New Results

Integrating gene expression with summary association statistics to identify susceptibility genes for 30 complex traits

Nicholas Mancuso, Huwenbo Shi, Pagé Goddard, Gleb Kichaev, Alexander Gusev, Bogdan Pasaniuc
doi: https://doi.org/10.1101/072967

Abstract

Although genome-wide association studies (GWASs) have identified thousands of risk loci for many complex traits and diseases, the causal variants and genes at these loci remain largely unknown. We leverage recently introduced methods to integrate gene expression measurements from 45 expression panels with summary GWAS data to perform 30 transcriptome-wide association studies (TWASs). We identify 1,196 susceptibility genes whose expression is associated with these traits; of these, 168 reside more than 0.5Mb away from any previously reported GWAS significant variant, thus providing new risk loci. Second, we find 43 pairs of traits with significant genetic correlation at the level of predicted expression; of these, 8 are not found through genetic correlation at the SNP level. Third, we use bi-directional regression to find evidence for BMI causally influencing triglyceride levels, and triglyceride levels causally influencing LDL. Taken together, our results provide insights into the role of expression to susceptibility of complex traits and diseases.

Introduction

Although genome-wide association studies (GWASs) have identified tens of thousands of common genetic variants associated with many complex traits1, with some notable exceptions2; 3, the causal variants and genes at these loci remain unknown. Multiple lines of evidence show that GWAS risk variants co-localize with genetic variants that regulate expression—i.e. expression quantitative trait loci (eQTL)4. This suggests that a substantial proportion of GWAS risk variants influence complex trait by regulating gene expression levels of their target genes4-7. Analyses of genotype, phenotype, and gene expression measurements from multiple tissues in the same set of individuals can directly investigate this plausible chain of causality. However, doing so is challenging due to cost and tissue availability; therefore, GWAS and eQTL data sets remain largely independent (i.e. no overlapping subjects)8; 9. Recent work demonstrated that using eQTL data to predict expression into the much larger GWAS followed by association testing can identify new susceptibility genes10-12. This approach, referred to as transcriptome-wide association study (TWAS), provides testable hypotheses under the molecular cascade of genetic variation impacting expression which in turn impacts complex trait.

In this work we connect TWAS to a test for non-zero genetic covariance between expression and trait, and extend it to estimate the genetic correlation between expression and trait. This interpretation enables us to develop new methods that characterize the relationship between complex traits using gene effects instead of single nucleotide polymorphism (SNP) effects. In particular, we estimate the genetic correlation between pairs of traits at the level of predicted expression; this is analogous to computing genome-wide genetic correlation between traits13, with correlations being determined over gene effects rather than SNP effects. Finally, we use a bi-directional regression approach14 to investigate putative causal direction for pairs of traits. This approach compares models that regress over estimated effects for identified susceptibility genes and is conceptually similar to recent work15 which uses effects of GWAS risk SNPs.

We analyze 30 GWASs spanning over 2.3 million phenotype measurements16-29 jointly with 45 expression panels sampled from more than 35 tissues and perform 30 TWASs to gain insights into the role of expression in complex trait etiology. First, we identify 1,196 genes associated with these complex traits and diseases resulting in 1,789 distinct gene-trait pairs. Of these pairs, 168 did not overlap (0.5Mb from TSS) a genome-wide significant SNP for that respective trait, which we consider to be novel risk loci. We also find 219 cases where association signal is stronger in TWAS suggesting that allelic heterogeneity plays a role in regulating expression. Consistent with previous reports11; 12, the vast majority of susceptibility genes were not proximal to the GWAS index SNP. Second, we estimate genetic correlation between these traits at the level of predicted expression and identify 43 pairs with significantly non-zero estimates; of these, 35 can be identified through genetic correlation analyses at the SNP level with 8 being identified only by analyzing predicted expression. These results suggest that a significant component of genetic correlation between complex traits can be explained by predicted expression. Lastly, we perform bi-directional analyses to provide evidence for putative causal effects between pairs of traits. Using this approach, we find evidence consistent with a causal model where body mass index (BMI) influences triglyceride levels, in line with earlier work15. We also report a novel result suggesting that triglyceride levels influence low-density lipoprotein (LDL) levels. Overall, our results shed light on shared biological mechanisms responsible for susceptibility to disease and complex trait, as well as potential downstream effects between traits.

Methods

Data Sets

We used summary association statistics from 30 large-scale (N>20,000 subjects) GWAS including various anthropometric16; 28; 29 (BMI, femoral neck bone mineral density (BMD), forearm BMD, height, lumbar spine), hematopoietic24; 27 (hemoglobin, HBA1C, mean cell hemoglobin (MCH), MCH concentration, mean cell volume, number of platelets, packed cell volume, red blood cell count), immune-related18; 20 (Crohn’s disease, inflammatory bowel disease, rheumatoid arthritis, and ulcerative colitis), metabolic17; 23 (age of menarche, fasting glucose, fasting insulin, high-density lipoprotein, HOMA-B, HOMA-IR, low-density lipoprotein, triglycerides, type 2 diabetes, and total cholesterol levels), neurological19 (schizophrenia), and social phenotypes22 (college and educational attainment; see Supplementary Table 1). We removed SNPs that were strand-ambiguous or those with minor allele frequency £ 1% (see Supplementary Table 1).

Gene expression data from RNA-Seq data were obtained from the CommonMind Consortium30 (CMC; brain; N=613), the Genotype-Tissue Expression project8 (GTEx; 41 tissues; see Supplementary Table 2 for sample size per tissue), Metabolic Syndrome in Men (METSIM; adipose; N=563)31; 32. Expression microarray data were obtained from the Netherlands Twins Registry33 (NTR; blood; N=1,247), and the Young Finns Study34; 35 (YFS; blood; N=1,264).

