Gene clusters-based pathway enrichment analysis identifies four pan-cancer subtypes with distinct molecular and clinical features

Gene clusters-based pathway enrichment analysis identifies four pan-cancer subtypes with distinct molecular and clinical features | 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 Gene clusters-based pathway enrichment analysis identifies four pan-cancer subtypes with distinct molecular and clinical features Mengli Xu, Hongjing Ai, Danni Wang, Xiaosheng Wang This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-5114169/v1 This work is licensed under a CC BY 4.0 License Status: Published Journal Publication published 28 Nov, 2024 Read the published version in Functional & Integrative Genomics → Version 1 posted 9 You are reading this latest preprint version Abstract Background Pathways-based clustering methods have been proposed to explore tumor heterogeneity. However, such methods are currently disadvantageous in that specific pathways need to be explicitly claimed. Methods We developed the PathClustNet algorithm, a pathway-based clustering method to identify cancer subtypes. This method first detects gene clusters and identifies overrepresented pathways associated with these gene clusters. Based on the enrichment scores of the pathways, reveals cancer subtypes by clustering analysis. We applied the method to TCGA pan-cancer to identify cancer subtypes. Results We identified four pan-cancer subtypes, termed C1, C2, C3 and C4. C1 exhibited high metabolic activity, favourable survival and the lowest TP53 mutation rate. C2 had high immune, developmental, and stromal pathway activities, the lowest tumor purity and intratumor heterogeneity. C3 overexpressed cell cycle and DNA repair pathways were most genomically instable and had the highest TP53 mutation rate. C4 overrepresented neuronal pathways, with the lowest response rate to chemotherapy, and the highest tumor purity and genomic stability. Furthermore, ages showed positive correlations with most pathways but a negative correlation with neuronal pathways. Smoking, Viral infections and alcohol use variously affect neuron, cell cycle, immune, stromal, developmental, and metabolic pathway activities. Conclusion The novel PathClustNet clustering algorithm unveils a new classification of pan-cancer based on metabolic, immune, stromal, developmental, cell cycle, and neuronal pathways. These subtypes display different molecular and clinical features to warrant the investigation of precision oncology. Pan-cancer Subtyping Gene clusters Pathway enrichment analysis PathClustNet Algorithm Tumor heterogeneity Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Figure 6 Figure 7 Figure 8 Figure 9 Figure 10 1 Introduction Tumor heterogeneity is one of the typical properties of cancer, reflecting molecular and phenotypic variations between tumors or within a tumor( 1 ). With the emergence of abundant cancer-associated bulk and single-cell transcriptomics data, unraveling transcriptome heterogeneity in cancer is viable( 2 ). Gene expression profiles-based clustering analysis is a commonly used avenue to decipher transcriptome heterogeneity in cancer, based on which myriad investigations into identification of cancer subtypes have been implemented( 3 , 4 ). For example, Thorsson et al. uncovered six immune subtypes of the TCGA pan-cancer based on transcriptome-based clustering analysis( 4 ). Another analysis of the TCGA pan-cancer transcriptomes identified three metabolic subtypes showing different clinical outcomes( 3 ). However, because outliers of individual genes’ expressions may compromise the stability of clustering results, pathways or gene sets-based clustering methods have been proposed( 5 – 7 ). For example, Li et al. identified three gastric cancer subtypes based on the enrichment levels of 15 pathways involved in immune, DNA repair, oncogenic, and stromal signatures( 5 ). He et al. uncovered three immune-specific subtypes of triple-negative breast cancer based on the enrichment levels of 29 immune signatures( 6 ). Lei et al. unveiled four subtypes of the pan-cancer of 33 TCGA cancer types based on the enrichment scores of six stemness-associated gene sets, which are markers of embryonic stem cells and hematopoietic stem cells, respectively( 7 ). Although these pathway-based clustering analyses are successful in identifying cancer subtypes, they are disadvantageous in that specific pathways need to be explicitly claimed in advance. To fill the knowledge gap, here we developed a pathway-based clustering method for unveiling cancer subtypes. Different from previous methods, this method establishes the matrix of expression correlations between gene pairs. Next, the genes are classified into different clusters using a consensus clustering analysis of the gene expression correlations matrix. For each cluster of genes, its associated overrepresented pathways are identified. Lastly, based on the enrichment scores of these pathways, hierarchal clustering analysis is performed to uncover cancer subtypes. We applied the method to TCGA pan-cancer and identified four subtypes and comprehensively characterized the molecular and clinical features of these subtypes. Furthermore, we explored the correlations between pathways’ activities and etiological and clinicopathologic parameters in pan‑cancer. 2 Methods 2.1 Overview of the algorithm The PathClustNet algorithm comprises several steps. First, for a cancer-associated gene expression n × m matrix M , where n and m denote the number of genes and samples, respectively, we calculate the standard deviation of each gene across all samples to assess its variability, we then select the top 5000 genes with the highest standard deviations in descending order and obtain the top 5000 gene expressions correlation matrix M f . Next, we calculate the Pearson correlation between each pair of genes’ expressions and obtain the gene expressions correlation matrix M g . By consensus clustering of M g , we attain gene clusters. For each gene cluster, we identify the pathways significantly associated with it by gene set enrichment analysis. Based on these pathways, we construct samples’ pathway enrichment matrix M p . Finally, we identify cancer subtypes by hierarchical clustering of M p . A schematic overview of the algorithm is displayed in Fig. 1 . 2.2 Data collection We downloaded data of RSEM normalized and batch effects adjusted gene expressions, somatic mutations, copy number alternations (CNA), protein expressions, and clinical information for the TCGA pan-cancer (PANCAN) cohort consisting of 33 cancer types from UCSC Xena ( https://xenabrowser.net/datapages/ ). The 33 cancer types included adrenocortical carcinoma (ACC), bladder urothelial carcinoma (BLCA), breast invasive carcinoma (BRCA), cervical squamous cell carcinoma and endocervical adenocarcinoma (CESC), cholangiocarcinoma (CHOL), colon adenocarcinoma (COAD), lymphoid neoplasm diffuse large B-cell lymphoma (DLBC), esophageal carcinoma (ESCA), glioblastoma multiforme (GBM), head and neck squamous cell carcinoma (HNSC), kidney chromophobe (KICH), kidney renal clear cell carcinoma (KIRC), kidney renal papillary cell carcinoma (KIRP), acute myeloid leukemia (LAML), brain lower grade glioma (LGG), liver hepatocellular carcinoma (LIHC), lung adenocarcinoma (LUAD), lung squamous cell carcinoma (LUSC), mesothelioma (MESO), ovarian serous cystadenocarcinoma (OV), pancreatic adenocarcinoma (PAAD), pheochromocytoma and paraganglioma (PCPG), prostate adenocarcinoma (PRAD), rectum adenocarcinoma (READ), sarcoma (SARC), skin cutaneous melanoma (SKCM), stomach adenocarcinoma (STAD), testicular germ cell tumors (TGCT), thyroid carcinoma (THCA), thymoma (THYM), uterine corpus endometrial carcinoma (UCEC), uterine carcinosarcoma (UCS), and uveal melanoma (UVM). A summary of these datasets is shown in Supplementary Table S1 . 2.3 Clustering analysis We utilized consensus clustering to identify gene clusters with the input of gene expression correlations matrix. Consensus clustering defines specified cluster counts by subsampling from a group of samples, which enhances the stability and reliability of clustering by aggregating results from multiple clustering iterations( 8 ). We performed this analysis using the ConsensusClusterPlus R package( 9 ) with the parameters: clusterAlg = “pam”, distance= “euclidean”, reps = 1000, pItem = 0.8, and pFeature = 1. In addition, we employed hierarchical clustering( 10 ) to identify cancer subtypes with the input of pathway enrichment matrix. This analysis was performed using the hclust R package with the parameters: method = “ward. D2” and members = NULL. Before hierarchical clustering, the data of pathway enrichment scores were transformed by z-score and translated into distance matrices by the dist function with the parameter: method = “Euclidean”. 2.4 Single‑sample gene‑set enrichment analysis We employed the single-sample gene-set enrichment analysis (ssGSEA) method( 11 ) to evaluate the enrichment level of a gene set in a sample; the gene set represents some pathway, biological process or phenotypic feature. The ssGSEA algorithm determines a gene set’s enrichment score in a sample with the input of gene expression profiles. We performed this analysis with the GSVA R package. The gene sets analyzed are presented in Supplementary Table S2. 2.5 Survival analysis We utilized the Kaplan-Meier (K-M)( 12 ) model to compare overall survival (OS), disease-specific survival (DSS), disease-free interval (DFI), and progression-free interval (PFI) among different subgroups of cancer patients. Log-rank tests were employed to assess the significance of survival time differences. The survival analysis was implemented with the survfit function in the survival R package. 2.6 Evaluating CNA, tumor mutation burden (TMB), homologous recombination deficiency (HRD), intratumor heterogeneity (ITH), and tumor immune infiltration levels To evaluate CNA and HRD in tumors, we obtained CNA and HRD scores of TCGA pan-cancer from the publication by Knijnenburg et al.( 13 ). In addition, we used GISTIC2( 14 ) to calculate tumors’ G-scores with the input of “SNP6” files. The G-score reflects the amplitude of the somatic CNA and the frequency of its occurrence across a class of samples( 14 ). We defined a tumor’s TMB as the total count of somatic mutations in the tumor. Besides, we utilized the DEPTH algorithm( 2 ) to quantify ITH. DEPTH evaluates ITH at mRNA level based on the gene expression perturbations’ heterogeneity To assess tumor immune infiltration levels, we calculated immune scores using the ESTIMATE algorithm( 15 ) with the input of gene expression profiles. 2.7 Identification of gene and protein markers for subtypes To identify gene or protein markers for a subtype, we first identified the significantly upregulated genes or proteins in the subtype relative to each of the other subtypes by differential expression analysis. The gene or protein markers for the subtype were defined as the common genes or proteins among the upregulated genes or proteins. The differential expression analysis for genes used the threshold of Student’s t test adjusted P value 2; for proteins, the threshold was Student’s t test adjusted P value 1. 2.8 Pathway enrichment analysis To identify enriched pathways in a specific subtype, we utilized the network tool g:Profiler( 16 ) with the input of gene expression profiles and the marker genes of the subtype, selecting the Biological Process (BP) from the Gene Ontology (GO) database( 17 ), and setting the threshold of adjusted P value < 0.05. We also performed pathway enrichment analysis based on protein expression profiles. This analysis used the enrichGO function in the clusterProfiler R package with the input of protein expression profiles and the marker proteins of the subtype. The cnetplot function was utilized for visualization of the enrichment analysis results( 18 ). The parameters for pathway enrichment analysis were: OrgDb = “org.Hs.eg.db”, ont = “BP”, pvalueCutoff = 0.05, qvalueCutoff = 0.05. 2.9 Statistical analysis In class comparisons, for non-normally distributed data, we utilized Mann–Whitney U tests (for two classes) or Kruskal–Wallis (K–W) tests( 19 ) (for more than two classes); for normally distributed data, we used Student’s t-tests (for two classes) or ANOVA tests( 20 ) (for more than two classes). In the analysis of contingency tables, we utilized Fisher’s exact tests or Chi-square tests. We employed the Pearson or Spearman method( 21 ) to evaluate the correlations between two variables. To adjust for P values in multiple tests, we utilized the Benjamini-Hochberg method( 22 ) to calculate the false discovery rate (FDR). All statistical analyses were conducted in the R programming environment (version 4.2.2). 