Skip to main content

Transcriptome profiling of patient-derived tumor xenografts suggests novel extracellular matrix-related signatures for gastric cancer prognosis prediction

Abstract

Background

A major obstacle to the development of personalized therapies for gastric cancer (GC) is the prevalent heterogeneity at the intra-tumor, intra-patient, and inter-patient levels. Although the pathological stage and histological subtype diagnosis can approximately predict prognosis, GC heterogeneity is rarely considered. The extracellular matrix (ECM), a major component of the tumor microenvironment (TME), extensively interacts with tumor and immune cells, providing a possible proxy to investigate GC heterogeneity. However, ECM consists of numerous protein components, and there are no suitable models to screen ECM-related genes contributing to tumor growth and prognosis. We constructed patient-derived tumor xenograft (PDTX) models to obtain robust ECM-related transcriptomic signatures to improve GC prognosis prediction and therapy design.

Methods

One hundred twenty two primary GC tumor tissues were collected to construct PDTX models. The tumorigenesis rate and its relationship with GC prognosis were investigated. Transcriptome profiling was performed for PDTX-originating tumors, and least absolute shrinkage and selection operator (LASSO) Cox regression analysis was applied to extract prognostic ECM signatures and establish PDTX tumorigenicity-related gene (PTG) scores. The predictive ability of the PTG score was validated using two independent cohorts. Finally, we combined PTG score, age, and pathological stage information to establish a robust nomogram for GC prognosis prediction.

Results

We found that PDTX tumorigenicity indicated a poor prognosis in patients with GC, even at the same pathological stage. Transcriptome profiling of PDTX-originating GC tissues and corresponding normal controls identified 383 differentially expressed genes, with enrichment of ECM-related genes. A robust prognosis prediction model using the PTG score showed robust performance in two validation cohorts. A high PTG score was associated with elevated M2 polarized macrophage and cancer-associated fibroblast infiltration. Finally, combining the PTG score with age and TNM stage resulted in a more effective prognostic model than age or TNM stage alone.

Conclusions

We found that ECM-related signatures may contribute to PDTX tumorigenesis and indicate a poor prognosis in GC. A feasible survival prediction model was built based on the PTG score, which was associated with immune cell infiltration. Together with patient ages and pathological TNM stages, PTG score could be a new approach for GC prognosis prediction.

Background

Gastric cancer (GC) is the third leading cause of cancer-related deaths in China. [1, 2] Approximately 44% of GC cases and 48.6% of GC mortalities have been reported in China [1, 3]. Additionally, over the past 35 years in China, the incidence and mortality of GC have remained high, with an overall 5-year survival rate of less than 50% [4]. One of the crucial barriers to GC-targeted therapies is intra-tumor, intra-patient, and inter-patient heterogeneity [5, 6]. Fortunately, patient-derived tumor xenograft (PDTX) models can recapitulate the molecular diversity of donor tumors, making in vivo translational research feasible [7,8,9]. Therefore, PDTX models are now widely used in biomarker development, preclinical drug screening, and personalized clinical decision making [7,8,9]. In addition, PDTX models can be used to investigate the biological functions of donor tumors.

The tumor microenvironment (TME) is comprised of extracellular matrix (ECM), blood vessels, neurons, immune cells, cancer-associated fibroblasts (CAFs), and other non-malignant cells, and it plays a key role in regulating tumor cell proliferation, drug resistance, and metastasis [10,11,12]. Among the various components of the TME, the ECM plays a major role. The ECM is a complex system that provides mechanical support for the TME, modulates growth factor secretion, and mutually interacts with tumor and immune cells in the TME [13,14,15,16]. Specifically, the ECM contains various macromolecules with distinctive physical, biochemical, and biomechanical properties. During tumorigenesis, the ECM system is deregulated to favor the generation of a tumorigenic microenvironment that enhances tumor-associated angiogenesis and inflammation [15]. The ECM goes through remodeling during tumor development. This remodeling contributes to the establishment of a premetastatic niche by reorganizing or degrading the pre-existing matrix architecture or by stimulating local matrix secretion [17]. Accordingly, the investigation of ECM-related molecular characteristics is of great value for personalized GC treatment.

In general, GC stages and histotypes are diagnosed based on morphological characteristics at the cellular or tissue level [2]. In clinical practice, only a few molecular markers are commonly examined using immunohistochemistry (IHC) and fluorescence in situ hybridization (FISH) for treatment guidance. According to the National Comprehensive Cancer Network (NCCN) and The Chinese Society of Clinical Oncology (CSCO) clinical guidelines [18, 19], human epidermal growth factor receptor 2 (HER2) status testing is recommended for all patients with GC; programmed death protein-1 (PDL1) expression and microsatellite instability (MSI)/mismatch repair (MMR) status evaluation is recommended for patients with GC who are to undergo immunotherapy; and neurotrophic tropomysin-related kinase (NTRK) gene fusion testing is recommended for patients with GC who have failed to respond to standard treatment. In addition to these routinely tested molecules, several genomic classifications have been reported, including Epstein-Barr virus (EBV)-positive GC, characterized as EBV-encoded-RNA-positive [20]. However, such a small number of molecular tests is not sufficient to help clinicians fully understand the transcriptomic characteristics of patients with GC. In addition, IHC and FISH can only achieve a semi-quantitative determination of molecular expression levels, and they do not provide precise expression information for these markers. However, GC is well known for its intra-tumor and inter-individual heterogeneity [5], and prognosis predictions that simply rely on morphological and histological information may result in inaccurate prognosis prediction and failure in post-surgical treatment.

Recent research has shown that, even in patients at the same stage and subtype, the transcriptomic landscapes of primary tumors can be distinguished from one another [21]. We noted a divergence in tumorigenic potential among GC patients with the same clinical characteristics and pathological stage diagnosis. Because PDTX models recapitulate the molecular diversity of donor tumors, they mimic the inter-individual heterogeneity seen in patients with GC. Therefore, we investigated whether heterogeneity exists at the transcriptomic level among these patients and whether gene expression signatures can be adopted to improve GC prognosis. Our findings will be helpful in the prediction of GC prognosis and will provide new insights into research on therapeutic targets for GC.

Methods

PDTX model construction

Fresh GC and adjacent normal control tissues (at least 2 cm from the matched GC tissues) were surgically resected. According to previously reported methods [22], GC tissues were cut into 3 mm3 pieces and implanted subcutaneously in 6-week-old non-obese diabetes/severe combined immunodeficiency (Nod/Scid) mice. We observed and recorded the tumor size in the mice every 3 days. When the tumor size increased to 1000 mm3, we stopped our observation and harvested the tumor for further study. Implanted tumors that showed no growth within 6 months were categorized into the non-tumorigenesis group.

Sample and data collection

In total, 1003 GC samples with RNA expression data and clinicopathological information from four cohorts (n = 122, 300, 387, and 194) were included in this study. Of these, 122 samples used for PDTX model construction were treated at PKUCH (Beijing, China). The Asian Cancer Research Group (ACRG) cohort, which consists of 300 cancer samples, was used as the training cohort [23], and all 300 samples were used for model construction. The Cancer Genome Atlas (TCGA) stomach adenocarcinoma (STAD) cohort and 198 patients with GC treated at PKUCH were used as validation cohorts [24]. All 387 samples in the TCGA cohort and 194 samples in the PKUCH validation sets, with available overall survival (OS) data, were used for OS analysis and 304 samples in the TCGA cohort and 194 samples in PKUCH, with available progression-free survival (PFS) data, were used for PFS analysis. RNA expression data and clinical information for patients in the TCGA cohort were downloaded from the University of California Santa Cruz (UCSC) website (https://xenabrowser.net/datapages/, accessed on February 28, 2023). Correlated data for the ACRG cohort were downloaded from the Gene Expression Omnibus (GEO) database (https://www.ncbi.nlm.nih.gov/geo/, accessed on February 28, 2023). The 5-year survival and clinical data for the TCGA cohort were collected from the UCSC website (https://xenabrowser.net/datapages/, accessed on February 28, 2023). The 5-year survival and clinical data for the ACRG cohort were collected from supplementary data files in the literature [23]. The 5-year survival and clinical data for the PKUCH cohort were provided by the corresponding author [24]. All patients provided written informed consent and the Institutional Review Board of PKUCH approved this study (2019KT11).

Whole tissue RNA-sequencing

  1. 1.