Performing TWAS using GWAS summary statistics

We estimated SNP heritability for observed expression levels partitioned into cis-Embedded Image (1 Mb region surrounding the gene) and trans-Embedded Image (rest of genome) components. We used the AI-REML algorithm implemented in GCTA36, which allows estimates to fall outside of the (0, 1) boundaries to maintain unbiasedness. To control for confounding, we included batch variables and the top 20 principal components estimated from genome-wide SNPs. Genes with significant cis-Embedded Image (p < 0.05 in a likelihood ratio test between the cis-only and joint model) in expression data were used for prediction. We performed a prediction-based transcriptome wide association study (TWAS) for each of the 30 GWAS using the summary approach described in ref11. In brief, we estimated the strength of association between predicted expression of gene and complex trait (zTWAS), as function of the vector of GWAS association summary Z-scores at a given cis locus zT and the LD-adjusted weights vector learned from the gene expression data wGE as Embedded Image where V is a covariance matrix across SNPs at the locus (i.e. LD). We estimated wGE using GBLUP37 from eQTL data and computed zTWAS using GWAS summary data for all 30 traits and the ∼36k gene expression measurements across all studies. We removed all loci in the human leukocyte antigen (HLA) region due to complex LD patterns. We conservatively account for multiple tests using trait-specific Bonferroni correction factors (see Supplementary Table 2).

Estimating the proportion of trait variance explained by predicted expression

We use an LD-Score38; 39 approach11 to quantify the heritability for a complex trait explained by predicted expression (denoted here as Embedded Image). The expected χ2 statistic under a polygenic trait is Embedded Image where NT is the number of individuals in the GWAS, M is the number of genes, ℓ is the LD-score, and a is the effect of population structure. We estimate ℓ for each gene by predicting expression for 503 European samples in 1000Genomes40 using the BLUP weights (see above) and then computing sample correlation. For each complex trait we perform LD-Score regression using Embedded Image (which is asymptotically equivalent to χ2) to infer Embedded Image. We estimate heritability for each expression study separately, to account for varying sample sizes and repeated gene measurements.

Estimating genetic correlation of expression and complex trait from summary data

Let expression and trait be modeled as a linear function of the genotypes in a ∼1Mb local region flanking the gene: yGE + XβGE + ∊GE and yT + XβT + ∊T where X is the standardized genotype matrix, βGE(βT) are the standardized effects, and ∊GE(∊T) is environmental noise for expression (trait). The local covariance between expression and complex trait is Embedded Image where V is the LD matrix. If no individuals are shared between studies then cov(∊GE, ∊T) = 0, (as in eQTL and GWAS studies). The local genetic correlation can be computed as Embedded Image where Embedded Image is the local SNP-heritability41 for expression (trait) estimated at the locus captured by X; however, this requires knowing the true effect sizes. Previous work41 describes a method to obtain unbiased estimates for βi using genome-wide association summary statistics (i.e. Z-scores) and reference LD. Given association statistics zT, an LD-adjusted effect size estimate is computed as Embedded Image. Hence, an estimate of the local genetic covariance42 is given by Embedded Image where bGE(bT) are the marginal (i.e. LD-unadjusted) effect sizes41; 43. It follows that Embedded Image

We standardize this estimate to obtain our final local genetic correlation estimate as Embedded Image

In practice we use the variance explained by the local index (i.e. smallest p-value) SNP as proxy for Embedded Image.

Local components of genetic correlation characterize the shared SNP effect between complex trait and expression; however, we can interpret ρg,local as the standardized effect of predicted expression on trait. Using this definition, we estimate the genetic correlation between two complex traits as the Pearson correlation across the vector of ρg,local across all genes; we term this estimate as ρGE. We test for significance assuming that Embedded Image where M is the number of genes. This procedure is unbiased in principle provided that effects of genes within single trait are not correlated. This assumption may be violated; hence, we computed trait correlation using one gene per 1Mb locus. To determine if estimates of ρGE were sensitive to changes in scale, we recomputed ρGE using non-standardized estimates of genetic covariance. We found our estimates to be highly correlated (r = 0.94; p < 2.2 × 10-16), indicating little importance in using correlation versus covariance. We report results using standardized effects for consistency across figures and tables.

Estimating putative casual relationships between pairs of traits

To glean insight into the underlying causal relationship between pairs of traits, we perform a bi-directional regression14 and estimate two different values of ρGE by varying gene sets. Before describing the approach, we first review several causal models that explain non-zero ρGE between two traits (see Figure 1). Models A and B depict causal relationships in which the effects of a gene set are mediated by one trait on the other. We can formally state model A (without loss of generality for B). Let T1 be defined as yT1 = GT1βT1 where GT1 denotes the matrix of predicted expression at the causal genes, βT1 are the effect sizes, and ∊T1 is environmental noise. We define T2 as, Embedded Image where γT1 is the causal effect of T1 on T2, GT1βT1 are the remaining causal genes and their effects for T2, and Embedded Image is the combined environment component. Under model A, the causal gene set for T1 will have a non-zero effect on T2 (i.e. γT1 ≠ 0); however, if T1 does not cause T2, this effect will be zero since unrelated genes have no downstream effect. Bi-directional regression provides a test to distinguish between models A and B by regressing estimated effect sizes for gene sets under model A (i.e. βT1 ∼ βT1γT1) and comparing to estimates under model B (i.e. βT2 ∼ βT2γT2). Since the causal gene sets for each trait are unknown, we use their identified susceptibility genes as proxy. We estimate ρGE conditional on the gene set for trait i and denote its value as ρj|i. This procedure is repeated by ascertaining the gene set for trait j to obtain ρi|j. We perform a Welch’s t-test44 to determine if estimates of ρi|j and ρj|i are significantly different, thus providing evidence consistent with a causal direction. This approach is conceptually similar to bi-directional regression analyses of estimated SNP effects on two complex traits15;45. We stress that while a bi-directional approach is capable of rejecting model A in favor of model B (or vice-versa), it cannot rule out model C, in which a shared pathway (or set of pathways) drive both traits independently.

Figure 1.
Figure 1.

Illustration of several causal models that explain expression correlation for traits T1 and T2 given their causal gene sets. Model A) trait 1 directly influences trait 2. In this case, the effect of genes Embedded Image on trait 2 is mediated by trait 1 which implies Embedded Image. Model B) trait 2 directly influences trait 1. Similarly, the effect of genes Embedded Image on trait 1 is mediated by trait 2 which implies Embedded Image. Model C) traits 1 and 2 are influenced independently through unobserved trait or traits.

