Increased expression of stemness markers and altered tumor stroma in hepatocellular carcinoma under TACE-induced hypoxia: A biopsy and resection matched study

Background Hepatocellular carcinomas (HCCs) expressing stemness markers are characterized by an aggressive behavior, which might be promoted by an altered tumor stroma. Transarterial chemoembolization (TACE) induces severe hypoxia, and its effect on stemness and tumor stroma of HCCs remains unclear. The purpose of this study was to evaluate the sequential changes of stemness and tumor stroma under TACE-induced hypoxia using biopsy and resection-matched HCCs. Methods Forty-six biopsy and resection matched HCCs including 10 cases with and 36 cases without preoperative TACE were selected. Immunohistochemistry for stemness (keratin 19 [K19], epithelial cell adhesion molecule [EpCAM], and CD133), hypoxia (carbonic anhydrase IX [CAIX] and vascular endothelial growth factor [VEGF]), and tumor stromal (α-smooth muscle actin [α-SMA] and fibroblast activation protein [FAP]) markers were performed and compared in matched biopsied and resected HCCs with and without TACE. Results The accuracy of K19, EpCAM, CD133, CAIX, VEGF, α-SMA and FAP detected on biopsied HCCs was 64% ∼ 86%, using the expression status in resected HCCs as a reference standard in non-TACE group. The sequential change of hypoxia, stemness and stromal marker expression in matched biopsied and resected HCC was greater in TACE group than in non-TACE group (P < 0.05 for all). The degree of stemness marker expression was well correlated with those of tumor stromal markers, and the degree of CAIX expression was well correlated with that of K19 (P < 0.05). Conclusions Stemness marker expression is considered to be increased along with tumor stromal alteration under TACE-induced hypoxia, which might promote the aggressive biology of HCC.


INTRODUCTION
Hepatocellular carcinoma (HCC) is the seventh most common malignancy worldwide, and the third greatest cause of cancer related mortality, especially in Asia and sub-Saharan Africa [1,2]. Transarterial chemoembolization (TACE) is a popular loco-regional therapy in downstaging or bridging to make curative treatments (e.g. resection, transplantation) of HCC possible [3]. Although TACE induces marked tumor necrosis by obstructing tumor-feeding arteries with chemotherapy emulsioned with lipiodol and embolic agents [4,5], a significant number (27 ~ 72%) of HCCs show residual viable tumor after TACE [6].
The biological behavior of tumors is reportedly determined by not only tumoral epithelial cells themselves but also by the tumor stroma, which is a complicated system composed of extracellular matrix proteins, proteolytic enzymes, blood vessels, and a variety of cellular components, including cancer-associated fibroblasts (CAFs) and immune cells, etc [18][19][20][21]. CAFs, histologically categorized as myofibroblasts or activated fibroblasts, were also reported to be associated with aggressive biological behavior, poor prognosis, and resistance to chemotherapy and radiation therapy in several tumors including HCC [22][23][24][25]. HCCs usually contain no or only a little amount of fibrous stroma; nevertheless, we previously reported that HCCs with stemness markers showed tumor stroma [9]. Interestingly, explanted HCCs after TACE were reported to be more fibrotic than those without [16].
Therefore, TACE-induced hypoxia might modulate stemness and tumor stroma of HCC, which are associated with poor clinical outcomes. However, the sequential changes in the expression of stemness markers and tumor stroma of HCCs after TACE remain unclear. In this study, we first checked whether the immunoprofiles of biopsied HCCs, which contain only small portions of tumor, can represent those of whole tumor, by comparing the immunoprofiles of stemness (K19, EpCAM, and CD133), hypoxia (carbonic anhydrase IX [CAIX] and vascular endothelial growth factor [VEGF]), and tumor stromal markers (α-smooth muscle actin [α-SMA] and fibroblast activation protein [FAP]) between matched biopsied and resected HCCs in non-TACE group. Then the sequential changes in the expression status of stemness, hypoxia, and tumor stromal markers were evaluated by comparing the change in expression status of these markers in matched preoperative biopsied and resected HCCs, between HCCs with preoperative TACE and those without.

