Clinicopathological characteristics

Clinical characteristics of 91 AFP-producing GC patients enrolled in this study are summarized in Table 1 (Supplementary). AFPGC is more common in the male population, with 73 males and 18 females, for a male-to-female ratio of 4.05:1. The age of this cohort ranged from 27 to 82 years. All 91 patients had elevated serum AFP levels preoperation, ranging from 16.17 to 60,500.00 ng/ml. At the tumor site, the AFPGC were mainly located in the cardia (41/91) and the gastric antrum (31/91), followed by the gastric body (14/91) and the esophagogastric junction (5/91). According to the depth of invasion, five cases invaded the mucosa and submucosa (T1), ten invaded the muscularis propria (T2), 70 invaded the subserosa (T3), four invaded the serosal layer but did not invade adjacent organs (T4a), and two invaded the serosal layer and invaded adjacent organs (T4b). The maximum diameter of the tumors ranged from 2.00 cm to 19.00 cm. Our cohort was classified into four histologic subtypes: (1) Common intestinal adenocarcinoma type (COM): This subtype accounted for 35 cases (38.46%) and including poorly, moderately, and well-differentiated types, exhibiting tubular and papillary structures (Fig. 1a, b). (2) Enteroblastic type (ENT): Consisting of 12 cases (13.19%) with clear cytoplasm composed of cuboidal or columnar cells resembling primitive intestinal cells, and forming tubular, papillary, or adenoid structures. (Fig. 1c, d). (3) Hepatoid type (HPT): Observed in 40 cases (43.95%), characterized by cells with clear cytoplasm or eosinophilic granules arranged in medullary or cord-like structures with vesicular nuclei, prominent nucleoli, reminiscent of liver tissue organization (Fig. 1e, f). (4) Yolk sac tumor type (YST): Identified in 4 cases (4.40%), exhibiting a network-like arrangement of interconnected cytoplasm with communicating cavities and cysts resembling a labyrinth structure. (Fig. 1g, h). In addition, we also found a small proportion of other types of adenocarcinoma components in 91 cases of AFPGC, with two cases exhibited mixed ENT and HPT components, three displayed mucinous adenocarcinoma components, and three combined with neuroendocrine carcinoma. All mixed components accounted for less than 10%. No statistical differences were observed in gender, age, serum AFP level, tumor location, depth of invasion, pTNM stage, tumor size, lymph node metastasis, nerve invasion, vascular invasion, and differentiation degree among the histologic subtypes (Table 1).

Fig. 1

HE staining of AFPGC four histologic subtypes. a, b Common intestinal adenocarcinoma type (COM). c, d Enteroblastic type (ENT). e–f: Hepatoid type (HPT). g, h Yolk sac tumor type (YST). Red arrow: Glom-like bodies (Shiller-Duval bodies). Box: to be enlarged. Bar = 100 μm

Expression of AFPGC immunohistochemical markers

We conducted IHC in this cohort to examine the expression of markers associated with the four subtypes. The IHC results revealed that the embryonic proteins of AFP (Fig. 2a, d), GPC3 (Fig. 2b, e), and SALL4 (Fig. 2c, f) expressed in AFPGC were 51.65%, 64.84%, and 58.24%, respectively. We found that the expression of the above three embryonic proteins in HPT was significantly higher than those in the other three subtypes (P<0.05). Moreover, there was a significant correlation between their expression and patients' serum AFP level (P<0.05). The expression of AFP was significantly positively associated with vascular invasion (P<0.05). The intestinal epithelial differentiation markers CDX-2 (Fig. 2g) and CD10 (Fig. 2h) were expressed in 85 (93.41%) and 81 (89.01%) cases of AFPGC patients, respectively. Meantime, we found CEA was expressed in 84 (92.31%) cases which were significantly correlated with lymph node metastasis (P<0.05).

Fig. 2

Immunohistochemical phenotyping of AFPGC. a, b, c AFP, GPC3, and SALL4 positive expressed in ENT. d, e, f AFP, GPC3, and SALL4 positive expressed in HPT. g, h CDX2, CD10 positive expressed in COM. i HNF-1β showed nuclear positive in YST. Bar = 100 μm

In addition, CLDN6 was expressed in 88 (96.70%) cases, while OCT3/4 was only expressed in 3 (3%) cases. Notably, OCT3/4 showed high expression specifically in ENT. We also found ATBF1 and HNF-1β (Figure 2i) expressed in 22 (24.18%) and 78 (85.71%) patients of AFPGC, respectively. Notably, the negative rate of ATBF1 was higher in the HPT subtype compared to other subtypes (P <0.05). The prevalence of mutant p53 was observed in 75 cases (82.42%), while p53 wild-type was in 16 cases (17.58%). Patients aged ≥60 years with mutant p53 exhibited significantly higher compared to those with p53 wild-type. Furthermore, 78 cases (85.71%) displayed proficient MMR (pMMR) and 13 (14.29%) as deficient MMR (dMMR, Table 1, Supplementary). Two cases exhibiting HER-2 (3+), and eight with HER-2 (2+) expression, which were further examined by FISH. The results indicated that five out of eight HER-2 (2+) cases with HER-2 amplification (Supplementary Figure 1). Only one AFPGC with diffuse EBER-ISH labeling was considered EBV-positive GC.

