Youjia Kong Tengfei Zhuang Ding Xvshen Sirui Cai ^* Weijie Ding ^* Xiaoxiao Zhang ^* Sun Yubo ^* Bingquan Zhou ^* Yuanjie Sun Shuya Yang Xiyang Zhang ^* Kun Yang ^* Dongbo Jiang ^*

• Air Force Medical University, Xi'an, China

Primary liver cancer, with 906,000 new cases, ranks sixth in growth among malignant tumors. Additionally, with 830,000 deaths, it ranks third in terms of mortality (1). Hepatocellular carcinoma (HCC), which is the most important form of primary liver cancer, accounts for approximately 90% of liver cancer cases (2). A variety of risk factors can contribute to the development of primary liver cancer, including hepatitis B virus (HBV) infection, hepatitis C virus (HCV) infection, fibrotic chronic liver damage, aflatoxin B1, and excessive alcohol consumption (3,4). The progression of HCC evolves continuously from dysplastic lesions with minor genetic mutations to the late stages of HCC, displaying significant molecular heterogeneity involving numerous molecules (5). The extensive tumor heterogeneity across multiple stages of HCC development hinders patient stratification for effective treatment (6). Therefore, exploring the tumor heterogeneity of HCC would help stratify patients for effective treatment.The tumor transformation of HCC usually originates from hepatocytes and progenitor cells, both of which are epithelial cell types. The plasticity changes in these epithelial cells commonly known as epithelial-to-mesenchymal transition (EMT) increase the complexity of cellular heterogeneity (6). The EMT program in cancer cells can be transiently or stably activated to varying extents during invasion and metastasis. High expression of adhesion molecules can increase cell migration ability and invasiveness. A significant body of evidence illustrates that the EMT plays an important role in cancer invasion and metastasis (7)(8)(9)(10). By analyzing various EMT phenotypes in malignant epithelial hepatocytes, researchers could estimate the complexity and cellular heterogeneity of HCC. Few studies have investigated several EMT markers in a large number of biopsies, making it difficult to identify the occurrence of EMT based on a single marker alone (11). E-cadherin together with occludin or cytokeratins represents the most commonly used markers for epithelial features, while Ncadherins and vimentin are markers for mesenchymal features (8). Hence, in this study, E-cadherin, N-cadherins, and vimentin (EMT triplet) were selected to characterize the occurrence of EMT. B-cell receptor-associated protein 31 (BAP31) is named for its association with the B-cell receptor component immunoglobulin D. BAP31 exhibits an apparent molecular weight of 31 kDa, a characteristic that is clearly observed on a denaturing electrophoresis gel (12). BAP31 has been identified as a cancer antigen (13). It is overexpressed in cancer tissues compared to healthy adjacent tissues, and it holds promise as a prognostic biomarker for several different types of cancer (14)(15)(16)(17). In our previous study, using mIHC and multispectral imaging techniques, we demonstrated that BAP31 promotes cell proliferation by interacting with Serpin Family E Member 2 (SERPINE2) in hepatocellular carcinoma (HCC) (18). Only a few studies have reported on the role of BAP31 in activating invasion and metastasis. Recent studies have demonstrated that BAP31 can induce epithelial-mesenchymal transition (EMT) by enhancing the expression of the EMT-related factor Snail and decreasing the content and membrane distribution of E-cadherin (19). However, it remains scarce how BAP31 induces the entire EMT process. The current study reported a multi-molecule staining dataset of EMT triplet and BAP31 to estimate tumor heterogeneity and preliminarily explored the relationship between BAP31 and EMT.Conventional immunohistochemistry techniques have several limitations. Multiplex immunohistochemistry(mIHC) technology allows for the simultaneous detection of multiple markers on a single tissue section, providing a new method for comprehensive studies of cell composition, cell function, and intercellular interactions.Moreover, the use of multiplex immunohistochemistry (mIHC) (20).Through the use of mIHC, EMT triplets and BAP31 in liver cancer can be elucidated together.Here, we present our dataset containing numerous images and analyzing data from a digitally scanned high-resolution tissue microarray (TMA) with 138 samples, termed LV138. The TMA was stained for HE and other specific biomarkers, such as BAP31, E-cadherins, N-cadherins, vimentin, and 4,6-diamidino-2-phenylindole (DAPI). Each sample was accompanied by clinical data, pathologist annotations, and staging information. Cell and tissue segmentation, as well as the expression of specific biomarkers, were performed using inForm Advanced image analysis software (inForm 2.6, Akoya). The results of each step were compiled and formed part of the dataset. The utility of our datasets was confirmed by preliminary statistical analysis. Application of dataset can prevent a multitude of repetitive operations and offers an insightful approach for investigating molecular heterogeneity of HCC. This patient group consisted of 138 primary liver cancer patients aged between 21 and 98 years, including 37 women and 101 men. We obtained 138 tissue samples from these individuals and stored them in a TMA-LV138. The study involving human participants was reviewed and approved by the Ethics Committee of the