Abstract
Background
Stomach adenocarcinoma (STAD) is one of the most common malignancies.
Infection of helicobacter pylori (H. pylori) is a major risk factor
that leads to the development of STAD. This study constructed a risk
model based on the H. pylori-related macrophages for predicting STAD
prognosis.
Methods
The single-cell RNA sequencing (scRNA-seq) dataset and the clinic
information and RNA-seq datasets of STAD patients were collected for
establishing a prognostic model and for validation. The “Seurat” and
“harmony” packages were used to process the scRNA-seq data. Key gene
modules were sectioned using the “limma” package and the “WGCNA”
package. Kaplan-Meier (KM) and Receiver Operating Characteristic Curve
(ROC) analyses were performed with “survminer” package. The “GSVA”
package was employed for single sample gene set enrichment analysis
(ssGSEA). Cell migration and invasion were measured by carrying out
wound healing and trans-well assays.
Results
A total of 17397 were screened and classified into 8 cell type
clusters, among which the macrophage cluster was closely associated
with the H. pylori infection. Macrophages were further categorized into
four subtypes (including C1, C2, C3, and C4), and highly variable genes
of macrophage subtype C4 could serve as an indicator of the prognosis
of STAD. Subsequently, we developed a RiskScore model based on six H.
pylori -associated genes (TNFRSF1B, CTLA4, ABCA1, IKBIP, AKAP5, and
NPC2) and observed that the high-risk patients exhibited poor
prognosis, higher suppressive immune infiltration, and were closely
associated with cancer activation-related pathways. Furthermore, a
nomogram combining the RiskScore was developed to accurately predict
the survival of STAD patients. ABCA1 in the RiskScore model
significantly affected the migration and invasion of tumor cells.
Conclusion
The gene expression profile served as an indicator of the survival for
patients with STAD and addressed the clinical significance of using H.
pylori-associated genes to treat STAD. The current findings provided
novel understandings for the clinical evaluation and management of
STAD.
Keywords: Stomach adenocarcinoma (STAD), Single cell RNA-Seq profile,
RiskScore, H. pylori infection phenotype, ssGSEA, WGCNA
1. Introduction
Stomach adenocarcinoma (STAD) is a gastrointestinal cancer with the
fifth highest incidence and the third highest mortality rate worldwide.
Early diagnosis and treatment can increase 5-year survival rate to
90–97 % but this figure drops below 30 % for patients with advanced or
metastatic STAD [[39]1,[40]2]. Previous study has classified gastric
carcinomas (GC) into intestinal and diffuse types, which all belong to
adenocarcinomas [[41]3] but with different pathological and
epidemiological features [[42]4]. The World Health Organization (WHO)
also provides a histopathology-based classification guideline that
divides the GC into mucinous, papillary, tubular and poorly cohesive
carcinoma [[43]5]. Although these classification methods are simple to
use, pathological results may vary differently due to complex
background (e.g. different pathogen infections) and subjective
discrimination factors [[44]6]. Currently, molecular subtypes
classified based on the biological behaviors of STAD are poorly
studied.
Studies have shown that the H. pylori is a gram-negative bacterium that
colonizes the gastric mucous environment of 60.3 % of the world's
population, especially in underdeveloped countries [[45]7]. This is
partly due to poor sanitation and high population density in these
countries, which can easily lead to oral and fecal-oral transmission of
H. pylori [[46]8]. H. pylori is largely associated with increased risk
of gastroduodenal disturbances (such as the gastroesophageal reflux
disease, peptic ulcerations disease, and inflammatory bowel disease),
particularly in mucosa-associated lymphoid tissue lymphoma, STAD and
peptic ulcer [[47]8,[48]9]. The H. pylori infects and induces the
normal mucosa to transfer chronic superficial gastritis, which further
evolves to invasive GC from chronic gastritis, metaplasia and dysplasia
[[49]10]. In addition, several factors such as high intake of dietary
salt, drinking and smoking [[50]11], genetically high gastric acid
secretion will all increase the risk of developing GC [[51]12,[52]13],
but the inflammatory response to the persistent infection of H. pylori
is a major etiological cause leading to carcinogenesis [[53]14].
