Primary biliary cholangitis (PBC) and primary sclerosing cholangitis (PSC) are hepatobiliary autoimmune diseases with strong genetic risk components [1]. PBC is characterized by progressive immune-mediated destruction of interlobular biliary ductules, and PSC by inflammatory and fibrotic processes affecting the intrahepatic and extrahepatic biliary tree; both diseases lead to liver cirrhosis [1–4]. PBC occurs more frequently in women than men and primarily in middle-age, with prevalences ranging from 40 to 400 patients per million and incidences ranging from 0.7 to 49 per million [5–8]. PSC affects 9 to 13 patients per million annually with a male-to-female ratio of 2:1 [3]. In Western countries, 60–80% of PSC cases are associated with inflammatory bowel disease (IBD), while PSC is present in 3–8% of all patients with ulcerative colitis and 1–3% of patients with Crohn’s disease [2, 9, 10]. Both PBC and PSC are classic polygenic disorders, with a genetic load represented by common, mostly non-protein-coding single nucleotide polymorphisms (SNPs), exhibiting similar small effect sizes [1].

Standard genome-wide association studies (GWASs) use high throughput SNP microarray technologies to discover SNPs associated with disease. Generally, the higher the sample size of a GWAS, the higher the number of the associations that reach the genome-wide significance threshold of P < 5 × 10^-8 [2]. By contrast, studies with small sample sizes typically fail to reach GWAS statistical significance. The largest genetic association study for IBDs employed over 75,000 patients and controls and uncovered 163 IBD susceptibility loci [3]. However, studies of the heritable component of PBC and PSC by GWAS have been rather limited, primarily because of the lower prevalence of hepatobiliary autoimmunity, particularly PSC. While several GWASs have been performed using well-characterized PBC patient cohorts from North America, Europe, or Japan, only a few have included PSC patients mainly from northern Europe [4]; however, all of these GWASs had relatively small sample sizes [2]. Despite this, in addition to multiple PSC and PBC associations within the HLA complex, and with non-HLA genes mapping to the chromosome 6p21 major histocompatibility complex (MHC) locus, several associations have been identified at non-MHC susceptibility loci with genome-wide significance [6–17].

Many of these are associated with other immune-mediated diseases, including multiple sclerosis, celiac disease, and type 1 diabetes (T1D) [17, 18]. However, while larger GWAS cohorts, combining patients from different populations, may miss some sub-population-specific risk variants, smaller GWASs generally suffer from lack of statistical power. As a consequence, in GWASs limited to hundreds of patients, few or no associations at the accepted genome-wide significance level have been identified.

To identify susceptibility loci for PBC and PSC in the Polish population, we employed a pooled-DNA GWAS approach, together with a new method for identifying SNP associations. Considering the limited sizes of the cohorts investigated in our GWAS, this approach was effective in identifying novel PBC and/or PSC susceptibility loci. Despite the fact that “index SNPs” were selected at P-values much lower than the standard genome-wide significance threshold, the majority were validated by individual TaqMan SNP genotyping assays.

For association screening, we adopted a cost-effective pooled-DNA GWAS design [20], which involved two separate hybridizations. Individual DNA samples from the same patient cohorts, that passed quality control, were combined in equimolar amounts, according to patient diagnosis, to obtain two sets of DNA pools from the same patients groups; the first set represented 420 PBC patients and 120 PSC patients and the second set, 407 and 111 of these two patient groups, respectively. In contrast, two different cohorts of 724 and 370 controls were employed. Selected SNPs, identified by GWAS, were genotyped using individual DNA samples from the same cohorts of 443 PBC patients, 120 PSC patients, and 934 control subjects in TaqMan SNP genotyping assays.

The method of selecting loci for validation is crucial for a successful GWAS. Our relatively small patient cohorts, particularly the PSC cohort, were statistically unpowered to detect SNPs associated at the standard threshold for genome-wide significance (P < 5 × 10^-8). Therefore, assuming that an “index SNP” at a given locus is usually not independent of neighboring SNPs, we focused on loci forming blocks of at least 10 SNPs in strong allele linkage disequilibrium, defined by the following criteria: the distance between each pair of adjacent SNPs in the block was < 30 kb, and each SNP was associated with a disease at a significance level of P < 0.005. Among blocks meeting such criteria, we selected “index SNPs” associated with disease at P < 10^-4, which were subsequently subjected to TaqMan SNP genotyping assays using individual PBC and PSC patient DNA samples. Although no extensive optimization of the number of required loci, P-value threshold, or window size was performed, the algorithm used was effective.

