Hepatoid adenocarcinoma of the stomach and non-hepatoid alpha-fetoprotein-producing gastric cancer exhibit a high degree of molecular similarity Authors: Liqiao Chen, Xuesong Yang, Peiyu Zhu, Han Bao, Yan Wu, Ke Ji, Ji Zhang, Xiaojiang Wu, Kai Zhou, Jieli Xu, Jiatian Tang, Anqiang Wang, Zhaode Bu Published: 2025 ABSTRACT AbstractBackgroundThere is currently no unified consensus on the diagnosis and treatment of hepatoid adenocarcinoma of the stomach (HAS) and non-hepatoid AFP-producing gastric cancer (AFPGC). This study aims to explore the molecular similarities between the two, providing a basis for the diagnosis and precision treatment of these patients.MethodsWe retrospectively collected tumor tissues, adjacent tissues, and peripheral blood samples from 83 patients for whole-exome sequencing or transcriptome sequencing. Spearman correlation analysis, unsupervised clustering analysis and so on were performed to assess the similarity between different sample groups, explore the molecular features of non-hepatoid AFPGC and HAS, and compare their correlations.ResultsAll the patient groups shared high-frequency mutated genes such as TP53, LRP1B, MUC16, CSMD3, and FAT4. Copy number variation analysis revealed similarities in the copy number variations between the two patient groups. The majority of patients in both groups exhibited amplification of the CCNE1 or ERBB2. PCA analysis based on transcriptomic data showed a clear clustering trend within the HAS and non-hepatoid AFPGC subgroups, which was distinct from conventional gastric adenocarcinoma. Moreover, unsupervised clustering analysis indicated that the samples within the different subgroups of the two groups had similar transcriptional expression patterns. Finally, we identified a potential therapeutic target, FAT4. Mutations in FAT4 further affect transcriptional expression and prognosis in gastric cancer patients, as well as influence immune infiltration and the response to immune checkpoint blockade therapy.ConclusionHAS and non-hepatoid AFPGC exhibit a high degree of similarity at both the genomic and transcriptomic levels.Clinical trial numberNot applicable.Graphical Abstract Supplementary InformationThe online version contains supplementary material available at 10.1007/s13402-025-01145-7.Keywords: Hepatoid adenocarcinoma of the stomach, Non-hepatoid AFP-producing gastric cancer, Whole-exome sequencing, Transcriptome sequencing Introduction Over the past several decades, the incidence and mortality rates of gastric cancer have decreased globally. However, it remains the fifth most common cancer and the third leading cause of cancer-related death worldwide [1, 2]. In high-incidence regions, particularly East Asian countries, the health and economic burden posed by gastric cancer remains significant. According to the latest data from the National Cancer Center of China, as of 2022, the incidence and mortality rates of gastric cancer in China follow the aforementioned trends, continuing to present a concerning issue [3]. Hepatoid adenocarcinoma of the stomach (HAS) is a rare but highly aggressive subtype of gastric cancer, with its incidence varying across different study populations and diagnostic criteria. Due to its high invasiveness and metastatic potential, patients with this subtype often develop liver metastasis, resulting in poor prognosis [4]. However, the diagnosis of HAS remains a contentious issue. According to the World Health Organization’s fifth edition of the Classification of Digestive System Tumours, the diagnosis primarily relies on the identification of hepatocellular differentiation areas in histopathological slides [5]. Additionally, patients frequently exhibit elevated serum alpha-fetoprotein (AFP) levels, though this is not always observed in all cases. In terms of prognosis, patients with hepatocellular-like gastric adenocarcinoma have a significantly worse outcome compared to those with non-hepatocellular adenocarcinoma. Liver metastasis and elevated serum AFP levels are often indicative of poor prognosis [6, 7]. Non-hepatoid alpha-fetoprotein-producing gastric cancer (AFPGC) also shares similar clinical and pathological features, including high invasiveness and metastatic potential [8]. However, while no hepatocellular differentiation areas are observed in histopathological slides of non-hepatoid alpha-fetoprotein-producing gastric cancer, the inability of pathological examination to capture all tumor aspects raises the possibility of missed diagnoses [9]. Furthermore, the definitions of HAS and non-hepatoid AFPGC overlap, and the currently adopted definitions and classification methods often yield differing results across various studies [10]. Some researchers have proposed that the prognosis of patients with non-hepatoid AFPGC