Patients with polymorphisms of encoding tumor necrosis factor-α,
interleukin-1β and IL-1β receptor antagonist [[54]15,[55]16] are at
higher risk of gastric atrophy expansion, hypochlorhydria and GC after
H. pylori infection [[56]17,[57]18]. Due to a high mutational rate, H.
pylori populations are highly genetically diverse, including large
number of deletions, insertions and chromosomal rearrangements that
affect the function of housekeeping genes [[58]19]. H. pylori strains
vary in their pathogenic properties, which could result in genetic
variation in virulence factors, explaining why some infected
individuals are asymptomatic and do not have severe pathological
changes, while others exhibit peptic ulcer disease or cancers
[[59]20,[60]21].
As STAD is often diagnosed at late stage, it is essential to identify
factors that lead to the occurrence of the malignancy [[61]22,[62]23].
In addition, the next generation sequencing (NGS) technique has been
widely used in molecular profiling research of many tumors. Fabio el.
revealed the genetic and epigenetic heterogeneity of STAD by conducting
large-scale genomic and transcriptomic studies [[63]24]. The
single-cell RNA sequencing (scRNA-seq) characterizes the
transcriptional states of a single cell [[64]25], allowing unbiased
analysis of cellular characteristics in tumor tissues for exploring
tumor ecosystem [[65]26] and cancer-immune heterogeneity [[66]27]. This
study performed an unbiased and systemic transcriptomic profile
analysis using the data from The Cancer Genome Atlas (TCGA) and Gene
Expression Omnibus (GEO) databases. The scRNA-seq landscape showed that
macrophages were closely associated with the phenotype of H. pylori
infection. We further used the differentially expressed genes (DEGs)
and macrophage-related module to develop a reliable prognosis model
[[67]28] applying WGCNA, un/multivariate and LASSO cox regression
analysis. The current findings provided novel insights into the
sell-cell molecular landscape, clinical diagnosis and prognosis of
STAD.
2. Material and methods
2.1. Data acquisition and preprocessing
The RNA-seq data and clinical follow-up information of STAD patients
were downloaded from TCGA ([68]https://portal.gdc.cancer.gov/) through
the TCGA GDC API tool [[69]29]. After excluding patients without
survival time or state from the TCGA-STAD expression profile, a total
of 353 STAD samples and 53 para-cancer control samples were obtained.
The expression matrix were converted to TPM format and log2-transformed
to select protein-encoding genes. Subsequently, [70]GSE66229 containing
the expression of 300 STAD samples was retrieved from the GEO
([71]https://www.ncbi.nlm.nih.gov/geo/) [[72]30]. The probes were
mapped to the genes based on the corresponding annotation information,
and the mean value of expression was taken when multiple probes matched
to one gene. In addition, through searching the keywords of
KEGG_EPITHELIAL_CELL_SIGNALING_IN_HELICOBACTER_PYLORI_INFECTION and
HP_HELICOBACTER_PYLORI_INFECTION, 73 H. pylori infection-related genes
and the cancer-related Hallmark gene sets were collected from The
Molecular Signatures Database (MSigDB,
[73]https://www.gsea-msigdb.org/gsea/msigdb) [[74]31].
2.2. Single-cell RNA profile analysis and enrichment analysis
The scRNA-seq expression dataset ([75]GSE167297) of STAD containing 10
tumor samples were collected from the GEO [[76]30]. Genes expressed in
fewer than three individual cells and containing less than 200 or more
than 2000 genes were removed, while those with over 15 % of
mitochondrial gene expression were retained. Subsequently, the
scRNA-seq data were normalized using the “Seurat” R package [[77]32]
and FindVariableFeatures function was used for searching highly
variable genes (logfc.threshold = 0.5 and min.pct = 0.25) [[78]32].
Then, the ScaleData function was used to scale all the genes and
principal component analysis (PCA) was preformed to find the clustering
anchor (dim = 10). Batch effect among samples were removed using the
“harmony” R package, followed by using the RunTSNE function to further
reduce dimension [[79]33]. Finally, the cells were clustered by the
FindNeighbors and FindClusters functions at resolution = 0.2, and
different cell types were annotated according to marker genes.
Each cell associated with H. pylori infection was assigned with an
enrichment score using the "AUCell" package [[80]34], with a higher
AUCell score indicating a closer biological relevance between the cell
cluster and H. pylori infection. The “CellChat” R package was used for
inferring the cell-to-cell communication. Finally, to assess the
prognostic differences among patients with low and high scores, ssGSEA
was performed using "GSVA" package [[81]35] to compute the scores for
macrophage subpopulations.
