Introduction

Gastric cancer (GC) is a major global health concern, ranking as the fifth most commonly diagnosed malignant tumor and the fourth leading cause of cancer-related deaths worldwide1. The development of Non-cardia GC is closely linked to chronic Helicobacter pylori (H. pylori) infection2,3, which can lead to sustained immune infiltration and contribute to the progression of gastric atrophy, intestinal metaplasia (IM), dysplasia, and intestinal-type GC4. As a result, intestinal-type GC is widely recognized as an inflammation-driven cancer5. A growing body of research has demonstrated that the inflammatory microenvironment, particularly macrophages, plays a critical role in the destruction and metaplasia of gastric mucosa glandular epithelial cells6,7.

Chronic inflammation can induce parietal cell loss, oxyntic cell atrophy, and metaplastic lineages arise, resulting in an increased risk of GC8. Macrophages, a key component of the immune system, have been shown to play a critical role in the progression of GC through their ability to be polarized by different microenvironment signals9. Studies have shown that macrophages display a high degree of plasticity when stimulated by cytokines, opsonic receptors, and other ligands10 and that the tissue and microenvironment context plays a significant role in determining macrophage function11. Transcriptome sequencing of gastric mucosa from L-635 treated mice6 revealed that macrophages were primarily of the M2a phenotype, and that depletion of macrophages significantly reduced the incidence of IM and advanced metaplasia. Another study12 found that macrophages were required for the growth of gastric premalignant lesions and were likely a rate-limiting factor for the development of gastric tumors in gp130FF mice. These studies highlight the crucial role that macrophages play in gastric tumorigenesis and suggest that specific phenotypes may play a key role in tumor progression. However, the characterization of macrophage subsets in the progression of precancerous lesions to gastric cancer remains incomplete due to the limitations of bulk sequencing methods.

Recently, single-cell RNA sequencing (scRNA-seq) has emerged as a powerful tool for resolving the complex cellular composition of the tumor microenvironment at high resolution. This approach has been used to identify common functional characteristics and key differential characteristics of macrophage subsets13. Macrophages were found to play a pivotal role in gastritis-induced gastric tumorigenesis by scRNA-seq, as they are the dominant source of interleukin-1B and PTGS2, which was reported to be associated with an increased risk of GC14. However, further analysis of macrophage subsets has not yet been performed. Among other diseases, Xingwu Liu and colleagues15 reported a specific macrophage subtype, IDO1 macrophages, that can predict response to immunotherapy in colorectal cancer patients. Aizhen Xiong and colleagues16 reported that glycoprotein non-metastatic B-high macrophages promote glioblastoma plasticity and prevent T-cell activation by dendritic cells. Therefore, we aimed to identify specific macrophage subtypes that can aim in identifying biomarkers and potential treatment targets for gastric tumorigenesis.

Here, we systematically analyzed scRNA-seq and microarray data in patients with gastritis, IM, dysplasia, and intestinal-type GC cascade from various human studies to identify the macrophages and related genes. The in-vitro experiment was performed to verify the mechanism of the key inflammatory factors. Our findings suggest a subtype of S100A8 and S100A9 + macrophages was associated with gastritis-induced pathological progression, possibly through the TLR4-NFκB signaling pathway.

Methods

Data sources

We obtained Whole Human Genome Microarray data from 19 samples of chronic gastritis, 19 samples of low-grade intraepithelial neoplasia (LGIN), 20 samples of high-grade intraepithelial neoplasia (HGIN), and 19 samples of early gastric cancer (EGC) from the Gene Expression Omnibus (GEO; RRID: SCR_005012) database(https://www.ncbi.nlm.nih.gov/geo/, GSE55696)17.

The single-cell transcriptome data from a total of 26 patients were included in the analysis. The data of 3 non-atrophic gastritis, 3 chronic atrophic gastritis(CAG), 6 IM, and 1 EGC were collected from the GEO database(GSE134520)14. The data of 8 intestinal-type GC (pat01-08), 2 CAG (pat25/pat29), and 3 NAG (pat26-28) were selected from the GEO database(GSE150290)18.

Estimation of the extent of immune cell infiltration

We applied an algorithm of CIBERSORT(RRID: SCR_016955) to estimate the relative proportions of immune cell subsets in tissues based on the input microarray gene expression matrix19. The absolute score of all the immune cells was used to determine the extent of immunity infiltrating.