may be worse than that of patients with HAS [11]. Therefore, integrated and comparative research on these two subtypes is urgently needed. In this study, we compared the genomic and transcriptomic correlations HAS and non-hepatoid AFPGC. These samples underwent genomic and transcriptomic sequencing, combined with data from the The Cancer Genome Atlas (TCGA) on conventional gastric adenocarcinoma, to describe the basic genomic and transcriptomic features of different types of non-hepatoid AFPGC and HAS. Based on these findings, the study aimed to assess the molecular difference of different subtypes of HAS and related gastric cancer to conventional gastric adenocarcinoma, and identify key differences at the molecular level. Finally, we identified a target that exhibits mutations and expression alterations in both HAS and non-hepatoid AFPGC. We conducted further analysis on this target with the aim of identifying potential therapeutic targets for these patients. Methods Patient inclusion and exclusion criteria This study intends to include patients with HAS and non-hepatoid AFPGC, who underwent gastric cancer resection surgery at the Gastrointestinal Oncology Center of Peking University Cancer Hospital from October 2009 to November 2021. Inclusion Criteria: surgical resection; pathologically confirmed primary gastric cancer; patients with elevated serum AFP before treatment, or postoperative pathological AFP immunohistochemistry positivity, or those pathologically confirmed as HAS. Exclusion Criteria: presence of preoperative metastasis (including peritoneal metastasis, positive or suspicious positive cytology of ascites); coexistence of other systemic malignant tumors (excluding low-grade malignant tumors such as thyroid papillary carcinoma or basal cell carcinoma); age over (and including) 80 years; incomplete clinical data; no preserved tumor tissue or normal tissue paraffin or fresh frozen specimens, or poor specimen preservation. Additional methodological content is described in the Supplementary material. Results Sample grouping and characteristics In this study, tumor samples from 83 patients with HAS and non-hepatoid AFPGC were collected for sequencing research, including 33 paraffin samples and 50 frozen samples. Corresponding para-tumor and peripheral blood specimens were also collected as controls. Whole exome sequencing was conducted on all tumor and normal samples, and transcriptome sequencing was performed on the frozen samples. Among the 83 patients, 29 were classified as non-hepatoid AFPGC and 54 were classified as HAS. For the non-hepatoid AFPGC samples: 13 samples had elevated serum AFP but were immunohistochemically AFP-negative; 13 samples were immunohistochemically AFP-positive but had normal serum AFP levels; 3 samples had both elevated serum AFP and were immunohistochemically AFP-positive. Finally, 16 cases were included in the serum AFP elevated group (AFPeGC), 16 cases in the immunohistochemically AFP-positive group (AFPiGC), and 54 cases in the HAS group. Somatic mutation distribution and gene mutation landscape Exome sequencing analysis of all tumor samples showed that a total of 11,266 somatic mutations were detected across the 83 tumor samples, with an average of 135.7 mutations per sample. Among the groups: The HAS group had 7,527 mutations, with an average of 139.4 mutations per sample. The AFPiGC group had 2,331 mutations, with an average of 145.7 mutations per sample. The AFPeGC group had 1,764 mutations, with an average of 110.3 mutations per sample. There was no significant difference in mutation counts between the HAS and non-hepatoid AFPGC groups (P = 0.214). Missense mutations accounted for the largest proportion of mutations in all three groups, with a total missense mutation rate of 79.3% (8,932/11,266). In the HAS group, 81.8% (6,158/7,527) of somatic mutations were missense mutations, while in the AFPiGC and AFPeGC groups, 74.0% (1,726/2,331) and 75.2% (1,327/1,764) of somatic mutations were missense mutations, respectively. The second and third most common types of somatic mutations in HAS and related gastric cancer were splice site mutations (8.8%, 997/11,266) and in-frame/frameshift mutations (6.4%, 726/11,266) (Fig. 1A). We further analyzed the known driver genes and high-frequency mutated genes in all samples. Figure 1B showed a panoramic view of the top 30 known driver genes by mutation frequency, with significantly mutated genes highlighted in bold. We found that the most common mutation type in the driver genes of the two patient groups was missense mutations, and there was no significant difference in the mutation frequency of each driver gene between the two groups. Subsequently, we further separately mapped the genomic mutation landscapes of the HAS, AFPeGC, and