2.3. Weighted gene co-expression network analysis (WGCNA)
The DEGs analysis between tumor and para-cancer control samples in the
TCGA cohort was used using the “limma” R package [[82]32] (setting
p < 0.05 and |log2Fold Change|>log2(1.5)). WGCNA was performed applying
the “WGCNA” R package to determine the gene module associated with the
phenotype of H. pylori infection. The “pickSoftThreshold” function in
the “WGCNA” R package was applied for determining the soft threshold β
(setting the sensibility to 2 and the module merge threshold to 0.3).
Gene modules with at least 50 genes were sectioned [[83]32]. The
correlation analysis between the modules and the phenotype was
performed using the Pearson method. The “clusterProfiler” R package
[[84]36] was used for the Gene Ontology (GO) and Kyoto Encyclopedia of
Genes and Genomes (KEGG) analysis.
2.4. Identifying risk genes for developing a risk model
Univariate cox regression analysis was used to identify significant
prognostic genes (p < 0.05). Then LASSO cox regression analysis was
performed applying the “glmnet” R package to primarily shrink the
number of candidate genes [[85]32]. Multivariable cox with regression
analysis and stepwise algorithm were employed to further reduce the
candidate genes and calculate the regression coefficient, respectively.
The risk model was established according to the formula of RiskScore =
[MATH: ∑(hazardcoxcoefficient*expressionofriskgene) :MATH]
.
2.5. Verification of model prognostic value
The RiskScore was calculated for all the patients and for classifying
low- and high-risk groups according to the median RiskScore value. KM
survival analysis and the ROC analysis with the Area Under Curve (AUC)
were performed using the “survminer” R package [[86]32].
2.6. Analysis of immune infiltration
CIBERSORT is an algorithm estimates the abundance of different cell
types in mixed tissues based on suppressed gene expression profiles
[[87]37]. The TIMER algorithm is mainly used to quantify various types
of immune cells in tumors [[88]38]. This study used both CIBERSORT and
TIMER algorithms to assess the differences in immune cell infiltration
in patients from different risk groups.
2.7. Cell culture and transfection
From the American Type Culture Collection (ATCC), we obtained the GC
cell line HGC27 and the gastric epithelial cell line GES-1. DEME medium
(Invitrogen) containing 10 % fetal bovine serum (Gibco, Thermo Fisher
Scientific), 50 μg/ml streptomycin and 100 U/ml penicillin was used for
cell culture in 5 % CO[2] at 37 °C [[89]39]. The siRNA was used to
silence ABCA1 applying the siRNA regent (Sangon, shanghai, china) with
forward sequence of 5′-GCGACTCCACATAGAAGAC-3′ and reversed sequence of
5′-GACGTATGTGCAGATCATA-3’. The Lipofectamine 3000 (Invitrogen) was used
for the cell transfection. After incubation for 12 h, the cell samples
were harvested for qPCR detection. Briefly, total RNA was extracted by
using the TRizol Reagent (Invitrogen) and the cDNA was synthesized by
the ReverTra Qpcr RT Master Mix kit (TOYOBO). Then the SYBR Green PCR
Master Mix (Biosystems) was applied for qRT-PCR on LightCycler 96
(Roche) according to the manufacturer's specification. Target gene
expression [[90]40] from three times sample and technique repetition
was calculated by the 2^–△△CT method, with gene β-actin as a reference.
The specific primers were listed in [91]Table S1.
2.8. Cell migration and invasion assays
Cell migration was measured by wound healing assay. 4 × 10^6 cells were
seeded into a 6-well plate (Corning) and incubated until confluent, and
then a rectilinear scratch was produced with a 100-μL pipette tip.
After 24 h, the cells were fixed by 4 % paraformaldehyde for 15 min
(min) and stained with 0.1 % crystal violet (Servicebio) for another
15 min. Next, wound closure was photographed with an inverted
microscope (Leica) [[92]40]. For invasion assays, a total of
4 × 10^4 cells were plated into the upper chamber well of 24-well
plates (Corning, 8-μm pore) containing 200 μL serum-free DMEM, while
the lower chamber was supplemented with 800 μL of DMEM containing 20 %
FBS (Thermo Fisher Scientific). After 48-h incubation, the migrating
cells were fixed by 4 % paraformaldehyde and stained by 0.1 % crystal
violet for 15 min and then imaged with an inverted microscope [[93]40].