Comparison of the immunoprofiles between matched preoperative biopsies and resected HCCs
To investigate whether the immunoprofiles of biopsied HCC tissue may represent those of the whole tumor in resected HCCs, we compared the expression status of stemness-, hypoxia-and stromal-markers between matched preoperative baseline biopsies and the resected HCCs in the non-TACE group (n = 36) ( Figure 1). Only the non-TACE group was analyzed in order to exclude the possible effects of preoperative TACE on the immunoprofiles of resected HCCs. No significant differences were found in the degrees of K19, EpCAM, CD133, CAIX, VEGF, α-SMA and FAP expression between biopsied and resected HCC tissues ( Figure 1B).
In addition, the diagnostic performance of immunohistochemical result based on biopsied HCCs was evaluated using the immunoprofiles of resected HCCs as a standard of reference, and the expression status of each marker was checked as binary classification (i.e. negative and positive). The accuracy of immunohistochemical stains on biopsies was 86%, 78% and 78% for K19, EpCAM, and CD133, respectively, and 72%, 81%, 72% and 64% for CAIX, VEGF, α-SMA and FAP, respectively (Table 1). Taken together, the immunoprofiles of biopsied HCCs may represent those of resected HCCs.
The sequential change in the expression of stemness, hypoxia and stromal markers after TACE No significant differences were seen in the expression status of K19, EpCAM, CD133, CAIX, VEGF, α-SMA and FAP in the preoperative baseline biopsied HCCs between the TACE and non-TACE groups. In contrast, for resected HCCs, the incidence of EpCAM and α-SMA expression was significantly higher in HCCs with preoperative TACE than in those without (P < 0.05 for both). The clinicopathological features of resected HCCs with and without preoperative TACE are summarized in Table 2. The resected HCCs with preoperative TACE showed younger age, poorer differentiation of HCC, higher incidence and greater extents of tumor necrosis, compared to those without preoperative TACE (P < 0.05 for all). Other clinicopathological features of resected HCCs, including TNM stage showed no significant difference between the two groups.
The sequential change in the degree of immunoexpression between biopsied and resected HCCs was evaluated and compared between HCCs with preoperative TACE and those without. The sum of score change in three stemness markers (K19, EpCAM, and CD133) showed a significant difference between HCCs with preoperative TACE and those without (P = 0.031), indicating an increased expression of stemness markers after TACE (Figure 2A). The sum of score change in hypoxia markers (CAIX and VEGF) showed a significant difference between HCCs with preoperative TACE and those without (P = 0.046), indicating an increased hypoxic tumor microenvironment after TACE ( Figure 2B). Similarly, the sum of score change in stromal markers (α-SMA and FAP) showed a significant difference between HCCs with preoperative TACE and those without (P = 0.047), suggesting an altered tumor stroma after TACE ( Figure 2C).
The relationship of stemness marker expression with those for tumor stroma and hypoxia was evaluated in all resected HCCs including TACE and non-TACE groups (n = 46) ( Figure 2D). The degree of K19 and EpCAM expression was well correlated with tumor stromal FAP expression (P = 0.042 and P = 0.004, respectively), and the degree of CD133 expression was well correlated with tumor stromal α-SMA expression (P = 0.005). The degree of CAIX expression was well correlated with that of K19 expression (P = 0.007), but not for those of EpCAM and CD133 (P > 0.05 for both). The degree of VEGF expression was not correlated with any stemness marker of K19, EpCAM and CD133 (P > 0.05 for all).

The prognostic significance of stemness, hypoxia and stromal marker expression in biopsied and resected HCCs in non-TACE group
The prognostic significance of clinicopathologic features, and expression of stemness, hypoxia and stromal markers was evaluated in non-TACE group to exclude the effect of TACE on HCCs. In the preoperative biopsied HCCs of the non-TACE group, univariate analysis of immunoprofiles performed on biopsied HCCs revealed that expression of K19 (P = 0.011), CAIX (P = 0.023), VEGF (P = 0.039) and α-SMA (P = 0.045) were adverse prognostic factors for overall survival (Table 3, Figure 3A). Other immunomarkers (including EpCAM, CD133 and FAP), and other preoperatively collected clinical data (including age, gender, etiology and serum markers) showed no significant impact on overall survival.