    RNA extraction

    Total RNA was extracted from the tissues using TRIzol Reagent (Invitrogen Life Technologies, Carlsbad, CA, USA) according to the manufacturer’s instructions. The RNA quality and integrity were determined using a NanoDrop spectrophotometer (Thermo Fisher Scientific, Waltham, MA, USA). Only high-quality RNA samples (OD260/280 = 1.8~2.2, OD260/230 ≥ 2.0) were used to construct the sequencing library.

  2. 2.

    Library preparation and sequencing

    The library was prepared using 3 μg of total RNA. First, mRNA was isolated according to the polyA selection method using oligo(dT) beads, and then fragmented using divalent cations at elevated temperatures in a proprietary fragmentation buffer (Illumina, San Deigo, CA, USA). Second, first-strand cDNA was synthesized using random oligonucleotides and Super Script II, and second-strand cDNA synthesis was subsequently performed using DNA polymerase I and RNase H. The remaining overhangs were then converted into blunt ends via exonuclease/polymerase activities, after which, the enzymes were removed. After adenylation of the 3′ ends of the DNA fragments, Illumina paired-end adapter oligonucleotides were ligated to prepare them for hybridization. Next, to select cDNA fragments of the preferred size (400–500 bp), the library fragments were purified using an AMPure XP system (Beckman Coulter, Brea, CA, USA). DNA fragments with ligated adaptor molecules at both ends were selectively enriched using an Illumina PCR Primer Cocktail in a 15 cycle PCR reaction. The products were purified (AMPure XP system) and quantified using an Agilent High-Sensitivity DNA Assay on a Bioanalyzer 2100 system (Agilent Technologies, Santa Clara, CA, USA). Finally, paired-end RNA-sequencing (RNA-seq) libraries were sequenced using a NovaSeq 6000 sequencer (Illumina; 2 × 150 bp read length).

  3. 3.

    Quality control and read mapping

    Quality control was performed on raw paired-end reads using FastQC (v0.11.9, https://www.bioinformatics.babraham.ac.uk/projects/fastqc/), with default parameters, and trimmed using Trim Galore (v0.6.7, https://www.bioinformatics.babraham.ac.uk/projects/trim_galore/), also with default parameters. The clean reads were aligned to the reference genome (hg38) using STAR software [25]. Next, genes were annotated using the “gencode.v38.primary_assembly.annotation.gtf” file downloaded from the GENCODE website (https://www.gencodegenes.org) and gene expression levels were quantified using RSEM software, with the default parameters [26]. Raw RNA-seq datasets (fastq files) were uploaded to the National Genomics Data Center (NGDC, https://ngdc.cncb.ac.cn) under accession number HRA004403.

  4. 4.

    Differential expression analysis and functional enrichment

    To identify differentially expressed genes (DEGs) between GC and normal samples, differential expression analysis was performed using the DESeq2 package. DEGs with |log2 fold change| ≥ 0.678 and P < 0.05 were considered to be significant. In addition, functional enrichment analysis of Gene Ontology (GO) terms was performed with a Benjamini-and-Hochberg-corrected P-value threshold of 0.01 and an overlapped gene number threshold of 10. GO functional enrichment analysis was performed using KOBAS [27].

STRING protein–protein interaction analysis

DEGs were submitted to the STRING database [28] to identify associated genes (90% confidence, 5% false discovery rate). Interactions were examined using curated databases, experimentally determined gene neighborhoods, gene fusions, gene co-occurrence, text mining, co-expression, and protein homology.

Establishment and validation of the PDTX-tumorigenicity-related gene score for GC prognosis prediction

In the training set, 383 unique DEGs between GC and normal tissues in the tumorigenesis samples were included to train a univariate Cox regression model using the “survival” R package [29]. In total, 116 candidate prognostic genes were identified. Next, least absolute shrinkage and selection operator (LASSO) Cox regression models were used to identify the most robust markers related to survival. Four PDTX tumorigenicity-related genes (PTGs) were integrated to construct a predictive signature for the PTG score.

$$PTG\,score=\sum \left(LASSO\,coefficient\,of\,RNAi\times\,RNAi\,expression\right).$$

The best PTG score cut-off value was determined by the median value to enhance accuracy. The samples were classified into low- and high-PTG groups based on their median values. We then compared OS and PFS between the two groups to validate the prognostic predictive values using the ACRG, TCGA and PKUCH validation cohorts.

Characterization of immune cell infiltration

To depict the immune cell infiltration landscape of GC patients with low and high PTG scores, the CIBERSORTx algorithm was used to assess the abundance of 22 immune-infiltrating cells [30]. The TIMER2.0 database (http://timer.cistrome.org) was used to estimate the proportion of immune cell infiltration.

Stromal and immune score analysis

To evaluate the stromal and immune scores of GC samples, we employed the ESTIMATE (Estimation of STromal and Immune cells in MAlignant Tumors using Expression data) algorithm using the “estimate” R package [31]. The correlation between the PTG score and stromal score, the ESTIMATE score, and tumor purity were visualized using the “ggplot2” R package.

Establishment and validation of the integrative nomogram

Multivariate Cox regression analysis was used to assess the correlation between clinicopathological features and PTG scores in the ACRG cohort. Based on the multivariate analysis results, the significant factors (P < 0.05) were subsequently used as inputs to develop a predictive nomogram utilizing the “rms” R package. The predictive accuracy of the nomogram scoring system was evaluated by receiver operating characteristic (ROC) analysis for 1-, 3-, and 5-year survival rates. Calibration curves were used to depict the consistency between predicted survival events and actual observations.

Statistical analysis

Student’s t-tests were used to compare continuous variables between the two groups, and chi-square tests were used to compare categorical variables. The Kaplan–Meier method was used to compare the OS and PFS between patient subgroups. Univariate and multivariate Cox regression models were used to evaluate the independent prognostic value of the PTG signature. A P ≤ 0.05 was considered a statistical significance. All statistical analyses were performed using R software (v4.2.2).

Results

PDTX tumorigenicity indicates poor prognosis in patients with GC

We collected 122 GC cases to construct the PDTX model in Nod/Scid mice. Of the 122 cases, 74 successfully developed tumors in mice, namely the tumorigenesis group, and 48 failed to develop tumors, namely the non-tumorigenesis group (Fig. 1A–E; Additional file 1: Fig. S1). We then investigated the patient composition of the two groups (Fig. 2A–G). However, there were no significant differences in the clinical characteristics between the groups. Interestingly, of the four patients at the T1 stage, only one was in the tumorigenesis group. This suggests that GC at the T1 stage may not be suitable for PDTX-model-based investigations. We explored the prognosis of all 122 patients. Patients whose tumor samples successfully developed the PDTX model exhibited a poor prognosis (Fig. 2H–I). Next, to rule out the effects of tumor stage and patient age on prognosis, we performed prognosis analysis for each TNM stage. The results showed that, even at the same stage or within the same age range, patients with successful tumorigenicity exhibited poorer prognosis (Additional file 1: Fig. S2). Univariate and multivariate Cox regression analyses were performed to explore the prognostic predictive ability of tumorigenicity. The results showed that tumorigenicity was an independent risk factor in patients (Fig. 2J; Additional file 2: Table S1). Therefore, we hypothesized that the PDTX model could be used as an in vivo tool for predicting GC prognosis (Table 1).

Fig. 1
figure 1

Outline of the study flow. A Schematic diagram of the study flow. B, D Representative hematoxylin–eosin (HE) staining of gastric cancer (GC) tissue from the tumorigenesis group. C, E Representative HE staining of patient-derived tumor xenograft (PDTX) tumor tissue

Fig. 2
figure 2

PDTX models of tumorigenicity correlate with GC patient survival. AG Correlation analysis of PDTX tumorigenesis and clinical characteristics. H, I Kaplan–Meier curves for overall survival (OS) and progression-free survival (PFS) of 122 GC patients. J Multivariate Cox regression analysis of tumorigenicity and clinical characteristics of GC. *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001

Table 1 Clinical characteristics of 122 patients with GC for PDTX models

However, owing to the limited amount of GC tissue available for diagnosis, it is not feasible to perform a PDTX model test for every GC patient. Therefore, we wondered if there were unique molecular signatures in PDTX-originating tumor (POT) tissues that regulate tumor cell tumorigenesis and can be applied for GC prognosis prediction.

