Abstract

   Alpha‐fetoprotein‐producing gastric cancer (AFPGC) is a rare but highly
   aggressive subtype of gastric cancer. Patients with AFPGC are at high
   risk of liver metastasis, and the tumor microenvironment (TME) is
   complex. A multicenter retrospective study is conducted from January
   2011 to December 2021 and included 317 AFPGC patients. Using a
   multivariable logistic regression model, a nomogram for predicting
   liver metastasis is built. By combining AFP and the
   neutrophil–lymphocyte ratio (NLR), we developed a novel and easily
   applicable predictive indicator, termed ANLiM score, for liver
   metastasis in AFPGC. An integrated multi‐omics analysis, including
   whole‐exome sequencing and proteomic analysis, is conducted and
   revealed an immunosuppressive TME in AFPGC with liver metastasis.
   Single‐cell RNA sequencing and multiplex immunofluorescence identified
   the potential roles of tumor‐associated neutrophils and tertiary
   lymphoid structures in shaping the immune microenvironment. These
   findings are validated in a real‐world cohort receiving anti‐programmed
   cell death 1 (anti‐PD‐1) therapy, which showed concordant
   effectiveness. In addition, the ANLiM score is also identified as a
   promising biomarker for predicting immunotherapy efficacy. Overall, a
   blood biomarker‐based predictive indicator is developed for liver
   metastasis and immunotherapy response in AFPGC. The findings on immune
   microenvironmental alterations for AFPGC with liver metastasis provide
   new insights for optimizing immunotherapy strategies.

   Keywords: alpha‐fetoprotein‐producing gastric cancer, immunotherapy,
   liver metastasis, tumor microenvironment
     __________________________________________________________________

   Alpha‐fetoprotein‐producing gastric cancer (AFPGC), a rare and
   aggressive subtype of gastric cancer, presents a high risk of liver
   metastasis. This study introduces the ANLiM score, a blood
   biomarker‐based predictive indicator for assessing liver metastasis and
   immunotherapy response. Multi‐omics analyses reveal critical immune
   microenvironmental alterations in AFPGC with liver metastasis, offering
   new insights for optimizing immunotherapy strategies.

   graphic file with name ADVS-12-e03499-g004.jpg

1. Introduction

   Gastric cancer (GC) is one of the leading causes of cancer‐related
   death worldwide.^[ [76]^1 ^] Alpha‐fetoprotein‐producing gastric cancer
   (AFPGC) is a rare and distinct subtype of GC that is characterized by
   the production of AFP and accounts for approximately 1.8% to 7.9% of
   all GCs.^[ [77]^2 , [78]^3 , [79]^4 ^] AFPGC is known to be highly
   aggressive, with a pronounced tendency for lymph node and liver
   metastasis.^[ [80]^5 ^] Previous studies have indicated that liver
   metastasis is an independent prognostic factor for patients with
   AFPGC,^[ [81]^3 ^] with an incidence of 40–60%,^[ [82]^2 , [83]^3 ,
   [84]^6 ^] which markedly exceeds the rate of 9.9–18.7% observed in
   conventional GC.^[ [85]^7 , [86]^8 ^] Currently, the treatment for
   AFPGC follows the general guidelines for GC, exhibiting a dismal
   prognosis.^[ [87]^4 ^] Given that liver metastasis is a hallmark
   feature of AFPGC, elucidating its underlying mechanisms is essential
   and will be pivotal for optimizing future treatment strategies and
   improving the prognosis.

   Cancer has long been considered a genetic disease, with the
   accumulation of gene mutations being the basis for cancer initiation
   and metastasis.^[ [88]^9 , [89]^10 ^] However, it is becoming
   increasingly apparent that genetics alone may not adequately explain
   the mechanisms of cancer metastasis, as few metastasis‐specific
   mutations have been definitively identified.^[ [90]^11 ^] Thus, cancer
   geneticists are increasingly focusing on nongenetic variations to seek
   explanations for metastasis.^[ [91]^12 ^] In the past decade, the tumor
   microenvironment (TME), a key non‐genetic factor, has been reported to
   play a crucial role in the metastasis of malignant tumors through the
   spatiotemporal dynamics of its various components (e.g., immune cells
   and the extracellular matrix).^[ [92]^12 , [93]^13 ^] Thus, identifying
   the early shifts associated with metastasis within the TME has the
   potential to revolutionize the early diagnosis and treatment of cancer
   metastasis, offering new strategies and hope.

   Previous studies have suggested that changes in the TME within the
   primary tumor site encompass early events associated with tumor
   metastasis.^[ [94]^13 ^] For instance, alterations in the local
   microenvironment of primary tumors, such as extracellular matrix
   remodeling and altered immune cell composition, establish conditions
   conducive to tumor dissemination.^[ [95]^13 ^] Additionally, during the
   initial stages of metastasis, the TME within primary tumors can induce
   systemic immune and inflammatory responses, thereby promoting tumor
   spread and colonization at distant sites.^[ [96]^13 , [97]^14 ^]
   Recently, the advent of high‐throughput technologies, including
   proteogenomics and single‐cell transcriptomics, has propelled cancer
   research from the single‐genome level to a comprehensive multi‐omics
   approach, offering new insights into the complexity of the TME.^[
   [98]^15 , [99]^16 ^] Therefore, utilizing multi‐omics approaches to
   characterize the features of TME changes in primary AFPGC lesions may
   aid in developing new strategies for the early identification and
   treatment of liver metastasis in AFPGC.