CIBERSORT is a computational method used to estimate the abundance of different immune cell types in a mixed cell population. The core of CIBERSORT is a deconvolution algorithm, which uses the signature matrix to estimate the proportion of each immune cell type in a mixed sample. This is done by comparing the gene expression profile of the mixed sample to the reference profiles. The “absolute score” refers to a measure of the overall immune cell content in a given sample, expressed in a way that reflects the absolute abundance of immune cells. The absolute score of all the immune cells is essentially the sum of the absolute scores for each immune cell type present in the sample. Each cell type’s absolute score represents its abundance, and when these are summed up, they give an overall indication of the immune cell load in the sample. By comparing the absolute scores across multiple samples, researchers can identify samples with relatively high or low immune content, which can be correlated with various clinical outcomes or biological conditions.

Identification of immune microenvironment-related key modules

The “WGCNA” R software package(RRID: SCR_003302) was employed for weighted correlation network analysis a method commonly used to identify biomarkers by clustering genes with similar expression patterns to establish modules, and analyzing the correlation between modules and specific characteristics such as clinical information of patients20. First, we checked for missing values in the expression matrix data and removed any outlier samples through clustering. We then selected a suitable soft threshold to construct a scale-free network, calculated the adjacency values between genes and constructed a topological overlap matrix. The genes with identical expression patterns were divided into distinct modules containing at least 50 genes based on the selected soft threshold. The modules with a correlation of less than 0.25 were merged by the dynamic tree pruning method, all genes were divided into several modules, and the correlation between each module and pathological category and the immune score was calculated separately. The module that was both positively correlated with pathological severity and the immune score is selected as the key module.

Single-cell RNA-seq analysis

Two single-cell transcriptome data were integrated to perform the downstream analysis21. The “Seurat” R software package(RRID: SCR_016341) was used for analysis22. First, the data was imported into R to generate a Gene-barcode matrix, cells with a gene number (nFeature_RNA) between 500 and 2000 are retained, and cells with mitochondrial genes greater than 25% are filtered. Then “Seurat” was used for log normalization of the data, “FindVariableFeatures” was used to find 2000 hypervariable genes, followed by data normalization, and PCA linear dimensionality reduction was performed on the normalized data. “t-Distributed Stochastic Neighbor Embedding(tSNE; RRID: SCR_024305)” was used for cluster visualization and to group cells. The “FindAllMarkers” function was used to obtain the eigengenes for each subset and the markers “CD163”, “CD14”, and “AIF1” were used to annotate macrophages. The original sparse matrix of macrophage expression values was extracted for re-clustering and analysis.

We analyzed the proportion of macrophage subsets in each sample, functional genes, pathway enrichment, and pathway activity. The “Monocle” and default parameters were used for pseudotime and cell trajectory analyses. The macrophage subset associated with gastric tumorigenesis that positively correlated with pathological severity was selected for downstream analysis. Overlap genes in hub genes and selected macrophage sub-cluster were selected as differentially expressed genes (DEGs).

Functional enrichment analysis of DEGs

All DEGs were subjected to Gene Ontology (GO; RRID: SCR_002811) and Kyoto Encyclopedia of Genes and Genomes (KEGG; RRID: SCR_012773)23analyses. Gene names were converted to gene IDs for GO and KEGG analysis using the R package “org.Hs.eg.db”. The “clusterProfliler”, “enrichplot”, and “ggplot” R packages were applied to perform the GO biological process and KEGG pathway analysis of DEGs. The gene IDs used in the KEGG pathway are retrieved and converted to Symbol genes again via “Perl”. A threshold of FDR < 0.05 was used as the cut-off criterion.

In vitro experiments

Human Gastric Adenocarcinoma Celine AGS (KG187; RRID: CVCL_0139) purchased from Jiangsu KGI Biotechnology Co., Ltd. were cultured in RPMI-1640 medium (Wisent, Montreal, Canada) at 37 °C in a humidified incubator with 5% CO2. The medium was supplemented with 10% fetal bovine serum (FBS) (PAN-Biotech, Germany) and 1% penicillin/streptomycin (Wisent, Montreal, Canada).