ECM-related gene signatures contribute to tumorigenicity

To uncover the underlying molecular signatures in POT tissues, we performed RNA-seq analysis of POT tissues (GC) and their corresponding normal tissues (control) from 19 patients (Fig. 1A; Additional file 2: Tables S2–S3). Compared to the normal control tissue, we identified 419 DEGs in the GC tissue of the tumorigenesis group and 41 DEGs in the GC tissue of the non-tumorigenesis group (Fig. 3A–D). By overlapping the two groups of DEGs, we found 383 unique DEGs in the tumorigenesis group, but only five unique DEGs in the non-tumorigenesis group. To explain the stark contrast between the numbers of unique DEGs in these two groups, we inferred that there was less molecular heterogeneity between GC and control tissues in the non-tumorigenesis group. Furthermore, we further investigated the five unique DEGs in the non-tumorigenesis group to identify possible connections among them. The five DEGs identified were CENPF (centromere protein F), LIPF (lipase F, gastric type), PELATON (plaque enriched long non-coding RNA (lncRNA) in atherosclerotic and inflammatory bowel macrophage regulation), PRRT3-AS1 (PRRT3 antisense RNA1), and SMKR1 (small lysine rich protein 1). We performed protein–protein interaction analysis of the three protein-coding genes [28], and the results showed no significant interaction between them (Additional file 1: Fig. S3A). We found no published evidence for the association of the lncRNAs, PELATON and PRRT3-AS1, with GC survival; therefore, we did not focus on these five DEGs that were only found in the non-tumorigenesis group. In addition, there were 36 DEGs in both the tumorigenic and non-tumorigenic groups. To examine the molecular functions of these 36 DEGs, we performed GO enrichment analysis of biological processes, cellular components, and molecular function terms. Nineteen terms showed significant corrected enrichment P-values (Additional file 1: Fig. S3B). The enriched biological processes, cellular components, and molecular functions can be summarized into the following five main categories: ion transport and homeostasis, steroid-regulated pathways, digestion, creatine metabolism, and aspartic-type endopeptidase activity. These terms are highly related to cell metabolism, which implies that these 36 DEGs may play important roles in maintaining homeostasis in primary GC cells, but do not determine the tumorigenic ability of these cells in mice.

Fig. 3
figure 3

Transcriptome profiling of PDTX-model-originating patients. A, B Differentially expressed genes (DEGs) between GC tissue and normal tissue in the tumorigenesis group (A) and non-tumorigenesis group (B). Red dots represent highly expressed genes in GC tissue with P < 0.05 and a fold change > 1.6, and blue dots represent down-regulated genes in GC tissue with P < 0.05 and a fold change < 0.625. C, D Heatmap of DEG expression levels in the tumorigenesis group (C) and non-tumorigenesis group (D). E Venn diagram of DEGs in the tumorigenesis and non-tumorigenesis group. FH Gene Ontology (GO) enrichment bar plots of 383 unique DEGs in the tumorigenesis group for cellular component (red), biological process (green), and molecular function (blue) terms. Colored titles are extracellular matrix (ECM)-related pathways

To further investigate the molecular signatures contributing to a poor prognosis of GC, we used the 383 unique DEGs in the tumorigenesis group (Fig. 3E) for GO enrichment analysis. As shown in the enrichment plots, of the top 10 enriched pathways in cellular component, biological process, and molecular function terms, 19 ECM-related pathways showed significant corrected enrichment P-values (Fig. 3F–H). ECM signatures in cancer have been reported to be responsible for regulating the immune response, metabolism, intracellular signaling, EMT, and metastasis [17], which strongly suggests that the ECM of GC may contribute to tumorigenesis in PDTX models. However, the GO terms of extracellular space, plasma membrane, and identical protein binding were also enriched for 36 DEGs (Additional file 1: Fig. S3B); therefore, rather than including all genes of the 19-ECM related pathways, we only selected 299 DEGs from the 383 unique DEGs enriched in the 19 ECM-related terms for further analysis.

Construction of PTG-related prognosis signatures in GC

Due to the limited sample size of our RNA-seq data, we chose the ACRG cohort dataset to perform Cox regression analysis to extract ECM-related gene expression signatures. Using a univariate Cox regression analysis, we identified 116 candidate genes associated with GC prognosis (Fig. 4A). The expression levels of 116 candidate genes were highly correlated (Additional file 1: Fig. S4A), which implied the probability of variable interaction effects among these genes and intermolecular influences in predicting prognosis. LASSO Cox regression analysis was performed to identify the most robust and non-redundant genes that predicted GC prognosis (Fig. 4B). Finally, four genes (RBPMS2, ORM1, ESM1, and PLEKHS2) with individual non-zero LASSO regression coefficients were screened and integrated to establish a PTG score model (Fig. 4C).

Fig. 4
figure 4

Construction of PTG signatures. A The hazard ratio of 116 candidate prognostic genes in the Asian Cancer Research Group (ACRG) cohort. B The most robust predictive genes were identified by the least absolute shrinkage and selection operator (LASSO) Cox regression algorithm. C An ensemble of four genes remained with non-zero LASSO Cox coefficients. PTG, PDTX-tumorigenicity-related gene

To measure the prognostic value of our PTG signature, we investigated the relationship between PTG scores and 5-year OS in an independent ACRG cohort. With the median PTG score set as the cutoff, patients were assigned to high-risk or low-risk group (Fig. 5A). In the validation cohort, high PTG scores were associated with poor survival (Fig. 5B). Kaplan–Meier curves showed that patients with higher PTG scores had poorer OS and PFS (Fig. 5C–D). The areas under the ROC curve (AUCs) of the PTG score for 1-, 3-, and 5-year OS were 0.72, 0.76, and 0.75, respectively (Fig. 5E). In addition, to highlight the differences in the expression patterns of PTG-related genes, we performed principal component analysis (PCA) based on the PTG-related genes of the low- and high-risk groups. The scatter plot showed that the PTG expression patterns were substantially different between the two groups (Fig. 5F).

Fig. 5
figure 5

Prognosis analysis in the training cohort. A, B The distribution and median values of PTG scores. C, D Kaplan–Meier curves for the OS and PFS of GC patients in the low and high PTG score groups. E Area under the curve (AUC) of time-dependent receiver operating characteristic (ROC) curves at 1, 3 and 5 years. F Principal component analysis (PCA) analysis showed different distribution patterns in the low and high PTG score groups. The ellipse represents the 95% confidence interval

Next, we used the TCGA STAD and PKUCH cohorts to further validate the feasibility of PTG score for predicting GC prognosis. To maintain consistency with the training cohort, the cutoff values for the low- and high-risk groups were determined based on the median PTG scores. The results of prognostic analysis were consistent with those of the ACRG training cohort. Kaplan–Meier survival curves showed that OS and PFS were poorer in the high-risk group than in the low-risk group (Fig. 6A, B, E and F). In the TCGA cohort, the AUCs for 1-, 3-, and 5-year OS were 0.61, 0.62, and 0.60, respectively (Fig. 6C). In the PKUCH cohort, the AUCs for 1-, 3-, and 5-year OS were 0.54, 0.63, and 0.61, respectively (Fig. 6G). These relatively lower AUC values may be caused by lower transcriptome differences between the PTG-high and PTG-low groups, higher intra-group variation, and RNA-seq batch effects. To investigate the batch effect, we performed PCA by combining the training and the two validation cohorts. As shown in Additional file 1: Fig. S5A, there was a batch effect between the training and validation cohorts, although the PTG score was predictive in either cohort. PCA analysis based on PTG-related genes suggested that the PTG expression pattern difference between the high and low PTG score groups were compromised in the TCGA and PKUCH validation cohorts (Fig. 6D, H). Furthermore, we performed PCA of the whole transcriptome and calculated the Euclidean distances between every two samples in the training and validation sets. The results showed that, when considering the whole transcriptome in the PCA analysis, the expression pattern differences between the high and low PTG score groups in the validation cohorts became less obvious than in the training cohort, but were still detectable (Additional file 1: Fig. S6A–C). More importantly, the intragroup Euclidean distance distribution curves of the training and validation cohorts demonstrated increased intragroup variations within the high PTG score groups, with intragroup variations being especially large in the two validation cohorts. This result suggested higher levels of transcriptomic heterogeneity in tumors with a high PTG score, especially those from the validation cohorts (Additional file 1: Fig. S6D–F).