AFIpGC groups, and calculated whether there were differences in their mutation frequencies. The results showed that, compared to the HAS group, the mutation frequencies of TP53, FAT4, CDH1, and RHOA were significantly different in the AFPeGC group. In the AFPiGC group, there were significant differences in the mutation frequencies of TP53, CNTNAP2, and TET2 (Supplementary Fig. 2–4). Genomic mutational landscape of HAS and AFPGC. A. The distribution of somatic mutation numbers and mutation types. B. Mutation landscape of HAS and non-hepatoid AFPGC based on high-frequency mutated genes and driver genes The TP53 gene was the most frequently mutated driver gene in two groups, with mutations detected in 56% of the HAS and non-hepatoid AFPGC. In the HAS group, the TP53 mutation rate was 66.7% (36/54), while in the AFPiGC and AFPeGC groups, it was 37.5% (6/16) in both. It was noteworthy that the TP53 mutation types in the AFPeGC group were significantly different from the other two groups. Among the six patients with TP53 mutations in the AFPeGC group, three samples had nonsense mutations, one sample had a frameshift mutation, one sample had an in-frame/frameshift mutation, and one had a splice site mutation, with no missense mutations. In contrast, in the AFPiGC and HAS groups, the TP53 mutations were predominantly missense mutations. Other high-frequency mutated driver genes include LRP1B (19%), MUC16 (17%), CSMD (15%), and FAT4 (14%), with LRP1B, CSMD, and FAT4 being significantly mutated genes. Figure 1B showed that all the groups share some high-frequency mutated genes and driver genes, exhibiting certain similarities. Mutational signatures of HAS and non-hepatoid AFPGC We used the 30 known COSMIC mutation signatures to analyze the mutation features and mutation frequencies of all included samples [12]. The results showed that the HAS group exhibited a high frequency of mutation feature 1 (Signature 1), the AFPeGC group had a high frequency of mutation feature 3 (Signature 3), and the AFPiGC group exhibited a mixture of high-frequency mutation feature 3 in some samples and high-frequency mutation feature 30 (Signature 30) in others (Fig. 2A). In mutation feature 1, mutations of cytosine to thymine (C > T) were significantly increased. Mutation feature 3 showed no clear preference for specific mutation types, while mutation feature 30 displayed a higher frequency of cytosine to thymine (C > T) mutations (Fig. 2B) [12]. Mutation frequency and characteristics of HAS and AFPGC. A. Overview of the mutation characteristics and mutation frequency in HAS and non-hepatoid AFPGC. B. Mutation patterns of features 1, 3, and 30, as referenced from COSMIC Mutational Signatures version 3.4 Copy number variation and reproducibility analysis We performed somatic copy number variation (CNV) analysis on the 83 included samples. The results showed certain similarities in CNVs between the two groups of samples, specifically with notable amplifications in regions on chromosomes Chr1q, Chr5p, Chr6p, Chr7p, Chr8q, Chr15q, Chr17, Chr19, and Chr20, and significant deletions in regions on chromosomes Chr4, Chr5q, Chr9p, and Chr14. There were no specific differences between the two groups in terms of CNVs (Fig. 3A). Using the G-Score (Gistic Score), which calculates the overall somatic CNV status of the samples, we generated a CNV landscape for HAS and non-hepatoid AFPGC. The results showed that certain regions of chromosomes, particularly Chr19q12, exhibited significant amplification (Fig. 3B). In terms of therapeutic relevance, the important target gene CCNE1 is located in the Chr19q12 region, and the ERBB2 gene is located in the Chr17q12 region. To further clarify the expression of critical genes such as CCNE1 and ERBB2 in HAS and non-hepatoid AFPGC, we analyzed the amplification status of these genes. The results showed that the proportions of copy number alterations for c19orf12 and CCNE1 in HAS and non-hepatoid AFPGC were 62% and 63%, respectively. Specifically, the amplification rate of CCNE1 in HAS was 77.8% (42/54), in AFPeGC was 25.0% (4/16), and in AFPiGC was 43.8% (7/16). The amplification rate of ERBB2 in HAS was 50.0% (27/54), in AFPeGC was 12.5% (2/16), and in AFPiGC was 37.5% (6/16) (Fig. 3C). After performing Fisher’s exact test to compare the amplification rates of CCNE1 and ERBB2 between HAS and non-hepatoid AFPGC, we found that there was a significant difference in the amplification rate of CCNE1 between the two groups (P = 0.003), while no significant difference was observed in the amplification rate of ERBB2 (P = 0.0633). Copy number variation analysis between HAS and AFPGC. A. The copy number variation distribution heatmap of HAS and non-hepatoid AFPGC. B. Schematic diagram of the copy number variation regions in