Component with abundant immune‐related cells in combined hepatocellular cholangiocarcinoma identified by cluster analysis

Abstract Combined hepatocellular cholangiocarcinoma (cHCC‐CCA) is a heterogeneous tumor sharing histological features with hepatocellular carcinoma (HCC) and intrahepatic cholangiocarcinoma (iCCA). The tumor immune microenvironment (TIME) of cHCC‐CCA is unclear. We compared the TIME of cHCC‐CCA with that of HCC and iCCA. Twenty‐three patients with cHCC‐CCA after hepatectomy were evaluated in this study. Twenty‐three patients with iCCA and HCC were also included. iCCA was matched for size, and HCC was matched for size and hepatitis virus infection with cHCC‐CCA. Immune‐related cells among the iCCA‐component of cHCC‐CCA (C‐com), HCC‐component of cHCC‐CCA (H‐com), iCCA, and HCC were assessed using multiplex fluorescence immunohistochemistry. Among C‐com, H‐com, iCCA, and HCC, multiple comparisons and cluster analysis with k‐nearest neighbor algorithms were performed using immunological variables. Although HCC had more T lymphocytes and lower PD‐L1 expression than iCCA (P < 0.05), there were no significant differences in immunological variables between C‐com and H‐com. C‐com tended to have more T lymphocytes than iCCA (P = 0.09), and C‐com and H‐com had fewer macrophages than HCC (P < 0.05). In cluster analysis, all samples were classified into two clusters: one cluster had more immune‐related cells than the other, and 12 of 23 H‐com and eight of 23 C‐com were identified in this cluster. The TIME of C‐com and H‐com may be similar, and some immunological features in these components were different from those in HCC and some iCCA. Cluster analysis identified components with abundant immune‐related cells in cHCC‐iCCA.


| INTRODUC TI ON
Combined hepatocellular cholangiocarcinoma (cHCC-CCA) is a rare tumor characterized by hepatocellular and glandular features. The incidence of cHCC-CCA has been reported as 0.7%-14.2% among patients with primary liver cancer. [1][2][3] The prognosis of cHCC-CCA is better than that of intrahepatic cholangiocarcinoma (iCCA) but poorer than that of hepatocellular carcinoma (HCC), 4 and surgical resection is the only treatment for cHCC-CCA. However, there is no consensus regarding systemic treatments for patients with unresectable cHCC-CCA, and these treatments are often selected from among those intended for patients with HCC or iCCA. [5][6][7] Recently, immunotherapies, including immunocheckpoint inhibitors, have emerged as alternative treatments for patients with advanced solid tumors. 8 In Japan, the combination of atezolizumab and bevacizumab has replaced sorafenib as the first-line treatment for unresectable HCC. 9,10 Moreover, peptide vaccine therapy 11,12 and adoptive cell transfer 13,14 have been reported to be effective against HCC.
Contrastingly, for iCCA, immunotherapy has resulted in antitumor responses, but only in a select group of patients; consequently, the introduction of suitable indications has been delayed. 15,16 However, patients with cHCC-CCA are mostly excluded from clinical trials on immunotherapies because of the rarity and specific features of this disease 6 ; there is only one case report on using immune therapy for treating cHCC-CCA. 17 Interpretation of the tumor immune microenvironment (TIME) is important for exploring the therapeutic indication of immunotherapy. 18 Recent studies have shown that high expression of PD-L1 19,20 and abundant tumor-infiltrating lymphocytes [21][22][23] are correlated with the treatment effectiveness of immunotherapy. In the present study, we investigate the composition of TIME in cHCC-iCCA via comparisons and clustering analyses in two contrasting tumors: HCC, a carcinoma that has been shown to respond suitably to immunotherapy, and iCCA, a carcinoma that responds poorly to immunotherapy. The results of the present study could facilitate the introduction of suitable immunotherapies for the disease in future.

| Patients
Twenty-three patients who underwent hepatectomy and were diagnosed with cHCC-CCA between June 2003 and June 2018 at the National Cancer Center Hospital East (Chiba, Japan) were investigated in this study. Using a one-to-one matching approach, 23 size-matched patients from among 55 patients who underwent hepatectomy and were diagnosed with iCCA were also investigated; we also randomly included 23 size-and hepatitis virus infection-  Table S1.

