The Clinicopathological and Prognostic Significance of Nrf2 and Keap1 Expression in Hepatocellular Carcinoma

Nuclear factor E2-related factor2 (Nrf2) activation is associated with both cytoprotective effects and malignant behavior of cancer cells. This study aimed to evaluate the clinicopathological implications of the expression of Nrf2, pNrf2, and its regulator Keap1 in human hepatocellular carcinomas (HCCs). Tissue microarrays consisting of 285 surgically resected HCCs were immunohistochemically stained with pNrf2, Nrf2, Keap1, stemness-related markers (keratin 19 (K19), epithelial cell adhesion molecule (EpCAM)), carbonic anhydrase IX (CAIX), epithelial–mesenchymal transition (EMT)-related markers (ezrin, uPAR, E-cadherin), and p53, and the results were correlated with the clinicopathological features. pNrf2 expression was significantly associated with increased proliferative activity, as well as EpCAM, ezrin, p53, and CAIX expression and E-cadherin loss (p < 0.05, all). Strong cytoplasmic Nrf2 expression was associated with CAIX and ezrin expression (p < 0.05, both). Keap1 was associated with increased proliferative activity, portal vein invasion, EMT-related markers, and p53 expression in CAIX-negative HCCs (p < 0.05, all). Both pNrf2 and cytoplasmic Nrf2 expression were associated with decreased overall survival (p < 0.05, both), and cytoplasmic Nrf2 expression was an independent predictor of decreased overall survival on multivariate analysis (hazard ratio 4.15, p < 0.001). Both pNrf2 and cytoplasmic Nrf2 expression were associated with poor survival and aggressive behavior of HCC. In addition, Keap1 expression was also associated with aggressive HCC behavior in CAIX-negative HCCs, suggesting that Keap1 expression should be interpreted in the context of hypoxia status.


Introduction
Hepatocellular carcinoma (HCC) is the fourth leading cause of cancer-related mortality worldwide [1]. The genomic landscape of HCC was unraveled during the past decade, with the

