Download Continuous Inflation Analysis for Genetic Overlap in Imaging Genetics and more Study notes Genetics in PDF only on Docsity!
Continuous inflation analysis: a threshold-free method to
estimate genetic overlap and boost power in imaging
genetics
Derrek P. Hibar^1 , Neda Jahanshad^1 , Sarah E. Medland^2 , Paul M. Thompson^1 (^1) Imaging Genetics Center, Keck School of Medicine of USC, Marina del Rey, CA (^2) QIMR Berghofer Medical Research Institute, Brisbane, Australia Abstract. Methods to quantify genetic overlap may elucidate relationships be- tween disparate traits, and provide Bayesian priors to guide the search for ge- netic influences on brain measures. Here we describe a threshold-free method called continuous inflation analysis (CIA), which we used to compare genome- wide association statistics (GWAS) for the volumes of eight brain regions, computed from brain MRI. Our goal was to understand the extent of pleiotropy (overlap in genetic influences) and concordance for the volumes of brain re- gions with different biological functions. We found significant pleiotropy among seven of the subcortical brain volumes. We found positive concordance across the seven subcortical structures and negative concordance between ge- netic influences on each subcortical structure and intracranial volume (ICV). Using a conditional FDR approach, we showed that a given brain volume GWAS could act as a Bayesian prior and improve the power to detect novel as- sociations in a related brain volume. When conditioning the putamen volume GWAS on the caudate volume GWAS, we identified 17 novel loci associated with putamen volume.
1 Introduction
Recent imaging genetics work in the Enhancing NeuroImaging Genetics through Meta-Analysis (ENIGMA) Consortium has focused on discovering common genetic variants associated with the volumes of seven subcortical brain structures (nucleus accumbens, amygdala, caudate, hippocampus, globus pallidus, putamen, thalamus) and one measure of global head size (intracranial volume; ICV) [1]. Hibar et al. ex- amined individual SNP associations with each of the eight brain volumes – each con- sidered as a single trait – but did not examine the overlapping genetic influence of the full set of common variants across structures. By examining the pleiotropy (common genetic influences) and concordance^1 across subcortical structures we should be able to (1) define Bayesian approaches to guide the search for genetic influences on the (^1) Concordance is an extension to the concept of pleiotropy that includes the direction of a pleiotropic SNP effect (i.e. a positive or negative correlation).
brain, and (2) better understand the underlying genetic pathways that may partially explain volumetric variations across different brain regions. Twin and family studies can estimate the genetic correlation between subcortical brain volume traits [2], i.e., the fraction of the observed correlation that is due to ge- netic factors. However, there may be little or no publicly available twin or family data for a given pair of traits. If genome-wide SNP data is available for a cohort, we can use GWAS summary statistics (i.e., regression coefficients relating each SNP to the traits of interest) to estimate the common genetic overlap. This is perhaps surprising, because almost all SNPs have no detectable effects, and even significantly associated SNPs generally have weak effects. So, without vast samples of data, it can be chal- lenging to pick up genetic overlap from SNP association data. A recent method called LDSC regression [3] uses GWAS summary statistics from two traits to estimate a genetic correlation driven by common genetic determinants. One limitation of this method (and genetic correlations calculated from twin and family studies as well) is that it is not possible to identify which specific variants overlap and contribute to the correlation. A related method, SNP Effect Concordance Analysis (SECA)[4], looks at pleiotropy and concordance and predefined, arbitrary thresholds. Even so, it is of great interest to try to narrow the search for genetic variants associated with brain measures, to avoid heavy multiple comparisons corrections and the vast sample sizes they currently imply (often requiring tens of thousands of subjects, e.g. in the ENIGMA studies). Here we describe a novel method to quantify the global enrichment (pleiotropy) and concordance between GWAS summary statistics from two traits. We apply this meth- od to examine the genetic overlap between brain structures examined in the ENIGMA Consortium. Our hypothesis is that brain regions will show genetic overlap with struc- tures similar to their functional groupings: limbic system (hippocampus, amygdala, thalamus) and basal ganglia (putamen, caudate, nucleus accumbens, and globus palli- dus). Further, we examine whether a conditional FDR framework can be used to boost power to detect novel associations.
