INTRODUCTION

The title represents pipeline for Epigenomewide association study (EWAS) and Functional Summary-based Imputation (FUSION) association analyses.

Transcriptomewide association statistic \(z_{TWAS}\) was originally proposed for gene expression data. For a given Trait of interest T for which GWAS summary statistics \(z_T\) is available, the corresponding Wald statistic for TWAS is defined such that $$ z_{TWAS} = \frac{w^T_{ge}z_T}{\sqrt{w^T_{ge}Vw_{ge}}} $$

where \(w_{ge}\) is a weight associated with gene expression and V covariance matrix for \(z_T\), respectively.

By analogy, an epigenomewide association statistic \(z_{EWAS}\) is defined through methylation data so that $$ z_{EWAS} = \frac{w^T_{me}z_T}{\sqrt{w^T_{me}Vw_{me}}} $$

where \(w_{me}\) is the weight associated with methylation. Both approaches allow for imputation using GWAS summary statistics. The derivation of these weights and imputation were done using methods as described in @gusev16 called TWAS as well as in @mancuso17 called Functional Summary-based Imputation (FUSION). The TWAS statistics from both approaches agreed very well.

graph TB; SNP["LD reference panel (bed,bim,fam)"] --> |"EWAS reference panel(top1, blup, lasso, enet, bslmm)"| Methylation; Methylation --> Trait; SNP --> |"GWAS summary statistics (SNP, A1, A2, Z)"| Trait;

A total of 442,920 CpG sites based on Illumina humanmethylation450 chips on 1,.146 individuals in EPIC-Norfolk study were available. Among these, 1,117 individuals also had genotype data from Affymetrix BioBank Axiom chips. HapMap2 SNPs from genetic data of these individuals were extracted via PLINK2 according to cis-positions of each probe and subsequently used to build weight analogous to gene expression data as implemented in computer software TWAS. We filtered probes according to their heritabilities estimated from software GCTA at significant level of 0.01. We then performed EWAS for given GWAS summary statistics. The weight generation and methylation imputation was implemented in software called TWAS-pipeline, which allows for whole epigenome computation. After filtering, 78,133 probes reached significant level 0.01.

The FUSION framework has several advantages: First, it integrates heritability estimation and covariate adjustment for whole-chromosomes with additional models such as LASSO, elastic net, BLUP. Second, it offers cross-validation, joint/conditional analyses with the output also informing top hit SNPs and inferred methylation quantitative trait locus (meQTL). Besides, the new software uses modified GCTA software (gcta_nr_robust) leading to higher yield of probes with heritabilities reaching statistical significance, GEMMA giving BSLMM estimates and ability to align strands with reference panels. As both the increased number of models and cross-validation led to excessive computing time, we dropped BSLMM models and conducted five cross-validations. As a result our reference panel for EWAS imputation contains 77,372 probes reaching the heritability p value threshold of 0.01. The association as well as joint/conditional analysis using our weights and LD panel is implemented in software called EWAS-fusion. Like the original TWAS, our implementation will enable a range of GWAS summary statistics to be used coupled with downstream analysis.

EWAS-fusion is reminiscent of Mendelian Randomisation as shown below,

graph TB; SNP --> |"EWAS reference panel(top1, blup, lasso, enet, bslmm)"| Methylation; Methylation --> Trait; SNP --> |"GWAS summary statistics (SNP, A1, A2, Z)"| Trait;

INSTALLATION

  • To begin, the software FUSION including dependencies such as plink2R and reshape is required. The latest version also requires jlimR. Other facilities to be required are Sun grid engine (sge) or GNU parallel for Linux clusters. A summary is also available from here: https://jinghuazhao.github.io/software-notes/AA/.

  • Install the repository on your system, you will need weights based on epigenetic data or to generate them as described in Weight generation below.

FILE Description
EWAS-weights/ directory for EWAS weights
glist-hg19 Probe list
LDREF/ Reference for LD
EWAS-weights.pos Definition of regions
EWAS-weights.profile* Probe profiles

* It contains information about the probes but not directly involved in the association analysis. Earlier version of EWAS-fusion used EWAS/, RDat.pos, and RDat.profile.

USAGE

The syntax is as follows,

ewas-fusion.sh input-file

These will send jobs to the Linux clusters. The sge error and output, if any, should be called EWAS.e and EWAS.o in your HOME directory.

Input

The input file contains GWAS summary statistics similar to .sumstats as in LDSC with the following columns.

Column Name Description
1 SNP RS id of SNPs
2 A1 Effect allele (first allele)
3 A2 Other allele (second allele)
4 Z Z-scores, taking sign with repect to A1

Output

The results will be in input-file.tmp/ directory.

