A Novel Liquid–Liquid Phase Separation Characteristic Model Associated with Prognosis and Immune Landscape of Gastric Cancer Patients

A Novel Liquid–Liquid Phase Separation Characteristic Model Associated with Prognosis and Immune Landscape of Gastric Cancer Patients | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Research Article A Novel Liquid–Liquid Phase Separation Characteristic Model Associated with Prognosis and Immune Landscape of Gastric Cancer Patients Renjie Miao, Yun Liu, Ruiyun Chen, Zili Sun, Wei Zhang, Rui Li, and 4 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-4546744/v1 This work is licensed under a CC BY 4.0 License Status: Posted Version 1 posted You are reading this latest preprint version Abstract Background Liquid-liquid phase separation (LLPS) refers to a phenomenon in which unique liquid condensates are formed due to weak interactions among biomolecules, including proteins and nucleic acids. In cellular environments, abnormal LLPS can induce aggregation of membrane-less organelles, disrupt intracellular signaling, alter chromatin structures, and cause aberrant gene expression. The significance of LLPS in gastric cancer (GC) cells is still poorly understood. This study aims to integrate multiple omics analysis and multiple machine learning algorithms to identify LLPS characteristic genes (LCGs) which can be used to develop a LLPS characteristic prognostic model. Methods Transcriptomic and single-cell data for GC patients were retrieved from the GEO and TCGA databases. The LLPS gene set was extracted from the PhaSepDB database. Initial cellular localization analysis of LLPS gene set-expressing cells was performed using single-cell data from GSE167297. Subsequently, we analyzed 797 GC samples from the TCGA-STAD and GSE84437 merged cohort using the ConsensusClusterPlus method, then we subdivided the merged cohort into two clusters based on the expression of the LLPS gene set for further prognostic and immune analyses. Characteristic genes of the LLPS gene set were identified by the best combination of four machine learning algorithms correlating with patient survival status and time, which were then validated across three independent GC patient cohorts. The differential expression of LCGs in the prognostic model was validated using the HPA and UALCAN databases, as well as western blotting. Additionally, a nomogram was developed to improve the effectiveness of the model in clinical application. Furthermore, differences in the tumor immune microenvironment (TME), immunotherapy response, and drug sensitivity between different risk groups were studied through a variety of immune algorithms. Mutational analysis of ten LLPS gene set genes was conducted based on mutation data from the TCGA-STAD cohort. Results A LLPS characteristic prognostic model based on a combination of four machine learning algorithms was established, identifying ten LCGs with high predictive value for the prognosis, TME, immunotherapy responses, and chemotherapy drug sensitivity of GC patients. Additionally, a specific nomogram was developed, incorporating clinical features to enhance the effectiveness of the LLPS clinical score, with AUC values of 0.722, 0.715, 0.707 at 1, 3, and 5 years, respectively. The LLPS prognostic model demonstrated good predictive value for survival status across different age groups, T stages, and N stages of GC patients. Risk scores calculated from LCGs showed linear correlations with stromal scores, immune scores, TME scores, Tumor Immune Dysfunction and Exclusion (TIDE) scores, epithelial-mesenchymal transition (EMT) scores, angiogenesis scores, and tumor purity scores. Furthermore, mutations in LCGs were found to impact the survival of GC patients. Conclusions The LLPS characteristic prognostic model provides a new perspective for assessing the prognosis of GC patients, their responses to immunotherapy, TME and chemotherapy drug usage. Gastric cancer liquid-liquid phase separation LLPS characteristic genes LLPS characteristic prognostic model prognosis tumor immune microenvironment chemotherapy drug sensitivity immunotherapy machine learning Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Figure 6 Figure 7 Figure 8 Figure 9 Figure 10 Figure 11 Figure 12 Figure 13 Figure 14 Introduction Gastric cancer (GC) presents a formidable challenge and is considered the fifth most common cancer worldwide [ 1 ]. While incidence rates vary geographically, countries in Eastern Asia such as China continue to record high numbers [ 2 ]. The incidence of GC is closely associated with numerous factors including genetic mutations, aberrant gene expression, infection with Helicobacter pylori and Epstein-Barr virus infection [ 3 – 5 ]. Additionally, novel immunotherapeutic strategies targeting the GC immune microenvironment are currently gaining interest [ 6 – 8 ]. Furthermore, by utilizing certain cellular markers such as fibroblasts, macrophages, and exhausted T cells, it is possible to accurately forecast the outlook of patients with GC and their reaction to immunotherapy, thus aiding in guiding precision medicine more effectively [ 9 – 11 ]. Liquid-liquid phase separation (LLPS) is an emerging field in tumor biology, recognized for its crucial role in organizing intracellular biomolecules into membrane-less organelles [ 12 ]. Often likened to oil droplets separating in water, LLPS facilitates the stable concentration of specific proteins and nucleic acids, influencing various biological functions within the cell. LLPS maintains a relatively stable equilibrium within cells [ 13 ]. Disruption of this equilibrium results in dysregulated LLPS, affecting key cellular processes such as chromatin cycling, DNA damage repair, and cell signaling pathways [ 14 ]. Crucially, abnormal LLPS can stabilize oncogenic proteins, facilitate the evasion of tumor suppressor factors, and promote tumor immune escape, thereby enhancing the oncogenic potential of tumor cells [ 15 ]. Current research on tumor-related abnormal LLPS suggests that targeting LLPS-induced protein or nucleic acid aggregates could represent a novel approach to cancer therapy. For instance, targeting Sentrin/SUMO-specific protease, which undergoes abnormal LLPS, can restore the recruitment of RNF168 to DNA damage sites, enhancing DNA repair functions to maintain genomic integrity [ 16 ]. Research in prostate cancer has revealed that the LLPS-associated gene ET516 disrupts the androgen receptor complex, inhibiting the growth of cells with androgen receptor-resistant mutations. Targeting ET516-induced protein aggregates post-abnormal LLPS shows promise in treating prostate cancer [ 17 ]. With the rise of machine learning and artificial intelligence (AI), combined with transcriptomics, single-cell genomics, and spatial transcriptomics, effective LLPS characteristic prognostic models have been developed for certain cancers like breast cancer and glioblastoma [ 18 , 19 ]. These models facilitate better stratification of patients based on risk levels, guiding prognosis more effectively. Nevertheless, the involvement of LLPS in GC is not well comprehended, necessitating further research to explore its implications and potential as a therapeutic target. In this study, we obtained a LLPS gene set from PhaSepDB and initiated our investigation by analyzing single-cell data from the GSE167297 cohort. Preliminary analysis focused on the cellular localization and differential gene pathway enrichment of cells across different LLPS scoring groups. Subsequent clustering analysis in a merged cohort of TCGA-STAD and GSE84437 revealed that samples of GC patients could be clearly separated into two clusters using LLPS gene set, with these clusters showing significant differences in pathway enrichment, prognostic features, tumor immune microenvironment (TME), and immunoreactivity. Following this, we employed a combination of Lasso, XGBoost, Random Forest (RF) and multivariate Cox regression analyses to successfully develop a novel LLPS-based prognostic model, incorporating ten LLPS characteristic genes (LCGs): HOMER3, CAPRIN1, PAK2, HSPA1A, MED19, DACT1, RNF2, ARID1A, YBX1 and TBP. This model demonstrated robust performance in guiding prognosis for GC patients. To verify the model's effectiveness, we conducted external validations using three independent cohorts, which confirmed the model's strong predictive capabilities. Particularly noteworthy is our exploration of the new model's application in analyzing the TME, immunoreactivity, tumor mutational burden (TMB), and chemotherapy drug sensitivity across different risk strata of GC patients, affirming its high clinical value. Finally, we utilized the HPA and UALCAN databases as well as western blotting to confirm the high expression state of ten LCGs in GC tissues compared to normal tissues. This comprehensive approach not only confirms the utility of the LLPS gene set in stratifying GC but also highlights the potential of our prognostic model in personalizing patient management and guiding therapeutic decisions. Through the rigorous validation and analysis, our findings support the integration of LLPS-associated biomarkers in predictive models, offering a novel pathway for improving the accuracy of GC therapy approaches. Materials and Methods Data acquisition In this research, transcriptome data, mutation data, and clinical information for 415 GC samples along with 34 normal samples were downloaded from TCGA database, accessed on March 8, 2024. Additionally, the transcriptomic data, single-cell data, and clinical information for cohorts GSE84437, GSE84426, GSE84433, and GSE167297 were obtained from the GEO database, viewed on March 12, 2024. Furthermore, a total of 818 genes associated with LLPS (Table 1) were downloaded from PhaSepDB ( http://db.phasep.pro/browse/ ) [ 20 ]. Utilizing R software, transcriptome data from the TCGA-STAD and GSE84437 cohorts were extracted and normalized using limma and sva packages, followed by batch correction. After removing samples with a survival time of less than zero days, expression data for 797 GC samples were prepared for subsequent combined analyses. Analysis of GSE167297 scRNA-seq data We obtained single-cell data from the GSE167297 dataset with two samples: GSM5101019, and GSM5101020. The quality control procedures were implemented as follows: 1) Cells that express less than 200 genes were deemed ineligible for analysis; 2) Cells with mitochondrial gene content exceeding 10% were removed, as elevated mitochondrial gene expression can indicate cell stress or apoptosis; 3) For the analysis, we employed the FindIntegrationAnchors function from the Seurat package with a setting of 2000 anchors, followed by data integration using the IntegrateData function. To address batch effects between samples, we applied the SCT method. Dimensionality reduction was carried out using PCA, choosing 15 principal components. Cells were clustered through the t-SNE method. The identification of cell types was carried out through cell surface marker genes, which were obtained from a CellMarker database ( https://http://bio-bigdata.hrbmu.edu.cn/CellMarker/ , accessed on 12 March 2024). Using the LLPS gene set, we scored cells via the AddModuleScore function in Seurat. ConsensusClusterPlus analysis The ConsensusClusterPlus analysis is a commonly used classification analysis method in cancer research, and stable clustering of samples can be achieved using the ConsensusClusterPlus package [ 21 ]. We applied this method to cluster GC patients based on LLPS gene set. Furthermore, agglomerative PAM clustering using Euclidean distance was employed, and the analysis was repeated on 80% of the samples 100 times. The ideal number of clusters, found to be two, was determined through the empirical cumulative distribution function. Subsequently, PCA was conducted with the prcomp function in the stats package to validate the reliability of the two identified cluster. Kaplan-Meier survival analysis was utilized on various LLPS clusters through the survival package and survminer package. Gene differential analysis was carried out with the limma package to identify genes that differ significantly between clusters. These DEGs were visualized using ggplot2, pheatmap, and ggVolcano packages, with a display limit of 50 genes. Finally, we obtained genes related to immune checkpoints from the MSigDB ( https://www.gsea-msigdb.org/gsea/msigdb ) and investigated the differential expression of these immune checkpoint-related genes across the two clusters [ 22 ]. In order to delve deeper into the biological processes of the clusters, we evaluated the biological differences between the two clusters using three pathway enrichment methods. First, we conducted Gene Set Enrichment Analysis (GSEA), which assesses the distribution pattern of genes from a given gene set in a ranked gene list according to their correlation with a phenotype, thereby determining their contribution to the phenotype [ 23 ]. We downloaded the KEGG gene set "c2.cp.kegg_legacy.v2023.2.Hs.symbols" from the MSigDB database, ranked the DEGs among clusters, and then performed GSEA. Second, we performed ssGSEA to study the enrichment of gene sets in individual samples using the GSVA package, analyzing the differences in gene set enrichment between different clusters. Finally, we conducted KEGG enrichment analysis on the DEGs between clusters utilizing the clusterProfiler package, compared the enrichment results with those from GSE167297, and visualized the results using a Venn diagram. Analysis of TME and response to immunotherapy in two LLPS-related clusters We downloaded gene sets related to immune cells and immune functions [ 24 ]. The ssGSEA method was employed to determine the immune gene set scores for each sample, enabling the evaluation of immune infiltration variations across different clusters. TIDE score is a valuable tool for evaluating the likelihood of immune therapy responses and the potential for tumor immune escape [ 25 ]. We utilized the TIDE website ( http://tide.dfci.harvard.edu/ ) to upload expression matrix data from a combined cohort that included TCGA-STAD and GSE84437 datasets. From this, we derived TIDE scores and predictions for immune therapy responses for each sample. An elevated TIDE score represents an increased chance of tumor immune escape, allowing us to differentiate between groups based on this criterion. Further analysis was conducted to assess the tumor purity using the ESTIMATE and XCELL packages. Within these tools, the stromal score represents the level of stromal cell infiltration in the tumor tissue, and the immune score reflects the extent of immune cell infiltration. Summing these scores provides both the ESTIMATE score and the TME assessment. The derived scores give an inference of tumor purity, closely linked with the malignancy level of the tumor [ 26 – 28 ]. Subsequent analysis involved computing scores for angiogenesis and EMT using gene sets from the MSigDB database and previous research [ 29 , 30 ]. These computations were made using the ssGSEA method within the GSVA package. The angiogenesis and EMT scores offer deeper insights into the proliferative and migratory capabilities of the tumor. Development of a LLPS characteristic prognostic model To identify characteristic genes associated with the LLPS characteristic prognostic model, we utilized four machine learning algorithms combined with survival duration, survival state, and gene expression matrix from 797 samples, thereby constructing a prognostic model. Initially, we employed the glmnet package for LASSO analysis and utilized 10-fold cross-validation to identify the most suitable genes. Subsequently, the XGBoost package was applied for XGBoost analysis. XGBoost, a robust classification algorithm, amalgamates various tree models to form an enhanced tree model. Through this algorithm, the top 50 optimal genes were identified. Following this, the randomForestSRC package was used for RF analysis to ascertain the optimal genes. RF, an ensemble learning method composed of multiple decision trees, is adept at handling datasets with nonlinear relationships. The intersection of genes identified by the aforementioned three algorithms was then exposed to multivariate COX stepwise regression procedure to pinpoint the ten most significant characteristic genes related to the LLPS prognostic model. Spearman correlation analysis is a conventional statistical method to determine the correlation between two non-normally distributed continuous variables. Upon identifying these characteristic genes, spearman correlation analysis was conducted to ascertain the interrelationships among these ten genes. According to their median risk scores, the samples were separated into high and low risk groups, which were calculated from LCGs expressions. The calculation equation for the risk score was shown as belows: Risk score = Expression mRNA1 × Coef mRNA1 + Expression mRNA2 × Coef mRNA2 +… Expression mRNAn × Coef mRNAn . Validation of a LLPS characteristic prognostic model We initiated with a dimensional reduction analysis using the prcomp function from the PCA package. This step was crucial for validating the reliability of our risk stratification. Subsequently, we compared risk scores between the consensus clustering groups C1 and C2 to examine their consistency. For visual analysis, we employed the pheatmap package to create heat maps displaying the risk scores, patient survival distribution, and expression of risk genes across different risk groups. This visual representation facilitated a clearer understanding of the survival outcomes and gene expression variations between groups. Further, we utilized the ggplot2 package to construct Sankey diagrams, elucidating the distribution differences of patients from varying T stages, survival statuses, and consensus clustering groups within the risk categories. Additionally, clinical information heatmaps were generated using the ComplexHeatmap package, comparing variables such as consensus cluster groups, T stages, N stages, gender, and survival status across different risk groups. Lastly, Kaplan-Meier survival analysis was conducted with the survival packages for diverse patient subsets. This included patients aged ≤ 60 years, > 60 years, those in stages T1-2, stages T3-4, and in N stages ranging from N0-1 and N2-3. These analyses were critical for further validation of the risk group stratification's reliability. Development and validation of nomograms Nomograms are developed based on multivariable regression analysis, where each influencing factor in the model is assigned a score based on its contribution to the outcome variable. These scores are then summed in order to forecast the probability of a patient's survival at 1, 3, and 5 years. We utilized Cox regression through the rms package to construct separate nomograms for risk genes and clinical information, based on independent prognostic outcomes including risk genes, age, gender, and clinical staging. The performance and predictive ability of the models were evaluated with the receiver operating characteristic (ROC) curve, which allowed for the calculation of the area under the curve (AUC). For this purpose, we employed the pROC and timeROC packages to plot the ROC curves for 1, 3, and 5 years, as well as for clinical information, to study the predictive capacity of the constructed models. Additionally, calibration curves were utilized to further evaluate the accuracy of the models. These tools allow for a comprehensive evaluation of the models, ensuring that they provide reliable and accurate predictions of patient outcomes based on a combination of genetic and clinical factors. Study of TME and immune response in the risk group To elucidate the interrelation between ten LCGs and immune cells as well as immune functions, we employed ssGSEA scoring followed by the use of the linkET and ggplot2 packages to generate a spearman correlation heatmap. In the heatmap, each cell displays the correlation between specific immune cell types or immune function types and specific genes. The intensity of color in each cell reflects the strength of the correlation, with darker shades representing stronger correlations. Positive correlations are represented by the color blue and negative correlations are represented by the color blue. Additionally, the color and thickness of the lines within the heatmap signify the strength and direction of the correlations. Subsequently, we employed the ssGSEA method from the GSVA package to compute the scores of the immune gene sets for each sample, in order to assess the infiltration levels and functions of immune cells across different clusters. Finally, we studied the TME and immune therapy responses of patients in different risk groups using algorithms and scores such as TIDE, ESTIMATE, XCELL, EMT scoring and angiogenesis scoring. Pearson linear correlation analysis Pearson linear correlation analysis is a traditional statistical technique used to quantify the magnitude and orientation of the linear correlation between two variables.Then the pearson linear correlation analysis was adopted to detect the linear relationships between risk scores and various other metrics including stromal scores, immune scores, ESTIMATE scores, TME scores, EMT scores, angiogenesis scores, tumor purity scores, and TIDE scores. Sensitivity analysis of chemotherapy drugs in the risk group Anticancer drug training set data were downloaded from the Genomics of Drug Sensitivity in Cancer (GDSC) website, and the chemotherapy drug sensitivity of 797 patients in the merged cohort was