3 Results 3.1 Clustering analysis identifies four subtypes of pan‑cancer We applied the algorithm to the TCGA pan-cancer transcriptomics data involving 10,323 cancer specimens and 20,531 genes. Consensus clustering of the gene expressions correlation matrix identified six gene clusters, which fell into six categories of pathways: neural, metabolic, stromal, developmental, immune, and cell cycle pathways. From each of the six pathway categories, we randomly selected five pathways. Based on the 30 pathways, we obtained a pathway enrichment matrix of the pan-cancer for which we preformed hierarchical clustering and unveiled four pan-cancer subtypes (Fig. 2 A). We termed the four subtypes C1, C2, C3, and C4, respectively. C1 showed high enrichment of metabolic pathways; C2 overexpressed the stromal, developmental, and immune pathways; C3 overrepresented cell cycle pathways; and C4 displayed upregulation of neuronal pathways. The comparisons of survival prognosis among the four subtypes revealed almost consistent pattern for the four endpoints (10-year OS, DSS, DFI, and PFI): C1 had the best OS, DFI followed the order of C4 > C1 > C2 > C3, and both DSS and PFI had the pattern of C1 > C2 > C3 > C4 (Fig. 2 B). These results underscore significant prognostic differences among these subtypes. To be specific, C1 exhibited the most favorable OS, DSS, and PFI, C4 had the worst DSS and PFI, and C3 displayed the worst DFI. It indicates that overrepresentation of metabolic pathways is a favorable prognostic indicator, while upregulation of neuronal and cell cycle pathways are adverse prognostic factors. Conforming to the survival outcomes, the proportion of early-stage (stage I-II) tumors in the four subtypes followed the pattern: C1 (64.0%) > C3 (50.0%) > C2 (43.3%) > C4 (0.0%) (Chi-square test, P = 4.81×10 –10 ) (Fig. 2 C). The proportion of low-grade tumors (G1-2) also followed the similar trend (Fig. 2 C) ( P = 7.26×10 − 6 ). Furthermore, we observed that the response rate of these subtypes to chemotherapy followed the pattern: C1 (85.7%) > C3 (84.4%) > C2 (81.8%) > C4 (57.8%) ( P = 1.60×10 –54 ) (Fig. 2 C). 3.2 Distribution of individual cancer types across the subtypes Among the four pan-cancer subtypes, C1 harbored the highest proportions of ACC (91.1%), LIHC(93.8%), PRAD(94.8%), THCA(77.4%), CHOL(50%), KICH(75.8%), KIRP(68.8%) and UVM(95.0%), C2 harbored the highest proportions of HNSC(48.1%), KIRC(89.3%), LUAD(70.6%), PAAD(79.9%) and MESO(65.6%), C3 harbored the highest proportions of BLCA(59.1%), CESC(82.3%), COAD(85.5%), DLBC(100.0%), STAD(53.5%), ESCA(74.1%), LAML(100.0%), OV(67.2%), READ(85.7%), SKCM(61.5%), TGCT(86.5%), THYM(73.3%), UCEC(57.1%), and UCS(86.0%), and C4 harbored the highest proportions of GBM(87.4%), LGG(99.8%), and PCPG(98.4%)(Fig. 3 A). Notably, most ACC (91.1%), LIHC (93.8%), PRAD (94.8%), and UVM (95.0%) tumors belonged to C1, indicating the high enrichment of metabolic pathways in these cancer types. C2 was dominated by most KIRC (89.3%) and PAAD (79.9%) tumors, suggesting that the stromal, developmental, and immune pathways are enriched in both cancer types. C3 harbored all DLBC and LAML and most of CESC (82.3%), COAD (85.5%), READ (85.7%), TGCT (86.5%) and UCS (86.0%) tumors, suggesting that cell cycle pathways are overrepresented in these cancer types. Finally, most LGG (99.8%), GBM (87.4%), and PCPG (98.4%) tumors belonged to C4; it is justified as these cancer types are neural system-related and C4 overexpressed neuronal pathways. Notably, about 80% of gastrointestinal cancers, such as COAD, ESCA, STAD, and READ, were grouped into C3 (Fig. 3 B), indicating aberrant activation of the cell cycle pathway in gastrointestinal cancers. Moreover, approximately 70% of kidney cancers, including KIRC and KIRP, were classified into C2 (Fig. 3 B). It highlights that kidney cancers are highly enriched in stromal, developmental, and immune pathways. 3.3 The pan-cancer subtypes have different tumor microenvironment As expected, C2 displayed the highest immune scores and stromal scores, and the lowest tumor purity among the four subtypes (K-W test, P < 0.001) (Fig. 4 ), aligning with this subtype overexpressing immune and stromal pathways. In contrast, C4, characterized by high enrichment of neuronal pathways, had the lowest immune scores while the highest tumor purity. In addition, C3, overrepresenting cell cycle pathways, displayed the lowest stromal scores and the second highest immune scores. Altogether, these results indicate significantly different tumor microenvironment among these subtypes. 3.4 The pan-cancer subtypes have different genomic features Genomic instability is a common property of cancer, embodied by increased TMB, CNA and HRD( 23 ). We observed that C3 had the highest TMB, CNA and HRD scores, whereas C4 displayed the lowest (K-W test, P C3 > C1 > C2 (K-W test, P = 0) (Fig. 5 D). Additionally, G-scores of copy number amplifications and deletions followed the same pattern (Fig. 5 E). DNA repair pathways are crucial in maintaining genomic instability( 24 ). The enrichment scores of four key DNA repair pathways (base excision repair, nucleotide excision repair, Fanconi Anemia, and homologous recombination) exhibited a uniform pattern across the four subtypes: C3 > C2 > C1 > C4 ( P = 0) (Fig. 5 F). In addition, the enrichment scores of non-homologous end joining, translesion synthesis, and damage sensor pathways were also highest in C3, and C4 had the lowest scores of non-homologous end joining and direct repair pathways ( P < 0.001) (Fig. 5 F). Collectively, these results suggested that the DNA damage response was strongest in C3 and weakest in C4. Taken together, these analyses revealed that C3 and C4 had the highest and lowest degree of genomic instability, respectively. We further compared somatic mutation profiles among these subtypes. Figure 5 G showed top 10 genes most frequently mutated in each subtype. Consistently, TP53 had the highest or second highest mutation rate within each subtype, and its mutation rate followed the pattern across the subtypes: C3 (50%) > C2 (39%) > C4 (36%) > C1 (17%). Considering that p53 plays a crucial role in inducing cell cycle arrest( 25 ), it is reasonable that TP53 has the highest mutation rate in C3, the subtype with the highest enrichment of cell cycle pathways. IDH1 showed the highest mutation rate (48%) in C4, the subtype characterized by the highest enrichment of neuronal pathways. Again, the result is justified since IDH1 is frequently mutated in gliomas( 26 ). 3.5 The pan-cancer subtypes have different cancer-associated molecular and phenotypic features We compared cancer-associated molecular and phenotypic features among these subtypes. We observed that the enrichment scores of the stemness and proliferation signatures followed the pattern: C3 > C2 > C1 > C4 ( P = 0) (Fig. 6 A). It indicated that C3 had the strongest stemness and proliferation signatures, consistent with the highest enrichment of cell cycle pathways in this subtype. In contrast, the enrichment scores of the epithelial-mesenchymal transition (EMT) signature( 27 ) followed the pattern: C4 > C2 > C1 > C3 ( P = 0) (Fig. 6 A), suggesting that C4 has the highest EMT activity. Several cancer progression-related signatures, including inflammation, invasion, and metastasis( 28 ), also displayed noticeably distinct enrichment levels across the subtypes: C2 > C3 > C1 > C4 ( P = 0) (Fig. 6 B). This highlights the highest activity of inflammation, invasion, and metastasis in C2, aligning with the high enrichment of immue, developmental, and stromal pathways in this subtype. 3.6 Pathway enrichment analysis reveals distinct functional patterns among the pan-cancer subtypes By differential gene expression analysis, we identified marker genes for each subtype (Supplementary Table S3). Based on the marker genes, we identified biological processes significantly enriched in each subtype using the clusterProfiler R package( 29 ). As expected, metabolism-related biological processes were predominantly enriched in C1, including peroxisome, retinol metabolism, chemical carcinogenesis - DNA adducts( 30 ), drug metabolism - cytochrome P450, metabolism of xenobiotics by cytochrome P450, tyrosine metabolism, complement and coagulation cascades( 31 ), bile secretion, valine-leucine and isoleucine degradation, and steroid hormone biosynthesis (Fig. 7 A). C2 was primarily enriched in biological processes related to cellular development, extracellular matrix, and immunity, including cytokine-cytokine receptor interaction, hematopoietic cell lineage, cell adhesion molecules, phagosome, viral protein interaction with cytokine and cytokine receptor, graft-versus-host disease, intestinal immune network for IgA production, allograft rejection, autoimmune thyroid disease, and type 1 diabetes mellitus (Fig. 7 B). C3 overrepresented biological processes associated with cell cycle and DNA repair, such as cell cycle, DNA replication, oocyte meiosis, cellular senescence, motor protein, Fanconi anaemia pathway, homologous recombination, p53 signaling, base excision repair, and mismatch repair( 32 ) (Fig. 7 C), with cell cycle and DNA replication pathways showing the highest enrichment levels (Fig. 7 C). Lastly, C4 mainly expressed neural pathways, including axon guidance, glutamatergic synapse, dopaminergic synapse, circadian entrainment, GABAergic synapse, and synaptic vesicle cycle (Fig. 7 D). Taken together, these results demonstrate the rationality of our pan-cancer subtyping method. 3.7 Proteins overexpressed in the pan‑cancer subtypes Based on the proteomics data( 33 ) of TCGA pan-cancer, we analyzed the expressions of 210 proteins across the four subtypes (Supplementary Table 4). We observed that INPP4B( 34 ), a crucial protein functioning in regulating the PI3K/AKT signaling pathway and energy metabolism and biosynthesis, exhibited high expression in C1, supporting that C1 shows a more pronounced metabolic profile than the other subtypes (Fig. 8 A). Additionally, a range of proteins involved in regulating extracellular stromal, development, and immune pathways showed significant upregulation in C2, such as ANNEXIN1, CAVEOLIN1, LCK, NDRG1PT346, and VEGFR2 (Fig. 8 B). For instance, LCK, a tyrosine kinase, is essential for T cell activation and the immune response( 35 ). Expectedly, a number of proteins involved in cell cycle regulation and DNA repair showed significantly higher expression levels in C3, such as CASPASE7 CLEAVEDD198, CHK2, CYCLINB1, FOXM1, and PCNA (Fig. 8 C). Of them, CYCLINB1 collaborates with CDK1 to transition cells from the G2 phase to mitosis( 36 ). Furthermore, C4 demonstrated significantly higher expression levels of several proteins, including BRAF, GSK3ALPHABETAPS21S9( 37 , 38 ), GSK3PS9, MTORPS2448, and PTEN( 39 ) (Fig. 8 D). These proteins have been recognized as key molecules in brain tumors. For example, glycogen synthase kinase-3 (GSK3), a serine/threonine kinase, has been identified as a regulator of GBM cell survival( 40 , 41 ). Phosphatase and tensin homolog (PTEN), a phosphatase in humans, has a role in impairing glioma stem cells( 39 ). 3.8 The pathways’ enrichment correlates with etiological and clinicopathologic parameters in pan‑cancer For each tumor sample, we defined its enrichment level of a specific type of pathways as the average enrichment score of all pathways belonging to this type. As anticipated, the enrichment levels of metabolic, cell cycle, and neuronal pathways were most enriched in C1, C3, and C4, respectively, and the enrichment levels of immune, developmental, and stromal pathways were most enriched in C2 (Fig. 9 A). The enrichment levels of metabolic pathways were significantly lower in late-stage than in early-stage cancers ( P = 0.008). In contrast, the enrichment levels of cell cycle, neuronal, developmental, and stromal pathways were significantly higher in late-stage cancers ( P < 0.01) (Fig. 9 B). Moreover, low-grade tumors showed higher enrichment of metabolic, neuronal, developmental, and stromal pathways than high-grade tumors, while high-grade tumors displayed higher enrichment of cell cycle and immune pathways ( P < 0.05) (Fig. 9 C). We further analyzed correlations between the pathways’ enrichment and several common cancer risk factors in pan-cancer, including aging, smoking, viral infection, and alcohol use. Interestingly, almost all these pathways showed positive correlations with ages ( P < 0.0001), except neuronal pathways whose enrichment had a significant negative correlation with ages ( P = 6.86 × 10 − 109 , ρ = -0.24) (Fig. 9 D). Moreover, almost all these pathways showed higher enrichment in smoking than in non-smoking patients, except metabolic pathways whose enrichment was higher in non-smoking patients ( P < 0.01) (Fig. 9 E). Viral infections are significant etiological factors responsible for several cancers( 42 ), including Epstein–Barr virus (EBV) for STAD and DLBC, hepatitis B virus (HBV) for LIHC, and human papillomavirus (HPV) for CESC and HNSC. Of note, virus-infected cancers consistently displayed higher enrichment of cell cycle pathways than non-infected cancers for EBV, HBV, and HPV ( P < 0.01) (Fig. 9 F). In addition, immune pathways were more enriched in EBV-positive than in EBV-negative cancers ( P = 1.79 × 10 − 8 ). In contrast, the enrichment levels of metabolic, neuronal, developmental, and stromal pathways were likely higher in virus-negative than in virus-positive cancers (Fig. 9 G). Interestingly, the patients with alcohol use history displayed significantly higher enrichment of cell cycle pathways while lower enrichment of metabolic pathways than those without alcohol use history ( P < 0.05) (Fig. 9 H). Strikingly, all these pathways exhibited noticeably distinct enrichment levels between male and female patients, with metabolic, cell cycle, and immune pathways being more enriched in male patients while neuronal, developmental, and stromal pathways more enriched in female patients (Fig. 9 I). Hypoxia is a typical property of cancer( 43 ). We found hypoxia scores to be significantly and positively correlated with cell cycle and immune pathways’ enrichment and negatively correlated with neuronal pathways’ enrichment ( ρ = 0.62, 0.21, and − 0.44 for cell cycle, immune, and neuronal pathways, respectively) (Fig. 9 J). In addition, we observed that cell cycle, neuronal, developmental, and stromal pathways were more enriched in the patients receiving radiation therapy than those not, while metabolic and immune pathways exhibited an opposite trend (Fig. 9 K). Taken together, these analyses reveal significant and distinct associations between these pathways’ activities and cancer-associated etiological and clinicopathologic factors. 