   In this study, we aimed to investigate the clinicopathological
   characteristics of AFPGC and developed a clinical prediction model for
   liver metastasis. Additionally, we compared the proteogenomic and
   single‐cell transcriptomics profiles of primary lesions of
   non‐metastatic AFPGC and those with liver metastasis to elucidate
   alterations in the TME. Our findings were validated in a real‐world
   clinical cohort. The insights gained from this study may provide a new
   theoretical foundation for the early detection and therapeutic
   intervention of liver metastasis in AFPGC.

2. Result

2.1. Liver Metastasis was Identified as a Key Clinical Feature of AFPGC, with
ANLiM Score Serving as a Predictive Indicator

   To better elucidate the clinicopathological characteristics of AFPGC,
   we included patients with AFPGC from five hospitals in China to form a
   multicenter cohort (N = 317, Figure [100] 1A; Table [101]S2, Supporting
   Information). Figure [102]1B illustrates the distribution of various
   clinicopathological characteristics in the multicenter cohort (Table
   [103]S3, Supporting Information). Additionally, we matched a non‐AFPGC
   cohort (serum AFP levels at diagnosis <20 ng mL^−1) at a ratio of
   ≈1:2.5, based on the admission time. The comparison revealed that AFPGC
   was characterized by a later stage at diagnosis, a higher proportion of
   male patients, and poorer differentiation (Table [104]S3, Supporting
   Information). The overall survival (OS) of the AFPGC cohort was
   significantly worse than that of the non‐AFPGC cohort (Hazard Ratio
   [HR] = 2.43, 95% Confidence Interval [CI] = [1.96, 3.02], p < 0.001,
   Figure [105]1C). Figure [106]1D–F illustrated that within each
   stage‐specific subgroup, the prognosis of AFPGC was also relatively
   poorer than that of non‐AFPGC.

Figure 1.

   Figure 1
   [107]Open in a new tab

   Clinicopathological characteristics of AFPGC cohort and the
   construction of ANLiM as a predictive biomarker for liver metastasis.
   A) Screening workflow for the multicenter AFPGC Cohort. B) Distribution
   of clinicopathological characteristics AFPGC cohort. The survival time
   and survival status of each patient are displayed above the figure.
   Comparison of overall survival between AFPGC cohort and non‐AFPGC
   cohort C), as well as across various TNM stages D–F). G) The incidence
   of liver metastasis in AFPGC and non‐AFPGC cohorts. H) Comparison of
   metastatic pattern between two cohorts. I) Comparison of overall
   survival between two cohorts. J) A nomogram for predicting synchronous
   liver metastasis of AFPGC. K) Development and calculation of the ANLiM
   Score. (e.g., serum AFP = 320 ng/mL, NLR = 4.2, yielding ANLiM = 2;
   serum AFP = 98 ng/mL, NLR = 2.8, yielding ANLiM = 1) L) Relationship
   between the ANLiM score and the incidence of synchronous liver
   metastasis. M) Comparison of the predictive ability of ANLiM score with
   AFP and NLR for Liver Metastasis in AFPGC. N) Survival curves of AFPGC
   stratified by different ANLiM scores. NA, not available. ^****
   p < 0.0001.

   Additionally, patients with AFPGC exhibited a significantly higher
   incidence of liver metastasis than did those without AFPGC (41.6% vs
   11.3%, p < 0.001, Figure [108]1G). Among the initial metastases, liver
   metastasis constituted 72.5% of cases in AFPGC, which was markedly
   higher than the 30.1% observed in non‐AFPGC cases (Figure [109]1H).
   Moreover, among patients diagnosed with liver metastasis, those with
   AFPGC exhibited worse OS than of those with non‐AFPGC (HR = 2.20, 95%
   CI = [1.55, 3.13], p < 0.001, Figure [110]1I). These results indicated
   that liver metastasis is a prominent clinical feature of AFPGC. Early
   identification of liver metastasis and prompt intervention may
   significantly improve the prognosis of patients with AFPGC. Currently,
   no clinical prediction model exists for liver metastasis in AFPGC.
   Based on the univariate analysis (Table [111]S4, Supporting
   Information), we developed a nomogram based on clinically feasible
   variables to predict synchronous liver metastasis (liver metastasis
   occurring at the time of diagnosis or within six months thereafter)
   (Figure [112]1J; Figure [113]S1, Supporting Information). To facilitate
   clinical application, we extracted the two most significant variables
   from the predictive model, AFP and NLR (neutrophil–lymphocyte ratio),
   to create a novel index termed the ANLiM (AFP and NLR in Liver
   Metastasis of AFPGC) (Figure [114]1K). A score of 0 is given if AFP is
   less than or equal to 136.4 ng mL^−1 and NLR is less than or equal to
   2.14. If AFP is less than or equal to 136.4 ng mL^−1 but NLR is greater
   than 2.14, the score is 1. When AFP exceeds 136.4 ng mL^−1 and NLR is
   less than or equal to 2.14, the score is also 1. Finally, a score of 2
   is assigned if both AFP and NLR are greater than their respective
   thresholds of 136.4 ng mL^−1 and 2.14. When the ANLiM score was 0, the
   likelihood of synchronous liver metastasis was 6.8%, which increased to
   31.8% and 54.1% when the ANLiM scores were 1 and 2, respectively
   (Figure [115]1L). The predictive performance of the ANLiM score for
   synchronous liver metastasis was significantly superior to that of AFP
   or NLR alone (Figure [116]1M). The predictive value of the ANLiM score
   for synchronous liver metastasis was further validated in the
   validation cohort (Table [117]S5; Figure [118]S2, Supporting
   Information). Additionally, a higher ANLiM score was associated with
   poorer OS (p = 0.004, Figure [119]1N).

