. PubMed.
  5. . Association of polymorphisms in the Angiotensin-converting enzyme gene with Alzheimer disease in an Israeli Arab community. Am J Hum Genet. 2006 May;78(5):871-7. PubMed.
  6. . Role of Sug1, a 19S proteasome ATPase, in the transcription of MHC I and the atypical MHC II molecules, HLA-DM and HLA-DO. Immunol Lett. 2012 Sep;147(1-2):67-74. Epub 2012 Jul 4 PubMed.
  7. . The G protein-coupled receptor GPRC5B contributes to neurogenesis in the developing mouse neocortex. Development. 2013 Nov;140(21):4335-46. Epub 2013 Oct 2 PubMed.
  8. . GPRC5B activates obesity-associated inflammatory signaling in adipocytes. Sci Signal. 2012 Nov 20;5(251):ra85. PubMed.
  9. . Case-control association mapping by proxy using family history of disease. Nat Genet. 2017 Mar;49(3):325-331. Epub 2017 Jan 16 PubMed.

Further Reading

Primary Papers

  1. . Meta-analysis of genetic association with diagnosed Alzheimer's disease identifies novel risk loci and implicates Abeta, Tau, immunity and lipid processing. bioRxiv 294629; April 5, 2018
  2. . Genetic meta-analysis identifies 9 novel loci and functional pathways for Alzheimers disease risk. bioRxiv 258533; February 22, 2018
  3. . GWAS on family history of Alzheimer's disease. bioRxiv 246223; January 15, 2018