A common regulatory variant in SLC35B4 influences the recurrence and survival of prostate cancer

Abstract Single nucleotide polymorphisms (SNPs) within the regulatory elements of a gene can alter gene expression, making these SNPs of prime importance for candidate gene association studies. We aimed to determine whether such regulatory variants are associated with clinical outcomes in three cohorts of patients with prostate cancer. We used RegulomeDB to identify potential regulatory variants based on in silico predictions and reviewed genome‐wide experimental findings. Overall, 131 putative regulatory SNPs with the highest confidence score on predicted functionality were investigated in two independent localized prostate cancer cohorts totalling 458 patients who underwent radical prostatectomy. The statistically significant SNPs identified in these two cohorts were then tested in an additional cohort of 504 patients with advanced prostate cancer. We identified one regulatory SNPs, rs1646724, that are consistently associated with increased risk of recurrence in localized disease (P = .003) and mortality in patients with advanced prostate cancer (P = .032) after adjusting for known clinicopathological factors. Further investigation revealed that rs1646724 may affect expression of SLC35B4, which encodes a glycosyltransferase, and that down‐regulation of SLC35B4 by transfecting short hairpin RNA in DU145 human prostate cancer cell suppressed proliferation, migration and invasion. Furthermore, we found increased SLC35B4 expression correlated with more aggressive forms of prostate cancer and poor patient prognosis. Our study provides robust evidence that regulatory genetic variants can affect clinical outcomes.

Number: KMUH104-4R47, KMUH106-6R57; E-Da Hospital, Grant/Award Number: EDAHP104053, EDPJ104059, EDPJ105054, EDPJ106038 functionality were investigated in two independent localized prostate cancer cohorts totalling 458 patients who underwent radical prostatectomy. The statistically significant SNPs identified in these two cohorts were then tested in an additional cohort of 504 patients with advanced prostate cancer. We identified one regulatory SNPs, rs1646724, that are consistently associated with increased risk of recurrence in localized disease (P = .003) and mortality in patients with advanced prostate cancer (P = .032) after adjusting for known clinicopathological factors. Further investigation revealed that rs1646724 may affect expression of SLC35B4, which encodes a glycosyltransferase, and that down-regulation of SLC35B4 by transfecting short hairpin RNA in DU145 human prostate cancer cell suppressed proliferation, migration and invasion. Furthermore, we found increased SLC35B4 expression correlated with more aggressive forms of prostate cancer and poor patient prognosis. Our study provides robust evidence that regulatory genetic variants can affect clinical outcomes.

K E Y W O R D S
multi-stage association study, prognosis, prostate cancer, regulatory variant, SLC35B4

| INTRODUCTION
Prostate cancer is one of the most common cancers worldwide, and its incidence is increasing in most countries. Despite widespread prostate cancer screening with prostate-specific antigen (PSA), 15% of patients are still diagnosed at an advanced stage of the disease. 1 Radical prostatectomy (RP) and androgen deprivation therapy (ADT) are the recommended primary management strategies for patients with localized and advanced prostate cancer, respectively. These therapies are effective in the early stage of the disease, but recurrence and resistance to therapy are the principal causes of morbidity and mortality from prostate cancer. Currently, the defined risk classification of prostate cancer consists of clinical features including PSA level, Gleason score and tumour stage.
Although classification is straightforward and useful, patient outcome varies widely. Therefore, additional biomarkers are needed for improved prognosis or early detection to better guide patient management.
Considerable progress towards understanding the genetic basis of prostate cancer has been made in recent years because of the development of high-throughput genotyping technologies. 2,3 However, most single nucleotide polymorphisms (SNPs) tested in genome-wide association studies are located outside of the coding regions of genes and therefore have been difficult to interpret. 4 To identify variants that do not directly affect exonic sequence, but may confer a regulatory effect on gene expression, many studies have mapped open chromatin and protein binding regions for a large number of factors across many cell types. 5 In addition, computational processing and machine learning have furthered our predictive capabilities in identifying regulatory variants. We hypothesized that such functional regulatory SNPs have biologically important functions that may have an effect on cancer progression. In this multi-stage study, we systematically evaluated a comprehensive panel of predicted and known regulatory SNPs to determine their roles in prostate cancer progression and patient mortality, as illustrated in Figure 1.

