A genome‐wide association study identifying SVEP1 variant as a predictor of response to tolvaptan for cirrhotic ascites

Tolvaptan, vasopressin V2‐receptor antagonist, has been used for patients with difficult‐to‐treat ascites in Japan. In this study, we conducted a genome‐wide association study (GWAS) in the Japanese population to identify genetic variants associated with tolvaptan's efficacy for patients with hepatic ascites.


| INTRODUC TI ON
Hepatic ascites, the accumulation of fluid in the abdominal cavity due to liver disease, is a difficult condition to treat. In European guidelines, high-dose diuretics (160 mg furosemide, spironolactone 400 mg) are used for uncontrolled hepatic ascites. 1 However, Japanese patients are usually intolerant to high-dose conventional diuretics (furosemide and spironolactone) because of dehydration or hyponatremia. Since December 2013, tolvaptan, an orally active vasopressin V2-receptor antagonist, has been used in Japan for patients with ascites that are difficult to treat with conventional diuretics. 2 Tolvaptan suppresses the expression of aquaporin (AQP)-2 and inhibits water reabsorption in the renal collecting ducts. Tolvaptan does not stimulate sodium channels, unlike other diuretics, and increases free water excretion without affecting urinary sodium levels. 3 Japanese guidelines recommend administering 20-80mg furosemide and/or 25-100 mg spironolactone 4 for mild ascites. When ascites is uncontrolled by conventional diuretics, tolvaptan is administered at a dose of 3.75 or 7.5 mg. Tolvaptan was first reported to increase serum sodium levels safely and effectively in patients with euvolemic and hypervolemic hyponatremia in a study that explored ascending doses of tolvaptan (SALT-1 and SALT-2 studies) in 2006. 5 Other studies have also demonstrated the efficacy and safety of tolvaptan. 6 In a Japanese multicenter retrospective study, responders to tolvaptan were defined as patients who had body weight (BW) loss of 1.5 kg/week which reflect the improvement of ascites volume and symptoms. 7 Since this paper was published, BW loss of 1.5 kg/week criterion has been applied to determine the efficacy of tolvaptan. About two-thirds of patients showed an extreme increase in urine volume and/or decrease in BW. Conversely, roughly one-third of patients showed no increase in urine volume and/or decrease in BW. Several studies showed that low serum sodium, 8 high blood urine nitrogen (BUN), 9 or high urine osmolality 10 can worsen the treatment response, and that creatine (Cre) does not necessarily influence the response to tolvaptan.
The genome-wide association study (GWAS) method has been used to predict the treatment response, such in the analysis of IL28B variants strongly associated with the response to pegylatedinterferon (IFN) plus ribavirin (RBV) therapy for chronic hepatitis C patients. 11 We hypothesized that the response to tolvaptan differs based on clinical characteristics and host genetics. The reason for the difference in the effect of tolvaptan with respect to clinical characteristics has been discussed previously. 6,7 In this study, to determine whether to extend administration to tolvaptan non-responders in the hope that they will eventually respond, we conducted a GWAS in the Japanese population to identify genetic variants associated with tolvaptan's efficacy in patients with difficult-to-treat ascites, and to identify non-responders to tolvaptan. were included in the GWAS and replication study.
Results: Genome-wide association study showed 5 candidate SNPs around the miR818, KIAA1109, and SVEP1 genes. After validation and performing a replication study, an SNP (rs2991364) located in the SVEP1 gene was found to have a significant genome-wide association (OR = 3.55, P = 2.01 × 10 -8 ). Multivariate analyses showed that serum sodium (Na), blood urea nitrogen (BUN) and SVEP1 SNP were significantly associated with the response (OR = 0.92, P = .003; OR = 1.02, P = .02 and OR = 3.98, P = .000008, respectively). Based on a prediction model of logistic regression analysis in a population with the rs2991364 risk allele, the failure probability (=exp (score: 22.234 + BUN*0.077 + Na*-0.179) (1 + exp (score)) was determined for the detection of non-responders. Assuming a cutoff of failure probability at 38.6%, sensitivity was 84.4%, specificity was 70% and AUC was 0.774.
Conclusion: SVEP1 rs2991364 was identified as the specific SNP for the tolvaptan response. The prediction score (>38.6%) can identify tolvaptan non-responders and help to avoid a lengthy period of futile treatment.