2.2. Characterization of Multi‐Omics Features Related to Liver Metastasis in
AFPGC and Key Findings on the Immunosuppressive Microenvironment

   Liver metastasis is a critical event in AFPGC progression; however, the
   underlying molecular mechanisms remain unclear. We leveraged specimens
   from an AFPGC cohort that had been collected over the past decade for
   genomic sequencing and proteomic analyses (Figure [120] 2A). Based on
   the presence or absence of liver metastasis, the AFPGC genomic cohort
   was divided into a liver metastasis group (AFPGC LM[+], N = 20) and a
   non‐tumor‐associated neutrophils liver metastasis group (AFPGC LM[−], N
   = 38). We also included a cohort of common GC with liver metastasis
   from Memorial Sloan Kettering Cancer Center (MSKCC) (MSKCC‐GC LM[+], N
   = 89). Figure [121]2B illustrates the landscape of the somatic
   mutations and copy number variations in several cancer‐related
   signaling pathways as well as in driver genes. The alteration
   frequencies of most cancer‐related genes were not significantly
   different among the three groups (Figure [122]2C; Table [123]S6,
   Supporting Information). However, CCNE1 alteration (primarily gene
   amplification) was significantly more frequent in the AFPGC LM (+)
   group. Chromosomal instability (CIN), to some extent, reflects the
   invasive ability of tumors. We found that the CIN score, measured using
   weighted genome integrity index (wGII, details in Supporting Methods),
   was significantly elevated in the AFPGC LM(−) group (Figure [124]2D).
   Additionally, the AFPGC LM(−) group exhibited a higher proportion of
   polyclonality (Figure [125]2E). Notably, higher ANLiM scores were
   significantly correlated with higher wGII values (Figure [126]2F) and a
   greater proportion of polyclonality (Figure [127]2G). Conversely, no
   significant differences were observed in tumor mutation burden (TMB)
   (Figure [128]2H) and tumor neoantigen burden (TNB) (Figure [129]2I)
   between the two groups.

Figure 2.

   Figure 2
   [130]Open in a new tab

   Characterization of molecular features and microenvironmental changes
   associated with liver metastasis in AFPGC. A) Schematic workflow for
   genomic sequencing and proteomic analysis. B) The landscape of gene
   alterations among AFPGC LM (−), AFPGC LM (+), and MSKCC‐GC LM (+)
   groups. The middle panel displays somatic mutations across genes (rows)
   and tumor samples (columns), while the left panel annotates
   cancer‐related signaling pathways for these genes. The bottom panel
   presents different clinicopathological features. C) Comparison of gene
   alteration frequencies among different groups. D,E) Comparison of wGII
   index D) and polyclonality E) between two groups. F,G) Comparison of
   wGII index (F) and polyclonality G) between high ANLiM score and low
   ANLiM score groups. H,I) Comparison of TMB H) and TNB I) between two
   groups. J) PCA analysis of AFPGC tumor samples and adjacent normal
   tissues. K) Volcano plot of DEPs between two groups. L) Pathway
   enrichment analyses of DEPs. M) Heatmap of DEPs associated with immune
   process or metastasis in AFPGC LM(−) and AFPGC LM(+) groups. N,O)
   Comparison of ImmuneScores N) and TLS scores O) between AFPGC LM(−) and
   AFPGC LM(+) groups. P,Q) Comparison of ssGSEA enrichment scores for
   STING P) and IFNG Q) pathways. LM, liver metastasis; NAT, normal
   adjacent tissue; FFT, fresh frozen tissue; FFPE, formalin‐fixed
   paraffin‐embedded tissue; LM(−), the absence of liver metastasis;
   LM(+), the presence of liver metastasis; RTK, Receptor Tyrosine Kinase;
   wGII, weighted genomic instability index; PCA, principal component
   analysis; ssGSEA, single‐sample gene set enrichment analysis; NA, not
   available. ^* p < 0.05; ^** p < 0.01; ns, no significance.

   Currently, no proteomics studies have specifically addressed AFPGC. In
   this study, we performed proteomics analyses of 52 samples (comprising
   cancerous and adjacent noncancerous tissues) from 26 patients with
   AFPGC (Figure [131]S3, Supporting Information). Using PCA, we clearly
   distinguished between cancerous and adjacent non‐cancerous AFPGC
   samples (Figure [132]2J). We compared the differences in protein
   expression levels between AFPGC LM(+) and LM(−) groups, identifying 518
   differentially expressed proteins (DEPs) (Figure [133]2K; Table
   [134]S7, Supporting Information). Pathway enrichment analyses revealed
   that the DEPs were mainly involved in cell migration and immune‐related
   pathways (Figure [135]2L). Figure [136]2M illustrates the differences
   in protein expression of immune‐related genes and genes associated with
   liver metastasis in digestive system tumors between the AFPGC LM(−) and
   AFPGC LM(+) groups (Figure [137]2M; Supporting Methods). The AFPGC
   LM(+) group exhibited significantly lower immune scores
   (Figure [138]2N) and reduced tertiary lymphoid structure (TLS) scores
   (Figure [139]2O), indicating the presence of a pronounced
   immunosuppressive microenvironment within the primary tumor lesions of
   the AFPGC LM(+) group. Recent studies have shown that the suppressed
   stimulator of interferon genes (STING)^[ [140]^17 ^] and
   interferon‐gamma (IFNG) pathways^[ [141]^18 ^] play crucial roles in
   the immunosuppressive microenvironment of tumors. We also observed
   reduced activity of these pathways in the AFPGC LM(+) group.
   (Figure [142]2P,Q).