Fourth Military Medical University. Our samples were derived from a commercial tissue microarray (TMA), and the supplier ensured that informed consent was obtained from all patients during the data collection process.The original primary liver cancer tissue microarray (TMA), which was stained with hematoxylin and eosin (HE), was independently evaluated and analyzed by at least two pathologists. This meticulous evaluation led to the identification of 138 viable core samples. Each of these core samples was annotated with essential details such as the diagnosis, grade, TNM classification, stage, and type of cancer. Additionally, each core includes basic patient information and a unique identifier for accurate reference.To enhance visualization, the LV138 was scanned using an Aperio GT 450 scanner from Leica Biosystems (USA). The high-resolution scanning process produced SVS images that were subsequently saved in TIF format, ensuring clear and detailed representation of the tissue samples. The studies involving humans were approved by the institutional review board of the Fourth Military Medical University. The studies were conducted in accordance with the local legislation and institutional requirements. The human samples used in this study were acquired from a by-product of routine care or industry. Multiplexed tyramide signal amplification (TSA) immunofluorescence was performed on the LV138 TMA to simultaneously visualize multiple biomarkers on a single section using Opal 7-plex technology (Akoya).Tyramide Signal Amplification (TSA) is an enzymatic detection method based on horseradish peroxidase (HRP). The principle involves the covalent binding of fluorescently labeled tyramine to the tyrosine residues of the target protein under the catalysis of HRP, thereby labeling the target protein with specific fluorescence (21). The original slides of the LV138 TMA were processed into 5-μm-thick sections. The slides were deparaffinized in xylene and rehydrated in an ethanol gradient.The mIHC staining includes four sequential cycles, each specifically targeting one of the following molecular markers: Vimentin, BAP31, E-cadherin, and N-cadherin. At the start of each staining cycle, the slides are first immersed in either EDTA buffer (pH 9.0) or citrate buffer (pH 6.0). The reaction container with the slides is then placed in a microwave for heat-induced antigen retrieval and stripping. To minimize non-specific binding, we apply a 5% Bovine Serum Albumin (BSA) blocking solution to the slides and allow it to rest for 15 minutes. According to a pre-established and optimized protocol, during each round of staining, the slides are first incubated with primary antibodies-anti-Vimentin (Proteintech; 10366-1-AP), anti-BAP31 (FMU-BAP31-2)(18), anti-E-cadherin (Proteintech; 60335-1-IG), anti-N-cadherin (Proteintech; 66219-1-IG)followed by incubation with HRP-conjugated secondary antibodies (Akoya; Opal Polymer HRP Ms + Rb; ARH1001EA) to detect specific molecular markers. Afterwards,during each round of staining, the slides are incubated with fluorophore-conjugated tyramide-Opal 520, Opal 570, Opal 620, and Opal 690-for 10 minutes to visualize the corresponding molecular markers. Once a TSA staining round is complete, the same method used previously is applied for heat-induced antibody stripping and antigen retrieval. Finally, the cell nuclei are stained with 4,6-diamidino-2-phenylindole (DAPI; Sigma-Aldrich, St. Louis, Missouri, USA; Catalog No. D9542), marking the completion of the entire staining process. The slides were scanned using Vectra 3.0 (Akoya) to acquire multispectral images. All the samples are shown in the resulting images and an example of these images is shown in Figure . 1(A). The multispectral images were unmixed by advanced image analysis software (inForm 2.6.0, Akoya). In this study, we analyzed a total of 138 core samples, comprising 118 hepatocellular carcinoma (HCC) cores, 10 cholangiocarcinoma cores, and 10 normal liver cores. All pertinent data were meticulously documented within the LV138 dataset. The data records were systematically organized into four distinct folders: clinical data, optical acquisition and imaging, Artificial Intelligence(AI) scoring, and colocalization. Each file was methodically named according to the LV138 system, incorporating the file ID and data type to facilitate comprehension of our data structure. To further elucidate our organizational approach, we developed a data result diagram showed in Figure . 2 and appended a file type suffix to each file name (e.g., LV138_ID_coloc_data.txt). The clinical data of the patients are stored in Clinical Data.xlsx, including pathologist annotations, age, sex, diagnosis, grade and unique identifiers for each sample. The images were further analyzed using machine learning for additional interpretation. The inForm 2.6.0 software performed tissue and cell segmentation as well as histochemical scoring of markers, producing outputs of both images and analysis data. To further understand the influence of joint function of biomarks, the colocalizations of biomarkers were analyzed, and the results were divided into two categories, BAP31-EMT markers and EMT markers. The BAP31-EMT marker category includes the colocalization analysis results of each EMT marker with BAP31. The results were divided into four groups: BAP31_E-cadherin, BAP31_Ncadherin, BAP31_Vimentin, and BAP31_E-cad_N-cad_Vimentin. The files, including