The pathological features of resected non-TACE group HCCs were also evaluated for the overall survival, and the expression of K19, CAIX and FAP, and microvascular invasion, and TNM stage were significant for poor overall survival (P < 0.05 for all). (Table 3, Figure 3B). Multivariate analysis for overall survival was performed using five postoperatively collected factors (K19, CAIX, and FAP expression, microvascular invasion and TNM stage). CAIX and FAP expression showed statistical significance (P = 0.001 and P = 0.022, respectively) for overall survival (Table 4).

DISCUSSION
Hypoxia can modulate tumor biology by activation of hypoxia adaptation pathways, especially hypoxia inducible factor-1 (HIF-1) [3,12,26]. CAIX, a direct transcriptional target of HIF-1α, is considered an endogenous marker for the transcriptional activity of HIF-1α [27,28], and can be more easily detected by immunohistochemical stain than HIF-1α, which has a short half-life (<5 min) [29]. VEGF expression is also induced by hypoxia response genes including HIF-1α, and promotes neovascularization [30]. Recently, hypoxia has been reported to be important in reprogramming to cancer stem cell phenotype and maintenance of cancer stem cells in several cancer types, including HCC [12-15, 31, 32], and increased expression of stemness markers was reported in resected/explanted HCCs after TACE treatment [16,17]. We also reported that the expression of K19, EpCAM and CAIX was significantly higher in residual viable HCCs with preoperative TACE compared to those without, and that K19, EpCAM, and CAIX were more frequently expressed in HCCs with a greater number of TACE sessions, suggesting that evaluation of these markers in biopsied HCC tissue might have an additional value in predicting HCC outcome, especially for TACE-treated cases [33].
To address the question of whether the immunoprofiles of small needle biopsy tissues are representative of the whole tumor, we compared the expression status of stemness-, hypoxia-and tumor stroma-related markers in matched preoperative biopsied and resected HCC tissues without TACE. Using the immunoprofiles of the resected HCCs as a standard of reference, the accuracies of marker expression ranged from 64% to 86%, suggesting that the immunoprofiles of biopsied HCCs could be considered to be representative of whole HCCs. A gene expression analysis study on different areas of the same tumor demonstrated little differences in the gene expression profiles according to the location [34]. In addition, the expression of K19 and CAIX in biopsied HCC tissues as well as their expression in resected HCCs was significant for poor overall survival in this study.  In addition, it remains unclear whether the extent of K19, EpCAM and CAIX expression is constant or changed under TACE-induced hypoxia. To address this question, the sequential change in the degree of immunoexpression between biopsied and resected HCCs was evaluated and compared between HCCs of the TACE and non-TACE groups. Although there was extensive tumor necrosis (up to 95%, median 80%) in HCCs of TACE group, the viable tumor area ranged from 0.8 cm 2 to 27.0 cm 2 (median, 2.6 cm 2 ). Therefore, the immunoprofiles of the viable tumor in TACE group were considered to be representative. While there was no significant difference in stemness-related marker expression status in the baseline preoperative biopsies of both groups, the sum of score change for stemness markers was significantly higher in the TACE group HCCs compared to the non-TACE group. Therefore, it is suggested that stemness marker expression may be increased in HCCs under TACE-induced hypoxic conditions. Taken together, we discerned that HCC tumor cells expressing stemness markers, which are considered as CSCs, may have a survival advantage in hypoxic tumor environment induced by TACE, and that these tumor cells may proliferate under TACE-induced hypoxia, as evidenced by the increased tumor extent of stemnessrelated marker expression in this study. In addition, the degree of K19 expression was correlated with that of CAIX expression in resected HCCs. Accordingly, CAIX was reported to increase extracellular acidity and result in metabolic reprogramming, and maintenance of CSCs in breast cancer and prostate cancer [14,35].