HAS and non-hepatoid AFPGC. C. Copy number variation landscape of selected genes in HAS and non-hepatoid AFPGC Correlation analysis of HAS and non-hepatoid AFPGC with gastric adenocarcinoma at the transcriptome level Transcriptomic profiling was performed on 50 samples, comprising 27 HAS samples and 23 non-hepatoid AFPGC samples. By integrating transcriptomic data from TCGA database, we identified 3232 differentially expressed genes (DEGs) in the HAS group compared to STAD, including 982 upregulated and 2250 downregulated genes. In the non-hepatoid AFPGC group, 3443 DEGs were detected, with 575 upregulated and 2868 downregulated genes (Fig. 4A-B). Comparative analysis revealed 1621 overlapping DEGs shared between HAS and non-hepatoid AFPGC, accounting for 50.2% (1621/3232) of the total DEGs in the HAS group. Among these overlapping DEGs, the majority exhibited consistent expression patterns in both groups. Specifically, 75.5% (1224/1621) were downregulated in both HAS and non-hepatoid AFPGC, while 12.5% (203/1621) were upregulated in both groups (Fig. 4C). Transcriptome similarity analysis between HAS and AFPGC. A-B. The volcano plot of differential gene expression between HAS, non-hepatoid AFPGC, and TCGA-STAD. C. The Venn diagram of overlapping differentially expressed genes between HAS and non-hepatoid AFPGC. D. PCA analysis of HAS, non-hepatoid AFPGC, and TCGA-STAD. E. Unsupervised clustering analysis of HAS, non-hepatoid AFPGC, and TCGA-STAD based on the top 2000 most highly expressed genes. F-G. GO and KEGG enrichment analysis based on differentially expressed genes between HAS and TCGA-STAD. H-I. GO and KEGG enrichment analysis based on differentially expressed genes between AFPGC and TCGA-STAD We then applied Principal Component Analysis (PCA) to perform dimensionality reduction and clustering of HAS, non-hepatoid AFPGC, and STAD samples. after incorporating RNA-seq data from the TCGA STAD cohort, both HAS and non-hepatoid AFPGC samples showed a clear clustering trend in the PCA plot, distinct from gastric adenocarcinoma (Fig. 4D). Further filtering of the RNA-seq data focused on the top 2000 genes with the highest expression levels in HAS and non-hepatoid AFPGC samples, while excluding samples from the TCGA-STAD cohort with low gene expression levels. Unsupervised clustering of the remaining samples based on these top 2000 genes revealed similar expression patterns in HAS and non-hepatoid AFPGC samples (Fig. 4E). Subsequently, we calculated the silhouette score to assess the coherence of the clusters, which further confirmed this conclusion (Supplementary Fig. 1). These results suggest that, at the transcriptomic level, there is a notable similarity between the HAS and non-hepatoid AFPGC groups. Subsequently, we conducted functional enrichment analysis. GO analysis revealed that HAS and non-hepatic-like AFPGC share several overlapping biological processes, including DNA-binding transcription activator activity, protein-DNA complex, and cytokine activity/cytokine receptor activity. Similarly, KEGG enrichment analysis showed that HAS and non-hepatic-like AFPGC also overlap in several signaling pathways, including Signal transduction, Signaling molecules and interactions, and the Immune system. These findings suggest potential common mechanisms between these two groups, highlighting the relevance of these pathways in the context of the disease (Fig. 4F-I). In addition, we compared the clinicopathological characteristics of HAS and non-hepatoid AFPGC (Table 1). The results revealed that there was a significant difference only in the Lauren classification, with no significant differences observed in other demographic and clinicopathological features, including gender, age, tumor location, tumor size, family history of tumors, pTNM staging, and various other aspects. This further confirms the high similarity between the two groups. Table 1Comparison of clinicopathological data between HAS and AFPGCCharacteristicTotal, nGroup, nχ²P-valueHASAFPGCGender1.6740.273 Male644420 Female19109Age at surgery (years)1.7630.211 <65594118 ≥65241311History of Tumor0.4690.603 Yes22139 No614120Tumor location4.6430.096 Upper third29218 Middle third1358 Lower third412813Tumour size (cm)0.3280.418 <8724626 ≥81183Lauren classificationa6.7720.032 Intestinal37289 Diffuse1147 Mixed281513Borrmann classificationa3.1860.094 I/II21183 III/IV543519pT stageb6.7610.063 pT1312 pT212102 pT3503515 pT418810pN stageb5.2590.263 pN01679 pN120164 pN2271710 pN3a1394 pN3b752pTNM stageb1.0760.557 I633 II332310 III442816Lymphovascular invasion0.5290.483 Yes533617 No301812Perineural invasiona0.0301.000 Yes573621 No23158HER22.8100.151 Amplification1091 No amplification684325aSome data missedbFisher exact test Comparison of clinicopathological data between