predicted using the oncopdict package [ 31 ]. Statistical tests were conducted between high and low risk groups to identify drugs with significant differences in sensitivity, characterized by a median IC50 < 1. Validation of the training set of the LLPS characteristic prognostic model Based on the LLPS characteristic prognostic model, risk scores were calculated for ten LCGs within the GSE84433, GSE84437, and TCGA-STAD cohorts. Subsequently, the stability of the model was validated through PCA. The risk scores, patient survival distributions, and expression of risk genes across different risk groups were visualized using the pheatmap package to observe variations in patient survival and risk gene expression among the groups. Finally, the performance of this novel LLPS prognostic model was validated by generating Kaplan-Meier survival analysis curves and the ROC curves for 1, 3, and 5 years using the survival, survminer, timeROC, and pROC packages. TMB analysis in the TCGA-STAD cohort The TCGA-STAD cohort provides detailed mutation data, so we employed the maftools package to analyze somatic mutation data obtained from TCGA-STAD, assessing the mutational differences in characteristic genes across different LLPS scoring groups. Copy number variation (CNV), which is closely associated with the activation of oncogenes, results from genomic rearrangements and generally refers to genes longer than 1 kb. We downloaded CNV data for TCGA-STAD and the "UCSC.HG19.Human.CytoBandIdeogram" from the UCSC database. The chromosomal locations and copy number variations of characteristic genes were analyzed using the RCircos and rtracklayer packages. Finally, Kaplan-Meier survival analysis for different TMB groups was analyzed using the survival package. Validation of LCGs expression of LLPS characteristic prognostic model We validated the expression of the characteristic genes through the HPA and UALCAN databases [ 32 ]. Cell Culture Gastric mucosal epithelial cells (GES-1) and three types of GC cells (HGC-27, MKN-45 and SGC-7901), were employed in this study. These cell lines originated from the School of Medicine, Jiangsu University. As for the cell culture process, it involved the use of DMEM augmented with ten percent FBS. The process took place in a humidified incubator maintained 37 degrees Celsius with five percent CO2 atmosphere. Western Blotting Analysis The RIPA lysis buffer was adopted to destruct cells (GES-1, HGC-2, MKN-45, SGC-7901). The buffer was fortified with PMSF along with phosphatase inhibitors. A 10% SDS-PAGE gel was utilized to separate proteins. And the protein was then transferred to the PVDF membrane. The membrane was then blocked and incubated with the following antibodies (Table 2 ) overnight, the GAPDH protein serves as an internal reference. The membrane was cleaned and incubated with rabbit secondary antibody at room temperature. Following another wash, ECL exposure system was employed to shoot membranes containing proteins, all conducted as per the instructions provided by the manufacturer. Table 2 The antibodies used in this study Names of antibodies Dilution rate Source of antibodies HOMER3 1:500 Wanleibio TBP 1:500 Wanleibio YBX1 1:500 Wanleibio MED19 1:500 Wanleibio HSPA1A 1:400 Wanleibio PAK2 1:500 Wanleibio RNF2 1:500 Wanleibio DACT1 1:500 Wanleibio RNF2 1:400 Wanleibio CAPRIN1 1:400 Wanleibio GAPDH 1:2000 Abcam Statistical analysis and visualization For this research, data analysis and visualization were carried out using R software, version 4.3.2. Correlation assessments were conducted using both pearson and spearman correlation analyses. Comparisons between two groups were carried out using the wilcoxon and limma tests. All data results were thought statistically significant at a threshold of P < 0.05. Results Research flow chart An overall flow chart of this study was plotted to make the study easy to understand (Fig. 1 ). Single-cell analysis of GC for localization of cells with different LLPS scores In the GSE167297 cohort, two samples, GSM5101019 and GSM5101020, were downloaded, representing superficial cancer and deep cancer from the same patient, respectively. The data underwent filtering, batch correction, and cell quality control, setting the threshold for mitochondrial gene content at less than 10%, hemoglobin gene content at less than 5%, the number of genes observed per cell ranges from 200 to 5000, while the total number of mRNA molecules detected within a cell falls between 200 and 30,000. This preprocessing identified a total of 8,062 cells (Fig. 2 A). PCA demonstrates the stable distribution of cells for two types of samples, with the optimal number of PC being 15 (Fig. 2 B,C), resulting in the classification of cells into 19 clusters (Fig. 2 D). Cells were then annotated into seven categories based on cellular markers, types of cells include GC cells, T cells, B cells, endothelial cells, fibroblasts, myeloid cells and mast cells (Fig. 2 G). Visualization of marker gene expression was achieved through heat maps (Fig. 2 E,F). Following cell annotation, cells were scored using the LLPS gene set. After scoring, cells were categorized into LLPS-up and LLPS-dwon scoring subgroups. It was noted that cells with LLPS-up scoring subgroups were mostly found in endothelial cells, T cells, B cells, GC cells, suggesting that LLPS may primarily occur in these cell types within GC, potentially affecting the EMT, TME and immune infiltration (Fig. 2 H,I). ConsensusClusterPlus analysis based on the transcriptomic profiles of LLPS gene set reveals heterogeneous subgroups within GC patients We performed consensus ConsensusClusterPlus analysis on 797 samples from a combined cohort of TCGA-STAD and GSE84437, using a LLPS gene set obtained from the PhaSepDB database. The optimal K value was determined to be two based on the cumulative distribution function (CDF) curve, effectively separating 797 samples into two clusters: C1 and C2 (Fig. 3 A-C). Subsequent PCA analysis further confirmed the stability of these two clusters (Fig. 3 D,E). Survival curves exhibited that GC patients of cluster C2 had significantly lower survival rates compared to those in cluster C1 (Fig. 3 F). DEGs between the two clusters was examined using the limma test, identifying 1,222 genes with significant differences, the chart displayed the 50 most noteworthy LCGs (Fig. 3 G,H). Additionally, variations in the expression of immune checkpoint genes were noted, indicating potential distinctions in the TME and reactions to immunotherapy across the clusters (Fig. 3 I). Differential enrichment of KEGG and GO pathways in two LLPS-related clusters DEGs underwent pathway enrichment analysis using KEGG and GO databases between clusters C1 and C2. The KEGG pathway enrichment result revealed significant differences across pathways related to cell cycle, transcriptional misregulation in cancer, IL-17, p53, mismatch repair, DNA replication and etc (Fig. 4 A-C). Changes in these pathways are closely related to human diseases, cellular processes, genetic information processing and etc. The GO pathway enrichment result showed significant differences across pathways related to cell cycle, cell cycle phase transition, cell division,chromosome and etc (Fig. 4 E-G). Finally, the pathway variations between the two clusters were also confirmed using both ssGSEA and GSVA methods, with consistent results across all three approaches regarding the differences in GO-KEGG pathways, the pathway changes in C2 cluster mainly focus on mitosis, cell cycle, DNA replication and so on (Fig. 4 D,H,I,J). Differences in TME, immune infiltration and immunotherapy response between two LLPS-related clusters After calculating the immune infiltration status of two clusters using the ssGSEA method, we observed significant differences between Cluster C1 and Cluster C2 in several parameters. These include the levels of immune cell infiltration, chemokine receptor levels, human leukocyte antigen, T-cell co-stimulation levels, and reactions to Type I and Type II interferons (Fig. 5 A-C). Additionally, We utilized the ESTIMATE and XCELL algorithms for tumor stromal score and immune score analyses. The sum of these scores provided the ESTIMATE score and the TME score, from which we inferred the tumor purity score. We found that, compared to Cluster C1, Cluster C2 had a significantly lower tumor purity score, suggesting a lower tumor purity and a higher degree of malignancy in Cluster C2 (Fig. 5 D-F). Further analysis of the EMT and angiogenesis scores revealed that Cluster C2 also significantly exceeded Cluster C1 in these aspects. This suggests enhanced migratory capabilities and angiogenic potential in the tumor cells of Cluster C2, consistent with the results of single-cell analysis where high-scoring LLPS cells were localized to endothelial cells (Fig. 5 I,J). Finally, through TIDE analysis of the two clusters, we noted that C2 contained a TIDE score that was markedly higher than the TIDE score of C1. Moreover, the immunotherapeutic responsiveness of C2 was significantly lower than that of C1, indicating a stronger capacity for immune evasion and lower sensitivity to immunotherapy in C2 (Fig. 5 G,H). The establishment and validation of the LLPS characteristic prognostic model Following the identification of the clustering function by the LLPS gene set, we conducted dimensionality reduction on the LLPS gene set using survival time and status data from 797 GC patients to identify prognostic model characteristic genes. Initially, Lasso regression was utilized to select LCGs, choosing a lambda value of 0.04 to identify 45 characteristic genes (Fig. 6 A-C). Subsequently, Subsequently, the RF algorithm was adopted to study the prognostic relevance of each gene, resulting in the selection of 71 LCGs (Fig. 6 D-F). Ultimately, the XGBoost algorithm was used to select the top 50 optimal genes, and the intersection of results from the three algorithms yielded 17 LCGs (Fig. 6 G-I). Conducting a multivariate Cox regression method of these 17 genes established an optimal combination of 10 LCGs (HOMER3, CAPRIN1, PAK2, HSPA1A, MED19, DACT1, RNF2, ARID1A, YBX1, TBP). Spearman analysis revealed significant correlations among these ten genes, with four genes (CAPRIN1, PAK2, RNF2, TBP) potentially acting as protective factors, highly expressed in tumor cells and associated with longer survival periods (Fig. 6 J,L). The LLPS characteristic prognostic model built with these ten genes categorized samples into high and low risk subgroups by the median risk score. The risk score was computed using the following formula: Risk score = (0.220851788535136) × HOMER3 + (-0.22193287559865) × CAPRIN1 + (-0.104355501176279) × PAK2 + (0.0900195727621234) × HSPA1A + (0.553288202545932) × MED19 + (0.115430780509586) × DACT1 + (-0.371052343828004) × RNF2 + (0.113397725961348) × ARID1A + (0.352859164860805) × YBX1 + (-0.291531356154402) × TBP After adjusting for age, gender and clinical stage in the multivariate Cox analysis, the LLPS risk score, N stage, and age were determined to be independent prognostic factors (Fig. 6 K). To further confirm whether LCGs is highly expressed in GC tissues compared to gastric normal tissues, we obtained immunohistochemical images of the ten LCGs from the HPA database to further confirm the expression of these genes in GC tissues, the findings indicated that all LCGs exhibited high levels of expression (Fig. 7 A). Similarly, expression data obtained from the UALCAN database for the ten LCGs showed results consistent with the immunohistochemical staining, ten genes show a significantly higher level of expression (Fig. 7 B). The western blotting analysis results indicated the level of LCG expression in gastric cancer cells (MKN-45, HGC-27, SGC-7901) was higher than that of GSE-1 in vitro cell experiments (Fig. 7 C,D). After constructing the LLPS prognostic model, PCA confirmed the model's stability, with risk genes consistently distributed between two groups (Fig. 8 A,B). The Sankey diagram and violin plot showed that samples in Cluster C2 had notably higher risk scores compared to those in Cluster C1, with the majority of samples in Cluster C2 falling into the high-risk category (Fig. 8 C,F). Risk score analysis combined with patient survival distribution and risk gene expression revealed that the majority of the deceased cases were found within the high-risk group. In the low-risk group, the expression of risk genes (CAPRIN1, PAK2, RNF2, TBP) were at higher levels, while the high-risk group showed the expression of risk genes (HOMER3, HSPA1A, MED19, DACT1, TBP, ARID1A) were at higher levels (Fig. 8 D,E). Finally, a chi-square test analyzed the allocation of samples across different clinical stages and risk subgroups, finding correlations between cluster distribution, different T stages, patient survival status, and risk groups (Fig. 8 G). Kaplan-Meier analysis further clarified the clinical worth of the prognostic model, revealing that individuals who were classified as high-risk experienced noticeably shorter survival times. Further analysis of subgroups categorized by age, N stage and T stage revealed that individuals in the high-risk category had notably shorter survival durations across all subgroups (age ≤ 60, age > 60, N0-1 stage, N2-3 stage, T1-2 stage, T2-3 stage). After constructing the LLPS prognostic model using the merged TCGA-STAD and GSE84437 cohorts, we further validated the stability and accuracy of the LLPS prognostic model using three independent cohorts (GSE84433, GSE84426, TCGA-STAD) as training sets. In these independent cohorts survival state and expression of LCGs were consistent with those in the merged TCGA-STAD and GSE84437 cohorts (Fig. 9 A-C). PCA confirmed the stability of the three independent models (Fig. 9 D-F). Subsequent Kaplan-Meier analysis exhibited that the high-risk subgroup had notably shorter survival periods compared to the low-risk subgroup across all three cohorts. ROC curves evaluated the predictive performance of three models in independent cohorts, all of which demonstrated good predictive capabilities (Fig. 9 G-I). Finally, ROC curve analysis of a nomogram that utilizes risk scores and clinical characteristics revealed an improved AUC value, indicating enhanced predictive performance of the model (Fig. 9 J-L). The construction of nomograms Using the outcomes from the multivariable Cox regression analysis, we constructed nomograms for the ten characteristic genes and clinical pathological staging to visually demonstrate the predicted probabilities of survival rates of patients at 1, 3 and 5 years. The nomogram for the LCGs indicated excellent accuracy in predicting patient survival, the mortality rates at 1, 3 and 5 years are 0.382, 0.182 and 0.116 respectively (Fig. 10 A). In the nomogram combining clinical pathological staging with LLPS risk scores, N stage, age, and risk score were found to have higher clinical net benefits compared to other factors, with mortality rates at 1, 3 and 5 years of 0.431, 0.214 and 0.137 respectively (Fig. 10 D). Calibration plots for the nomograms showed good consistency (Fig. 10 B,E). The models were further evaluated using ROC curves, which demonstrated excellent predictive performance based on AUC values, with the nomogram incorporating risk scores and clinical features showing superior predictive accuracy (Fig. 10 C,F). Immunological analysis derived from the LLPS characteristic prognostic model Spearman correlation analysis revealed significant associations between the ten LCGs and immune infiltration (Fig. 11 A,B). Using the ssGSEA method, Significant variations were observed in the levels of immune cell infiltration, chemokine receptor, responses to type I and type II interferons, antigen-presenting cell co-stimulation, and parainflammation between different risk groups (Fig. 11 C-E). Additionally, the ESTIMATE and XCELL algorithms were utilized for examining the tumor stromal score and immune score. The sum of these scores provided the ESTIMATE score and TME score, from which tumor purity was inferred. Our observation revealed that the high-risk group had a notably reduced tumor purity score comparing with the low-risk group, indicating lower tumor purity and a higher level of malignancy. (Fig. 11 F-H). Further analysis of EMT and angiogenesis scores revealed that the high risk group obviously exceeded the low risk group, indicating enhanced migratory capabilities and angiogenic potential (Fig. 11 L,M). TIDE analysis displayed that the high risk group contained a much higher TIDE score and a significantly lower responsiveness to immunotherapy compared to the low risk group, pointing out an increased probability of immune system evasion and reduced sensitivity to immunotherapy. Additionally, our analysis disclosed that groups with lower immune responsiveness had notably higher risk scores, suggesting a negative association between risk score and immune responsiveness (Fig. 11 I-K). To further clarify the model's accuracy, we went on an exploration on correlations between the risk score and various indices using pearson linear correlation analysis. We observed a positive correlation between the risk score and stromal, immune, ESTIMATE, TME, TIDE, angiogenesis, and EMT scores, and negatively correlated with tumor purity, pointing out that a higher risk score is linked to increased tumor stromal scores, stronger immune evasion capabilities, poorer immune responsiveness, stronger tumor proliferation and migration abilities, and lower tumor purity, indicating a higher degree of malignancy (Fig. 12 ). Chemotherapy drug sensitivity analysis Analysis of drug sensitivity employing the GDSC database as a training set revealed significant differences in sensitivity to 132 chemotherapy drugs between the high and low-risk groups (IC50 < 1), with the low-risk group benefiting more from drugs such as Carmustine, Sinularin, Wnt-C59, BMS-345541, and Gefitinib, while the high-risk group benefited more from BMS-754807, AZD8055, AZD2014, AZD8186, and Dasatinib (Fig. 13 ). LCGs mutation analysis The TCGA-STAD cohort provided detailed sample mutation information, allowing us to provide additional clarification on the clinical significance of LLPS prognostic model characteristic genes by analyzing gene mutations. It was discovered that the mutation rate of ten LCGs was elevated in the high risk group (Fig. 14 A,B). Analysis of gene mutation burden showed that PAK2, RNF2, HSPA1A, and CAPRIN1 showed a greater prevalence of gain-of-function mutations in GC, while HOMER3, MED19, DACT1, TBP, ARID1A, and YBX1 had a higher frequency of loss-of-function mutations (Fig. 14 C,D). Kaplan-Meier curves revealed that TMB similarly affected the survival periods of GC patients (Fig. 14 E). Discussion GC, originating from the epithelial cells of the gastric mucosa, consistently ranks among the leading causes of cancer incidence and mortality worldwide. Although external factors such as diet and lifestyle are significant contributors to GC, many cases and deaths are attributed to changes in the local tissue microenvironment [ 33 , 34 ]. Early-stage GC patients generally have a favorable prognosis following treatment, whereas those diagnosed at advanced stages often experience poor therapeutic outcomes and prognosis [ 35 ]. Consequently, developing new methods to guide prognosis and treatment for GC patients is crucial. Despite recent advances in understanding the mechanisms of GC progression and new treatment strategies, such as ferroptosis, cuproptosis, necroptosis, and disulfidptosis, many unknown mechanisms within the TME continue to promote tumor development [ 36 – 39 ]. LLPS was initially regarded as a purely physical phenomenon of interest primarily in the field of physical chemistry [ 40 ]. However, recent studies have demonstrated that LLPS is also a fundamental mechanism of biochemical organization within cells, where various biomolecules such as proteins and nucleic acids can form membrane-less organelles through LLPS. Dysregulation of LLPS can lead to pathological aggregation of proteins and cellular dysfunction, forming the basis for many diseases [ 41 ]. In non-tumor diseases such as retroviral infections and coronavirus infections, viral nucleocapsid proteins can form condensates through LLPS, enhancing viral replication or evading host immune surveillance [ 42 , 43 ]. In cancer, proteins encoded by certain genes undergo LLPS to form condensates, leading to chromosomal inactivation, DNA damage, and autophagy, thereby promoting tumor progression [ 44 ]. Targeting these condensates to correct abnormal LLPS states has emerged as a new approach in precision oncology. However, research on LLPS in GC remains scarce. This study focuses on LLPS, utilizing the TCGA and GEO databases along with various machine learning algorithms and multi-omics analyses to comprehensively assess the role of LLPS in the TME, immune infiltration, immunotherapy, drug sensitivity, and prognosis of GC patients, thereby developing a novel LLPS prognostic model to guide patient prognosis and treatment. Initially, single-cell analysis was employed to preliminarily analyze the LLPS gene set in GC cells, identifying that cells with high LLPS gene set scores were predominantly T cells and B cells. This suggests potential dysregulation of LLPS in these cells, which could negatively impact the immune response, enhancing the tumor's ability to evade immune surveillance and clearance [ 45 – 48 ]. Subsequent KEGG-GO analysis through three enrichment methods in different clustering groups revealed activation of several inflammatory pathways such as the IL-17 