Introduction
Ankylosing spondylitis (AS), an autoinflammatory disorder, is an unusual but well-known cause of chronic back pain. Common signs of AS are joint pain and stiffness, typically occurring at joints of the spine, as well as in the pelvis, shoulders, or hips. AS may progress to symptoms such as deformed joints, limited lumbar movement, and reduced thoracic vertebral activity. Endochondral ossification slowly progressing to fusions of spinal segments is a major cause of the symptoms. The extreme pattern can lead to the bony fusion of vertebral joints and eventually become a disability. Currently, AS is incurable and thought to be caused mainly by uncharacterized genetic factors [
1].
The prevalence of AS per ten thousand is approximately 18.6 in Europe, 18.0 in Asia, 10.2 in Latin America and 7.4 in Africa (South Africa). However, the prevalence of AS is extremely high in certain countries, such as Turkey (11.9–49.0), China (37.1 specifically in the Shenzhen area), Italy (37.0), Taiwan (33.7) and the USA (31.9) [
2]. AS has a significant correlation with human leukocyte antigen B27 (HLA-B27). The prevalence of HLA-B27 in the AS population is higher than 90% [
3]. A recent study suggested that upregulation of the tissue-nonspecific alkaline phosphatase (TNAP)-related pathway caused by misfolding of HLA-B27 may contribute to the abnormal osteogenesis of AS in both a cell model and animal model. Additionally, the therapeutic potential of agents inhibiting TNAP was shown, though some adverse effects have been reported [
4,
5].
HLA-B27 is considered one of the most important genetic factors contributing to AS. However, only 1–2% of HLA-B27 carriers develop AS, so HLA-B27 is not always reliable as a diagnostic or prediction criterion [
3]. Furthermore, populations with a higher AS prevalence do not have a significantly higher ratio of HLA-B27 carriers. The prevalence of HLA-B27 carriers is approximately 10% among Caucasians, 8% among Han Chinese and 6% among the general population in Taiwan [
6]. Previous familial aggregation studies indicated that heritability affects a considerable proportion of individuals with AS susceptibility [
7]. Furthermore, ethnicity-specific genetic factors might be associated with disease severity and the high prevalence in certain populations [
8,
9].
Genome-wide association studies (GWASs) are used to investigate correlations between genetic variants and traits of interest, especially associations between SNPs and diseases [
10]. In the past decade, several GWASs have investigated the risk SNPs associated with AS, and hundreds of risk SNPs have been identified [
11‐
20]. One risk AS-associated SNP, rs17192932, is specific to the Turkish population, which has a low prevalence of HLA-B27; this suggests the existence of ethnicity-specific risk SNPs in certain populations [
19]. To date, no GWAS has been performed to study AS in individuals of Taiwanese descent.
The Taiwan Precision Medicine Initiative (TPMI) has recruited volunteers to collect Taiwanese genetic data and develop precision-based medicine since 2018. We used data from Tri-Service General Hospital (TSGH), which has joined the TPMI, to perform a GWAS to investigate risk SNPs associated with AS in the Taiwanese population.
We aimed to explore the race-specific AS susceptibility SNPs in Taiwanese individuals and to investigate the association between HLA-B27 and the AS susceptibility SNPs in Taiwan.
Discussion
This is the first GWAS of a Taiwanese AS population analyzed with the TPMI database. The association study followed a typical protocol for GWAS. We conducted two association studies based on two batches of genotyping data and one based on a dataset merging the two batches of genotyping data. Only the SNPs (n = 147) raised in three associated studies were considered AS-associated SNPs. The SNPs located on the sixth chromosome had higher susceptibility in the AS group. There was no overlap between our results and the susceptibility SNPs found in people of other races. These 147 AS-associated SNPs were assigned to 12 haplotype blocks. The SNP with the lowest p value among every haplotype block was considered the tag SNP. Nine tag SNPs corresponded to genes, and 11 tag SNPs had statistically significant associations between HLA-B27 and genotypes with minor alleles.