coloc_data.txt, coloc_quant_data.txt, and quant_data.txt, were stored in the data subfolders, and coloc.tif was stored in the image subfolders.The EMT markers category includes the colocalization analysis results between each pair of EMT markers. The results were divided into the following groups: N-cadherin_E-cadherin, Vimentin_Ecadherin, and Vimentin_N-cadherin. The files, including coloc _data.txt, coloc_quant_data.txt, and quant_data.txt, were stored in the data subfolders, and coloc.tif was stored in the image subfolders. For the preliminary analysis of the data, the datasets were analyzed using GraphPad Prism 10 software (San Diego, CA, United States). The Pearson correlation test was conducted to explore the correlations between four biomarkers, during which the correlation coefficients and P values were calculated, and scatter diagrams were generated. Analysis of variance (ANOVA) was performed to test for the presence of significant differences in different grades, and the H-score was used to represent expression levels. For non-normally distributed rank/ordered variables and data, the Kruskal-Wallis test was used. All P values were two-sided, and a P values less than 0.05 were considered to indicate statistical significance. ANOVA was used to compare the expression levels of BAP31 across different grades of cancer and in normal liver tissue. Notably, BAP31 expression was significantly greater in cancerous tissue than in normal liver tissue, indicating a strong correlation between elevated BAP31 expression and poor prognosis. Pearson correlation analysis revealed a significant relationship between the expression of N-cadherin and E-cadherin in liver cancer and normal liver tissues, with a Pearson's r value of 0.7749 and a P value less than 0.0001. This finding aligns with the results from GEPIA2 (Figure GEPIA 2 -Copyright © 2018) (http://gepia2.cancer-pku.ac.cn), where the Pearson's r value was 0.4 with a P value of 0. To further investigate the correlation between N-cadherin and E-cadherin expression in various tissues, we examined their relationship in normal liver and cancer tissues using the same methodology. The results indicated a significant correlation between a normal liver (Pearson's r = 0.7204, P = 0.0188) and liver cancer (Pearson's r = 0.7723, P < 0.0001). However, no significant correlation was found between the two biomarkers in cholangiocarcinoma (P = 0.7475). These findings were corroborated by data from GEPIA2 (http://gepia2.cancer-pku.ac.cn).Contrary to the traditional view of decreased E-cadherin and increased N-cadherin, the strong correlation between N-cadherin and E-cadherin suggests that the conventional epithelialmesenchymal transition (EMT) model may not be applicable for characterizing hepatocellular carcinoma, but its applicability to cholangiocarcinoma remains to be explored. Recent multicenter study have shown that the expression of N-and E-cadherin are markers for normal hepatocytes and cholangiocytes, respectively, and that the expression of E-and N-cadherin is retained in HCC and intrahepatic cholangiocarcinoma (iCCA) (22,23). his research aligns with our findings, further validating the credibility and value of our dataset.Additionally, the colocalization of E-cadherin and BAP31 was analyzed using ANOVA across four grades of cancer, yielding a p value of 0.0104. The colocalization analysis showed the extent of the overlapping area between the two biomarkers within the entire tissue core, suggesting a potential relationship between the degree of interaction between the biomarkers and the progression of cancer. Notably, survival data from the TCGA database (https://www.cancer.gov/aboutnci/organization/ccg/research/structural-genomics/tcga) indicate a trend toward diminished overall survival with increased transcription of the corresponding mRNA. This observation supports the validity of our database for survival assessments.Our findings suggest that multispectral analysis can significantly contribute to the diagnosis of liver cancer. Further studies based on a larger patient cohort are warranted. Moreover, incorporating survival outcomes could provide a critical indicator for predicting survival rates. Figure1(A): Four biomarkers of normal cores and HCC cores at different stages stained by mIHC technology; a darker color indicates higher biomarker expression. (B) The stained core was multispectral split, and the merged image was split into five images: Opal520-stained vimentin (green), Opal570-stained BAP31 (brown), Opal620-stained E-cadherin (orange), Opal690-stained Ncadherin (red) and DAPI (blue). (C) Colocalization of E-cadherin and N-cadherin. Colocalization analysis was subsequently performed. The percentage colocalization area (golden) was calculated. (D) All three steps to train the inForm for segmentation and the calculation of the H-score. From left to right are the tissue segmentation, cell segmentation and H-score. Each step is further analyzed based on the previous step. Different tissues were first divided into cancer (red), stromal (green), and background (blue) tissues. Then, the cells were segmented. According to the biomarker expression level, we set thresholds to divide cells into 4 grades (blue indicates no expression, yellow indicates +1 expression, orange indicates +2 and brown indicates +3), and the H-score was calculated. The percentage of the colocalization area (golden) was calculated.Figure2: Detailed structure drawing of the data record.