| Discrimination between iCCA and HCC component of cHCC-CCA via H&E staining
Formalin-fixed paraffin-embedded tumor tissues were sectioned into 4µm thick serial sections for each patient with cHCC-CCA, HCC, and iCCA. The sections were subjected to H&E staining. The iCCA-like component and HCC-like component were defined as described previously, 4 i.e., the iCCA component (C-com) was defined as "an area characterized by glandular differentiation with mucin production and abundant fibrous stroma" and the HCC component (H-com) was defined as "an area characterized by trabecular growth with bile production, abundant eosinophilic cytoplasm, and prominent nucleoli." For the mixed subtype, H-com and C-com were defined as areas where an HCC and iCCA area accounted for more than 90%, respectively. Intermediate regions were excluded from evaluation. Each component was differentiated under the supervision of an experienced pathologist (M.K.) for all patients with cHCC-CCA. Figure S1 shows an example of the differentiation procedure for each component via H&E staining of the cHCC-CCA tissues.

| Multiplex fluorescence immunohistochemistry
Immunohistochemistry was performed on 4μm thick tissue sections of cHCC-CCA, iCCA, and HCC. The sections were subjected to multiplex fluorescence immunohistochemistry (MFIH) using a PerkinElmer Opal Kit. Two patterns of MFIH combined with various antibodies were prepared. Table S2 lists the primary antibody conditions used for staining. MFIH images were acquired using an automated multisector imaging system (Vectra version 3.0; PerkinElmer). A maximum of 20 regions of interest (ROIs) (699 × 500 µm) were randomly selected from the C-com and H-com, which were discriminated using H&E staining of the cHCC-CCA specimens ( Figure S2).
Twenty ROIs (699 × 500 µm) were randomly selected for HCC and iCCA. PerkinElmer) was used to evaluate the ROI of each tumor, stroma, and nonevaluated tissue, such as the intraluminal structures, as well as to detect immune-related cells with specific phenotypes; the distribution of immune-related cells was analyzed. Figure S3 illustrates the trainable tissue segmentation in cHCC-CCA. Training sessions for tissue segmentation and phenotype recognition were performed repeatedly until the algorithm attained the level of confidence recommended by the manufacturer (at least 90% accuracy). 24,25 Numerical data of tissues and immune-related cells calculated from   Table S4.  Table 1 shows the comparison of histological variables between iCCA and HCC. The area proportion of the stroma tissue (20.6% vs. 5.9%, P < 0.001) and CK7 positivity (90.1% vs. 0.2%, P < 0.001) was higher, and that of the tumor (78.8% vs. 93.9%, P = 0.002) and GPC3 positivity (5.2% vs. 92.1%, P < 0.001) was lower in iCCA than in HCC. Table 2 shows a comparison of the histological variables between C-com and H-com. Ccom exhibited a higher area proportion of the stroma tissue (41.2% vs.

| Comparison of histological variables
13.7%, P < 0.001) and CK7 positivity (75.6% vs. 18.1%, P = 0.038), and a lower area proportion of the tumor tissue (55.9% vs. 86.3%, P < 0.001) and GPC3 positivity (4.9% vs. 40.7%, P = 0.005) compared to H-com.   Based on the immunological variables of C-com, the distribution of each patient is shown on the heat map associated with dendrograms by hierarchical clustering based on Ward's method ( Figure 2).

TA B L E 2 Comparison of histological variables between C-com and H-com
Although some cases exhibited differences in immunological features between C-com and H-com, almost all cases showed similar trends in immunological features between these components. HCC cases tended to have more immune-related cells than iCCA, C-com, and Hcom tissues. In contrast, some iCCA cases showed fewer lymphocytes compared to in C-com and H-com tissue. Figure 3 shows the t-SNE plots of immunological variables among iCCA, HCC, C-com, and H-com ( Figure 3A). HCC was distributed separately from iCCA. C-com and H-com were distributed randomly. In cluster analysis using the k-NN parameter, all samples were categorized into two clusters, clusters 0 and 1 ( Figure 3B). Table 4  The t-SNE plot of each group is shown in Figure 3C. HCC was identified more frequently in cluster 0 compared to iCCA (82.6% vs.  H&E staining with immunohistochemistry used as an adjunct. 6,27 Our previous study showed that GPC3 and CK7 are pathological markers for H-com and C-com, respectively. [28][29][30] In this study, we also defined the tissue components of cHCC-CCA as C-com and H-com, and then compared the various tissue features of these components. As expected, C-com and H-com exhibited histological characteristics and immunohistochemistry findings similar to those of iCCA and HCC, respectively (Tables 1 and 2).

