Cancer stem cell marker-positive tumour nests in intrahepatic cholangiocarcinoma are related to epithelial type CD1a-positive dendritic cell inltration and to high Sox9 expression

It is yet a mystery whether dendritic cells (DCs) contact cancer stem cells (CSCs) and uptake CSC antigens in intrahepatic cholangiocarcinoma (ICC). The aim of this study was to examine the histological relationship between tumour cells expressing CSC markers and DCs inltrating ICC. In addition, clinicopathological factors, including progression-free survival (PFS) and overall survival (OS), were compared between cases with many and few CSC marker-positive cells. method (log-rank test) was used to compare PFS and OS. PFS was from the date of surgery to the date of recurrence on the image or the last observation date, and OS was the date of surgery to the date of death or last observation. Univariate or multivariate analysis was performed on the effect of clinicopathological factors on survival using the Cox proportional hazards model. In all tests, P < 0.05 was determined to be signicant. group had signicantly shorter PFS and OS than the Sox9 high group. Sox9 plays an important role in the embryonic formation of several tissues and organs, such as the testis, heart, lung, pancreas, biliary tract, and central nervous system(18). Although Sox9-highly expressed ICC is considered to have a poor prognosis(12), it has also been reported that Sox9 is independent of tumour differentiation in biliary tract cancers, including extrahepatic cholangiocarcinoma, and OS is shortened in patients without positive cytoplasmic expression(34). The latter report suggests that Sox9 is involved in the differentiation of bile duct epithelium, and its loss of expression causes malignant transformation. The present study demonstrated that Sox9-negative cases had no nuclear expression, but negative cases had poor prognosis, as shown by both univariate and multivariate analyses of the Cox proportional hazards model between the Sox9 high and Sox9 low groups.


Results
CD1a + immature DCs in ltrated ICC tumour nests signi cantly more than the other types of DCs. Furthermore, they were frequently localized within CD44v9 − and EpCAM high ICC tumour nests. According to Kaplan-Meier method, PFS and OS were longer in the Sox9 high group than in the Sox9 low group, and according to both univariate and multivariate analyses, Sox9 expression was an independent factor.

Conclusion
The present study rst demonstrated that CD1a + immature DCs frequently in ltrated CD44v9 − and EpCAM high ICC tumour nests, suggesting the possibility of cell-to-cell contact between epithelial type immature DCs and CSCs in ICC. Furthermore, it was suggested that Sox9 expression in ICC may be a prognostic factor. Background Dendritic cells (DCs) are functionally classi ed into immature, mature, or activated cells (1). Immature DCs become mature or activated DCs by pathogens and in ammatory cytokines and can stimulate and activate T cells. In antitumour immunity, immature DCs in ltrate tumour tissues, and when they uptake tumour antigens, they migrate to the draining lymph nodes while maturing and activating to present tumour antigens to T cells. Antigen-speci c T cells leave the lymph nodes and in ltrate the tumour again to exert antitumour immunity (2). Indeed, many carcinomas contain various levels of migrating immature DCs, mature/activated DCs, and plasmacytoid DCs (3), and it has been reported that frequent CD1a + DCs also in ltrate intrahepatic cholangiocarcinoma (ICC) tissue (4).
Cholangiocarcinoma is a relatively rare malignant and lethal tumour derived from bile duct epithelium, and the morbidity is now increasing worldwide. This type of carcinoma is closely related to the tumour microenvironment and can be regulated by the interaction between cancer stem cells (CSCs) (5). CSCs are small cell populations of cancer cells capable of self-renewal and pluripotency and bring morphological and other types of diversity to tumours (6). CSCs also contribute to tumour initiation, malignant growth, and chemoresistance (7). Various markers expressed on CSCs include CD133(8), CD44(9), epithelial cell adhesion molecule (EpCAM), MOC31, CD326(10), aldehyde dehydrogenase 1 (11), sex determining region Y-box2 (Sox2) (9), and Sox9 (12). CD44 is a cell surface adhesion molecule that binds to extracellular matrix molecules, such as hyaluronic acid and osteopontin (13), and has ten subtypes (14). In particular, CD44 variant 9 (CD44v9) regulates the redox potential and suppresses the accumulation of reactive oxygen species in cells, thereby maintaining CSC survival (15). EpCAM is a cell surface marker for endodermal progenitor cells (16) and serves as an adhesion molecule between cells. Although EpCAM is involved in the control of cell proliferation, EpCAM-positive ICC cases have a poor prognosis (17). Sox9 is an endodermal transcription factor(16) that is involved in the maintenance of bile duct progenitor cells in the normal adult liver. However, Sox9-positive ICC has the potential to undergo metastasis, invasion, and epithelial-mesenchymal transition, resulting in poor prognosis (18).
Recent studies using CSCs as target cells for the treatment of cancers have been published (19), but it is yet mysterious whether DCs can recognize CSC antigens in human cancer tissues. Furthermore, information about the relationship between the frequency of CSC marker-expressing tumour cells and the survival of ICC patients is still limited. The aim of this study was to examine the histological relationship between tumour cells expressing CSC markers, such as CD44v9, EpCAM, and Sox9, and DCs in ltrating ICC. In addition, clinicopathological factors, including progression-free survival (PFS) and overall survival (OS), were compared between cases with many and fewer CSC marker-positive cells.