HAS and AFPGC aSome data missed bFisher exact test FAT4 is a potential therapeutic target and biomarker After filtering the data, we found that FAT4 exhibited a relatively high mutation frequency in both groups, with an overall mutation rate of 14%. Due to the lack of relevant data for HAS and non-hepatoid AFPGC, we conducted a detailed analysis of this target in the TCGA-STAD dataset. FAT4 encoded a member of the protocadherin family, playing a role in regulating planar cell polarity. First, we divided the TCGA-STAD samples into FAT4 mutation and wild type and assessed the expression of FAT4. The results showed that mutations significantly downregulated FAT4 transcriptional levels (Fig. 5A). Survival analysis revealed that FAT4 overexpression was associated with significantly poorer survival in gastric cancer patients (Fig. 5B). Next, we evaluated the impact of FAT4 mutations on immune cell infiltration in gastric cancer patients. The results showed that FAT4 mutations led to a significant increase in the infiltration level of M1 tumor-associated macrophages (Fig. 5C). We also analyzed the expression of immune checkpoints, and found that in the FAT4 mutation group, the expression level of ITPRIPL1 was significantly decreased, suggesting that FAT4 mutations may be linked to immune therapy targeting ITPRIPL1 (Fig. 5D). Finally, we used the TIDE algorithm to assess the response of the two patient groups to immune checkpoint blockade (ICB) therapy. We found that patients in the FAT4 mutation group had a lower TIDE score, indicating a significantly better response to ICB therapy and a longer survival period (Fig. 5E). These findings provided a potential therapeutic target and molecular biomarker for gastric cancer patients, which could be further validated in HAS and non-hepatoid AFPGC patients in future studies, offering insights for precision treatment in these patient groups. FAT4 is a potential therapeutic target and molecular marker for gastric cancer. A. Expression differences of FAT4 between different groups in the TCGA database (G1: FAT4 mutation group, G2: FAT4 wild group). B. Kaplan-Meier survival curves for gastric cancer patients with high or low expression of FAT4. C. The correlation between FAT4 expression and immune infiltration in gastric cancer across different groups (G1: FAT4 mutation group, G2: FAT4 wild group). D. The correlation between FAT4 expression and immune checkpoint expression across different groups (G1: FAT4 mutation group, G2: FAT4 wild group). E. Evaluation of patient responses to immunotherapy across different groups based on TIDE scores (G1: FAT4 mutation group, G2: FAT4 wild group) Discussion We performed whole exome and transcriptome sequencing to delineate the molecular characteristics of HAS and non-hepatoid AFPGC. Our findings confirmed the genomic and transcriptomic similarities between the two, as well as their distinct differences when compared to conventional gastric adenocarcinoma. Furthermore, we identified FAT4 as a potential therapeutic target and molecular biomarker, providing a foundation for subsequent research. Our findings reveal that TP53 is the gene with the highest mutation frequency in both HAS and non-hepatoid AFPGC, with a significantly higher mutation rate in the HAS group compared to other genes. TP53 is one of the most commonly mutated genes in gastric cancer and various other tumors, with mutations present in approximately 47.6% of gastric cancer patients, playing a crucial role in the initiation and progression of gastric cancer [13, 14]. A study has shown that TP53 is highly expressed in both the hepatoid and non-hepatoid differentiation regions of HAS, suggesting that TP53 may play a role in the phenotypic transformation of HAS. The functional abnormalities of TP53 could potentially serve as the initiating factor for the bidirectional differentiation of different components in HAS [15]. In the AFPeGC group, we found that TP53 did not harbor missense mutations, yet several other common mutation types were still present. This finding is striking, but its potential biological implications remain to be further explored. However, the remaining TP53 mutations still affect the tumor biological function of gastric cancer. Nonsense and frameshift mutations in the TP53 gene lead to the production of truncated, non-functional p53 proteins, which are unable to effectively regulate cell cycle arrest and apoptosis. Recent studies have shown significant differences in the effectiveness of nivolumab treatment in advanced gastric cancer patients depending on the type of TP53 mutation. Research indicates that patients carrying TP53 frameshift mutations have a significantly higher objective response rate to nivolumab compared to those with other mutation types, suggesting