A Single Nucleotide Polymorphism rs1010816 Predicts Sorafenib Therapeutic Outcomes in Advanced Hepatocellular Carcinoma

Sorafenib is currently a targeted agent widely used in the treatment of advanced hepatocellular carcinoma (aHCC). However, to date there is still a lack of a reliable marker capable of predicting sorafenib therapeutic responses. Here, we conducted a genome-wide association study (GWAS) to identify candidate single-nucleotide polymorphism outcome predictors in aHCC patients. A total of 74 real-world sorafenib-treated aHCC patients were enrolled for GWAS and outcome analysis. GWAS showed that rs1010816 (p = 2.2 × 10−7) was associated with sorafenib therapeutic response in aHCC patients. Kaplan–Meier analysis indicated that the “TT” genotype was significantly associated with a favorable therapeutic response but not significantly associated with overall survival (OS). Univariate followed by multivariate Cox proportional hazard analysis showed that ascites, main portal vein thrombosis, lower platelet count, lower total sorafenib doses, higher PALBI score in model A and higher ALBI grade in model B were significantly associated with a shorter OS. Subgroup analysis showed that only in alcoholic aHCC patients treated by sorafenib, rs1010816 “TT” genotype was significantly associated with longer OS (p = 0.021). Sorafenib had a favorable therapeutic outcome in alcoholic aHCC patients carrying rs1010816 “TT” genotype.


Introduction
Globally, hepatocellular carcinoma (HCC) accounts for more than 80% of the histological subtypes of primary liver cancer, and it occurs more frequently in male gender, with one in 45 men compared with one in 113 women. HCC ranks the sixth in cancer incidence and the fourth in cancer-related death in the world [1,2]. Multiple etiological factors for HCC are involved, including chronic hepatitis B and C virus infections, alcohol addiction, metabolic liver diseases, and exposure to dietary toxins such as aflatoxin and aristolochic acid [3][4][5][6]. If HCC diagnosis is made in an early stage, curative surgical resection or non-surgical local ablation therapy (such as radiofrequency ablation) can be performed to achieve complete remission. Liver transplantation is considered the best therapeutic modality in selected patients who fulfill designated criteria, resulting in total removal of the tumor and adjacent microinvasion, and the cirrhotic liver entirely replaced [7,8]. In contrast, in patients with incurable HCC, such as those in advanced stages or with distant 2 of 14 metastases, the recommended treatments are largely palliative, including transcatheter arterial chemoembolization for intermediate stage (Barcelona clinic liver cancer [BCLC] stage B) patients, and radiotherapy, chemotherapy, targeted therapy, and immunotherapy for patients in a further advanced stage (BCLC stage C) [9][10][11]. Recently, atezolizumab was identified as an immune checkpoint inhibitor through interaction of programmed death-ligand 1. Bevacizumab was a monoclonal antibody to block vascular endothelial growth factor (VEGF) activity [12]. Atezolizumab plus bevacizumab treatment has been suggested as a first-line therapy in HCC [13]. A comparative clinical study demonstrated that the combined atezolizumab plus bevacizumab treatment was correlated with better survival outcomes of advanced unresectable HCC patients than those with only sorafenib treatment [14]. To date, if immunotherapy is not available, tyrosine kinase inhibitors such as sorafenib and lenvatinib are still recommended as the first-line treatments for advanced HCC (aHCC), and regorafenib and cabozantinib are recommended as the second-line therapies [15][16][17].
Among these targeted agents in the current treatment guidelines, sorafenib was the first approved and is the most widely used targeted drug for patients with advanced-stage HCC. According to the results of two large-scale randomized controlled studies, SHARP (Sorafenib HCC assessment randomized protocol) and Asia Pacific (AP), sorafenib can prolong the overall survival (OS) in patients with incurable HCC, albeit the therapeutic outcomes and long-term prognosis vary widely among patients and the outcomes remain unsatisfactory [18,19]. In fact, sorafenib failed to show statistically significant benefits in aHCC patients with extrahepatic metastasis or macroscopic vascular invasion in a recent subgroup and exploratory analysis [20]. Additionally, unlike occasional complete remission (0-44%) in HCC patients receiving systemic chemotherapy or hepatic arterial infusion chemotherapy, sorafenib rarely resulted in complete remission in a recent study in which only 12 of 1119 (1%) patients were classified as complete responders [11,[21][22][23][24][25]. Therefore, to overcome these problems, it is necessary to determine a reliable marker capable of predicting sorafenib therapeutic responses so that clinicians can carefully select favorable patients to receive sorafenib therapy and, to advise unfavorable patients to choose other novel effective treatments.
In this study, in order to achieve this purpose, we have conducted a pilot genome-wide association study (GWAS) to identify candidate single-nucleotide polymorphism (SNP) markers as a prognosis predictor for sorafenib therapeutic response in BCLC stage C HCC patients.