We found that the sequential change in sum of hypoxia marker (CAIX and VEGF) expression in matched biopsied and resected HCC was greater in TACE group than in non-TACE group, suggesting that hypoxic condition increased after TACE. However, there were no significant differences in CAIX and VEGF expression between resected HCCs of TACE and non-TACE group. It could be speculated that large sized HCCs in the TACE group had already adapted to the hypoxic condition caused by the imbalance between rapid growth and blood supply before TACE. Accordingly, CD34-positive/ VEGF-negative HCCs have been reported to be resistant to TACE-induced hypoxia, as they had already developed a sufficient vascular network without requiring further neoangiogenesis [36]. stromal markers (α-SMA and FAP) checked in matched biopsied and resected HCC are compared between HCCs with preoperative TACE and those without TACE. The red bar represents that the expression of markers is higher in resected HCCs compared biopsied HCCs, in contrast that the blue bar represents decreased expression in resected HCCs than in biopsied HCCs. Deeper color indicates greater change, and the grey bar represents no change. (D) The correlation between expression of stemness markers (K19, EpCAM, and CD133) and those of stromal markers (α-SMA and FAP) and hypoxia markers (CAIX and VEGF). The black, grey, and white bars represent high, low, and no expression, respectively. * P < 0.05 Abundant intratumoral fibrous stroma is not a typical feature of HCC; we previously reported that HCCs with this feature exhibited high expression of stemness markers with upregulated transforming growth factor-β (TGF-β) signaling and epithelial-mesenchymal transition (EMT) regulators [37]. Recently, tumor stromal cells were reported to promote cancer cells to gain CSC properties through production of IL-6 or TGF-β1-induced EMT [38,39]. FAP was shown to increase stromal cell proliferation and invasiveness, to reduce apoptosis, and to be associated with worse prognosis in colon cancer and pancreatic cancer [40,41]. In this study, the expression score difference of stromal markers (α-SMA and FAP) between preoperative biopsied and resected HCCs was significantly higher in TACE group HCCs compared with the non-TACE group. α-SMA expression was also higher in resected TACE group HCCs compared to non-TACE group HCCs, while there was no significant difference in α-SMA expression status in the baseline preoperative biopsied HCCs between two groups. Taken together, TACE-induced hypoxia is considered to  alter tumor stroma, which may in turn contribute to the aggressiveness of HCCs. Crosstalk between the tumoral epithelia and stroma has been shown to facilitate tumor growth and cancer progression through EMT, in which TGF-β/ platelet-derived growth factor signaling plays a crucial role [42]. Interestingly, we found a positive correlation between the expression of stemness markers and stromal markers. Therefore, CSCs are not considered to be a fixed cell population and their plasticity might be regulated by tumor stromal factors, and the altered tumor stroma in hypoxia might provide a niche for CSC to proliferate in the residual HCC after TACE. In fact, CD133 transcription has been reported to be induced by IL-6/STAT3 signaling through functional cooperation with NF-κB and HIF-1α during liver carcinogenesis [43]. This is the first study to comprehensively evaluate the changes in immunoprofiles of tumor cells and tumor stroma after TACE using matched biopsied and resected HCCs. One of the limitations of this study is the small number of biopsy-resection matched cases. It is difficult to procure a large cohort of biopsyresection matched HCCs, as biopsy confirmation is not recommended for HCC when imaging findings are diagnostic according to the current guidelines. Survival analysis was not conducted separately for the TACE and non-TACE groups, due to the small number of biopsy-resection matched cases. Further studies based on a larger number of cases from multiple centers are required.
In conclusion, the expression of stemness markers is considered to be increased along with alteration of tumor stroma under TACE-induced hypoxia, which might promote the aggressive biology of HCC. Therefore, checking the expression status of these markers in biopsied HCCs may help to predict a poor outcome of HCCs, especially for TACE treatment.

Case selection and clinicopathologic analysis
The study subjects comprised 46 patients who were diagnosed as HCC by preoperative biopsies and underwent subsequent surgical resection at Severance Hospital from January 2001 to February 2014. All preoperative baseline biopsies were taken prior to any treatment. After the biopsy, 10 cases underwent preoperative TACE ("TACE group") and the remaining 36 cases received no TACE ("non-TACE group"). In all cases with multiple tumors (5 cases in non-TACE group, none in TACE group), the biopsy was performed from the largest nodule. Subsequently, the 46 patients received surgical resection of the HCCs. TACE, containing lipiodol with adriamycin or doxorubicin, was performed once in seven cases, twice in two cases, and three times in one case. The average intervals between biopsy and resection for the non-TACE group and TACE group were 72 days (range, 6 ~ 1039 days) and 65 days (range, 16 ~ 247 days), respectively. Cases that underwent other types of treatment, such as radiofrequency ablation or chemotherapy, were excluded, and cases with total necrosis after TACE were also not included in this study, as their tumor immunoprofiles could not be assessed.