that frameshift mutations may retain certain exploitable functional characteristics [16]. This finding emphasizes the importance of analyzing the specific mutation types in the TP53 gene, as they not only affect treatment outcomes but also provide crucial information for developing personalized treatment strategies. Furthermore, for gastric cancer patients with TP53 frameshift mutations, studies have found that LSD1 protein may become a key target for targeted therapy. This discovery further reveals the biological consequences of related gene mutations. Research suggests that frameshift mutations in TP53, especially those lacking nuclear localization sequences, significantly increase the expression of histone demethylase LSD1—an enzyme that drives tumor formation. Following the inhibition of LSD1 with GSK690, researchers observed significant growth inhibition in gastric cancer cells harboring these mutations, offering new insights into targeted therapy for this specific mutation subtype [17]. Furthermore, the overall frequency of ERBB2 amplification in both HAS and non-hepatoid AFPGC was 41%, with the rate reaching 50% in the HAS group. This suggested that anti-HER2 therapy may be effective in both HAS and non-hepatoid AFPGC. In 2024, a study reported that five patients with locally advanced HAS who received Herceptin treatment showed no deaths or recurrence/metastasis during a three-year follow-up period. Antibody-drug conjugates (ADC) are targeted biologic agents composed of an antibody, a linker, and a cytotoxic drug. Recent studies have indicated that ADC drugs targeting HER2 demonstrate similar efficacy in gastric cancer patients with HER2 IHC scores of 2 + and 3+ [18–20]. Given the significant heterogeneity in HER2 expression status in HAS and non-hepatoid AFPGC, ADCs targeting HER2 may lead to breakthroughs in the treatment of such tumors [21]. FAT4, a member of the cadherin-related family, acts as a tumor suppressor and plays an important role in various cancer types. It is involved in the regulation of multiple signaling pathways, including the Hippo and Wnt/β-catenin pathways, highlighting its crucial role in regulating cell proliferation, migration, and invasion across different cancer types. In gastric cancer, FAT4 exerts its tumor suppressive effect by modulating Wnt/β-catenin signaling, which is essential for maintaining cellular homeostasis and inhibiting tumor progression. Additionally, the role of FAT4 in gastric cancer is linked to its epigenetic inactivation, which promotes carcinogenesis in the gastric region, making the tissue more prone to malignancy [22–24]. Additionally, FAT4 mutations have been associated with tumor mutational burden and favorable prognosis in gastric cancer, suggesting that FAT4 could serve as a useful biomarker for immunotherapy in these patients, which is consistent with our research findings [25]. In addition, our study also found that LRP1B is one of the genes with the highest mutation frequency in HAS and AFPGC aside from TP53. LRP1B encodes a member of the low-density lipoprotein (LDL) receptor family, and it has been reported that this gene is often disrupted in various types of cancer. LRP1B has been confirmed as a potential tumor suppressor gene in gastric cancer, and its mutation frequency is closely associated with tumor mutation burden (TMB) and prognosis. Its mutations have been identified as an independent risk factor affecting disease-free survival (DFS) in patients. Studies show that LRP1B expression levels are correlated with increased numbers of tumor-infiltrating immune cells (such as CD4 + and CD8 + T cells), suggesting that this gene plays an important role in regulating the immune function of the gastric cancer microenvironment [26]. In another rare subtype of gastric cancer, gastric neuroendocrine carcinoma and mixed adenoneuroendocrine carcinoma, it was also found that LRP1B could serve as a prognostic marker for patients [27]. It is worth noting that a study in 2025 found that LRP1B could serve as a molecular marker for the prognosis of AFPGC. After LRP1B mutation, patient survival was significantly reduced [28]. These results highlight the importance of further exploring the molecular regulatory mechanisms and potential therapeutic value of LRP1B as a critical target in GHA and AFPGC. Above all, through correlation analysis, this study found a significant association between certain HAS samples and non-hepatoid AFPGC. An ideal molecular classification method should ensure high internal consistency within different subtypes, while maintaining low genomic similarity between subtypes. The results of this study suggest that the method of classifying HAS and non-hepatoid AFPGC based on the hepatoid differentiation region does not accurately