Genetic Associations with Respect to Objective Responses to Sorafenib Therapy
A genome-wide screen was conducted on this retrospective cohort of 74 subjects, who either had positive sorafenib therapeutic response (n = 24) or no response (n = 50). Figure 1 shows that the top hit rs1010816 (p = 2.2 × 10 −7 ) was associated with genome-wide significance for sorafenib therapeutic response in aHCC patients. Accordingly, rs1010816 was selected for further investigation. Based on the Hardy-Weinberg equilibrium, there are 6, 33 and 35 persons with the GG, GT and TT genotypes at rs1010816, respectively. Accordingly, Kaplan-Meier analysis was used to examine the significance of the therapeutic response, progression-free survival (PFS) and OS for the rs1010816 genotype in aHCC patients receiving sorafenib treatment ( Figure 2). It was found that the "TT" genotype was significantly associated with shorter time-to-therapeutic response (i.e., a better response) when compared with those in "GT" and "GG" genotypes (p < 0.001) (Figure 2a). Subsequently, because the number of patients with "GG" genotype was small, the rs1010816 genotype was reclassified as rs1010816 "TT" and "non-TT". As expected, the "TT" genotype was significantly associated with shorter time-to-response when compared with that of the "non-TT" genotype (p < 0.001) (Figure 2b). However, there was no significant difference among the rs1010816 genotypes in PFS (Figure 2c; p = 0.909) and in OS (Figure 2e; p = 0.213). In addition, no significant difference was found between the "TT" and "non-TT" genotypes in PFS (Figure 2d; p = 0.726) and OS (Figure 2f; p = 0.184). On the other hand, Kaplan-Meier analysis showed that the Child-Pugh grade was significantly correlated with progression free survival and overall survival of aHCC patients, but not in sorafenib therapeutic response (Supplementary Figure S1). tients with "GG" genotype was small, the rs1010816 genotype was reclassified as rs1010816 "TT" and "non-TT". As expected, the "TT" genotype was significantly associated with shorter time-to-response when compared with that of the "non-TT" genotype (p < 0.001) (Figure 2b). However, there was no significant difference among the rs1010816 genotypes in PFS (Figure 2c; p = 0.909) and in OS (Figure 2e; p = 0.213). In addition, no significant difference was found between the "TT" and "non-TT" genotypes in PFS (Figure 2d; p = 0.726) and OS (Figure 2f; p = 0.184). On the other hand, Kaplan-Meier analysis showed that the Child-Pugh grade was significantly correlated with progression free survival and overall survival of aHCC patients, but not in sorafenib therapeutic response (Supplementary Figure S1).

Identification of Genetic and Clinical Predictors for Sorafenib Therapeutic Response
Univariate followed by multivariate Cox proportional hazard analysis was performed to analyze the association between genetic and clinical factors and the therapeutic response (calculated as time-to-response) after receiving sorafenib (Table 2). It was found that the rs1010816 genotype (GG vs. GT vs. TT) (p < 0.001), rs1010816 "TT" (p < 0.001), age (p = 0.033), and sorafenib doses use (p = 0.002) were associated with time-to-sorafenib therapeutic response. However, after being adjusted for confounding factors, multivariate analysis revealed that only the rs1010816 genotype (p = 0.003 in model A and p = 0.002 in model B) was an independent predictor.

Identification of Genetic and Clinical Predictors for Progression-Free Survival
Subsequently, univariate followed by multivariate Cox proportional hazard analysis was performed to understand the genetic and clinical factors associated with PFS (Table 3). It was found that the presence of ascites (p = 0.030), bilirubin (p = 0.002), total sorafenib doses (p < 0.001), Child-Pugh score (p = 0.003), Child-Pugh grade (p = 0.001), and PALBI grade (p = 0.011) were associated with PFS. However, after being adjusted for confounding factors, multivariate analysis revealed that the independent predictors were the presence of ascites (p = 0.042) and sorafenib total doses (p < 0.001) in model A and total sorafenib doses (p < 0.001) in model B.

Subgroup Analysis to Identify Patient Subgroups wherein rs1010816 Genotype Effectively Predicted Overall Survival
To gain more insight into why the rs1010816 genotype was associated with therapeutic response but not OS, we divided patients into different clinical subgroups for further analysis (Figure 3). It was discovered that only in aHCC patients with alcoholism, was the "TT" genotype significantly associated with longer OS (p = 0.021), implying that alcoholrelated aHCC might affect the predictive role of rs1010816 in OS.

