Are preoperative high soluble programmed death ligand 1 levels responsible for patients’ poor prognoses in hepato-biliary-pancreatic cancer?

Background: Several reports showed that high soluble programmed death-ligand 1(sPD-L1) level was a risk factor for poor prognosis in various tumors. To date, the clinicopathologic and prognostic impact of sPD-L1 level in patients with hepato-biliary-pancreatic cancer have not been determined. Methods: A total of 119 patients (66 patients with hepatocellular carcinoma, 23 patients with cholangiocarcinoma, 30 patients with pancreatic cancer) who were treated at the Toho University Omori Hospital (Tokyo, Japan) from 2008 to 2016 were retrospectively analyzed. sPD-L1 levels were measured using an enzyme-linked immunosorbent assay for PD-L1 to evaluate clinicopathologic and prognostic impact. Results: sPD-L1 levels were signicantly higher in low-albumin group than normal albumin group. According to stages in hepatocellular carcinoma and cholangiocarcinoma, there were no signicant differences in sPD-L1 levels, which gradually increased according to stage in pancreatic cancer. Using a cut-off value of 81.6pg/ml for sPD-L1level, the high sPD-L1 group showed signicantly worse prognosis than the low sPD-L1 group in patients with pancreatic cancer. Multivariate analysis identied sPD-L1 level ≥ 81.6 mg/dl ( p = 0.047) as an independent predictor of poor overall survival in patients with pancreatic cancer. Conclusion: Using a cut-off value of 81.6pg/ml for sPD-L1level, high sPD-L1 levels were independently associated with poor prognosis in patients with pancreatic cancer. However, this association in hepatocellular carcinoma or cholangiocarcinoma was not clear.

Are preoperative high soluble programmed death ligand 1 levels responsible for patients' poor prognoses in hepato-biliary-pancreatic cancer? Background Programmed death-ligand 1 (PD-L1) is an immune checkpoint protein within the cancer-immunity cycle that is expressed on the surface of tumor cells and tumor in ltrating immune cells to downregulate T-cell function [1]. Recently, soluble PD-L1 (sPD-L1) has been detected in the blood of cancer patients [2].
In patients with HCC, high sPD-L1 level and high PD-L1 expression in tumors has been identi ed as a poor prognostic factor [5,11]. In patients with cholangiocarcinoma (CCA), high PD-L1 expression in tumors has been identi ed as a poor prognostic factor [12]. On the other hand, the clinicopathologic signi cance of high sPD-L1 level has not been reported. In patients with pancreatic cancer (PC), high sPD-L1 level was not signi cantly associated with overall survival [13]. To date, clinicopathological and prognostic impact of sPD-L1 levels in patients with hepato-biliary-pancreatic cancer have not been determined [4,5,14]. disease included distant lymph node metastasis (n = 2), organ metastasis (n = 4), and peritoneal metastasis or cancer cells on peritoneal cytology (n = 9). Using the tumor-node-metastasis (TNM) classi cation system of malignant tumors of the Union for International Cancer Control (UICC), 8th edition [15], the nal stage of HBP cancer was assessed pathologically.
This retrospective study was approved by the Institutional Ethics Committee of the Toho University (IRB no. A18103). All patients were followed-up until the end of April 2020 or death.
Sample collection and enzyme-linked immunosorbent assay All serum samples were collected before surgery, and sPD-L1 levels were measured using an enzymelinked immunosorbent assay (ELISA) for PD-L1 (R&D Systems, Inc., Minneapolis, MN, USA) as previously described [3].

Study design and serum biomarker analysis
In order to evaluate the relationship with sPD-L1 level, C-reactive protein (CRP), albumin, and plasma brinogen were analyzed before surgery. White blood cell, neutrophil, and lymphocyte counts were also analyzed. Cut-off values were determined in accordance with the institutional standards for white blood cell count (7000 cells/mm 3 ), CRP levels (0.3 mg/mL), and albumin (3.5 mg/dl). Then, we analyzed the association between sPD-L1 level and clinicopathologic factors with HBP cancer.