Results

TWAS identifies 1,196 susceptibility genes for 30 complex traits and diseases

We integrated the 30 GWAS summary data with gene expression to identify 1,196 susceptibility genes (i.e. gene with at least one significant trait association) comprising 5,490 total associations (after Bonferroni correction; see Methods). Of these associations, we observed 1,789 distinct gene-trait pairs with 783 found in anthropometric traits, 423 in metabolic traits, 215 in immune-related traits, 213 in hematopoietic traits, 137 in neurological traits (i.e. schizophrenia), and 18 in social traits (see Table 1; see Supplementary Tables 3-4). For example, the 137 susceptibility genes found for schizophrenia included SNX19 (cerebellum; p=2.2 × 10-8) and NMRAL1 (muscle; p=9.7 × 10-7); this is consistent with a previously reported study12 that used different methods and expression data (see Supplementary Table 5). We did not find susceptibility genes for forearm bone mineral density (BMD), HOMA-B, and mean cell hemoglobin concentration, which is consistent with low GWAS signal for these traits (see Table 1). Indeed, the number of GWAS risk loci strongly correlated with the number of identified susceptibility genes (r=0.99; p < 2.2 × 10-16) which reflects the underlying polygenicity of these traits. We explored putative molecular function and pathways enriched with identified susceptibility genes using the PANTHER database46, but were underpowered to detect molecular function for most individual traits (see Supplementary Note).

View this table:
Table 1.

Summary of GWAS and TWAS results. The majority (92%) of GWAS risk loci overlap with at least one eGene, of which 40% contain at least one susceptibility gene. We report 168 (9%) of identified gene-trait pairs do not overlap a GWAS variant, which provide novel risk loci for follow up.

Next, we quantified the overlap of susceptibility genes and GWAS signals. Of the 1,789 identified gene-trait pairs, 168 (9%) were not proximal (more than 0.5Mb from TSS) to any genome-wide significant SNP for that respective trait thus yielding new risk loci. Conversely, of the 1,526 GWAS risk loci, 1,405 (92%) overlapped with at least one eGene (i.e. gene with heritable expression levels in at least one of the considered expression panels) and 551 (36%) overlapping at least one susceptibility gene (see Table 1). Focusing on the 1,621 associations that overlapped a genome-wide significant SNP, we observed 1,488 (83%) genes that were not nearest, suggesting that the traditional heuristic of prioritizing genes closest to GWAS SNPs is typically not supported by evidence from eQTL data (see Supplementary Figure 1). While GWAS SNPs provide the majority of the power in this approach, the flexibility of TWAS to leverage allelic heterogeneity provides a significant gain11. We found 219 instances across 19 traits where association signal was stronger in TWAS compared to GWAS, with an average 1.2 × increase in χ2 statistics. For example, predicted expression in CCDC88B (a gene involved in T-cell maturation and inflammation47) exhibited strong association with Crohn’s disease (pTWAS=6.32 × 10-8) whereas the index SNP (i.e. top overlapping GWAS SNP) at site rs11231774 was only suggestive (pGWAS=2.47 × 10-6). This effect was most dramatic for height, with 108 susceptibility genes having stronger signal than GWAS index SNPs. We observed a 2.6 × increase in χ2 statistics for predicted expression in CRELD1 (pTWAS=1.55 × 10-10) compared to the index SNP rs1473183 (pGWAS=6.33 × 10-5).

Recent work48 applied a similar approach12 using summary eQTL from blood and GWAS data to identify 71 genes for 28 complex traits48. Of the investigated traits, 12 overlapped our study. Surprisingly, despite using independent methods and expression data we were able to validate 40 out of 51 associations (see Supplementary Table 6). Overall, we identified 564 genes for these traits in contrast to 63 genes reported in that study. This increase in power can be attributed to two reasons. First, we integrate multiple expression panels sampled from many tissues, which assays many more genes. Second, we use a method that jointly tests the entire locus, rather than index SNPs. We have shown that many identified susceptibility genes contain signals of allelic heterogeneity; therefore, using individual SNPs will decrease power.

Genes associated to multiple traits

We investigated the degree of pleiotropic susceptibility genes (i.e. gene associated with more than one trait) in our data and found 380 (32%) identified genes associated with multiple traits (see Supplementary Figure 2). For example, the gene IKZF3 displayed strong associations in Crohn’s disease (blood; p=1.6 × 10-9), HDL levels (blood; p=6.6 × 10-15), IBD (blood; p=7.9 × 10-16), rheumatoid arthritis (blood; p=6.0 × 10-8), and ulcerative colitis (blood; p=9.2 × 10-10). Indeed, IKZF3 has been shown to influence lymphocyte development and differentiation49; 50. These traits are known to have a strong autoimmune component51; hence, association with predicted IKZF3 expression levels is consistent with a model where cis-regulated variation in IKZF3 product levels contributes to risk. Similarly, we observed three susceptibility genes shared between education years and height (see Figure 2): ABCB9 (heart; pheight=1.38 × 10-15, pey=1.28 × 10-6), BTN2A3P (adipose; pheight=3.82 × 10-12, pey=1.90 × 10-7), and MPHOSPH9 (thyroid; pheight=5.84 × 10-18, pey=1.30 × 10-6). This is consistent with a recent study13 that reported a nonzero genetic correlation between height and education years (ρg = 0.13, p=3.82 × 10-6).

Figure 2.
Figure 2.

Susceptibility genes shared for education years and height. We indicate −log10 p-values for eQTLs in green and trait-specific GWAS in black using separate axes to simplify illustration. Their respective TWAS p-values are ABCB9 (heart; pheight=1.38 × 10-15, pey=1.28 × 10-6), BTN2A3P (adipose; pheight=3.82 × 10-12, pey=1.90 × 10-7), and MPHOSPH9 (thyroid; pheight=5.84 × 10-18, pey=1.30 × 10-6).

Effect of cis expression on trait is consistent across tissues

Having established the importance of individual predicted gene expression levels for these traits, we next estimated the amount of trait variance explained by predicted expression using all examined genes, including those not significantly associated, using an LD-Score regression approach (see Methods). We found 108 tissue-trait pairs across 17 traits and 33 tissues where the cumulative effect of all measured genes on trait was significantly greater (p < 0.05 / 45) than the significant-only set. For example, in height we estimated Embedded Image (Jack-knife SE=0.02; p=5.6 × 10-4; see Supplementary Table 7) using all 3,733 measured genes in YFS and Embedded Image (Jack-knife SE=6.9 × 10-3; p=0.026) using the 169 YFS susceptibility genes (pALL>SIG=5.6 × 10-3). This suggests that there exist additional susceptibility genes for height, which we are underpowered to detect. However, for most trait-tissue pairs we did not observe a significant difference at our given sample sizes. Indeed, we measured a significant association between expression study sample size and number of eGenes (r=0.2; SE=0.05; p=6.4 × 10-8), which indicates that smaller studies lack power to find eGenes, thus underestimating the total Embedded Image.

