Multivariate discovery and replication of five novel loci associated with Immunoglobulin G N-glycosylation

Abstract

Joint modeling of a number of phenotypes using multivariate methods has often been neglected in genome-wide association studies and if used, replication has not been sought. Modern omics technologies allow characterization of functional phenomena using a large number of related phenotype measures, which can benefit from such joint analysis. Here, we report a multivariate genome-wide association studies of 23 immunoglobulin G (IgG) N-glycosylation phenotypes. In the discovery cohort, our multi-phenotype method uncovers ten genome-wide significant loci, of which five are novel (IGH, ELL2, HLA-B-C, AZI1, FUT6-FUT3). We convincingly replicate all novel loci via multivariate tests. We show that IgG N-glycosylation loci are strongly enriched for genes expressed in the immune system, in particular antibody-producing cells and B lymphocytes. We empirically demonstrate the efficacy of multivariate methods to discover novel, reproducible pleiotropic effects.

Decreased IgG core fucosylation, a player in antibody-dependent cell-mediated cytotoxicity, is associated with autoimmune thyroid diseases

“…We and others identified several loci associated with IgG and plasma glycosylation and autoimmune conditions in the previous GWAS 24,25,31,32 , so we wanted to relate those findings to the genetic background of the subjects in this study to determine whether there is a shared heritability of AITD and IgG N-glycan traits. We estimated the proportions of genetic and environmental variance in the phenotypic variation of AITD status and different IgG N-glycan traits using the twin data from the TwinsUK cohort and the maximum likelihood variance component model fitting, also known as a structural equation-modeling or ADCE model 33 Supplementary Table 6), as previously estimated in a Danish twin cohort [25][26][27] .…”

“…On the other hand, it was observed that among HT, GD, and PTC groups, HT patients had significantly lower core fucose content on TgAb than the other two groups 22 . Furthermore, since recent genome-wide association studies (GWAS) identified novel loci associated with IgG glycosylation, which were known to be strongly associated with autoimmune conditions 23,24 and the heritability of AITD is estimated to 55-75% [25][26][27] , the next logical step was to examine if there was any common genetic background between those…”

Beyond power: Multivariate discovery, replication, and interpretation of pleiotropic loci using summary association statistics

“…Cichonska et al [11] 55 get CCA results using summary-level data, but how to interpret and replicate CCA 56 results are not thoroughly discussed. Based on CCA, using individual-level data, we 57 introduced a combined phenotype score to assess the genetic effect so that the 58 replication can be meaningful [12]. More importantly, what geneticists want to acquire 59 from multi-trait analyses is additional knowledge about pleiotropy, while such an aim is 60 February 22, 2019 4/27 hardly reached by current multivariate methods.…”

“…For the 23 loci replicated by the MVA-significance test, to test the consistency of 284 pleiotropic effects, we implemented two novel replication strategies as follow-up methods 285 of the MVA-significance replication: MVA-score replication, which we previously 286 developed for individual-level data analysis [12], and MC-based correlation replication 287 (S1 Fig). The MVA-score replication integrates the multivariate genetic effects into the 288 genetic effect on MVA-score, and aims to replicate the discovered locus pleiotropic 289 profile by testing this integrated effect; while the MC-based correlation replication tests 290 the consistency of multivariate genetic effects directly.…”

“…The replication results from these two methods can 372 provide additional evidence for the existence of pleiotropy. The first method is 373 MVA-score replication [12], which aims to replicate the association between SNP and 374 the optimal linear combination of traits. This method is closely related to CCA (S2 Fig).…”