4 Discussion Pathway-based clustering analyses are advantageous in uncovering cancer subtypes and exploring tumor heterogeneity. First of all, pathway (or gene set) scores can enhance the stability of clustering results that otherwise could be degraded due to outliers of individual genes’ expressions. Second, pathway-based clustering results are easier to elucidate based on specific pathways’ enrichment levels. However, most previous pathway-based clustering methods need to claim specific pathways prior to clustering analysis, as may result in subjectivity and selectivity. To overcome the limitation, here we proposed a novel pathway-based clustering method to discover cancer subtypes. Unlike previous methods, this method first detects gene clusters based on gene expression correlations matrix and identifies overrepresented pathways associated with these gene clusters. Based on the enrichment scores of the overrepresented pathways, this method reveals cancer subtypes by clustering analysis. Applying this method to TCGA pan-cancer, we identified four cancer subtypes: C1, C2, C3, and C4. C1 was characterized by high metabolic activity, the best OS, DSS, and PFI outcomes, the highest response rate to chemotherapy, and the lowest TP53 mutation rate. C2 was characterized by high activities of immune, developmental, and stromal pathways, the lowest tumor purity and ITH. C3 overexpressed cell cycle and DNA repair pathways, had the worst DFI outcome, the highest degree of genomic instability, the highest TP53 mutation rate, and the strongest stemness and proliferation signatures. C4 overrepresented neuronal pathways, with the worst DSS and PFI outcomes, the lowest immune activity, the highest tumor purity, the lowest degree of genomic instability and DNA repair activity, the highest ITH, the weakest stemness and proliferation signatures, the highest IDH1 mutation rate while the lowest TP53 mutation rate, and the lowest response rate to chemotherapy (Fig. 10 ). We further revealed that individual cancer types exhibited significantly distinct distribution across the four subtypes. C1 harbored most adrenocortical, liver, prostate, and uveal melanoma tumors, suggesting these cancer types to be enriched in metabolic pathways. Most kidney renal clear cell carcinoma and pancreatic adenocarcinoma tumors were grouped into C2; this is reasonable since renal cell carcinoma is a highly immunogenic tumor( 44 ) and pancreatic adenocarcinoma is a stroma-enriched tumor( 45 ). C3 included most gastrointestinal cancers, highlighting a high cell cycle activity and genomic instability in this class of cancers. Lastly, C4 was exclusively dominated by neural system-related tumors, most of which were gliomas, aligning with this subtype characterized by neuronal pathway enrichment. For the first time, we comprehensively explored associations between multiple pathways and various etiological and clinicopathologic parameters in pan‑cancer. Ages showed significant positive correlations with the activities of most classes of pathways but a negative correlation with the activity of neuronal pathways. It is imaginable that neuronal activity reduces with increasing age. Our analysis indicates that smoking may enhance the activity of various oncogenic pathways, such as cell cycle, stromal, and developmental pathways, while it could inhibit the activity of metabolic pathways. Our data suggest that viral infections may increase the activity of cell cycle and immune pathways. It is consistent with previous findings that viral infections promote genomic instability and anti-tumor immune responses( 46 ). In addition, alcohol use may increase the activity of cell cycle pathways but decrease the activity of metabolic pathways. Our analyses also suggest that the differences in pathways activity exist between male and female cancer patients. Therapies, such as radiation therapy, could have an impact on the activities of various pathways. A major limitation of the present work is that the pan-cancer cohort encompasses 33 cancer types that involve different size of samples. As a result, the cancer types with large sample size, such as breast, lung and colorectal cancer, would have exerted greater impact on the results. Another limitation is that our findings of the significant associations between pathway activities and etiological and clinicopathologic features, albeit interesting, need to be verified by additional data. In conclusion, we developed a novel clustering algorithm to unveil a new classification of pan-cancer based on metabolic, immune, stromal, developmental, cell cycle, and neuronal pathways. We identified four pan-cancer subtypes distinguished by tumor microenvironment, genome instability, DNA repair activity, ITH, oncogenic signatures, somatic mutation and CNA profiles, protein expression profiles, response to chemotherapy, and clinical outcomes. The identification of pan-cancer subtypes provides novel insights into tumor biology and has potential clinical implications for cancer management. Declarations Ethics approval and consent to participate Not applicable. Consent for publication Not applicable. Availability of data and materials All data associated with this study are available within the paper and its supplementary data. The algorithm and analysis workflow code have also been uploaded to GitHub and can be accessed via the following link: https://github.com/WangX-Lab/PathClustNet. This repository includes the complete code and instructions necessary to reproduce the results presented in this study. Competing interests The authors declare that they have no competing interests. Funding This work was supported by the China Pharmaceutical University (grant number 3150120001 to XW). Acknowledgments Not applicable. Author Contributions Mengli Xu : Software, Formal analysis, Investigation, Data curation, Visualization, Writing - original draft. Hongjing Ai : Formal analysis, Investigation, Visualization. Danni Wang : Investigation. Xiaosheng Wang : Conceptualization, Methodology, Resources, Investigation, Writing - original draft, Supervision, Project administration. References Caswell DR, Swanton C (2017) The role of tumour heterogeneity and clonal cooperativity in metastasis, immune evasion and clinical outcome. BMC Med 15(1):133 Li M, Zhang Z, Li L et al (2020) An algorithm to quantify intratumor heterogeneity based on alterations of gene expression profiles. Commun Biol 3(1):505 Peng X, Chen Z, Farshidfar F et al (2018) Molecular Characterization and Clinical Relevance of Metabolic Expression Subtypes in Human Cancers. Cell Rep 23(1):255–269 Thorsson V, Gibbs DL, Brown SD et al (2018) The Immune Landscape of Cancer. Immunity 48(4):812–830 Li L, Wang X (2021) Identification of gastric cancer subtypes based on pathway clustering. npj Precis Onc 5(1):46 He Y, Jiang Z, Chen C et al (2018) Classification of triple-negative breast cancers based on Immunogenomic profiling. J Exp Clin Cancer Res 37(1):327 Lei J, Luo J, Liu Q et al (2024) Identifying cancer subtypes based on embryonic and hematopoietic stem cell signatures in pan-cancer. Cell Oncol 47(2):587–605 Monti S, Tamayo P, Mesirov J et al (2003) Consensus Clustering: A Resampling-Based Method for Class Discovery and Visualization of Gene Expression Microarray Data. Mach Learn 52(1):91–118 Brière G, Darbo É, Thébault P et al (2021) Consensus clustering applied to multi-omics disease subtyping. BMC Bioinformatics 22(1):361 Guess MJ, Wilson SB (2002) Introduction to hierarchical clustering. J Clin Neurophysiol 19(2):144–151 Hänzelmann S, Castelo R, Guinney J (2013) GSVA: gene set variation analysis for microarray and RNA-seq data. BMC Bioinformatics 14:7 Bland JM, Altman DG (1998) Survival probabilities (the Kaplan-Meier method). BMJ 317(7172):1572 Burrell RA, McGranahan N, Bartek J et al (2013) The causes and consequences of genetic heterogeneity in cancer evolution. Nature 501(7467):338–345 Mermel CH, Schumacher SE, Hill B et al (2011) GISTIC2.0 facilitates sensitive and confident localization of the targets of focal somatic copy-number alteration in human cancers. Genome Biol 12:R41 Yoshihara K, Shahmoradgoli M, Martínez E et al (2013) Inferring tumour purity and stromal and immune cell admixture from expression data. Nat Commun 4:2612 Reimand J, Arak T, Adler P et al (2016) g:Profiler-a web server for functional interpretation of gene lists (2016 update). Nucleic Acids Res 44(W1):W83–89 Gene Ontology Consortium (2015) Gene Ontology Consortium: going forward. Nucleic Acids Res 43(Database issue):D1049–1056 Merico D, Isserlin R, Stueker O et al (2010) Enrichment map: a network-based method for gene-set enrichment visualization and interpretation. PLoS ONE 5(11):e13984 Vrbin CM (2022) Parametric or nonparametric statistical tests: Considerations when choosing the most appropriate option for your data. Cytopathology 33(6):663–667 Liu Q, Wang L (2021) t-Test and ANOVA for data with ceiling and/or floor effects. Behav Res Methods 53(1):264–277 Pripp AH (2018) Pearson’s or Spearman’s correlation coefficients. Tidsskr Nor Laegeforen. ;138(8) Benjamini Y, Hochberg Y (1995) Controlling the False Discovery Rate: A Practical and Powerful Approach to Multiple Testing. J Roy Stat Soc: Ser B (Methodol) 57(1):289–300 Negrini S, Gorgoulis VG, Halazonetis TD (2010) Genomic instability–an evolving hallmark of cancer. Nat Rev Mol Cell Biol 11(3):220–228 Groelly FJ, Fawkes M, Dagg RA et al (2023) Targeting DNA damage response pathways in cancer. Nat Rev Cancer 23(2):78–94 Chen J (2016) The Cell-Cycle Arrest and Apoptotic Functions of p53 in Tumor Initiation and Progression. Cold Spring Harb Perspect Med 6(3):a026104 Yan H, Parsons DW, Jin G et al (2009) IDH1 and IDH2 mutations in gliomas. N Engl J Med 360(8):765–773 Craene BD, Berx G (2013) Regulatory networks defining EMT during cancer initiation and progression. Nat Rev Cancer 13(2):97–110 Grivennikov SI, Greten FR, Karin M (2010) Immunity, Inflammation, and Cancer. Cell 140(6):883–899 Yu G, Wang LG, Han Y et al (2012) clusterProfiler: an R Package for Comparing Biological Themes Among Gene Clusters. OMICS 16(5):284–287 Guengerich FP (2000) Metabolism of chemical carcinogens. Carcinogenesis 21(3):345–351 Copenhaver M, Yu CY, Hoffman RP (2019) Complement Components, C3 and C4, and the Metabolic Syndrome. Curr Diabetes Rev 15(1):44–48 MARINUS MG (2012) DNA Mismatch Repair. EcoSal Plus. ;5(1) Zhang Y, Chen F, Chandrashekar DS et al (2022) Proteogenomic characterization of 2002 human cancers reveals pan-cancer molecular subtypes and associated pathways. Nat Commun 13(1):2669 Zhang M, Ceyhan Y, Kaftanovskaya EM et al (2021) INPP4B protects from metabolic syndrome and associated disorders. Commun Biol 4(1):1–15 Palacios EH, Weiss A (2004) Function of the Src-family kinases, Lck and Fyn, in T-cell development and activation. Oncogene 23(48):7990–8000 Porter L, Donoghue D (2003) Cyclin B1 and CDK1: Nuclear localization and upstream regulators. Prog Cell Cycle Res 5:335–347 Hetman M, Cavanaugh JE, Kimelman D et al (2000) Role of Glycogen Synthase Kinase-3β in Neuronal Apoptosis Induced by Trophic Withdrawal. J Neurosci 20(7):2567–2574 Hur EM, Zhou FQ (2010) GSK3 signalling in neural development. Nat Rev Neurosci 11(8):539–551 Benitez JA, Ma J, D’Antonio M et al (2017) PTEN regulates glioblastoma oncogenesis through chromatin-associated complexes of DAXX and histone H3.3. Nat Commun 8(1):15223 Bruning-Richardson A, Droop A, Tams D et al (2018) Identification of transcriptional targets of GSK3 involved in glioblastoma invasion. Neuro Oncol 20(Suppl 1):i26 Kotliarova S, Pastorino S, Kovell LC et al (2008) Glycogen Synthase Kinase 3 inhibition Induces Glioma Cell Death