GWASs provide information about common variants associated with disease susceptibility. Although GWASs allow the identification of disease risk alleles without prior knowledge of their position or biological function, they require large study cohorts to identify associations at the genome-wide significance threshold (5 × 10^-8). By contrast, smaller GWASs using the same threshold may generate false-negative results. For association screening, we performed pooled-DNA GWASs and applied novel selection criteria that took into account strong allele linkage disequilibrium. We selected SNP blocks, defined by a distance of less than 30 kb between each pair of at least 10 SNPs associated with disease at P < 5 × 10^-3, with an “index SNP” in the block for which the association was significant at P ≤ 10^-4. All of the associations identified using this approach would have been missed using the standard genome-wide significance threshold. Of 22 and 29 associations with PBC and PCS, respectively, selected for validation, 19 and 21 SNPs were verified using TaqMan SNP genotyping assays of individual patient and control samples. In total, 19 SNPs reached the stringent (corrected) significance threshold, while the other 21 reached a nominal level of significance (P < 0.05 with OR > 1.2 or < 0.83), demonstrating at least suggestive evidence for association (Table 2). However, the expected number of false-positive results for 50 independent tests, with a significance threshold of 0.05, is < 3, while appropriate correction for multiple comparisons would reduce this to < 1. In our validation studies, correction for multiple testing reduced the number of associations at the nominal level of significance from 40 to 19, thus likely generating a large number of false-negative results and only slightly reducing the expected number of false-positive results.

This study identified 57 SNPs associated with either PBC or PSC or both disorders, which represent 38 genetic regions (Table 2). As expected, there were higher numbers of associations with HLA and non-HLA loci mapping to the MHC region (chromosome 6p21). Of 13 SNPs from this region, two, five, and six SNPs were associated with PBC, PSC, and both disorders, respectively. Of the SNPs shared between PBC and PSC, only two exhibited the same direction of effect.

As both PBC and PSC are hepatobiliary autoimmune diseases with low prevalence, the largest GWASs of these diseases have recruited individuals from different populations, which has certainly introduced a level of heterogeneity into the results, arising from the different genetic backgrounds of the geographically distinct populations. Consequently, these large cohort studies may have missed some subtle, sub-population-specific risk variants that may account for missing heritability. Conversely, studies with smaller sample sizes typically reveal a smaller fraction of the heritability of a complex disease, as they fail to detect associations because any found do not reach statistical significance thresholds [26]. The relative homogeneity of the Polish population may explain, at least in part, why our investigation identified so many SNPs significantly associated with PBC and/or PSC.

A GWAS including 536 North American PBC patients uncovered disease associations for several gene variants in the HLA class II region and coding variants in the interleukin-12a (IL12A) and IL12 receptor b2 (IL12RB2) genes [11]. Further GWASs have replicated these findings in a European population, and identified additional risk genes overlapping with other autoimmune diseases [17, 18]. In six GWASs, 27 non-HLA risk loci associated with PBC were identified [4]. Most have also been implicated in other autoimmune diseases, highlighting different immunoregulatory pathways. Our findings indicated the possible involvement of six previously described regions (1p31.3, rs3790567 [11]; 3q13 [13, 16]; 7q32.1, rs10488631 [12, 14]; 11q23.3 [13, 16]; 17q12, rs9303277 [18]; and 19q13.33, rs3745516 [14]) and the HLA-containing 6p21 locus in the development of PBC in Polish patients (Table 2).

Genomic studies of PSC have uncovered 18 associated genetic regions: 1p36 (TNFRSF14, MMEL1); 2q13 (BCL2L1); 2q33 (CD28); 2q35 (GPBAR1); 2q37.3 (GPR35); 3p21 (USP4, MST1); 4q27 (IL2, IL21); 6q15 (BACH2); 6p21 (HLA region); 10p15 (IL2RA); 11q23 (SIK2); 12q13 (HDAC7); 12q24 (SH2B3, ATXN2); 13q31 (GPC5/6); 18q21.1 (TCF4); 18q22 (CD226); 19q13 (PRKD2, STRN4); and 21q22 (PSMG1) [5, 6, 8, 27, 28]. Of these, only the MHC region was replicated in our PSC patients (Table 2).

The MHC region, which contains more than 224 genes and is highly polymorphic, is known to be associated with more than 100 different autoimmune and infectious diseases [29–31]. In European-based GWASs, the most pronounced MHC associations with PSC were with class I (HLA-B and -C) rather than class II (HLA-DRB1 and -DQB1) loci [27]. A meta-analysis of three independent PBC cohorts identified HLA class II alleles (HLA-DRB1, HLA-DQA1, and HLA-DQB1) achieving genome-wide significance levels, with similar allele frequencies in Canadian, US, and Italian PBC cohorts [15]. Outside of the MHC region, our investigation confirmed six and zero genetic regions uncovered by previous GWASs as associated with PBC and PSC, respectively [27]. Of 30 chromosomal regions representing novel susceptibility loci, 13, 9, and 8 were associated with PBC, PSC, and both disorders, respectively. Of these, 17 SNPs have a shared genetic association with IBD, three with rheumatoid arthritis, two with lupus erythematosus, and single SNPs with psoriasis, lateral sclerosis, T1D, and intrahepatic cholestasis of pregnancy.