AGS cells were seeded at a concentration of 1 × 106 cells/mL. After overnight culture AGS were stimulated for 12 h with 1.5 µg/ml S100A9 (#9254-S9-050, R&D Systems), 0.5 µM Paquinimod(CM01386, Prointech), 1.5 µg/ml S100A9 plus 0.5 µM Paquinimod, 1.5 µg/ml S100A9 plus 1 µM CLI-095(tlrl-cli95, Invivogen), and 1.5 µg/ml S100A9 plus 10µM RAGE antagonistic peptide (6259/1, Tocris).

Western blot analysis

AGS cells were inoculated into 6 cm dishes, and the culture medium was sucked off after adhering to the wall. After washing with PBS three times, S100A9, Paquinimod, and S100A9 + Paquinimod were added to intervene for 24 h. Proteins were extracted and quantified using a BCA kit. Proteins were denatured in a boiling water bath and stored for later use. Protein separation was performed using 10% SDS-PAGE gel electrophoresis, and the target protein was transferred from the gel to a PVDF membrane. Place the transferred membrane in the incubation tank with TBST and rinse once quickly. Then block with skim milk for 30 min at room temperature. After washing with TBST, the membrane was incubated with the specific primary antibody overnight at 4 °C. After another TBST wash, the membrane was incubated with the secondary antibody for 1 h at room temperature and then developed in the dark.

The primary antibodies include Toll-like receptor 4 TLR4 (66350-1-Ig-50ul, RRID: AB_1930596) and Nuclear Factor kappa-B p65(80979-1-RR; RRID: AB_895724). The secondary antibodies include antirabbit HRP-IgG (CST, Danvers, MA; RRID: AB_1026778) and antimouse HRP-IgG(CST, Danvers, MA).

Cell proliferation assay

Cells were seeded and cultured in 96-well plates with an equal number. The cell proliferation rate was determined with cell counting kit-8 (Dojindo, Kumamoto, Japan) reagent. The plates were incubated in a cell incubator containing 5% CO2 at 37 °C for 1.5 h. Then the absorbance at 450 nm was measured.

Wound healing assay

Inoculate the same number of cells after drawing 4 horizontal lines at the same interval on the back of the 6-well plate. Scratch the surface of each well in the 6-well plate as evenly as possible with a 200 µL pipette tip. The wells were then rinsed three times with PBS and incubated in 1640 medium containing 1% FBS at 37 °C for 48 h. Images of the initial wound and the movement of cells into the scratched area were taken at 0 h and 24 h using a Nikon microscope (Olympus Corporation, Tokyo, Japan).

Statistical analysis

R (version 4.3) and GraphPad Prism software (version 8.0) were used for statistical analysis and plotting. Image J software were used for image processing and evaluation. Student’s t test and one-way ANOVA test were used to determine the significance of differences, and p-values less than 0.05 were considered to be significant.

Result

Key-module identification related to the immune microenvironment and Correa cascade

Seventy-seven samples were clustered by the hclust clustering algorithm, and the clustering tree (Supplementary Figure S1) was drawn. No obvious outliers were found, and all samples entered the downstream analysis. We calculated the soft threshold and built a scale-free topology model (soft-thresholding power 9). The scale-independent topological and mean connectivity networks were plotted (Fig. 1A, B). A cluster dendrogram was drawn according to the weighted values, and branches containing similarly expressed genes are regarded as gene modules. A total of 18 gene modules were obtained by fusing similar modules (Fig. 1C).

Fig. 1
figure 1

WGCNA analysis of the gastric premalignant lesions. (A) Estimation of the scale independence index of the 1–20 soft threshold power (β = 9). (B) Determination of the mean connectivity of the 1–20 soft threshold power. (C) Identification and establishment of gene co-expression modules that are shown as different colors in hierarchical clustering. (D) Module-trait relationships of gastritis, LGIN, HGIN, EGC and the absolute score of immune cells. LGIN: low-grade intraepithelial neoplasia; HGIN: high-grade intraepithelial neoplasia; EGC: early gastric cancer.

Full size image

Among the 18 modules (Fig. 1D), the greenyellow module had the highest correlation with the absolute score (r = 0.83, P = 1e−0), which also correlated with pathological severity (EGC r = 0.59, P = 1e−8; HGIN r = − 0.014, P = 0.9; LGIN r = − 0.25, P = 0.03; gastritis r = − 0.33, P = 0.04). Therefore, the greenyellow module was selected as a key module, and 362 hub genes with a gene significance value greater than 0.6 and a module membership value greater than 0.7 were identified in the key module (Fig. 2A).

