LeBorgne J, Gomez L, Heikkinen S, Amin N, Ahmad S, Choi SH, Bis J, Grenier-Boley B, Rodriguez OG, Kleineidam L, Young J, Tripathi KP, Wang L, Varma A, vanderLee S, Damotte V, deRojas I, Palmal S, EADB, GR@ACE, DEGESCO, EADI, GERAD, DemGene, FinnGen, ADGC, CHARGE, Giedraitis V, Ghidoni R, Fernandez V, Kehoe PG, Frikke-Schmidt R, Tsolaki M, Sanchez-Juan P, Sleegers K, Ingelsson M, Haines J, Farrer L, Mayeux R, Wang L-S, Sims R, DeStefano A, Schellenberg GD, Seshadri S, Amouyel P, Wil. X-chromosome-wide association study for Alzheimer's disease. 2024 May 03 10.1101/2024.05.02.24306739 (version 1) medRxiv.
Recommends
Please login to recommend the paper.
Comments
University of California, San Francisco
In their timely and well-executed study, Belloy et al. conducted a large-scale X-chromosome-wide association study of the genetics of Alzheimer’s disease. Their approach is novel and much-anticipated because the X chromosome has been largely excluded in genome-wide studies due to technical challenges, despite the high relevance of X-linked genes in the brain and in neurological conditions. Those technical challenges have been related to X hemizygosity in male individuals, random X inactivation and baseline X escape in female individuals, shared sequences between the X and Y, and limited representation of the X on SNP arrays. Expanding tool kits in genome-wide association studies, alongside other sequencing approaches in examining the X, are advancing a dedicated study of this sex chromosome and leading to potentially meaningful discoveries.
Analysis of over one million individuals revealed four loci with genome-wide significance, with a lead variant within SLC9A7, a transporter molecule that contributes to pH homeostasis within the Golgi apparatus. It will be particularly interesting to know how genetic variation alters cell type-specific SLC9A7 levels and function, and how that links to AD risk—an important mechanistic charge for basic and translational bench research.
It is notable that the two similar studies, while significantly smaller, also report X-chromosome wide signals. Collectively, these studies highlight the high value of the X as a contributor to neural-related functions and as a source of sex difference.
While genetic variation of the X chromosome is an important broad-stroke approach to examining this sex chromosome, X biology may contribute to risk and resilience of AD in several ways, including through gene expression and epigenetic alterations. This is particularly important because females harbor two X chromosomes, and while one is epigenetically inactivated compared to males, the “silent X” partially escapes inactivation and therefore increases the “dose” of the X in females.
In up-and-coming advances, we will understand more about how aging and Alzheimer’s modulate the inactive X, and how that influences sex-based risk and resilience. At the end of the day, the new XWASes and other X studies are pivotal because they could pave the way to new therapeutic targets that benefit men, women, or both sexes.
View all comments by Dena DubalUniversity of Lille, Inserm, Institute Pasteur Lille
Belloy et al. performed an XWAS for AD by analyzing 15,081 clinical-AD cases, 41,091 registry-AD and Alzheimer’s disease and dementia (ADD) cases and 82,386 proxy-ADD cases. They identified a genome-wide significant (P<5x10-8) signal in the SLC9A7 locus, with a low effect size (OR=1.054 (1.035-1.075)). They additionally identified five X-chromosome-wide significant (defined as P < 1x10-5) signals. This work addresses an important gap in the genetics of AD, as the X chromosome was excluded from the large-scale GWAS on AD.
In the European Alzheimer & Dementia Biobank (EADB), the International Genomics of Alzheimer’s Project (IGAP), and two biobanks, we also performed a large-scale XWAS for AD on 52,214 clinical-AD cases, 7,759 registry-AD cases and 55,868 proxy-ADD cases. Even though we considered two additional models of X-chromosome inactivation compared to the Belloy study, we did not identify any genome-wide significant signals, but did identify seven X-chromosome-wide significant loci, considering a stricter threshold of P ≤ 1.6×10−6 than did Belloy et al.