2.9. Statistical analysis
All statistical analysis and data visualization was performed in the R
software (version 4.3.1). The Pearson method was used for the
correlation analysis. A p-value <0.05 was defined as statistically
significant. SangerBox ([94]http://sangerbox.com/home.html) provided
certain data analysis.
3. Results
3.1. The landscape of single-cell RNA of STAD samples
After cell filtering ([95]Figs. S1A and B), normalization and
dimensionality reduction clustering ([96]Fig. S1C), a total of
17397 cells from the scRNA-seq expression dataset were clustered into 8
clusters ([97]Fig. 1A) including B cells, epithelial cells, T cells,
dendritic cells, fibroblasts, macrophages, endothelial cells, and mast
cells ([98]Fig. 1B) according to the expression of marker genes
([99]Fig. 1C–[100]Table S2). The T cells and dendritic cells had a
higher proportion in each sample ([101]Fig. 1D). Analysis on the H.
pylori infection-related genes showed that macrophages had the highest
AUCell score ([102]Fig. 1E), suggesting that macrophages were closely
associated with the H. pylori infection in STAD.
Fig. 1.
[103]Fig. 1
[104]Open in a new tab
Single cell atlas of stomach adenocarcinoma. (A) TSEN plot of single
cell clustering. (B) TSEN plot of the annotated cell clusters. (C) The
bubble plot of the expression of marker genes in each cell cluster. (D)
The proportion of cell cluster in different samples. (E) The AUCell
score of H. pylori infection in each cell cluster.
3.2. Identifying H. pylori infection-associated macrophage subtypes
The macrophage population was selected to perform the t-distributed
stochastic neighbor embedding (TSNE) clustering (resolution = 0.4). The
macrophages were divided into 6 subclusters (cluster1-6, [105]Fig. 2A),
among which the C4 cluster exhibited the highest AUCell score of H.
pylori infection ([106]Fig. 2B). Subsequently, we identified the highly
variable genes between these macrophage populations and found that
several chemokines genes including the CCL3L3, CXCL5 and CCL3 were
high-expressed in the C4 subpopulation ([107]Fig. 2C). These
protein-encoding genes promoted inflammatory response through
regulating the migration and activation of leukocytes. KEGG pathway
enrichment analysis further showed that these genes were enriched in
inflammatory response, cytokine-cytokine receptor interaction, TNF and
IL-17 signaling pathways ([108]Fig. 2D).
Fig. 2.
[109]Fig. 2
[110]Open in a new tab
Macrophage clustering. (A) The t-NSE plot of macrophage clustering. (B)
The AUCell score of H. pylori infection in different macrophage
cluster. (C) The bubble plot of the expression of marker genes in
different macrophage cluster. (D) The KEGG analysis of highly variable
gene of macrophage (C4).
3.3. C4 subcluster of macrophage-mediated cell communication
In multicellular organisms, cell communication plays an important part
in cell life activity. According to the results of enrichment analysis,
it was found that the C4 macrophages functioned crucially in regulating
STAD progression. Further cell communication analysis revealed that
macrophages exhibited obvious interaction relationship with other cell
clusters ([111]Fig. 3A), with the C4 macrophages having greater
interaction intensity with other cell clusters ([112]Fig. 3B). Further
analysis of ligand-receptor information between different cell clusters
showed that the C4 macrophages affected other cell clusters, especially
some immune cell clusters (T cells and mast cells) through the
SPP1-CD44 and MIF(CD74+CD44) interaction ([113]Fig. 3C), while
epithelial cells and fibroblasts clusters affected the C4 macrophages
through the MDK-SDC2 interaction ([114]Fig. 3D).
Fig. 3.
[115]Fig. 3
[116]Open in a new tab
Cell communication analysis. (A) The interaction relationship between
the macrophage and other cell cluster. (B) The interaction strength
analysis between the macrophage and other cell cluster. (C) The
receptor-ligand interaction ways of macrophage to others cell cluster.
(D) The receptor-ligand interaction ways of immune cell clusters to
macrophage.