Statistical analyses
Statistical analysis was performed using JMP statistical software (version 12; SAS Institute, Cary, NC, USA). Serum biomarker levels were expressed as the mean ± standard deviation. Comparisons between unpaired groups for these variables were conducted with the Mann-Whitney U test. In addition, the survival rate was calculated by the Kaplan-Meier product limit estimate. Differences between groups regarding survival were analyzed by the log-rank test. Multivariate analysis using the Cox proportional hazards model was used to assess signi cant predictors identi ed by univariate analysis. We considered p < 0.05 as statistically signi cant.

Results
Correlation and association of serum programmed death ligand 1 level with serum biomarkers The associations of sPD-L1 levels with several blood tests were evaluated (Table 1). sPD-L1 level was signi cantly higher in the low-albumin group than in the normal albumin group (p = 0.035, Table 1). There were no signi cant differences between sPD-L1 levels and serum biomarkers in the other parameters: white blood cell, neutrophils, lymphocytes, C-reactive protein, total bilirubin, and amylase.  1a). There were no signi cant differences in sPD-L1 levels among HBP cancer (HCC vs CCA p = 0.895, HCC vs PC p = 0.818, CCA vs PC p = 0.686). There were no signi cant differences in sPD-L1 levels according to UICC stages in HCC and CCA (Fig. 1b, 1c). On the other hand, sPD-L1 levels gradually increased according to UICC stage in PC (Fig. 1d). sPD-L1 levels of stage IV PC was signi cantly higher than that of stage I/II/III PC (Fig. 1c, p = 0.039).
Comparisons of soluble programmed death ligand 1 levels according to clinicopathologic factors and various biomarkers with HBP cancer sPD-L1 levels were not associated with any of the clinicopathologic factors, such as gender, age, stage, tumor size, and conventional tumor markers, in patients with HCC and CCA (Table 2a, 2b, 2c). Except, sPD-L1 levels were signi cantly higher in PC patients with high CA19-9 levels (p = 0.028, Table 2c).
Overall survival curves according to soluble programmed death ligand 1 levels in the patients with HBP cancer All 119 cases were divided by quartiles according to sPD-L1 levels as follows: the range of sPD-L1 levels with Q1 was 15.8 pg/ml to 49.3 pg/ml, Q2 was 49.8 pg/ml to 64.3 pg/ml, Q3 was 64.5 pg/ml to 81.5 pg/ml, and Q4 was 81.7 pg/ml to 188.0 pg/ml. Although there were no signi cant differences between each group (p =0.878), the Q4 group showed the worst survival rate (Fig. 2a). Therefore, we decided the cut-off value of 81.6 pg/ml sPD-L1 level to evaluate prognostic signi cance of sPD-L1 in further analyses. Although the Q4 group showed worse prognosis than the other groups (Q1Q2Q3), the difference was not statistically signi cant (p = 0.333, Fig. 2b).
In order to evaluate the prognostic signi cance of sPD-L1 in each cancer type, overall survival (OS) was compared between Q1Q2Q3 group vs Q4 group in each cancer types (Fig. 3).
When each types cancer was divided by quartiles according to each sPD-L1 levels, compared between the Q1Q2Q3 group and the Q4 group in each cancer types, the cutoff value for sPD-L1 was also 81.6 pg/ml in common (Fig. 3a-f). There was no prognostic signi cance in HCC (p = 0.977, hazard ratio=1.017, Fig. 3b) and CCA (p = 0.665, hazard ratio = 0.717, Fig. 3d). On the other hand, Q4 group showed signi cantly worse prognosis than Q1Q2Q3 group in patients with PC (p = 0.005, hazard ratio = 5.059, Fig. 3f).
Univariate and multivariate analysis of risk factors for overall survival with pancreatic cancer Several prognostic factors, including sPD-L1 levels, were evaluated in patients with PC (

