We obtained summary statistics of hematological traits from a previous genome-wide association study (GWAS) conducted in 173,480 participants without any blood cancer or other major blood disorder from the UK Biobank (N = 132,959) and the INTERVAL studies (N = 40,521) [10]. Blood samples for full blood count analysis were collected by venipuncture in EDTA tubes, and measured by a combination of fluorescence and impedance flow cytometry at the centralized processing laboratory of UK Biocentre (Stockport, UK) within 36 h. Eighteen blood cell traits were analyzed, including 12 red blood cell traits (red blood cell count, mean corpuscular volume, hematocrit, hemoglobin concentration, mean corpuscular hemoglobin, mean corpuscular hemoglobin concentration, red cell distribution width, reticulocyte count, reticulocyte fraction of red cells, immature fraction of reticulocytes, high light scatter reticulocyte count, and high light scatter reticulocyte percentage of red cells), as well as six white blood cell traits (neutrophil count, lymphocyte count, monocyte count, basophil count, eosinophil count and leukocyte count). Details of the summary data from all GWAS were listed in Supplementary Table 1. Single nucleotide polymorphisms (SNP) that passed the generally accepted genome-wide significance threshold (P < 5E−08) were chosen as instrumental variables. Then we clumped the instrumental variables with the clump_data function of the TwoSampleMR package (version 0.5.5). We employed a clumping window of 10,000 kb and linkage disequilibrium cutoff of 0.001, and used the European population in the 1000 Genome Phase 3v5 dataset to identify the leading SNPs. Furthermore, we used the PhenoScanner v2 tool to check for variants associated with other phenotypes (P < 5E−08) which might affect the risk of PBC independent of hematological traits.

Summary statistics of PBC was from genome-wide meta-analysis among 24,510 individuals of European ancestry (Ncase = 8021, Ncontrol = 16,489) [11]. Harmonization was undertaken to rule out strand mismatches and ensure alignment of SNP effect sizes.

We hypothesized that hematological traits as risk factors could influence the susceptibility of PBC, and the following assumptions were satisfied: the genetic variants as instrumental variables are associated with hematological traits; the genetic variants are not associated with any confounders; the genetic variants are associated with risk of PBC through hematological traits (namely horizontal pleiotropy should not be present).

To evaluate the causative effect of hematological traits on the risk of PBC, we performed two-sample MR analysis using the random effects inverse variance weighted (IVW) method, which is widely used in MR studies and could provide robust causal estimates under the absence of directional pleiotropy. A P value below 2.78E−03 (0.05/18) was considered statistically significant after the Bonferroni correction. For the significant association, we further verified the results using another three MR methods, including MR-Egger, weighted median, and weighted mode.

We conducted comprehensive sensitivity analyses to estimate potential violations of the model assumptions in the MR analysis. We performed Mendelian randomization pleiotropy residual sum and outlier (MR-PRESSO) analysis and leave-one-out analysis to detect outlier instrumental variables. Outlier instrumental variables identified by the MR-PRESSO outlier test were removed step-by-step to reduce the effect of horizontal pleiotropy. Cochran’s Q test was executed to check the heterogeneity across the individual causal effects. MR-Egger regression was performed to evaluate the directional pleiotropy of the instrumental variables. We computed the F-statistic of each SNP to evaluate the strength of each instrumental variable. The statistical power was calculated using an online tool at http://cnsgenomics.com/shiny/mRnd/. The statistical analyses were conducted using the R package TwoSampleMR 0.5.5.