Fig. 6
figure 6

Validation of the PTG score in The Cancer Genome Atlas and Peking University Cancer Hospital cohorts. A, B Kaplan–Meier curves for the OS and PFS of GC patients in the low and high PTG score groups from The Cancer Genome Atlas (TCGA) cohort. C AUC of time-dependent ROC curves at 1, 3 and 5 years for the TCGA cohort. D PCA analysis showed different distribution patterns in the low and high PTG score groups from the TCGA cohort. The ellipse represents the 95% confidence interval. E, F Kaplan–Meier curves for the OS and PFS of GC patients in the low group and high PTG score groups from the Peking University Cancer Hospital (PKUCH) cohort. G AUC of time-dependent ROC curves at 1, 3 and 5 years from the PKUCH cohort. H PCA showed different distribution patterns in the low and high PTG score groups from the PKUCH cohort. The ellipse represents the 95% confidence interval

Although the prognostic predictive value of the PTG score was effectively validated by ROC analysis in patients with GC from both the training and validation cohorts, tumor transcriptomic heterogeneity and RNA-seq batch effects still posed limitations to this study. To further test the robustness of the PTG score at predicting GC prognosis, prospective studies with larger and more representative cohorts are needed.

The PTG score correlates with the immune cell infiltration landscape

The ECM of tumors plays a key role in immune cell infiltration [17] and contributes to tumor ECM remodeling [10, 13]. Next, we performed immune cell infiltration estimation analysis. First, we calculated the stromal score, ESTIMATE score, and tumor purity using ESTIMATE [31]. The results showed a positive correlation between the PTG score and stromal score and the ESTIMATE score, and a negative correlation between the PTG score and tumor purity (Fig. 7A–C). We analyzed 22 types of immune cell infiltration landscapes using the CIBERSORTx algorithm [30]. The patients in the three cohorts were divided into two groups based on the median PTG score. The 22 immune cell proportion plots indicated different immune microenvironmental patterns (Fig. 7D–F). TIMER has previously been used to estimate the degree of immune cell infiltration [32]. Of all immune cells analyzed, macrophages and CAFs showed the most significant differential infiltration. Specifically, M0 macrophage infiltration was similar in the low PTG score and high PTG score groups, classically activated macrophage (M1) infiltration was lower in the high PTG score group, alternatively activated macrophage (M2) infiltration was higher in the high PTG score group, and the proportion of CAFs was higher in the high PTG score group (Fig. 7G–I). Since M1 macrophages are regarded as having an antitumor phenotype, and M2 macrophages have been reported to contribute to immune suppression and tumor cell metastasis [33,34,35,36], we inferred that, as the PTG score increases, macrophages are more inclined to be M2-polarized, leading to a more friendly microenvironment for tumor cells to escape immune surveillance. Additionally, CAFs showed a higher proportion in patients in the high PTG score group. CAFs are known for their key roles in the TME [37]. They act as tumor-promoting cells by remodeling the ECM; producing growth factors; and regulating tumor cell progression, metabolism, and angiogenesis [38,39,40]. Based on the above results, we propose that a high PTG score indicates a protumor immune microenvironment, which leads to a poor prognosis in patients with GC.

Fig. 7
figure 7

The PTG score correlates with the immune cell infiltration pattern. AC Association between the PTG signature and stromal score, ESTIMATE score, and tumor purity. DF Stacked histogram of the ratio of 22 immune cell types in GC patients in the low- or high PTG score groups from the ACRG, TCGA, and PKUCH cohorts. GI The different distributions of immune cells in the low and high PTG score groups from the ACRG, TCAG, and PKUCH cohorts

Correlation between the PTG score and GC clinical characteristics

Next, we investigated the clinical characteristics, including 5-year survival, age, T stage, N stage, M stage, and American Joint Committee on Cancer (AJCC) pTNM stage, in the low and high PTG score groups (Fig. 8A). There were significant differences in the distributions of clinical characteristics between the two groups. Consistent with the above findings, the 5-year survival rate was significantly lower in the high PTG score group (Fig. 8B). Moreover, GC patients aged greater than 63 years exhibited higher PTG scores (Fig. 8C). Additionally, a high PTG score was correlated with a more advanced TNM stage (Fig. 8D–G). These results indicate that the PTG score is associated with a higher risk of poor clinical outcomes. Next, to explore whether the PTG score could be used as an independent predictor of GC prognosis, we performed univariate and multivariate Cox regression analyses. The results showed that, together with age, T stage, N stage, and M stage, the PTG score was an independent prognostic factor for OS (Fig. 8H; Additional file 2: Table S4).

Fig. 8
figure 8

Correlations between PTG scores and clinical characteristics in GC. A Heatmap of the clinicopathological characteristics and PTG scores. BG The PTG scores in different groups classified based on clinical characteristics from the ACRG, TCGA, and PKUCH sets. H Multivariate Cox regression analyses of OS in the ACRG cohort. *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001

Construction and validation of an integrative nomogram

As the independent prognostic analyses described above revealed that age, T stage, N stage, M stage, and PTG score were independent prognostic factors for GC, we established an integrative nomogram to predict the 1-, 3-, and 5-year OS of patients with GC (Fig. 9A). The calibration charts illustrated that the OS probability predicted by the nomogram approximated the actual probability well, and the 3-year OS prediction agreed best with the actual OS (Fig. 9B–D). We analyzed the relationship between survival status and nomogram scores in both the training and validation sets. Kaplan–Meier survival analysis showed that GC patients with higher nomogram scores had poorer OS in the ACRG (P < 0.001), TCGA (P < 0.0001), and PKUCH cohorts (P < 0.0001; Fig. 9E–G). In the training set, the AUCs for OS at 1, 3, and 5 years, according to the nomogram model, indicated high prognostic validity. Specifically, the AUC values of the ACRG cohort for 1-, 3-, and 5-year OS were 0.843, 0.837, and 0.824, respectively (Fig. 9H). In the TCGA validation set, the AUC values for 1-, 3-, and 5-year OS were 0.667, 0.694, and 0.649, respectively (Fig. 9I). In the PKUCH validation set, the AUC values for 1-, 3-, and 5-year OS were 0.753, 0.745, and 0.739, respectively (Fig. 9J). These results illustrated the effectiveness of the nomogram. Furthermore, to compare the predictive potential of the nomogram and other clinical characteristics, we constructed ROC curves of the nomogram, age, T stage, N stage, M stage, and pTNM stage at 3 years OS. In the training set, the AUC value of the nomogram was 0.837, which was higher than the AUC values for age (0.551), T stage (0.633), N stage (0.717), M stage (0.578), and pTNM stage (0.744; Additional file 1: Fig. S7A–B). In the validation set, the conclusion was similar to that of the training set. Specifically, in the TCGA validation set, the AUC values of the nomogram, age, T stage, N stage, M stage, and pTNM stage were 0.694, 0.578, 0.559, 0.624, 0.527, and 0.608, respectively (Additional file 1: Fig. S7C–D). In the PKUCH validation set, the AUC values of the nomogram, age, T stage, N stage, M stage, and pTNM stage were 0.745, 0.543, 0.551,0.736, 0.540, and 0.695, respectively (Additional file 1: Fig. S7E–F). These results showed that our nomogram combining PTG score, age, T stage, N stage, and M stage was more effective than age or TNM stage alone at predicting GC prognosis.

Fig. 9
figure 9

Nomogram combining AJCC stage and PTG score to predict the OS of GC patients. A Nomogram integrating the PTG score, age, T stage, N stage, and M stage for predicting the 1-, 3-, and 5-year OS of GC patients in the training set. BD Calibration curves of the nomogram for predicting the 1-, 3- and 5-year OS in the ACRG, TCGA, and PKUCH cohorts. EG Kaplan–Meier survival plots of OS in the ACRG, TCGA, and PKUCH cohorts. HJ ROC curves for predicting the 1-, 3-, and 5-year OS in the ACRG, TCGA, and PKUCH cohorts

Discussion

PDTX models have been widely used in clinical research [41]. To create a PDTX model, cancer tissue was cut into small pieces of approximately 3 mm3, and then implanted subcutaneously in mice according to published methods [22]. Thus, PDTX models not only provide an in vivo microenvironment for investigating the tumorigenic potential of tumor cells, but also help maintain the interaction between tumor cells and other cells within the TME. Therefore, in addition to their application in drug screening, PDTX models can act as biological-function-testing models to measure the tumorigenic ability of tumor tissues. We noticed a significant variation in the tumorigenicity of tumor tissues from 122 patients with GC, even for those in the same age group and pathological stage. Further analysis revealed a correlation between tumorigenicity and GC prognosis. Notably, GC at the T1 stage show only a 25% tumorigenesis rate. This suggests that patients with T1 stage GC might not be suitable for PDTX model construction. However, owing to the limited number of T1 samples (four samples in this study), the low tumorigenesis rate was not statistically significant. Whether GC tissues at the T1 stage possess unique molecular characteristics that hinder tumorigenesis requires further investigation.