We next asked whether any tissues are burdened with increased levels of risk for a given trait. To test this hypothesis, we examined the difference between estimated trait variance explained per gene with the average. Our results did not suggest tissue-specific enrichment at current sample sizes (see Supplementary Table 8). Given no observable difference in tissue-specific risk, we expect local estimates of genetic correlation to be highly similar across tissues. When estimating ρg,local, we observed consistent effect size estimates in both sign and magnitude estimates across tissues (mean tissue-tissue r=0.82; see Figure 3). These results are compatible with earlier work that found cis effects on expression is largely consistent across tissues52. To obtain a meta estimate of local genetic correlation for gene-trait pairs with measurements in multiple tissues, we use the mean genetic correlation across all expression panels in all following analyses.

Figure 3.
Figure 3.

Histogram and density estimate for correlation of ρg,local across tissues. We computed the correlation across pairs of different tissues using local estimates of genetic correlation between expression on trait. The majority of tissues exhibited high correlation over the underlying gene effects on trait with an estimated mean r = 0.82

Genetic correlation between traits using predicted expression

To evaluate the shared contribution of predicted expression on pairs of traits, we computed expression correlation (ρGE; see Methods) using nominally significant (pTWAS < 0.05) genes. This approach is similar to estimating genetic correlation (ρG) between two complex traits13; however, it differs in that correlation is computed through predicted components of gene expression rather than SNP effects. For 435 distinct pairs, we discovered 43 significant expression correlations, 22 of which had previously reported non-zero genetic correlations13 (see Figure 4; see Supplementary Table 9). For example, age of menarche and BMI had an estimated ρGE = −0.32 (95% CI [-0.32, -0.21]; p=7.97 × 10-8). This negative correlation is consistent with estimates published in epidemiological studies53 in addition to studies probing genetic correlation across complex traits13. Using estimates of ρGE, we clustered traits and observed groups forming naturally in the trait-trait matrix (see Figure 4). Interestingly, BMI clustered with insulin-related traits (HOMA-B, HOMA-IR, and fasting insulin). Our estimates were highly consistent with LD-Score regression results (see Figure 4; Supplementary Table 9). Out of 435 pairs of traits, 35 demonstrated significance for ρGE and ρg, whereas 8 and 27 were exclusive for ρGE and ρg, respectively. Given the high degree of concordance between estimates of ρGE and ρg, we tested if any were significantly different and found four insulin-related pairs of traits and three blood-related pairs with more extreme values for ρGE (see Supplementary Table 9). Differences for these pairs of traits can be partially explained by overconfident standard errors in ρGE (see Supplementary Table 10). Overall, we found ρGE to explain the majority of variation in ρg (r2 = 0.72).

Figure 4.
Figure 4.

Estimates of genetic correlation ρg obtained from LD-Score vs estimates of expression ρGE using nominally significant TWAS results. A) Correlation matrix for 30 traits. The lower triangle contains ρGE and the upper triangle contains ρg estimates. Estimates of correlation that are significantly non-zero (p < 0.05 / 435) are marked with a star (*). Strength and direction of correlation is indicated by size and color. We found 43 significantly correlated traits using cis expression and 62 using genome-wide SNPs. B) Linear relationship between estimates of ρGE and ρg. We indicate whether individual estimates were significant in either approach by color. Non-significant trait pairs are reduced in size for visibility.

Bi-directional regression suggests putative causal relationships

Given pairs of traits with significant estimates of ρGE, we aimed to distinguish among possible causal explanations by performing bi-directional regression analyses (see Methods). To empirically validate our approach, we regressed HDL, LDL, and triglycerides with total cholesterol. Total cholesterol (TC) is the direct consequence of summing over triglyceride, HDL, and LDL levels, thus we expect to observe increased signal for ρTC|Lipid compared to ρLipid|TC. Of these three, we found evidence for triglycerides influencing total cholesterol (p=2.34 × 10-3). We observed consistent, but not significant, evidence for the effect of LDL on TC (p=6.79 × 10-2) and HDL on TC (p=5.56× 10-1; see Figure 5). These results suggest that point-estimates from the bi-directional approach favor the correct model, but may not have adequate power required for significance.

Figure 5.
Figure 5.

Estimates of expression correlation ρGE for HDL, LDL, and TG with total cholesterol. Column A) Estimates of ρGE using nominally significant genes (p < 0.05). Column B) We repeated the analysis using only susceptibility genes found in the x-axis trait but not found in the y-axis trait. Column C) Same analysis as Column B, but using the other trait’s susceptibility genes. All three analyses resulted in stronger point estimates for ρTC|Lipid when conditioning on HDL/LDL/TG genes compared to ρLipid|TC; however, significance was only observed for ρTC|TG (p=2.34 × 10-3).

We tested the 43 pairs of traits identified above (see Table 3) while ascertaining on susceptibility genes and observed asymmetric effects at p < 0.05 for BMI-triglycerides and LDL-triglycerides (see Figure 6). For example, in the bi-directional analysis on BMI and triglycerides, we observed a significant effect for ρTC|BMI = 0.62 (95% CI [0.27, 0.83]; p=2.06 × 10-3). By contrast, the reverse analysis estimate overlapped with zero at ρBMI|TG = −0.04 (95% CI [-0.49, 0.42]; p=0.86). Individual estimates for ρTG|BMI and ρBMI|TG were significantly different (p=0.01; Welch’s t-test), which is consistent with a model where BMI directly influences triglyceride levels. In practice, we used susceptibility genes found through TWAS (p ∼ 1 × 10-6), but this may be too strict an inclusion threshold for genes which we lack power to detect. We report analyses using weaker thresholds and observe similar results (see Supplementary Tables 11, 12). Our result reinforces previous estimates of putative causal effect where BMI influences triglyceride levels15; 54.