3.4. Identifying gene module related to C4 macrophages
We calculated the ssGSEA score in the TCGA cohort according to the
expression of the highly variable genes of C4 macrophages, and found
that STAD patients with a higher score tended to have a worse prognosis
([117]Fig. 4A). Next, WGCNA was used to identify gene module related to
C4 macrophages based on the DEGs. The soft threshold β was set at 12 to
ensure a scale-free network ([118]Fig. 4B). After hierarchical
clustering and module merging, a total of 5 co-expression modules were
obtained ([119]Fig. 4C). As the grey module cannot be merged with other
modules, it was considered as an invalid module. Further correlation
analysis showed that the green module had a strongcorrelation with the
C4 subcluster ([120]Fig. 4D). KEGG enrichment analysis revealed that
the genes in the green module were enriched in cytokine receptor
interaction, inflammatory response pathways, and cell adhesion
molecules ([121]Fig. S2A). GO analysis showed that these genes were
closely related to immune cell activation and immune response
regulation pathways in biological process ([122]Fig. S2B), protein and
immune complex pathways in cell component ([123]Fig. S2C) and receptor
bind and activity regulation pathways in molecular function ([124]Fig.
S2D).
Fig. 4.
[125]Fig. 4
[126]Open in a new tab
WGCNA for Macrophage(C4)-related gene module searching. (A) The KM
survival analysis of patients with different AUCell score of highly
variable gene of macrophage (C4). (B) Analysis of the mean connectivity
for various soft-thresholding powers for WGCNA. (C) Dendrogram of genes
clustered based on a dissimilarity measure (1-TOM). (D) The correlation
between module and feature.
3.5. Establishment of a risk classification model
To establish risk model, the samples in the TCGA-STAD cohort were
divided into the training set and test set at the ratio of 7:3, with
the [127]GSE66229 cohort as an independent validation set. The clinical
information of training set and test set was listed in [128]Table 1.
Chi-square test showed no significant difference between varying
clinical groups, indicating than our grouping was random and reliable.
Univariate, LASSO and multivariate cox regression analysis filtered 6
key prognostic genes and used them to establish a risk model:
RiskScore =
[MATH: (−0.25*T
NFRSF1B<
/mrow>)+(−0.2*C<
mi>TLA4)+(0.384*ABCA1)+(0.343*IKBIP)+(−0.564*AKAP5)+(0.4*NP<
mi>C2)
:MATH]
.
Table 1.
The clinical information of training set and test set.
Characteristics Train cohort(N = 247) Test cohort(N = 106)
Total(N = 353) pvalue FDR
Age
Mean ± SD 65.52 ± 10.53 65.50 ± 10.88 65.51 ± 10.62
Median[min-max] 67.00[35.00,90.00] 68.00[41.00,90.00]
67.00[35.00,90.00]
Gender 0.63 1
FEMALE 85(24.08 %) 40(11.33 %) 125(35.41 %)
MALE 162(45.89 %) 66(18.70 %) 228(64.59 %)
AJCC stage 0.43 1
I 29(8.22 %) 19(5.38 %) 48(13.60 %)
II 74(20.96 %) 35(9.92 %) 109(30.88 %)
III 107(30.31 %) 39(11.05 %) 146(41.36 %)
IV 25(7.08 %) 10(2.83 %) 35(9.92 %)
unknown 12(3.40 %) 3(0.85 %) 15(4.25 %)
Grade 0.54 1
G1 8(2.27 %) 1(0.28 %) 9(2.55 %)
G2 86(24.36 %) 42(11.90 %) 128(36.26 %)
G3 147(41.64 %) 60(17.00 %) 207(58.64 %)
unknown 6(1.70 %) 3(0.85 %) 9(2.55 %)
Status 0.92 1
Alive 146(41.36 %) 64(18.13 %) 210(59.49 %)
Death 101(28.61 %) 42(11.90 %) 143(40.51 %)
OS.time
Mean ± SD 608.00 ± 527.65 625.69 ± 594.02 613.31 ± 547.63
Median[min-max] 476.00[3.00,3540.00] 406.00[20.00,3720.00]
468.00[3.00,3720.00]
[129]Open in a new tab
3.6. Validation of the effectiveness of the model classification
Based on the optimal cutoff point, the patients in training set were
divided into high-risk and low-risk groups. It was found that high-risk
patients had a poor prognosis ([130]Fig. 5A), with an AUC value of
0.67, 0.72, 0.71 and 0.65 for 1-, 2-, 3- and 4-year survival,
respectively, which indicated that a high accuracy of the RiskScore in
long- and short-term prediction and classification ([131]Fig. 5A).
Similar survival results were also observed in the test set ([132]Fig.