K E Y W O R D S
blood urine nitrogen (BUN), genome-wide association study (GWAS), hepatic ascites, liver cirrhosis, non-responder, Polydom 2 | MATERIAL AND ME THODS

| Patients
This study was conducted nationwide in Japan from 2014 through 2018. The protocol was registered to the clinical trials registry managed by the University Hospital Medical Information Network in Japan (registration no. UMIN000025905). Genomic DNA samples were obtained before or after tolvaptan administration from 550 patients who had been treated with tolvaptan add on to conventional diuretics (20-80 mg/day furosemide and/ or 25-100 mg/day spironolactone) for hepatic ascites at each of the participating hospitals (40 hospital liver units with hepatologists). As per previous reports, 7 responders to tolvaptan were defined as those who had a greater than 1.5-kg decrease in BW after 1-week tolvaptan treatment; and non-responders, as those with an increase in BW after 1-week tolvaptan treatment. A total of 80 cases (non-responders) and 333 controls (responders) were included in the GWAS and replication study, and 137 borderline patients whose BW change after 1-week tolvaptan treatment was 0 kg to −1.49 kg, were excluded from the study. To identify genetic variants associated with the response to tolvaptan in patients with cirrhotic ascites, we performed a cohort GWAS recruiting 181 patients (25 cases and 156 controls) from 2014 to 2016 for SNP candidates with P < 10 -4 . A replication study recruiting 232 patients (55 cases and 177 controls) from 2017 to 2018 was also performed.

| Statistical analysis
To assess risk factors associated with tolvaptan response, univariate and multivariate logistic regression was performed. Factors with P < .1 in univariate analysis were included in multivariate analysis with backward elimination (P < .05). Receiver operating characteristics (ROC) analysis was used to assess the prediction ability of the model. All analyses were carried out using SPSS for Windows version 25.0 (IBM, Armonk, NY, USA). The prediction model for the risk allele group was constructed based on a logistic regression model in which backward elimination (P < .05) was applied after including age, sex, height, weight, Child-Pugh classification, presence of HCC, platelet, PT activity, Alb, AST, ALT, T-bil, BUN, Cre, Na, NH3 and aetiology (HBV, HCV, alcohol, NASH, AIH, PBC and others). For these factors, missing rate was less than five percent and frequency in categorical factors was less biased (proportion was more than 10% even in the lesser category). The prediction score was then calculated using coefficients of selected factors based on the definition of the logistic regression.
In the GWAS (including genome-wide imputation data) and replication study, the chi-square test was applied to a 2-by-2 contingency table in the allele frequency model. The odds ratio (OR) and the confidence interval (CI) were calculated using the major alleles as references. We considered P < 5 × 10 -8 as the threshold for genome-wide significance in the combined analysis.
In the replication stage, we selected 134 SNPs with P values <10 -5 and linkage disequilibrium (LD) <0.9 from the results of the chi-square test in the GWAS using genome-wide imputed data. We additionally selected 49 SNPs located on the functionally interested genes. DigiTag2 assay and TaqMan SNP genotyping assay (Applied Biosystems, Foster City, CA, USA) were used to confirm the genotypes at each SNP. We genotyped 80 cases and 333 controls to validate the GWAS results and for the replication study.  and Na (P = .03).

| Genetic variants associated with response to tolvaptan
We conducted a GWAS using 25 cases (non-responders) and 156 controls (responders) by analysing 411,709 autosomal SNPs.

| Imputation-based GWAS and replication study
We performed genome-wide imputation-based GWAS in order to find additional candidate SNPs associated with the response to tolvaptan ( Figure S2). There were 2,127 SNPs with P values <10 SNPs on AXIOM ASI1 array, and 49 SNPs located on the functionally interested genes. The original GWAS set of 181 samples (25 cases and 156 controls) and an independent set of patients (55 cases and 177 controls) were genotyped and used in a subsequent replication analysis. Of the candidate SNPs, three (rs2991364, rs9299186 and rs4978937 on SVEP1 intron region) were validated and consistent associations were observed between the GWAS set and replication set ( Table 2). One SNP showed a genome-wide significant association (rs2991364: OR = 3.55, P = 2.01 × 10 -8 ) using the combined set (80 cases and 333 controls) ( Table 2).

| Risk factors for response to tolvaptan
The results obtained by multivariate logistic regression analysis are shown in Table 3. Univariate analysis showed that AFP, BUN, Cre, Na and SVEP1 SNPs (rs2991364) were significantly associated with the response to tolvaptan, and multivariate analysis showed that BUN, Na and SVEP1 SNPs were also associated with the response to tolvaptan (OR = 1.02, P = .02, OR = 0.92, P = .003 and OR = 3.98, P < .00001 respectively). Table S1 shows that the highest sum of sensitivity and specificity of Na was 0.44 and 0.73 respectively.  combination of Na, BUN and SVEP1 SNP was 0.69, and 95%CI was 0.61-0.77. The difference in AUC between Na and the combination of Na, BUN and SVEP1 SNP was significant (P < .01). As well, the difference in AUC between SVEP1 SNP alone and the combination of Na, BUN and SVEP1 SNP was also significant (P < .05) However, there were no difference in AUC between BUN alone and the combination of Na, BUN and SVEP1 SNP (P = .12) (Supplement Figure S3).
These results suggest that the combination of Na, BUN and SVEP1 SNP could predict the response to tolvaptan more accurately and identify non-responders than a single entry.