Figure 6.
Figure 6.

Estimates of expression correlation ρGE for triglycerides with BMI and triglycerides with LDL. We present results for pairs of traits that displayed a significant difference (p < 0.05; Welch’s t-test) in their conditional estimates. These results are consistent with a causal model where BMI influences TG and TG influences LDL.

View this table:
Table 2.

Novel risk loci. Identified susceptibility genes that do not overlap a genome-wide significant SNP (p < 5 × 10-8) within 0.5Mb for the tested trait.

View this table:
Table 3.

Significant estimates of ρGF for 43 pairs of traits. We performed bi-directional regression and obtain conditional estimates of ρGF, which provides evidence for a putative causal direction. We observed three pairs of traits with a significant difference between their directional estimates; namely, BMI influencing TG, TG influencing LDL, and TG influencing TC. We mark entries with M < 3 with “-“. a determined by Welch–Satterthwaite equation

Discussion

In this work we used GWAS summary statistics from 30 complex traits and diseases jointly with expression data sampled across 45 expression panels to identify susceptibility genes for complex traits. We identified 1,196 susceptibility genes for 27 of the 30 complex traits. We use estimates of local genetic correlation between gene expression and trait to compute ρGE, which quantifies the shared effect of predicted expression levels between two complex traits. Using this definition, we found 43 pairs of traits to be significantly correlated, of which 8 were novel. To provide evidence of possible causal direction, we adapted a recently proposed causality test15 to operate at the gene level. Our results support triglycerides (TG) influencing LDL, and BMI influencing triglycerides. As more GWAS and eQTL summary results become publicly available, we expect additional studies to integrate cross-trait information to make inferences about mechanistic bases for complex trait.

Assuming gene expression mediates the effect of genetics on complex trait, testing for association between the predicted component of expression and trait is equivalent with a two-sample Mendelian randomization test for a causal effect of expression on trait55; 56. This test for causality is valid provided SNPs do not exhibit pleiotropic effects; therefore, the TWAS associations are not proof of causal relationships between expression and complex trait. This set of assumptions extends to our bi-directional approach to infer causal direction. A bi-directional regression is capable of distinguishing between direction of effect, but cannot rule out pleiotropy.

We conclude with several caveats. First, we note that using estimates of genetic correlation to find susceptibility genes may still be biased due to confounding. The expression weights used for TWAS may tag variants that are causal through other genes or non-genic mechanisms. In principle, this can be partially remedied by jointly testing multiple genes with trait; however, a correctly specified model would require covariance estimates between observed, not predicted, expression levels—which is not available in summary data. In this work we combined estimates across tissues by taking the mean effect to compute the genetic correlation between trait and expression. This approach is unbiased, but may be inefficient. Recent work57 describes a random-effects model to combine estimates across tissues to increase power. Finally, our method to estimate correlation between traits using the genetically predicted component of gene expression makes several simplifying assumptions. We remedied the non-independence of genes by sampling single genes within a 1Mb region, an approach which has been used previously45. However, a more powerful approach may take correlations across genes into account.

Acknowledgements

We would like to thank Valerie Arboleda, Robert Brown, Kathy Burch, and Malika Kumar for helpful discussions and feedback. We also thank Dr. Nicole Soranzo for sharing summary data for the platelet traits.

CMC: Data were generated as part of the CommonMind Consortium supported by funding from Takeda Pharmaceuticals Company Limited, F. Hoffman-La Roche Ltd and NIH grants R01MH085542, R01MH093725, P50MH066392, P50MH080405, R01MH097276, RO1-MH-075916, P50M096891, P50MH084053S1, R37MH057881 and R37MH057881S1, HHSN271201300031C, AG02219, AG05138 and MH06692. Brain tissue for the study was obtained from the following brain bank collections: the Mount Sinai NIH Brain and Tissue Repository, the University of Pennsylvania Alzheimer’s Disease Core Center, the University of Pittsburgh NeuroBioBank and Brain and Tissue Repositories and the NIMH Human Brain Collection Core. CMC Leadership: Pamela Sklar, Joseph Buxbaum (Icahn School of Medicine at Mount Sinai), Bernie Devlin, David Lewis (University of Pittsburgh), Raquel Gur, Chang-Gyu Hahn (University of Pennsylvania), Keisuke Hirai, Hiroyoshi Toyoshiba (Takeda Pharmaceuticals Company Limited), Enrico Domenici, Laurent Essioux (F. Hoffman-La Roche Ltd), Lara Mangravite, Mette Peters (Sage Bionetworks), Thomas Lehner, Barbara Lipska (NIMH)