2.3. Single‐Cell Transcriptomics Revealed the Involvement of TANs in Shaping
the Immunosuppressive Microenvironment in AFPGC Patients with Liver
Metastasis

   To gain a deeper understanding of the changes occurring in the TME of
   primary AFPGC lesions during liver metastasis, we analyzed single‐cell
   sequencing samples from four AFPGC cases (Table [143]S8, Supporting
   Information). The UMAP plot shows the identified cell types, including
   epithelial cells (malignant/normal epithelial cells), fibroblast cells,
   CD4+ T cells, CD8+ T cells, macrophage cells, and neutrophil cells
   (Figure [144] 3A). The expression of representative genes across the
   different cell types is shown in Figure [145]3B. We also demonstrated
   the expression levels of AFP in different cell types and found that it
   is primarily expressed in malignant epithelial cells (Figure [146]S4,
   Supporting Information). Figure [147]3C illustrates the distribution of
   cells from the AFPGC LM(+) and AFPGC LM(−) groups. The total number and
   intensity of interactions between cell clusters were identified using
   CellChat (Figure [148]S5, Supporting Information). Additionally, we
   showed the differences in the number and intensity of interactions
   between the AFPGC LM(+) and LM(−) groups (Figure [149]3D; Figure
   [150]S6, Supporting Information). All communication probabilities in
   the information network were summarized, and receptor‐ligand pathways
   that were more active in the AFPGC LM(+) group were shown
   (Figure [151]3E). In addition, we specifically examined the STING and
   IFNG pathways and confirmed that the STING pathway was notably inactive
   in the AFPGC LM(+) group (Figure [152]3E).

Figure 3.

   Figure 3
   [153]Open in a new tab

   Single‐cell analysis reveals immune cells in the tumor microenvironment
   of AFPGC. A) UMAP plots depicting distinct cell types present in AFPGC.
   B) Gene expression heatmap in distinct cell types. C) UMAP plots
   showing cell clustering, colored with AFPGC LM(−) and AFPGC LM(+). D)
   The differences in the number of interactions between AFPGC LM(+) group
   and LM(−) group. Red (blue) edges indicate that the number of
   interactions increases (decreases) in the LM(+) group. E) Active
   Receptor‐Ligand Pathways in AFPGC LM(+) and AFPGC LM(−). F) UMAP plots
   showing cell clustering, colored with immune cells and non‐immune
   cells. G) Proportional distribution of different immune cells in AFPGC.
   H) Immune‐inhibitory scores of each cell type in AFPGC. I) Neutrophils
   were divided into 2 clusters (N2 and non‐N2) using UMAP plot. J,K)
   Expression of VEGFA, MMP9, and OLR1 in N2 and non‐N2 clusters. L)
   Differential distribution of N2 and non‐N2 subsets among TANs in AFPGC
   with and without LM. M) The comparison of N2/N1 ratio between AFPGC
   LM(−) and AFPGC LM(+) based on ssGSEA analysis using proteomics data.
   UMAP, uniform manifold approximation and projection; TANs,
   tumor‐associated neutrophils; ^* p < 0.05.

   The AFPGC LM(+) group exhibited a lower immune score (Figure [154]S7,
   Supporting Information). Based on the cell clustering results, the
   cells were classified as immune or non‐immune cells (Figure [155]3F).
   Among the immune cells, neutrophils, CD4+ T cells, and CD8+ T cells
   accounted for 40.5%, 21.1%, and 19.8% of the cells, respectively
   (Figure [156]3G; Table [157]S9, Supporting Information). Neutrophils
   presented the highest immune‐inhibitory scores (details in Supporting
   Methods) among the tumor‐infiltrating immune cells, followed by
   macrophages and CD4+ T cells (Figure [158]3H). We then re‐clustered the
   neutrophils and divided them into N2 (pro‐tumor neutrophil) and non‐N2
   neutrophils (Figure [159]3I). VEGFA, MMP9, and OLR1, which are classic
   markers of N2 neutrophils, were mainly expressed in the N2 clusters
   (Figure [160]3J,K). Additionally, we mapped the distribution and
   expression patterns of CD54 (ICAM1) and CD177 as markers associated
   with N1 neutrophils, as well as IFIT1 and RSAD2 as markers of
   interferon‐stimulated neutrophils (Figure [161]S8, Supporting
   Information). Notably, the immune inhibitory scores for the N2
   neutrophils that we identified were significantly higher than those for
   the non‐N2 neutrophils (Figure [162]S9, Supporting Information). This
   suggests that N2 has an important role in shaping the immunosuppressive
   environment in AFPGC. The relative proportion of N2, as a subset of
   tumor‐associated neutrophils (TANs), was higher in AFPGC patients with
   liver metastasis (Figure [163]3L). We also performed ssGSEA analysis
   using the proteomics data and confirmed that the N2/N1 ratio was higher
   in the AFPGC LM(+) group (Figure [164]3M).