Tissue samples
Samples from twenty-two ICC cases that had been pathologically diagnosed at Yamagata University Hospital and Yonezawa Municipal Hospital from 2003 to 2018 were used. The diagnosis of ICC was made based on the WHO classi cation (20). One case of breast cancer and colon cancer was used as positive controls for the CSC markers CD44v9 and EpCAM (breast cancer) and Sox9 (colon cancer) in immunohistochemistry and reverse transcription-polymerase chain reaction (RT-PCR) (21)(22)(23). One tonsillar tissue with chronic tonsillitis was used as a positive control for DC markers in immunohistochemistry. All the excised tissues were xed with 10% buffered neutral formalin at room temperature for 12 to 48 hrs and embedded in para n. This study was approved by the Yamagata University School of Medicine Ethics Committee (H29-302).
First, the number of positive cells for ve different DC markers within tumour nests of ICC were counted. Ten tumour nests in ltrated by the higher number of DCs positive for each DC marker, such as CD1a, DC-SIGN, DEC205, DC-LAMP, and CD123, were photographed under a high-power eld (HPF, 400x in magni cation). The number of DCs in ltrating tumour nests was counted on the picture screen, and the ratio of tumour nests simultaneously occupying the screen was de ned as the tumour nest occupancy rate (Supplemental Fig. 1a & 1b). Finally, "the corrected DC number" was calculated as "the number of DCs within the tumour nest per HPF"/"tumour nest occupancy".
Second, the percentage of CSC marker-positive tumour cells per total of tumour cells in a section was scored as follows: score 0, 0%; score 1+, 1-25%; score 2+, 26-50%; and score 3+, > 50%. Furthermore, cases positive for CD44v9 were divided into CD44v9 + group for scores 1 + to 3 + and CD44v9 − group for score 0, those positive for EpCAM were divided into EpCAM high group for score 3 + and EpCAM low group for scores 0 to 2+, and those positive for Sox9 were divided into the Sox9 high group for score 3 + and Sox9 low group for scores 0 to 2+.
Third, the number of each of the ve types of DCs (CD1a, DC-SIGN, DEC205, DC-LAMP, and CD123) in ltrating the tumour nests of each of the CD44v9 + and CD44v9 − , EpCAM high and EpCAM low , and Sox9 high and Sox9 low groups was compared as follows. Serial sections were prepared from formalin-xed and para n-embedded tissues, and a pair of sections among them were immunostained for the DC marker and the CSC marker. Tumour nests immunopositive for the CSC marker in one section were encircled by lines on the HPF photographs, and the same nests immunostained for the DC marker in another section were encircled by lines on HPF photographs. The numbers of DC marker-positive cells in the encircled tumour nests were counted, and the same method was used to calculate "the corrected DC number" (CD44v9 + group) (Supplemental Fig. 1c & d). Haematoxylin and eosin staining of a serial section assisted in the calculation of "the corrected DC number" within the CSC marker-negative tumour nests (CD44v9 − group). Similarly, in the cases of EpCAM and Sox9 immunostaining, both tumour nests exhibiting a score 3+ (EpCAM high and Sox9 high groups) and scores 0 to 2+ (EpCAM low and Sox9 low groups) were independently marked, and "the corrected DC number" within each tumour nest was calculated. One to 10 tumour nests in ltrated by a high number of DCs positive for each DC marker were evaluated. These counts and calculations were performed independently by two pathologists (U.A. and K.T.).
Finally, the correlation between the frequency of the expression of CSC markers on ICC tumour cells and PFS and OS was evaluated. That is, the PFS and OS were compared between the CD44v9 + and CD44v9 − groups, the EpCAM high and EpCAM low groups, and the Sox9 high and Sox9 low groups by the Kaplan-Meier method (log-rank test).