References

  1. ↵
    Welter, D., MacArthur, J., Morales, J., Burdett, T., Hall, P., Junkins, H., Klemm, A., Flicek, P., Manolio, T., Hindorff, L., et al. (2014). The NHGRI GWAS Catalog, a curated resource of SNP-trait associations. Nucleic Acids Research 42, D1001–D1006.
    OpenUrlCrossRefPubMedWeb of Science
  2. ↵
    Claussnitzer, M., Dankel, S.N., Kim, K.-H., Quon, G., Meuleman, W., Haugen, C., Glunk, V., Sousa, I.S., Beaudry, J.L., Puviindran, V., et al. (2015). FTO Obesity Variant Circuitry and Adipocyte Browning in Humans. New England Journal of Medicine 373, 895–907.
    OpenUrlCrossRefPubMed
  3. ↵
    Consortium, T.I.M.S.G.C.T.W.T.C.C. (2011). Genetic risk and a primary role for cell-mediated immune mechanisms in multiple sclerosis. Nature 476, 214–219.
    OpenUrlCrossRefPubMedWeb of Science
  4. ↵
    Nicolae, D.L., Gamazon, E., Zhang, W., Duan, S., Dolan, M.E., and Cox, N.J. (2010). Trait-Associated SNPs Are More Likely to Be eQTLs: Annotation to Enhance Discovery from GWAS. PLoS Genet 6, e1000888.
    OpenUrlCrossRefPubMed
  5. Emilsson, V., Thorleifsson, G., Zhang, B., Leonardson, A.S., Zink, F., Zhu, J., Carlson, S., Helgason, A., Walters, G.B., Gunnarsdottir, S., et al. (2008). Genetics of gene expression and its effect on disease. Nature 452, 423–428.
    OpenUrlCrossRefPubMedWeb of Science
  6. Nica, A.C., Montgomery, S.B., Dimas, A.S., Stranger, B.E., Beazley, C., Barroso, I., and Dermitzakis, E.T. (2010). Candidate Causal Regulatory Effects by Integration of Expression QTLs with Complex Trait Genetic Associations. PLoS Genet 6, e1000895.
    OpenUrlCrossRefPubMed
  7. ↵
    Albert, F.W., and Kruglyak, L. (2015). The role of regulatory variation in complex traits and disease. Nat Rev Genet 16, 197–212.
    OpenUrlCrossRefPubMed
  8. ↵
    Lonsdale, J., Thomas, J., Salvatore, M., Phillips, R., Lo, E., Shad, S., Hasz, R., Walters, G., Garcia, F., Young, N., et al. (2013). The Genotype-Tissue Expression (GTEx) project. Nat Genet 45, 580–585.
    OpenUrlCrossRefPubMed
  9. ↵
    Lappalainen, T., Sammeth, M., Friedlander, M.R., t Hoen, P.A.C., Monlong, J., Rivas, M.A., Gonzalez-Porta, M., Kurbatova, N., Griebel, T., Ferreira, P.G., et al. (2013). Transcriptome and genome sequencing uncovers functional variation in humans. Nature 501, 506–511.
    OpenUrlCrossRefPubMedWeb of Science
  10. ↵
    Gamazon, E.R., Wheeler, H.E., Shah, K.P., Mozaffari, S.V., Aquino-Michaels, K., Carroll, R.J., Eyler, A.E., Denny, J.C., Consortium, G.T., Nicolae, D.L., et al. (2015). A gene-based association method for mapping traits using reference transcriptome data. Nat Genet 47, 1091–1098.
    OpenUrlCrossRefPubMed
  11. ↵
    Gusev A, K.A., Shi H, Bhatia G, Chung W, Penninx B, Jansen R, de Geus E, Boomsma DI, Wright FA, Sullivan PF, Nikkola E, Alvarez M, Civelek M, Lusis AJ, Lehtimäki T, Raitoharju E, Kähönen M, Seppälä I, Raitakari OT, Kuusisto J, Laakso M, Price AL, Pajukanta P, Pasaniuc B. (2016). Integrative approaches for large-scale transcriptome-wide association studies. Nature Genetics.
  12. ↵
    Zhu, Z., Zhang, F., Hu, H., Bakshi, A., Robinson, M.R., Powell, J.E., Montgomery, G.W., Goddard, M.E., Wray, N.R., Visscher, P.M., et al. (2016). Integration of summary data from GWAS and eQTL studies predicts complex trait gene targets. Nat Genet advance online publication.
  13. ↵
    Bulik-Sullivan, B., Finucane, H.K., Anttila, V., Gusev, A., Day, F.R., Loh, P.-R., ReproGen, C., Psychiatric Genomics, C., Genetic Consortium for Anorexia Nervosa of the Wellcome Trust Case Control, C., Duncan, L., et al. (2015). An atlas of genetic correlations across human diseases and traits. Nat Genet 47, 1236–1241.
    OpenUrlCrossRefPubMed
  14. ↵
    Davey Smith, G., and Hemani, G. (2014). Mendelian randomization: genetic anchors for causal inference in epidemiological studies. Human Molecular Genetics 23, R89–R98.
    OpenUrlCrossRefPubMedWeb of Science
  15. ↵
    Pickrell, J.K., Berisa, T., Liu, J.Z., Segurel, L., Tung, J.Y., and Hinds, D.A. (2016). Detection and interpretation of shared genetic influences on 42 human traits. Nat Genet advance online publication.
  16. ↵
    Zheng, H., Forgetta, V., Hsu, Y., Estrada, K., RoselloDiez, A., Leo, P.J., Dahia, C.L., ParkMin, K.H., Tobias, J.H., Kooperberg, C., et al. (2015). Whole-genome sequencing identifies EN1 as a determinant of bone density and fracture. Nature 526, 112–117.
    OpenUrlCrossRefPubMed
  17. ↵
    Morris AP, V.B., Teslovich TM, Ferreira T, Segre AV, Steinthorsdottir V, Strawbridge RJ, Khan H, Grallert H, Mahajan A. (2012). Large-scale association analysis provides insights into the genetic architecture and pathophysiology of type 2 diabetes. Nat Genet 44, 981–990.
    OpenUrlCrossRefPubMed
  18. ↵
    Liu, J.Z., van Sommeren, S., Huang, H., Ng, S.C., Alberts, R., Takahashi, A., Ripke, S., Lee, J.C., Jostins, L., Shah, T., et al. (2015). Association analyses identify 38 susceptibility loci for inflammatory bowel disease and highlight shared genetic risk across populations. Nat Genet 47, 979–986.
    OpenUrlCrossRefPubMed
  19. ↵
    Schizophrenia Working Group of the Psychiatric Genomics, C. (2014). Biological insights from 108 schizophrenia-associated genetic loci. Nature 511, 421–427.
    OpenUrlCrossRefPubMedWeb of Science
  20. ↵
