C-reactive protein upregulates the whole blood expression of CD59 - an integrative analysis

Fig 1. Pairwise modelling pipeline of whole genome sequencing (WGS), RNA sequencing (RNA-seq) and C-reactive protein (CRP) data. (A) First, we identified genes whose expression levels (denoted by E) were significantly associated with CRP. Second, we used these genes to perform a cis-eQTL analysis and extract SNPs (denoted by G) that act on the expression of those genes. Third, for each triplet (G, E, CRP), we used maximum likelihood to select the best supported model out of a limited number of possible models–given that G is correlated with E, E is correlated with CRP and assuming directed acyclic graphs. The dashed edge in model IV indicates that either E acts on CRP or vice versa–these two models are Markov equivalent so we cannot differentiate between them. Fourth, we ensured that the best candidate models fulfilled necessary partial correlation criteria. Fifth, we subjected the best candidates to MR analysis where the instruments were chosen from published GWAS summary statistics. Finally, we validated the findings using cell culture stimulation assays. (B) Venn diagram of available sample sizes.

Fig 2. Analysis of ΔAIC values. (A) Differences in the Akaike information criterion values between the causal and colliding models (ΔAIC) in triplets best supported by either model. On average, the colliding models have higher ΔAIC values. This indicates that we are more likely to identify genes whose expression is in some way regulated by CRP. (B) Scatter plot of ΔAIC values against the CRP-expression association p-values with a linear trend and 95% confidence interval. Despite a small positive trend, we can observe that higher correlation does not necessarily translate to more evidence of a causal effect.

Fig 3. Analytical validation of the causal CRP and CD59 link. (A) QQ-plot of p-values from the association analysis between CRP-associated SNPs and CD59 expression. The empirical quantiles are not in line with the theoretical quantiles of the uniform distribution (Kolmogorov-Smirnov p = 0.026) and there is some enrichment of small p-values. (B) Funnel plot of minor allele frequency corrected genetic effects on CRP against causal effect estimates between CRP and CD59 expression for each CRP-associated SNP. (C) Scatter plot of the genetic effect on CD59 expression against the genetic effect on CRP. Causal effect slope estimates from the TSLS solutions with the GRSCRP instrument and with all the 16 CRP-associated SNPs as instruments (both forced through zero) are coloured in blue and green, respectively. The bias-corrected slope from the MR-Egger regression is shown in red.

Fig 4. Upregulation of CD59 surface expression by CRP in cell culture experiments. Peripheral blood cells from two donors were treated with five increasing doses of CRP protein. For negative controls, the cells were not treated with CRP or were treated with additive NaN3 only. The CD59 antigen values were measured after 48 hours and are shown in mean fluorescent intensity units as the arbitrary values of flow cytometry. Black dots represent individual measurements in different replicates, red dots are the averages and whiskers represent ±1 standard errors.