5B) and the TCGA cohort ([133]Fig. 5C), and ABCA1, IKBIP and NPC2 were
high-expressed in high-risk group ([134]Fig. 5C). We further evaluated
the accuracy and robustness of the model in the [135]GSE66229
validation set. The results showed that the patients with a higher
RiskScore had the worst prognosis and shorter survival time, with an
AUC of 1- to 5-year survival rate higher than 0.65 ([136]Fig. 5D),
which demonstrated that the RiskScore was highly effective in
predicting long- and short-term prognosis of STAD ([137]Fig. 5D).
Fig. 5.
[138]Fig. 5
[139]Open in a new tab
Validation of model prognostic performance. (A) KM survival and ROC
analysis of varying patients in training set. (B) KM survival and ROC
analysis of varying patients in test set. (C) KM survival, ROC and
living time analysis of varying patients in TCGA cohort. (D) KM
survival, ROC and living time analysis of varying patients in
validation cohort.
3.7. Identifying independent prognostic factors and establishing a nomogram
The RiskScore and other clinical factors were incorporated into
univariate cox regression analysis, which showed that the Age, American
Joint Committee on Cancer (AJCC) stage and RiskScore were significant
influencing factors for STAD prognosis (p < 0.05, [140]Fig. 6A).
Multivariate cox regression analysis also proved that these three
factors were independent prognostic factors (p < 0.05, [141]Fig. 6B).
To further improve the risk assessment and survival prediction for
STAD, the Age, AJCC stage and RiskScore were combined to develop a
nomogram model ([142]Fig. 6C). The results showed that the RiskScore
had the greatest influence on predicting patients’ survival ([143]Fig.
6C). The calibration curve of nomogram presented that the 1-, 3- and 5-
year calibration curve was close to the standard curve ([144]Fig. 6D),
suggesting that the nomogram model had an excellent prognostic
prediction performance. In the decision curve analysis (DCA), the net
benefit of the nomogram and RiskScore was obviously higher than the
extreme curve ([145]Fig. 6E), indicating that the current prognostic
model had the strongest survival prediction ability.
Fig. 6.
[146]Fig. 6
[147]Open in a new tab
Independent factor and nomogram developing. (A) Univariate cox
regression analysis for significant prognostic factors. (B)
Multivariate cox regression analysis for significant independent
prognostic factors. (C) A nomogram developing. (D) Calibration curve of
nomogram model. (E) Decision curve analysis (DCA) of nomogram model.
3.8. Immune infiltration and pathway activation difference
Differences in tumor microenvironment (TME) were compared based on the
immune infiltration score. CIBERSOER analysis showed that naïve B
cells, CD4^+ memory activated T cells, CD8^+ T cells were infiltrated
in low-risk group, while neutrophils and macrophage M2 were infiltrated
in high-risk group ([148]Fig. S3A). Higher macrophage M2 infiltration
suggests the presence of immunosuppressive TME [[149]41]. Similar
results were observed in TIMER analysis, as CD4^+ T cells, B cells, and
CD8^+ T cells were enriched in the low-risk group ([150]Fig. S3B).
Further analysis showed that the RiskScore was positively related to
tumorigenesis and development pathways such as angiogenic hypoxia,
Notch signaling, epithelial mesenchymal transition (EMT) and negatively
correlated with immune and cell cycle pathways ([151]Fig. S3C). This
suggested that STAD patients with a higher RiskScore were more prone to
activate these typical cancer activation-related pathways.
3.9. ABCA1 mediated the migration and invasion of tumor cells
Finally, we examined the expression of model genes and the role of
ABCA1 in cell migration and invasion. The results of qPCR showed that
these genes including ABCA1, CTLA4, IKBIP, NPC2 and TNFRSF1B were
significantly overexpressed in the HGC27 cancer cells (p < 0.05,
[152]Fig. 7A–E), while AKAP5 was significantly downregulated in the
HGC27 cells in comparison to that in the epithelial cell line GES-1
([153]Fig. 7F). In addition, the invasion ability of cancer was greatly
inhibited after ABCA1 silencing and the number of blue migrated cells
in the si-ABCA1 was significantly lower than that in the si-NC groups
(p < 0.05, [154]Fig. 7G). Wound healing assay revealed that silencing
ABCA1 significantly affected the rate of wound closure, which was
significantly reduced in the si-ABCA1 groups ([155]Fig. 7H).