pathway, NF-kappa pathway, and TNF pathway, which play crucial roles in tumor growth regulation [ 49 – 51 ]. This further clarifies the diverse regulatory mechanisms of LLPS in the GC immune microenvironment. After establishing the potential regulatory roles of LLPS in GC through single-cell omics, the study continued with transcriptomic analysis combined with various machine learning algorithms. The integration of machine learning with medical data has become a popular research method, effectively predicting diseases and assessing risks, guiding personalized medicine, and drug development. The larger the sample size included in machine learning, the more accurate the algorithm's predictions [ 52 , 53 ]. Therefore, by merging the TCGA-STAD and GSE84437 cohorts, we significantly increased the sample size, enhancing the accuracy of machine learning predictions. In the transcriptomic study, the LLPS gene set was initially analyzed using consensus clustering, clearly dividing 797 samples into two clusters. Analysis of DEGs between the subgroups revealed significant differences in the expression of multiple immune checkpoints (CTLA4, PDL1, LAG3 and etc), which are crucial for cancer therapy. Inhibiting these immune checkpoints can restore the immune system's ability to recognize and eliminate tumors, thus guiding the use of immunosuppressants [ 54 , 55 ]. Recent studies have also shown that interactions between tumor cells, immune cells, and stromal cells in the TME can influence tumor development, making the TME a new target for cancer therapy [ 56 , 57 ]. Therefore, assessing the TME can effectively guide targeted immunotherapy for GC patients to correct the immune microenvironment. Subsequently, various immune algorithms were used to further study the immune infiltration, immunotherapy responsiveness, immune evasion capabilities, and tumor purity of the two clusters, with significant differences observed between them. In summary, this study preliminarily clarified the role of the LLPS gene set in GC. However, due to the large number of genes in the LLPS gene set, it was not feasible to construct an effective clinical prediction model for practical medical use. Therefore, four machine learning regression algorithms were combined to perform dimensionality reduction on the gene set, selecting LCGs that could guide patient prognosis. The XGBoost and RF algorithms were efficient and accurate but had fitting issues, while the Lasso algorithm provided better fitting results. The combination of these four machine learning algorithms effectively integrated their respective advantages and compensated for their shortcomings, allowing the selected characteristic genes to constitute a more precise prognostic model to guide the prognosis. Ultimately, ten LCGs were identified (HOMER3, CAPRIN1, PAK2, HSPA1A, MED19, DACT1, RNF2, ARID1A, YBX1, TBP). YBX1 and CAPRIN1 have been proven in liver cancer research to form protein condensates that cause cellular pathway changes and promote tumor development [ 58 , 59 ], while the other eight genes have not been studied in tumors and may become key entry points for LLPS research in GC. Based on these ten LCGs, a prognostic model was constructed to divide GC samples into high and low-risk groups. The assessments of immune infiltration, immunotherapy responsiveness, immune evasion capabilities, and tumor purity in the two groups were consistent with the cluster assessments, and the prognostic model also performed well in predicting patient survival. The risk score combined with clinical information in the nomogram prediction model showed better predictive performance, and the ROC curve further verified the model's accuracy. Prognostic models constructed in three independent training cohorts all demonstrated good predictive performance. In conclusion, the LLPS prognostic model constructed in this study has high clinical value, effectively stratifying GC patients and guiding their prognosis and immunotherapy. In terms of chemotherapy drug use, early-stage GC is often treated with endoscopic surgical resection, while advanced GC is treated with continuous chemotherapy regimens [ 60 , 61 ]. Therefore, the choice of chemotherapy drugs is crucial for GC patients, and the LLPS prognostic model developed in this study can effectively group GC patients, predicting chemotherapy drugs with better sensitivity for different groups, providing valuable guidance for clinical medication. CNV are also important factors in tumor development, with increased copy numbers of oncogenes or decreased copy numbers of tumor suppressor genes promoting tumor progression [ 62 , 63 ]. In this study, sample mutation information provided by the TCGA database was employed for analysis of the copy number mutations of the ten LCGs and the gene mutation frequency in different risk groups, further clarifying the clinical predictive value of these ten genes. However, whether mutations in the characteristic genes are driving factors for abnormal LLPS has not been studied. The limitations of this study are twofold. First, the GC samples included in the study encompass all pathological stages, treatment stages, and different metastatic states. However, the TME of GC patients varies at different stages. Therefore, in subsequent research, we will meticulously divide GC samples, such as by T stage, in situ cancer versus metastatic cancer, and pre- and post-immunosuppressant treatment relapse. By comparing the immune infiltration and survival situations of GC patients in these different categories, we aim to further reveal at which stage LLPS has a more significant impact on the disease. Additionally, KEGG enrichment results showed significant changes in the P53 pathway in the prognostic model groups. Therefore, subsequent studies could also divide GC patients based on P53 mutation status and non-mutation status to investigate whether there is mutual regulation between LLPS and the P53 pathway. The second limitation is the limited single-cell data included in the study. Due to existing equipment and funding constraints, personalized single-cell sequencing analyses could not be conducted to better understand the biological roles of LLPS in tumor cells and the TME. Additionally, the lack of complete clinical information in the samples provided by the GEO database prevents better classification. Subsequent studies could simulate the TME in vitro and further clarify the impact of LCGs in the GC tumor microenvironment through single-cell sequencing and cell functional experiments. It also remains to be studied whether the characteristic genes will form protein aggregates after undergoing LLPS, affecting cellular biological functions, and whether targeting these protein condensates encoded by the characteristic genes could become a new method for treating GC through in vivo and in vitro researches. Conclusion This study developed a novel prognostic model containing ten characteristic genes for LLPS, which demonstrates good predictive capacity for the overall survival of GC patients. Combining this model with patient clinical pathological features to construct a nomogram can better assist in clinical decision-making. Based on this model, effective risk stratification of patients, prediction of patient immune responses, and guidance on the use of chemotherapy drugs can be achieved, thus also having high clinical value in guiding the precision treatment of GC patients. Abbreviations GC gastric cancer LLPS liquid-liquid phase separation LCGs LLPS characteristic genes TME tumor immune microenvironment EMT epithelial mesenchymal transformation TIDE Tumor Immune Dysfunction and Exclusion AI artificial intelligence RF Random Forest TMB tumor mutational burden PCA Principal Component Analysis DEGs differential expression genes GSEA Gene Set Enrichment Analysis ssGSEA single-sample Gene Set Enrichment Analysis ROC receiver operating characteristic AUC area under the curve GDSC Genomics of Drug Sensitivity in Cancer CNV copy number variation GES-1 gastric mucosal epithelial cell Declarations Ethics approval and consent to participate Not applicable. Consent for publication Not applicable. Availability of data and materials All information of GC patients used during the research was acquired from the TCGA and GEO databases. Competing Interests Not Applicable. Funding The funding for this project was supported by Social Development Project of Zhenjiang (Grant No. SH2022101), Jiangsu Funding Program for Excellent Postdoctoral Talent (No. 2023ZB180) and Postdoctoral Fellowship Program of CPSF (No. GZC20230992). Author Contributions RM and YL were in charge of in the main information analysis and experiments. RC, ZS, WZ, RL, RS and XW were responsible for the data acquisition and classification. JW and SS took charge of technical guidance and the modification of the paper. Acknowledgements We are grateful for the support provided by the TGCA and GEO databases, as well as Jiangsu University. Conflicts of Interest No conflict of interest. References Bray F, Laversanne M, Sung HYA, Ferlay J, Siegel RL, Soerjomataram I, Jemal A. Global cancer statistics 2022: GLOBOCAN estimates of incidence and mortality worldwide for 36 cancers in 185 countries. CA-Cancer J Clin 2024:35. Xia CF, Dong XS, Li H, Cao MM, Sun DAQ, He SY, Yang F, Yan XX, Zhang SL, Li N, Chen WQ. Cancer statistics in China and United States, 2022: profiles, trends, and determinants. Chin Med J. 2022;135(5):584–90. Li S, Yu WB, Xie F, Luo HT, Liu ZM, Lv WW, Shi DB, Yu DX, Gao P, Chen C, et al. Neoadjuvant therapy with immune checkpoint blockade, antiangiogenesis, and chemotherapy for locally advanced gastric cancer. Nat Commun. 2023;14(1):16. Chia NY, Tan P. Molecular classification of gastric cancer. Ann Oncol. 2016;27(5):763–9. Yang J, Liu Z, Zeng B, Hu G, Gan R. Epstein-Barr virus-associated gastric cancer: A distinct subtype. Cancer Lett. 2020;495:191–9. Amieva M, Peek RM Jr.. Pathobiology of Helicobacter pylori-Induced Gastric Cancer. Gastroenterology. 2016;150(1):64–78. Lin Y, Jing X, Chen Z, Pan X, Xu D, Yu X, Zhong F, Zhao L, Yang C, Wang B, et al. Histone deacetylase-mediated tumor microenvironment characteristics an d synergistic immunotherapy in gastric cancer. Theranostics. 2023;13(13):4574–600. Kim R, An M, Lee H, Mehta A, Heo YJ, Kim K-M, Lee S-Y, Moon J, Kim ST, Min B-H, et al. Early Tumor-Immune Microenvironmental Remodeling and Response to First -Line Fluoropyrimidine and Platinum Chemotherapy in Advanced Gastric C ancer. Cancer Discov. 2022;12(4):984–1001. Mak TK, Li X, Huang H, Wu K, Huang Z, He Y, Zhang C. The cancer-associated fibroblast-related signature predicts prognosis and indicates immune microenvironment infiltration in gastric cancer. Front Immunol. 2022;13:951214. Li Y, Hu X, Lin R, Zhou G, Zhao L, Zhao D, Zhang Y, Li W, Zhang Y, Ma P, et al. Single-cell landscape reveals active cell subtypes and their interacti on in the tumor microenvironment of gastric cancer. Theranostics. 2022;12(8):3818–33. Chen HY, Sun Q, Zhang CA, She JJ, Cao S, Cao M, Zhang NA, Adiila AV, Zhong JJ, Yao CY, et al. Identification and Validation of CYBB, CD86, and C3AR1 as the Key Genes Related to Macrophage Infiltration of Gastric Cancer. Front Mol Biosci. 2021;8:15. Mehta S, Zhang J. Liquid-liquid phase separation drives cellular function and dysfunctio n in cancer. Nat Rev Cancer. 2022;22(4):239–52. Liu Z, Qin Z, Liu Y, Xia X, He L, Chen N, Hu X, Peng X. Liquid–liquid phase separation: roles and implications in future cance r treatment. Int J Biol Sci. 2023;19(13):4139–56. Li R-H, Tian T, Ge Q-W, He X-Y, Shi C-Y, Li J-H, Zhang Z, Liu F-Z, Sang L-J, Yang Z-Z, et al. A phosphatidic acid-binding lncRNA SNHG9 facilitates LATS1 liquid-liqu id phase separation to promote oncogenic YAP signaling. Cell Res. 2021;31(10):1088–105. Ahn JH, Davis ES, Daugird TA, Zhao S, Quiroga IY, Uryu H, Li J, Storey AJ, Tsai Y-H, Keeley DP, et al. Phase separation drives aberrant chromatin looping and cancer developm ent. Nature. 2021;595(7868):591–5. Wei M, Huang X, Liao L, Tian Y, Zheng X. SENP1 Decreases RNF168 Phase Separation to Promote DNA Damage Repair a nd Drug Resistance in Colon Cancer. Cancer Res. 2023;83(17):2908–23. Xie J, He H, Kong W, Li Z, Gao Z, Xie D, Sun L, Fan X, Jiang X, Zheng Q, et al. Targeting androgen receptor phase separation to overcome antiandrogen resistance. Nat Chem Biol. 2022;18(12):1341–50. Ning L, Zhao G, Xie C, Lan H, Chen J, Tan H, Wei C, Zhou Z. Development and Validation of a Liquid-Liquid Phase Separation-Related Gene Signature as Prognostic Biomarker for Low-Grade Gliomas. Dis Markers 2022, 2022:1487165. Xie J, Chen L, Wu D, Liu S, Pei S, Tang Q, Wang Y, Ou M, Zhu Z, Ruan S, et al. Significance of liquid-liquid phase separation (LLPS)-related genes in breast cancer: a multi-omics analysis. Aging. 2023;15(12):5592–610. Hou C, Wang X, Xie H, Chen T, Zhu P, Xu X, You K, Li T. PhaSepDB in 2022: annotating phase separation-related proteins with dr oplet states, co-phase separation partners and other experimental info rmation. Nucleic Acids Res. 2023;51(D1):D460–5. Wilkerson MD, Hayes DN. ConsensusClusterPlus: a class discovery tool with confidence assessmen ts and item tracking. Bioinformatics. 2010;26(12):1572–3. Liberzon A, Birger C, Thorvaldsdóttir H, Ghandi M, Mesirov JP, Tamayo P. The Molecular Signatures Database (MSigDB) hallmark gene set collectio n. Cell Syst. 2015;1(6):417–25. Subramanian A, Tamayo P, Mootha VK, Mukherjee S, Ebert BL, Gillette MA, Paulovich A, Pomeroy SL, Golub TR, Lander ES, Mesirov JP. Gene set enrichment analysis: a knowledge-based approach for interpret ing genome-wide expression profiles. Proc Natl Acad Sci U S A. 2005;102(43):15545–50. Hänzelmann S, Castelo R, Guinney J. GSVA: gene set variation analysis for microarray and RNA-seq data. BMC Bioinformatics. 2013;14:7. Zheng K, Hai Y, Chen H, Zhang Y, Hu X, Ni K. Tumor immune dysfunction and exclusion subtypes in bladder cancer and pan-cancer: a novel molecular subtyping strategy and immunotherapeutic prediction model. J Transl Med. 2024;22(1):365. Chen Y, Meng Z, Zhang L, Liu F. CD2 Is a Novel Immune-Related Prognostic Biomarker of Invasive Breast Carcinoma That Modulates the Tumor Microenvironment. Front Immunol. 2021;12:664845. Yoshihara K, Shahmoradgoli M, Martínez E, Vegesna R, Kim H, Torres-Garcia W, Treviño V, Shen H, Laird PW, Levine DA, et al. Inferring tumour purity and stromal and immune cell admixture from exp ression data. Nat Commun. 2013;4:2612. Aran D, Sirota M, Butte AJ. Systematic pan-cancer analysis of tumour purity. Nat Commun. 2015;6:8971. Masiero M, Simões FC, Han HD, Snell C, Peterkin T, Bridges E, Mangala LS, Wu SY-Y, Pradeep S, Li D, et al. A core human primary tumor angiogenesis signature identifies the endot helial orphan receptor ELTD1 as a key regulator of angiogenesis. Cancer Cell. 2013;24(2):229–41. Lamouille S, Xu J, Derynck R. Molecular mechanisms of epithelial-mesenchymal transition. Nat Rev Mol Cell Biol. 2014;15(3):178–96. Yang W, Soares J, Greninger P, Edelman EJ, Lightfoot H, Forbes S, Bindal N, Beare D, Smith JA, Thompson IR, et al. Genomics of Drug Sensitivity in Cancer (GDSC): a resource for therapeu tic biomarker discovery in cancer cells. Nucleic Acids Res. 2013;41(Database issue):D955–961. Chandrashekar DS, Karthikeyan SK, Korla PK, Patel H, Shovon AR, Athar M, Netto GJ, Qin ZS, Kumar S, Manne U, et al. UALCAN: An update to the integrated cancer data analysis platform. Neoplasia. 2022;25:18–27. Chen Y, Jia K, Sun Y, Zhang C, Li Y, Zhang L, Chen Z, Zhang J, Hu Y, Yuan J, et al. Predicting response to immunotherapy in gastric cancer via multi-dimen sional analyses of the tumour immune microenvironment. Nat Commun. 2022;13(1):4851. Jin G, Lv J, Yang M, Wang M, Zhu M, Wang T, Yan C, Yu C, Ding Y, Li G, et al. Genetic risk, incident gastric cancer, and healthy lifestyle: a meta-a nalysis of genome-wide association studies and prospective cohort stud y. Lancet Oncol. 2020;21(10):1378–86. Yang W-J, Zhao H-P, Yu Y, Wang J-H, Guo L, Liu J-Y, Pu J, Lv J. Updates on global epidemiology, risk and prognostic factors of gastric cancer. World J Gastroenterol. 2023;29(16):2452–68. Zhang H, Deng T, Liu R, Ning T, Yang H, Liu D, Zhang Q, Lin D, Ge S, Bai M, et al. CAF secreted miR-522 suppresses ferroptosis and promotes acquired chem o-resistance in gastric cancer. Mol Cancer. 2020;19(1):43. Sun L, Zhang Y, Yang B, Sun S, Zhang P, Luo Z, Feng T, Cui Z, Zhu T, Li Y, et al. Lactylation of METTL16 promotes cuproptosis via m 6 A-modific ation on FDX1 mRNA in gastric cancer. Nat Commun. 2023;14(1):6523. Wu N, Liu F, Huang Y, Su X, Zhang Y, Yu L, Liu B. Necroptosis Related Genes Predict Prognosis and Therapeutic Potential in Gastric Cancer. Biomolecules. 2023;13(1):101. Li J, Yu T, Sun J, Ma M, Zheng Z, He Y, Kang W, Ye X. Integrated analysis of disulfidptosis-related immune genes signature t o boost the efficacy of prognostic prediction in gastric cancer. Cancer Cell Int. 2024;24(1):112. Chen D, Lyu M, Kou X, Li J, Yang Z, Gao L, Li Y, Fan L-M, Shi H, Zhong S. Integration of light and temperature sensing by liquid-liquid phase se paration of phytochrome B. Mol Cell. 2022;82(16):3015–e30293016. Alberti S, Gladfelter A, Mittag T. Considerations and Challenges in Studying Liquid-Liquid Phase Separati on and Biomolecular Condensates. Cell. 2019;176(3):419–34. Chau B-A, Chen V, Cochrane AW, Parent LJ, Mouland AJ. Liquid-liquid phase separation of nucleocapsid proteins during SARS-Co V-2 and HIV-1 replication. Cell Rep. 2023;42(1):111968. Wei W, Bai L, Yan B, Meng W, Wang H, Zhai J, Si F, Zheng C. When liquid-liquid phase separation meets viral infections. Front Immunol. 2022;13:985622. Liu Q, Li J, Zhang W, Xiao C, Zhang S, Nian C, Li J, Su D, Chen L, Zhao Q, et al. Glycogen accumulation and phase separation drives liver tumor initiati on. Cell. 2021;184(22):5559–e55765519. Park J, Hsueh P-C, Li Z, Ho P-C. Microenvironment-driven metabolic adaptations guiding CD8 + T cell anti-tumor immunity. Immunity. 2023;56(1):32–42. Baldominos P, Barbera-Mourelle A, Barreiro O, Huang Y, Wight A, Cho J-W, Zhao X, Estivill G, Adam I, Sanchez X, et al. Quiescent cancer cells resist T cell attack by forming an immunosuppre ssive niche. Cell. 2022;185(10):1694–e17081619. Downs-Canner SM, Meier J, Vincent BG, Serody JS. B Cell Function in the Tumor Microenvironment. Annu Rev Immunol. 2022;40:169–93. Liu Y, Zhou X, Wang X. Targeting the tumor microenvironment in B-cell lymphoma: challenges an d opportunities. J Hematol Oncol. 2021;14(1):125. Li X, Bechara R, Zhao J, McGeachy MJ, Gaffen SL. IL-17 receptor-based signaling and implications for disease. Nat Immunol. 2019;20(12):1594–602. Yu H, Lin L, Zhang Z, Zhang H, Hu H. Targeting NF-κB pathway for the therapy of diseases: mechanism and cli nical study. Signal Transduct Target Ther. 2020;5(1):209. Yuan S, Carter P, Bruzelius M, Vithayathil M, Kar S, Mason AM, Lin A, Burgess S, Larsson SC. Effects of tumour necrosis factor on cardiovascular disease and cancer: A two-sample Mendelian randomization study. EBioMedicine. 2020;59:102956. Greener JG, Kandathil SM, Moffat L, Jones DT. A guide to machine learning for biologists. Nat Rev Mol Cell Biol. 2022;23(1):40–55. Haug CJ, Drazen JM. Artificial Intelligence and Machine Learning in Clinical Medicine, 202 3. N Engl J Med. 2023;388(13):1201–8. Galluzzi L, Humeau J, Buqué A, Zitvogel L, Kroemer G. Immunostimulation with chemotherapy in the era of immune checkpoint in hibitors. Nat Rev Clin Oncol. 2020;17(12):725–41. Heinhuis KM, Ros W, Kok M, Steeghs N, Beijnen JH, Schellens JHM. Enhancing antitumor response by combining immune checkpoint inhibitors with chemotherapy in solid tumors. Ann Oncol. 2019;30(2):219–35. Xiao Y, Yu D. Tumor microenvironment as a therapeutic target in cancer. Pharmacol Ther. 2021;221:107753. Jin M-Z, Jin W-L. The updated landscape of tumor microenvironment and drug repurposing. Signal Transduct Target Ther. 2020;5(1):166. Chen S, Cao X, Zhang J, Wu W, Zhang B, Zhao F. circVAMP3 Drives CAPRIN1 Phase Separation and Inhibits Hepatocellular Carcinoma by Suppressing c-Myc Translation. Adv Sci (Weinh). 2022;9(8):e2103817. Liu B, Shen H, He J, Jin B, Tian Y, Li W, Hou L, Zhao W, Nan J, Zhao J, et al. Cytoskeleton remodeling mediated by circRNA-YBX1 phase separation supp resses the metastasis of liver cancer. Proc Natl Acad Sci U S A. 2023;120(30):e2220296120. Smyth EC, Nilsson M, Grabsch HI, van Grieken NC, Lordick F. Gastric cancer. Lancet. 2020;396(10251):635–48. Joshi SS, Badgwell BD. Current treatment and recent progress in gastric cancer. CA Cancer J Clin. 2021;71(3):264–79. Ramtohul T, Djerroudi L, Lissavalid E, Nhy C, Redon L, Ikni L, Djelouah M, Journo G, Menet E, Cabel L, et al. Multiparametric MRI and Radiomics for the Prediction of HER2-Zero, -Lo w, and -Positive Breast Cancers. Radiology. 2023;308(2):e222646. Tashakori M, Kadia T, Loghavi S, Daver N, Kanagal-Shamanna R, Pierce S, Sui D, Wei P, Khodakarami F, Tang Z, et al. TP53 copy number and protein expression inform mutation status across risk categories in acute myeloid leukemia. Blood. 