Discussion
In this study, we examined the expression of Nrf2, pNrf2, and Keap1 in HCCs, and we correlated the findings with the clinicopathological parameters and expression status of stemness, hypoxia, and EMT-related markers. The association between Nrf2 expression and poor prognosis was previously demonstrated in various solid malignancies, including in two meta-analysis studies [18,21]; however, the prognostic value of Nrf2/Keap1 in HCCs was only reported in two studies so far [16,22]. Zhang et al. demonstrated an association between Nrf2 expression in HCCs and decreased survival, poor histological differentiation, increased proliferation, and more frequent metastasis [22]. Although we observed a significant association between strong cytoplasmic Nrf2 expression in HCC with decreased OS and EMT-related marker expression, other clinicopathological factors such as proliferative activity were not significantly influenced by Nrf2 expression status. The different antibodies used may partly explain the differences; Zhang et al. described nucleocytoplasmic Nrf2 expression with predominantly nuclear localization in their study, while Nrf2 expression in the current study was cytoplasmic. As it is the nuclear form of Nrf2 that is recognized to be the active pro-tumorigenic form, we also examined nuclear pNrf2 expression in HCC, and we found that pNrf2-high HCCs were associated with poor prognosis, increased proliferative activity, and more frequent expression of markers associated with stemness, EMT, and hypoxia. The association between pNrf2 expression and decreased survival in HCCs is in line with a previous study by Chen et al., although we did not find the same expected positive correlation between pNrf2 and Nrf2 expression or an inverse correlation between pNrf2 and Keap1 expression that they observed [16]. It should be noted that Chen et al. reported a larger proportion of pNrf2-positive HCCs (50%) which could be explained by the different cut-offs; we evaluated pNrf2 using a quantitative approach and defined "pNrf2-high" HCCs (~25%) as those with strong nuclear labeling exceeding 10%, while the former study adopted a semiquantitative approach for both pNrf2 and Keap1, counting any labeling for pNrf2 and Keap1 (in the nucleus and cytoplasm, respectively) that was more than focal weak positive as "positive" [16]. Nevertheless, despite the differences in interpretation methods, it is interesting that pNrf2 expression was associated with CAIX, EpCAM, K19, and ezrin expression in this study.
A subset of HCCs that are morphologically compatible with HCC but express immunohistochemical markers associated with stemness (e.g., K19, EpCAM, and SAL-like protein 4 (SALL4)) recently received increasing attention, as these HCCs demonstrate more aggressive behavior compared to conventional HCCs, expression of EMT-related markers, poor prognosis, and resistance to locoregional and systemic treatment [23][24][25]. HCCs expressing stemness-related markers were shown to more frequently result in residual tumors after transarterial chemoembolization (TACE), compared to conventional HCCs, and the residual tumors after TACE often expressed stemness-related markers that were co-localized with CAIX expression, thus suggesting a relationship between TACE-induced hypoxia and stemness [26]. In this study, we found positive correlations between pNrf2 expression and hypoxia, stemness, and EMT-related markers, suggesting a possible role for Nrf2 in the link between tumor hypoxia, stemness, and aggressive behavior of HCC. pNrf2 may be upregulated in HCCs in response to oxidative stress, and pNrf2 may further enhance the aggressive behavior of hypoxic HCCs; however, functional studies would be necessary for validation.
Although Keap1 expression status was not correlated with clinicopathological factors or patient survival, Keap1-high HCCs more frequently demonstrated p53 expression and higher proliferative activity. In addition, a positive correlation was found between Keap1 and pNrf2 expression status. These results are somewhat counterintuitive, as Keap1 was described to be a protein that suppresses Nrf2 activation. Interestingly, when we examined the transcriptomics data provided by The Human Protein Atlas (https://www.proteinatlas.org/) and UALCAN (http://ualcan.path.uab.edu/), which are based on The Cancer Genome Atlas (TCGA) data, we found that KEAP1 messenger RNA (mRNA) expression was in fact higher in HCCs than in non-neoplastic liver tissues, and high KEAP1 mRNA was associated with poor prognosis in HCCs ( Figure S2) [27,28]. An association between Keap1 expression and poor prognosis was also demonstrated in endometrial cancers, and the authors postulated that the overexpression of Keap1 may reflect Keap1 induction in the setting of oxidative stress [15]. On the other hand, when we analyzed the Keap1 expression status separately in CAIX-positive and CAIX-negative HCCs, we interestingly found that Keap1 expression was associated with more frequent portal vein invasion, higher proliferative activity, frequent p53 expression, and a tendency for decreased DFS in CAIX-negative HCCs, while Keap1 was not associated with any of the clinicopathological or immunohistochemical factors in the CAIX-positive HCCs. Instead, Nrf2 and/or pNrf2 were the determinants of aggressive behavior and poor survival in the setting of CAIX-positive HCCs. Thus, it is possible that, under hypoxic stress, Nrf2 is released from Keap1 to contribute to the aggressive behavior of HCC, while, under normoxic conditions, Keap1 is also a determinant of aggressive behavior. In addition, the association between strong cytoplasmic Nrf2 expression and decreased OS in this study suggests that, although the nuclear pNrf2 is the active form of Nrf2, strong overexpression of Nrf2 in the cytoplasm is also abnormal and has prognostic implications.
There are a few limitations to this study. Firstly, although we analyzed a large number of HCCs, this is a retrospective cohort study using formalin-fixed paraffin-embedded tissue samples and the results are based on immunohistochemical results. As all the HCCs enrolled in this study were surgically resected HCCs, it is possible that early well-differentiated HCCs-which are often treated by loco-regional modalities-were less likely to be included in this study, resulting in a biased cohort predominantly consisting of moderately to poorly differentiated HCCs. Functional studies would need to be performed to further evaluate the functional relationships among Nrf2/Keap1, hypoxia, and EMT. Secondly, as the majority of the HCCs in this study have an HBV-related etiology, it would be necessary to perform a similar analysis on an independent cohort of HCCs from regions where HBV is not as predominant.