2.4. Dynamics of the Immune Microenvironment of AFPGC Verified by mIF

   To further verify the correlation between the tumor immune
   microenvironment and liver metastasis of AFPGC, we conducted multiplex
   immunofluorescence (mIF) on 18 clinical samples, including 11 cases
   without liver metastasis (LM[−]) and seven cases with liver metastasis
   (LM[+]). We established two mIF panels to profile T‐cell
   subpopulations, B‐cell subpopulations, TANs subtypes (N1 and N2),
   dendritic cells, TLSs (Figure [165] 4A,[166]B). Figure [167]4C compares
   the differences in cell density between the AFPGC LM(−) and AFPGC LM(+)
   groups. No significant difference in CD4+/CD8+ was observed between the
   AFPGC LM(−) and LM(+) groups (Figure [168]4D); however, the N2/N1 ratio
   was higher in AFPGC LM(+) than that in LM(−) (Figure [169]4E).
   Proteomic analysis revealed a reduced TLS score in AFPGC LM(+), and
   using mIF, we assessed TLS in a more visual and quantitative manner.
   Figure [170]4F shows a typical image of a mature TLS. The TLS density
   and presence ratio in the AFPGC LM(+) group were both lower than those
   in the AFPGC LM(−) group (Figure [171]4G,H). To better characterize the
   immune microenvironment in AFPGC, we designed a new mIF‐based metric
   called the “Immune Score”, using the N2/N1 ratio and TLS as indicators
   (Figure [172]4I). The liver metastasis group exhibited a lower Immune
   Score compared to the non‐liver metastasis group (Figure [173]4J), and
   patients with lower Immune Scores had significantly poorer OS
   (Figure [174]4K). Moreover, we observed that the high Immune Score
   group tended to have lower ANLiM scores (p = 0.054; Figure [175]4L).

Figure 4.

   Figure 4
   [176]Open in a new tab

   Comparison of tumor immune microenvironment of AFPGC using mIF. A)
   Representative mIF images in AFPGC stained for antibodies in panel 1.
   Panel 1 consisted of CD66b, HLA‐DR, CD11c, MMP9, TGF‐β and PD‐L1. B)
   Representative mIF images in AFPGC stained for antibodies in panel 2.
   Panel 2 consisted of CD4, CD8, FoxP3, CD20, CD21 and CD23. C) Box plots
   showing the cell density (number of cells per mm^2) between AFPGC LM(−)
   and AFPGC LM(+). D,E) Comparison of CD4/CD8 ratio D) and N2/N1 ratio E)
   between AFPGC LM(−) and AFPGC LM(+). F) A typical image of a mature
   tertiary lymphoid structure using mIF. G,H) Comparison of the density
   G) and TLS positivity H) between AFPGC LM(−) and AFPGC LM(+). I)
   Calculation of the mIF‐based immune score. * represents N2/N1 > 0.8,
   which refers to the result of ssGSEA analysis using proteomics data.
   (e.g., The presence of TLS, N2/N1 ratio = 0.9, yielding score = 1) J)
   Comparison of mIF‐based immune score between AFPGC LM(−) and AFPGC
   LM(+) groups. K) Comparison of overall survival between high and low
   immune score groups. L) The relation between mIF‐based immune score and
   ANLiM scores. mIF, multiplex immunofluorescence; ^* p < 0.05.

2.5. Real‐World Survival Analysis Reveals the Efficacy of Immunotherapy in
AFPGC Patients and ANLiM Score can Serve as a Predictive Marker