2022;140(1):58–72. Tables Table 1 is available in the Supplementary Files section. Additional Declarations No competing interests reported. Supplementary Files Table1LLPSgeneset.xlsx WBOriginalImages.png Cite Share Download PDF Status: Posted Version 1 posted You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-4546744","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":320258901,"identity":"c7eca339-dd6d-4a4f-be2b-0da07723125f","order_by":0,"name":"Renjie Miao","email":"","orcid":"","institution":"The Affiliated Third Hospital of Zhenjiang to Jiangsu University","correspondingAuthor":false,"prefix":"","firstName":"Renjie","middleName":"","lastName":"Miao","suffix":""},{"id":320258904,"identity":"fc91fec5-86bb-4db0-8dac-ccddcb31cfc4","order_by":1,"name":"Yun Liu","email":"","orcid":"","institution":"Jiangsu University","correspondingAuthor":false,"prefix":"","firstName":"Yun","middleName":"","lastName":"Liu","suffix":""},{"id":320258905,"identity":"a6e4ff21-3f24-4347-87e4-564b881152a0","order_by":2,"name":"Ruiyun Chen","email":"","orcid":"","institution":"University of Health and Rehabilitation Sciences","correspondingAuthor":false,"prefix":"","firstName":"Ruiyun","middleName":"","lastName":"Chen","suffix":""},{"id":320258907,"identity":"afb98874-36f0-4d68-98bc-3f3bfdb95629","order_by":3,"name":"Zili Sun","email":"","orcid":"","institution":"The Affiliated Third Hospital of Zhenjiang to Jiangsu University","correspondingAuthor":false,"prefix":"","firstName":"Zili","middleName":"","lastName":"Sun","suffix":""},{"id":320258909,"identity":"d9e37255-2453-47b0-b0eb-3767c1cb95aa","order_by":4,"name":"Wei Zhang","email":"","orcid":"","institution":"The Affiliated Third Hospital of Zhenjiang to Jiangsu University","correspondingAuthor":false,"prefix":"","firstName":"Wei","middleName":"","lastName":"Zhang","suffix":""},{"id":320258911,"identity":"341a3c27-06dc-431e-950c-a05463b53a23","order_by":5,"name":"Rui Li","email":"","orcid":"","institution":"The Affiliated Third Hospital of Zhenjiang to Jiangsu University","correspondingAuthor":false,"prefix":"","firstName":"Rui","middleName":"","lastName":"Li","suffix":""},{"id":320258913,"identity":"b4b767c2-992a-4dc2-8758-a5354edff8a5","order_by":6,"name":"Rong Song","email":"","orcid":"","institution":"The Affiliated Third Hospital of Zhenjiang to Jiangsu University","correspondingAuthor":false,"prefix":"","firstName":"Rong","middleName":"","lastName":"Song","suffix":""},{"id":320258915,"identity":"7002633b-a2bb-4ae0-be2b-35256ab962b9","order_by":7,"name":"Xuemei Wang","email":"","orcid":"","institution":"The Affiliated Third Hospital of Zhenjiang to Jiangsu University","correspondingAuthor":false,"prefix":"","firstName":"Xuemei","middleName":"","lastName":"Wang","suffix":""},{"id":320258917,"identity":"69964398-a806-42ed-9662-925ed1f8ec40","order_by":8,"name":"Jinlan Wang","email":"","orcid":"","institution":"The Affiliated Third Hospital of Zhenjiang to Jiangsu University","correspondingAuthor":false,"prefix":"","firstName":"Jinlan","middleName":"","lastName":"Wang","suffix":""},{"id":320258919,"identity":"98f66ab4-3460-4c6b-b9a0-5742ce7355dc","order_by":9,"name":"Shuo Shen","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA0klEQVRIiWNgGAWjYBACNvn3Hx98qLDhkedvPkCcFj6GBGPDGWfSZAxnHEsgToscQ4KZNG/bYRuGAzkGRDqM4UCC5Aw2Zh7GhjMfb7xhsJPTbSCkhbHhgMEHHjYedubezZZzGJKNzQ4Q0sLM2JA4Q4IHaMvZbdI8DAcStxHUwsbMcJjHQAKoOOcZkVp42BibeRIMQFrYiNQiwcPMOONAAg8wkI0t5xgQ4Rf5GTzsPz7++28PjMqHN95U2MkR1IICJHiIjBpkLaTqGAWjYBSMghEBABFfPgkpXGsAAAAAAElFTkSuQmCC","orcid":"","institution":"The Affiliated Third Hospital of Zhenjiang to Jiangsu University","correspondingAuthor":true,"prefix":"","firstName":"Shuo","middleName":"","lastName":"Shen","suffix":""}],"badges":[],"createdAt":"2024-06-07 14:33:59","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-4546744/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-4546744/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":60344065,"identity":"e6909d01-d4f3-4fe2-9a73-cdc314f3f356","added_by":"auto","created_at":"2024-07-15 19:20:28","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":642220,"visible":true,"origin":"","legend":"\u003cp\u003eThe flowchart of this study\u003c/p\u003e","description":"","filename":"FIGURE1.png","url":"https://assets-eu.researchsquare.com/files/rs-4546744/v1/e31f3882096737c40faca37b.png"},{"id":60344839,"identity":"8e63a357-584b-47c2-b98c-d06110dd07cc","added_by":"auto","created_at":"2024-07-15 19:28:28","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":2047014,"visible":true,"origin":"","legend":"\u003cp\u003eSingle-cell analysis based on the LLPS gene set. \u003cstrong\u003eA\u003c/strong\u003e Cell quality control based on mitochondrial content, hemoglobin content, number of genes, and total mRNA molecules per cell. \u003cstrong\u003eB\u003c/strong\u003e and \u003cstrong\u003eC\u003c/strong\u003e PCA shows the uniform distribution of cells across three kinds of samples and the optimal PC value. \u003cstrong\u003eD \u003c/strong\u003eand\u003cstrong\u003eG \u003c/strong\u003eAnnotation of cells following clustering. \u003cstrong\u003eE \u003c/strong\u003eand\u003cstrong\u003e F\u003c/strong\u003e Visualization of marker gene expression in annotated cells. \u003cstrong\u003eH \u003c/strong\u003eand \u003cstrong\u003eI\u003c/strong\u003e Localization of cells with different LLPS scores.\u003c/p\u003e","description":"","filename":"FIGURE2.png","url":"https://assets-eu.researchsquare.com/files/rs-4546744/v1/b842c5a63fdbfef8a7d8a2f8.png"},{"id":60344066,"identity":"9d86c412-eb1b-4d47-9ab4-1af17cf1b0ce","added_by":"auto","created_at":"2024-07-15 19:20:28","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":1234217,"visible":true,"origin":"","legend":"\u003cp\u003eIdentification of subgroups of GC patients based on LLPS-related genes. \u003cstrong\u003eA \u003c/strong\u003eThe samples in the combined cohort of TCGA-STAD and GSE84437 are divided into two distinct clusters when \u003cem\u003ek \u003c/em\u003eequals 2.\u003cstrong\u003e \u003c/strong\u003e\u0026nbsp;\u003cstrong\u003eB\u003c/strong\u003eCDF curves for\u003cem\u003e k \u003c/em\u003eranging from 2 to 9. \u003cstrong\u003eC \u003c/strong\u003eThe proportional differences in the area beneath the CDF curves from \u003cem\u003ek \u003c/em\u003eranging from 2 to 9.\u003cstrong\u003e D \u003c/strong\u003eand\u003cstrong\u003e E\u003c/strong\u003e PCA conducted to validate the stability of the clustering. \u003cstrong\u003eF \u003c/strong\u003eThe Kaplan-Meier survival curves are depicted for clusters C1 and C2. \u003cstrong\u003eG\u003c/strong\u003e and\u003cstrong\u003e H\u003c/strong\u003e Heatmaps and volcano plots showing differential gene expression among the two clusters. \u003cstrong\u003eI\u003c/strong\u003eComparison of immune checkpoint gene level. *\u003cem\u003eP\u003c/em\u003e \u0026lt; 0.05, **\u003cem\u003eP \u003c/em\u003e\u0026lt; 0.01, ***\u003cem\u003eP\u003c/em\u003e \u0026lt; 0.001.\u003c/p\u003e","description":"","filename":"FIGURE3.png","url":"https://assets-eu.researchsquare.com/files/rs-4546744/v1/e0ad9878bff871536b82a8e9.png"},{"id":60344078,"identity":"3a9789ab-187e-48ee-96a2-7a5b1c0ebdfe","added_by":"auto","created_at":"2024-07-15 19:20:29","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":8112662,"visible":true,"origin":"","legend":"\u003cp\u003eAlterations in KEGG and GO pathways among GC patients in the two clusters. \u003cstrong\u003eA \u003c/strong\u003eto \u003cstrong\u003eC \u003c/strong\u003eEnrichment of KEGG pathways through DEGs in the two clusters. \u003cstrong\u003eE \u003c/strong\u003eto \u003cstrong\u003eG \u003c/strong\u003eEnrichment of GO pathways through DEGs in the two clusters. \u003cstrong\u003eD \u003c/strong\u003eand \u003cstrong\u003eH \u003c/strong\u003eThe heatmap displaying the outcomes of ssGSEA analysis on changes in KEGG and GO pathways within the two clusters. \u003cstrong\u003eI \u003c/strong\u003eand\u003cstrong\u003e J\u003c/strong\u003e Heatmap displaying the results of GSVA analysis of KEGG and GO pathway alterations within the two clusters.\u003c/p\u003e","description":"","filename":"FIGURE4.png","url":"https://assets-eu.researchsquare.com/files/rs-4546744/v1/c5e540062ea0555db74be21d.png"},{"id":60344838,"identity":"059b31c8-2ba7-41b2-bab4-16185a7fb0b7","added_by":"auto","created_at":"2024-07-15 19:28:28","extension":"png","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":1153907,"visible":true,"origin":"","legend":"\u003cp\u003eComparison in TME, tumor purity, and immunotherapy response among GC patients within two clusters. \u003cstrong\u003eA \u003c/strong\u003eand\u003cstrong\u003e B\u003c/strong\u003e Comparison of immune cell infiltration and heatmap visualization between the two clusters. \u003cstrong\u003eC \u003c/strong\u003eComparison of immune function infiltration within the two clusters. \u003cstrong\u003eD \u003c/strong\u003eto\u003cstrong\u003e F\u003c/strong\u003e Assessment of tumor purity and degree of malignancy in the two clusters using the ESTIMATE and XCELL algorithms. \u003cstrong\u003eG\u003c/strong\u003e and\u003cstrong\u003e H\u003c/strong\u003e Evaluation of immune escape capabilities and immunotherapy responses in the two clusters using the TIDE algorithm. \u003cstrong\u003eI \u003c/strong\u003eAssessment of tumor cell migratory capabilities in the two clusters using EMT scoring. \u003cstrong\u003eJ \u003c/strong\u003eEvaluation of tumor cell proliferation capabilities within the two clusters using angiogenesis scoring.\u003cstrong\u003e \u003c/strong\u003e*\u003cem\u003eP\u003c/em\u003e \u0026lt; 0.05, **\u003cem\u003eP \u003c/em\u003e\u0026lt; 0.01, ***\u003cem\u003eP\u003c/em\u003e \u0026lt; 0.001, ****\u003cem\u003eP\u003c/em\u003e\u0026lt; 0.0001.\u003c/p\u003e","description":"","filename":"FIGURE5.png","url":"https://assets-eu.researchsquare.com/files/rs-4546744/v1/b9d01a8a1adf251f138b2baf.png"},{"id":60344069,"identity":"e3c851e9-b75a-4289-99d2-6b47cfccd692","added_by":"auto","created_at":"2024-07-15 19:20:28","extension":"png","order_by":6,"title":"Figure 6","display":"","copyAsset":false,"role":"figure","size":2430422,"visible":true,"origin":"","legend":"\u003cp\u003eThe construction of the LLPS characteristic prognostic model and analysis of \u0026nbsp;LCGs based on four algorithms. \u003cstrong\u003eA \u003c/strong\u003eto\u003cstrong\u003e C\u003c/strong\u003e Lasso regression was used to visualize the selected genes after selecting the optimal lambda value. \u003cstrong\u003eB \u003c/strong\u003eto\u003cstrong\u003e F\u003c/strong\u003e RF regression analysis and visualization of the selected genes. \u003cstrong\u003eG \u003c/strong\u003eand\u003cstrong\u003e H\u003c/strong\u003e XGBoost regression analysis followed by visualization of the selected genes. \u003cstrong\u003eI\u003c/strong\u003e Venn diagram showing the intersection of genes across the three algorithms. \u003cstrong\u003eJ\u003c/strong\u003e Multivariable Cox regression analysis based on 17 intersecting genes. \u003cstrong\u003eK\u003c/strong\u003e Multivariable cox regression analysis incorporating clinical pathological features. \u003cstrong\u003eL\u003c/strong\u003e Spearman correlation analysis to display the correlations among ten LCGs. *\u003cem\u003eP\u003c/em\u003e \u0026lt; 0.05, **\u003cem\u003eP \u003c/em\u003e\u0026lt; 0.01, ***\u003cem\u003eP\u003c/em\u003e \u0026lt; 0.001.\u003c/p\u003e","description":"","filename":"FIGURE6.png","url":"https://assets-eu.researchsquare.com/files/rs-4546744/v1/8217cfe40d16b4e0a35cee4a.png"},{"id":60344080,"identity":"f29565ab-4962-4e85-a097-8f2c9dbafd4f","added_by":"auto","created_at":"2024-07-15 19:20:29","extension":"png","order_by":7,"title":"Figure 7","display":"","copyAsset":false,"role":"figure","size":50586725,"visible":true,"origin":"","legend":"\u003cp\u003eExpression of LCGs in the prognostic model. \u003cstrong\u003eA\u003c/strong\u003e Analysis of immunohistochemical images of LCGs from the HPA database: HOMER3 (antibody HPA040999, 10\u003csup\u003e×\u003c/sup\u003e ), CAPRIN1 (antibody HPA018126, 10\u003csup\u003e×\u003c/sup\u003e ), PAK2 (antibody CAB007794, 10\u003csup\u003e×\u003c/sup\u003e ), HSPA1A (antibody CAB032815, 10\u003csup\u003e×\u003c/sup\u003e ), MED19 (antibody HPA040860, 10\u003csup\u003e×\u003c/sup\u003e), DACT1 (antibody HPA003016, 10\u003csup\u003e×\u003c/sup\u003e), RNF2 (antibody HPA026803, 10\u003csup\u003e×\u003c/sup\u003e ), ARID1A (antibody HPA005456, 10\u003csup\u003e×\u003c/sup\u003e ), YBX1 (antibody CAB005875, 10\u003csup\u003e×\u003c/sup\u003e ), TBP (antibody HPA049805, 10\u003csup\u003e×\u003c/sup\u003e ). \u003cstrong\u003eB\u003c/strong\u003e Evaluation of the gene expression levels of LCGs retrieved from the UALCAN database. \u003cstrong\u003eC\u003c/strong\u003e Verification of the expression levels of LCGs using western blot analysis. *\u003cem\u003eP\u003c/em\u003e \u0026lt; 0.05, **\u003cem\u003eP \u003c/em\u003e\u0026lt; 0.01, ***\u003cem\u003eP\u003c/em\u003e \u0026lt; 0.001.\u003c/p\u003e","description":"","filename":"FIGURE7.png","url":"https://assets-eu.researchsquare.com/files/rs-4546744/v1/548e2ff4b74821f30b950e63.png"},{"id":60344067,"identity":"b13b3768-4716-42f1-ae40-34522123b45e","added_by":"auto","created_at":"2024-07-15 19:20:28","extension":"png","order_by":8,"title":"Figure 8","display":"","copyAsset":false,"role":"figure","size":1363991,"visible":true,"origin":"","legend":"\u003cp\u003eVerification of the stability and clinical value of the LLPS characteristic prognostic model. \u003cstrong\u003eA \u003c/strong\u003eand\u003cstrong\u003e B \u003c/strong\u003ePCA verification of the stability of clustering belonging to the high-risk category. \u003cstrong\u003eC \u003c/strong\u003eand\u003cstrong\u003eF\u003c/strong\u003e Relation between various clustering groups, risk scores, and risk groupings. \u003cstrong\u003eD\u003c/strong\u003e Heatmap displaying the relation of patient survival state and risk scores. \u003cstrong\u003eE \u003c/strong\u003eHeatmap of risk gene expression in risk groups.\u003cstrong\u003e G \u003c/strong\u003eHeatmap illustrating the correlation between patient clinical data and risk group categorization. \u003cstrong\u003eH \u003c/strong\u003eKaplan-Meier survival analysis of two risk groups.\u003cstrong\u003e I \u003c/strong\u003eto\u003cstrong\u003e N\u003c/strong\u003e Kaplan-Meier survival curves for patients in different subgroups based on age, N and T stages. *\u003cem\u003eP\u003c/em\u003e \u0026lt; 0.05, **\u003cem\u003eP \u003c/em\u003e\u0026lt; 0.01, ***\u003cem\u003eP\u003c/em\u003e\u0026lt; 0.001.\u003c/p\u003e","description":"","filename":"FIGURE8.png","url":"https://assets-eu.researchsquare.com/files/rs-4546744/v1/8c2c01778fef2efa5e564a9f.png"},{"id":60344072,"identity":"a715d720-e3fb-454e-a6bc-bdb850df9868","added_by":"auto","created_at":"2024-07-15 19:20:28","extension":"png","order_by":9,"title":"Figure 9","display":"","copyAsset":false,"role":"figure","size":1727848,"visible":true,"origin":"","legend":"\u003cp\u003eVerification of the LLPS characteristic prognostic model in three training cohorts. \u003cstrong\u003eA \u003c/strong\u003eto\u003cstrong\u003e C \u003c/strong\u003eRisk scores correlation with patient survival distribution and gene expression heatmaps in the GSE84433, GSE84426, and TCGA-STAD cohorts. \u003cstrong\u003eD \u003c/strong\u003eto\u003cstrong\u003eF \u003c/strong\u003ePCA validation confirms the reliability of the developed predictive model in the GSE84433, GSE84426, and TCGA-STAD cohorts.\u003cstrong\u003e G \u003c/strong\u003eto\u003cstrong\u003e I \u003c/strong\u003eKaplan-Meier survival curves and ROC curves are employed to evaluate the predictive accuracy of the prognostic model in the GSE84433, GSE84426, and TCGA-STAD cohorts. \u003cstrong\u003eJ \u003c/strong\u003eto\u003cstrong\u003eL \u003c/strong\u003eCalibration curves and ROC curves that vary over time for foreseeing survival at 1, 3 and 5 years in the GSE84433, GSE84426 and TCGA-STAD datasets.\u003c/p\u003e","description":"","filename":"FIGURE9.png","url":"https://assets-eu.researchsquare.com/files/rs-4546744/v1/8b36ba5ed44f38101d3c0195.png"},{"id":60344079,"identity":"5e977e6c-5c74-4ee5-965c-dadf55e50195","added_by":"auto","created_at":"2024-07-15 19:20:29","extension":"png","order_by":10,"title":"Figure 10","display":"","copyAsset":false,"role":"figure","size":636934,"visible":true,"origin":"","legend":"\u003cp\u003eConstruction and evaluation of nomograms. \u003cstrong\u003eA \u003c/strong\u003eand\u003cstrong\u003e D \u003c/strong\u003eDevelopment of a nomogram using LCGs, LLPS risk scores and additional clinical characteristics. \u003cstrong\u003eB\u003c/strong\u003eand\u003cstrong\u003e E \u003c/strong\u003eCalibration curves. \u003cstrong\u003eC \u003c/strong\u003eand \u003cstrong\u003eF\u003c/strong\u003e ROC curves plotting for evaluating the predictive performance of nomograms. *\u003cem\u003eP\u003c/em\u003e \u0026lt; 0.05, **\u003cem\u003eP \u003c/em\u003e\u0026lt; 0.01, ***\u003cem\u003eP\u003c/em\u003e\u0026lt; 0.001.\u003c/p\u003e","description":"","filename":"FIGURE10.png","url":"https://assets-eu.researchsquare.com/files/rs-4546744/v1/3172aefbe710472d63da4e92.png"},{"id":60344840,"identity":"f7a28e80-bbda-4011-9c2d-bb820984e333","added_by":"auto","created_at":"2024-07-15 19:28:28","extension":"png","order_by":11,"title":"Figure 11","display":"","copyAsset":false,"role":"figure","size":3695680,"visible":true,"origin":"","legend":"\u003cp\u003eComparison in TME, tumor purity, and immunotherapy response among GC patients in low and high groups. \u0026nbsp;\u003cstrong\u003eA \u003c/strong\u003eand\u003cstrong\u003e B \u003c/strong\u003eVisualization of spearman correlation analysis between feature genes and immune infiltration. \u003cstrong\u003eC \u003c/strong\u003eand\u003cstrong\u003e D\u003c/strong\u003e Analyzing the infiltration of immune cells in two groups by heatmap visualization. \u003cstrong\u003eE \u003c/strong\u003eAnalysis of immune system infiltration differences in two groups. \u003cstrong\u003eF \u003c/strong\u003eto\u003cstrong\u003e H\u003c/strong\u003e Assessment of tumor purity and degree of malignancy in the two groups using the ESTIMATE and XCELL algorithms. \u003cstrong\u003eI\u003c/strong\u003e and\u003cstrong\u003e K\u003c/strong\u003e TIDE algorithm evaluation of immune escape capabilities and immune therapy response in two groups. \u003cstrong\u003eL \u003c/strong\u003eEMT score assessment of tumor cell migration capability in two groups. \u003cstrong\u003eM \u003c/strong\u003eAngiogenesis score evaluation of tumor cell proliferation capacity among two groups. *\u003cem\u003eP\u003c/em\u003e \u0026lt; 0.05, **\u003cem\u003eP \u003c/em\u003e\u0026lt; 0.01, ***\u003cem\u003eP\u003c/em\u003e \u0026lt; 0.001.\u003c/p\u003e","description":"","filename":"FIGURE11.png","url":"https://assets-eu.researchsquare.com/files/rs-4546744/v1/1359ad415400e6dcd2cb93e3.png"},{"id":60344077,"identity":"33ca2b44-4d75-4fb7-a622-6eb3d033c875","added_by":"auto","created_at":"2024-07-15 19:20:29","extension":"png","order_by":12,"title":"Figure 12","display":"","copyAsset":false,"role":"figure","size":1172414,"visible":true,"origin":"","legend":"\u003cp\u003eVisualization involves calculating the pearson linear interrelation between the risk score and the scores for stromal, immune, ESTIMATE, TME,TIDE, angiogenesis, EMT and tumor purity.\u003c/p\u003e","description":"","filename":"FIGURE12.png","url":"https://assets-eu.researchsquare.com/files/rs-4546744/v1/08f91892dc01270ffb776c3b.png"},{"id":60344076,"identity":"ad2ec54e-0db7-4b65-9005-0ddcf9488cf6","added_by":"auto","created_at":"2024-07-15 19:20:29","extension":"png","order_by":13,"title":"Figure 13","display":"","copyAsset":false,"role":"figure","size":935451,"visible":true,"origin":"","legend":"\u003cp\u003eVisualization of drug sensitivity (IC50 \u0026lt; 1). *\u003cem\u003eP\u003c/em\u003e \u0026lt; 0.05, **\u003cem\u003eP \u003c/em\u003e\u0026lt; 0.01, ***\u003cem\u003eP\u003c/em\u003e \u0026lt; 0.001.\u003c/p\u003e","description":"","filename":"FIGURE13.png","url":"https://assets-eu.researchsquare.com/files/rs-4546744/v1/6fc7ed7744d73f44ad54bade.png"},{"id":60344075,"identity":"8733219a-6e75-4e3b-a2dc-728a50791bd4","added_by":"auto","created_at":"2024-07-15 19:20:28","extension":"png","order_by":14,"title":"Figure 14","display":"","copyAsset":false,"role":"figure","size":492479,"visible":true,"origin":"","legend":"\u003cp\u003eTCGA-STAD cohort tumor mutation analysis. \u003cstrong\u003eA \u003c/strong\u003eand\u003cstrong\u003e B \u003c/strong\u003eVisualization of the rates of mutation of ten LCGs in the high and low risk groups. \u003cstrong\u003eC \u003c/strong\u003eChromosomal locations of ten LCGs. \u003cstrong\u003eD \u003c/strong\u003eCNV of the ten LCGs.\u003cstrong\u003e \u003c/strong\u003e\u0026nbsp;\u003cstrong\u003eE \u003c/strong\u003eKaplan-Meier curves are plotted for various subgroups based on their TMB levels and risk scores.