| Patient recruitment and data collection
This study included 458 patients with localized prostate cancer who underwent RP as initial therapy and 504 patients with advanced prostate cancer who received primary ADT, as described previously. [6][7][8][9][10][11] The localized prostate cancer cohort consisted of participants from two independent data sets. The discovery data set was composed of 246 patients from the National Taiwan University Hospital located in northern Taiwan, and the replication data set was composed of 212 patients from the Kaohsiung Medical University Hospital, the E-Da Hospital and the Kaohsiung Veterans General Hospital, all located in southern Taiwan. Clinical and follow-up data were obtained from patient medical records. The primary end-point was biochemical recurrence (BCR), which was defined as two consecutive PSA values of at least 0.2 ng/mL after RP. 12,13 Patients in the advanced prostate cancer cohort were recruited from all four medical centres. Overall survival (OS) was defined as the duration from diagnosis to death of any cause or to last follow-up. This study was approved by the Institutional Review Boards of the National Taiwan University Hospital, the Kaohsiung Medical University Hospital, the E-Da Hospital and the Kaohsiung Veterans General Hospital.
Written informed consent was obtained from each patient prior to their participation in the study, which was performed in accordance with the approved guidelines.

| SNP selection and genotyping
The RegulomeDB database (http://www.regulomedb.org) uses a scoring system to guide interpretation of putatively regulatory variants based on their functional confidence. 14  Genomic DNA was extracted from peripheral blood samples from participants using the QIAamp DNA Blood Mini Kit (Qiagen, Valencia, CA) and stored at À80°C. Genotyping was performed at the National Center for Genome Medicine (Taiwan) using the Agena Bioscience MassARRAY iPLEX system (Agena Bioscience, San Diego, CA), as described previously. 9 The average genotype call rate for these SNPs was 98.9%, and the concordance rate was 100% among 10 blind duplicated quality control samples. Any SNP that failed the assay design (N = 18), deviated from Hardy-Weinberg equilibrium (P < .005, N = 13), or had a genotyping call rate <85% (N = 2), was removed. Therefore, 131 SNPs underwent statistical analysis.

| Bioinformatics analysis
We used several bioinformatics tools to assess whether rs1646724 or linked genetic variants were associated with a putative function that may have an effect on patient outcome. Tests of association between rs1646724 and SLC35B4 expression were evaluated using mRNA data of lymphoblastoid cell lines derived from HapMap and consisting of 90 Utah residents with ancestry from northern and western Europe (CEU), 45 CHB, 45 Japanese in Tokyo, Japan (JPT) and 90 Yoruba in Ibadan, Nigeria (YRI). 16