2 Methods
2.1 Estimating the genetic overlap between two traits We developed a data-driven, threshold-free method, called continuous inflation analy- sis (CIA), to assess global enrichment (pleiotropy) and concordance based on GWAS summary statistics from any two pairwise traits. Here we were interested in assessing the genetic overlap across the volumes of eight different brain regions: the nucleus accumbens, amygdala, caudate, hippocampus, globus pallidus, putamen, thalamus, and intracranial volume (ICV). We performed all pairwise combinations of overlap tests between the eight traits. Before comparing two traits, we designated one dataset the reference dataset and the other the test dataset. This designation is important be- cause the CIA procedure is not symmetric. To begin, we performed a clumping pro-
data from 3,314 participants (twins and their family members) using genotypes im- puted to the 1000 Genomes phase 1, version 3 reference panel [ 8 ]. 2.3 Boosting power to detect novel gene variants using conditional FDR For pairwise comparisons that show significant overlap, we can boost the power to detect individual SNPs associated with a given test trait by conditioning on the refer- ence GWAS dataset. From the CIA model for a given pairwise comparison, we can choose the step-based cutoff that results in the most significant enrichment over all possible cutoffs. Next, we can apply the BH-FDR to the SNP P - values from the sub- setted test dataset with q = 0.05. For comparison, we applied the BH-FDR to the full set of SNP P - values from the test dataset with q = 0.05. SNPs that pass BH-FDR in the subsetted dataset but not in the full dataset are considered to be detected with in- creased power when conditioning on the reference dataset.
3 Results
3.1 Pleiotropic gene variants influence multiple brain regions We found significant evidence for pleiotropy between all pairwise comparisons of seven subcortical brain volumes (see Fig. 1 ). None of the pairwise comparisons with ICV showed significant overlap. The most significant comparison showing the high- est evidence of pleiotropy occurred between the putamen and caudate ( q = 0.0058). This relationship makes intuitive sense given the strong functional and histological evidence linking the two basal ganglia brain structures together. Fig. 1. Global evidence of pleiotropy for pairwise comparisons of eight brain traits. Compari- sons were made using CIA and were considered significant at a BH-FDR threshold q = 0.00625). The seven subcortical brain structures were tightly linked in terms of pleiotropy, but no structures showed evidence of pleiotropy with ICV. The most significant comparison (Pu- tamen | Caudate) is marked with a white star.
3.2 Evidence of a positive concordance between subcortical brain structures We found significant positive concordance in each of the pairwise comparisons of subcortical brain traits (see Fig. 2 ). In other words, genetic variants associated with an increase in a given brain volume also tend to be associated with an increase in the volume of another subcortical trait (and vice versa ). Here there is no detectable evi- dence of positive concordance between the subcortical brain structures and ICV. Fig. 2. Global evidence of positive concordance for pairwise comparisons of eight brain traits. Comparisons were made using CIA and were considered significant at a BH-FDR threshold q = 0.00625). The seven subcortical brain structures were tightly linked in terms of positive con- cordance, whereas none of the structures showed evidence of positive concordance with ICV.
3.3 Finger whorl pattern as a negative control for enrichment tests in brain We found no evidence of pleiotropy between putamen volume and the dermatoglyph- ic negative control (presence of whorl on the left thumb) at an FDR q - value = 0.05. 3.4 Conditioning enrichment tests on another brain prior can boost power to detect effects in the original trait Several of the pairwise comparisons of pleiotropy were significant, so, for purposes of illustration of the method, here we give the conditional FDR results for the “most significant” comparison (putamen volume GWAS conditioned on caudate volume GWAS). We identified 17 additional significant variants influencing putamen volume that were previously undetected without conditioning on the caudate volume GWAS (see Table 1 ).