\u003c/p\u003e","description":"","filename":"FIGURE14.png","url":"https://assets-eu.researchsquare.com/files/rs-4546744/v1/0c6c3f6b91358d71e851660c.png"},{"id":60344071,"identity":"176b0260-6b43-4ec6-a66b-279ead8f72ab","added_by":"auto","created_at":"2024-07-15 19:20:28","extension":"xlsx","order_by":0,"title":"","display":"","copyAsset":false,"role":"supplement","size":35346,"visible":true,"origin":"","legend":"","description":"","filename":"Table1LLPSgeneset.xlsx","url":"https://assets-eu.researchsquare.com/files/rs-4546744/v1/5d2c670ea4c5f67c40cd9da3.xlsx"},{"id":60344074,"identity":"8c6d18c3-72bb-482c-9e47-3f15cfae4a90","added_by":"auto","created_at":"2024-07-15 19:20:28","extension":"png","order_by":17,"title":"","display":"","copyAsset":false,"role":"supplement","size":5164390,"visible":true,"origin":"","legend":"","description":"","filename":"WBOriginalImages.png","url":"https://assets-eu.researchsquare.com/files/rs-4546744/v1/66dffd3d9887ecaad2f20da6.png"}],"financialInterests":"No competing interests reported.","formattedTitle":"A Novel Liquid–Liquid Phase Separation Characteristic Model Associated with Prognosis and Immune Landscape of Gastric Cancer Patients","fulltext":[{"header":"Introduction","content":"\u003cp\u003eGastric cancer (GC) presents a formidable challenge and is considered the fifth most common cancer worldwide [\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e]. While incidence rates vary geographically, countries in Eastern Asia such as China continue to record high numbers [\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e]. The incidence of GC is closely associated with numerous factors including genetic mutations, aberrant gene expression, infection with Helicobacter pylori and Epstein-Barr virus infection [\u003cspan additionalcitationids=\"CR4\" citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e]. Additionally, novel immunotherapeutic strategies targeting the GC immune microenvironment are currently gaining interest [\u003cspan additionalcitationids=\"CR7\" citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e]. Furthermore, by utilizing certain cellular markers such as fibroblasts, macrophages, and exhausted T cells, it is possible to accurately forecast the outlook of patients with GC and their reaction to immunotherapy, thus aiding in guiding precision medicine more effectively [\u003cspan additionalcitationids=\"CR10\" citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eLiquid-liquid phase separation (LLPS) is an emerging field in tumor biology, recognized for its crucial role in organizing intracellular biomolecules into membrane-less organelles [\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e]. Often likened to oil droplets separating in water, LLPS facilitates the stable concentration of specific proteins and nucleic acids, influencing various biological functions within the cell. LLPS maintains a relatively stable equilibrium within cells [\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e]. Disruption of this equilibrium results in dysregulated LLPS, affecting key cellular processes such as chromatin cycling, DNA damage repair, and cell signaling pathways [\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e]. Crucially, abnormal LLPS can stabilize oncogenic proteins, facilitate the evasion of tumor suppressor factors, and promote tumor immune escape, thereby enhancing the oncogenic potential of tumor cells [\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e]. Current research on tumor-related abnormal LLPS suggests that targeting LLPS-induced protein or nucleic acid aggregates could represent a novel approach to cancer therapy. For instance, targeting Sentrin/SUMO-specific protease, which undergoes abnormal LLPS, can restore the recruitment of RNF168 to DNA damage sites, enhancing DNA repair functions to maintain genomic integrity [\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e]. Research in prostate cancer has revealed that the LLPS-associated gene ET516 disrupts the androgen receptor complex, inhibiting the growth of cells with androgen receptor-resistant mutations. Targeting ET516-induced protein aggregates post-abnormal LLPS shows promise in treating prostate cancer [\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e]. With the rise of machine learning and artificial intelligence (AI), combined with transcriptomics, single-cell genomics, and spatial transcriptomics, effective LLPS characteristic prognostic models have been developed for certain cancers like breast cancer and glioblastoma [\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e, \u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e]. These models facilitate better stratification of patients based on risk levels, guiding prognosis more effectively. Nevertheless, the involvement of LLPS in GC is not well comprehended, necessitating further research to explore its implications and potential as a therapeutic target.\u003c/p\u003e \u003cp\u003eIn this study, we obtained a LLPS gene set from PhaSepDB and initiated our investigation by analyzing single-cell data from the GSE167297 cohort. Preliminary analysis focused on the cellular localization and differential gene pathway enrichment of cells across different LLPS scoring groups. Subsequent clustering analysis in a merged cohort of TCGA-STAD and GSE84437 revealed that samples of GC patients could be clearly separated into two clusters using LLPS gene set, with these clusters showing significant differences in pathway enrichment, prognostic features, tumor immune microenvironment (TME), and immunoreactivity. Following this, we employed a combination of Lasso, XGBoost, Random Forest (RF) and multivariate Cox regression analyses to successfully develop a novel LLPS-based prognostic model, incorporating ten LLPS characteristic genes (LCGs): HOMER3, CAPRIN1, PAK2, HSPA1A, MED19, DACT1, RNF2, ARID1A, YBX1 and TBP. This model demonstrated robust performance in guiding prognosis for GC patients. To verify the model's effectiveness, we conducted external validations using three independent cohorts, which confirmed the model's strong predictive capabilities. Particularly noteworthy is our exploration of the new model's application in analyzing the TME, immunoreactivity, tumor mutational burden (TMB), and chemotherapy drug sensitivity across different risk strata of GC patients, affirming its high clinical value. Finally, we utilized the HPA and UALCAN databases as well as western blotting to confirm the high expression state of ten LCGs in GC tissues compared to normal tissues. This comprehensive approach not only confirms the utility of the LLPS gene set in stratifying GC but also highlights the potential of our prognostic model in personalizing patient management and guiding therapeutic decisions. Through the rigorous validation and analysis, our findings support the integration of LLPS-associated biomarkers in predictive models, offering a novel pathway for improving the accuracy of GC therapy approaches.\u003c/p\u003e"},{"header":"Materials and Methods","content":"\u003cdiv id=\"Sec3\" class=\"Section2\"\u003e \u003ch2\u003eData acquisition\u003c/h2\u003e \u003cp\u003eIn this research, transcriptome data, mutation data, and clinical information for 415 GC samples along with 34 normal samples were downloaded from TCGA database, accessed on March 8, 2024. Additionally, the transcriptomic data, single-cell data, and clinical information for cohorts GSE84437, GSE84426, GSE84433, and GSE167297 were obtained from the GEO database, viewed on March 12, 2024. Furthermore, a total of 818 genes associated with LLPS (Table\u0026nbsp;1) were downloaded from PhaSepDB (\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttp://db.phasep.pro/browse/\u003c/span\u003e\u003cspan address=\"http://db.phasep.pro/browse/\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e) [\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e]. Utilizing R software, transcriptome data from the TCGA-STAD and GSE84437 cohorts were extracted and normalized using limma and sva packages, followed by batch correction. After removing samples with a survival time of less than zero days, expression data for 797 GC samples were prepared for subsequent combined analyses.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec4\" class=\"Section2\"\u003e \u003ch2\u003eAnalysis of GSE167297 scRNA-seq data\u003c/h2\u003e \u003cp\u003eWe obtained single-cell data from the GSE167297 dataset with two samples: GSM5101019, and GSM5101020. The quality control procedures were implemented as follows: 1) Cells that express less than 200 genes were deemed ineligible for analysis; 2) Cells with mitochondrial gene content exceeding 10% were removed, as elevated mitochondrial gene expression can indicate cell stress or apoptosis; 3) For the analysis, we employed the FindIntegrationAnchors function from the Seurat package with a setting of 2000 anchors, followed by data integration using the IntegrateData function.\u003c/p\u003e \u003cp\u003eTo address batch effects between samples, we applied the SCT method. Dimensionality reduction was carried out using PCA, choosing 15 principal components. Cells were clustered through the t-SNE method.\u003c/p\u003e \u003cp\u003eThe identification of cell types was carried out through cell surface marker genes, which were obtained from a CellMarker database (\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://http://bio-bigdata.hrbmu.edu.cn/CellMarker/\u003c/span\u003e\u003cspan address=\"https://http://bio-bigdata.hrbmu.edu.cn/CellMarker/\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e, accessed on 12 March 2024). Using the LLPS gene set, we scored cells via the AddModuleScore function in Seurat.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec5\" class=\"Section2\"\u003e \u003ch2\u003eConsensusClusterPlus analysis\u003c/h2\u003e \u003cp\u003eThe ConsensusClusterPlus analysis is a commonly used classification analysis method in cancer research, and stable clustering of samples can be achieved using the ConsensusClusterPlus package [\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e]. We applied this method to cluster GC patients based on LLPS gene set. Furthermore, agglomerative PAM clustering using Euclidean distance was employed, and the analysis was repeated on 80% of the samples 100 times. The ideal number of clusters, found to be two, was determined through the empirical cumulative distribution function. Subsequently, PCA was conducted with the prcomp function in the stats package to validate the reliability of the two identified cluster.\u003c/p\u003e \u003cp\u003eKaplan-Meier survival analysis was utilized on various LLPS clusters through the survival package and survminer package. Gene differential analysis was carried out with the limma package to identify genes that differ significantly between clusters. These DEGs were visualized using ggplot2, pheatmap, and ggVolcano packages, with a display limit of 50 genes. Finally, we obtained genes related to immune checkpoints from the MSigDB (\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://www.gsea-msigdb.org/gsea/msigdb\u003c/span\u003e\u003cspan address=\"https://www.gsea-msigdb.org/gsea/msigdb\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e) and investigated the differential expression of these immune checkpoint-related genes across the two clusters [\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eIn order to delve deeper into the biological processes of the clusters, we evaluated the biological differences between the two clusters using three pathway enrichment methods. First, we conducted Gene Set Enrichment Analysis (GSEA), which assesses the distribution pattern of genes from a given gene set in a ranked gene list according to their correlation with a phenotype, thereby determining their contribution to the phenotype [\u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e]. We downloaded the KEGG gene set \"c2.cp.kegg_legacy.v2023.2.Hs.symbols\" from the MSigDB database, ranked the DEGs among clusters, and then performed GSEA. Second, we performed ssGSEA to study the enrichment of gene sets in individual samples using the GSVA package, analyzing the differences in gene set enrichment between different clusters. Finally, we conducted KEGG enrichment analysis on the DEGs between clusters utilizing the clusterProfiler package, compared the enrichment results with those from GSE167297, and visualized the results using a Venn diagram.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec6\" class=\"Section2\"\u003e \u003ch2\u003eAnalysis of TME and response to immunotherapy in two LLPS-related clusters\u003c/h2\u003e \u003cp\u003eWe downloaded gene sets related to immune cells and immune functions [\u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e]. The ssGSEA method was employed to determine the immune gene set scores for each sample, enabling the evaluation of immune infiltration variations across different clusters.\u003c/p\u003e \u003cp\u003eTIDE score is a valuable tool for evaluating the likelihood of immune therapy responses and the potential for tumor immune escape [\u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e]. We utilized the TIDE website (\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttp://tide.dfci.harvard.edu/\u003c/span\u003e\u003cspan address=\"http://tide.dfci.harvard.edu/\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e) to upload expression matrix data from a combined cohort that included TCGA-STAD and GSE84437 datasets. From this, we derived TIDE scores and predictions for immune therapy responses for each sample. An elevated TIDE score represents an increased chance of tumor immune escape, allowing us to differentiate between groups based on this criterion. Further analysis was conducted to assess the tumor purity using the ESTIMATE and XCELL packages. Within these tools, the stromal score represents the level of stromal cell infiltration in the tumor tissue, and the immune score reflects the extent of immune cell infiltration. Summing these scores provides both the ESTIMATE score and the TME assessment. The derived scores give an inference of tumor purity, closely linked with the malignancy level of the tumor [\u003cspan additionalcitationids=\"CR27\" citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eSubsequent analysis involved computing scores for angiogenesis and EMT using gene sets from the MSigDB database and previous research [\u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e, \u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e]. These computations were made using the ssGSEA method within the GSVA package. The angiogenesis and EMT scores offer deeper insights into the proliferative and migratory capabilities of the tumor.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec7\" class=\"Section2\"\u003e \u003ch2\u003eDevelopment of a LLPS characteristic prognostic model\u003c/h2\u003e \u003cp\u003eTo identify characteristic genes associated with the LLPS characteristic prognostic model, we utilized four machine learning algorithms combined with survival duration, survival state, and gene expression matrix from 797 samples, thereby constructing a prognostic model. Initially, we employed the glmnet package for LASSO analysis and utilized 10-fold cross-validation to identify the most suitable genes. Subsequently, the XGBoost package was applied for XGBoost analysis. XGBoost, a robust classification algorithm, amalgamates various tree models to form an enhanced tree model. Through this algorithm, the top 50 optimal genes were identified. Following this, the randomForestSRC package was used for RF analysis to ascertain the optimal genes. RF, an ensemble learning method composed of multiple decision trees, is adept at handling datasets with nonlinear relationships. The intersection of genes identified by the aforementioned three algorithms was then exposed to multivariate COX stepwise regression procedure to pinpoint the ten most significant characteristic genes related to the LLPS prognostic model.\u003c/p\u003e \u003cp\u003eSpearman correlation analysis is a conventional statistical method to determine the correlation between two non-normally distributed continuous variables. Upon identifying these characteristic genes, spearman correlation analysis was conducted to ascertain the interrelationships among these ten genes. According to their median risk scores, the samples were separated into high and low risk groups, which were calculated from LCGs expressions. The calculation equation for the risk score was shown as belows:\u003c/p\u003e \u003cp\u003eRisk score\u0026thinsp;=\u0026thinsp;Expression\u003csub\u003emRNA1\u003c/sub\u003e \u0026times; Coef\u003csub\u003emRNA1\u003c/sub\u003e + Expression\u003csub\u003emRNA2\u003c/sub\u003e \u0026times; Coef\u003csub\u003emRNA2\u003c/sub\u003e +\u0026hellip; Expression\u003csub\u003emRNAn\u003c/sub\u003e \u0026times; Coef\u003csub\u003emRNAn\u003c/sub\u003e.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec8\" class=\"Section2\"\u003e \u003ch2\u003eValidation of a LLPS characteristic prognostic model\u003c/h2\u003e \u003cp\u003eWe initiated with a dimensional reduction analysis using the prcomp function from the PCA package. This step was crucial for validating the reliability of our risk stratification. Subsequently, we compared risk scores between the consensus clustering groups C1 and C2 to examine their consistency. For visual analysis, we employed the pheatmap package to create heat maps displaying the risk scores, patient survival distribution, and expression of risk genes across different risk groups. This visual representation facilitated a clearer understanding of the survival outcomes and gene expression variations between groups. Further, we utilized the ggplot2 package to construct Sankey diagrams, elucidating the distribution differences of patients from varying T stages, survival statuses, and consensus clustering groups within the risk categories. Additionally, clinical information heatmaps were generated using the ComplexHeatmap package, comparing variables such as consensus cluster groups, T stages, N stages, gender, and survival status across different risk groups. Lastly, Kaplan-Meier survival analysis was conducted with the survival packages for diverse patient subsets. This included patients aged\u0026thinsp;\u0026le;\u0026thinsp;60 years, \u0026gt; 60 years, those in stages T1-2, stages T3-4, and in N stages ranging from N0-1 and N2-3. These analyses were critical for further validation of the risk group stratification's reliability.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec9\" class=\"Section2\"\u003e \u003ch2\u003eDevelopment and validation of nomograms\u003c/h2\u003e \u003cp\u003eNomograms are developed based on multivariable regression analysis, where each influencing factor in the model is assigned a score based on its contribution to the outcome variable. These scores are then summed in order to forecast the probability of a patient's survival at 1, 3, and 5 years. We utilized Cox regression through the rms package to construct separate nomograms for risk genes and clinical information, based on independent prognostic outcomes including risk genes, age, gender, and clinical staging. The performance and predictive ability of the models were evaluated with the receiver operating characteristic (ROC) curve, which allowed for the calculation of the area under the curve (AUC). For this purpose, we employed the pROC and timeROC packages to plot the ROC curves for 1, 3, and 5 years, as well as for clinical information, to study the predictive capacity of the constructed models. Additionally, calibration curves were utilized to further evaluate the accuracy of the models. These tools allow for a comprehensive evaluation of the models, ensuring that they provide reliable and accurate predictions of patient outcomes based on a combination of genetic and clinical factors.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec10\" class=\"Section2\"\u003e \u003ch2\u003eStudy of TME and immune response in the risk group\u003c/h2\u003e \u003cp\u003eTo elucidate the interrelation between ten LCGs and immune cells as well as immune functions, we employed ssGSEA scoring followed by the use of the linkET and ggplot2 packages to generate a spearman correlation heatmap. In the heatmap, each cell displays the correlation between specific immune cell types or immune function types and specific genes. The intensity of color in each cell reflects the strength of the correlation, with darker shades representing stronger correlations. Positive correlations are represented by the color blue and negative correlations are represented by the color blue. Additionally, the color and thickness of the lines within the heatmap signify the strength and direction of the correlations. Subsequently, we employed the ssGSEA method from the GSVA package to compute the scores of the immune gene sets for each sample, in order to assess the infiltration levels and functions of immune cells across different clusters. Finally, we studied the TME and immune therapy responses of patients in different risk groups using algorithms and scores such as TIDE, ESTIMATE, XCELL, EMT scoring and angiogenesis scoring.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec11\" class=\"Section2\"\u003e \u003ch2\u003ePearson linear correlation analysis\u003c/h2\u003e \u003cp\u003ePearson linear correlation analysis is a traditional statistical technique used to quantify the magnitude and orientation of the linear correlation between two variables.Then the pearson linear correlation analysis was adopted to detect the linear relationships between risk scores and various other metrics including stromal scores, immune scores, ESTIMATE scores, TME scores, EMT scores, angiogenesis scores, tumor purity scores, and TIDE scores.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec12\" class=\"Section2\"\u003e \u003ch2\u003eSensitivity analysis of chemotherapy drugs in the risk group\u003c/h2\u003e \u003cp\u003eAnticancer drug training set data were downloaded from the Genomics of Drug Sensitivity in Cancer (GDSC) website, and the chemotherapy drug sensitivity of 797 patients in the merged cohort was predicted using the oncopdict package [\u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e31\u003c/span\u003e]. Statistical tests were conducted between high and low risk groups to identify drugs with significant differences in sensitivity, characterized by a median IC50\u0026thinsp;\u0026lt;\u0026thinsp;1.