While well-designed GWASs should be conducted with groups of at least 1,000 patients and 1,000 controls, the appropriate level of statistical power to test for genetic associations (at P < 5 × 10^-8) often relates to higher effect sizes [32]. However, since loci with a high effect size have generally been efficiently removed from the human population by natural selection, the identification of a common polymorphic susceptibility locus strongly associated with disease, with an OR > 2 or < 0.5, is unlikely [33]. Instead, a large number of previously identified loci associated with different disorders exhibit relatively small effect sizes, with ORs < 1.3. The present study uncovered only one SNP (rs35730843, POLR2G, P = 1.2 × 10^-5, OR = 0.393) strongly associated with PBC and 11 SNPs strongly associated with PSC (rs3822659, coding in WWC1, P = 0.0051, OR = 0.236; rs9686714, intron of WWC1, P = 0.00077, OR = 0.195; rs13191240, intron of ADGRB3, P = 0.0095, OR = 0.2; rs7454108, intergenic between LOC100294145 and C4B_2, P = 0.0013, OR = 0.326; rs2524163, intron of HLA-B, P = 5.9 × 10^-7, OD = 2.02; rs2187668, intron of HLA-DQA1, P = 1.5 × 10^-7, OR = 2.47; rs3130484, intron of MSH5-SAPCD1, P = 5.1 × 10^-11, OR = 3.23; rs1264377, intergenic between PSORS1C3 − MIR877, P = 8 × 10^-8, OR = 2.39; rs3130626, intron of PRRC2A, P = 1.5 × 10^-6, OR = 2.10; rs419788, intron of SKIV2L, P = 1.2 × 10^-6, OR = 2.03; and rs34708188, intergenic close to SENCR, P = 0.0056, OR = 2.27). Of these, nine SNPs map to the MHC region (6p21), while POLR2G, WWC1, and ADGRB3 are located in other genomic regions.

Our results indicated that a rare variant in the POLR2G gene promoter is associated with decreased risk of PBC, with a high effect size in the Polish population. POLR2G encodes one of the subunits in the polymerase 2 RNA complex, which is responsible for transcribing protein coding genes, miRNAs, and some classes of non-coding RNAs [34] and maps to the 11q12.3 locus, within which variants associated with chronic obstructive pulmonary disease [35] and asthma [36] have been identified. We also identified a decreased risk for PCS (OR < 0.25) conferred by a single rare variant in the ADGRB3 gene and two rare variants (rs3822659 and rs9686714) in the WWC1 gene. ADGRB3 encodes transmembrane adhesion G protein-coupled receptor B3 (BAI3), which is broadly expressed in the brain and involved in the regulation of excitatory synapse connectivity [37]. Furthermore, BAI3 can promote myoblast fusion in vertebrates [38]. A previous GWAS identified SNPs at the ADGRB3 locus as associated with early-onset venous thromboembolism [39]. WWC1 encodes the KIBRA protein that plays versatile roles including in the regulation of cellular signaling, cell polarity, vesicular trafficking, and cell migration and division [40]. Specifically, KIBRA is a regulator of the Hippo signaling pathway, which controls tissue growth and tumorigenesis by inhibiting cell proliferation and promoting apoptosis [41]. Notably, WWC1 hypermethylation occurs in 70% of B-cell acute lymphocytic leukemias [42] and its epigenetic silencing is also associated with unfavorable prognostic parameters in chronic lymphocytic leukemia [43]. Interestingly, the triggering of IL-6 trans-signaling, a process of aggregation of extracellular soluble IL-6 receptor and IL-6 associated with rheumatoid arthritis [44] and IBD [45], significantly increased WWC1 expression in human airway smooth muscle cells [46], suggesting a link between its expression and inflammatory diseases. Importantly, rs3822659 is a missense variant (Ser735Ala) that alters the interaction of KIBRA with phosphatidylinositol 3-phosphate [47]. Other GWASs have implicated SNPs in WWC1 as associated with memory performance and cognition [48], as well as Alzheimer’s disease [49].

To the best of our knowledge, we have performed the first GWAS of PBC and PSC patients from the Polish population. Our cost-effective GWAS approach, followed by individual genotyping, allowed us to confirm several previously described associations and discover new susceptibility loci associated with both diseases. Although GWASs allow scanning of the entire genome to identify disease risk alleles without prior knowledge of their position or biological function, the use of statistical, rather than biological, criteria for selection of association findings greatly limits our understanding of the role of newly identified variants related to disease development. More importantly, the contribution of the genetic landscape in the context of environmental factors, and interactions between these two influences in conferring susceptibility to disease, has yet to be elucidated.