through c-MYC, NF-κB and Glucose Regulation. Cancer Res 68(16):6643–6651 Morris JDH, Eddleston ALWF, Crook T (1995) Viral infection and cancer. Lancet 346(8977):754–758 Ruan K, Song G, Ouyang G (2009) Role of hypoxia in the hallmarks of human cancer. J Cell Biochem 107(6):1053–1062 de Vivar Chevez AR, Finke J, Bukowski R (2014) The role of inflammation in kidney cancer. Adv Exp Med Biol 816:197–234 Hosein AN, Brekken RA, Maitra A (2020) Pancreatic cancer stroma: an update on therapeutic targeting strategies. Nat Rev Gastroenterol Hepatol 17(8):487–505 Wu Z, Dong Z, Luo J et al (2024) A comprehensive comparison of molecular and phenotypic profiles between hepatitis B virus (HBV)-infected and non-HBV-infected hepatocellular carcinoma by multi-omics analysis. Genomics 116(3):110831 Additional Declarations No competing interests reported. Supplementary Files Supplementarydata.xlsx Cite Share Download PDF Status: Published Journal Publication published 28 Nov, 2024 Read the published version in Functional & Integrative Genomics → Version 1 posted Editorial decision: Revision requested 23 Oct, 2024 Reviews received at journal 23 Oct, 2024 Reviewers agreed at journal 23 Oct, 2024 Reviews received at journal 22 Oct, 2024 Reviewers agreed at journal 22 Oct, 2024 Reviewers invited by journal 03 Oct, 2024 Editor assigned by journal 24 Sep, 2024 Submission checks completed at journal 24 Sep, 2024 First submitted to journal 19 Sep, 2024 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-5114169","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":369732690,"identity":"6a28cd17-733d-4fca-9ffe-6f8f158b6a82","order_by":0,"name":"Mengli Xu","email":"","orcid":"","institution":"China Pharmaceutical University","correspondingAuthor":false,"prefix":"","firstName":"Mengli","middleName":"","lastName":"Xu","suffix":""},{"id":369732691,"identity":"86b746df-0d17-4785-82b3-5ffc1bea9d6d","order_by":1,"name":"Hongjing Ai","email":"","orcid":"","institution":"China Pharmaceutical University","correspondingAuthor":false,"prefix":"","firstName":"Hongjing","middleName":"","lastName":"Ai","suffix":""},{"id":369732692,"identity":"453c1ac8-cb1c-48ea-81a3-34593b9a198d","order_by":2,"name":"Danni Wang","email":"","orcid":"","institution":"China Pharmaceutical University","correspondingAuthor":false,"prefix":"","firstName":"Danni","middleName":"","lastName":"Wang","suffix":""},{"id":369732693,"identity":"d6c26c6f-c6a7-4d63-ba86-0a0f7571ddf7","order_by":3,"name":"Xiaosheng Wang","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAABEElEQVRIiWNgGAWjYDACZiBmbIBzbSAUDwla0ojQwoCq5TBhLQbHmR8+/LnDJk/egcfwc8Gv8/LmEgmMD962Mcib49Ai2cxmbMx7Jq3Y8ACPsfTMvtuGO2ckMBvObWMw3NmAXQs/M4OZNGPb4cSNDbwbpHl7bicY3Ehgk+ZtY0gwOIBdCxsz+zfJn23/QVo2/+btOQfSwv4bnxZ+Zh4zCd62A4nzGXi3SfP8OAC2hRmfFslmnmJj3rbkxA3M/N+seRuSDXf2PGyWnHNOwnADDi0G549vfPizzS5xfntb8m2eP3by5uzJBz+8KbORx2ULQi8oRhjbgAxINEkQUA8E8iB1DH9AWkbBKBgFo2AUoAIAN2FX7qqVVNkAAAAASUVORK5CYII=","orcid":"","institution":"China Pharmaceutical University","correspondingAuthor":true,"prefix":"","firstName":"Xiaosheng","middleName":"","lastName":"Wang","suffix":""}],"badges":[],"createdAt":"2024-09-19 05:53:33","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-5114169/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-5114169/v1","draftVersion":[],"editorialEvents":[{"content":"https://doi.org/10.1007/s10142-024-01501-0","type":"published","date":"2024-11-28T15:57:17+00:00"}],"editorialNote":"","failedWorkflow":false,"files":[{"id":68192307,"identity":"d8b912e1-5e69-4ee9-831e-30b3cc549e9e","added_by":"auto","created_at":"2024-11-04 14:00:19","extension":"jpg","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":260811,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eA schematic overview of the algorithm.\u003c/strong\u003e\u003c/p\u003e","description":"","filename":"Figure1.jpg","url":"https://assets-eu.researchsquare.com/files/rs-5114169/v1/df8af68fbb706bedea59cd77.jpg"},{"id":68189174,"identity":"a0a76a0f-f55f-4dac-aa29-9ae7bb6cc577","added_by":"auto","created_at":"2024-11-04 13:36:19","extension":"jpg","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":542969,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003ePan-cancer subtypes identified by clustering analysis.(A) \u003c/strong\u003eFour subtypes (C1, C2, C3, C4) identified using a novel clustering algorithm in TCGA pan-cancer data. Ad: Adenocarcinoma, GI: Gastrointestinal, GYN: Gynecological, NA: Data not available. \u003cstrong\u003e(B)\u003c/strong\u003e Kaplan-Meier curves showing differences in 10-year survival rates (OS, DSS, DFI, PFI) among the four subtypes. The log-rank test \u003cem\u003eP\u003c/em\u003e values are shown. OS: overall survival, DSS: disease-specific survival, PFI: progression-free interval, DFI: disease-free interval. \u003cstrong\u003e(C)\u003c/strong\u003eComparison of clinical and pathological features among pan-cancer subtypes in TCGA, including the proportion of early-stage (stage I-II) tumors and low-grade (G1-2) tumors, and chemotherapy response rates. Statistical analysis by chi-square test, \u003cem\u003eP\u003c/em\u003e \u0026lt; 0.001.\u003c/p\u003e","description":"","filename":"Figure2.jpg","url":"https://assets-eu.researchsquare.com/files/rs-5114169/v1/07da158f5e965c2d115490d7.jpg"},{"id":68189180,"identity":"efa2621d-5eee-4c6e-a551-022e1157c6ac","added_by":"auto","created_at":"2024-11-04 13:36:19","extension":"jpg","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":339037,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eDistribution of subtypes across\u003c/strong\u003e \u003cstrong\u003eindividual cancer types. (A) \u003c/strong\u003eProportions of each subtype in 33 TCGA cancer types. Only the proportions over 3% are shown in the bars.\u003cstrong\u003e (B)\u003c/strong\u003e Sankey diagram illustrating subtype composition in pan-gastric and pan-renal cancers.\u003c/p\u003e","description":"","filename":"Figure3.jpg","url":"https://assets-eu.researchsquare.com/files/rs-5114169/v1/c5bf7b65f6a2ead860209d4a.jpg"},{"id":68190927,"identity":"3dfd526b-1832-4830-a5e1-eba26dbfdce3","added_by":"auto","created_at":"2024-11-04 13:52:19","extension":"jpg","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":147440,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eComparisons of immune, stromal, and tumor purity scores among pan-cancer subtypes. \u003c/strong\u003eComparisons of stromal (left), immune (middle), and tumor purity (right) scores across the four subtypes.\u003c/p\u003e","description":"","filename":"Figure4.jpg","url":"https://assets-eu.researchsquare.com/files/rs-5114169/v1/5ca5dfbcdd24386a28f26d82.jpg"},{"id":68189185,"identity":"7e91dd37-eebc-4026-8af4-32819fe64356","added_by":"auto","created_at":"2024-11-04 13:36:19","extension":"jpg","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":1125920,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eComparisons of genomic instability among pan-cancer subtypes. \u003c/strong\u003eComparison of TMB \u003cstrong\u003e(A)\u003c/strong\u003e,\u003cstrong\u003e \u003c/strong\u003eCNA scores \u003cstrong\u003e(B)\u003c/strong\u003e, HRD scores \u003cstrong\u003e(C)\u003c/strong\u003e, and ITH scores \u003cstrong\u003e(D) \u003c/strong\u003eamong the four subtypes.\u003cstrong\u003e (E) \u003c/strong\u003eThe\u003cstrong\u003e \u003c/strong\u003eG-scores are calculated by GISTIC2, representing gains and losses in copy number. \u003cstrong\u003e(F) \u003c/strong\u003eEnrichment scores of DNA damage repair pathways.\u003cstrong\u003e (G) \u003c/strong\u003eThe oncoplots showing the top ten mutated genes in each subtype.\u003cstrong\u003e \u003c/strong\u003eTMB: Tumor Mutational Burden, CNA: Copy Number Alteration, ITH: Intratumoral Heterogeneity, HRD: Homologous Recombination Deficiency. The Kruskal-Wallis test \u003cem\u003eP\u003c/em\u003e values are shown.\u003c/p\u003e","description":"","filename":"Figure5.jpg","url":"https://assets-eu.researchsquare.com/files/rs-5114169/v1/db1850701984e74021f5e61b.jpg"},{"id":68189176,"identity":"0ca9927c-2b85-4089-b6a4-69db93907acf","added_by":"auto","created_at":"2024-11-04 13:36:19","extension":"jpg","order_by":6,"title":"Figure 6","display":"","copyAsset":false,"role":"figure","size":168180,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eComparisons of molecular and phenotypic features among pan-cancer subtypes. (A) \u003c/strong\u003eComparisons of the enrichment scores of biological processes (EMT, proliferation, and stemness) among the four subtypes. \u003cstrong\u003e(B)\u003c/strong\u003e Comparisons of the enrichment scores of clinical features (metastasis, differentiation, inflammation, and invasion) among the four subtypes. The Kruskal-Wallis test \u003cem\u003eP\u003c/em\u003evalues are shown. * \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.jpg","url":"https://assets-eu.researchsquare.com/files/rs-5114169/v1/fad7d774435f958f03446a64.jpg"},{"id":68190472,"identity":"6b6ff496-2a8f-4387-b502-870a0c729c39","added_by":"auto","created_at":"2024-11-04 13:44:19","extension":"jpg","order_by":7,"title":"Figure 7","display":"","copyAsset":false,"role":"figure","size":313380,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eComparisons of pathway enrichment among pan-cancer subtypes. (A) \u003c/strong\u003eBiological processes enriched in subtype C1. \u003cstrong\u003e(B)\u003c/strong\u003e Biological processes enriched in subtype C2. \u003cstrong\u003e(C)\u003c/strong\u003e Biological processes enriched in subtype C3. \u003cstrong\u003e(D) \u003c/strong\u003eBiological processes enriched in subtype C 4.\u003c/p\u003e","description":"","filename":"Figure7.jpg","url":"https://assets-eu.researchsquare.com/files/rs-5114169/v1/d8b278880cca995eaa7d597a.jpg"},{"id":68189181,"identity":"a36e5285-9acf-4470-99a8-b7bcf10accb2","added_by":"auto","created_at":"2024-11-04 13:36:19","extension":"jpg","order_by":8,"title":"Figure 8","display":"","copyAsset":false,"role":"figure","size":579093,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eComparisons of protein expression in pan-cancer subtypes. (A)\u003c/strong\u003e Metabolism-related protein expression levels in subtypes. \u003cstrong\u003e(B)\u003c/strong\u003e Stromal, development, and immune-related protein expression levels in subtypes.\u003cstrong\u003e (C)\u003c/strong\u003e Cell cycle-related protein expression levels in subtypes. \u003cstrong\u003e(D)\u003c/strong\u003e Neural-related protein expression levels in subtypes. The Kruskal-Wallis test \u003cem\u003eP\u003c/em\u003evalues are shown.\u003c/p\u003e","description":"","filename":"Figure8.jpg","url":"https://assets-eu.researchsquare.com/files/rs-5114169/v1/1af61884413ce9017189918c.jpg"},{"id":68189183,"identity":"c84398e3-003d-4a51-84c4-c6c5ecaf6199","added_by":"auto","created_at":"2024-11-04 13:36:19","extension":"jpg","order_by":9,"title":"Figure 9","display":"","copyAsset":false,"role":"figure","size":975040,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eCorrelations\u003c/strong\u003e \u003cstrong\u003eof pathway enrichment levels with etiological etiological factors\u003c/strong\u003e \u003cstrong\u003ein pan-cancer. (A)\u003c/strong\u003e Pathway enrichment levels in four cancer subtypes.The Kruskal-Wallis test \u003cem\u003eP\u003c/em\u003e values are shown. Comparisons of pathway enrichment levels between late-stage (stage III-IV) and early-stage (stage I-II) tumors \u003cstrong\u003e(B)\u003c/strong\u003e,\u003cstrong\u003e \u003c/strong\u003ehigh-grade (G3-4) and low-grade (G1-2) tumors \u003cstrong\u003e(C)\u003c/strong\u003e. \u003cstrong\u003e(D)\u003c/strong\u003e Correlations between pathway enrichment levels and ages of patients.\u003cstrong\u003e (E)\u003c/strong\u003e Comparisons of pathway enrichment levels between smokers and non-smokers. \u003cstrong\u003e(F)\u003c/strong\u003e Comparisons of cell cycle pathway enrichment levels between HPV/HBV/EBV positive and negative patients. \u003cstrong\u003e(G)\u003c/strong\u003e Comparisons of metabolic, neuronal, and immune-related pathway enrichment levels between EBV positive and negative patients.\u003cstrong\u003e (H)\u003c/strong\u003e Comparisons of cell cycle and metabolic pathway enrichment levels in patients with and without alcohol history.