Fig. 7.
[156]Fig. 7
[157]Open in a new tab
qPCR, wound healing and invasion assay. (A–F) The expression of model
genes in cancer cells and epithelial cells. (G) Wound healing assay for
cell migration. (H) Trans-well assay for cell invasion. *p < 0.05,
**p < 0.01, ***p < 0.001.
4. Discussion
Infection of H. pylori could easily cause corpus-predominant gastritis
and has been considered a high-risk factor that significantly promotes
the occurrence of STAD [[158]42]. Some infected patients develop
gastritis, while others with H. pylori infection may have gastric
cancer [[159]43]. Many patients with gastric inflammation are
asymptomatic, and gastritis symptoms in certain types of patients are
more persistent or will recur after eradication treatment [[160]44].
Varied clinical outcomes could be explained by multiple factors
[[161]28] such as virulence factors, but studies demonstrated that the
pathogenic mechanisms of H. pylori infection are more complex than
generally accepted [[162]45]. This study established a risk prognosis
model according to the H. pylori infection phenotype to explore the
potential pathogenic mechanism of H. pylori risk factor at molecular
level and the model was able to achieve precise risk stratification for
STAD patients. The nomogram model developed based on the RiskScore,
AJCC stage and age can be used to accurately assess the survival
probability of patients with STAD, meanwhile, the RiskScore also acted
as an indicator of cancer activation-related pathways. Our finding
could assist the clinical diagnosis and treatment of STAD patients.
Using the AUCell algorithm, the current results demonstrated that H.
pylori was closely associated with macrophages. The interaction between
H. pylori and macrophages plays a significant role in the progression,
pathogenesis, and suppression of the infection [[163]46]. In a mouse
model, Zhuang et al. found that the NF-kappa B pathway is involved in
the macrophage response to H. pylori and produces IL-6, IL-23, and
CCL20, which in turn inhibit the NF-kappa-B pathway in macrophages and
ultimately lead to reduced differentiation of Th17 cells [[164]47]. Wen
et al. discovered that in a co-culture model of macrophages and H.
pylori, the use of γ-secretase to inhibit Notch signaling causes a
downregulation of the expression of inducible nitric oxide synthase
(iNOS) and its product, nitric oxide (NO). Such an intervention reduces
the secretion of pro-inflammatory cytokines and suppresses the
phagocytic and bactericidal functions of macrophages against H. pylori
[[165]48]. These results further strengthened a strong association
between H. pylori and macrophages in STAD, providing new insights into
the clinical management of STAD patients. Based on the coefficient of
genes in the RiskScore model, TNFRSF1B and CTLA4 were regarded as the
protective factors (coefficient <0), while the ABCA1, IKBIP, AKAP5 and
NPC2 were regarded as the risk factors (coefficient >0). TNFRSF1B is a
tumor necrosis factor receptor that recognizes their cognate ligands
(TNF) and promotes the differentiation, clonal expansion and survival
of antigen-primed CD8 and CD4 T cells, mediating adaptive immunity to
kill cancer cells [[166]49]. Cytotoxic T lymphocyte antigen 4 (CTLA4)
is an immune checkpoint molecule. Targeting CTLA4 is widely used to
activate anti-cancer immune response through stimulating T cell
activation [[167]50]. Guan et al. indicated that CTLA4 enhances
macrophage recruitment and increases macrophage proportion in
glioblastoma [[168]51]. This indicates that CTLA4 may influence tumor
immunosuppression and progression of STAD by affecting macrophage
polarization and, in turn, tumor immunosuppression. ATP binding
cassette protein A1 (ABCA1) is a crucial molecule in cholesterol
homeostasis, and the expression of ABCA1 is upregulated during the EMT
of breast cancer to promote the metastatic capacity of tumor [[169]52].