Statistical analysis
Statistical analyses were performed with JMP version 14 (SAS Institute, Tokyo, Japan) and Microsoft Excel 2013 (Microsoft, Redmond, WA). The Mann-Whitney U test was performed for comparison between the two groups. The Kaplan-Meier method (log-rank test) was used to compare PFS and OS. PFS was from the date of surgery to the date of recurrence on the image or the last observation date, and OS was the date of surgery to the date of death or last observation. Univariate or multivariate analysis was performed on the effect of clinicopathological factors on survival using the Cox proportional hazards model. In all tests, P < 0.05 was determined to be signi cant.

Results
Comparison of the number of DCs in ltrating the ICC tumour nests "The corrected DC number" within 10 tumour nests was measured in 22 cases with ICC. The numbers of positive DCs for CD1a, DC-SIGN (immature DCs), DEC205, DC-LAMP (mature DCs), and CD123 (plasmacytoid DC) were 4.89 ± 5.97, 0.53 ± 2.74, 0.29 ± 1.71, 1.48 ± 5.28, and 0 ± 0, respectively (Fig. 1a-e). CD1a + immature DCs were the most abundant among any type of DCs in ltrating the tumour nests (P < 0.01) ( Table 1). Comparison of "the corrected DC number" in ltrating tumour nests of ICC between the CD44v9 + and CD44v9 − groups, the EpCAM high and EpCAM low groups, and the Sox9 high and Sox9 low groups Fourteen cases were included in the CD44v9 + group (score 1+, 7 cases; 2+, 6 cases; and 3+, 1 case) and 8 cases were included in the CD44v9 − group (score 0); 18 cases were included in the = EpCAM high group (score 3+) and 4 cases were included in the EpCAM low group (score 0, 2 cases; 1+, 1 case; and 2+, 1 case); and 16 cases were included in the Sox9 high group (score 3+) and 6 cases were included in the Sox9 low group (score 0, 2 cases; 1+, 0 case; and 2 +, 4 cases) (Fig. 1f, g, & h). The comparison of "the corrected DC number" in ltrating the tumour nests between the CD44v9 + and CD44v9 − groups, the EpCAM high and EpCAM low groups, and the Sox9 high and Sox9 low groups is shown in Table 2. The number of CD1a + DCs was higher in the CD44v9 − group (4.39 ± 5.30) than in the CD44v9 + group (2.66 ± 4.54) (P < 0.01) and higher in the EpCAM high group (5.52 ± 7.64) than in the EpCAM low group (2.50 ± 3.80) (P < 0.01), although there was no signi cant difference between the Sox9 high (4.00 ± 6.01) and Sox9 low groups (2.74 ± 3.57). There were, however, no signi cant differences in the numbers of any DCs positive for DC-SIGN, DEC205, DC-LAMP, and CD123 between the CD44v9 + and CD44v9 − groups, the EpCAM high and EpCAM low groups, and the Sox9 high and Sox9 low groups. Expression of mRNA of CSC markers by RT-PCR Positive mRNA bands for CD44v9 and EpCAM in breast cancer and for Sox9 in colorectal cancer were con rmed for use as positive controls (Fig. 2). All 4 ICC cases expressed any mRNAs for CD44v9 at 52 bp, for EpCAM at 100 bp, and for Sox9 at 132 bp.
Comparison of PFS and OS between the CD44v9 − and CD44v9 + groups, the EpCAM low and EpCAM high groups, and the Sox9 low and Sox9 high groups Comparison between the CD44v9 + and CD44v9 − groups and the EpCAM low and EpCAM high groups by the Kaplan-Meier method (log-rank test) showed no signi cant difference in PFS (P = 0.437) and OS (P = 0.790) (Fig. 3a & b) or in PFS (P = 0.084) and OS (P = 0.095) (Fig. 3c & d), respectively. However, the Sox9 low group had signi cantly shorter PFS (P = 0.029) and OS (P = 0.012) than the Sox9 high group (Fig. 3e & f).