Discussion
In the present study, our goal was to identify pretherapeutic biomarkers capable of predicting the outcomes of sorafenib treatment, so that only individuals with a higher probability of responses and favorable outcomes should be treated with this drug. Through GWAS, it was found that there was a linkage disequilibrium in the SNPs surrounding the rs1010816 associated with sorafenib treatment response. There were 6, 33 and 35 persons with the GG, GT and TT genotypes at rs1010816, respectively. Further-

Discussion
In the present study, our goal was to identify pretherapeutic biomarkers capable of predicting the outcomes of sorafenib treatment, so that only individuals with a higher probability of responses and favorable outcomes should be treated with this drug. Through GWAS, it was found that there was a linkage disequilibrium in the SNPs surrounding the rs1010816 associated with sorafenib treatment response. There were 6, 33 and 35 persons with the GG, GT and TT genotypes at rs1010816, respectively. Furthermore, multivariate analysis also showed that the rs1010816 genotype was the only independent predictor for time-to-sorafenib therapeutic response. However, the rs1010816 "TT" genotype was not significantly associated with favorable PFS or OS, as compared with those with rs1010816 "non-TT" genotype. Multivariate analysis revealed that the presence of ascites or sorafenib total dose was significantly associated with PFS. So far, information on the rs1010816 genotype is limited. The SNP rs1010816 was located on chromosome 4 and between the LOC105377364 and RNU6-289P genes. Accordingly, we proposed that the SNP rs1010816 was considered as intergenic SNP. Based on the single-nucleotide polymorphism database at the NCBI, the data showed that the frequency of the rs1010816 TT genotype in a global population (n = 147080) was approximately 74%. The presence of ascites, main portal vein thrombosis, platelet count, total sorafenib doses, PALBI score, or ALBI grade were independent predictors for OS. We were puzzled by this unexpected result. Therefore, a subgroup analysis was subsequently performed, and it was found that the "TT" genotype was significantly associated with longer OS in patients with alcoholism. In other words, the rs1010816 "TT" genotype in alcoholic aHCC patients receiving sorafenib therapies was associated with a favorable sorafenib therapeutic outcome (in both therapeutic response and OS). Previous studies showed that SNPs could modulate several cellular functions through alteration of mRNA secondary structure, modulation of RNA alternative splicing and regulation of target-gene-translation efficiency [26]. We proposed that SNP rs1010816 may regulate the expression of LOC105377364 via potential underlying mechanisms mentioned above, leading to regulated hepatic alcohol metabolism. The functional roles of SNP rs1010816/LOC105377364/sorafenib response axis in HCC need to be determined in the future.
To date, although immunotherapy has been recommended as the first-line treatment for advanced HCC patients with vascular invasion and/or extrahepatic metastasis for some time, sorafenib is still the most widely used standard-of-care first-line treatment for these patients [7,8]. Sorafenib significantly improved OS in aHCC patients but the outcomes varied greatly among different individuals. Several clinical prognostic factors have been proposed. The recently published GIDEON (global investigation of therapeutic decisions in hepatocellular carcinoma and of its treatment with sorafenib) was a largescale global prospective observational registration study that recruited 3202 patients with unresectable HCC treated in real-life practice conditions to evaluate the use and tolerability of sorafenib [27]. It was reported that the type and incidence of drug-related adverse events and those leading to sorafenib discontinuation were similar in patients with Child-Pugh A and B liver function. However, survival was longer in patients with Child-Pugh A liver function (13.6 months) than Child-Pugh B (5.2 months) and Child-Pugh C patients (2.6 months) in the intent-to-treat analysis. Subsequently, among the individual components of the Child-Pugh score, albumin levels, ascites, and bilirubin levels all have important prognostic values for OS. In addition, McNamara et al. [28] have reported a meta-analysis study that included four randomized controlled trials and twenty-six cohort studies with a total of 8678 patients. Multivariable meta-regression analysis revealed that patients with ECOG performance status of 2 (p = 0.04) and Child-Pugh B (p = 0.001) were significantly negatively associated with OS.
However, in our real-world cohort study, although the Child-Pugh score/grade remained significant in univariate analysis (p < 0.001, Table 4), the Child-Pugh stage/grade and ECOG performance status were not significant predictors of OS. Furthermore, a multivariate analysis showed that the presence of ascites, main portal vein thrombosis, lower platelet count, lower total sorafenib doses, and higher PALBI score in model A and higher ALBI grade in model B were significantly associated with a shorter OS, while the Child-Pugh score was not included as an independent predictor. Therefore, the Child-Pugh score was not considered a good predictor for OS in our patients, but the ALBI grade or PALBI grade may appear to be a better marker. Johnson et al. [29] proposed the ALBI grade in 2015, which used a different liver-dysfunction-assessment model to better select HCC patients receiving sorafenib and only involved the serum albumin and bilirubin levels from 1313 patients in Japan initially, and subsequently validated the results in 5097 patients in six cohorts from other geographic regions. In addition, the ALBI grade could eliminate the need for subjective variables such as ascites and encephalopathy, which is the requirement of the standard Child-Pugh grade. Regarding the comparison with the sub-categories according to points [5][6][7][8][9][10][11][12][13][14][15] in the Child-Pugh classification in the original publication, the combination of Child-Pugh A as the inclusion criteria and ALBI as the stratification factor in trials of systemic therapy may allow HCC patients to have more extraordinary discrimination ability when treated with sorafenib [30]. The PALBI grade was proposed by Roayaie based on the ALBI grade in 2015 [31]. By contrast, although the calculation of ALBI or PALBI grade was considered complicated, they avoided the disadvantages of the "ceiling effect" or "floor effect" through arbitrary cut-off values. Moreover, both ALBI and PALBI could ensure accurate assessment of liver function and predict the prognosis in HCC patients [32,33].
Our study has some limitations. First, the predictive value of rs1010816 genotypes in sorafenib therapeutic response or in atezolizumab plus bevacizumab treatment should be evaluated using independent cohorts of patients in the future. Second, this study was a retrospective investigation in a single center. These clinical findings should be verified by prospective studies. Recently, deep learning has become a powerful tool for patient stratification in precision medicine [34]. If we have more clinical information about rs1010816 genotypes on sorafenib therapeutic response, effective deep learning methods can be applied for evaluating the prognosis of aHCC patients based on rs1010816 genotypes.
In this study, we discovered a single-nucleotide polymorphism rs1010816, which could effectively predict the therapeutic response of sorafenib in patients with aHCC. However, it could not predict OS. It was possible that its inability to predict OS was largely due to the heterogeneity of cancer etiologies and patient populations. A reasonable approach to solve this problem was to determine the pre-treatment biomarker in various patient subgroups. Hence, it was found that only in patients with alcoholic aHCC, was the "TT" genotype significantly associated with longer OS during sorafenib treatment.