   To validate the clinical implications identified by multi‐omics
   analysis, we conducted an efficacy analysis of immunotherapy (anti‐PD‐1
   therapy) in real‐world patients with AFPGC (N = 49, Table [177]S10,
   Supporting Information and Supporting Methods). The waterfall plot in
   Figure [178] 5A shows the best response to immunotherapy‐based
   neoadjuvant and first‐line therapies in two groups, respectively.
   Additionally, Figure [179]5B,C illustrate the changes in the sum of the
   diameters of target lesions (tumor burden) over time during treatment
   for AFPGC LM(−) and AFPGC LM(+), respectively. We observed that the
   decreased tumor burden in AFPGC patients with liver metastasis was not
   well maintained over time. In addition, the objective response rate
   (ORR) for AFPGC LM (−) was higher than that for AFPGC LM (+)
   (Figure [180]5D). From another perspective, AFPGC LM (−) patients
   undergoing combined neoadjuvant immunotherapy exhibited a higher
   pathological complete response (pCR) rate than that of non‐AFPGC LM (−)
   patients (Figure [181]5E). Additionally, AFPGC LM (+) patients
   demonstrated poorer progression‐free survival (PFS) in first‐line
   combined immunotherapy compared to non‐AFPGC LM (+) patients
   (Figure [182]5F). This finding not only suggests that the efficacy of
   immunotherapy differs in AFPGC cases compared to non‐AFPGC cases, but
   also indirectly supports our hypothesis that, based on our
   understanding of the immune microenvironment in AFPGC, the
   effectiveness of immunotherapy in AFPGC LM (+) patients is inferior to
   that in AFPGC LM (−) patients. Figure [183]5G depicts the case of an
   AFPGC LM (−) patient who experienced a reduction in tumor size
   following neoadjuvant immunotherapy and subsequently achieved a pCR
   after surgery. In contrast, Figure [184]5H presents the case of an
   AFPGC LM (+) patient in whom both liver metastases and tumor markers
   progressed after two cycles of treatment. We evaluated the capacity of
   the mismatch repair (MMR) status and programmed cell death‐ligand 1
   (PD‐L1) expression to predict immunotherapy efficacy and found that the
   correlation was relatively weak (Figure [185]5I). Therefore,
   identification of valuable predictive biomarkers is urgently required.
   Based on the AFPGC proteomic sequencing cohort in this study, a more
   in‐depth analysis was conducted. Our findings revealed a significant
   correlation, showing that a high proportion of patients with elevated
   ANLiM scores (specifically, ANLiM score = 2) were predicted to be
   non‐responders according to the Tumor Immune Dysfunction and Exclusion
   (TIDE) analysis (Figure [186]5J, the details of TIDE analysis can be
   found in Supporting Methods). Among patients with AFPGC who were
   undergoing neoadjuvant immunotherapy, we observed that those who
   achieved pCR tended to exhibit a lower proportion of high ANLiM scores
   (Figure [187]5K) and lower overall ANLiM scores (Figure [188]5L);
   however, these findings were not statistically significant. Notably, we
   observed that with first‐line combined immunotherapy for AFPGC LM(+),
   the proportion of patients with high ANLiM scores was significantly
   higher in the poor response group (PFS < 6 months) than in the good
   response group (PFS ≥ 6 months) (Figure [189]5M). Additionally, the
   overall ANLiM score was also higher in the poor response group
   (Figure [190]5N). This result suggests that the ANLiM score may have
   potential predictive value for the immunotherapy efficacy in AFPGC
   patients, particularly in cases involving liver metastasis.

Figure 5.

   Figure 5
   [191]Open in a new tab

   The efficacy of immunotherapy in AFPGC patients and the role of ANLiM
   as a predictive marker. A) Efficacy of immunotherapy (anti‐PD‐1
   therapy) in combination with chemotherapy in AFPGC LM(−) and AFPGC
   LM(+) groups, respectively. B,C) Spider plots showing treatment
   responses for all patients in AFPGC LM(−) group. B) and in AFPGC LM(+)
   group C). D) The ORR for immunotherapy in combination with chemotherapy
   in AFPGC LM (−) and AFPGC LM (+) patients, respectively. E) Comparison
   of pCR rate of neoadjuvant chemoimmunotherapy between AFPGC LM(−) and
   non‐AFPGC LM(−). F) Comparison of PFS between AFPGC LM(+) and non‐AFPGC
   LM(+) patients receiving first‐line ICI combination therapy. G)
   Representative case showing reduced tumor lesions of AFPGC LM(−) after
   immunotherapy‐based neoadjuvant treatment and achieving pCR. ’BASELINE’
   indicates the starting point for comparison. H) Representative case
   showing tumor progression of AFPGC LM(+) after 2 cycles
   immunotherapy‐based treatment. I) Sankey diagram illustrating the
   relationships between different MMR statuses, PD‐L1 expression levels,
   and therapeutic efficacy. J) Association between ANLiM score and TIDE
   predicted response. K,L) Association between ANLiM score and response
   to neoadjuvant chemoimmunotherapy in AFPGC LM(−). M,N) Association
   between ANLiM score and response to first‐line ICI combination therapy
   in AFPGC LM(−). PD, progressive disease; SD, stable disease; PR,
   partial response; ORR, objective response rate; pMMR, proficient
   mismatch repair; dMMR, deficient mismatch repair; CPS, combined
   positive score; ICI: immune checkpoint inhibitor; SOD, sum of longest
   diameters; ^* p < 0.05; ^** p < 0.01.

3. Discussion

   AFPGC is a unique subtype of GC, and our preliminary research pioneered
   the reporting of its genomic characteristics of AFPGC.^[ [192]^2 ^] In
   this study, we observed that the incidence of liver metastasis in AFPGC
   was significantly higher than that in non‐AFPGC. This observation
   aligns with the results of other studies.^[ [193]^3 , [194]^6 ^]
   Currently, the diagnosis and treatment of AFPGC follow the guidelines
   established for conventional GC. Surgery is the primary treatment
   modality for localized AFPGC. However, a substantial number of patients
   with AFPGC experience rapid postoperative liver recurrence, posing
   significant clinical challenges.^[ [195]^4 ^] Previous studies have
   indicated that plasma AFP levels may serve as an effective marker for
   predicting liver metastasis in AFPGC^[ [196]^4 , [197]^19 ^] In this
   study, we introduced the ANLiM score for predicting synchronous liver
   metastasis (including those diagnosed with occult liver metastasis),
   which demonstrated significantly higher predictive accuracy than that
   of the plasma AFP levels. Accurate risk prediction of liver metastasis
   is crucial for making informed clinical decisions and improving
   prognosis. However, the underlying mechanisms of liver metastasis
   remain elusive. Therefore, urgent and in‐depth mechanistic
   investigations are warranted to develop precise therapeutic strategies
   and enhance the prognosis of AFPGC.