| Constructing a prediction model for detecting non-responders
In the non-risk allele group, 86.0% of the patients achieved a treatment response, which is sufficient to consider the risk to be low. The percentage of unsuccessful treatment cases was around 40% in the risk allele group (red area), and the number of successful cases was as high as 60% ( Figure 1A). Therefore, it is important to conduct a detailed risk assessment in the risk allele group. A logistic model, using the variable reduction method, structured the prediction model based on BUN and TA B L E 2 Three SNPs associated with tolvaptan's efficacy

TA B L E 3 Univariate and multivariate analysis for tolvaptan response
Na, and the predictive score (score) = 22.234 + BUN*0.077 + Na * −0.179 was calculated. The probability score (failure probability =exp (score)/ (1 + exp (score)) was identified for detecting non-responders ( Figure 1B). Assuming a cutoff of failure probability of 38.6%, sensitivity was 84.4%, specificity was 70%, and the AUC was 0.774 ( Figure 1C). When the failure probability was more than 38.6%, the treatment failure rate was 64.3%, while the treatment response was observed in 87.5% of the patients with a failure probability ≤38.6% (Figure 2). There was a statistically significant difference between the groups (P < .001).

| Functional analysis of SNPs rs2991364 in SVEP1 gene
To determine the effect of the rs2991364 genotype on SVEP1 mRNA expression levels, we evaluated the expression quantitative trait loci (eats) of rs2991364 using the Genotype-Tissue Expression (GITEX Release V8) project dataset via the website gtexportal.org. 12 In the GITEX data, rs2991364 minor CT/TT-allele variants led to lower SVEP1 gene expression in brain tissue [brain -cerebellar hemisphere: P = 1.40 × 10 -6 , brain -cerebellar: P = 2.00 × 10 -6 ].
We next used a public database to determine whether SNP rs2991364 influences the expression of SVEP1 downstream signal genes. We analysed WGS and RNA-seq data from hepatitis C virus-associated cancer patients in ICGC -RIKEN, Japan (LIRI-JP) cohorts. 13 After searching for SNP rs2991364 in germline data from HCC patients, we selected SNPs rs2991364 major TT-allele patients These results indicate that SNPs rs2991364 can influence the expression of SVEP1 and its downstream signal genes, resulting in vascular network fragility ( Figure 3D).

| D ISCUSS I ON
In this study, we found SVEP1 SNPs correlated most closely with the response to tolvaptan. This is the first report showing a strong corre- and plays a critical role in epidermal differentiation. 14 Previously known as Polydom, SVEP1 is also a high-affinity ligand for integrin a9b1. Polydom knockout mice show severe oedema and die immediately after birth as a result of respiratory failure due to dysfunction of fluid drainage; these mice also fail to undergo remodelling and formation of collecting lymphatic vessels. 14 SVEP1 is expressed in the heart, lung, skeletal tissue, placenta, stomach, intestine, stromal osteogenic tissues and so on. A recent report showed that missense variants of SVEP1 were significantly related to coronary arterial disease. 16 With respect to clinical data, BUN and Na were identified as a response marker based on multivariate analysis ( We propose the following strategy for the use of tolvaptan in clinical practice ( Figure S4).  Figure 4). A significant association between coronary artery disease and missense variants in the SVEP1 gene was reported based on a large-scale exome-wide association study. 16 To elucidate the relationship between the tolvaptan response and rs2991364 allele in cardiac oedema patients, further studies are required.
This study had several limitations. First, a limited number of SVEP1 SNP variants were studied. Second, the number of nonresponders was relatively small and difficult to draw conclusions, as we excluded patients with a borderline response (BW change −1.5 < to <0 kg) to tolvaptan. Third, it was not possible to evaluate urine osmolarity or urine AQP-2. Fourth, portal hypertension (PH) would influence ascites or the response to treatment of ascites, as well as serum albumin. Unfortunately, we did not collect data on the existence of PH and the portal vein invasion/thrombosis in this study.
Lastly, all enrolled patients in this study were Japanese and further studies are needed to confirm the data in other populations.
In conclusion, we identified an association between SVEP1 SNPs and the response to tolvaptan among patients with difficult-to-treat hepatic ascites in the Japanese population. The combination of Na, BUN and SVEP1 SNP was predictive of the response to tolvaptan, and the use of the predictive score can further help to identify nonresponders and avoid prolonged use of tolvaptan in patients who will not ultimately benefit.

CO N S ENT S TATEM ENT
The study protocol was conducted in accordance with the ethical guidelines of the Declaration of Helsinki and was approved by each of the participating Institutional Ethics Review Committees for our human genome projects. Written informed consent was obtained from all individual participants.