\u003cstrong\u003e (I)\u003c/strong\u003e Comparisons of pathway enrichment levels between male and female patients. \u003cstrong\u003e(J)\u003c/strong\u003e Correlations between pathway enrichment levels and hypoxia score of tumors. The Spearman correlation coefficients and \u003cem\u003eP\u003c/em\u003e values are shown. \u003cstrong\u003e(K)\u003c/strong\u003e Comparisons of pathway enrichment levels between patients treated with and without radiation therapy.The Mann-Whitney \u003cem\u003eU\u003c/em\u003e test \u003cem\u003eP\u003c/em\u003e values are shown. * \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":"Figure9.jpg","url":"https://assets-eu.researchsquare.com/files/rs-5114169/v1/66103f347fe45ba2a66bf475.jpg"},{"id":68190471,"identity":"918e4817-30c1-40f5-b445-39600e4b1f47","added_by":"auto","created_at":"2024-11-04 13:44:19","extension":"jpg","order_by":10,"title":"Figure 10","display":"","copyAsset":false,"role":"figure","size":323398,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eA summary of molecular and clinical features of the four pan-cancer subtypes. \u003c/strong\u003eThe four subtypes display significantly different molecular and clinical features. The figure was created with BioRender.com\u003c/p\u003e","description":"","filename":"Figure10.jpg","url":"https://assets-eu.researchsquare.com/files/rs-5114169/v1/66e62f13f881b27e9296f964.jpg"},{"id":70389475,"identity":"036f20cd-6fee-43c0-b36a-36e468cdc4e5","added_by":"auto","created_at":"2024-12-02 17:28:47","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":5706372,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-5114169/v1/934f6d48-83ff-478b-b82e-31eef148153f.pdf"},{"id":68190475,"identity":"c1df9078-eb28-450e-bebf-c902c8722fa4","added_by":"auto","created_at":"2024-11-04 13:44:19","extension":"xlsx","order_by":12,"title":"","display":"","copyAsset":false,"role":"supplement","size":92355,"visible":true,"origin":"","legend":"","description":"","filename":"Supplementarydata.xlsx","url":"https://assets-eu.researchsquare.com/files/rs-5114169/v1/8c7d66763c9b1aa8e6a7da36.xlsx"}],"financialInterests":"No competing interests reported.","formattedTitle":"Gene clusters-based pathway enrichment analysis identifies four pan-cancer subtypes with distinct molecular and clinical features","fulltext":[{"header":"1 Introduction","content":"\u003cp\u003eTumor heterogeneity is one of the typical properties of cancer, reflecting molecular and phenotypic variations between tumors or within a tumor(\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e). With the emergence of abundant cancer-associated bulk and single-cell transcriptomics data, unraveling transcriptome heterogeneity in cancer is viable(\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e). Gene expression profiles-based clustering analysis is a commonly used avenue to decipher transcriptome heterogeneity in cancer, based on which myriad investigations into identification of cancer subtypes have been implemented(\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e, \u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e). For example, Thorsson et al. uncovered six immune subtypes of the TCGA pan-cancer based on transcriptome-based clustering analysis(\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e). Another analysis of the TCGA pan-cancer transcriptomes identified three metabolic subtypes showing different clinical outcomes(\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e). However, because outliers of individual genes\u0026rsquo; expressions may compromise the stability of clustering results, pathways or gene sets-based clustering methods have been proposed(\u003cspan additionalcitationids=\"CR6\" citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e). For example, Li et al. identified three gastric cancer subtypes based on the enrichment levels of 15 pathways involved in immune, DNA repair, oncogenic, and stromal signatures(\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e). He et al. uncovered three immune-specific subtypes of triple-negative breast cancer based on the enrichment levels of 29 immune signatures(\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e). Lei et al. unveiled four subtypes of the pan-cancer of 33 TCGA cancer types based on the enrichment scores of six stemness-associated gene sets, which are markers of embryonic stem cells and hematopoietic stem cells, respectively(\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e). Although these pathway-based clustering analyses are successful in identifying cancer subtypes, they are disadvantageous in that specific pathways need to be explicitly claimed in advance.\u003c/p\u003e \u003cp\u003eTo fill the knowledge gap, here we developed a pathway-based clustering method for unveiling cancer subtypes. Different from previous methods, this method establishes the matrix of expression correlations between gene pairs. Next, the genes are classified into different clusters using a consensus clustering analysis of the gene expression correlations matrix. For each cluster of genes, its associated overrepresented pathways are identified. Lastly, based on the enrichment scores of these pathways, hierarchal clustering analysis is performed to uncover cancer subtypes. We applied the method to TCGA pan-cancer and identified four subtypes and comprehensively characterized the molecular and clinical features of these subtypes. Furthermore, we explored the correlations between pathways\u0026rsquo; activities and etiological and clinicopathologic parameters in pan‑cancer.\u003c/p\u003e"},{"header":"2 Methods","content":"\u003cdiv id=\"Sec3\" class=\"Section2\"\u003e \u003ch2\u003e2.1 Overview of the algorithm\u003c/h2\u003e \u003cp\u003eThe PathClustNet algorithm comprises several steps. First, for a cancer-associated gene expression \u003cem\u003en\u003c/em\u003e \u0026times; \u003cem\u003em\u003c/em\u003e matrix \u003cem\u003eM\u003c/em\u003e, where \u003cem\u003en\u003c/em\u003e and \u003cem\u003em\u003c/em\u003e denote the number of genes and samples, respectively, we calculate the standard deviation of each gene across all samples to assess its variability, we then select the top 5000 genes with the highest standard deviations in descending order and obtain the top 5000 gene expressions correlation matrix \u003cem\u003eM\u003c/em\u003e\u003csub\u003e\u003cem\u003ef\u003c/em\u003e\u003c/sub\u003e. Next, we calculate the Pearson correlation between each pair of genes\u0026rsquo; expressions and obtain the gene expressions correlation matrix \u003cem\u003eM\u003c/em\u003e\u003csub\u003e\u003cem\u003eg\u003c/em\u003e\u003c/sub\u003e. By consensus clustering of \u003cem\u003eM\u003c/em\u003e\u003csub\u003e\u003cem\u003eg\u003c/em\u003e\u003c/sub\u003e, we attain gene clusters. For each gene cluster, we identify the pathways significantly associated with it by gene set enrichment analysis. Based on these pathways, we construct samples\u0026rsquo; pathway enrichment matrix \u003cem\u003eM\u003c/em\u003e\u003csub\u003e\u003cem\u003ep\u003c/em\u003e\u003c/sub\u003e. Finally, we identify cancer subtypes by hierarchical clustering of \u003cem\u003eM\u003c/em\u003e\u003csub\u003e\u003cem\u003ep\u003c/em\u003e\u003c/sub\u003e. A schematic overview of the algorithm is displayed in Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec4\" class=\"Section2\"\u003e \u003ch2\u003e2.2 Data collection\u003c/h2\u003e \u003cp\u003eWe downloaded data of RSEM normalized and batch effects adjusted gene expressions, somatic mutations, copy number alternations (CNA), protein expressions, and clinical information for the TCGA pan-cancer (PANCAN) cohort consisting of 33 cancer types from UCSC Xena (\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://xenabrowser.net/datapages/\u003c/span\u003e\u003cspan address=\"https://xenabrowser.net/datapages/\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e). The 33 cancer types included adrenocortical carcinoma (ACC), bladder urothelial carcinoma (BLCA), breast invasive carcinoma (BRCA), cervical squamous cell carcinoma and endocervical adenocarcinoma (CESC), cholangiocarcinoma (CHOL), colon adenocarcinoma (COAD), lymphoid neoplasm diffuse large B-cell lymphoma (DLBC), esophageal carcinoma (ESCA), glioblastoma multiforme (GBM), head and neck squamous cell carcinoma (HNSC), kidney chromophobe (KICH), kidney renal clear cell carcinoma (KIRC), kidney renal papillary cell carcinoma (KIRP), acute myeloid leukemia (LAML), brain lower grade glioma (LGG), liver hepatocellular carcinoma (LIHC), lung adenocarcinoma (LUAD), lung squamous cell carcinoma (LUSC), mesothelioma (MESO), ovarian serous cystadenocarcinoma (OV), pancreatic adenocarcinoma (PAAD), pheochromocytoma and paraganglioma (PCPG), prostate adenocarcinoma (PRAD), rectum adenocarcinoma (READ), sarcoma (SARC), skin cutaneous melanoma (SKCM), stomach adenocarcinoma (STAD), testicular germ cell tumors (TGCT), thyroid carcinoma (THCA), thymoma (THYM), uterine corpus endometrial carcinoma (UCEC), uterine carcinosarcoma (UCS), and uveal melanoma (UVM). A summary of these datasets is shown in Supplementary Table \u003cspan refid=\"MOESM1\" class=\"InternalRef\"\u003eS1\u003c/span\u003e.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec5\" class=\"Section2\"\u003e \u003ch2\u003e2.3 Clustering analysis\u003c/h2\u003e \u003cp\u003eWe utilized consensus clustering to identify gene clusters with the input of gene expression correlations matrix. Consensus clustering defines specified cluster counts by subsampling from a group of samples, which enhances the stability and reliability of clustering by aggregating results from multiple clustering iterations(\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e). We performed this analysis using the \u003cem\u003eConsensusClusterPlus\u003c/em\u003e R package(\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e) with the parameters: clusterAlg = \u0026ldquo;pam\u0026rdquo;, distance= \u0026ldquo;euclidean\u0026rdquo;, reps\u0026thinsp;=\u0026thinsp;1000, pItem\u0026thinsp;=\u0026thinsp;0.8, and pFeature\u0026thinsp;=\u0026thinsp;1. In addition, we employed hierarchical clustering(\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e) to identify cancer subtypes with the input of pathway enrichment matrix. This analysis was performed using the \u003cem\u003ehclust\u003c/em\u003e R package with the parameters: method = \u0026ldquo;ward. D2\u0026rdquo; and members\u0026thinsp;=\u0026thinsp;NULL. Before hierarchical clustering, the data of pathway enrichment scores were transformed by z-score and translated into distance matrices by the \u003cem\u003edist\u003c/em\u003e function with the parameter: method = \u0026ldquo;Euclidean\u0026rdquo;.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec6\" class=\"Section2\"\u003e \u003ch2\u003e2.4 Single‑sample gene‑set enrichment analysis\u003c/h2\u003e \u003cp\u003eWe employed the single-sample gene-set enrichment analysis (ssGSEA) method(\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e) to evaluate the enrichment level of a gene set in a sample; the gene set represents some pathway, biological process or phenotypic feature. The ssGSEA algorithm determines a gene set\u0026rsquo;s enrichment score in a sample with the input of gene expression profiles. We performed this analysis with the \u003cem\u003eGSVA\u003c/em\u003e R package. The gene sets analyzed are presented in Supplementary Table S2.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec7\" class=\"Section2\"\u003e \u003ch2\u003e2.5 Survival analysis\u003c/h2\u003e \u003cp\u003eWe utilized the Kaplan-Meier (K-M)(\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e) model to compare overall survival (OS), disease-specific survival (DSS), disease-free interval (DFI), and progression-free interval (PFI) among different subgroups of cancer patients. Log-rank tests were employed to assess the significance of survival time differences. The survival analysis was implemented with the \u003cem\u003esurvfit\u003c/em\u003e function in the \u003cem\u003esurvival\u003c/em\u003e R package.\u003c/p\u003e \u003cp\u003e \u003cb\u003e2.6 Evaluating CNA, tumor mutation burden (TMB), homologous recombination deficiency (HRD), intratumor heterogeneity (ITH), and tumor immune infiltration levels\u003c/b\u003e \u003c/p\u003e \u003cp\u003eTo evaluate CNA and HRD in tumors, we obtained CNA and HRD scores of TCGA pan-cancer from the publication by Knijnenburg et al.(\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e). In addition, we used GISTIC2(\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e) to calculate tumors\u0026rsquo; G-scores with the input of \u0026ldquo;SNP6\u0026rdquo; files. The G-score reflects the amplitude of the somatic CNA and the frequency of its occurrence across a class of samples(\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e). We defined a tumor\u0026rsquo;s TMB as the total count of somatic mutations in the tumor. Besides, we utilized the DEPTH algorithm(\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e) to quantify ITH. DEPTH evaluates ITH at mRNA level based on the gene expression perturbations\u0026rsquo; heterogeneity To assess tumor immune infiltration levels, we calculated immune scores using the ESTIMATE algorithm(\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e) with the input of gene expression profiles.