Fig. 2
figure 2

The module membership in the green-yellow module and enrichment analysis of the hub genes. (A) Identification of the hub gene in the green-yellow module. (B) GO analysis of hub genes. (C) KEGG analysis of hub genes.

Full size image

Enrichment analysis of hub genes

The biological process (BP), cellular component (CC), and molecular function (MF) of hub genes were identified and explored by GO analysis (Fig. 2B). The top eight enrichment terms of BP were all related to immune response: T cell activation, leukocyte cell-cell adhesion, immune response-regulating signaling pathway, leukocyte activation involved in immune response, regulation of leukocyte cell-cell adhesion, activation of immune response, and regulation of cell-cell adhesion. Moreover, the external side of plasma membrane, tertiary granule, and secretory granule membrane were the top three enriched CCs in GO analysis. Immune receptor activity, pattern recognition receptor activity, and cytokine receptor activity were the top three enriched MFs in GO analysis.

KEGG pathway analysis of hub genes identified 40 pathways, including Osteoclast differentiation, Leukocyte transendothelial migration, Leishmaniasis, Neutrophil extracellular trap formation, Hematopoietic cell lineage, NF -kappa B signaling pathway, and Toll -like receptor signaling pathway, etc. (Fig. 2C).

Macrophage characteristics in the Correa cascade through single-cell RNA analysis

We used the tSNE algorithm for non-linear dimensionality reduction, revealing 30 distinct cell clusters from the single-cell analysis of 26 patients (Fig. 3A). Among these, the 21st cluster was identified as macrophages, specifically marked by AIF1 and CD14 (Fig. 3B, C). We extracted the macrophages for further analysis, and differential expression analysis of core genes within this cluster highlighted the top 10 variable genes: CCL20, S100A8, CCL22, HIST1H4C, CXCL1, S100A9, APOC1, CXCL10, TFF1, and CCR7 (Fig. 3D). A PCA was conducted, visualizing the top 30 feature genes of the first two principal components (Fig. 3E).

Fig. 3
figure 3

Identification, extraction, and dimensionality reduction of macrophage single-cell data. (A) Thirty cell clusters were identified by performing tSNE. (B) AIF expression distribution map. (C) CD14 expression distribution map. (D) The 10 variable genes were marked in the macrophage cluster. (E) The top two PCs of all macrophages.

Full size image

Further single-cell analysis of macrophages from the 26 patients revealed four distinct cell clusters (Fig. 4A). The proportions of these macrophage subsets varied significantly across different pathological stages of the Correa cascade, with Cluster 3 notably increased during the early gastric cancer (EGC) stage (Fig. 4B). A heatmap of the top 5 differential genes in each of the four macrophage subsets showed that S100A8, FCN1, S100A9, S100A12, and IGJ were the top genes in Cluster 3 (Fig. 4C). Pseudotime analysis using cell cluster markers suggested potential differentiation trajectories for these macrophage subtypes (Fig. 4D). Over time, macrophages can differentiate into Cluster 3 and Cluster 4 subtypes, with S100A8/S100A9 being highly expressed during differentiation into Cluster 3 (Fig. 4E, F, G).

Fig. 4
figure 4

Proportional, differential genes, and pseudotime analysis of macrophage. (A) Four macrophage clusters were identified by performing tSNE. (B) The proportions of four macrophage clusters in different pathological stages. (C) The heat map of the top 5 differential genes of the 4 macrophage clusters. (D) The macrophage differentiation trajectory plot. (E) Map of macrophage differentiation trajectories colored in order of cell progression. The deeper the color, the earlier the beginning of cell progression. (F) & (G) The expression of S100A8/S100A9 in the cell differentiation trajectory map.