Univariate And Multivariate Analyses For Clinicopathological Factors
There were no signi cant associations between any clinicopathological factors and OS in either the univariate or multivariate analysis (Table 3). On the other hand, only Sox9 among the clinicopathological factors in both the univariate and multivariate analyses was signi cantly associated with PFS (P = 0.016 and 0.025, respectively) ( Table 4).  we demonstrated that the frequency of CD1a + immature DCs was signi cantly higher in the CD44v9 − group than in the CD44v9 + group and in the EpCAM high group than in the EpCAM low group ( Table 2). The mRNA expression of any CSC marker was con rmed by RT-PCR (Fig. 3). CD44v9 interacts with the rst apoptosis signal (Fas) and suppresses apoptosis of tumour cells (27), suggesting that CD44v9-positive CSCs may hardly cause apoptosis, so very few tumour antigens may be released.
Even if CD1a + DCs in ltrate the CSC nests, few antigens may be captured. Conversely, it may be possible that CD1a + DCs in ltrate CD44v9-negative tumour nests relatively easily, as shown in this study, and thus easily capture tumour antigens. EpCAM is a 40 kD glycoprotein and a homophilic cell-cell adhesion molecule (28). Epithelial cells overexpressing EpCAM are able to downregulate E-cadherin, and increasing expression of EpCAM in cadherin-positive cells leads to the gradual abrogation of adhere junctions (29). Therefore, it may be possible that CD1a + DCs may in ltrate tumour nests more easily because the adhesion ability between tumour cells may be weaker in EpCAM-overexpressing tumours.
The CD44v9 + group had a poor prognosis for hepatocellular carcinoma, breast cancer, plasmacytoma, gastric cancer, bladder cancer, gallbladder cancer and ICC (15,21,(30)(31)(32)(33). There was, however, no signi cant difference in the PFS and OS between the CD44v9 + and CD44v9 − groups by the Kaplan-Meier method (log-rank test) and Cox proportional hazards model in the present study (Tables 3 & 4, Fig. 3). The amount of EpCAM expression on ICC tumour cells was also not associated with survival in the Cox proportional hazards model. It is suggested that one reason for this is because the number of cases may be small in this study. The Sox9 low group had signi cantly shorter PFS and OS than the Sox9 high group. Sox9 plays an important role in the embryonic formation of several tissues and organs, such as the testis, heart, lung, pancreas, biliary tract, and central nervous system (18). Although Sox9-highly expressed ICC is considered to have a poor prognosis (12), it has also been reported that Sox9 is independent of tumour differentiation in biliary tract cancers, including extrahepatic cholangiocarcinoma, and OS is shortened in patients without positive cytoplasmic expression (34). The latter report suggests that Sox9 is involved in the differentiation of bile duct epithelium, and its loss of expression causes malignant transformation. The present study demonstrated that Sox9negative cases had no nuclear expression, but negative cases had poor prognosis, as shown by both univariate and multivariate analyses of the Cox proportional hazards model between the Sox9 high and Sox9 low groups.

Conclusion
The present study rst demonstrated that epithelial type CD1a + immature DCs frequently in ltrated CD44v9 − and EpCAM high ICC tumour nests, suggesting the possibility of cell-to-cell contact between this type of DC and CSCs in ICC. Furthermore, the present study also suggested that the frequent expression of Sox9 in ICC may be an independent prognostic factor.

Figure 3
Relationship between the frequency of cancer stem cell marker expression and Kaplan-Meier curve in ICC. There was no signi cant difference in the PFS (P = 0.314) or OS (P = 0.751) between the CD44v9-and CD44v9+ groups (3a & 3b) or in the PFS (P = 0.107) or OS (P = 0.219) between the EpCAMhigh and EpCAMlow groups of ICC (3c & 3d). However, both the PFS (P =