To investigate the mechanism underlying the prognosis predicting potential of PDTX tumorigenicity, we performed RNA-seq analysis of PDTX-originating tumor tissues and their corresponding normal tissues from 14 patients in the tumorigenesis group and five in the non-tumorigenesis group. By profiling the transcriptomes of these samples, we discovered that multiple ECM-related signaling pathway gene sets were enriched in the tumorigenesis group. This suggests that the ECM may play key roles in tumorigenesis in PDTX models. In addition, ECM-related prognostic signatures have been identified across multiple solid tumor types, including GC [42,43,44,45,46], non-small cell lung cancer, pancreatic ductal adenocarcinoma, ovarian cancer, hepatocellular carcinoma, and triple-negative breast cancer [47, 48]. However, these analyses focused only on prognosis and lacked animal models to confirm the in vivo role of the ECM signatures in tumor tissues.

Based on the ECM-related genes discovered in the tumorigenesis group, we constructed a robust GC prediction model using LASSO Cox regression analysis, namely, the PTG score model. This model had AUC values of 0.72, 0.76, and 0.75 for 1-, 3-, and 5-year OS rates, respectively. The predictive ability of the PTG score was validated using the TCGA STAD dataset and our PKUCH cohort. Furthermore, GC patients with different PTG scores showed heterogeneous pathological characteristics and immune landscapes. In the high PTG score group, macrophages were more inclined to be M2 polarized, which has been reported to be a tumor-promoting phenotype. Additionally, the proportion of patients with CAF infiltration was higher in the high PTG score group.

The ECM consists of a variety of macromolecules that maintain the architecture, integrity, development, and homeostasis of normal tissues [49, 51]. The dynamic reciprocity between the ECM and the cells within it has long been observed [52]. The loss of correct ECM organization and homeostasis is a hallmark of solid tumors, in which both tumor and non-malignant stromal cells contribute to, and in turn are affected by, the deposition and remodeling of the ECM [16]. Biochemical and biomechanical changes in the ECM result in dysregulation of intracellular signaling in tumor cells and the promotion of proliferation, survival, and resistance to chemotherapy [53]. Conversely, to induce surrounding non-malignant cells to support tumor cell migration and invasion, tumor cells often subvert nearby stromal cells to remodel the ECM, constituting a crosstalk pattern that differs from the pattern in non-malignant conditions [17]. In addition, CAFs can remodel the ECM by degrading normal ECM components; secreting multiple matrix proteins; producing MMPs, including MMP-1 and MMP-3; increasing ECM stiffness; and facilitating tumor progression [54,55,56,57,58]. The remodeled ECM then promotes CAF activation [59]. Moreover, there is an interaction between tumor-associated macrophages (TAMs) and CAFs [60]. For instance, CAFs promote monocyte migration and polarization into the M2 phenotype in breast and prostate carcinomas [61,62,63,64]. Reciprocally, TAMs in the M2 polarization state facilitate CAF activation and progression [65].

Therefore, we suggest that, in our study, the ECM-related signaling pathways were dysregulated in patients whose originating tumor tissue successfully formed subcutaneous tumors in the PDTX model. This dysregulation results in changes to tumor intracellular signaling and immune cell infiltration, especially by CAFs and TAMs. In turn, the tumor cells and infiltrated immune cells promote remodeling of the ECM, forming a positive feedback loop and facilitating primary tumor development and progression in patients and tumorigenesis in mice. Thus, by post-surgery pathological diagnosis, we could not identify patients with a highly remodeled ECM. However, although the GC tissue was surgically removed, the ECM of the highly remodeled tumor may have already released numerous tumor cells for metastasis before surgery. Previous studies have shown that a large fraction of disseminated tumor cells are solitary, mitotically quiescent cells that are often referred to as dormant cells [66]. Dormant tumor cells can re-activate to re-enter the cell cycle after months or years, and rapidly develop lesions [67]. This may be the underlying mechanism of the poor prognosis in the PDTX tumorigenesis group of GC patients. Importantly, the remodeled ECM is a potential target for inhibiting GC metastasis and recurrence. To target the remodeled ECM, the use of multi-vesicular vesicles is a promising alternative to existing therapeutics [68]. Specifically, extracellular vesicles can be applied in drug delivery and release [69], with the potential to target cancer cells and organelles, including the mitochondria. The high plasticity and load capacity of engineered extracellular vesicles make it possible to achieve various combinations of cancer treatment methods [70]. Nanomaterials have been widely studied as vectors to improve drug delivery in cancer therapy [71]. For instance, in lung cancer, by encapsulating doxorubicin in superparamagnetic iron oxide nanoparticles, the drug can be released in a controlled manner, which may become a powerful chemopreventive and chemotherapeutic system for patients [72]. However, further studies are required to confirm these hypotheses in patients with GC.

Finally, we integrated the PTG score, age, and pathological stage information to construct an effective nomogram for predicting GC prognosis. The AUC values for 1-, 3-, and 5-year OS in the training sets were 0.843, 0.837, and 0.824, respectively. The prognosis-predicting ability of the nomogram was also validated in the TCGA STAD and PKUCH cohorts. In the TCGA validation set, the AUC values for 1-, 3-, and 5-year OS were 0.667, 0.694, and 0.649, respectively. In the PKUCH validation set, the AUC values for 1-, 3-, and 5-year OS were 0.753, 0.745, and 0.739, respectively. We further compared the prognostic predictive ability of the nomogram with the predictive ability of age and pathological characteristics. The results showed that our integrative nomogram combining the PTG score, age, T stage, N stage, and M stage was more effective than age or TNM stage alone at predicting GC prognosis.

Although our study provides new insights into prognostic prediction and targeted therapy for GC, it has several limitations. First, owing to the limited number of GC samples used for RNA-seq, the prognostic PTG score model was established and validated using retrospective datasets. Large-scale prospective clinical cohorts are required to test the robustness of the model. Second, we applied bulk RNA-seq to GC tissues and their corresponding normal tissues. Further studies should include single-cell RNA-seq to validate the expression signatures of stromal and immune cells in the ECM. Finally, in vitro and in vivo assays are needed to validate the correlation between the PTG score and immune cell infiltration, which will provide new insights into GC prognosis prediction and individualized therapy.

Conclusions

In the present study, we used PDTX models in 122 GC cases and found that PDTX tumorigenicity was an independent prognostic factor. By investigating the transcriptome of PDTX-originating tumor cells, we found that ECM-related genes were strongly associated with tumorigenicity. Thus, we established a feasible model, named the PTG score, to predict GC prognosis. The predictive ability of the model was more robust when combined with patient age and pathological TNM stage. In addition, we revealed the relationship between PTG score and immune cell infiltration in tumors. Specifically, a high PTG score was significantly associated with CAF infiltration and macrophage M2 polarization, which may be promising targets for individual GC treatment. Overall, together with patient age and pathological TNM stage, the PTG score may be used as an effective tool for predicting GC prognosis.

Availability of data and materials

The RNA-seq raw data (fastq files) for the 19 PDTX-originating GC tissues and their corresponding normal tissues generated and analyzed during the current study are available from the NGDC repository (https://ngdc.cncb.ac.cn/, accession ID: HRA004403). The gene expression dataset of the ACRG cohort analyzed in the current study is available at the GEO database repository, https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE62254 [23]. The gene expression dataset of TCGA cohort analyzed in the current study is available at the UCSC repository, https://xenabrowser.net/datapages/. The gene expression dataset of the PKUCH cohort analyzed in the current study is available from the corresponding author upon reasonable request [24].