The precise pathogenesis of AS is still unknown. However, this autoimmune disease is related to multifactor interactions, such as genetic background, immune response, environmental factors, and microbial infection [
3]. Since 1961, AS has been known to involve a non-sex-linked dominant hereditary mechanism [
35]. Genetic effects have been identified as causative factors, accounting for more than 90% of the population variation [
36]. Previous studies indicated that the major histocompatibility complex (MHC) on chromosome arm 6p and HLA-B27, one of the MHC-1 molecules, is strongly linked to and associated with AS [
37,
38]. This is compatible with our results. Although approximately 95% of Caucasian patients with AS are HLA-B27 positive, only 8% of HLA-B27-positive individuals in the population develop the disease [
39‐
41]. This means that HLA-B27 is essential for family inheritance but that there are still other genetic risk factors. HLA-B27 has a high degree of genetic polymorphism, and more than 100 known subtypes have been identified. The distinct subtypes are related to the prevalence of AS in the different regions of the world. The most significant subtypes associated with AS are HLA-B*27:05 (Caucasians), HLA-B*27:04 (Chinese), and HLA-B*27:02 (Mediterranean populations) [
42,
43]. Laval’s whole-genome screening study indicated that genes localized to chromosomes 1p, 2q, 6p, 9q, 10q, 16q, and 19q were associated with AS [
44].
GWASs have been used to map the patterns of inheritance for the SNP, the most common form of genomic variation [
45,
46]. A GWAS in 2010 surveying AS in a large population of European descent revealed that multiple gene variants, including ARTS1, IL23R, ANTXR2 and IL1R2, confer AS risk [
47]. In the past decade, the following GWASs have identified 113 SNPs affecting the risk of developing AS. Furthermore, an ongoing GWAS will likely identify more than 100 new risk loci [
14,
17,
48]. However, GWASs of the Han Chinese AS population are few [
12,
49], and no GWAS has been performed among the Han Taiwanese AS population. A previous study indicated that ethnic differences would lead to genetic heterogeneity in AS susceptibility. Some genes, including those in the 2p15, ERAP1, and NPEPPS–TBKBP1 regions, may still play a critical role in AS pathogenesis across diverse populations [
50].
Our results revealed that the AS-associated SNPs were clustered around HLA-B27. While many of them were located in intergenetic regions (30%), the others could be mapped to a group of genes. Among them, 38 SNPs were mapped to HLA-B, and some were mapped to HCP5 (13 SNPs), AL671883.3 (9 SNPs), POU5F1 (7 SNPs), CCHCR1 (6 SNPs), LINC02571 (5 SNPs), MICB (5 SNPs), MUC22 (4 SNPs), TCF19 (4 SNPs), and MICA (4 SNPs), among others. (Fig.
3).
Data collected from GTEx Portal reveal that SNPs are associated with the expression level of their mapped genes. The tag SNP, rs142577772, of Haplotype block 1 is located in the 3 prime UTR of the GNL1 gene. The mutation position of rs142577772 is the CCCTC-binding factor binding site. This multifunctional transcription regulator might affect the expression of multiple epigenes [
51]. The GNL1 gene and HLA-E gene present a high degree of linkage disequilibrium. There is a strong association between the HLA-E gene and AS haplotype [
52]. It could be inferred that the GNL1 gene might be associated with AS. The tag SNP, rs2073716, of haplotype block 3 is located in the intron of the CCHCR1 gene. The CCHR1 locus may be protective against AS [
53]. rs7766452 of haplotype block 5, rs9368671 of haplotype block 6 and rs28862571 of haplotype block 7 are located in introns of the HLB-B gene. The HLA-B gene is listed as an AS-related gene in the GWAS Catalog database. A GWAS of Turks and Iranians indicated that rs17192932, HLA-B*2705, HLA-B*2702 and HLA-B2707 are variants of HLA-B related to AS [
19]. rs6936035 of haplotype block 9 is located in the intron of the AL671883.3 gene. The AL671883.3 gene was shown to be an AS-related gene in the GWAS Catalog database and previous studies [
11,
14]. rs2251396 of haplotype block 10 is located in the intron of the MICA gene. The MICA gene is listed as an AS-related gene in the GWAS Catalog database and a previous study [
13]. rs3094228 of haplotype block 11 and rs9688839 of haplotype block 12 are located in the introns of the HCP5 gene. Coit’s study indicated that the genetic variant present in the CpG methylation site in HCP5 determines its methylation status and is linked to HLA-B*27 status in AS patients [
54].