Patients
This retrospective cohort study was conducted under the approval of the Institutional Review Board of Chang Gung Medical Hospital. From February 2008 to January 2018, a total of 74 consecutive aHCC patients with BCLC stage C were enrolled, who were not suitable for curative therapy and received sorafenib therapy in Linkou Chang Gung Medical Hospital, and their samples were retrievable from a tissue bank. The diagnosis criteria of HCC included dynamic computed tomography or dynamic magnetic resonance imaging plus angiography when the HCC size in cirrhotic liver exceeded 2 cm with typical HCC characteristics. When such typical features were not present, a liver biopsy or aspiration cytology diagnosis was required for diagnosis.
In order to evaluate the therapeutic response of sorafenib, alpha-fetoprotein (AFP) changes and modified response evaluation criteria in solid tumors (mRECIST) were applied as previously described [37][38][39]. Accordingly, a 20% reduction of AFP was also considered as achieving treatment response [38]. OS was calculated from the date of sorafenib treatment to the date of death or the last follow-up.

Genome-Wide Variant Assessment
The Affymetrix Axiom Genome-Wide TWB 2.0 array plates, designed jointly by the Taiwan Biobank, the National Center for Genome Medicine (Nankang, Taipei, Taiwan) and the Thermo Fisher Scientific (Waltham, MA, USA), were used for assessing~750,000 genome-wide variants in this GWAS study. The base-calling algorithm was Affymetrix APT. The hybridization of samples onto the microarrays and the washing procedure were conducted following the manufacturer's protocol. The Affymetrix scanner scanned the fluorescent signals and then stored them as digital images for further data processing. Associations with the phenotype were evaluated using Fisher's exact test of allele counts.

Statistical Analysis
Parametric data were recorded as mean ± standard deviation and compared using a two-sample t-test. Dichotomized data were recorded as numbers and percentages (%) and compared using the Chi-square test or Fisher's exact tests, where appropriate. Univariate followed by multivariate Cox proportional hazard models were performed to estimate the therapeutic response and progression-free survival (PFS) and OS for clinical factors and genotypic variables. In this study, significant factors determined from the univariate analysis were included for multivariate Cox proportional hazards. The Kaplan-Meier method was performed to estimate the survival probability between the different groups, and the logrank test was performed to compare survival. All tests were two-tailed, and a p < 0.05 was considered statistically significant. All statistical analyses were performed using Statistical Package for the Social Sciences (SPSS) statistics Version 20 (SPSS, Chicago, IL, USA).

Conclusions
In summary, we demonstrated that the rs1010816 "TT" genotype could be an effective predictor of favorable therapeutic response and OS in alcoholic aHCC treated by sorafenib. Prospective studies should be conducted to verify this result. Informed Consent Statement: Informed consent was obtained from all subjects involved in the study.

Data Availability Statement:
The data presented in this study are available upon request from the corresponding author.