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec13\" class=\"Section2\"\u003e \u003ch2\u003eValidation of the training set of the LLPS characteristic prognostic model\u003c/h2\u003e \u003cp\u003eBased on the LLPS characteristic prognostic model, risk scores were calculated for ten LCGs within the GSE84433, GSE84437, and TCGA-STAD cohorts. Subsequently, the stability of the model was validated through PCA. The risk scores, patient survival distributions, and expression of risk genes across different risk groups were visualized using the pheatmap package to observe variations in patient survival and risk gene expression among the groups. Finally, the performance of this novel LLPS prognostic model was validated by generating Kaplan-Meier survival analysis curves and the ROC curves for 1, 3, and 5 years using the survival, survminer, timeROC, and pROC packages.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec14\" class=\"Section2\"\u003e \u003ch2\u003eTMB analysis in the TCGA-STAD cohort\u003c/h2\u003e \u003cp\u003eThe TCGA-STAD cohort provides detailed mutation data, so we employed the maftools package to analyze somatic mutation data obtained from TCGA-STAD, assessing the mutational differences in characteristic genes across different LLPS scoring groups. Copy number variation (CNV), which is closely associated with the activation of oncogenes, results from genomic rearrangements and generally refers to genes longer than 1 kb. We downloaded CNV data for TCGA-STAD and the \"UCSC.HG19.Human.CytoBandIdeogram\" from the UCSC database. The chromosomal locations and copy number variations of characteristic genes were analyzed using the RCircos and rtracklayer packages. Finally, Kaplan-Meier survival analysis for different TMB groups was analyzed using the survival package.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec15\" class=\"Section2\"\u003e \u003ch2\u003eValidation of LCGs expression of LLPS characteristic prognostic model\u003c/h2\u003e \u003cp\u003eWe validated the expression of the characteristic genes through the HPA and UALCAN databases [\u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e32\u003c/span\u003e].\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec16\" class=\"Section2\"\u003e \u003ch2\u003eCell Culture\u003c/h2\u003e \u003cp\u003eGastric mucosal epithelial cells (GES-1) and three types of GC cells (HGC-27, MKN-45 and SGC-7901), were employed in this study. These cell lines originated from the School of Medicine, Jiangsu University. As for the cell culture process, it involved the use of DMEM augmented with ten percent FBS. The process took place in a humidified incubator maintained 37 degrees Celsius with five percent CO2 atmosphere.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec17\" class=\"Section2\"\u003e \u003ch2\u003eWestern Blotting Analysis\u003c/h2\u003e \u003cp\u003eThe RIPA lysis buffer was adopted to destruct cells (GES-1, HGC-2, MKN-45, SGC-7901). The buffer was fortified with PMSF along with phosphatase inhibitors. A 10% SDS-PAGE gel was utilized to separate proteins. And the protein was then transferred to the PVDF membrane. The membrane was then blocked and incubated with the following antibodies (Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e2\u003c/span\u003e) overnight, the GAPDH protein serves as an internal reference. The membrane was cleaned and incubated with rabbit secondary antibody at room temperature. Following another wash, ECL exposure system was employed to shoot membranes containing proteins, all conducted as per the instructions provided by the manufacturer.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab1\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 2\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003e The antibodies used in this study\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"3\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colspan=\"3\" nameend=\"c3\" namest=\"c1\"\u003e \u003cp\u003eNames of antibodies Dilution rate Source of antibodies\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eHOMER3\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e1:500\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eWanleibio\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eTBP\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e1:500\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eWanleibio\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eYBX1\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e1:500\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eWanleibio\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMED19\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e1:500\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eWanleibio\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eHSPA1A\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e1:400\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eWanleibio\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003ePAK2\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e1:500\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eWanleibio\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eRNF2\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e1:500\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eWanleibio\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eDACT1\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e1:500\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eWanleibio\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eRNF2\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e1:400\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eWanleibio\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eCAPRIN1\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e1:400\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eWanleibio\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eGAPDH\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e1:2000\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eAbcam\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec18\" class=\"Section2\"\u003e \u003ch2\u003eStatistical analysis and visualization\u003c/h2\u003e \u003cp\u003eFor this research, data analysis and visualization were carried out using R software, version 4.3.2. Correlation assessments were conducted using both pearson and spearman correlation analyses. Comparisons between two groups were carried out using the wilcoxon and limma tests. All data results were thought statistically significant at a threshold of \u003cem\u003eP\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;0.05.\u003c/p\u003e \u003c/div\u003e"},{"header":"Results","content":"\u003cdiv id=\"Sec20\" class=\"Section2\"\u003e \u003ch2\u003eResearch flow chart\u003c/h2\u003e \u003cp\u003eAn overall flow chart of this study was plotted to make the study easy to understand (Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e).\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec21\" class=\"Section2\"\u003e \u003ch2\u003eSingle-cell analysis of GC for localization of cells with different LLPS scores\u003c/h2\u003e \u003cp\u003eIn the GSE167297 cohort, two samples, GSM5101019 and GSM5101020, were downloaded, representing superficial cancer and deep cancer from the same patient, respectively. The data underwent filtering, batch correction, and cell quality control, setting the threshold for mitochondrial gene content at less than 10%, hemoglobin gene content at less than 5%, the number of genes observed per cell ranges from 200 to 5000, while the total number of mRNA molecules detected within a cell falls between 200 and 30,000. This preprocessing identified a total of 8,062 cells (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003eA). PCA demonstrates the stable distribution of cells for two types of samples, with the optimal number of PC being 15 (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003eB,C), resulting in the classification of cells into 19 clusters (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003eD). Cells were then annotated into seven categories based on cellular markers, types of cells include GC cells, T cells, B cells, endothelial cells, fibroblasts, myeloid cells and mast cells (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003eG). Visualization of marker gene expression was achieved through heat maps (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003eE,F). Following cell annotation, cells were scored using the LLPS gene set. After scoring, cells were categorized into LLPS-up and LLPS-dwon scoring subgroups. It was noted that cells with LLPS-up scoring subgroups were mostly found in endothelial cells, T cells, B cells, GC cells, suggesting that LLPS may primarily occur in these cell types within GC, potentially affecting the EMT, TME and immune infiltration (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003eH,I).\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003e \u003cb\u003eConsensusClusterPlus analysis based on the transcriptomic profiles of LLPS gene set reveals heterogeneous subgroups within GC patients\u003c/b\u003e \u003c/p\u003e \u003cp\u003eWe performed consensus ConsensusClusterPlus analysis on 797 samples from a combined cohort of TCGA-STAD and GSE84437, using a LLPS gene set obtained from the PhaSepDB database. The optimal K value was determined to be two based on the cumulative distribution function (CDF) curve, effectively separating 797 samples into two clusters: C1 and C2 (Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003eA-C). Subsequent PCA analysis further confirmed the stability of these two clusters (Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003eD,E). Survival curves exhibited that GC patients of cluster C2 had significantly lower survival rates compared to those in cluster C1 (Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003eF). DEGs between the two clusters was examined using the limma test, identifying 1,222 genes with significant differences, the chart displayed the 50 most noteworthy LCGs (Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003eG,H). Additionally, variations in the expression of immune checkpoint genes were noted, indicating potential distinctions in the TME and reactions to immunotherapy across the clusters (Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003eI).\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec22\" class=\"Section2\"\u003e \u003ch2\u003eDifferential enrichment of KEGG and GO pathways in two LLPS-related clusters\u003c/h2\u003e \u003cp\u003eDEGs underwent pathway enrichment analysis using KEGG and GO databases between clusters C1 and C2. The KEGG pathway enrichment result revealed significant differences across pathways related to cell cycle, transcriptional misregulation in cancer, IL-17, p53, mismatch repair, DNA replication and etc (Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003eA-C). Changes in these pathways are closely related to human diseases, cellular processes, genetic information processing and etc. The GO pathway enrichment result showed significant differences across pathways related to cell cycle, cell cycle phase transition, cell division,chromosome and etc (Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003eE-G). Finally, the pathway variations between the two clusters were also confirmed using both ssGSEA and GSVA methods, with consistent results across all three approaches regarding the differences in GO-KEGG pathways, the pathway changes in C2 cluster mainly focus on mitosis, cell cycle, DNA replication and so on (Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003eD,H,I,J).\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cdiv id=\"Sec23\" class=\"Section3\"\u003e \u003ch2\u003eDifferences in TME, immune infiltration and immunotherapy response between two LLPS-related clusters\u003c/h2\u003e \u003cp\u003eAfter calculating the immune infiltration status of two clusters using the ssGSEA method, we observed significant differences between Cluster C1 and Cluster C2 in several parameters. These include the levels of immune cell infiltration, chemokine receptor levels, human leukocyte antigen, T-cell co-stimulation levels, and reactions to Type I and Type II interferons (Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003eA-C). Additionally, We utilized the ESTIMATE and XCELL algorithms for tumor stromal score and immune score analyses. The sum of these scores provided the ESTIMATE score and the TME score, from which we inferred the tumor purity score. We found that, compared to Cluster C1, Cluster C2 had a significantly lower tumor purity score, suggesting a lower tumor purity and a higher degree of malignancy in Cluster C2 (Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003eD-F). Further analysis of the EMT and angiogenesis scores revealed that Cluster C2 also significantly exceeded Cluster C1 in these aspects. This suggests enhanced migratory capabilities and angiogenic potential in the tumor cells of Cluster C2, consistent with the results of single-cell analysis where high-scoring LLPS cells were localized to endothelial cells (Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003eI,J). Finally, through TIDE analysis of the two clusters, we noted that C2 contained a TIDE score that was markedly higher than the TIDE score of C1. Moreover, the immunotherapeutic responsiveness of C2 was significantly lower than that of C1, indicating a stronger capacity for immune evasion and lower sensitivity to immunotherapy in C2 (Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003eG,H).\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv id=\"Sec24\" class=\"Section2\"\u003e \u003ch2\u003eThe establishment and validation of the LLPS characteristic prognostic model\u003c/h2\u003e \u003cp\u003eFollowing the identification of the clustering function by the LLPS gene set, we conducted dimensionality reduction on the LLPS gene set using survival time and status data from 797 GC patients to identify prognostic model characteristic genes. Initially, Lasso regression was utilized to select LCGs, choosing a lambda value of 0.04 to identify 45 characteristic genes (Fig.\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003eA-C). Subsequently, Subsequently, the RF algorithm was adopted to study the prognostic relevance of each gene, resulting in the selection of 71 LCGs (Fig.\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003eD-F). Ultimately, the XGBoost algorithm was used to select the top 50 optimal genes, and the intersection of results from the three algorithms yielded 17 LCGs (Fig.\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003eG-I).\u003c/p\u003e \u003cp\u003eConducting a multivariate Cox regression method of these 17 genes established an optimal combination of 10 LCGs (HOMER3, CAPRIN1, PAK2, HSPA1A, MED19, DACT1, RNF2, ARID1A, YBX1, TBP). Spearman analysis revealed significant correlations among these ten genes, with four genes (CAPRIN1, PAK2, RNF2, TBP) potentially acting as protective factors, highly expressed in tumor cells and associated with longer survival periods (Fig.\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003eJ,L).\u003c/p\u003e \u003cp\u003eThe LLPS characteristic prognostic model built with these ten genes categorized samples into high and low risk subgroups by the median risk score. The risk score was computed using the following formula:\u003c/p\u003e \u003cp\u003eRisk score =\u003c/p\u003e \u003cdiv id=\"Sec25\" class=\"Section3\"\u003e \u003ch2\u003e(0.220851788535136) \u0026times; HOMER3 + (-0.22193287559865) \u0026times; CAPRIN1\u003c/h2\u003e \u003cdiv id=\"Sec26\" class=\"Section4\"\u003e \u003ch2\u003e+ (-0.104355501176279) \u0026times; PAK2 + (0.0900195727621234) \u0026times; HSPA1A\u003c/h2\u003e \u003cp\u003e+ (0.553288202545932) \u0026times; MED19 + (0.115430780509586) \u0026times; DACT1\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv id=\"Sec27\" class=\"Section2\"\u003e \u003ch2\u003e+ (-0.371052343828004) \u0026times; RNF2 + (0.113397725961348) \u0026times; ARID1A\u003c/h2\u003e \u003cdiv id=\"Sec28\" class=\"Section3\"\u003e \u003ch2\u003e+ (0.352859164860805) \u0026times; YBX1 + (-0.291531356154402) \u0026times; TBP\u003c/h2\u003e \u003cp\u003eAfter adjusting for age, gender and clinical stage in the multivariate Cox analysis, the LLPS risk score, N stage, and age were determined to be independent prognostic factors (Fig.\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003eK).\u003c/p\u003e \u003cp\u003eTo further confirm whether LCGs is highly expressed in GC tissues compared to gastric normal tissues, we obtained immunohistochemical images of the ten LCGs from the HPA database to further confirm the expression of these genes in GC tissues, the findings indicated that all LCGs exhibited high levels of expression (Fig.\u0026nbsp;\u003cspan refid=\"Fig7\" class=\"InternalRef\"\u003e7\u003c/span\u003eA). Similarly, expression data obtained from the UALCAN database for the ten LCGs showed results consistent with the immunohistochemical staining, ten genes show a significantly higher level of expression (Fig.\u0026nbsp;\u003cspan refid=\"Fig7\" class=\"InternalRef\"\u003e7\u003c/span\u003eB). The western blotting analysis results indicated the level of LCG expression in gastric cancer cells (MKN-45, HGC-27, SGC-7901) was higher than that of GSE-1 in vitro cell experiments (Fig.\u0026nbsp;\u003cspan refid=\"Fig7\" class=\"InternalRef\"\u003e7\u003c/span\u003eC,D).\u003c/p\u003e \u003cp\u003eAfter constructing the LLPS prognostic model, PCA confirmed the model's stability, with risk genes consistently distributed between two groups (Fig.\u0026nbsp;\u003cspan refid=\"Fig8\" class=\"InternalRef\"\u003e8\u003c/span\u003eA,B). The Sankey diagram and violin plot showed that samples in Cluster C2 had notably higher risk scores compared to those in Cluster C1, with the majority of samples in Cluster C2 falling into the high-risk category (Fig.\u0026nbsp;\u003cspan refid=\"Fig8\" class=\"InternalRef\"\u003e8\u003c/span\u003eC,F). Risk score analysis combined with patient survival distribution and risk gene expression revealed that the majority of the deceased cases were found within the high-risk group. In the low-risk group, the expression of risk genes (CAPRIN1, PAK2, RNF2, TBP) were at higher levels, while the high-risk group showed the expression of risk genes (HOMER3, HSPA1A, MED19, DACT1, TBP, ARID1A) were at higher levels (Fig.\u0026nbsp;\u003cspan refid=\"Fig8\" class=\"InternalRef\"\u003e8\u003c/span\u003eD,E). Finally, a chi-square test analyzed the allocation of samples across different clinical stages and risk subgroups, finding correlations between cluster distribution, different T stages, patient survival status, and risk groups (Fig.\u0026nbsp;\u003cspan refid=\"Fig8\" class=\"InternalRef\"\u003e8\u003c/span\u003eG). Kaplan-Meier analysis further clarified the clinical worth of the prognostic model, revealing that individuals who were classified as high-risk experienced noticeably shorter survival times. Further analysis of subgroups categorized by age, N stage and T stage revealed that individuals in the high-risk category had notably shorter survival durations across all subgroups (age\u0026thinsp;\u0026le;\u0026thinsp;60, age\u0026thinsp;\u0026gt;\u0026thinsp;60, N0-1 stage, N2-3 stage, T1-2 stage, T2-3 stage).\u003c/p\u003e \u003cp\u003eAfter constructing the LLPS prognostic model using the merged TCGA-STAD and GSE84437 cohorts, we further validated the stability and accuracy of the LLPS prognostic model using three independent cohorts (GSE84433, GSE84426, TCGA-STAD) as training sets. In these independent cohorts survival state and expression of LCGs were consistent with those in the merged TCGA-STAD and GSE84437 cohorts (Fig.\u0026nbsp;\u003cspan refid=\"Fig9\" class=\"InternalRef\"\u003e9\u003c/span\u003eA-C). PCA confirmed the stability of the three independent models (Fig.\u0026nbsp;\u003cspan refid=\"Fig9\" class=\"InternalRef\"\u003e9\u003c/span\u003eD-F). Subsequent Kaplan-Meier analysis exhibited that the high-risk subgroup had notably shorter survival periods compared to the low-risk subgroup across all three cohorts. ROC curves evaluated the predictive performance of three models in independent cohorts, all of which demonstrated good predictive capabilities (Fig.\u0026nbsp;\u003cspan refid=\"Fig9\" class=\"InternalRef\"\u003e9\u003c/span\u003eG-I). Finally, ROC curve analysis of a nomogram that utilizes risk scores and clinical characteristics revealed an improved AUC value, indicating enhanced predictive performance of the model (Fig.\u0026nbsp;\u003cspan refid=\"Fig9\" class=\"InternalRef\"\u003e9\u003c/span\u003eJ-L).\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e \u003c/div\u003e\n\u003ch3\u003eThe construction of nomograms\u003c/h3\u003e\n\u003cp\u003eUsing the outcomes from the multivariable Cox regression analysis, we constructed nomograms for the ten characteristic genes and clinical pathological staging to visually demonstrate the predicted probabilities of survival rates of patients at 1, 3 and 5 years. The nomogram for the LCGs indicated excellent accuracy in predicting patient survival, the mortality rates at 1, 3 and 5 years are 0.382, 0.182 and 0.116 respectively (Fig.\u0026nbsp;\u003cspan refid=\"Fig10\" class=\"InternalRef\"\u003e10\u003c/span\u003eA). In the nomogram combining clinical pathological staging with LLPS risk scores, N stage, age, and risk score were found to have higher clinical net benefits compared to other factors, with mortality rates at 1, 3 and 5 years of 0.431, 0.214 and 0.137 respectively (Fig.\u0026nbsp;\u003cspan refid=\"Fig10\" class=\"InternalRef\"\u003e10\u003c/span\u003eD). Calibration plots for the nomograms showed good consistency (Fig.