    Okada, Y., Wu, D., Trynka, G., Raj, T., Terao, C., Ikari, K., Kochi, Y., Ohmura, K., Suzuki, A., Yoshida, S., et al. (2014). Genetics of rheumatoid arthritis contributes to biology and drug discovery. Nature 506, 376–381.
    OpenUrlCrossRefPubMedWeb of Science
  21. Perry, J.R.B., Day, F., Elks, C.E., Sulem, P., Thompson, D.J., Ferreira, T., He, C., Chasman, D.I., Esko, T., Thorleifsson, G., et al. (2014). Parent-of-origin-specific allelic associations among 106 genomic loci for age at menarche. Nature 514, 92–97.
    OpenUrlCrossRefPubMedWeb of Science
  22. ↵
    Rietveld, C.A., Medland, S.E., Derringer, J., Yang, J., Esko, T., Martin, N.W., Westra, H.-J., Shakhbazov, K., Abdellaoui, A., Agrawal, A., et al. (2013). GWAS of 126,559 Individuals Identifies Genetic Variants Associated with Educational Attainment. Science 340, 1467–1471.
    OpenUrlAbstract/FREE Full Text
  23. ↵
    Global Lipids Genetics, C. (2013). Discovery and refinement of loci associated with lipid levels. Nat Genet 45, 1274–1283.
    OpenUrlCrossRefPubMed
  24. ↵
    Soranzo, N., Sanna, S., Wheeler, E., Gieger, C., Radke, D., Dupuis, J., Bouatia-Naji, N., Langenberg, C., Prokopenko, I., Stolerman, E., et al. (2010). Common Variants at 10 Genomic Loci Influence Hemoglobin A(1C) Levels via Glycemic and Nonglycemic Pathways. Diabetes 59, 3229–3239.
    OpenUrlAbstract/FREE Full Text
  25. Dupuis, J., Langenberg, C., Prokopenko, I., Saxena, R., Soranzo, N., Jackson, A.U., Wheeler, E., Glazer, N.L., Bouatia-Naji, N., Gloyn, A.L., et al. (2010). New genetic loci implicated in fasting glucose homeostasis and their impact on type 2 diabetes risk. Nat Genet 42, 105–116.
    OpenUrlCrossRefPubMedWeb of Science
  26. Gieger, C., Radhakrishnan, A., Cvejic, A., Tang, W., Porcu, E., Pistis, G., Serbanovic-Canic, J., Elling, U., Goodall, A.H., Labrune, Y., et al. (2011). New gene functions in megakaryopoiesis and platelet formation. Nature 480, 201–208.
    OpenUrlCrossRefPubMedWeb of Science
  27. ↵
    van der Harst, P., Zhang, W., Mateo Leach, I., Rendon, A., Verweij, N., Sehmi, J., Paul, D.S., Elling, U., Allayee, H., Li, X., et al. (2012). Seventy-five genetic loci influencing the human red blood cell. Nature 492, 369–375.
    OpenUrlCrossRefPubMedWeb of Science
  28. ↵
    Locke, A.E., Kahali, B., Berndt, S.I., Justice, A.E., Pers, T.H., Day, F.R., Powell, C., Vedantam, S., Buchkovich, M.L., Yang, J., et al. (2015). Genetic studies of body mass index yield new insights for obesity biology. Nature 518, 197–206.
    OpenUrlCrossRefPubMed
  29. ↵
    Wood, A.R., Esko, T., Yang, J., Vedantam, S., Pers, T.H., Gustafsson, S., Chu, A.Y., Estrada, K., Luan, J.a., Kutalik, Z., et al. (2014). Defining the role of common variation in the genomic and biological architecture of adult human height. Nat Genet 46, 1173–1186.
    OpenUrlCrossRefPubMed
  30. ↵
    Fromer, M., Roussos, P., Sieberts, S.K., Johnson, J.S., Kavanagh, D.H., Perumal, T.M., Ruderfer, D.M., Oh, E.C., Topol, A., Shah, H.R., et al. (2016). Gene Expression Elucidates Functional Impact of Polygenic Risk for Schizophrenia. bioRxiv.
  31. ↵
    Stančáková, A., Civelek, M., Saleem, N.K., Soininen, P., Kangas, A.J., Cederberg, H., Paananen, J., Pihlajamäki, J., Bonnycastle, L.L., Morken, M.A., et al. (2012). Hyperglycemia and a Common Variant of GCKR Are Associated With the Levels of Eight Amino Acids in 9,369 Finnish Men. Diabetes 61, 1895–1902.
    OpenUrlAbstract/FREE Full Text
  32. ↵
    Stančáková, A., Javorský, M., Kuulasmaa, T., Haffner, S.M., Kuusisto, J., and Laakso, M. (2009). Changes in Insulin Sensitivity and Insulin Release in Relation to Glycemia and Glucose Tolerance in 6,414 Finnish Men. Diabetes 58, 1212–1221.
    OpenUrlAbstract/FREE Full Text
  33. ↵
    Wright, F.A., Sullivan, P.F., Brooks, A.I., Zou, F., Sun, W., Xia, K., Madar, V., Jansen, R., Chung, W., Zhou, Y.-H., et al. (2014). Heritability and genomics of gene expression in peripheral blood. Nat Genet 46, 430–437.
    OpenUrlCrossRefPubMed
  34. ↵
    Nuotio, J., Oikonen, M., Magnussen, C.G., Jokinen, E., Laitinen, T., Hutri-Kähönen, N., Kähönen, M., Lehtimäki, T., Taittonen, L., Tossavainen, P., et al. (2014). Cardiovascular risk factors in 2011 and secular trends since 2007: The Cardiovascular Risk in Young Finns Study. Scandinavian Journal of Public Health 42, 563–571.
    OpenUrlCrossRefPubMed
  35. ↵
    Raitakari, O.T., Juonala, M., Rönnemaa, T., Keltikangas-Järvinen, L., Räsänen, L., Pietikäinen, M., Hutri-Kähönen, N., Taittonen, L., Jokinen, E., Marniemi, J., et al. (2008). Cohort Profile: The Cardiovascular Risk in Young Finns Study. International Journal of Epidemiology 37, 1220–1226.
    OpenUrlCrossRefPubMedWeb of Science
  36. ↵
    Yang, J., Lee, S.H., Goddard, M.E., and Visscher, P.M. GCTA: A Tool for Genome-wide Complex Trait Analysis. The American Journal of Human Genetics 88, 76–82.
  37. ↵
    de los Campos, G., Vazquez, A.I., Fernando, R., Klimentidis, Y.C., and Sorensen, D. (2013). Prediction of Complex Human Traits Using the Genomic Best Linear Unbiased Predictor. PLoS Genet 9, e1003608.
    OpenUrlCrossRefPubMed
  38. ↵
    Bulik-Sullivan, B.K., Loh, P.-R., Finucane, H.K., Ripke, S., Yang, J., Schizophrenia Working Group of the Psychiatric Genomics, C., Patterson, N., Daly, M.J., Price, A.L., and Neale, B.M. (2015). LD Score regression distinguishes confounding from polygenicity in genome-wide association studies. Nat Genet 47, 291–295.