However, the loci we and Belloy identified do not overlap. Though we both found signals in the NLGN4X region, they are different: the two index variants (rs150798997 in Belloy et al., rs4364769 in our study) are located 270,925 bp away, and there is no linkage disequilibrium as determined in the EADB-core dataset. It is noteworthy that, even if we do not replicate the signal seen by Belloy at the SLC9A7 index variant (P=1.36x10-2), we did observe a signal in the locus at another variant, but with a lower absolute effect size than in (P=5.2x10-5).
The lack of overlap between the two studies could be due to several reasons, including:
a) some of the loci are false positives; a higher rate of false positives is expected among signals with X-chromosome-wide significance rather than genome-wide significance;
b) the winner’s curse: signals are expected to be slightly inflated in the first study which identified them;
c) a difference in power;
d) the phenotype definition. The proportion of clinical-AD, registry-AD, registry-ADD and proxy-ADD cases is very different between the two studies. Considering that four proxy cases effectively provide the same power as one diagnosed case, the clinical-AD, registry-AD/ADD, and proxy-ADD cases represent 20 percent, 54 percent, and 27 percent, respectively, of the effective number of cases in Belloy et al study, but 71 percent, 10 percent, and 19 percent in our study. Since a higher proportion of non-AD dementia cases is expected in the registry and proxy-ADD cases, this could lead to different genetic signals. Additionally, the proxy-ADD cases definition also differs in the two studies.
In conclusion, these XWAS did not find common genetic risk factors of large effect for AD on the non-pseudoautosomal region of the X-chromosome but identified signals which warrant further investigations, in particular to delineate their impact on AD versus ADD risk. Also, both studies were based on genotyping data, which leads to technical difficulties—for example lower coverage, in particular of the X-chromosome pseudoautosomal regions, lower call rate or lower imputation quality compared to autosomes. Future analyses of sequencing data will help to address some of those issues, and will allow to study the impact of X-chromosome rare or structural variants on AD risk.
View all comments by Céline BellenguezWashington University in St. Louis
After decades of the X chromosome remaining in the dark, it is wonderful now to have, in addition to ours, two additional Alzheimer’s disease XWAS papers see the light of day (Le Borgne et al., 2024; Simmonds et al., 2024). I believe this corroborates the timeliness of the research question to study the X chromosome in Alzheimer’s disease genetics, as well as the fact that our technologies and sample sizes have now advanced far enough to begin to tackle it.
Across these three studies, there are differences in approaches, methods, phenotypes, and sample sizes, which in sum produced a set of suggestively associated loci (variants passing P-values <1e-5 in Belloy et al., corresponding to a typical suggestive threshold; P-values < ~2e-5 in Simmonds et al., corresponding to an FDR-correction for X chromosome SNPs; and P-values < 1.7e-6 in Le Borgne et al., corresponding to a Bonferroni correction for X chromosome SNPs) and one genome-wide significant locus (the conventional threshold of P-values < 5e-8) on SLC9A7 in Belloy et al. If the suggestive P-value < 1.7e-6 threshold from Le Borgne, which marked four loci in their work, was considered across all studies, this would retain three out of four common variant loci in Belloy et al. and one common variant locus in Simmonds et al., for a total of eight independent lead variants across seven loci.
In trying to make sense of these signals, it is important to understand that power in XWAS is more challenging than in autosomal genome-wide association studies (GWAS). Power in genetic association studies relates to how much of the variance a variant can explain in a phenotype. Sidorenko et al., 2019, nicely illustrate that in men, due to hemizygosity (one X chromosome copy), there is half the power compared to the autosomes. In women, when there is random X chromosome inactivation (r-XCI), which is expected for ~70 percent of the X chromosome, the power is even further reduced, down to 1/4 compared to the autosomes. This emphasizes why one may be more lenient in paying attention to the suggestive signals on the X chromosome, as they may still be highly promising, while at the same time remaining mindful that those signals may harbor false positive associations.
Another way to assess the potential relevance of a GWAS or XWAS association signal is to look for functional support at that locus. Typically, GWAS or XWAS require post hoc analyses to address this, and one of the most common approaches currently is to run genetic co-localization analyses with “QTL data.” These represent genetic association analyses with, e.g., expression levels of a given gene in a given tissue. By assessing whether the genetic signal for Alzheimer’s disease at a locus overlaps with the genetic signal for the expression of a gene in that locus, we get an initial insight into a potential causal relationship (genetic regulation of that gene’s expression may in turn affect risk for Alzheimer’s disease). In Belloy et al. this approach was implemented and supported all four common variant loci, including the prioritization of SLC9A7 at that locus. It would be interesting to see if the studies by Le Borgne et al. and Simmonds et al. could similarly identify functional support for their loci.