Annotation

This is furnished with contribution from Dr Alexia Cardona, alexia.cardona@mrc-epid.cam.ac.uk, as follows,

Rscript ewas-annotate.R input-file.tmp

It is assumed that HumanMethylation450_15017482_v1-2.csv is available from the directory containing ewas-annotate.Rbut this can be at different location

Rscript ewas-annotate.R input-file.tmp manifest_location=/at/different/location

Q-Q and Manhattan plots using R/gap can be obtained from

Rscript ewas-plot.R input-file.tmp

Example

The script test.sh uses data reported in @wood14. It downloads and generates an input file called height to ewas-fusion.sh.

ewas-fusion.sh height

The results will be in height.tmp/ once it is done.

The annotation is done with

Rscript ewas-annotate.R height.tmp

The Q-Q and Manhattan plots are generated with

Rscript ewas-plot.R height.tmp

Weight generation

This is a revised and much simplified implementation of codes available from TWAS-pipeline. Under our sge it is furnished with

qsub get_weight.qsub

or

qsub get_weight.qsub 22

for chromosome 22.

Inputs to these are summarised as follows,

File Description
FUSION.pheno PLINK phenotype file containing data for all probes
FUSION.covar PLINK covariate file containing covariates such as PCs
CpG.txt CpG ID, chromosome and position

In addition, PLINK binary pedigree file for each CpG also requires to be prepared, as in files. Although it was not done, it is possible to use code as in 1KG.sh to get around gerneration of these individual files by using a combined one. Note the setup takes advantage of the compact storage of non-genetic data.

The results will be available from the EWAS-fusion directory to be profiled and used for association analysis above. As the number of files is fairly large, cp_weight.qsub is written to put weights from their temporary directories in place while ewas-profile.sh profiles these weights as well as prepares for LDREF. For the version with FUSION, it can be done as follows,

wget -qO- https://data.broadinstitute.org/alkesgroup/FUSION/LDREF.tar.bz2 | tar xfj - --strip-components=1
seq 22|awk -vp=1000G.EUR. '{print p $1}' > merge-list
plink-1.9 --merge-list merge-list --make-bed --out FUSION
rm 1000G.EUR.* merge-list
sort -k2,2 FUSION.bim > EUR.bim

To mirror FUSION, which uses glist-hg19, an equivalent for EWAS needs to be built.

ACKNOWLEDGEMENTS

We wish to thank colleagues and collaborators for their invaluable contributions to make this work possible.

APPENDIX

Additional information for Illumina infinium humanmethylation450 beadchip as in Illumina website

Column Name Description
Index Probe Index
TargetID Identifies the probe name. Also used as a key column for data import.
ProbeID_A Illumina identifier for probe sequence A
ProbeID_B Illumina identifier for probe sequence B
IlmnID Unique CpG locus identifier from the Illumina CG database
Name Unique CpG locus identifier from the Illumina CG database
AddressA_ID Address of probe A
AlleleA_ProbeSeq Sequence for probe A
AddressB_ID Address of probe B
AlleleB_ProbeSeq Sequence for probe B
Infinium_Design_Type Defines Assay type - Infinium I or Infinium II
Next_Base Base added at SBE step - Infinium I assays only
Color_Channel Color of the incorporated baseá (Red or Green) - Infinium I assays only
Forward_Sequence Sequence (in 5'-3' orientation) flanking query site
Genome_Build Genome build on which forward sequence is based
CHR Chromosome - genome build 37
MAPINFO Coordinates - genome build 37
SourceSeq Unconverted design sequence
Chromosome_36 Chromosome - genome build 36
Coordinate_36 Coordinates - genome build 36
Strand Design strand
Probe_SNPs Assays with SNPs present within probe >10bp from query site
Probe_SNPs_10 Assays with SNPs present within probe ?10bp from query site (HM27 carryover or recently discovered)
Random_Loci Loci which were chosen randomly in the design proccess
Methyl27_Loci Present or absent on HumanMethylation27 array
UCSC_RefGene_Name Gene name (UCSC)
UCSC_RefGene_Accession Accession number (UCSC)
UCSC_RefGene_Group Gene region feature category (UCSC)
UCSC_CpG_Islands_Name CpG island name (UCSC)
Relation_to_UCSC_CpG_Island Relationship to Canonical CpG Island: Shores - 0-2 kb from CpG island; Shelves - 2-4 kb from CpG island.
Phantom FANTOM-derived promoter
DMR Differentially methylated region (experimentally determined)
Enhancer Enhancer element (informatically-determined)
HMM_Island Hidden Markov Model Island
Regulatory_Feature_Name Regulatory feature (informatically determined)
Regulatory_Feature_Group Regulatory feature category
DHS DNAse hypersensitive site (experimentally determined)