   Although certain individual genetic changes have been implicated in the
   development of metastasis, few appear to be metastasis‐specific.^[
   [198]^11 , [199]^20 ^] Previous studies suggest that metastasis may
   rely more on non‐genetic mechanisms, such as the TME, than previously
   recognized.^[ [200]^12 , [201]^21 ^] The TME encompasses diverse immune
   cell types, fibroblasts, the endothelium, and various other
   tissue‐resident cell types.^[ [202]^13 ^] Tumors and their surrounding
   microenvironment are increasingly being conceptualized as a
   comprehensive ecosystem and are explored at three different levels:
   primary (the primary site), regional (metastatic lymph node), and
   distal (metastatic lesion) onco‐spheres.^[ [203]^13 , [204]^22 ,
   [205]^23 ^] A comparative analysis of the ecosystems of metastatic and
   primary tumor sites is necessary for revealing metastasis‐related
   spatiotemporal heterogeneity.^[ [206]^13 , [207]^24 , [208]^25 ^] Such
   insights are crucial for developing novel therapeutic targets and for
   tackling challenges associated with primary and secondary drug
   resistance. Tumor metastasis is a complex, multi‐step process that
   involves multiple stages. During the early stages of tumor metastasis,
   significant features that promote metastasis often accumulate at the
   primary site as the primary tumor continues to evolve^[ [209]^13 ,
   [210]^25 ^] Therefore, investigating the alterations in the
   microenvironment of the primary tumor before and after metastasis may
   assist with identifying earlier metastasis‐related events at the
   primary site. Additionally, this will provide valuable insights for the
   early prediction of tumor metastasis and intervention. In this study,
   we observed that primary tumors with liver metastasis exhibited higher
   CIN. CIN arises from persistent errors in chromosomal segregation
   during mitosis and plays a significant role in cancer metastasis.^[
   [211]^26 , [212]^27 ^] The proteome is more closely aligned with
   biological phenotypes than are either the genome or the transcriptome.
   Our proteomic analysis revealed that proteins related to metastasis,
   such as METTL3^[ [213]^28 ^] and IGF2BP2^[ [214]^29 ^] were highly
   expressed in the primary lesions of AFPGC with liver metastasis.
   Interestingly, METTL3 and IGF2BP2 are key members of the m6A
   methylation machinery, with METTL3 acting as a methylation writer and
   IGF2BP2 as a reader^[ [215]^30 , [216]^31 ^] Dysregulated m6A
   methylation has been reported to play a significant role in the
   remodeling of the TME and promoting tumor metastasis.^[ [217]^32 ^] In
   this study, in addition to the metastasis‐related pathways, we observed
   that additional immune‐related pathways were enriched through the
   pathway enrichment analysis of DEPs. Integrated analysis revealed that
   AFPGC patients with liver metastases exhibited not only significantly
   lower immunoscores in primary tumors but also higher CIN levels,^[
   [218]^26 , [219]^27 ^] both of which converge toward a possible
   immunosuppressive microenvironment.

   To further investigate the immune microenvironment within AFPGC
   patients, we analyzed the scRNA‐seq data from primary tumors with or
   without liver metastasis. Our analysis of immune cell components
   revealed that TANs were represented the predominant cell type within
   the TME. TANs exhibit multifaceted characteristics and can affect the
   tumors in different ways depending on their polarization states.^[
   [220]^33 ^] Analogous to the M1/M2 classification of macrophages, the
   N1 and N2 have been introduced to define neutrophils with antitumor and
   pro‐tumor functions, respectively.^[ [221]^33 , [222]^34 ^] The role of
   macrophages within the immunosuppressive microenvironment has been
   extensively characterized and well‐documented.^[ [223]^35 ^] Our study
   showed that TANs, especially the N2 subtype (pro‐tumor neutrophil),
   exhibited higher immune inhibitory scores than macrophages in AFPGC.
   Recent studies have revealed that TANs, especially the N2 subtype, play
   a crucial role in establishing an immunosuppressive microenvironment
   within tumors.^[ [224]^36 , [225]^37 , [226]^38 ^] The stimulator of
   interferon genes (STING) pathway plays a crucial role in antiviral host
   defense mechanisms and antitumor immunity.^[ [227]^39 , [228]^40 ^]
   Notably, we observed an increased proportion of N2 and suppression of
   STING signaling in primary tumors with liver metastasis. The biological
   significance and potential mechanistic role of this finding remain to
   be further elucidated. Hao et al. utilized neutrophils as carriers to
   deliver STING agonists for penetrating solid tumors, effectively
   activating the STING pathway.^[ [229]^41 ^] This activation reprograms
   the TME by converting neutrophils and macrophages into antitumor
   phenotypes (N1/M1), thereby enhancing antitumor immunity. Lu et al.
   found that activation of the STING pathway by MnO₂ nanoparticles
   promotes IRF3 phosphorylation and stimulates IFN‐β secretion in breast
   cancer xenograft model, thereby inducing a shift in neutrophil
   polarization from the N2 to the N1 phenotype and enhancing antitumor
   immunity.^[ [230]^42 ^] Additionally, previous studies have indicated
   that the STING pathway can induce changes in the downstream TGF‐β
   pathway,^[ [231]^43 ^] which has been reported to mediate the N1‐N2
   conversion.^[ [232]^34 ^] However, the potential association between
   STING pathway activation and N1/N2 neutrophil polarization dynamics in
   AFPGC with liver metastases remains to be experimentally explored. For
   example, future studies could explore the polarization of neutrophils
   in models with activated STING pathways and assess the resulting
   alterations of cytokine profiles in AFPGC with liver metastases.
   Additionally, these studies could investigate the interplay between
   STING and related signaling pathways, such as TGF‐β, in modulating
   neutrophil function and phenotype. TLSs are critical indicators of the
   TME status and serve as essential sites for the initiation of
   anti‐tumor immune responses.^[ [233]^44 , [234]^45 ^] TLSs modulate the
   immune microenvironment by actively recruiting circulating immune cells
   and enhancing local immune responses.^[ [235]^44 , [236]^45 , [237]^46
   ^] Our study showed that the density of TLSs was relatively low in the
   AFPGC with liver metastasis, indicating limited benefit from
   immunotherapy. Chelvanambi et al. found that intratumoral injection of
   the STING agonist can slow melanoma growth, improve vascular
   normalization, and promote the formation of TLS within the tumor
   microenvironment, thereby enhancing the anti‐tumor immune response.^[
   [238]^47 ^] In future studies, the relationship between the formation
   of TLS and the efficacy of immunotherapy in AFPGC patients with liver
   metastasis warrants further exploration.