| Statistical analysis
Patient clinicopathological characteristics were summarized as either the number and proportion (%) of patients or as the median and interquartile range. Evidence of association between clinicopathological characteristics and either BCR or OS was assessed using the logrank test or Cox regression. Multivariate Cox proportional hazards regression analyses were used to assess the effect of each SNP on clinical outcomes with adjustments for clinicopathological variables, as described previously. 13 Of the 131 regulatory SNPs that were genotyped, we found 12 SNPs were associated with BCR in the discovery data set of patients with localized prostate cancer (all P < .05, Table S1). This subset of significant SNPs was tested using the replication data set, and two SNPs, rs1646724 and rs4239504, were validated in the replication cohort (P ≤ .047) and in the combined analysis (P ≤ .005; Table 2).
After adjusting for age, PSA, pathological Gleason score and stage, we found that rs1646724 and rs4239504 remained significant (P ≤ .009). The minor G allele of rs1646724 was associated with an increased risk of BCR in a dose-dependent manner [HR 1.38, 95% confidence interval (CI) 1.09-1.74, P = .007; Table 2 and Figure 2A].
In addition, patients with the rs4239504 CC genotype had a 1.78fold increased risk of BCR (95% CI 1.15-2.75, P = .009; Table 2 and Figure 2B). Concordant with our findings in patients with localized prostate cancer, we found that rs1646724 was also significantly associated with OS in patients with advanced prostate cancer after adjusting for age, PSA at ADT initiation, Gleason score, stage, PSA nadir and treatment modality (HR 1.35, 95% CI 1.03-1.78, P = .032; Table 3 and Figure 2C).
The regulatory variant rs1646724 is located 71 base pair (bp) upstream to the SLC35B4 start site. Thus, we performed an in silico analysis using genetic variation and gene expression data from Hap-Map populations to determine whether rs1646724 could affect SLC35B4 expression. From our analysis, a positive Spearman correlation coefficient indicated that the rs1646724 risk allele G is correlated with increased SLC35B4 expression (P < .001, Figure 3A). As an initial step towards understanding the role of SLC35B4 in prostate cancer, we evaluated whether SLC35B4 has a role in proliferation, migration and invasion of prostate cancer cells. First, using Western blot analysis, we examined the protein level of SLC35B4 in cells from two widely used metastatic human prostate cancer cell lines, PC-3 and DU145. As shown in Figure 3B, we found increased SLC35B4 protein in DU145 cells compared to that found in PC-3 cells. Next, using a short hairpin RNA (shRNA) approach, we performed SLC35B4 knockdown in DU145 cells and found that the protein level of SLC35B4 was effectively down-regulated in SLC35B4 knockdown cells (shSLC35B4) compared to cells with a scrambled control (shScramble; Figure 3B). We then evaluated cell proliferation and migration using a real-time xCELLigence biosensor system for 96 hours. We found that SLC35B4 depletion significantly reduced DU145 cell proliferation ( Figure 3C) and migration ( Figure 3D). In addition, we used an invasion assay to measure the capability of cancer cells to degrade and invade a basement membrane matrix.
We found that the number of invading cells was dramatically decreased in shSLC35B4 cells compared to shScramble control cells ( Figure 3E). These findings demonstrate that SLC35B4 may be important in prostate cancer progression.
To understand the clinical relevance of SLC35B4 expression in prostate cancer, we examined SLC35B4 expression in a large number of prostate carcinoma cases using tissue microarrays. Representative images of SLC35B4 immunostaining are shown in Figure 4A. We found a trend towards increased SLC35B4 expression with more aggressive forms of prostate cancer (P < .001, Figure 4B), and that up-regulation of SLC35B4 was correlated with higher Gleason tumour scores (P = .048, Figure 4B). Moreover, our analysis of an independent public data set showed that high SLC35B4 mRNA levels also signified poor BCR-free survival ( Figure 4C). Our results suggest that SLC35B4 warrants further investigation as a potential prognostic marker of prostate cancer.

| DISCUSSION
We systematically analysed a large panel of SNPs residing in regulatory regions that could modulate transcription factor binding, and subsequently, affect target gene expression. In our investigation The expression of cancer-associated glycans, such as sialyl-Lewis X , Thomsen-nouvelle (Tn) antigen and sialyl-Tn antigen, has been detected in many types of cancer, including prostate cancer. 30 Glycosylation is a frequent post-translational protein modification, and the degree of protein glycosylation depends on the glycosylation sites and the expression and activity of glycosyltransferases such as SLC35B4, and glycosidase enzymes. It is established that glycomodification can modulate activity of numerous growth factor receptors including EGFR, PDGFR and IGFR, all of which can sustain proliferative signaling. 31 Cancer cells often have high levels of sialylated glycans, 32 and increased sialylation can increase local negative charges to disrupt cell-cell adhesion. 33 Expression of the cancer-associated sialyl-Tn antigen reduces cell adhesion and increases migration and invasion in prostate cancer. 34 Because the prostate is a secretory gland involved in the secretion of many types of glycoproteins, alterations in glycans are of particular interest as potential biomarkers and therapeutic targets for prostate cancer. Consistent with these observations, we found that increased SLC35B4 expression was associated with more aggressive forms of prostate cancer (Figure 4), and that silencing SLC35B4 markedly suppresses cancer cell proliferation, migration and invasion ( Figure 3). Based on these findings, we posit that SLC35B4 may alter glycosylation thereby modifying some biological processes involved in prostate cancer progression.
The main limitations of this study were that the sample size of our separate cohorts was modest and that multiple statistical comparisons were made. Even with independent discovery and replication cohorts study design to validate our findings and reduce the possibility of a false discovery, our results should still be interpreted with caution owing to the number of statistical tests performed. In addition, using functional assays provided confirmatory support of gene-disease associations. Another limitation of our study was that all participants were Taiwanese, and because of homogeneity of the Taiwanese population, our findings may be less generalizable to other ethnic groups. However, the same homogeneity also enables the identification of unique ethnic-or population-specific culprit genes that may not be found in association studies of other groups.
Additional research is warranted to investigate the biological mechanisms that underlie our observed associations.
In summary, this study provides evidence for an association between the regulatory variant SLC35B4 rs1646724 and the clinical outcomes of patients with prostate cancer, which may be of prog-