Mutations and signatures of AFPGC

Next, we performed NGS methods to investigate the mutational characteristics of 91 AFPGC patients in this study, with a total of 1271 variants successfully detected, including 744 (58.54%) SNVs, 510 (40.13%) CNVs, and 17 (1.34%) SVs, with a mean number of variants of 14. The vast majority of harbored at least one mutation (89/91, 97.8%), and only two cases tested negative. The most frequently altered genes in AFPGC were TP53 (84%), MYC (59%), IRS2 (38%), PCK1 (38%), and GNAS (38%). Other mutation frequencies in more than 10% were LRP1B (34%), CCNE1 (27%), CEBPA (22%), MLL3 (16%), EGFR (12%) and ERBB2 (HER-2, 12%). In addition, TMB values were also analyzed, with a median of 5.76 muts/Mb (Fig. 3a).

Fig. 3

Mutations signatures and potential therapeutic targets of AFPGC. a: TMB and Top 20 somatic mutations of 91 AFPGC tissues. b The TP53 mutation rate in AFPGC and CGA. *P < 0.05. c Comparative analysis of the approved target genes of drugs in the OncoKB website. d: Mutually exclusive co-mutations in 21 target genes. *P < 0.05; ▪P < 0.1

Then, we analyzed different mut-types, starting with the SNVs analysis showed TP53 mutation was the most frequent (76/91, 83.52%), followed by LRP1B (31/91, 34.07%), MLL3 (15/91, 16.48%), PTPRD (11/91, 12.09%), EPHA3 (9/91, 9.89%), ATM (9/91, 9.89%), and APC (7/91, 7.69%) (Fig. 3a). The frequencies of TP53 mutation in AFPGC were significantly higher than that in common gastric adenocarcinoma (CGA) from the Cancer Genome Atlas database [4] with rates of 83.52% (76/91) versus 46.77% (138/295, P < 0.05, Fig. 3b). Second, we found that CNVs were observed in 80 patients. However, most CNVs involve more than two genes occurring simultaneously (68/80, 85%). The most frequently gene amplification included GNAS (32/91, 35.16%), PCK1 (31/91, 34.07%), IRS2 (31/91, 34.07%), CCNE1 (24/91, 26.37%), PTEN (24/91, 26.37%), CEBPA (19/91, 20.88%), TOP1 (17/91, 18.68%), SRC (11/91, 12.09%), BCL2L1 (10/91, 10.99%), VEGFA (10/91, 10.99%), EGFR (8/91, 8.79%) and ERBB2 (7/91, 7.69%). Corresponding to amplification, the most common gene deletions were MYC (53/91, 58.24%) and SOX2 (25/91, 27.47%) (Fig. 3a). Third, SV mutation was detected in 16 patients (≥ 1 SVs), with one patient having SVs in both the BRCA1-LOC100507425 and MET genes. The rest have one gene SV involved the following genes: AKAP9-BRAF, IRF2BP1-MYC, C14orf177-GNAS, PDGFRA rearrangement, MAGT1-ATRX, EPHA6-ROS1, FANCM-GRIA4, PREX2-ROS1, FGFR3-TACC3, EPHA6-ROS1, EGFR rearrangement, FGFR2-UBE2D1, ERC1-BRAF, RPS6KA4-BRCA1, AIM1-FGFR3.

Potential therapeutic targets

To further explore the potential therapeutic targets of AFPGC, we conducted a comparative analysis of the approved target genes of drugs in the OncoKB (OncoKB™—MSK's Precision Oncology Knowledge Base) website. We found the cumulative incidence of 21 genes was 59.34%, of which CCNE1, ERBB2, and EGFR were the highest (Fig. 3c). The genes associated with targeted drugs were often co-occurrence (Fig. 3d).

Therewith, we compared the differences in molecular characteristics between different histologic subtypes and between subgroups with or without metastasis. The results showed that the frequencies of CREBBP and FGF14 in the yolk sac type were significantly higher than those in other subtypes. We also found that the incidence of CEBPA mutation was more likely to occur in patients with metastasis, whereas MLL3 mutation was more likely to occur in patients without metastasis.