HLA-B27 is a necessary factor for the development of AS. However, only 77.6% of subjects in our first batch and 73.7% of subjects in our second batch were HLA-B27 positive. The minor allele frequency in the case group was significantly higher than that in the South Han Chinese population (Additional file
2: Table S5). The genotype distribution of the three SNPs (rs2524069, rs2524067 and rs7766452) with the smallest p value in haplotype block 5 showed that the proportion of HLA-B27-positive people with minor alleles was higher than that of HLA-B27-negative people. Most people with HLA-B27 positivity carry only one minor allele. Most people who are HLA-B27 negative carry the major allele. This might infer that the proportion with minor alleles in the SNP is relatively high in the Taiwanese population. This would cause the prevalence of AS to be higher than that in other regions of the world, but the proportion of HLA-B27 is not higher.
In addition, we used six PRS methods to estimate the risks of developing AS in Taiwanese populations. Overall, six PRS models yielded good performance with an AUROC of approximately 0.76 (Fig.
4A), and the top 10% of PRSs showed at least a fivefold increase in developing AS compared to the remaining lower risk groups (Fig.
4C). It is noteworthy that we employed the train-test split method to evaluate the PRS models using the independent testing cohort to avoid the overfitting problem. Among the six AS-PRS models, the GenEpi model achieved the highest performance in terms of its AUROC value (Fig.
4A). In contrast to the other five methods, GenEpi applies a machine learning approach to identify the epistasis effect of joint genetic effects associated with AS. Indeed, GenEpi identified 110 significant SNP-SNP interactions across entire genomic loci harboring many different genes (Additional file
2: Table S2). The most significant interaction effect on AS was found between two SNPs, rs2844532 (near HLA-S gene) and rs2904599 (near MICA gene), with a
p value of 1.628 × 10
–125. Both the HLA-S and MICA genes have been implicated to have SNPs associated with AS in a previous study, whereas the joint effect of two SNPs within these two genes has not yet been suggested [
13]. On the other hand, we identified several genomic loci, including 23 genes, as having consensus SNPs that contributed considerably to at least three PRS models (Fig.
3D). Among 23 genes, many have been suggested to be associated with AS and other related autoimmune diseases, such as psoriasis and vitiligo, in GWASs. For example, WASF5P and LINC02571 have both been reported to be significantly associated with vitiligo in the Chinese Han population by GWAS [
55]. Additionally, HLA-B, HLA-C, and MICA are indeed AS-related genes, as discussed in the aforementioned paragraphs. Since six different PRS methods apply different algorithms to evaluate the importance of SNPs to estimate the risk of developing AS, the genes with consensus SNPs could be considered critical genetic information related to AS. Therefore, the genomic loci in Fig.
4D might potentially be used as amplicon-based genes for the prediction of AS.
Several limitations exist. First, the data are from the division of rheumatology of a single medical center of the TPMI. As a result, there might be some AS patients going to other hospitals due to non-AS medical problems and joining the TPMI. These AS patients might be classified as a control group, which would cause bias. Second, we obtained the data from medical records. These data lack detailed basic demographic variables, personal health behaviors and living environment exposure. In addition, this study lacks HLA-B27 genotyping data in a normal population because only patients with symptoms of AS had HLA-B27 detected. Therefore, we could not analyze the association between HLA-B27 and SNPs in the general population. While SNPs are likely to have an effect on their mapped genes, SNPs located in intergenic regions were not assigned to any genes to avoid error of prediction. Third, the AS susceptibility SNPs found in this study almost did not overlap with the related SNPs in the GWAS Catalog database. For some SNPs, there is no relevant information or gene expression data for the corresponding gene in the database of the GTEx portal. However, all data in the GTEx portal are from donors in the United States. This discrepancy might be due to ethnic differences. In addition, few GWASs have been performed among the Han Chinese population [
12,
49]. Currently, there is no GWAS of AS populations in Taiwan. We have few previous data to compare.
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.