    OpenUrlCrossRefPubMed
  39. ↵
    Finucane, H.K., Bulik-Sullivan, B., Gusev, A., Trynka, G., Reshef, Y., Loh, P.-R., Anttila, V., Xu, H., Zang, C., Farh, K., et al. (2015). Partitioning heritability by functional annotation using genome-wide association summary statistics. Nat Genet 47, 1228–1235.
    OpenUrlCrossRefPubMed
  40. ↵
    The Genomes Project, C. (2015). A global reference for human genetic variation. Nature 526, 68–74.
    OpenUrlCrossRefPubMed
  41. ↵
    Shi, H., Kichaev, G., and Pasaniuc, B. (2016). Contrasting the Genetic Architecture of 30 Complex Traits from Summary Association Data. The American Journal of Human Genetics.
  42. ↵
    Shi, H.M., Nicholas; Pasaniuc, Bogdan;. (2016). Identifying genetic overlap among 30 complex traits from GWAS summary data. (In preperation).
  43. ↵
    Yang, J., Ferreira, T., Morris, A.P., Medland, S.E., Madden, P.A.F., Heath, A.C., Martin, N.G., Montgomery, G.W., Weedon, M.N., Loos, R.J., et al. (2012). Conditional and joint multiple-SNP analysis of GWAS summary statistics identifies additional variants influencing complex traits. Nat Genet 44, 369–375.
    OpenUrlCrossRefPubMed
  44. ↵
    Welch, B.L. (1947). The Generalization of ‘Student’s’ Problem when Several Different Population Variances are Involved. Biometrika 34, 28–35.
    OpenUrlCrossRefPubMedWeb of Science
  45. ↵
    Do, R., Willer, C.J., Schmidt, E.M., Sengupta, S., Gao, C., Peloso, G.M., Gustafsson, S., Kanoni, S., Ganna, A., Chen, J., et al. (2013). Common variants associated with plasma triglycerides and risk for coronary artery disease. Nat Genet 45, 1345–1352.
    OpenUrlCrossRefPubMed
  46. ↵
    Mi, H., Muruganujan, A., Casagrande, J.T., and Thomas, P.D. (2013). Large-scale gene function analysis with the PANTHER classification system. Nat Protocols 8, 1551–1566.
    OpenUrl
  47. ↵
    Kennedy, J.M., Fodil, N., Torre, S., Bongfen, S.E., Olivier, J.-F., Leung, V., Langlais, D., Meunier, C., Berghout, J., Langat, P., et al. (2014). CCDC88B is a novel regulator of maturation and effector functions of T cells during pathological inflammation. The Journal of Experimental Medicine 211, 2519–2535.
    OpenUrlAbstract/FREE Full Text
  48. ↵
    Pavlides, J.M.W., Zhu, Z., Gratten, J., McRae, A.F., Wray, N.R., and Yang, J. (2016). Predicting gene targets from integrative analyses of summary data from GWAS and eQTL studies for 28 human complex traits. Genome Medicine 8, 1–6.
    OpenUrl
  49. ↵
    Hosokawa, Y., Maeda, Y., Takahashi, E.-i., Suzuki, M., and Seto, M. (1999). Human Aiolos, an Ikaros-Related Zinc Finger DNA Binding Protein: cDNA Cloning, Tissue Expression Pattern, and Chromosomal Mapping. Genomics 61, 326–329.
    OpenUrlCrossRefPubMedWeb of Science
  50. ↵
    Quintana, F.J., Jin, H., Burns, E.J., Nadeau, M., Yeste, A., Kumar, D., Rangachari, M., Zhu, C., Xiao, S., Seavitt, J., et al. (2012). Aiolos promotes TH17 differentiation by directly silencing Il2 expression. Nat Immunol 13, 770–777.
    OpenUrlCrossRefPubMed
  51. ↵
    Farh, K.K.-H., Marson, A., Zhu, J., Kleinewietfeld, M., Housley, W.J., Beik, S., Shoresh, N., Whitton, H., Ryan, R.J.H., Shishkin, A.A., et al. (2015). Genetic and epigenetic fine mapping of causal autoimmune disease variants. Nature 518, 337–343.
    OpenUrlCrossRefPubMed
  52. ↵
    Gutierrez-Arcelus, M., Ongen, H., Lappalainen, T., Montgomery, S.B., Buil, A., Yurovsky, A., Bryois, J., Padioleau, I., Romano, L., Planchon, A., et al. (2015). Tissue-Specific Effects of Genetic and Epigenetic Variation on Gene Regulation and Splicing. PLoS Genet 11, e1004958.
    OpenUrlCrossRefPubMed
  53. ↵
    Parsons, T.J., Power, C., Logan, S., and Summerbelt, C.D. (1999). Childhood predictors of adult obesity: a systematic review. International Journal of Obesity 23.
  54. ↵
    Fall, T., Hägg, S., Mägi, R., Ploner, A., Fischer, K., Horikoshi, M., Sarin, A.-P., Thorleifsson, G., Ladenvall, C., Kals, M., et al. (2013). The Role of Adiposity in Cardiometabolic Traits: A Mendelian Randomization Analysis. PLoS Medicine 10, e1001474.
    OpenUrl
  55. ↵
    Pickrell, J. (2015). Fulfilling the promise of Mendelian randomization. bioRxiv.
  56. ↵
    Davey Smith, G., and Ebrahim, S. (2003). ‘Mendelian randomization’: can genetic epidemiology contribute to understanding environmental determinants of disease? International Journal of Epidemiology 32, 1–22.
    OpenUrlCrossRefPubMedWeb of Science
  57. ↵
    Wang, J., Gamazon, Eric R., Pierce, Brandon L., Stranger, Barbara E., Im, Hae K., Gibbons, Robert D., Cox, Nancy J., Nicolae, Dan L., and Chen, Lin S. (2016). Imputing Gene Expression in Uncollected Tissues Within and Beyond GTEx. The American Journal of Human Genetics 98, 697–708.
    OpenUrlCrossRef
Back to top
Posted September 01, 2016.
Download PDF
Email
Integrating gene expression with summary association statistics to identify susceptibility genes for 30 complex traits
Nicholas Mancuso, Huwenbo Shi, Pagé Goddard, Gleb Kichaev, Alexander Gusev, Bogdan Pasaniuc
bioRxiv 072967; doi: https://doi.org/10.1101/072967
Twitter logo Facebook logo LinkedIn logo Mendeley logo
Citation Tools

Subject Area