Bioconductor packages

These are IlluminaHumanMethylation450kanno.ilmn12.hg19 and IlluminaHumanMethylation450kmanifest as shown in minfiDataEPIC.

library(IlluminaHumanMethylation450kanno.ilmn12.hg19)
data(IlluminaHumanMethylation450kanno.ilmn12.hg19)
options(width=200)
annotation.table <- getAnnotation(IlluminaHumanMethylation450kanno.ilmn12.hg19)
head(Locations)
head(Manifest)
head(annotation.table)

with IlluminaHumanMethylation450kanno.ilmn12.hg19 as follows,

IlluminaMethylationAnnotation object
Annotation
  array: IlluminaHumanMethylation450k
  annotation: ilmn12
  genomeBuild: hg19
Available annotation
  Islands.UCSC
  Locations
  Manifest
  Other
  SNPs.132CommonSingle
  SNPs.135CommonSingle
  SNPs.137CommonSingle
  SNPs.138CommonSingle
  SNPs.141CommonSingle
  SNPs.142CommonSingle
  SNPs.144CommonSingle
  SNPs.146CommonSingle
  SNPs.147CommonSingle
  SNPs.Illumina
Defaults
  Locations
  Manifest
  SNPs.137CommonSingle
  Islands.UCSC
  Other

REFERENCES

Freund MK, Burch KS, Shi H, Mancuso N, Kichaev G, Garske KM, Pan DZ, Miao Z, Mohlke KL, Laakso M, Pajukanta P, Pasaniuc B, Arboleda VA. 2018. Phenotype-specific enrichment of mendelian disorder genes near GWAS regions across 62 complex traits. Am J Hum Genet 103: 535–552.

Gusev A, Ko A, Shi H, Bhatia G, Chung W, Penninx BW, Jansen R, Geus EJ de, 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. Nat Genet 48: 245–52.

Mancuso N, Shi H, Goddard P, Kichaev G, Gusev A, Pasaniuc B. 2017. Integrating gene expression with summary association statistics to identify genes associated with 30 complex traits. Am J Hum Genet 100: 473–487.

Raj T, Li YI, Wong G, Humphrey J, Wang M, Ramdhani S, Wang YC, Ng B, Gupta I, Haroutunian V, Schadt EE, Young-Pearse T, Mostafavi S, Zhang B, Sklar P, Bennett DA, De Jager PL. 2018. Integrative transcriptome analyses of the aging brain implicate altered splicing in alzheimer’s disease susceptibility. Nat Genet 50: 1584–1592.

Turner SD. 2018. Qqman: An r package for visualizing GWAS results using q-q and manhattan plots. Journal of Open Source Software 3: 731.

Wood AR, Esko T, Yang J, Vedantam S, Pers TH, Gustafsson S, Chu AY, Estrada K, Luan J, Kutalik Z, Amin N, Buchkovich ML, Croteau-Chonka DC, Day FR, Duan Y, Fall T, Fehrmann R, Ferreira T, Jackson AU, Karjalainen J, Lo KS, Locke AE, Mägi R, Mihailov E, Porcu E, Randall JC, Scherag A, Vinkhuyzen AA, Westra HJ, Winkler TW, Workalemahu T, Zhao JH, Absher D, Albrecht E, Anderson D, Baron J, Beekman M, Demirkan A, Ehret GB, Feenstra B, Feitosa MF, Fischer K, Fraser RM, Goel A, Gong J, Justice AE, Kanoni S, Kleber ME, Kristiansson K, Lim U, Lotay V, Lui JC, Mangino M, Mateo Leach I, Medina-Gomez C, Nalls MA, Nyholt DR, Palmer CD, Pasko D, Pechlivanis S, Prokopenko I, Ried JS, Ripke S, Shungin D, Stancáková A, Strawbridge RJ, Sung YJ, Tanaka T, Teumer A, Trompet S, Laan SW van der, Setten J van, Van Vliet-Ostaptchouk JV, Wang Z, Yengo L, Zhang W, Afzal U, Arnlöv J, Arscott GM, Bandinelli S, Barrett A, Bellis C, Bennett AJ, Berne C, Blüher M, Bolton JL, Böttcher Y, Boyd HA, Bruinenberg M, Buckley BM, Buyske S, Caspersen IH, Chines PS, Clarke R, Claudi-Boehm S, Cooper M, Daw EW, De Jong PA, et al. 2014. Defining the role of common variation in the genomic and biological architecture of adult human height. Nat Genet 46: 1173–86.

Zhao JH. 2007. Gap: Genetic analysis package. 2007 23: 18.

Zhao JH. 2021. Weights of EWAS-fusion. https://doi.org/10.5281/zenodo.5069993.