Discussion
In the present study, preoperative sPD-L1 levels were analyzed in a total of 119 patients with HCC, CCA, or PC. High sPD-L1 level was associated with low albumin level in all HBP cancer patients and not in the speci c cancer type. In patients with PC, high sPD-L1 levels were associated with high CA19-9 levels but not with tumor stage. Using a cut-off value of 81.6 pg/ml, high sPD-L1 level was an independent risk factor for poor OS of patients with PC.
sPD-L1 levels have been reported to increase in patients with cancer and/or systemic inflammation in HCC, gastric cancer and PC [3][4][5]. Although previous studies showed the signi cant association between sPD-L1 and CRP, there was no signi cant relationship between sPD-L1 level and CRP in our current study. Such discrepancy could be partly explained by the fact that there was big difference between the ratio of stage IV in previous studies and the ratio in our present study: stage IV cases accounted for 14%~85% in previous studies and 12% in our study. In stage IV disease, the tumor microenvironment contains many cells producing in ammatory cytokines and promotion of metastatic disease.
Although previous reports did not show that high sPD-L1 levels had signi cant negative effect on OS in PC, our present data showed that high sPD-L1 level was a signi cant risk factor reducing patients' survival. A discrepancy such as this could be explained by the aforementioned ratio of stage IV disease in previous reports being higher than in our present study. Too much stage IV disease might mask prognostic impact of PD-L1 expression and the effects of checkpoint blockade in PC [16,17]. Even for resectable PC, PD-L1 blockade may be effective for the patients with high PD-L1 expression. In other words, PD-L1 blockade may effectively improve the prognosis of resectable PC with high PD-L1 levels.
Given the small number of patients included in our study, further studies are necessary to confirm this observation.
Although sPD-L1 level was signi cantly associated with poor prognosis [5,11] in previous reports in HCC, our present series did not show such association. Two previous reports used different ELISA kits than the one used in our present analysis. Moreover, the detection limits and the cut-off values were completely different from those in our series. Such differences might partly explain these discrepancies. Regarding the correlation between high sPD-L1 level and high CRP, 2 previous reports showed positive correlations, in HCC [5] and in PC [4]. They suspected that an increased activation of innate immunity is part of the immunosuppressive environment which hampers activity of the adaptive anti-tumor response. We could not con rm such correlations in our present study. The impact of sPD-L1 levels might be independent to the activity of in ammatory cytokines in HCC.
One of the limitations of the present study was retrospective nature and the number of patients in each cancer type was too small to draw any solid conclusions. Secondly, this study was the lack of healthy controls or the patients with in ammatory diseases like pancreatitis and liver in ammation to compare sPD-L1 levels with those of patients with HBP cancer. To conclude and validate this observation without a shadow of doubt, a larger cohort with control group is needed. Using the same ELISA kit that was used in the present study, Chen et al reported that the median serum PD-L1 level in healthy controls was 48.15 pg/ml [18]. The median serum PD-L1 level reported in our present study was slightly higher than this value. Furthermore, no immunohistochemical analysis was performed to evaluate the impact of tissue PD-L1 expression on serum PD-L1 levels. In cancer, the relationship between sPD-L1 and PD-L1 expression in tissues remains unclear. sPD-L1 may be produced by multiple sources via distinct mechanisms from both tumor and immune cells [19,20].

Conclusion
High sPD-L1 levels might be associated with tumor progression in PC. Moreover, high sPD-L1 levels were independently associated with poor patient survival in the PC group. However, these associations in HCC and CCA were limited. The scatter plot of sPD-L1 level for 66 patients with hepatocellular carcinoma, 23 patients with cholangiocarcinoma, and 30 patients with pancreatic cancer(a). Comparison of sPD-L1 levels according to UICC stages in hepatocellular carcinoma(b), cholangiocarcinoma(c) and pancreatic cancer(d).