Looping back to power, the studies by Belloy et al. and Le Borgne et al. are more similar in their approach and aim. The best way to compare these studies is to consider “effective sample sizes” (a more comparable measure that estimates sample size under a balanced sample design: 50/50 cases and controls). We calculated this while factoring in that datasets with proxy samples should have sample sizes divided by four to account for reduced power (Liu et al., 2016). This would indicate in Belloy et al. that the effective sample size was N=273 ,815, while in Le Borgne et al. it was N=235,757. As noted earlier, however, power on the X chromosome is further decreased relative to autosomes, notably in female samples. The study by Belloy et al. was able to leverage a large sample from the Million Veterans Program (MVP), which is primarily male-skewed when using the health registry phenotype subset (MVP-1). This likely contributed to an additional power gain in Belloy et al. relative to Le Borgne et al.
A final factor to consider is the phenotypes used. In Belloy et al., there was a larger fraction of proxy cases relative to Le Borgne et al., which may raise concern about the validity of the associations and specificity to Alzheimer’s disease rather than dementia generally. However, sensitivity analyses in Belloy et al. confirmed that all associated loci had near-identical effect sizes even when proxy datasets were excluded. This is where the smaller study by Simmonds et al. provides a nice addition, in that it may identify loci that are more specifically associated with Alzheimer’s disease pathology. It can also be leveraged to verify the Alzheimer’s disease specificity of those identified in the studies by Belloy et al. and Le Borgne et al. The top variant at the SCL9A7 locus, under the random XCI model (where the effect size corresponds to half an active allele), in Belloy et al., indicated an odds ratio (OR) = 1.027. This same variant under the random XCI model, in Simmonds et al., indicated OR = 1.046, suggesting the association may be more pronounced with Alzheimer’s disease pathology confirmed individuals.
Taken together, these three XWAS represent an exciting initial foray into the X chromosome in Alzheimer’s disease genetics and pave the work for new studies, including mapping endophenotypes and integration with multi-omics datasets to corroborate these newly associated loci and, ultimately, to identify novel drug targets.
References:
LeBorgne J, Gomez L, Heikkinen S, Amin N, Ahmad S, Choi SH, Bis J, Grenier-Boley B, Rodriguez OG, Kleineidam L, Young J, Tripathi KP, Wang L, Varma A, vanderLee S, Damotte V, deRojas I, Palmal S, EADB, GR@ACE, DEGESCO, EADI, GERAD, DemGene, FinnGen, ADGC, CHARGE, Giedraitis V, Ghidoni R, Fernandez V, Kehoe PG, Frikke-Schmidt R, Tsolaki M, Sanchez-Juan P, Sleegers K, Ingelsson M, Haines J, Farrer L, Mayeux R, Wang L-S, Sims R, DeStefano A, Schellenberg GD, Seshadri S, Amouyel P, Wil. X-chromosome-wide association study for Alzheimer's disease. 2024 May 03 10.1101/2024.05.02.24306739 (version 1) medRxiv.
Simmonds E, Leonenko G, Yaman U, Bellou E, Myers A, Morgan K, Brookes K, Hardy J, Salih D, Escott-Price V. Chromosome X-wide association study in case control studies of pathologically confirmed Alzheimer's disease in a European population. Transl Psychiatry. 2024 Sep 4;14(1):358. PubMed.
Sidorenko J, Kassam I, Kemper KE, Zeng J, Lloyd-Jones LR, Montgomery GW, Gibson G, Metspalu A, Esko T, Yang J, McRae AF, Visscher PM. The effect of X-linked dosage compensation on complex trait variation. Nat Commun. 2019 Jul 8;10(1):3009. PubMed.
Liu JZ, Erlich Y, Pickrell JK. Case-control association mapping by proxy using family history of disease. Nat Genet. 2017 Mar;49(3):325-331. Epub 2017 Jan 16 PubMed.
View all comments by Michael BelloyMake a Comment
To make a comment you must login or register.