| Visualization of immunological variables among iCCA, HCC, C-com, and H-com using t-SNE plots
Next, we evaluated the TIME of C-com, H-com, HCC, and iCCA through multiple comparison analysis (Tables 3 and S4). HCC had more CD3 + cells and lower PD-L1 expression compared to iCCA in the tumor tissue, suggesting that the TIME is more active in HCC than in iCCA. We speculated that these results may be related to the fact that immunotherapy against iCCA is not as effective against HCC. In cHCC-iCCA, although some patients showed different immunological features between each component, we observed no significant differences in immunological features between H-com and C-com (Tables 3 and S4, and Figure 2). A previous report suggested that H-com and C-com in cHCC-CCA have different immune microenvironments, with more CD3 + and CD8 + cells observed in H-com than in C-com. 31 The differences between our results and the previous report may be related to the small number of cases, the fact that as many as 82.6% of cases were a mixed type, and the analysis of statistical significance using multiple comparison tests. These results suggest that the infiltration of immune cells is related to some intrinsic features of cHCC-CCA, including tumor antigenicity, rather than the histological characteristics of the tumor tissue. In the comparison of HCC, C-com, and H-com, the density of CD163 + CD204 + cells, known as tumor-associated macrophages (TAMs) of the immunosuppressive cells, 32,33 was lower in C-com and H-com than in HCC (Tables 3 and   S4). In the comparison of iCCA, C-com, and H-com, C-com tended to have more CD3 + cells than iCCA. Moreover, some cases in C-com and H-com had more lymphocytes than in iCCA. Thus, C-com and H-com may have distinct TIME in both HCC and iCCA.
Visualization of the immunological features of the TIME in each case revealed that the distribution of iCCA greatly differed from that In addition, 34.8% of cases with iCCA were also classified in this cluster, suggesting that components with abundant immune-related cells were also present within iCCA. Contrastingly, since cluster 1 comprised cold tumors with a few lymphocytes, immunotherapies such as adoptive cell transfer of T cells might be proposed. 13,14 The immunogenicity of tumors is associated with genetic mutations. 37 Mutations in TP53 and CTNNB1, which are common in HCC, and the IDH1 mutation, which is common in iCCA, are also reported to be associated with the immune microenvironment. 31,[38][39][40][41][42] To examine the relationship among TIME analyzed by MFIH and the characteristics of genetic mutations and the transcriptome in tumor tissue, we also collected other cases prospectively. Unfortunately, we have not yet determined key factors to be able to discuss the biological significance of clusters from MFIH using these genetic analyses with this prospective cohort ( Figure S5). We believe that this is mainly due to the following reason: it is difficult to stratify the immunological status of each component by genetic analyses with bulk tissue samples, especially in a mixed type of cHCC-iCCA.
Immunohistochemistry can be used simultaneously to identify the  (Figures S6 and S7). It was speculated that the comprehensive evaluation of TIME via cluster analysis using the k-NN parameter with multiple variables from MFIH, including not only lymphocyte infiltration but also PD-L1 expression and TAM activation, identified hot tumors independent of viral infection.
Furthermore, accurate prognostic analysis was challenging because the cohort of patients in the present study was small due to the rarity of the disease, and some patients underwent repeat hepatectomy or volume reduction surgery. Although Figure S8 shows that cluster 0 tends to have better prognosis than cluster 1, further studies based on large cohorts at multiple centers, as well as prospective studies, are required to further explore the clinical or research implications of the conclusions of the clustering analysis in the present study.
In conclusion, we could determine the composition of the TIME of cHCC-iCCA using an index for HCC, a carcinoma for which immunotherapy is currently being introduced, and an index of iCCA, a carcinoma for which there are only a few effective immunotherapies. Considering most of the cHCC-iCCA in the present cohort were the mixed type, the boundaries of the tissue regions were very unclear under macroscopy. Approaches such as omics analysis with entire tissue were difficult, therefore we performed an MFIH assay, which is more suitable for the analyzing the TIME of each tissue region clearly. We found no significant