Table 1. Conditional False Discovery Rate (FDR) analysis of putamen GWAS conditioned on caudate GWAS. Shown here are variants that pass FDR at q = 0.05 in the putamen volume GWAS when prioritizing SNPs based on their significance in the caudate GWAS, but do not pass FDR when considering the full set of putamen GWAS variants.
SNP
Raw P-‐value in
Putamen GWAS Subset FDR
FDR of Full
Sample
rs4888010 4.92E-‐ 07 0.025 0.07 1
rs11150623 1.25E-‐ 06 0.027 0.
rs17388257 1.41E-‐ 06 0.027 0.
rs62394265 1.42E-‐ 06 0.027 0.
rs76647989 7.70E-‐ 07 0.027 0.
rs7873504 1.37E-‐ 06 0.027 0.
rs10963102 2.77E-‐ 06 0.02 9 0.0 80
rs12487861 3.16E-‐ 06 0.02 9 0.0 80
rs184917581 3.10E-‐ 06 0.02 9 0.0 80
rs6135525 3.04E-‐ 06 0.02 9 0.0 80
rs62022639 2.12E-‐ 06 0.02 9 0.0 80
rs6869844 2.26E-‐ 06 0.02 9 0.0 80
rs7325851 2.47E-‐ 06 0.02 9 0.0 80
rs80258284 2.30E-‐ 06 0.02 9 0.0 80
rs842389 1.92E-‐ 06 0.02 9 0.0 80
rs10033333 4.72E-‐ 06 0.04 1 0.11 3
rs115186168 5.00E-‐ 06 0.04 1 0.
4 Conclusions
We discovered evidence of significant pleiotropy between gene variants influencing different subcortical brain volumes, using continuous inflation analysis (CIA). This agrees with findings from twin and family heritability studies, which show that there is significant genetic correlation for volumetric measures of the subcortical structures [2]. The CIA analysis builds on the twin and family heritability estimates, because the overlap between traits is estimated from genome-wide association statistics only, and does not require a family or twin design – it can be applied to imaging genetic studies of unrelated individuals, which are more common. The most significant evidence of pleiotropy came from the putamen volume GWAS conditioned on the caudate volume GWAS. The close relationship between gene variants effecting both structures is intu-
6 References
- Hibar, Derrek P., et al. "Common genetic variants influence human subcortical brain struc- tures." Nature (2015).
- Rentería, Miguel E., et al. "Genetic architecture of subcortical brain regions: common and region-specific genetic contributions." Genes, Brain and Behavior 13.8 (2014): 821-830.
- Bulik-Sullivan, Brendan K., et al. "LD Score regression distinguishes confounding from polygenicity in genome-wide association studies." Nature genetics 47.3 (2015): 291-295.
- Nyholt, Dale R. "SECA: SNP effect concordance analysis using genome-wide association summary results." Bioinformatics (2014): btu171.
- Purcell, Shaun, et al. "PLINK: a tool set for whole-genome association and population- based linkage analyses." The American Journal of Human Genetics 81.3 (2007): 559-575.
- Benjamini, Yoav, and Yosef Hochberg. "Controlling the false discovery rate: a practical and powerful approach to multiple testing." Journal of the Royal Statistical Society. Series B (Methodological) (1995): 289-300.
- Ho, Y.Y.W. et al. Variants within ADAMTS9-AS2 influence whorls in fingerprint pat- terns. (Submitted).
- 1000 Genomes Project Consortium. "An integrated map of genetic variation from 1, human genomes." Nature 491.7422 (2012): 56-65.
- Yelnik, Jérôme, et al. "A three-dimensional, histological and deformable atlas of the hu- man basal ganglia. I. Atlas construction based on immunohistochemical and MRI data." Neuroimage 34.2 (2007): 618-638.
- Brun, Caroline C., et al. "Sex differences in brain structure in auditory and cingulate re- gions." Neuroreport 20.10 (2009): 930.
- http://www.broadinstitute.org/mammals/haploreg/haploreg.php