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec8\" class=\"Section2\"\u003e \u003ch2\u003e2.7 Identification of gene and protein markers for subtypes\u003c/h2\u003e \u003cp\u003eTo identify gene or protein markers for a subtype, we first identified the significantly upregulated genes or proteins in the subtype relative to each of the other subtypes by differential expression analysis. The gene or protein markers for the subtype were defined as the common genes or proteins among the upregulated genes or proteins. The differential expression analysis for genes used the threshold of Student\u0026rsquo;s \u003cem\u003et\u003c/em\u003e test adjusted \u003cem\u003eP\u003c/em\u003e value\u0026thinsp;\u0026lt;\u0026thinsp;0.01 and mean expression fold change\u0026thinsp;\u0026gt;\u0026thinsp;2; for proteins, the threshold was Student\u0026rsquo;s \u003cem\u003et\u003c/em\u003e test adjusted \u003cem\u003eP\u003c/em\u003e value\u0026thinsp;\u0026lt;\u0026thinsp;0.05 and mean expression fold change\u0026thinsp;\u0026gt;\u0026thinsp;1.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec9\" class=\"Section2\"\u003e \u003ch2\u003e2.8 Pathway enrichment analysis\u003c/h2\u003e \u003cp\u003eTo identify enriched pathways in a specific subtype, we utilized the network tool g:Profiler(\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e) with the input of gene expression profiles and the marker genes of the subtype, selecting the Biological Process (BP) from the Gene Ontology (GO) database(\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e), and setting the threshold of adjusted \u003cem\u003eP\u003c/em\u003e value\u0026thinsp;\u0026lt;\u0026thinsp;0.05. We also performed pathway enrichment analysis based on protein expression profiles. This analysis used the \u003cem\u003eenrichGO\u003c/em\u003e function in the \u003cem\u003eclusterProfiler\u003c/em\u003e R package with the input of protein expression profiles and the marker proteins of the subtype. The \u003cem\u003ecnetplot\u003c/em\u003e function was utilized for visualization of the enrichment analysis results(\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e). The parameters for pathway enrichment analysis were: OrgDb = \u0026ldquo;org.Hs.eg.db\u0026rdquo;, ont = \u0026ldquo;BP\u0026rdquo;, pvalueCutoff\u0026thinsp;=\u0026thinsp;0.05, qvalueCutoff\u0026thinsp;=\u0026thinsp;0.05.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec10\" class=\"Section2\"\u003e \u003ch2\u003e2.9 Statistical analysis\u003c/h2\u003e \u003cp\u003eIn class comparisons, for non-normally distributed data, we utilized Mann\u0026ndash;Whitney \u003cem\u003eU\u003c/em\u003e tests (for two classes) or Kruskal\u0026ndash;Wallis (K\u0026ndash;W) tests(\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e) (for more than two classes); for normally distributed data, we used Student\u0026rsquo;s t-tests (for two classes) or ANOVA tests(\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e) (for more than two classes). In the analysis of contingency tables, we utilized Fisher\u0026rsquo;s exact tests or Chi-square tests. We employed the Pearson or Spearman method(\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e) to evaluate the correlations between two variables. To adjust for \u003cem\u003eP\u003c/em\u003e values in multiple tests, we utilized the Benjamini-Hochberg method(\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e) to calculate the false discovery rate (FDR). All statistical analyses were conducted in the R programming environment (version 4.2.2).\u003c/p\u003e \u003c/div\u003e"},{"header":"3 Results","content":"\u003cdiv id=\"Sec12\" class=\"Section2\"\u003e \u003ch2\u003e3.1 Clustering analysis identifies four subtypes of pan‑cancer\u003c/h2\u003e \u003cp\u003eWe applied the algorithm to the TCGA pan-cancer transcriptomics data involving 10,323 cancer specimens and 20,531 genes. Consensus clustering of the gene expressions correlation matrix identified six gene clusters, which fell into six categories of pathways: neural, metabolic, stromal, developmental, immune, and cell cycle pathways. From each of the six pathway categories, we randomly selected five pathways. Based on the 30 pathways, we obtained a pathway enrichment matrix of the pan-cancer for which we preformed hierarchical clustering and unveiled four pan-cancer subtypes (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003eA). We termed the four subtypes C1, C2, C3, and C4, respectively. C1 showed high enrichment of metabolic pathways; C2 overexpressed the stromal, developmental, and immune pathways; C3 overrepresented cell cycle pathways; and C4 displayed upregulation of neuronal pathways.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eThe comparisons of survival prognosis among the four subtypes revealed almost consistent pattern for the four endpoints (10-year OS, DSS, DFI, and PFI): C1 had the best OS, DFI followed the order of C4\u0026thinsp;\u0026gt;\u0026thinsp;C1\u0026thinsp;\u0026gt;\u0026thinsp;C2\u0026thinsp;\u0026gt;\u0026thinsp;C3, and both DSS and PFI had the pattern of C1\u0026thinsp;\u0026gt;\u0026thinsp;C2\u0026thinsp;\u0026gt;\u0026thinsp;C3\u0026thinsp;\u0026gt;\u0026thinsp;C4 (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003eB). These results underscore significant prognostic differences among these subtypes. To be specific, C1 exhibited the most favorable OS, DSS, and PFI, C4 had the worst DSS and PFI, and C3 displayed the worst DFI. It indicates that overrepresentation of metabolic pathways is a favorable prognostic indicator, while upregulation of neuronal and cell cycle pathways are adverse prognostic factors.\u003c/p\u003e \u003cp\u003eConforming to the survival outcomes, the proportion of early-stage (stage I-II) tumors in the four subtypes followed the pattern: C1 (64.0%)\u0026thinsp;\u0026gt;\u0026thinsp;C3 (50.0%)\u0026thinsp;\u0026gt;\u0026thinsp;C2 (43.3%)\u0026thinsp;\u0026gt;\u0026thinsp;C4 (0.0%) (Chi-square test, \u003cem\u003eP\u003c/em\u003e\u0026thinsp;=\u0026thinsp;4.81\u0026times;10\u003csup\u003e\u0026ndash;10\u003c/sup\u003e) (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003eC). The proportion of low-grade tumors (G1-2) also followed the similar trend (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003eC) (\u003cem\u003eP\u003c/em\u003e\u0026thinsp;=\u0026thinsp;7.26\u0026times;10\u003csup\u003e\u0026minus;\u0026thinsp;6\u003c/sup\u003e). Furthermore, we observed that the response rate of these subtypes to chemotherapy followed the pattern: C1 (85.7%)\u0026thinsp;\u0026gt;\u0026thinsp;C3 (84.4%)\u0026thinsp;\u0026gt;\u0026thinsp;C2 (81.8%)\u0026thinsp;\u0026gt;\u0026thinsp;C4 (57.8%) (\u003cem\u003eP\u003c/em\u003e\u0026thinsp;=\u0026thinsp;1.60\u0026times;10\u003csup\u003e\u0026ndash;54\u003c/sup\u003e) (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003eC).\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec13\" class=\"Section2\"\u003e \u003ch2\u003e3.2 Distribution of individual cancer types across the subtypes\u003c/h2\u003e \u003cp\u003eAmong the four pan-cancer subtypes, C1 harbored the highest proportions of ACC (91.1%), LIHC(93.8%), PRAD(94.8%), THCA(77.4%), CHOL(50%), KICH(75.8%), KIRP(68.8%) and UVM(95.0%), C2 harbored the highest proportions of HNSC(48.1%), KIRC(89.3%), LUAD(70.6%), PAAD(79.9%) and MESO(65.6%), C3 harbored the highest proportions of BLCA(59.1%), CESC(82.3%), COAD(85.5%), DLBC(100.0%), STAD(53.5%), ESCA(74.1%), LAML(100.0%), OV(67.2%), READ(85.7%), SKCM(61.5%), TGCT(86.5%), THYM(73.3%), UCEC(57.1%), and UCS(86.0%), and C4 harbored the highest proportions of GBM(87.4%), LGG(99.8%), and PCPG(98.4%)(Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003eA). Notably, most ACC (91.1%), LIHC (93.8%), PRAD (94.8%), and UVM (95.0%) tumors belonged to C1, indicating the high enrichment of metabolic pathways in these cancer types. C2 was dominated by most KIRC (89.3%) and PAAD (79.9%) tumors, suggesting that the stromal, developmental, and immune pathways are enriched in both cancer types. C3 harbored all DLBC and LAML and most of CESC (82.3%), COAD (85.5%), READ (85.7%), TGCT (86.5%) and UCS (86.0%) tumors, suggesting that cell cycle pathways are overrepresented in these cancer types. Finally, most LGG (99.8%), GBM (87.4%), and PCPG (98.4%) tumors belonged to C4; it is justified as these cancer types are neural system-related and C4 overexpressed neuronal pathways. Notably, about 80% of gastrointestinal cancers, such as COAD, ESCA, STAD, and READ, were grouped into C3 (Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003eB), indicating aberrant activation of the cell cycle pathway in gastrointestinal cancers. Moreover, approximately 70% of kidney cancers, including KIRC and KIRP, were classified into C2 (Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003eB). It highlights that kidney cancers are highly enriched in stromal, developmental, and immune pathways.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec14\" class=\"Section2\"\u003e \u003ch2\u003e3.3 The pan-cancer subtypes have different tumor microenvironment\u003c/h2\u003e \u003cp\u003eAs expected, C2 displayed the highest immune scores and stromal scores, and the lowest tumor purity among the four subtypes (K-W test, \u003cem\u003eP\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;0.001) (Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003e), aligning with this subtype overexpressing immune and stromal pathways. In contrast, C4, characterized by high enrichment of neuronal pathways, had the lowest immune scores while the highest tumor purity. In addition, C3, overrepresenting cell cycle pathways, displayed the lowest stromal scores and the second highest immune scores. Altogether, these results indicate significantly different tumor microenvironment among these subtypes.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec15\" class=\"Section2\"\u003e \u003ch2\u003e3.4 The pan-cancer subtypes have different genomic features\u003c/h2\u003e \u003cp\u003eGenomic instability is a common property of cancer, embodied by increased TMB, CNA and HRD(\u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e). We observed that C3 had the highest TMB, CNA and HRD scores, whereas C4 displayed the lowest (K-W test, \u003cem\u003eP\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;0.001) (Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003eA, Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003eB and Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003eC). ITH is also a typical feature of cancer. The ITH scores calculated by DEPTH(\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e) were highest in C4 and lowest in C2, following the pattern: C4\u0026thinsp;\u0026gt;\u0026thinsp;C3\u0026thinsp;\u0026gt;\u0026thinsp;C1\u0026thinsp;\u0026gt;\u0026thinsp;C2 (K-W test, \u003cem\u003eP\u003c/em\u003e\u0026thinsp;=\u0026thinsp;0) (Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003eD). Additionally, G-scores of copy number amplifications and deletions followed the same pattern (Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003eE). DNA repair pathways are crucial in maintaining genomic instability(\u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e). The enrichment scores of four key DNA repair pathways (base excision repair, nucleotide excision repair, Fanconi Anemia, and homologous recombination) exhibited a uniform pattern across the four subtypes: C3\u0026thinsp;\u0026gt;\u0026thinsp;C2\u0026thinsp;\u0026gt;\u0026thinsp;C1\u0026thinsp;\u0026gt;\u0026thinsp;C4 (\u003cem\u003eP\u003c/em\u003e\u0026thinsp;=\u0026thinsp;0) (Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003eF). In addition, the enrichment scores of non-homologous end joining, translesion synthesis, and damage sensor pathways were also highest in C3, and C4 had the lowest scores of non-homologous end joining and direct repair pathways (\u003cem\u003eP\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;0.001) (Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003eF). Collectively, these results suggested that the DNA damage response was strongest in C3 and weakest in C4. Taken together, these analyses revealed that C3 and C4 had the highest and lowest degree of genomic instability, respectively. We further compared somatic mutation profiles among these subtypes. Figure\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003eG showed top 10 genes most frequently mutated in each subtype. Consistently, \u003cem\u003eTP53\u003c/em\u003e had the highest or second highest mutation rate within each subtype, and its mutation rate followed the pattern across the subtypes: C3 (50%)\u0026thinsp;\u0026gt;\u0026thinsp;C2 (39%)\u0026thinsp;\u0026gt;\u0026thinsp;C4 (36%)\u0026thinsp;\u0026gt;\u0026thinsp;C1 (17%). Considering that p53 plays a crucial role in inducing cell cycle arrest(\u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e), it is reasonable that \u003cem\u003eTP53\u003c/em\u003e has the highest mutation rate in C3, the subtype with the highest enrichment of cell cycle pathways. \u003cem\u003eIDH1\u003c/em\u003e showed the highest mutation rate (48%) in C4, the subtype characterized by the highest enrichment of neuronal pathways. Again, the result is justified since \u003cem\u003eIDH1\u003c/em\u003e is frequently mutated in gliomas(\u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e).