Full size image

Enrichment analysis of cluster 3 macrophage markers and hub genes overlap key genes

Cluster 3 macrophage subset, marked specifically by S100A8/S100A9 (Fig. 5A–C), may indicate progression from gastric precancerous lesions to EGC. Subsequent analysis of cluster 3 macrophage single-cell genes and hub genes identified 69 overlap genes as potential key genes (Fig. 5D). GO analysis of the potential key genes revealed that the top three enriched biological processes were negative regulation of cell activation, positive regulation of response to external stimulus, and negative regulation of leukocyte activation. The top three enriched cellular components were tertiary granule, actin filament, and secretory granule membrane. The top three enriched molecular functions were RAGE receptor binding, Toll-like receptor binding, and antioxidant activity. (Fig. 5E). KEGG pathway analysis demonstrated that the top three enriched pathways were Lipid and atherosclerosis, NF-kappa B signaling pathway, and Phagosome (Fig. 5F).

Fig. 5
figure 5

Enrichment analysis of cluster 3 macrophages single-cell genes and hub genes overlap key genes and GSEA. (A) Four macrophage clusters were identified by performing tSNE. (B) S100A8 expression distribution map. (C) S100A9 expression distribution map. (D) Venn diagram of cluster 3 macrophages single-cell gene and hub genes. (E) GO analysis of key genes. (F) KEGG analysis of key genes.

Full size image

S100A9 promotes the proliferation and migration of AGS cells through the TLR4-NFκB signaling pathway

The cytokine S100A9 significantly promoted the migration of AGS cells (p = 0.045), and the inhibitor of S100A9, Paquinimod, could prevent the migration of AGS cells (p = 0.009, Fig. 6A). The intervention of S100A9 significantly promoted the proliferation of AGS cells (Ctr vs. S100A9, p < 0.0001, MD= − 22.82, 95%CI − 31.33 to 14.31), but the addition of Paquinimod could inhibit the proliferation (S100A9 vs. S100A9 + Paquinimod, p < 0.0001, MD = 42.47, 95%CI 33.96–50.97, Fig. 6B). Western blot found that the expression of TLR-4 protein in AGS cells increased after cytokine S100A9 intervention, and Paquinimod could reduce the expression of TLR-4(Ctr vs. S100A9, P = 0.031, Fig. 6C). There was no difference in NFκB-P65 protein expression, but phosphorylated NFκB-P65 increased after S100A9 intervention (p = 0.036) and decreased after Paquinimod intervention(S100A9 vs. Paquinimod, P = 0.044, Fig. 6C). According to literature reports, both TLR-4 and RAGE are receptors of S100A924,25. Therefore, we conducted TLR-4 and RAGE blocking experiments, and the results showed that both of them can reduce the pro-proliferation effect of S100A9, but neither can achieve the blocking effect similar to that of Paquinimod(Ctr vs. S100A9, p < 0.0001, Ctr vs. S100A9 + paquinimod, p < 0.0001,Ctr vs. S100A9 + CLI-095, p = 0.012, Ctr vs. S100A9 + Anti-RAGE, p = 0.008, S100A9 vs. S100A9 + Paquinimod, p < 0.0001, S100A9 vs. S100A9 + CLI-095 p = 0.009, S100A9 vs. S100A9 + Anti-RAGE, p = 0.006, S100A9 + Paquinimod vs. S100A9 + CLI-095, p < 0.0001, S100A9 + Paquinimod vs. S100A9 + Anti-RAGE, p < 0.0001, Fig. 6D).

Fig. 6
figure 6

Cytokine S100A9 promotes cell proliferation and migration. (A) Cytokines S100A9 and Paquinimod interfered with AGS cells for 24 h and observed the migration area. Scale bar: 200µM. (B) Cell proliferation of cytokines S100A9, Paquinimod, and S100A9 plus Paquinimod intervened AGS cells for 24 h. (C) The expressions of TLR-4, NFκB-P65, and Phosphorylated NFκB-P65 proteins in AGS cells after cytokines S100A9 and Paquinimod intervened for 24 h. (D) Cell proliferation of cytokines S100A9, Paquinimod, S100A9 plus Paquinimod, S100A9 plus CLI-095(TLR-4 inhibitor), and S100A9 plus Anti-RAGE intervened AGS cells for 24 h.

Full size image

Discussion

Chronic inflammation is a well-established risk factor that can lead to the formation of an immunosuppressive microenvironment, which can in turn promote GC formation and progression26. Macrophages are key participants within the immune microenvironment and are known to play a crucial role in promoting the persistence and progression of gastric inflammation. However, most studies on the subject have been incomplete, due to a lack of detailed characterization of the various macrophage subsets6,7,12.