\u0026nbsp;\u003cspan refid=\"Fig10\" class=\"InternalRef\"\u003e10\u003c/span\u003eB,E). The models were further evaluated using ROC curves, which demonstrated excellent predictive performance based on AUC values, with the nomogram incorporating risk scores and clinical features showing superior predictive accuracy (Fig.\u0026nbsp;\u003cspan refid=\"Fig10\" class=\"InternalRef\"\u003e10\u003c/span\u003eC,F).\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cdiv id=\"Sec30\" class=\"Section2\"\u003e \u003ch2\u003eImmunological analysis derived from the LLPS characteristic prognostic model\u003c/h2\u003e \u003cp\u003eSpearman correlation analysis revealed significant associations between the ten LCGs and immune infiltration (Fig.\u0026nbsp;\u003cspan refid=\"Fig11\" class=\"InternalRef\"\u003e11\u003c/span\u003eA,B). Using the ssGSEA method, Significant variations were observed in the levels of immune cell infiltration, chemokine receptor, responses to type I and type II interferons, antigen-presenting cell co-stimulation, and parainflammation between different risk groups (Fig.\u0026nbsp;\u003cspan refid=\"Fig11\" class=\"InternalRef\"\u003e11\u003c/span\u003eC-E). Additionally, the ESTIMATE and XCELL algorithms were utilized for examining the tumor stromal score and immune score. The sum of these scores provided the ESTIMATE score and TME score, from which tumor purity was inferred. Our observation revealed that the high-risk group had a notably reduced tumor purity score comparing with the low-risk group, indicating lower tumor purity and a higher level of malignancy. (Fig.\u0026nbsp;\u003cspan refid=\"Fig11\" class=\"InternalRef\"\u003e11\u003c/span\u003eF-H). Further analysis of EMT and angiogenesis scores revealed that the high risk group obviously exceeded the low risk group, indicating enhanced migratory capabilities and angiogenic potential (Fig.\u0026nbsp;\u003cspan refid=\"Fig11\" class=\"InternalRef\"\u003e11\u003c/span\u003eL,M). TIDE analysis displayed that the high risk group contained a much higher TIDE score and a significantly lower responsiveness to immunotherapy compared to the low risk group, pointing out an increased probability of immune system evasion and reduced sensitivity to immunotherapy. Additionally, our analysis disclosed that groups with lower immune responsiveness had notably higher risk scores, suggesting a negative association between risk score and immune responsiveness (Fig.\u0026nbsp;\u003cspan refid=\"Fig11\" class=\"InternalRef\"\u003e11\u003c/span\u003eI-K).\u003c/p\u003e \u003cp\u003eTo further clarify the model's accuracy, we went on an exploration on correlations between the risk score and various indices using pearson linear correlation analysis. We observed a positive correlation between the risk score and stromal, immune, ESTIMATE, TME, TIDE, angiogenesis, and EMT scores, and negatively correlated with tumor purity, pointing out that a higher risk score is linked to increased tumor stromal scores, stronger immune evasion capabilities, poorer immune responsiveness, stronger tumor proliferation and migration abilities, and lower tumor purity, indicating a higher degree of malignancy (Fig.\u0026nbsp;\u003cspan refid=\"Fig12\" class=\"InternalRef\"\u003e12\u003c/span\u003e).\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec31\" class=\"Section2\"\u003e \u003ch2\u003eChemotherapy drug sensitivity analysis\u003c/h2\u003e \u003cp\u003eAnalysis of drug sensitivity employing the GDSC database as a training set revealed significant differences in sensitivity to 132 chemotherapy drugs between the high and low-risk groups (IC50\u0026thinsp;\u0026lt;\u0026thinsp;1), with the low-risk group benefiting more from drugs such as Carmustine, Sinularin, Wnt-C59, BMS-345541, and Gefitinib, while the high-risk group benefited more from BMS-754807, AZD8055, AZD2014, AZD8186, and Dasatinib (Fig.\u0026nbsp;\u003cspan refid=\"Fig13\" class=\"InternalRef\"\u003e13\u003c/span\u003e).\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cdiv id=\"Sec32\" class=\"Section3\"\u003e \u003ch2\u003eLCGs mutation analysis\u003c/h2\u003e \u003cp\u003eThe TCGA-STAD cohort provided detailed sample mutation information, allowing us to provide additional clarification on the clinical significance of LLPS prognostic model characteristic genes by analyzing gene mutations. It was discovered that the mutation rate of ten LCGs was elevated in the high risk group (Fig.\u0026nbsp;\u003cspan refid=\"Fig14\" class=\"InternalRef\"\u003e14\u003c/span\u003eA,B). Analysis of gene mutation burden showed that PAK2, RNF2, HSPA1A, and CAPRIN1 showed a greater prevalence of gain-of-function mutations in GC, while HOMER3, MED19, DACT1, TBP, ARID1A, and YBX1 had a higher frequency of loss-of-function mutations (Fig.\u0026nbsp;\u003cspan refid=\"Fig14\" class=\"InternalRef\"\u003e14\u003c/span\u003eC,D). Kaplan-Meier curves revealed that TMB similarly affected the survival periods of GC patients (Fig.\u0026nbsp;\u003cspan refid=\"Fig14\" class=\"InternalRef\"\u003e14\u003c/span\u003eE).\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e \u003c/div\u003e"},{"header":"Discussion","content":"\u003cp\u003eGC, originating from the epithelial cells of the gastric mucosa, consistently ranks among the leading causes of cancer incidence and mortality worldwide. Although external factors such as diet and lifestyle are significant contributors to GC, many cases and deaths are attributed to changes in the local tissue microenvironment [\u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e33\u003c/span\u003e, \u003cspan citationid=\"CR34\" class=\"CitationRef\"\u003e34\u003c/span\u003e]. Early-stage GC patients generally have a favorable prognosis following treatment, whereas those diagnosed at advanced stages often experience poor therapeutic outcomes and prognosis [\u003cspan citationid=\"CR35\" class=\"CitationRef\"\u003e35\u003c/span\u003e]. Consequently, developing new methods to guide prognosis and treatment for GC patients is crucial. Despite recent advances in understanding the mechanisms of GC progression and new treatment strategies, such as ferroptosis, cuproptosis, necroptosis, and disulfidptosis, many unknown mechanisms within the TME continue to promote tumor development [\u003cspan additionalcitationids=\"CR37 CR38\" citationid=\"CR36\" class=\"CitationRef\"\u003e36\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR39\" class=\"CitationRef\"\u003e39\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eLLPS was initially regarded as a purely physical phenomenon of interest primarily in the field of physical chemistry [\u003cspan citationid=\"CR40\" class=\"CitationRef\"\u003e40\u003c/span\u003e]. However, recent studies have demonstrated that LLPS is also a fundamental mechanism of biochemical organization within cells, where various biomolecules such as proteins and nucleic acids can form membrane-less organelles through LLPS. Dysregulation of LLPS can lead to pathological aggregation of proteins and cellular dysfunction, forming the basis for many diseases [\u003cspan citationid=\"CR41\" class=\"CitationRef\"\u003e41\u003c/span\u003e]. In non-tumor diseases such as retroviral infections and coronavirus infections, viral nucleocapsid proteins can form condensates through LLPS, enhancing viral replication or evading host immune surveillance [\u003cspan citationid=\"CR42\" class=\"CitationRef\"\u003e42\u003c/span\u003e, \u003cspan citationid=\"CR43\" class=\"CitationRef\"\u003e43\u003c/span\u003e]. In cancer, proteins encoded by certain genes undergo LLPS to form condensates, leading to chromosomal inactivation, DNA damage, and autophagy, thereby promoting tumor progression [\u003cspan citationid=\"CR44\" class=\"CitationRef\"\u003e44\u003c/span\u003e]. Targeting these condensates to correct abnormal LLPS states has emerged as a new approach in precision oncology. However, research on LLPS in GC remains scarce.\u003c/p\u003e \u003cp\u003eThis study focuses on LLPS, utilizing the TCGA and GEO databases along with various machine learning algorithms and multi-omics analyses to comprehensively assess the role of LLPS in the TME, immune infiltration, immunotherapy, drug sensitivity, and prognosis of GC patients, thereby developing a novel LLPS prognostic model to guide patient prognosis and treatment.\u003c/p\u003e \u003cp\u003eInitially, single-cell analysis was employed to preliminarily analyze the LLPS gene set in GC cells, identifying that cells with high LLPS gene set scores were predominantly T cells and B cells. This suggests potential dysregulation of LLPS in these cells, which could negatively impact the immune response, enhancing the tumor's ability to evade immune surveillance and clearance [\u003cspan additionalcitationids=\"CR46 CR47\" citationid=\"CR45\" class=\"CitationRef\"\u003e45\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR48\" class=\"CitationRef\"\u003e48\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eSubsequent KEGG-GO analysis through three enrichment methods in different clustering groups revealed activation of several inflammatory pathways such as the IL-17 pathway, NF-kappa pathway, and TNF pathway, which play crucial roles in tumor growth regulation [\u003cspan additionalcitationids=\"CR50\" citationid=\"CR49\" class=\"CitationRef\"\u003e49\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR51\" class=\"CitationRef\"\u003e51\u003c/span\u003e]. This further clarifies the diverse regulatory mechanisms of LLPS in the GC immune microenvironment.\u003c/p\u003e \u003cp\u003eAfter establishing the potential regulatory roles of LLPS in GC through single-cell omics, the study continued with transcriptomic analysis combined with various machine learning algorithms. The integration of machine learning with medical data has become a popular research method, effectively predicting diseases and assessing risks, guiding personalized medicine, and drug development. The larger the sample size included in machine learning, the more accurate the algorithm's predictions [\u003cspan citationid=\"CR52\" class=\"CitationRef\"\u003e52\u003c/span\u003e, \u003cspan citationid=\"CR53\" class=\"CitationRef\"\u003e53\u003c/span\u003e]. Therefore, by merging the TCGA-STAD and GSE84437 cohorts, we significantly increased the sample size, enhancing the accuracy of machine learning predictions. In the transcriptomic study, the LLPS gene set was initially analyzed using consensus clustering, clearly dividing 797 samples into two clusters. Analysis of DEGs between the subgroups revealed significant differences in the expression of multiple immune checkpoints (CTLA4, PDL1, LAG3 and etc), which are crucial for cancer therapy. Inhibiting these immune checkpoints can restore the immune system's ability to recognize and eliminate tumors, thus guiding the use of immunosuppressants [\u003cspan citationid=\"CR54\" class=\"CitationRef\"\u003e54\u003c/span\u003e, \u003cspan citationid=\"CR55\" class=\"CitationRef\"\u003e55\u003c/span\u003e]. Recent studies have also shown that interactions between tumor cells, immune cells, and stromal cells in the TME can influence tumor development, making the TME a new target for cancer therapy [\u003cspan citationid=\"CR56\" class=\"CitationRef\"\u003e56\u003c/span\u003e, \u003cspan citationid=\"CR57\" class=\"CitationRef\"\u003e57\u003c/span\u003e]. Therefore, assessing the TME can effectively guide targeted immunotherapy for GC patients to correct the immune microenvironment. Subsequently, various immune algorithms were used to further study the immune infiltration, immunotherapy responsiveness, immune evasion capabilities, and tumor purity of the two clusters, with significant differences observed between them.\u003c/p\u003e \u003cp\u003eIn summary, this study preliminarily clarified the role of the LLPS gene set in GC. However, due to the large number of genes in the LLPS gene set, it was not feasible to construct an effective clinical prediction model for practical medical use. Therefore, four machine learning regression algorithms were combined to perform dimensionality reduction on the gene set, selecting LCGs that could guide patient prognosis. The XGBoost and RF algorithms were efficient and accurate but had fitting issues, while the Lasso algorithm provided better fitting results. The combination of these four machine learning algorithms effectively integrated their respective advantages and compensated for their shortcomings, allowing the selected characteristic genes to constitute a more precise prognostic model to guide the prognosis. Ultimately, ten LCGs were identified (HOMER3, CAPRIN1, PAK2, HSPA1A, MED19, DACT1, RNF2, ARID1A, YBX1, TBP). YBX1 and CAPRIN1 have been proven in liver cancer research to form protein condensates that cause cellular pathway changes and promote tumor development [\u003cspan citationid=\"CR58\" class=\"CitationRef\"\u003e58\u003c/span\u003e, \u003cspan citationid=\"CR59\" class=\"CitationRef\"\u003e59\u003c/span\u003e], while the other eight genes have not been studied in tumors and may become key entry points for LLPS research in GC. Based on these ten LCGs, a prognostic model was constructed to divide GC samples into high and low-risk groups. The assessments of immune infiltration, immunotherapy responsiveness, immune evasion capabilities, and tumor purity in the two groups were consistent with the cluster assessments, and the prognostic model also performed well in predicting patient survival. The risk score combined with clinical information in the nomogram prediction model showed better predictive performance, and the ROC curve further verified the model's accuracy. Prognostic models constructed in three independent training cohorts all demonstrated good predictive performance. In conclusion, the LLPS prognostic model constructed in this study has high clinical value, effectively stratifying GC patients and guiding their prognosis and immunotherapy.\u003c/p\u003e \u003cp\u003eIn terms of chemotherapy drug use, early-stage GC is often treated with endoscopic surgical resection, while advanced GC is treated with continuous chemotherapy regimens [\u003cspan citationid=\"CR60\" class=\"CitationRef\"\u003e60\u003c/span\u003e, \u003cspan citationid=\"CR61\" class=\"CitationRef\"\u003e61\u003c/span\u003e]. Therefore, the choice of chemotherapy drugs is crucial for GC patients, and the LLPS prognostic model developed in this study can effectively group GC patients, predicting chemotherapy drugs with better sensitivity for different groups, providing valuable guidance for clinical medication.\u003c/p\u003e \u003cp\u003eCNV are also important factors in tumor development, with increased copy numbers of oncogenes or decreased copy numbers of tumor suppressor genes promoting tumor progression [\u003cspan citationid=\"CR62\" class=\"CitationRef\"\u003e62\u003c/span\u003e, \u003cspan citationid=\"CR63\" class=\"CitationRef\"\u003e63\u003c/span\u003e]. In this study, sample mutation information provided by the TCGA database was employed for analysis of the copy number mutations of the ten LCGs and the gene mutation frequency in different risk groups, further clarifying the clinical predictive value of these ten genes. However, whether mutations in the characteristic genes are driving factors for abnormal LLPS has not been studied.\u003c/p\u003e \u003cp\u003eThe limitations of this study are twofold. First, the GC samples included in the study encompass all pathological stages, treatment stages, and different metastatic states. However, the TME of GC patients varies at different stages. Therefore, in subsequent research, we will meticulously divide GC samples, such as by T stage, in situ cancer versus metastatic cancer, and pre- and post-immunosuppressant treatment relapse. By comparing the immune infiltration and survival situations of GC patients in these different categories, we aim to further reveal at which stage LLPS has a more significant impact on the disease. Additionally, KEGG enrichment results showed significant changes in the P53 pathway in the prognostic model groups. Therefore, subsequent studies could also divide GC patients based on P53 mutation status and non-mutation status to investigate whether there is mutual regulation between LLPS and the P53 pathway.\u003c/p\u003e \u003cp\u003eThe second limitation is the limited single-cell data included in the study. Due to existing equipment and funding constraints, personalized single-cell sequencing analyses could not be conducted to better understand the biological roles of LLPS in tumor cells and the TME. Additionally, the lack of complete clinical information in the samples provided by the GEO database prevents better classification. Subsequent studies could simulate the TME in vitro and further clarify the impact of LCGs in the GC tumor microenvironment through single-cell sequencing and cell functional experiments. It also remains to be studied whether the characteristic genes will form protein aggregates after undergoing LLPS, affecting cellular biological functions, and whether targeting these protein condensates encoded by the characteristic genes could become a new method for treating GC through in vivo and in vitro researches.\u003c/p\u003e"},{"header":"Conclusion","content":"\u003cp\u003eThis study developed a novel prognostic model containing ten characteristic genes for LLPS, which demonstrates good predictive capacity for the overall survival of GC patients. Combining this model with patient clinical pathological features to construct a nomogram can better assist in clinical decision-making. Based on this model, effective risk stratification of patients, prediction of patient immune responses, and guidance on the use of chemotherapy drugs can be achieved, thus also having high clinical value in guiding the precision treatment of GC patients.\u003c/p\u003e"},{"header":"Abbreviations","content":"\u003cdiv class=\"DefinitionList\"\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eGC\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003egastric cancer\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eLLPS\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eliquid-liquid phase separation\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eLCGs\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eLLPS characteristic genes\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eTME\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003etumor immune microenvironment\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eEMT\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eepithelial mesenchymal transformation\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eTIDE\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eTumor Immune Dysfunction and Exclusion\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eAI\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eartificial intelligence\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eRF\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eRandom Forest\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eTMB\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003etumor mutational burden\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003ePCA\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003ePrincipal Component Analysis\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eDEGs\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003edifferential expression genes\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eGSEA\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eGene Set Enrichment Analysis\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003essGSEA\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003esingle-sample Gene Set Enrichment Analysis\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eROC\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003ereceiver operating characteristic\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eAUC\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003earea under the curve\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eGDSC\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eGenomics of Drug Sensitivity in Cancer\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eCNV\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003ecopy number variation\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eGES-1\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003egastric mucosal epithelial cell\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003c/div\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003cstrong\u003eEthics approval and consent to participate\u0026nbsp;\u003c/strong\u003eNot applicable.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eConsent for publication\u0026nbsp;\u003c/strong\u003eNot applicable.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAvailability of data and materials\u0026nbsp;\u003c/strong\u003eAll information of GC patients used during the research was acquired from the TCGA and GEO databases.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eCompeting Interests\u003c/strong\u003e Not Applicable.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFunding\u003c/strong\u003e The funding for this project was supported by Social Development\u0026nbsp;Project of Zhenjiang\u0026nbsp;(Grant No. SH2022101), Jiangsu Funding Program for Excellent Postdoctoral Talent (No.\u0026nbsp;2023ZB180) and Postdoctoral Fellowship Program of CPSF (No. GZC20230992).