Abbreviations

GC:

Gastric cancer

PDTX:

Patient-derived tumor xenograft

ECM:

Extracellular matrix

CAF:

Cancer-associated fibroblasts

TME:

Tumor microenvironment

IHC:

Immunohistochemistry

FISH:

Fluorescence in situ hybridization

NCCN:

National Comprehensive Cancer Network

CSCO:

Chinese Society of Clinical Oncology

HER2:

Human epidermal growth factor receptor 2

PD-L1:

Programmed death protein-1

MSI:

Microsatellite instability

MMR:

Mismatch repair

NTRK:

Neurotrophic tropomysin-related kinase

EBV:

Epstein-Barr virus

PKUCH:

Peking University Cancer Hospital

ACRG:

Asian Cancer Research Group

TCGA:

The Cancer Genome Atlas

STAD:

Stomach adenocarcinoma

UCSC:

University of California Santa Cruz

OS:

Overall survival

PFS:

Progression free survival

GEO:

Gene expression omnibus

RNA-seq:

RNA-sequencing

DEG:

Differentially expressed gene

GO:

Gene ontology

LASSO:

Least absolute shrinkage and selection operator

PTG:

PDTX tumorigenicity-related genes

ESTIMATE:

Estimation of stromal and immune cells in malignant tumors using expression data

Nod/Scid:

Non-obese diabetes/severe combined immune-deficiency

POT:

PDTX originating tumor

CC:

Cellular component

BP:

Biological process

MF:

Molecular function

ROC:

Receiver-operating characteristic

AUC:

Area under the ROC curve

PCA:

Principal components analysis

AJCC:

American Joint Committee on Cancer

HE:

Hematoxylin–eosin

NGDC:

National Genomics Data Center

lncRNA:

Long non-coding RNA

TAM:

Tumor-associated macrophage

References

  1. Sung H, Ferlay J, Siegel RL, Laversanne M, Soerjomataram I, Jemal A, et al. Global Cancer Statistics 2020: GLOBOCAN estimates of incidence and mortality worldwide for 36 cancers in 185 countries. CA Cancer J Clin. 2021;71(3):209.

    Article  PubMed  Google Scholar 

  2. Health Commission Of The People’s Republic Of China N. National guidelines for diagnosis and treatment of gastric cancer 2022 in China (English version). Chin J Cancer Res. 2022;34(3):207–37.

  3. Fan X, Qin X, Zhang Y, Li Z, Zhou T, Zhang J, et al. Screening for gastric cancer in China: advances, challenges and visions. Chin J Cancer Res. 2021;33(2):168–80.

    CAS  PubMed  PubMed Central  Google Scholar 

  4. Qiu H, Cao S, Xu R. Cancer incidence, mortality, and burden in China: a time-trend analysis and comparison with the United States and United Kingdom based on the global epidemiological data released in 2020. Cancer Commun (London, England). 2021;41(10):1037–48.

    Google Scholar 

  5. Smyth EC, Nilsson M, Grabsch HI, van Grieken NC, Lordick F. Gastric cancer. Lancet (London, England). 2020;396(10251):635–48.

    CAS  PubMed  Google Scholar 

  6. Yang H, Ji K, Ji J. Current status and perspectives of conversion therapy for advanced gastric cancer. Chin J Cancer Res. 2022;34(2):109–14.

    CAS  PubMed  PubMed Central  Google Scholar 

  7. Cassidy JW, Caldas C, Bruna A. Maintaining tumor heterogeneity in patient-derived tumor xenografts. Cancer Res. 2015;75(15):2963–8.

    CAS  PubMed  PubMed Central  Google Scholar 

  8. Aparicio S, Hidalgo M, Kung AL. Examining the utility of patient-derived xenograft mouse models. Nat Rev Cancer. 2015;15(5):311–6.

    CAS  PubMed  Google Scholar 

  9. Bruna A, Rueda OM, Greenwood W, Batra AS, Callari M, Batra RN, et al. A biobank of breast cancer explants with preserved intra-tumor heterogeneity to screen anticancer compounds. Cell. 2016;167(1):260-274.e22.

    CAS  PubMed  PubMed Central  Google Scholar 

  10. de Visser KE, Joyce JA. The evolving tumor microenvironment: from cancer initiation to metastatic outgrowth. Cancer Cell. 2023;41(3):374–403.

    PubMed  Google Scholar 

  11. Xiao Y, Yu D. Tumor microenvironment as a therapeutic target in cancer. Pharmacol Ther. 2021;221: 107753.

    CAS  PubMed  Google Scholar 

  12. Li C, Teixeira AF, Zhu HJ, Ten Dijke P. Cancer associated-fibroblast-derived exosomes in cancer progression. Mol Cancer. 2021;20(1):154.

    PubMed  PubMed Central  Google Scholar 

  13. Sutherland TE, Dyer DP, Allen JE. The extracellular matrix and the immune system: a mutually dependent relationship. Science. 2023;379(6633):eabp8964.

    CAS  PubMed  Google Scholar 

  14. Huang J, Zhang L, Wan D, Zhou L, Zheng S, Lin S, et al. Extracellular matrix and its therapeutic potential for cancer treatment. Signal Transduct Target Ther. 2021;23(6):153.

    Google Scholar 

  15. Belli C, Trapani D, Viale G, D’Amico P, Duso BA, Della Vigna P, et al. Targeting the microenvironment in solid tumors. Cancer Treat Rev. 2018;65:22–32.

    CAS  PubMed  Google Scholar 

  16. Tian C, Clauser KR, Öhlund D, Rickelt S, Huang Y, Gupta M, et al. Proteomic analyses of ECM during pancreatic ductal adenocarcinoma progression reveal different contributions by tumor and stromal cells. Proc Natl Acad Sci USA. 2019;116(39):19609–18.

    CAS  PubMed  PubMed Central  Google Scholar 

  17. Cox TR. The matrix in cancer. Nat Rev Cancer. 2021;21(4):217–38.

    CAS  PubMed  Google Scholar 

  18. Ajani JA, D’Amico TA, Bentrem DJ, Chao J, Cooke D, Corvera C, et al. Gastric Cancer, Version 2.2022, NCCN Clinical Practice Guidelines in Oncology. J Natl Compr Canc Netw. 2022;20(2):167–92.

    CAS  PubMed  Google Scholar 

  19. Wang F, Zhang X, Li Y, Tang L, Qu X, Ying J, et al. The Chinese Society of Clinical Oncology (CSCO): clinical guidelines for the diagnosis and treatment of gastric cancer, 2021. Cancer Commun (Lond). 2021;41(8):747–95.

    PubMed  Google Scholar 

  20. Bai Y, Xie T, Wang Z, Tong S, Zhao X, Zhao F, et al. Efficacy and predictive biomarkers of immunotherapy in Epstein-Barr virus-associated gastric cancer. J Immunother Cancer. 2022;10(3): e004080.

    PubMed  PubMed Central  Google Scholar 

  21. Sun K, Xu R, Ma F, Yang N, Li Y, Sun X, et al. scRNA-seq of gastric tumor shows complex intercellular interaction with an alternative T cell exhaustion trajectory. Nat Commun. 2022;13(1):4943.

    CAS  PubMed  PubMed Central  Google Scholar 

  22. Tentler JJ, Tan AC, Weekes CD, Jimeno A, Leong S, Pitts TM, et al. Patient-derived tumour xenografts as models for oncology drug development. Nat Rev Clin Oncol. 2012;9(6):338–50.

    CAS  PubMed  PubMed Central  Google Scholar 

  23. Cristescu R, Lee J, Nebozhyn M, Kim KM, Ting JC, Wong SS, et al. Molecular analysis of gastric cancer identifies subtypes associated with distinct clinical outcomes. Nat Med. 2015;21(5):449–56.

    CAS  PubMed  Google Scholar 

  24. Huang HC, Wen XZ, Xue H, Chen RS, Ji JF, Xu L. Phosphoglucose isomerase gene expression as a prognostic biomarker of gastric cancer. Chin J Cancer Res. 2019;31(5):771–84.

    CAS  PubMed  PubMed Central  Google Scholar 

  25. Dobin A, Davis CA, Schlesinger F, Drenkow J, Zaleski C, Jha S, et al. STAR: ultrafast universal RNA-seq aligner. Bioinformatics. 2013;29(1):15–21.

    CAS  PubMed  Google Scholar 

  26. Li B, Dewey CN. RSEM: accurate transcript quantification from RNA-Seq data with or without a reference genome. BMC Bioinformatics. 2011;4(12):323.

    Google Scholar 

  27. Bu D, Luo H, Huo P, Wang Z, Zhang S, He Z, et al. KOBAS-i: intelligent prioritization and exploratory visualization of biological functions for gene enrichment analysis. Nucleic Acids Res. 2021;49(W1):W317–25.

