Patients and samples

We searched the electronic pathology database in the Department of Pathology of the Nanjing Drum Tower Hospital in China over the period from April 2008 to May 2022 and identified a total of 403 consecutive tumors diagnosed with ICCs in radical resections (Fig. 1). The inclusion criteria were as follows: (1) pathologically diagnosed adenocarcinoma, mucinous adenocarcinoma according to the latest WHO classification; (2) radical surgical resection with nodal dissection; (3) complete clinical, radiological and pathological data. Exclusion criteria included: (1) palliative resection without nodal dissection; (2) preoperative local or systematic anticancer neoadjuvant therapy; (3) incomplete clinical and radiologic data; (4) samples can’t be tested by FISH or IHC. Finally, 304 cases were included in this study. Medical records of all patients were reviewed. Tabulated was preoperative information, including the general information, HBV infection, dipsomania, hepatic steatosis, biliary hamartoma, clonorchiasis. The pathological data, such as tumor number, maximum dimension, histologicalclassification, mVI, differentiation (L, low differentiated; M, moderately differentiated; H, high differentiated; U, undifferentiated), and G (grade)/S (stage)/T (tumor) stage were recorded. Tumor TNM staging was determined according to the eighth edition of the American Joint Committee on Cancer (AJCC)/Union for International Cancer Control (UICC) TNM Classification and Stage Groups for ICC. Moreover, because of the scattered large geographic location of the patients and long follow-up duration, only 60.2% (183/304) of patients in researches were obtained the follow-up information. This study was supported by Nanjing Drum Tower Hospital ethics committee.

Fig. 1

Flowchart of study design

Tissue microarray construction

Each tissue sample underwent immediate fixation in 10% neutral buffered formalin for duration of 12–48 h, followed by paraffin embedding. The processed samples were subjected to routine deparaffinization and rehydration procedures. A tissue microarray (TMA) was created using the Grand Master automated array (3DHISTECH Ltd., Budapest, Hungary), with 2 mm punch size obtained from representative tumor blocks of each case. The tumor core was extracted from the invasive front of the deepest tumor invasion portion, with avoidance of necrotic areas. Before constructing tissue microarrays, each sample was subjected to hematoxylin and eosin (HE) staining, and the tumor areas were circled. Subsequently, two cores were randomly selected from the circled tumor areas to create tissue microarrays. ual representations of TMAs were constructed and then sectioned into 4-μm-thick sections for histological, immunohistochemical, and FISH detection procedures. To address the issue of heterogeneity in tissue microarrays, samples with discordant results from the two cores were re-evaluated across the entire slide. This was done to ensure that the selected cores could represent the overall status of the entire tissue section.

Immunohistochemical staining and analysis

The anti-HER2/neu monoclonal antibody (CST, cot:CST4290) was used for HER2 immunohistochemistry (IHC) staining through an automated protocol validated on the Ventana Bench Mark Ultra system (Roche Diagnostics, Basel, Switzerland). Two pathologists independently interpreted the HER2 IHC results in the absence of clinical or pathological information, and any disagreements were adjudicated through consultation. In cases of inconsistent duplication, the entire slide was stained for final interpretation. HER2 scores were determined by multiplying the positive intensity with the positive percentage. Intensity was scored as follows: 0 = no staining; 1 +  = weak, incomplete membrane staining in > 10% of tumor cells; 2 +  = weak to moderatelycomplete membrane staining in > 10% of tumor cells; 3 +  = strong, complete membrane staining in > 30% of tumor cells. Finally, based on the HER2 scores, we adopted the American Society of Clinical Oncology/College of American Pathologists (ASCO/CAP) recommended criteria (Schalper et al. 2014) for breast cancer and categorized IHC values into 1 + for 1–50, 2 + for 51–150, and 3 + for values greater than 150.

In addition to HER2 evaluation, immunophenotypic analysis of tumor cells and tumor-infiltrating lymphocytes (TILs) was performed using IHC staining for CD20 (Abcam, cot:Ab78237), CD3 (Abcam, cot:Ab135372), CD68 (CST, cot:CST76437), CD8 (CST, cot:CST30706), CD4 (ZSGB-BIO, cot:ZM-0418), and CD163 (ZSGB-BIO, cot:ZM-0428), FOXP3 (Biolegend, cot:BLG320202), and PD1 (ZSGB-BIO, cot:ZM-0381) in whole slide of 19 cases. The IHC scores for CD20, CD3, CD68, CD8, CD4, and CD163 were calculated separately based on the percentage of positive cells in the tumor center and tumor periphery. On the other hand, the percentage of FOXP3 or PD1-positive cells was calculated considering both the tumor center and tumor periphery. 19 samples were randomly selected to immunophenotypic analysis using whole tumor sections. Among these, 7 samples exhibited HER2 amplification, while 12 samples with no HER2 amplification as controls.