   With the widespread use of immune checkpoint inhibitors (ICIs),
   immune‐based combination therapies have shown significant efficacy in
   GC, and their application is now being extended from advanced stages to
   the perioperative setting.^[ [239]^48 ^] Our real‐world cohort analysis
   showed that the efficacy of immune combination therapy was suboptimal
   in the AFPGC liver metastasis group compared with that in the LM(−)
   group. This phenomenon can be partially attributed to the
   characteristics of the immunosuppressive microenvironment. Currently,
   the most commonly used predictive biomarkers for ICI efficiency are
   PD‐L1, MMR, TMB, and TNB, all of which are tissue‐based biomarkers.^[
   [240]^49 ^] Holder et al. previously highlighted the challenges
   associated with tissue‐based biomarkers, such as invasive procedures,
   high costs, and heterogeneous sampling. Additionally, our data also
   indicated that the correlation between PD‐L1, MMR, and the efficacy of
   immunotherapy for AFPGC was relatively poor. Blood‐based biomarkers
   might represent promising avenues for future research.^[ [241]^49 ^]
   ANLiM, a novel blood‐based biomarker, was initially proposed as a
   potential predictor of liver metastasis. Our findings suggest that
   ANLiM is a promising biomarker for predicting the efficacy of
   immunotherapy in patients with AFPGC. By integrating AFP, a
   well‐established tumor marker with confirmed relevance in AFPGC^[
   [242]^2 , [243]^19 ^] that reflects immune system dynamics,^[ [244]^50
   ^] with NLR, an indicator of systemic inflammation,^[ [245]^51 ^] ANLiM
   offers a comprehensive view of the tumor's biological aggressiveness
   and the host's immune response. Similarly, Scheiner et al. have devised
   a CRAFITY scoring model that leveraged two clinically established
   biomarkers: C‐reactive protein and AFP.^[ [246]^52 ^] This model offers
   a rapid and precise means of predicting the therapeutic response to
   immunotherapy in patients with liver cancer. Therefore, our proposed
   ANLiM as a robust predictive biomarker for both liver metastasis and
   the immunotherapy response. This study provides valuable insights for
   screening patients with AFPGC who may be sensitive to immunotherapy.
   Additionally, it suggests potential approaches for designing clinical
   trials and optimizing personalized treatment strategies.

4. Limitations

   First, this study involved a relatively small number of sequencing
   samples, encompassing both proteomics and single‐cell sequencing. This
   is partly attributed to the rarity of AFPGC. However, by focusing on
   specific clinical issues, we made notable discoveries with the limited
   available samples. Second, this study did not address interventional
   research on the immunosuppressive microenvironment in AFPGC patients
   with liver metastasis, which is a promising direction for future
   research. In our previous studies, we have constructed multiple
   humanized patient‐derived xenograft (PDX) models for AFPGC^[ [247]^2 ^]
   and also achieved partial immune reconstruction in PDX mice using
   peripheral blood mononuclear cells (PBMCs) engraftment from adult
   donors.^[ [248]^53 ^] We are currently developing a patient‐derived
   xenograft model of AFPGC with a humanized immune system, which will be
   crucial for future studies. Third, considering that the construction
   and validation of the ANLiM score are based on retrospective cohorts,
   it is essential to conduct multi‐center prospective validation in the
   future. Furthermore, the applicability of the ANLiM score in AFPGC
   patients accompanied by acute infections requires further
   investigation.

5. Conclusion

   This study established a practical framework for leveraging clinically
   accessible information and multi‐omics to gain insights into the
   clinical implications of rare cancers. We developed a blood
   biomarker‐based predictive indicator for liver metastasis and
   immunotherapy response in AFPGC. Our findings on immune
   microenvironmental alterations for AFPGC with liver metastasis provide
   new insights for optimizing immunotherapy strategies.

6. Experimental Section

   Details about experimental procedures and data analysis are provided in
   the Supporting Methods.

Conflict of Interest

   The authors declare no conflict of interest.

Ethics Approval

   This study was approved by the Institutional Review Board of the First
   Affiliated Hospital of Zhejiang University School of Medicine (approval
   number: 2024‐0734‐Fast).

Supporting information

   Supporting Information
   [249]ADVS-12-e03499-s001.pdf^ (2.6MB, pdf)

Acknowledgements