    CAS  PubMed  PubMed Central  Google Scholar 

  28. Szklarczyk D, Kirsch R, Koutrouli M, Nastou K, Mehryary F, Hachilif R, et al. The STRING database in 2023: protein–protein association networks and functional enrichment analyses for any sequenced genome of interest. Nucleic Acids Res. 2023;51(D1):D638–46.

    CAS  PubMed  Google Scholar 

  29. Therneau TM, Grambsch PM. Modeling survival data: extending the cox model. New York: Springer; 2000.

    Google Scholar 

  30. Newman AM, Steen CB, Liu CL, Gentles AJ, Chaudhuri AA, Scherer F, et al. Determining cell type abundance and expression from bulk tissues with digital cytometry. Nat Biotechnol. 2019;37(7):773–82.

    CAS  PubMed  PubMed Central  Google Scholar 

  31. Yoshihara K, Shahmoradgoli M, Martínez E, Vegesna R, Kim H, Torres-Garcia W, et al. Inferring tumour purity and stromal and immune cell admixture from expression data. Nat Commun. 2013;4:2612.

    PubMed  Google Scholar 

  32. Li T, Fu J, Zeng Z, Cohen D, Li J, Chen Q, et al. TIMER2.0 for analysis of tumor-infiltrating immune cells. Nucleic Acids Res. 2020;48(W1):W509–14.

    CAS  PubMed  PubMed Central  Google Scholar 

  33. Boutilier AJ, Elsawa SF. Macrophage polarization states in the tumor microenvironment. Int J Mol Sci. 2021;22(13):6995.

    CAS  PubMed  PubMed Central  Google Scholar 

  34. Mantovani A, Locati M. Tumor-associated macrophages as a paradigm of macrophage plasticity, diversity, and polarization: lessons and open questions. Arterioscler Thromb Vasc Biol. 2013;33(7):1478–83.

    CAS  PubMed  Google Scholar 

  35. Mantovani A, Sozzani S, Locati M, Allavena P, Sica A. Macrophage polarization: tumor-associated macrophages as a paradigm for polarized M2 mononuclear phagocytes. Trends Immunol. 2002;23(11):549–55.

    CAS  PubMed  Google Scholar 

  36. Zhang X, Ji L, Li MO. Control of tumor-associated macrophage responses by nutrient acquisition and metabolism. Immunity. 2023;56(1):14–31.

    CAS  PubMed  Google Scholar 

  37. Costa A, Kieffer Y, Scholer-Dahirel A, Pelon F, Bourachot B, Cardon M, et al. Fibroblast heterogeneity and immunosuppressive environment in human breast cancer. Cancer Cell. 2018;33(3):463-479.e10.

    CAS  PubMed  Google Scholar 

  38. Mhaidly R, Mechta-Grigoriou F. Role of cancer-associated fibroblast subpopulations in immune infiltration, as a new means of treatment in cancer. Immunol Rev. 2021;302(1):259–72.

    CAS  PubMed  PubMed Central  Google Scholar 

  39. Rimal R, Desai P, Daware R, Hosseinnejad A, Prakash J, Lammers T, et al. Cancer-associated fibroblasts: origin, function, imaging, and therapeutic targeting. Adv Drug Deliv Rev. 2022;189: 114504.

    CAS  PubMed  Google Scholar 

  40. Vitale I, Manic G, Coussens LM, Kroemer G, Galluzzi L. Macrophages and metabolism in the tumor microenvironment. Cell Metab. 2019;30(1):36–50.

    CAS  PubMed  Google Scholar 

  41. Hidalgo M, Amant F, Biankin AV, Budinská E, Byrne AT, Caldas C, et al. Patient-derived xenograft models: an emerging platform for translational cancer research. Cancer Discov. 2014;4(9):998–1013.

    CAS  PubMed  PubMed Central  Google Scholar 

  42. Yuzhalin AE, Urbonas T, Silva MA, Muschel RJ, Gordon-Weeks AN. A core matrisome gene signature predicts cancer outcome. Br J Cancer. 2018;118(3):435–40.

    CAS  PubMed  PubMed Central  Google Scholar 

  43. Wang Z, Wang Z, Hu X, Han Q, Chen K, Pang G. Extracellular matrix-associated pathways promote the progression of gastric cancer by impacting the dendritic cell axis. Int J Gen Med. 2021;14:6725–39.

    CAS  PubMed  PubMed Central  Google Scholar 

  44. Yang Z, Xue F, Li M, Zhu X, Lu X, Wang C, et al. Extracellular matrix characterization in gastric cancer helps to predict prognosis and chemotherapy response. Front Oncol. 2021;27(11): 753330.

    Google Scholar 

  45. Moreira AM, Ferreira RM, Carneiro P, Figueiredo J, Osório H, Barbosa J, et al. Proteomic identification of a gastric tumor ECM signature associated with cancer progression. Front Mol Biosci. 2022;1(9): 818552.

    Google Scholar 

  46. Alatan H, Chen Y, Zhou J, Wang L. Extracellular matrix-related hubs genes have adverse effects on gastric adenocarcinoma prognosis based on bioinformatics analysis. Genes (Basel). 2021;12(7):1104.

    CAS  PubMed  Google Scholar 

  47. Pearce OMT, Delaine-Smith RM, Maniati E, Nichols S, Wang J, Böhm S, et al. Deconstruction of a metastatic tumor microenvironment reveals a common matrix response in human cancers. Cancer Discov. 2018;8(3):304–19.

    CAS  PubMed  Google Scholar 

  48. Bin Lim S, Chua MLK, Yeong JPS, Tan SJ, Lim WT, Lim CT. Pan-cancer analysis connects tumor matrisome to immune response. NPJ Precis Oncol. 2019;3:15.

    PubMed  Google Scholar 

  49. Eble JA, Niland S. The extracellular matrix in tumor progression and metastasis. Clin Exp Metastasis. 2019;36(3):171–98.

    CAS  PubMed  Google Scholar 

  50. Ozbek S, Balasubramanian PG, Chiquet-Ehrismann R, Tucker RP, Adams JC. The evolution of extracellular matrix. Mol Biol Cell. 2010;21(24):4300–5.

    PubMed  PubMed Central  Google Scholar 

  51. Malandrino A, Mak M, Kamm RD, Moeendarbary E. Complex mechanics of the heterogeneous extracellular matrix in cancer. Extreme Mech Lett. 2018;21:25–34.

    PubMed  PubMed Central  Google Scholar 

  52. Bissell MJ, Hall HG, Parry G. How does the extracellular matrix direct gene expression? J Theor Biol. 1982;99(1):31–68.

    CAS  PubMed  Google Scholar 

  53. Wei L, Ye H, Li G, Lu Y, Zhou Q, Zheng S, et al. Cancer-associated fibroblasts promote progression and gemcitabine resistance via the SDF-1/SATB-1 pathway in pancreatic cancer. Cell Death Dis. 2018;9(11):1065.

    PubMed  PubMed Central  Google Scholar 

  54. Najafi M, Farhood B, Mortezaee K. Extracellular matrix (ECM) stiffness and degradation as cancer drivers. J Cell Biochem. 2019;120(3):2782–90.

    CAS  PubMed  Google Scholar 

  55. Miles FL, Sikes RA. Insidious changes in stromal matrix fuel cancer progression. Mol Cancer Res. 2014;12(3):297–312.

    CAS  PubMed  PubMed Central  Google Scholar 

  56. Erdogan B, Webb DJ. Cancer-associated fibroblasts modulate growth factor signaling and extracellular matrix remodeling to regulate tumor metastasis. Biochem Soc Trans. 2017;45(1):229–36.

    CAS  PubMed  PubMed Central  Google Scholar 

  57. Lopez JI, Kang I, You WK, McDonald DM, Weaver VM. In situ force mapping of mammary gland transformation. Integr Biol (Camb). 2011;3(9):910–21.

    CAS  PubMed  Google Scholar 

  58. Murphy G, Nagase H. Progress in matrix metalloproteinase research. Mol Aspects Med. 2008;29(5):290–308.

    CAS  PubMed  PubMed Central  Google Scholar 

  59. Calvo F, Ege N, Grande-Garcia A, Hooper S, Jenkins RP, Chaudhry SI, et al. Mechanotransduction and YAP-dependent matrix remodelling is required for the generation and maintenance of cancer-associated fibroblasts. Nat Cell Biol. 2013;15(6):637–46.