\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAuthor Contributions\u0026nbsp;\u003c/strong\u003eRM and YL were in charge of in the main information analysis and experiments. RC, ZS, WZ, RL, RS and XW were responsible for the data acquisition and classification. JW and SS took charge of technical guidance and the modification of the paper.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAcknowledgements\u0026nbsp;\u003c/strong\u003eWe are grateful for the support provided by the TGCA and GEO databases, as well as Jiangsu University.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eConflicts of Interest\u0026nbsp;\u003c/strong\u003eNo conflict of interest.\u0026nbsp;\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\u003cli\u003e\u003cspan\u003eBray F, Laversanne M, Sung HYA, Ferlay J, Siegel RL, Soerjomataram I, Jemal A. Global cancer statistics 2022: GLOBOCAN estimates of incidence and mortality worldwide for 36 cancers in 185 countries. CA-Cancer J Clin 2024:35.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eXia CF, Dong XS, Li H, Cao MM, Sun DAQ, He SY, Yang F, Yan XX, Zhang SL, Li N, Chen WQ. Cancer statistics in China and United States, 2022: profiles, trends, and determinants. Chin Med J. 2022;135(5):584\u0026ndash;90.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLi S, Yu WB, Xie F, Luo HT, Liu ZM, Lv WW, Shi DB, Yu DX, Gao P, Chen C, et al. Neoadjuvant therapy with immune checkpoint blockade, antiangiogenesis, and chemotherapy for locally advanced gastric cancer. Nat Commun. 2023;14(1):16.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eChia NY, Tan P. Molecular classification of gastric cancer. Ann Oncol. 2016;27(5):763\u0026ndash;9.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eYang J, Liu Z, Zeng B, Hu G, Gan R. Epstein-Barr virus-associated gastric cancer: A distinct subtype. Cancer Lett. 2020;495:191\u0026ndash;9.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eAmieva M, Peek RM Jr.. Pathobiology of Helicobacter pylori-Induced Gastric Cancer. Gastroenterology. 2016;150(1):64\u0026ndash;78.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLin Y, Jing X, Chen Z, Pan X, Xu D, Yu X, Zhong F, Zhao L, Yang C, Wang B, et al. Histone deacetylase-mediated tumor microenvironment characteristics an d synergistic immunotherapy in gastric cancer. Theranostics. 2023;13(13):4574\u0026ndash;600.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKim R, An M, Lee H, Mehta A, Heo YJ, Kim K-M, Lee S-Y, Moon J, Kim ST, Min B-H, et al. Early Tumor-Immune Microenvironmental Remodeling and Response to First -Line Fluoropyrimidine and Platinum Chemotherapy in Advanced Gastric C ancer. Cancer Discov. 2022;12(4):984\u0026ndash;1001.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMak TK, Li X, Huang H, Wu K, Huang Z, He Y, Zhang C. The cancer-associated fibroblast-related signature predicts prognosis and indicates immune microenvironment infiltration in gastric cancer. Front Immunol. 2022;13:951214.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLi Y, Hu X, Lin R, Zhou G, Zhao L, Zhao D, Zhang Y, Li W, Zhang Y, Ma P, et al. Single-cell landscape reveals active cell subtypes and their interacti on in the tumor microenvironment of gastric cancer. Theranostics. 2022;12(8):3818\u0026ndash;33.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eChen HY, Sun Q, Zhang CA, She JJ, Cao S, Cao M, Zhang NA, Adiila AV, Zhong JJ, Yao CY, et al. Identification and Validation of CYBB, CD86, and C3AR1 as the Key Genes Related to Macrophage Infiltration of Gastric Cancer. Front Mol Biosci. 2021;8:15.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMehta S, Zhang J. Liquid-liquid phase separation drives cellular function and dysfunctio n in cancer. Nat Rev Cancer. 2022;22(4):239\u0026ndash;52.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLiu Z, Qin Z, Liu Y, Xia X, He L, Chen N, Hu X, Peng X. Liquid\u0026ndash;liquid phase separation: roles and implications in future cance r treatment. Int J Biol Sci. 2023;19(13):4139\u0026ndash;56.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLi R-H, Tian T, Ge Q-W, He X-Y, Shi C-Y, Li J-H, Zhang Z, Liu F-Z, Sang L-J, Yang Z-Z, et al. A phosphatidic acid-binding lncRNA SNHG9 facilitates LATS1 liquid-liqu id phase separation to promote oncogenic YAP signaling. Cell Res. 2021;31(10):1088\u0026ndash;105.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eAhn JH, Davis ES, Daugird TA, Zhao S, Quiroga IY, Uryu H, Li J, Storey AJ, Tsai Y-H, Keeley DP, et al. Phase separation drives aberrant chromatin looping and cancer developm ent. Nature. 2021;595(7868):591\u0026ndash;5.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWei M, Huang X, Liao L, Tian Y, Zheng X. SENP1 Decreases RNF168 Phase Separation to Promote DNA Damage Repair a nd Drug Resistance in Colon Cancer. Cancer Res. 2023;83(17):2908\u0026ndash;23.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eXie J, He H, Kong W, Li Z, Gao Z, Xie D, Sun L, Fan X, Jiang X, Zheng Q, et al. Targeting androgen receptor phase separation to overcome antiandrogen resistance. Nat Chem Biol. 2022;18(12):1341\u0026ndash;50.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eNing L, Zhao G, Xie C, Lan H, Chen J, Tan H, Wei C, Zhou Z. Development and Validation of a Liquid-Liquid Phase Separation-Related Gene Signature as Prognostic Biomarker for Low-Grade Gliomas. \u003cem\u003eDis Markers\u003c/em\u003e 2022, 2022:1487165.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eXie J, Chen L, Wu D, Liu S, Pei S, Tang Q, Wang Y, Ou M, Zhu Z, Ruan S, et al. Significance of liquid-liquid phase separation (LLPS)-related genes in breast cancer: a multi-omics analysis. Aging. 2023;15(12):5592\u0026ndash;610.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHou C, Wang X, Xie H, Chen T, Zhu P, Xu X, You K, Li T. PhaSepDB in 2022: annotating phase separation-related proteins with dr oplet states, co-phase separation partners and other experimental info rmation. Nucleic Acids Res. 2023;51(D1):D460\u0026ndash;5.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWilkerson MD, Hayes DN. ConsensusClusterPlus: a class discovery tool with confidence assessmen ts and item tracking. Bioinformatics. 2010;26(12):1572\u0026ndash;3.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLiberzon A, Birger C, Thorvaldsd\u0026oacute;ttir H, Ghandi M, Mesirov JP, Tamayo P. The Molecular Signatures Database (MSigDB) hallmark gene set collectio n. Cell Syst. 2015;1(6):417\u0026ndash;25.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSubramanian A, Tamayo P, Mootha VK, Mukherjee S, Ebert BL, Gillette MA, Paulovich A, Pomeroy SL, Golub TR, Lander ES, Mesirov JP. Gene set enrichment analysis: a knowledge-based approach for interpret ing genome-wide expression profiles. Proc Natl Acad Sci U S A. 2005;102(43):15545\u0026ndash;50.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eH\u0026auml;nzelmann S, Castelo R, Guinney J. GSVA: gene set variation analysis for microarray and RNA-seq data. BMC Bioinformatics. 2013;14:7.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZheng K, Hai Y, Chen H, Zhang Y, Hu X, Ni K. Tumor immune dysfunction and exclusion subtypes in bladder cancer and pan-cancer: a novel molecular subtyping strategy and immunotherapeutic prediction model. J Transl Med. 2024;22(1):365.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eChen Y, Meng Z, Zhang L, Liu F. CD2 Is a Novel Immune-Related Prognostic Biomarker of Invasive Breast Carcinoma That Modulates the Tumor Microenvironment. Front Immunol. 2021;12:664845.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eYoshihara K, Shahmoradgoli M, Mart\u0026iacute;nez E, Vegesna R, Kim H, Torres-Garcia W, Trevi\u0026ntilde;o V, Shen H, Laird PW, Levine DA, et al. Inferring tumour purity and stromal and immune cell admixture from exp ression data. Nat Commun. 2013;4:2612.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eAran D, Sirota M, Butte AJ. Systematic pan-cancer analysis of tumour purity. Nat Commun. 2015;6:8971.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMasiero M, Sim\u0026otilde;es FC, Han HD, Snell C, Peterkin T, Bridges E, Mangala LS, Wu SY-Y, Pradeep S, Li D, et al. A core human primary tumor angiogenesis signature identifies the endot helial orphan receptor ELTD1 as a key regulator of angiogenesis. Cancer Cell. 2013;24(2):229\u0026ndash;41.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLamouille S, Xu J, Derynck R. Molecular mechanisms of epithelial-mesenchymal transition. Nat Rev Mol Cell Biol. 2014;15(3):178\u0026ndash;96.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eYang W, Soares J, Greninger P, Edelman EJ, Lightfoot H, Forbes S, Bindal N, Beare D, Smith JA, Thompson IR, et al. Genomics of Drug Sensitivity in Cancer (GDSC): a resource for therapeu tic biomarker discovery in cancer cells. Nucleic Acids Res. 2013;41(Database issue):D955\u0026ndash;961.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eChandrashekar DS, Karthikeyan SK, Korla PK, Patel H, Shovon AR, Athar M, Netto GJ, Qin ZS, Kumar S, Manne U, et al. UALCAN: An update to the integrated cancer data analysis platform. Neoplasia. 2022;25:18\u0026ndash;27.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eChen Y, Jia K, Sun Y, Zhang C, Li Y, Zhang L, Chen Z, Zhang J, Hu Y, Yuan J, et al. Predicting response to immunotherapy in gastric cancer via multi-dimen sional analyses of the tumour immune microenvironment. Nat Commun. 2022;13(1):4851.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eJin G, Lv J, Yang M, Wang M, Zhu M, Wang T, Yan C, Yu C, Ding Y, Li G, et al. Genetic risk, incident gastric cancer, and healthy lifestyle: a meta-a nalysis of genome-wide association studies and prospective cohort stud y. Lancet Oncol. 2020;21(10):1378\u0026ndash;86.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eYang W-J, Zhao H-P, Yu Y, Wang J-H, Guo L, Liu J-Y, Pu J, Lv J. Updates on global epidemiology, risk and prognostic factors of gastric cancer. World J Gastroenterol. 2023;29(16):2452\u0026ndash;68.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZhang H, Deng T, Liu R, Ning T, Yang H, Liu D, Zhang Q, Lin D, Ge S, Bai M, et al. CAF secreted miR-522 suppresses ferroptosis and promotes acquired chem o-resistance in gastric cancer. Mol Cancer. 2020;19(1):43.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSun L, Zhang Y, Yang B, Sun S, Zhang P, Luo Z, Feng T, Cui Z, Zhu T, Li Y, et al. Lactylation of METTL16 promotes cuproptosis via m\u0026thinsp;\u0026lt;\u0026thinsp;sup\u0026thinsp;\u0026gt;\u0026thinsp;6\u0026thinsp;A-modific ation on FDX1 mRNA in gastric cancer. Nat Commun. 2023;14(1):6523.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWu N, Liu F, Huang Y, Su X, Zhang Y, Yu L, Liu B. Necroptosis Related Genes Predict Prognosis and Therapeutic Potential in Gastric Cancer. Biomolecules. 2023;13(1):101.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLi J, Yu T, Sun J, Ma M, Zheng Z, He Y, Kang W, Ye X. Integrated analysis of disulfidptosis-related immune genes signature t o boost the efficacy of prognostic prediction in gastric cancer. Cancer Cell Int. 2024;24(1):112.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eChen D, Lyu M, Kou X, Li J, Yang Z, Gao L, Li Y, Fan L-M, Shi H, Zhong S. Integration of light and temperature sensing by liquid-liquid phase se paration of phytochrome B. Mol Cell. 2022;82(16):3015\u0026ndash;e30293016.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eAlberti S, Gladfelter A, Mittag T. Considerations and Challenges in Studying Liquid-Liquid Phase Separati on and Biomolecular Condensates. Cell. 2019;176(3):419\u0026ndash;34.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eChau B-A, Chen V, Cochrane AW, Parent LJ, Mouland AJ. Liquid-liquid phase separation of nucleocapsid proteins during SARS-Co V-2 and HIV-1 replication. Cell Rep. 2023;42(1):111968.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWei W, Bai L, Yan B, Meng W, Wang H, Zhai J, Si F, Zheng C. When liquid-liquid phase separation meets viral infections. Front Immunol. 2022;13:985622.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLiu Q, Li J, Zhang W, Xiao C, Zhang S, Nian C, Li J, Su D, Chen L, Zhao Q, et al. Glycogen accumulation and phase separation drives liver tumor initiati on. Cell. 2021;184(22):5559\u0026ndash;e55765519.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePark J, Hsueh P-C, Li Z, Ho P-C. Microenvironment-driven metabolic adaptations guiding CD8\u0026thinsp;\u0026lt;\u0026thinsp;sup\u0026gt;+\u0026thinsp;T cell anti-tumor immunity. Immunity. 2023;56(1):32\u0026ndash;42.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBaldominos P, Barbera-Mourelle A, Barreiro O, Huang Y, Wight A, Cho J-W, Zhao X, Estivill G, Adam I, Sanchez X, et al. Quiescent cancer cells resist T cell attack by forming an immunosuppre ssive niche. Cell. 2022;185(10):1694\u0026ndash;e17081619.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDowns-Canner SM, Meier J, Vincent BG, Serody JS. B Cell Function in the Tumor Microenvironment. Annu Rev Immunol. 2022;40:169\u0026ndash;93.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLiu Y, Zhou X, Wang X. Targeting the tumor microenvironment in B-cell lymphoma: challenges an d opportunities. J Hematol Oncol. 2021;14(1):125.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLi X, Bechara R, Zhao J, McGeachy MJ, Gaffen SL. IL-17 receptor-based signaling and implications for disease. Nat Immunol. 2019;20(12):1594\u0026ndash;602.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eYu H, Lin L, Zhang Z, Zhang H, Hu H. Targeting NF-κB pathway for the therapy of diseases: mechanism and cli nical study. Signal Transduct Target Ther. 2020;5(1):209.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eYuan S, Carter P, Bruzelius M, Vithayathil M, Kar S, Mason AM, Lin A, Burgess S, Larsson SC. Effects of tumour necrosis factor on cardiovascular disease and cancer: A two-sample Mendelian randomization study. EBioMedicine. 2020;59:102956.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGreener JG, Kandathil SM, Moffat L, Jones DT. A guide to machine learning for biologists. Nat Rev Mol Cell Biol. 2022;23(1):40\u0026ndash;55.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHaug CJ, Drazen JM. Artificial Intelligence and Machine Learning in Clinical Medicine, 202 3. N Engl J Med. 2023;388(13):1201\u0026ndash;8.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGalluzzi L, Humeau J, Buqu\u0026eacute; A, Zitvogel L, Kroemer G. Immunostimulation with chemotherapy in the era of immune checkpoint in hibitors. Nat Rev Clin Oncol. 2020;17(12):725\u0026ndash;41.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHeinhuis KM, Ros W, Kok M, Steeghs N, Beijnen JH, Schellens JHM. Enhancing antitumor response by combining immune checkpoint inhibitors with chemotherapy in solid tumors. Ann Oncol. 2019;30(2):219\u0026ndash;35.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eXiao Y, Yu D. Tumor microenvironment as a therapeutic target in cancer. Pharmacol Ther. 2021;221:107753.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eJin M-Z, Jin W-L. The updated landscape of tumor microenvironment and drug repurposing. Signal Transduct Target Ther. 2020;5(1):166.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eChen S, Cao X, Zhang J, Wu W, Zhang B, Zhao F. circVAMP3 Drives CAPRIN1 Phase Separation and Inhibits Hepatocellular Carcinoma by Suppressing c-Myc Translation. Adv Sci (Weinh). 2022;9(8):e2103817.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLiu B, Shen H, He J, Jin B, Tian Y, Li W, Hou L, Zhao W, Nan J, Zhao J, et al. Cytoskeleton remodeling mediated by circRNA-YBX1 phase separation supp resses the metastasis of liver cancer. Proc Natl Acad Sci U S A. 2023;120(30):e2220296120.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSmyth EC, Nilsson M, Grabsch HI, van Grieken NC, Lordick F. Gastric cancer. Lancet. 2020;396(10251):635\u0026ndash;48.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eJoshi SS, Badgwell BD. Current treatment and recent progress in gastric cancer. CA Cancer J Clin. 2021;71(3):264\u0026ndash;79.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eRamtohul T, Djerroudi L, Lissavalid E, Nhy C, Redon L, Ikni L, Djelouah M, Journo G, Menet E, Cabel L, et al. Multiparametric MRI and Radiomics for the Prediction of HER2-Zero, -Lo w, and -Positive Breast Cancers. Radiology. 2023;308(2):e222646.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eTashakori M, Kadia T, Loghavi S, Daver N, Kanagal-Shamanna R, Pierce S, Sui D, Wei P, Khodakarami F, Tang Z, et al. TP53 copy number and protein expression inform mutation status across risk categories in acute myeloid leukemia. Blood. 2022;140(1):58\u0026ndash;72.\u003c/span\u003e\u003c/li\u003e\u003c/ol\u003e"},{"header":"Tables","content":"\u003cp\u003eTable 1 is available in the Supplementary Files section.\u003c/p\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":true,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":false,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true},"keywords":"Gastric cancer, liquid-liquid phase separation, LLPS characteristic genes, LLPS characteristic prognostic model, prognosis, tumor immune microenvironment, chemotherapy drug sensitivity, immunotherapy, machine learning","lastPublishedDoi":"10.21203/rs.3.rs-4546744/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-4546744/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003ch2\u003eBackground\u003c/h2\u003e \u003cp\u003eLiquid-liquid phase separation (LLPS) refers to a phenomenon in which unique liquid condensates are formed due to weak interactions among biomolecules, including proteins and nucleic acids. In cellular environments, abnormal LLPS can induce aggregation of membrane-less organelles, disrupt intracellular signaling, alter chromatin structures, and cause aberrant gene expression. The significance of LLPS in gastric cancer (GC) cells is still poorly understood. This study aims to integrate multiple omics analysis and multiple machine learning algorithms to identify LLPS characteristic genes (LCGs) which can be used to develop a LLPS characteristic prognostic model.\u003c/p\u003e\u003ch2\u003eMethods\u003c/h2\u003e \u003cp\u003eTranscriptomic and single-cell data for GC patients were retrieved from the GEO and TCGA databases. The LLPS gene set was extracted from the PhaSepDB database. Initial cellular localization analysis of LLPS gene set-expressing cells was performed using single-cell data from GSE167297. Subsequently, we analyzed 797 GC samples from the TCGA-STAD and GSE84437 merged cohort using the ConsensusClusterPlus method, then we subdivided the merged cohort into two clusters based on the expression of the LLPS gene set for further prognostic and immune analyses. Characteristic genes of the LLPS gene set were identified by the best combination of four machine learning algorithms correlating with patient survival status and time, which were then validated across three independent GC patient cohorts. The differential expression of LCGs in the prognostic model was validated using the HPA and UALCAN databases, as well as western blotting. Additionally, a nomogram was developed to improve the effectiveness of the model in clinical application. Furthermore, differences in the tumor immune microenvironment (TME), immunotherapy response, and drug sensitivity between different risk groups were studied through a variety of immune algorithms. Mutational analysis of ten LLPS gene set genes was conducted based on mutation data from the TCGA-STAD cohort.\u003c/p\u003e\u003ch2\u003eResults\u003c/h2\u003e \u003cp\u003eA LLPS characteristic prognostic model based on a combination of four machine learning algorithms was established, identifying ten LCGs with high predictive value for the prognosis, TME, immunotherapy responses, and chemotherapy drug sensitivity of GC patients. Additionally, a specific nomogram was developed, incorporating clinical features to enhance the effectiveness of the LLPS clinical score, with AUC values of 0.722, 0.715, 0.707 at 1, 3, and 5 years, respectively. The LLPS prognostic model demonstrated good predictive value for survival status across different age groups, T stages, and N stages of GC patients. Risk scores calculated from LCGs showed linear correlations with stromal scores, immune scores, TME scores, Tumor Immune Dysfunction and Exclusion (TIDE) scores, epithelial-mesenchymal transition (EMT) scores, angiogenesis scores, and tumor purity scores. Furthermore, mutations in LCGs were found to impact the survival of GC patients.\u003c/p\u003e\u003ch2\u003eConclusions\u003c/h2\u003e \u003cp\u003eThe LLPS characteristic prognostic model provides a new perspective for assessing the prognosis of GC patients, their responses to immunotherapy, TME and chemotherapy drug usage.\u003c/p\u003e","manuscriptTitle":"A Novel Liquid–Liquid Phase Separation Characteristic Model Associated with Prognosis and Immune Landscape of Gastric Cancer Patients","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2024-07-15 19:20:22","doi":"10.21203/rs.3.rs-4546744/v1","editorialEvents":[{"type":"communityComments","content":0}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"53bd7e87-0b7f-47ff-a02e-cc611e5865fd","owner":[],"postedDate":"July 15th, 2024","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[],"tags":[],"updatedAt":"2024-09-17T07:02:31+00:00","versionOfRecord":[],"versionCreatedAt":"2024-07-15 19:20:22","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-4546744","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-4546744","identity":"rs-4546744","version":["v1"]},"buildId":"qtupq5eGEP_6zYnWcrvyt","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}