Pathway enrichment analysis

To further clarify whether these genomic alterations lead to differences in cancer-related signaling pathways, we performed pathway enrichment analysis using the KEGG database (Fig. 4). The results revealed several signaling pathways were enriched, including disease signal transduction, receptor tyrosine kinase signaling, intracellular second messenger signaling, PIP3-Akt signaling pathway, MAPK family signaling cascade, nuclear receptor signaling, FLT3 signaling, MAPK1/MAPK3 signaling, PI3KI-AKT Signaling as well as ESR-mediated signaling. We found that these signaling pathways were mainly enriched in pathways related to cell proliferation, invasion, and metastasis. However, no significant differences between histologic subtypes were observed (Fig. 4b).

Fig. 4

KEGG pathway enrichment analysis of AFPGC. a AFPGC patients. b Patients of COM, ENT, HPT and YST subtypes

Survival analyses of AFPGC

Our next step aims to conduct a ROC curve to identify the optimal cut-off points of serum AFP values for survival analysis. As shown, 510 ng/ml was the optimal cut-off, the sensitivity was 52.8%, the specificity was 72.7%, and the area under the curve was 0.628 (P = 0.039) (Supplementary Fig. 2). Therefore, we chose “cut-off = 500 ng/ml” to perform the follow-up analysis. The results indicate that patients with serum AFP ≥ e500 ng/ml had poor prognosis. The analysis of OS included 91 cases, a univariate Cox analysis based on OS as clinical outcomes revealed that clinical stage, venous invasion, nerve invasion, distant metastasis, serum AFP, and LRP1B mutation were prognostic factors for AFPGC (P < 0.05), Furthermore, a multivariate Cox proportional hazards regression model revealed that clinical stage and nerve invasion were still independent prognostic factors (P < 0.05, Fig. 5, Supplementary Fig. 3, Table 2). The DFS analysis included 82 patients who underwent a radical operation. A univariate Cox analysis based on PFS as clinical outcomes revealed that tumor size, lymph node-positive, clinical stage, venous invasion, nerve invasion, serum AFP, ARID1A mutation were prognostic factors for AFPGC (P < 0.05). In the multivariate Cox proportional hazards regression model, clinical stage, serum AFP and nerve invasion were still independent prognostic factors (P < 0.05, Fig. 5, Supplementary Fig. 3, Table 3).

Fig. 5

Kaplan–Meier curves for overall survival (OS) and Disease-Free Survival (DFS). a-g: Patients with high clinical stage, nerve invasion, serum AFP ≥ 500 ng/ml, and LRP1B mutation had poor OS. e–h. Patients with high clinical stage, nerve invasion, serum AFP ≥ 500 ng/ml, and ARID1A mutation had poor DFS

To further explore the influence of molecular characteristics on prognostic, we analyzed the relationship between different gene mutations and survival. We found that the LRP1B was a risk factor for OS, while ARID1A acted as a protective factor for DFS. Subsequently, we analyzed the effects of mutations on OS and DFS with mutation frequencies greater than 8%. The results indicated that patients with the LRP1B mutation experienced significantly decreased OS (P = 0.016, HR = 2.18, 95% CI: 1.04–4.56), while patients carried ARID1A mutation with significantly increased DFS (P = 0.021).

Molecular typing analysis in AFPGC

Based on the molecular characteristics of TCGA typing of GC [4], we typed the patients in our cohort (Fig. 6). A total of 64 patients (70.33%) were classified as CIN type, primarily occurring in cardia and antrum, and most of them were diffuse type in Lauren’s classification. The mutation frequency of TP53 in CIN type was as high as 93.75%, and chromosome amplification was found in cell cycle regulatory genes (CCNE1, CCND1, CDK6) and some RTKs genes (EGFR, ERBB2, FGFR2, MET, RAS). 13 patients (14.29%) were classified as MSI type, the predilection location of which were gastric antrum and body, and most of them belong to mixed type in Lauren’s classification. The typical molecular events were PIK3CA, ERBB2, and EGFR mutation, along with some gene chromosome amplification. Only one patient (1.1%) was EBV-positive GC, which also occurred in the gastric antrum and belonged to the diffuse type in Lauren’s classification. Notably, there were no CDKN2A mutation, compared to the TCGA cohort. 13 patients (14.29%) were classified as GS type, the frequency of CDH1 (15.38%) mutation was high in this type, and TP53 mutation frequency was lower than CIN type. Finally, we performed a survival analysis for different molecular subtypes, and the results showed that the prognosis of GS was the worst (Fig. 6).

Fig. 6

The mutations landscape of AFPGC