We have revealed a subtype of macrophages characterized by the expression of S100A8 and S100A9 that are differentially expressed in gastric precancerous and GC progression, using a combination of microarray and single-cell transcriptome data in the Correa cascade. These macrophages have not previously been described in gastric tumor progression, and S100A8/S100A9 may serve as potential markers for the diagnosis and risk assessment of GC. Our findings suggest that the NFκB signaling pathway is a major pathway through which S100A8/S100A9 promotes tumor progression and that these genes can also promote the malignant proliferation of monocyte-macrophages. Real et al. identified S100A8 + macrophages as a distinct subtype of macrophages that act as the primary reservoir for viral mucosal infections27. Their data also showed that S100A8 + macrophages can reactivate HIV viral production upon stimulation of autocrine/paracrine S100A8-mediated glycolysis. Therefore, while S100A8/A9 is secreted by macrophages, it also regulates the phenotype and function of macrophages28. However, this subtype of macrophages has not been previously reported in gastric precancerous diseases or GC. Our study is the first to report that S100A8/A9 + macrophages are increased in the EGC stage. We found that macrophages differentiated into newly appeared S100A8/A9 + macrophages subtype or the subtype that expressed LSP1, CD1C, CD1E, etc. These two macrophage subtypes may represent two modes of disease progression. S100A8/A9 + macrophages gradually increase as the Correa cascade progresses, indicating an increased risk of GC. Therefore, S100A8 and S100A9 are potential markers for the diagnosis and risk assessment of GC.

S100A8 and S100A9 are members of the S100 protein family (S100s), which are mainly secreted by the inflammatory, tumor, and stromal cells29. They can bind together to form a heterodimer that acts as a pro-inflammatory damage-associated molecular pattern (DAMP)30. Intracellular S100A8/A9 complexes are involved in cytoskeleton modulation, while extracellular S100A8/A9 acts as a chemoattractant and promotes leukocyte recruitment31. S100A8/A9 respond quickly in the early stage of inflammation and play a key role in promoting further inflammation. In this study, we identified S100A8 and S100A9 as robust markers of gastric carcinogenesis progression as they are both hub genes of the greenyellow module, which is negatively associated with the Correa cascade, and key genes of Cluster 3 macrophages, which are increased at the EGC stage. Previous research by Zhao Z et al. analyzed the expression levels of S100A9 in 200 patients with gastric atrophy, IM, dysplasia, GC, and healthy controls and found that higher S100A9 expression levels indicate more severe pathological stages32. As a biomarker for EGC, their data also reported that the diagnostic sensitivity and specificity of S100A9 were 61.4% and 81.3%, respectively.

Through enrichment analysis, we discovered that the NF-κB signaling pathway plays a crucial role in the promotion of gastric tumorigenesis by S100A8/S100A9. This is evident from its ranking as the 6th hub gene in KEGG analysis and the 2nd key gene in KEGG analysis. It is known that S100A8/S100A9 heterodimer can bind to Toll-like receptor 4 (TLR4) to regulate inflammatory responses33. Previous studies have suggested that TLR-4 regulates inflammation via the NF-κB signaling pathway in conditions such as ulcerative colitis34, obesity35, and intervertebral disc degeneration36. However, to our knowledge, the S100A8/S100A9- NFκB signaling pathway has not yet been reported in gastric tumor progression. Through the preliminary exploration of in vitro experiments, we found that S100A9 has multiple targets in gastric cancer cells, including TLR-4 and RAGE, which can further activate NFκB and promote AGS cell proliferation and migration.

Although our study has yielded new insight into gastric tumor progression through extensive bioinformatics analysis, it is important to note that it has some limitations. Firstly, the findings of this study should be validated through examination of the tissues and blood samples from patients. Additional follow-up data would also strengthen the credibility of the results. Secondly, the identified genes and pathways required further validation. Finally, with the advancement of bioinformatics technology, understanding the spatial distribution of these macrophages and genes will aid in the comprehension of gastric tumorigenesis.

In conclusion, our study identified a subtype of S100A8 and S100A9-positive macrophages that significantly increased in the EGC. Our results suggest that S100A8 and S100A9 may be potential markers for the diagnosis and risk assessment of GC, and they can promote gastric cancer cell proliferation and migration through the S100A8/S100A9- NFκB signaling pathway, ultimately leading to gastric tumor progression.