\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 \u003cdiv id=\"Sec16\" class=\"Section2\"\u003e \u003ch2\u003e3.5 The pan-cancer subtypes have different cancer-associated molecular and phenotypic features\u003c/h2\u003e \u003cp\u003eWe compared cancer-associated molecular and phenotypic features among these subtypes. We observed that the enrichment scores of the stemness and proliferation signatures followed the pattern: C3\u0026thinsp;\u0026gt;\u0026thinsp;C2\u0026thinsp;\u0026gt;\u0026thinsp;C1\u0026thinsp;\u0026gt;\u0026thinsp;C4 (\u003cem\u003eP\u003c/em\u003e\u0026thinsp;=\u0026thinsp;0) (Fig.\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003eA). It indicated that C3 had the strongest stemness and proliferation signatures, consistent with the highest enrichment of cell cycle pathways in this subtype. In contrast, the enrichment scores of the epithelial-mesenchymal transition (EMT) signature(\u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e) followed the pattern: C4\u0026thinsp;\u0026gt;\u0026thinsp;C2\u0026thinsp;\u0026gt;\u0026thinsp;C1\u0026thinsp;\u0026gt;\u0026thinsp;C3 (\u003cem\u003eP\u003c/em\u003e\u0026thinsp;=\u0026thinsp;0) (Fig.\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003eA), suggesting that C4 has the highest EMT activity. Several cancer progression-related signatures, including inflammation, invasion, and metastasis(\u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e), also displayed noticeably distinct enrichment levels across the subtypes: C2\u0026thinsp;\u0026gt;\u0026thinsp;C3\u0026thinsp;\u0026gt;\u0026thinsp;C1\u0026thinsp;\u0026gt;\u0026thinsp;C4 (\u003cem\u003eP\u003c/em\u003e\u0026thinsp;=\u0026thinsp;0) (Fig.\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003eB). This highlights the highest activity of inflammation, invasion, and metastasis in C2, aligning with the high enrichment of immue, developmental, and stromal pathways in this subtype.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec17\" class=\"Section2\"\u003e \u003ch2\u003e3.6 Pathway enrichment analysis reveals distinct functional patterns among the pan-cancer subtypes\u003c/h2\u003e \u003cp\u003eBy differential gene expression analysis, we identified marker genes for each subtype (Supplementary Table S3). Based on the marker genes, we identified biological processes significantly enriched in each subtype using the \u003cem\u003eclusterProfiler\u003c/em\u003e R package(\u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e). As expected, metabolism-related biological processes were predominantly enriched in C1, including peroxisome, retinol metabolism, chemical carcinogenesis - DNA adducts(\u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e), drug metabolism - cytochrome P450, metabolism of xenobiotics by cytochrome P450, tyrosine metabolism, complement and coagulation cascades(\u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e31\u003c/span\u003e), bile secretion, valine-leucine and isoleucine degradation, and steroid hormone biosynthesis (Fig.\u0026nbsp;\u003cspan refid=\"Fig7\" class=\"InternalRef\"\u003e7\u003c/span\u003eA). C2 was primarily enriched in biological processes related to cellular development, extracellular matrix, and immunity, including cytokine-cytokine receptor interaction, hematopoietic cell lineage, cell adhesion molecules, phagosome, viral protein interaction with cytokine and cytokine receptor, graft-versus-host disease, intestinal immune network for IgA production, allograft rejection, autoimmune thyroid disease, and type 1 diabetes mellitus (Fig.\u0026nbsp;\u003cspan refid=\"Fig7\" class=\"InternalRef\"\u003e7\u003c/span\u003eB). C3 overrepresented biological processes associated with cell cycle and DNA repair, such as cell cycle, DNA replication, oocyte meiosis, cellular senescence, motor protein, Fanconi anaemia pathway, homologous recombination, p53 signaling, base excision repair, and mismatch repair(\u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e32\u003c/span\u003e) (Fig.\u0026nbsp;\u003cspan refid=\"Fig7\" class=\"InternalRef\"\u003e7\u003c/span\u003eC), with cell cycle and DNA replication pathways showing the highest enrichment levels (Fig.\u0026nbsp;\u003cspan refid=\"Fig7\" class=\"InternalRef\"\u003e7\u003c/span\u003eC). Lastly, C4 mainly expressed neural pathways, including axon guidance, glutamatergic synapse, dopaminergic synapse, circadian entrainment, GABAergic synapse, and synaptic vesicle cycle (Fig.\u0026nbsp;\u003cspan refid=\"Fig7\" class=\"InternalRef\"\u003e7\u003c/span\u003eD). Taken together, these results demonstrate the rationality of our pan-cancer subtyping method.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec18\" class=\"Section2\"\u003e \u003ch2\u003e3.7 Proteins overexpressed in the pan‑cancer subtypes\u003c/h2\u003e \u003cp\u003eBased on the proteomics data(\u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e33\u003c/span\u003e) of TCGA pan-cancer, we analyzed the expressions of 210 proteins across the four subtypes (Supplementary Table\u0026nbsp;4). We observed that INPP4B(\u003cspan citationid=\"CR34\" class=\"CitationRef\"\u003e34\u003c/span\u003e), a crucial protein functioning in regulating the PI3K/AKT signaling pathway and energy metabolism and biosynthesis, exhibited high expression in C1, supporting that C1 shows a more pronounced metabolic profile than the other subtypes (Fig.\u0026nbsp;\u003cspan refid=\"Fig8\" class=\"InternalRef\"\u003e8\u003c/span\u003eA). Additionally, a range of proteins involved in regulating extracellular stromal, development, and immune pathways showed significant upregulation in C2, such as ANNEXIN1, CAVEOLIN1, LCK, NDRG1PT346, and VEGFR2 (Fig.\u0026nbsp;\u003cspan refid=\"Fig8\" class=\"InternalRef\"\u003e8\u003c/span\u003eB). For instance, LCK, a tyrosine kinase, is essential for T cell activation and the immune response(\u003cspan citationid=\"CR35\" class=\"CitationRef\"\u003e35\u003c/span\u003e). Expectedly, a number of proteins involved in cell cycle regulation and DNA repair showed significantly higher expression levels in C3, such as CASPASE7 CLEAVEDD198, CHK2, CYCLINB1, FOXM1, and PCNA (Fig.\u0026nbsp;\u003cspan refid=\"Fig8\" class=\"InternalRef\"\u003e8\u003c/span\u003eC). Of them, CYCLINB1 collaborates with CDK1 to transition cells from the G2 phase to mitosis(\u003cspan citationid=\"CR36\" class=\"CitationRef\"\u003e36\u003c/span\u003e).\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eFurthermore, C4 demonstrated significantly higher expression levels of several proteins, including BRAF, GSK3ALPHABETAPS21S9(\u003cspan citationid=\"CR37\" class=\"CitationRef\"\u003e37\u003c/span\u003e, \u003cspan citationid=\"CR38\" class=\"CitationRef\"\u003e38\u003c/span\u003e), GSK3PS9, MTORPS2448, and PTEN(\u003cspan citationid=\"CR39\" class=\"CitationRef\"\u003e39\u003c/span\u003e) (Fig.\u0026nbsp;\u003cspan refid=\"Fig8\" class=\"InternalRef\"\u003e8\u003c/span\u003eD). These proteins have been recognized as key molecules in brain tumors. For example, glycogen synthase kinase-3 (GSK3), a serine/threonine kinase, has been identified as a regulator of GBM cell survival(\u003cspan citationid=\"CR40\" class=\"CitationRef\"\u003e40\u003c/span\u003e, \u003cspan citationid=\"CR41\" class=\"CitationRef\"\u003e41\u003c/span\u003e). Phosphatase and tensin homolog (PTEN), a phosphatase in humans, has a role in impairing glioma stem cells(\u003cspan citationid=\"CR39\" class=\"CitationRef\"\u003e39\u003c/span\u003e).\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec19\" class=\"Section2\"\u003e \u003ch2\u003e3.8 The pathways\u0026rsquo; enrichment correlates with etiological and clinicopathologic parameters in pan‑cancer\u003c/h2\u003e \u003cp\u003eFor each tumor sample, we defined its enrichment level of a specific type of pathways as the average enrichment score of all pathways belonging to this type. As anticipated, the enrichment levels of metabolic, cell cycle, and neuronal pathways were most enriched in C1, C3, and C4, respectively, and the enrichment levels of immune, developmental, and stromal pathways were most enriched in C2 (Fig.\u0026nbsp;\u003cspan refid=\"Fig9\" class=\"InternalRef\"\u003e9\u003c/span\u003eA). The enrichment levels of metabolic pathways were significantly lower in late-stage than in early-stage cancers (\u003cem\u003eP\u003c/em\u003e\u0026thinsp;=\u0026thinsp;0.008). In contrast, the enrichment levels of cell cycle, neuronal, developmental, and stromal pathways were significantly higher in late-stage cancers (\u003cem\u003eP\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;0.01) (Fig.\u0026nbsp;\u003cspan refid=\"Fig9\" class=\"InternalRef\"\u003e9\u003c/span\u003eB). Moreover, low-grade tumors showed higher enrichment of metabolic, neuronal, developmental, and stromal pathways than high-grade tumors, while high-grade tumors displayed higher enrichment of cell cycle and immune pathways (\u003cem\u003eP\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;0.05) (Fig.\u0026nbsp;\u003cspan refid=\"Fig9\" class=\"InternalRef\"\u003e9\u003c/span\u003eC). We further analyzed correlations between the pathways\u0026rsquo; enrichment and several common cancer risk factors in pan-cancer, including aging, smoking, viral infection, and alcohol use. Interestingly, almost all these pathways showed positive correlations with ages (\u003cem\u003eP\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;0.0001), except neuronal pathways whose enrichment had a significant negative correlation with ages (\u003cem\u003eP\u003c/em\u003e\u0026thinsp;=\u0026thinsp;6.86 \u0026times; 10\u003csup\u003e\u0026minus;\u0026thinsp;109\u003c/sup\u003e, \u003cem\u003eρ\u003c/em\u003e = -0.24) (Fig.\u0026nbsp;\u003cspan refid=\"Fig9\" class=\"InternalRef\"\u003e9\u003c/span\u003eD). Moreover, almost all these pathways showed higher enrichment in smoking than in non-smoking patients, except metabolic pathways whose enrichment was higher in non-smoking patients (\u003cem\u003eP\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;0.01) (Fig.\u0026nbsp;\u003cspan refid=\"Fig9\" class=\"InternalRef\"\u003e9\u003c/span\u003eE). Viral infections are significant etiological factors responsible for several cancers(\u003cspan citationid=\"CR42\" class=\"CitationRef\"\u003e42\u003c/span\u003e), including Epstein\u0026ndash;Barr virus (EBV) for STAD and DLBC, hepatitis B virus (HBV) for LIHC, and human papillomavirus (HPV) for CESC and HNSC. Of note, virus-infected cancers consistently displayed higher enrichment of cell cycle pathways than non-infected cancers for EBV, HBV, and HPV (\u003cem\u003eP\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;0.01) (Fig.\u0026nbsp;\u003cspan refid=\"Fig9\" class=\"InternalRef\"\u003e9\u003c/span\u003eF). In addition, immune pathways were more enriched in EBV-positive than in EBV-negative cancers (\u003cem\u003eP\u003c/em\u003e\u0026thinsp;=\u0026thinsp;1.79 \u0026times; 10\u003csup\u003e\u0026minus;\u0026thinsp;8\u003c/sup\u003e). In contrast, the enrichment levels of metabolic, neuronal, developmental, and stromal pathways were likely higher in virus-negative than in virus-positive cancers (Fig.\u0026nbsp;\u003cspan refid=\"Fig9\" class=\"InternalRef\"\u003e9\u003c/span\u003eG). Interestingly, the patients with alcohol use history displayed significantly higher enrichment of cell cycle pathways while lower enrichment of metabolic pathways than those without alcohol use history (\u003cem\u003eP\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;0.05) (Fig.\u0026nbsp;\u003cspan refid=\"Fig9\" class=\"InternalRef\"\u003e9\u003c/span\u003eH). Strikingly, all these pathways exhibited noticeably distinct enrichment levels between male and female patients, with metabolic, cell cycle, and immune pathways being more enriched in male patients while neuronal, developmental, and stromal pathways more enriched in female patients (Fig.\u0026nbsp;\u003cspan refid=\"Fig9\" class=\"InternalRef\"\u003e9\u003c/span\u003eI). Hypoxia is a typical property of cancer(\u003cspan citationid=\"CR43\" class=\"CitationRef\"\u003e43\u003c/span\u003e). We found hypoxia scores to be significantly and positively correlated with cell cycle and immune pathways\u0026rsquo; enrichment and negatively correlated with neuronal pathways\u0026rsquo; enrichment (\u003cem\u003eρ\u003c/em\u003e\u0026thinsp;=\u0026thinsp;0.62, 0.21, and \u0026minus;\u0026thinsp;0.44 for cell cycle, immune, and neuronal pathways, respectively) (Fig.\u0026nbsp;\u003cspan refid=\"Fig9\" class=\"InternalRef\"\u003e9\u003c/span\u003eJ). In addition, we observed that cell cycle, neuronal, developmental, and stromal pathways were more enriched in the patients receiving radiation therapy than those not, while metabolic and immune pathways exhibited an opposite trend (Fig.\u0026nbsp;\u003cspan refid=\"Fig9\" class=\"InternalRef\"\u003e9\u003c/span\u003eK). Taken together, these analyses reveal significant and distinct associations between these pathways\u0026rsquo; activities and cancer-associated etiological and clinicopathologic factors.