We also found that the ABCA1 was overexpressed in the gastric cancer
cells, and that its silencing affected the migration and invasion of
tumors. Notably, a significantly positive correlation between the
RiskScore and EMT indicated that EMT increased tumor spread and
metastasis of STAD, and that H. pylori infection further exacerbated
the EMT process through the activating EMT-related signaling pathways
with ABCA1 as a crucial contributor during the process. I kappa B
kinase interacting protein (IKBIP) is a biomarker that maintains
abnormal proliferation of the glioblastoma cells through suppressing
the ubiquitination and degradation of CDK4 [[170]53]. Also, IKPIP is an
immunosuppressive microenvironment biomarker of digestive system
malignancies [[171]54]. Zhong el. revealed that the expression of
protein kinase A-anchoring protein 5 (AKAP5) is upregulated and is
closely associated with the clinical stages in the STAD, whereas
low-expressed AKAP5 can act as a protective factor [[172]55]. In
addition, some researchers found that the recruitment of immature
macrophages to the TME in lung cancer is inhibited by NPC2, and
confirmed that NPC2 is secreted by tumor cells and absorbed by immature
macrophages [[173]56]. Macrophages play a dual role in gastritis
induced by H. pylori, ulcers and gastric cancer. On one hand, H. pylori
induces macrophage polarization to promote inflammatory responses and
eliminate H. pylori, on the other hand, macrophages are key cells in
the primary immune response against H. pylori infection [[174]46].
Analytical method validation is the process used to prove that a test
method consistently yields what it is expected to do, and its purpose
is to establish that an accurate, precise, and rugged method has been
developed [[175]57]. Further experimentals showed that ABCA1 promoted
cancer cell migration and invasion. Taken together, the six prognostic
genes identified in this study could influence the process by promoting
macrophage-tumor cell interactions and their interactions with other
immune cells.
Previous studies have developed long non-coding RNA (lncRNA) signatures
related to H. pylori infection for predicting the prognosis of gastric
cancer patients [[176]58]. Zheng et al. integrated multiple gastric
cancer cohorts and identified 28 key prognostic genes for evaluating
the risk for patients [[177]59]. However, our model only contained 6
prognostic genes with a high classification effectiveness, indicating
that our model had greater potential for clinical application. Although
the current model had certain prognostic value, there were also some
limitations. Firstly, all the datasets were public databases and more
sample data from different populations are required to achieve greater
generalization. Secondly, the mechanisms through which these model
genes affected macrophages in the STAD microenvironment and their role
in H. pylori infection have not been validated. Therefore, in the
future, in vivo experiments should be carried out to explore the
specific biological functions of these genes. In addition, bias was
inevitable due to the heterogeneity of individual tumor, therefore
large-scale clinical trials are needed to provide individualized
treatment for STAD patients.
5. Conclusion
This study analyzed the single-cell profile of gastric cancer and
annotated a total of 8 cell clusters, among which T cells and dendritic
cells accounted for the highest proportion in tumor tissues and the C4
macrophages were closely associated with the H. pylori infection
phenotype. Furthermore, a risk prognostic model related to H. pylori
infection was developed based on the DEGs and the module genes related
to C4 subtype. Our model exhibited short- and long-term prognostic
value and patients with higher RiskScore were prone to activate typical
cancer activation-related pathways. A nomogram model was developed to
accurately predict the survival probability of STAD patients. The
present discoveries provided new insights for the diagnosis and
management of patients with STAD.
Ethics approval and consent to participate
Not applicable.
Consent for publication
Not applicable.
Availability of data and material
The datasets generated during and/or analyzed during the current study
are available in the GSE repository [[178]GSE66229]
([179]https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=
[180]GSE66229) and repository [[181]GSE16279]
([182]https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=
[183]GSE16279).
Funding
This work was supported by Shaanxi Provincial Health Research Fund
(2021D036).
CRediT authorship contribution statement
Jing Zhou: Writing – review & editing, Writing – original draft,
Software, Resources, Project administration, Formal analysis, Data
curation, Conceptualization. Li Guo: Writing – review & editing,
Writing – original draft, Validation, Software, Resources, Methodology,
Investigation. Yuzhen Wang: Supervision, Resources, Project
administration, Investigation, Formal analysis. Lina Li: Validation,
Resources, Project administration, Methodology, Formal analysis. Yahuan
Guo: Visualization, Validation, Resources, Methodology, Investigation.
Lian Duan: Visualization, Software, Project administration,
Methodology, Investigation. Mi Jiao: Visualization, Supervision,
Project administration, Investigation, Formal analysis, Data curation.
Pan Xi: Visualization, Supervision, Resources, Methodology, Formal
analysis, Data curation. Pei Wang: Writing – review & editing,
Visualization, Validation, Supervision, Software, Funding acquisition,
Formal analysis, Conceptualization.
Declaration of competing interest
The authors declare that they have no known competing financial
interests or personal relationships that could have appeared to
influence the work reported in this paper.
Acknowledgements