distinguish disease subtypes with different molecular features. At the genomic level, HAS and non-hepatoid AFPGC share certain similarities and may belong to the same type of tumor. Furthermore, at the transcriptomic level, non-hepatoid AFPGC and HAS share most of the differentially expressed genes, and PCA analysis based on the top 2000 genes by expression abundance also suggest a close correlation between the transcriptomic characteristics of HAS and non-hepatoid AFPGC. Therefore, this study proposes that the method of classifying HAS and non-hepatoid AFPGC based on the presence of hepatoid morphology in pathological tissues also has limitations at the molecular level, and molecular subtyping based on multi-omics warrants further exploration. Of course, there are some limitations in this study. First, although a relatively large sample size was included, all samples were from a single center. Sequencing data from a single center have certain advantages, as the diagnostic standards for HAS can vary between different centers. Single-center sampling will contribute to improving diagnostic accuracy, especially in the case of the rare occurrence of hepatoid adenocarcinoma, where there are clear and unified diagnostic criteria for hepatoid differentiation. Meanwhile, we will also explore the feasibility of conducting prospective multicenter studies for validation in the future to further improve the accuracy of our research. Secondly, further in-depth work on molecular subtyping and target validation is needed. Conclusion HAS and non-hepatoid AFPGC exhibit significant molecular similarities and show notable differences from conventional gastric adenocarcinoma. Currently, diagnosing HAS solely based on the hepatoid differentiation region may have several limitations. Furthermore, since HAS and non-hepatoid AFPGC are likely to be highly similar, it may be worth considering grouping them into a larger category for further in-depth research. Supplementary Information Below is the link to the electronic supplementary material. Supplementary Material 1 Acknowledgements Thanks for the support and assistance by Figdraw. (https://www.figdraw.com/#/). Author contributions Liqiao Chen performed data visualization, statistical analysis and writing, review and revision of the paper; Xuesong Yang and Peiyu Zhu contributed to sample collection; Han Bao, Yan Wu, Ke Ji, Ji Zhang, Xiaojiang Wu, Kai Zhou, Jieli Xu, Jiatian Tang, Zhongwu Li contributed to data analysis method consultation and data quality control; Anqiang Wang and Zhaode Bu contributed to study design and funding support. Funding This study was supported in Beijing Physician Scientist Training Project, BJPSTP (BJPSTP-2025-11), Science Foundation of Peking University Cancer Hospital (PY202329), Clinical Medicine Plus X-Young Scholars Project, Peking University, the Fundamental Research Funds for the Central Universities (PKU2024LCXQ022), Clinical Research Fund for Distinguished Young Scholars of Peking University Cancer Hospital (QNJJ202306), Clinical Research Fund for Distinguished Young Scholars of Peking University Cancer Hospital (QNJJ202415), The National Nature Science Foundation of China (No.82173151), National Key Research and Development Program of China (No.2023YFF1204700, branch No.2023YFF1204702). Data availability All patient records and information were anonymized and deidentified prior to analysis. Some of the data analyzed in this study were derived from the previous research work of our team [15]. The datasets analyzed during the current study are not publicly available, but are available from the corresponding author on reasonable request. Declarations Ethics approval and consent to participate The study of human samples was performed according to the Declaration of Helsinki and Good Clinical Practice. The protocol of this research and the license for usage of clinical data were given by the Ethics Committee of Peking University Cancer Hospital and Institute (code 2024KT174). All patients signed informed consent forms. Competing interests The authors declare no competing interests. Footnotes Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Liqiao Chen, Xuesong Yang and Peiyu Zhu contributed equally to this work. Contributor Information Anqiang Wang, Email: drwang726@hsc.pku.edu.cn. Zhaode Bu, Email: buzd@cjcrcn.org. References Associated Data Supplementary Materials Supplementary Material 1 Data Availability Statement All patient records and information were anonymized and deidentified prior to analysis. Some of the data analyzed in this study were derived from the previous research work of our team [15]. The datasets analyzed during the current study are not publicly available, but are available from the corresponding author on reasonable request.