Mapping the serum proteome to neurological diseases using whole genome sequencing

Abstract

Despite the increasing global burden of neurological disorders, there is a lack of effective diagnostic and therapeutic biomarkers. Proteins are often dysregulated in disease and have a strong genetic component. Here, we carry out a protein quantitative trait locus analysis of 184 neurologically-relevant proteins, using whole genome sequencing data from two isolated population-based cohorts (N = 2893). In doing so, we elucidate the genetic landscape of the circulating proteome and its connection to neurological disorders. We detect 214 independently-associated variants for 107 proteins, the majority of which (76%) are cis-acting, including 114 variants that have not been previously identified. Using two-sample Mendelian randomisation, we identify causal associations between serum CD33 and Alzheimer's disease, GPNMB and Parkinson's disease, and MSR1 and schizophrenia, describing their clinical potential and highlighting drug repurposing opportunities.

Fig. 1. pQTL signals for 107 serum proteins from Olink neurology and neuro-exploratory panels.

a 3D Manhattan plot of detected pQTLs. The x axis represents each of the 107 proteins; the y axis represents the chromosome location of each signal; and the z axis represents the −log10 p-values of each association signal. b Scatterplot of pQTL variant location against the location of the gene encoding the target protein. Each dot represents an independent variant. Cis-pQTLs are coloured in teal, while trans-pQTLs are in orange.

Fig. 2. Overall genetic architecture of 107 serum proteins of neurological relevance.

a A total of 214 independent variants were detected. Cis-acting variants were defined as variants lying within 1 Mb upstream and downstream of the gene encoding the target protein, while trans-acting variants are variants that lie outside of this region. Most severe consequence was determined by Ensembl’s variant effect predictor (VEP). Effects more than missense included ‘stop_gained’, ‘frameshift_variant’, and ‘splice_acceptor_variant’ in our dataset; ‘Regulatory region’ variants include ‘[3/5]_primeUTR_variant’, ‘TF_binding_site_variant’, ‘splice_region_variant’, and ‘regulatory_region_variant’; while’Others’ comprises mostly intergenic and intronic variants. Novelty was assessed by cross-referencing published summary statistics from other pQTL studies (Supplementary Data 2). Known pleiotropic loci were not considered novel. Rare, low-frequency and common variants were defined as variants with minor allele frequency (MAF) < 1%, MAF 1–5%, and MAF > 5%, respectively. b Number of proteins for which we detected only cis-pQTLs, trans-pQTLs, or both.

Fig. 3. Causal protein-disease associations identified using two-sample Mendelian randomisation.

We investigated the causal effect of serum proteins (exposure) on various neurological traits (outcome), indicated in the first two columns in the plot. PubMed IDs (PMIDs) are given where manually downloaded summary statistics were used; other IDs are those as given in MRBase (https://gwas.mrcieu.ac.uk/). The number of variants used in the analysis are given in the ‘nSNP’ column. The ‘pBH’ column contains the FDR-adjusted (Benjamini–Hochberg) P-value for each test. Protein–trait pairs with only one variant were analysed using the Wald ratio method, while those with more than one variant were analysed using the inverse variance-weighted (IVW) method. Data are represented as mean odds ratio ± SEM. *Additional signal arising from analysis using only cis-pQTLs as instrumental variables.

Fig. 4. Colocalisation plots. Each plot shows the association signal and the −log10 P-values.

The lead pQTL variant is represented by a black diamond, while other points are variants that are coloured according to the extent of linkage disequilibrium with the lead variant. The location of the genes of interest are also shown in yellow at the bottom of each plot. Significance thresholds used for each respective study are shown using a dotted red line. a Left: Protein QTL signal for serum CD33; right: GWAS signal for Alzheimer’s disease. b Left: Protein QTL signal for serum GPNMB; right: GWAS signal for Parkinson’s disease.

Fig. 5. Serum MSR1 is causally associated with schizophrenia.

a Genetic architecture of serum MSR1. Each of the three independent variants and their LD variants are represented in orange, teal, and purple, respectively; the intensity of the colours indicates the strength of linkage disequilibrium (r²). A rare deletion is also indicated in purple, and is in complete LD with the independent variant rs182190568. Below the signal plot, the location of the variants respective to the gene are indicated using coloured points for the SNVs and a dotted box for the deletion. b Proposed mechanism of how decreased MSR1 may lead to neuronal damage, resulting in neuropsychiatric disease. c Association signal plots at the MSR1 locus for (clockwise) serum MSR1, schizophrenia, gene expression of MSR1 in nucleus accumbens tissue, aorta, oesophagus muscularis and tibial artery. The lead pQTL variant is denoted by a black diamond, while variants in LD are coloured according to the strength of LD with the lead variant (red [r² > 0.8]; orange [0.5 < r² > 0.8]; blue [0.2 < r² > 0.5]; grey [r² < 0.2]).