Histopathologic analysis was performed for tumor number, diameter of entire tumor, viable tumor area, differentiation (Edmondson-Steiner grade), tumor capsule formation, serosal invasion, microvascular invasion, presence of tumor necrosis, and proportion of tumor necrosis area to entire tumor region. In the cases with multiple tumors, histopathologic analysis was performed from the largest nodule. Tumornode-metastasis (TNM) classification was analyzed according to the 7th American Joint Committee on Cancer/International Union against Cancer (AJCC/ UICC) staging system. Clinical data, including age, sex, etiology, cirrhosis, serum aspartate aminotransferase (AST), serum alanine aminotransferase (ALT), serum alpha-fetoprotein (AFP), and serum protein induced by vitamin K absence or antagonist II (PIVKA-II), and follow up data were obtained from the electronic medical records. The mean follow-up period after resection was 60.5 months (range, 2.5 ~ 142.4 months). This study was approved by the Institutional Review Board of Severance Hospital, Yonsei University College of Medicine (Seoul, Korea), and the need for patient consent was waived (4-2016-0826).

Immunohistochemical evaluation
Representative blocks of formalin-fixed paraffin embedded tissues were used for immunohistochemical stains. In the case of multiple nodules, we performed the immunohistochmical staining in the largest nodule, which was matched with biopsied HCC. Details of the antibodies used are summarized in Table 5, and immunohistochemistry was performed as previously described [9]. K19, EpCAM, CD133, CAIX and VEGF were stained in the tumor epithelial cells, whereas α-SMA and FAP were stained in the cancer associated fibroblasts (CAFs) of tumor stroma. The positive expression area was defined as a percentage of the total tumor area, and the intensity of staining was evaluated as follows: 1, weak; 2, moderate; or 3, strong. Final scores were obtained by multiplying the positive expression area (%) by intensity, and divided into negative group (score 0), and positive group of low expression (1+) and high expression (2+). For stemness-related and hypoxia markers (K19, EpCAM, CD133, CAIX and VEGF), negative (0) was defined as scores <1 in biopsied and <5 in resected HCCs, low expression (1+) was defined as scores 1 ~ 50 in biopsied and 5 ~ 50 in resected HCCs, and high expression (2+) was defined as score >50 in both biopsies and resection. For stromal markers (α-SMA and FAP), negative (0) was defined as scores <5 in biopsied and <10 in resected HCCs, low expression (1+) was defined as scores 5 ~ 20 in biopsies and 10 ~ 20 in resected HCCs, while high expression (2+) was defined as score >20 in both types of specimen.
In order to evaluate the sequential change in the expression status of each marker between the biopsies and resected specimens, we calculated the differences between the expression scores of each marker in the biopsies and matched resected specimens ("score change"), and then added the differences of each markers to yield the "sum of score change". For example, if K19 expression was low (1+) in the biopsy and high (2+) in the matched resected specimen, the score change was 1. In a similar manner, if the same patient showed low EpCAM (1+) and low CD133 (1+) expression in the biopsy, and high EpCAM (2+) and low CD133 (1+) in the resected tumors, the score changes were 1 and 0, respectively. The sum of score change was therefore 2 (1 + 1 + 0). The same method was applied for the hypoxia-related markers and stromal markers. The staining results were assessed by two pathologists unaware of the clinicopathological data for each case ( Figure 1A).

Statistical methods
Statistical analyses were carried out using SPSS software (version 21.0, SPSS Inc., Chicago, Illinois). Chi-square test or Fisher's exact test was used. Univariate survival analyses were performed by the Kaplan-Meier method with log-rank test, and multivariate survival analyses were conducted by Cox regression with forward conditional method. Statistical significance was assumed when P < 0.05.