Patients
A total of 285 cases of primary HCCs that were surgically resected between 2003 and 2012 at Seoul National University Bundang Hospital (Seongnam, Korea) were evaluated in this study. Clinicopathological information was obtained from the patients' electronic medical records, pathology database, and a review of the archived glass slides, and it included sex, age at operation, underlying etiology (including hepatitis B virus (HBV), hepatitis C virus (HCV), alcohol and non-alcoholic fatty liver disease (NAFLD)), tumor size, multiplicity, tumor differentiation (Edmonson-Steiner grade), presence of microvascular and portal vein invasion, mitotic index, extrahepatic metastasis, and pathological T stages according to the American Joint Committee on Cancer TNM staging system (eighth edition).
The mitotic index was expressed as the total number of mitotic figures in 10 high-power fields (HPFs).
The clinical follow-up information, including the dates of surgery, disease recurrence, death, and most recent follow-up, was retrieved from the electronic medical records. Overall survival (OS) was defined as the number of months between surgery and patient death, and disease-free survival (DFS) was defined as the number of months between surgery and disease recurrence (local recurrence and/or distant metastasis). In addition, a separate independent tissue microarray cohort of 293 resected HCCs from Seoul National University Hospital was evaluated to further examine the prognostic significance of pNrf2 expression. This study was approved by the Institutional Review Board of Seoul National University Hospital and Seoul National University Bundang Hospital (joint IRB# H-1808-073-965), and patient consent was waived due to the retrospective nature of this study.

Tissue Samples and Immunohistochemistry
Tissue microarrays with tissue cores measuring 2 mm in diameter were constructed from formalin-fixed paraffin-embedded HCC tissue, and they were arranged in recipient tissue microarray blocks using a trephine apparatus (Superbiochips Laboratories, Seoul, Korea). Immunohistochemistry was performed on 4-µm-thick tissue sections using antibodies for pNrf2, Nrf2, Keap1, K19, EpCAM, E-cadherin, CAIX, p53, and Ki-67, using an automated platform (Ventana BenchMark GX, Ventana Medical Systems, Oro Valley, AZ, USA). Immunohistochemical staining for ezrin and uPAR was performed manually, with antigen retrieval using the citrate buffer (pH 6.0) and the Envision Kit (Dako, Glostrup, Denmark). Detailed information on the antibodies is summarized in Table 4. For Nrf2 and Keap1, the intensity of cytoplasmic expression in the tumor cells was recorded as 0 (negative), 1+ (weak), 2+ (moderate), and 3+ (strong). Nrf2 expression was cytoplasmic, and strong positive (3+) staining was designated as "Nrf2-high" (overexpression). "Keap1-high" was defined as moderate to strong cytoplasmic staining in the tumor cells, and the staining intensity was similar to or more intense than that seen in non-neoplastic hepatocytes, cholangiocytes, and stromal cells. pNrf2 staining was evaluated in a quantitative manner, using the ImageJ software (National Institutes of Health, Bethesda, MD, USA); the percentage of nuclear pNrf2 labeling was calculated by dividing the number of positively stained tumor cell nuclei by the total number of tumor cell nuclei, in an HPF (400× magnification). "pNrf2-high" was defined as pNrf2 labeling in ≥10% of tumor cells. K19, EpCAM, CAIX, ezrin, and uPAR expression was designated positive when there was cytoplasmic (K19, CAIX, ezrin) or membranous (EpCAM and uPAR) expression in ≥5% of the tumor cells. Loss of E-cadherin expression was defined as complete absence of membranous E-cadherin expression in the tumor cells. p53 expression was defined as positive when strong nuclear labeling was present in more than 10% HCC cells. Two pathologists (K.L. and H.K.) independently evaluated the immunohistochemical stains and reached an agreement using a multihead microscope when there were discrepancies in the interpretation results.

Statistical Analysis
All data were analyzed using IBM SPSS statistics software version 25.0 (SPSS Inc., Chicago, IL, USA) and GraphPad Prism software version 7.0 (San Diego, CA, USA), and a p-value <0.05 was considered statistically significant. The associations between the immunohistochemical stain results and clinicopathological parameters were assessed using the χ 2 test, Fisher exact test, and Student's t-test, as deemed appropriate. OS and DFS were analyzed using the Kaplan-Meier method and the log-rank test. Multivariate analysis using the Cox regression model was performed for clinicopathological and immunohistochemical parameters with p < 0.05 on univariate analysis.

Conclusions
In summary, we demonstrate that pNrf2 expression in HCC was associated with increased proliferative activity, decreased survival, and more frequent expression of hypoxia, stemness, and EMT-related markers, suggesting a possible role for pNrf2 as a marker of aggressive behavior and poor prognosis in HCC. In addition, contrary to previous knowledge, strong cytoplasmic Nrf2 expression was also an independent predictor of decreased survival in HCC, and Keap1 expression was associated with aggressive HCC behavior in CAIX-negative tumors, suggesting that Keap1 expression in HCCs should be interpreted in the context of the metabolic status. Further prognostic studies in independent cohorts and functional studies would be required to validate the current findings.