Multiplex immunofluorescence staining and evaluation

Multiplex immunofluorescence staining was performed using the Opal 5-Color Manual IHC Kits (Panovue Biotechnology) following the manufacturer's instructions. Two panels, consisting of CD20 (Abcam, cot:Ab78237), CD3 (Abcam, cot:Ab135372), CD68 (CST, cot:CST76437), HER2 (CST, cot:CST4290), and CK (Sigma, cot:C2562), as well as CK, PD-L1 (CST, cot:CST13684), CD8 (CST, cot:CST30706), FOXP3 (Biolegend, cot:BLG320202), and CTLA4 (Abcam, cot:ab237712), were visualized using the Olyvia System and analyzed with Qupath software (Bainuo, China). The Olyvia System quantitatively assessed the average density (cells/mm2)of each lymphocyte subset or merged lymphocyte subsets across the entire slide. The expression of PD-L1 in tumor-infiltrating lymphocytes (TILs) was evaluated using an immunoreactivity scoring system (IRS). PD-L1 staining was considered positive if any perceptible membrane staining, whether partial or complete, of any intensity distinct from cytoplasmic staining was present. The intensity of staining was categorized as weak (score 1), moderate (score 2), or strong (score 3). PD-L1 expression immunoreactive score (IRS) was calculated by multiplying the positive intensity by the positive percentage. All multiplex immunofluorescence staining was conducted on whole slides. In order to evaluate different HER2 statuses in cholangiocarcinoma samples, a total of 28 cases were randomly included in the multiplex immunofluorescence staining analysis, comprising 11 samples with HER2 amplification and 17 samples without HER2 amplification.

Fluorescence in situ hybridization

For fluorescent in situ hybridization (FISH), a commercially available, locus-specific HER2 probe and CEP17 probe were used as recommended by the manufacturer (Kanglu, Wuhan, China). At least 20 non-overlapping nuclei of tumor cells per sample were evaluated for HER2 probe (red) and CEP17 probe (green) signals, and the signal counting results of HER2 and CEP17 were recorded for further evaluation. Similar to IHC, FISH was performed in both duplicates of each case, and in case of disagreement between the duplicates in FISH results, the entire slide was used for a final decision.

Concordance index between IHC and FISH

We used the FISH results as the final outcome to assess the concordance between IHC and FISH results.

Next-generation sequencing

Out of the 304 cases of ICCs included in the study, 283 cases underwent Whole-Exome Sequencing (WES) analysis. DNA and RNA from tumor tissues were extracted, and sequencing libraries were prepared. DNA-based target sequencing was performed on a panel of 1021 cancer-related genes. Complete DNA and RNA sequencingwas performed on a Gene + Seq2000 (Beijing Gene Plus, Beijing, China.) or DNBSEQ-T7 (Beijing Genomics Institute, Beijing, China.) instrument. Sequencing data were analyzed using the default parameters. Adaptor sequences and low-quality reads were removed before aligning to the reference human genome (hg19) using the Burrows‒Wheeler Aligner (BWA; version 0.7.12-r1039). Realignment and recalibration were performed by using GATK (version 3.4-46-gbc02625). Single nucleotide variants (SNVs) were called using MuTect (version 1.1.4) and NChot, software developed in-house to review hotspot variants. Small insertions and deletions (Indel) were determined using GATK. Somatic copy number variations (CNVs) were identified using CONTRA (v2.0.8). The fusion genes were identified with the NCsv program (in-house) using split reads, discordant pair reads, and single unmapped reads in the alignment file. The final candidate variants were all manually verified using the Integrative Genomics Viewer.

In each group, the frequency of gene mutations was calculated, retaining only genes with a mutation frequency of 10% or higher. Tumor mutation burden was defined as the sum of synonymous and non-synonymous mutations. A waterfall plot was generated using the complexHeatmap function in R (v 4.2.2).

The Wilcoxon signed-rank sum test was employed to assess the differences in gene expression between the two groups. Differentially expressed genes were identified based on a criteria of p value < 0.05 and absolute log2 fold change > 1. Expression profiles of the selected differentially expressed genes were extracted from the original expression data. The data was first normalized by samples, followed by normalization by genes. Subsequently, a heatmap was generated using the pheatmap function in R (v 4.2.2), with sample clustering based on the default parameters of pheatmap.

Statistical analysis

Overall survival (OS) time was defined as the time in months from surgery to death or the last available follow-up. Patients who were alive at follow-up were censored. Statistical analyses were conducted using IBM SPSS Statistics for Windows, version 23.0. Categorical variables were analyzed using the Wilcoxon rank-sum, chi-square or Fisher's exact test, while continuous variables were analyzed using the t test. OS was analyzed using the Kaplan–Meier method and p value relates to the log rank analysis. Univariate analysis was performed for prognostic factors using the log-rank test. All graphs were generated using GraphPad Prism version 9 (GraphPad Software Inc., San Diego, CA). A p value of less than 0.05 was considered statistically significant.