    CAS  PubMed  Google Scholar 

  60. Mao X, Xu J, Wang W, Liang C, Hua J, Liu J, et al. Crosstalk between cancer-associated fibroblasts and immune cells in the tumor microenvironment: new findings and future perspectives. Mol Cancer. 2021;20(1):131.

    CAS  PubMed  PubMed Central  Google Scholar 

  61. Ksiazkiewicz M, Gottfried E, Kreutz M, Mack M, Hofstaedter F, Kunz-Schughart LA. Importance of CCL2-CCR2A/2B signaling for monocyte migration into spheroids of breast cancer-derived fibroblasts. Immunobiology. 2010;215(9–10):737–47.

    CAS  PubMed  Google Scholar 

  62. Cohen N, Shani O, Raz Y, Sharon Y, Hoffman D, Abramovitz L, et al. Fibroblasts drive an immunosuppressive and growth-promoting microenvironment in breast cancer via secretion of Chitinase 3-like 1. Oncogene. 2017;36(31):4457–68.

    CAS  PubMed  PubMed Central  Google Scholar 

  63. Hirz T, Mei S, Sarkar H, Kfoury Y, Wu S, Verhoeven BM, et al. Dissecting the immune suppressive human prostate tumor microenvironment via integrated single-cell and spatial transcriptomic analyses. Nat Commun. 2023;14(1):663.

    CAS  PubMed  PubMed Central  Google Scholar 

  64. Taddei ML, Cavallini L, Comito G, Giannoni E, Folini M, Marini A, et al. Senescent stroma promotes prostate cancer progression: the role of miR-210. Mol Oncol. 2014;8(8):1729–46.

    CAS  PubMed  PubMed Central  Google Scholar 

  65. Gunaydin G. CAFs interacting with TAMs in tumor microenvironment to enhance tumorigenesis and immune evasion. Front Oncol. 2021;11: 668349.

    CAS  PubMed  PubMed Central  Google Scholar 

  66. Pantel K, Alix-Panabières C, Riethdorf S. Cancer micrometastases. Nat Rev Clin Oncol. 2009;6(6):339–51.

    CAS  PubMed  Google Scholar 

  67. Yang L, Liu Q, Zhang X, Liu X, Zhou B, Chen J, et al. DNA of neutrophil extracellular traps promotes cancer metastasis via CCDC25. Nature. 2020;583(7814):133–8.

    CAS  PubMed  Google Scholar 

  68. Khalilov R. A comprehensive review of advanced nano-biomaterials in regenerative medicine and drug delivery. Adv Biol. 2023.

  69. Cabeza L, Perazzoli G, Peña M, Cepero A, Luque C, Melguizo C, et al. Cancer therapy based on extracellular vesicles as drug delivery vehicles. J Control Release. 2020;10(327):296–315.

    Google Scholar 

  70. Ma Y, Dong S, Li X, Kim BYS, Yang Z, Jiang W. Extracellular vesicles: an emerging nanoplatform for cancer therapy. Front Oncol. 2020;10: 606906.

    PubMed  Google Scholar 

  71. Eftekhari A, Kryschi C, Pamies D, Gulec S, Ahmadian E, Janas D, et al. Natural and synthetic nanovectors for cancer therapy. Nanotheranostics. 2023;7(3):236–57.

    PubMed  PubMed Central  Google Scholar 

  72. Hasanzadeh A, Khalilov R, Abasi E, Saghfi S, Nasibova A, Akbarzadeh A. development of doxorubicin—adsorbed magnetic nanoparticles modified with biocompatible copolymers for targeted drug delivery in lung cancer. advances in biology. 2017.

Download references

Acknowledgements

Not applicable.

Funding

This work was supported by the National Natural Science Foundation of China (No. 81972758, 82073312, 82272889); the Natural Science Foundation of Beijing (No. 7222023); the joint fund for key projects of National Natural Science Foundation of China (U20A20371); Open Project of Key Laboratory of Ministry of Education, Peking University Cancer Hospital (2022-3); Peking University Medicine Sailing Program for Young Scholars’ Scientific & Technological Innovation (BMU2023YFJHMX005); Science Foundation of Peking University Cancer Hospital (XKFZ2308).

Author information

Authors and Affiliations

Authors

Contributions

ZD, TG, XX, and JJ designed the work. ZD and TG performed the statistical work and bioinformatics analyses. TG, YH and HD conducted experiments. ZD and TG drafted the work. JJ, XX, SJ, TG and YZ supervised the study and revised the manuscript. JB, GW participated in clinical data collection. All authors read and approved the final manuscript.

Corresponding authors

Correspondence to Yuan Zhou, Shuqin Jia, Xiaofang Xing or Jiafu Ji.

Ethics declarations

Ethics approval and consent to participate

This study was conducted in accordance with the Declaration of Helsinki and approved by the Institutional Review Board of Peking University Cancer Hospital (2019KT11).

Consent for publication

Not applicable.

Competing interests

The authors declare that they have no competing interests.

Additional information

Publisher's Note

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Supplementary Information

Additional file 1: Figure S1

. Representative HE (hematoxylin-eosin) staining of GC tissue from non-tumorigenesis group. (A-B) Representative HE staining of GC tissue from non-tumorigenesis group. Figure S2. Survival analysis of patients for PDTX models. (A-P) Kaplan-Meier curves for OS and DFS of 122 GC patients grouped by clinical characteristics. Figure S3. (A) The protein-protein interaction (PPI) evidence among coding DEGs in non-tumorigenesis group. (B) GO enrichment bar plots of 36 DEGs for biological process (green), cellular component (red), and molecular function (blue). Figure S4. (A) The Pearson correlation analysis among 116 candidate prognostic genes. Figure S5. PCA analysis of training cohort and validation cohorts. (A) PCA analysis showed batch effect among ACRG training set and both validation sets. Ellipse represents 95% confidence interval. Figure S6. PCA analysis and Euclidean distances distribution of PTG score-low groups and PTG-score high groups in training cohort and validation cohorts. (A) PCA analysis showed different gene expression distribution patterns in the PTG score-low group and PTG score-high group from ACRG training cohort. Ellipse represents 95% confidence interval. (B-C) PCA analysis showed different gene expression distribution patterns in the PTG score-low groups and PTG score-high groups from TCGA cohort and PKUCH cohort. Ellipses represents 95% confidence interval. (D-F) Distribution plot of Euclidean distances between every two samples in the PTG score-low group and PTG score-high group from ACRG, TCGA and PKUCH cohorts. Figure S7. Comparison of AUCs of nomogram and age/TNM stages. (A) AUCs of the nomogram, age, T stage, N stage and M stage to predict OS at 1 year using ACRG cohort. (B) AUCs of the nomogram, age and TNM stage to predict OS at 1 year using ACRG cohort. (C) AUCs of the nomogram, age, T stage, N stage and M stage to predict OS at 1 year using TCGA cohort. (D) AUCs of the nomogram, age and TNM stage to predict OS at 1 year using TCGA cohort. (E) AUCs of the nomogram, age, T stage, N stage and M stage to predict OS at 1 year using PKUCH cohort. (F) AUCs of the nomogram, age and TNM stage to predict OS at 1 year using PKUCH cohort.

Additional file 2: Table S1

. Cox regression analysis of overall survival and PDTX tumorigenicity together with clinical characteristics. Table S2. Clinical characteristics of 14 tumorigenesis GC patients for RNA-seq. Table S3. Clinical characteristics of 5 non-tumorigenesis tumorigenesis GC patients for RNA-seq. Table S4. Cox regression analysis of overall survival and PTG score together with clinical characteristics.

Rights and permissions

Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/. The Creative Commons Public Domain Dedication waiver (http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated in a credit line to the data.

Reprints and permissions

About this article

Check for updates. Verify currency and authenticity via CrossMark

Cite this article

Deng, Z., Guo, T., Bi, J. et al. Transcriptome profiling of patient-derived tumor xenografts suggests novel extracellular matrix-related signatures for gastric cancer prognosis prediction. J Transl Med 21, 638 (2023). https://doi.org/10.1186/s12967-023-04473-0

Download citation

  • Received:

  • Accepted:

  • Published:

  • DOI: https://doi.org/10.1186/s12967-023-04473-0

Keywords