\u003c/p\u003e \u003cp\u003e \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"},{"header":"4 Discussion","content":"\u003cp\u003ePathway-based clustering analyses are advantageous in uncovering cancer subtypes and exploring tumor heterogeneity. First of all, pathway (or gene set) scores can enhance the stability of clustering results that otherwise could be degraded due to outliers of individual genes\u0026rsquo; expressions. Second, pathway-based clustering results are easier to elucidate based on specific pathways\u0026rsquo; enrichment levels. However, most previous pathway-based clustering methods need to claim specific pathways prior to clustering analysis, as may result in subjectivity and selectivity. To overcome the limitation, here we proposed a novel pathway-based clustering method to discover cancer subtypes. Unlike previous methods, this method first detects gene clusters based on gene expression correlations matrix and identifies overrepresented pathways associated with these gene clusters. Based on the enrichment scores of the overrepresented pathways, this method reveals cancer subtypes by clustering analysis.\u003c/p\u003e \u003cp\u003eApplying this method to TCGA pan-cancer, we identified four cancer subtypes: C1, C2, C3, and C4. C1 was characterized by high metabolic activity, the best OS, DSS, and PFI outcomes, the highest response rate to chemotherapy, and the lowest \u003cem\u003eTP53\u003c/em\u003e mutation rate. C2 was characterized by high activities of immune, developmental, and stromal pathways, the lowest tumor purity and ITH. C3 overexpressed cell cycle and DNA repair pathways, had the worst DFI outcome, the highest degree of genomic instability, the highest \u003cem\u003eTP53\u003c/em\u003e mutation rate, and the strongest stemness and proliferation signatures. C4 overrepresented neuronal pathways, with the worst DSS and PFI outcomes, the lowest immune activity, the highest tumor purity, the lowest degree of genomic instability and DNA repair activity, the highest ITH, the weakest stemness and proliferation signatures, the highest \u003cem\u003eIDH1\u003c/em\u003e mutation rate while the lowest \u003cem\u003eTP53\u003c/em\u003e mutation rate, and the lowest response rate to chemotherapy (Fig.\u0026nbsp;\u003cspan refid=\"Fig10\" class=\"InternalRef\"\u003e10\u003c/span\u003e). We further revealed that individual cancer types exhibited significantly distinct distribution across the four subtypes. C1 harbored most adrenocortical, liver, prostate, and uveal melanoma tumors, suggesting these cancer types to be enriched in metabolic pathways. Most kidney renal clear cell carcinoma and pancreatic adenocarcinoma tumors were grouped into C2; this is reasonable since renal cell carcinoma is a highly immunogenic tumor(\u003cspan citationid=\"CR44\" class=\"CitationRef\"\u003e44\u003c/span\u003e) and pancreatic adenocarcinoma is a stroma-enriched tumor(\u003cspan citationid=\"CR45\" class=\"CitationRef\"\u003e45\u003c/span\u003e). C3 included most gastrointestinal cancers, highlighting a high cell cycle activity and genomic instability in this class of cancers. Lastly, C4 was exclusively dominated by neural system-related tumors, most of which were gliomas, aligning with this subtype characterized by neuronal pathway enrichment.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eFor the first time, we comprehensively explored associations between multiple pathways and various etiological and clinicopathologic parameters in pan‑cancer. Ages showed significant positive correlations with the activities of most classes of pathways but a negative correlation with the activity of neuronal pathways. It is imaginable that neuronal activity reduces with increasing age. Our analysis indicates that smoking may enhance the activity of various oncogenic pathways, such as cell cycle, stromal, and developmental pathways, while it could inhibit the activity of metabolic pathways. Our data suggest that viral infections may increase the activity of cell cycle and immune pathways. It is consistent with previous findings that viral infections promote genomic instability and anti-tumor immune responses(\u003cspan citationid=\"CR46\" class=\"CitationRef\"\u003e46\u003c/span\u003e). In addition, alcohol use may increase the activity of cell cycle pathways but decrease the activity of metabolic pathways. Our analyses also suggest that the differences in pathways activity exist between male and female cancer patients. Therapies, such as radiation therapy, could have an impact on the activities of various pathways.\u003c/p\u003e \u003cp\u003eA major limitation of the present work is that the pan-cancer cohort encompasses 33 cancer types that involve different size of samples. As a result, the cancer types with large sample size, such as breast, lung and colorectal cancer, would have exerted greater impact on the results. Another limitation is that our findings of the significant associations between pathway activities and etiological and clinicopathologic features, albeit interesting, need to be verified by additional data.\u003c/p\u003e \u003cp\u003eIn conclusion, we developed a novel clustering algorithm to unveil a new classification of pan-cancer based on metabolic, immune, stromal, developmental, cell cycle, and neuronal pathways. We identified four pan-cancer subtypes distinguished by tumor microenvironment, genome instability, DNA repair activity, ITH, oncogenic signatures, somatic mutation and CNA profiles, protein expression profiles, response to chemotherapy, and clinical outcomes. The identification of pan-cancer subtypes provides novel insights into tumor biology and has potential clinical implications for cancer management.\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003cstrong\u003eEthics approval and consent to participate\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eNot applicable.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eConsent for publication\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eNot applicable.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAvailability of data and materials\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eAll data associated with this study are available within the paper and its supplementary data. The algorithm and analysis workflow code have also been uploaded to GitHub and can be accessed via the following link: https://github.com/WangX-Lab/PathClustNet. This repository includes the complete code and instructions necessary to reproduce the results presented in this study.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eCompeting interests\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe authors declare that they have no competing interests.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFunding\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThis work was supported by the China Pharmaceutical University (grant number\u0026nbsp;3150120001\u0026nbsp;to XW).\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAcknowledgments\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eNot applicable.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAuthor Contributions\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eMengli Xu\u003c/strong\u003e:\u0026nbsp;Software, Formal analysis, Investigation, Data curation, Visualization, Writing - original draft. \u003cstrong\u003eHongjing Ai\u003c/strong\u003e: Formal analysis, Investigation, Visualization. \u003cstrong\u003eDanni Wang\u003c/strong\u003e: Investigation. \u003cstrong\u003eXiaosheng Wang\u003c/strong\u003e: Conceptualization, Methodology, Resources, Investigation, Writing - original draft, Supervision, Project administration.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\u003cli\u003e\u003cspan\u003eCaswell DR, Swanton C (2017) The role of tumour heterogeneity and clonal cooperativity in metastasis, immune evasion and clinical outcome. BMC Med 15(1):133\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLi M, Zhang Z, Li L et al (2020) An algorithm to quantify intratumor heterogeneity based on alterations of gene expression profiles. Commun Biol 3(1):505\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePeng X, Chen Z, Farshidfar F et al (2018) Molecular Characterization and Clinical Relevance of Metabolic Expression Subtypes in Human Cancers. Cell Rep 23(1):255\u0026ndash;269\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eThorsson V, Gibbs DL, Brown SD et al (2018) The Immune Landscape of Cancer. Immunity 48(4):812\u0026ndash;830\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLi L, Wang X (2021) Identification of gastric cancer subtypes based on pathway clustering. npj Precis Onc 5(1):46\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHe Y, Jiang Z, Chen C et al (2018) Classification of triple-negative breast cancers based on Immunogenomic profiling. J Exp Clin Cancer Res 37(1):327\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLei J, Luo J, Liu Q et al (2024) Identifying cancer subtypes based on embryonic and hematopoietic stem cell signatures in pan-cancer. Cell Oncol 47(2):587\u0026ndash;605\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMonti S, Tamayo P, Mesirov J et al (2003) Consensus Clustering: A Resampling-Based Method for Class Discovery and Visualization of Gene Expression Microarray Data. Mach Learn 52(1):91\u0026ndash;118\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBri\u0026egrave;re G, Darbo \u0026Eacute;, Th\u0026eacute;bault P et al (2021) Consensus clustering applied to multi-omics disease subtyping. BMC Bioinformatics 22(1):361\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGuess MJ, Wilson SB (2002) Introduction to hierarchical clustering. J Clin Neurophysiol 19(2):144\u0026ndash;151\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eH\u0026auml;nzelmann S, Castelo R, Guinney J (2013) GSVA: gene set variation analysis for microarray and RNA-seq data. BMC Bioinformatics 14:7\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBland JM, Altman DG (1998) Survival probabilities (the Kaplan-Meier method). BMJ 317(7172):1572\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBurrell RA, McGranahan N, Bartek J et al (2013) The causes and consequences of genetic heterogeneity in cancer evolution. Nature 501(7467):338\u0026ndash;345\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMermel CH, Schumacher SE, Hill B et al (2011) GISTIC2.0 facilitates sensitive and confident localization of the targets of focal somatic copy-number alteration in human cancers. Genome Biol 12:R41\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eYoshihara K, Shahmoradgoli M, Mart\u0026iacute;nez E et al (2013) Inferring tumour purity and stromal and immune cell admixture from expression data. Nat Commun 4:2612\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eReimand J, Arak T, Adler P et al (2016) g:Profiler-a web server for functional interpretation of gene lists (2016 update). Nucleic Acids Res 44(W1):W83\u0026ndash;89\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGene Ontology Consortium (2015) Gene Ontology Consortium: going forward. Nucleic Acids Res 43(Database issue):D1049\u0026ndash;1056\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMerico D, Isserlin R, Stueker O et al (2010) Enrichment map: a network-based method for gene-set enrichment visualization and interpretation. PLoS ONE 5(11):e13984\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eVrbin CM (2022) Parametric or nonparametric statistical tests: Considerations when choosing the most appropriate option for your data. Cytopathology 33(6):663\u0026ndash;667\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLiu Q, Wang L (2021) t-Test and ANOVA for data with ceiling and/or floor effects. Behav Res Methods 53(1):264\u0026ndash;277\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePripp AH (2018) Pearson\u0026rsquo;s or Spearman\u0026rsquo;s correlation coefficients. Tidsskr Nor Laegeforen. ;138(8)\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBenjamini Y, Hochberg Y (1995) Controlling the False Discovery Rate: A Practical and Powerful Approach to Multiple Testing. J Roy Stat Soc: Ser B (Methodol) 57(1):289\u0026ndash;300\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eNegrini S, Gorgoulis VG, Halazonetis TD (2010) Genomic instability\u0026ndash;an evolving hallmark of cancer. Nat Rev Mol Cell Biol 11(3):220\u0026ndash;228\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGroelly FJ, Fawkes M, Dagg RA et al (2023) Targeting DNA damage response pathways in cancer. Nat Rev Cancer 23(2):78\u0026ndash;94\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eChen J (2016) The Cell-Cycle Arrest and Apoptotic Functions of p53 in Tumor Initiation and Progression. Cold Spring Harb Perspect Med 6(3):a026104\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eYan H, Parsons DW, Jin G et al (2009) IDH1 and IDH2 mutations in gliomas. N Engl J Med 360(8):765\u0026ndash;773\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eCraene BD, Berx G (2013) Regulatory networks defining EMT during cancer initiation and progression. Nat Rev Cancer 13(2):97\u0026ndash;110\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGrivennikov SI, Greten FR, Karin M (2010) Immunity, Inflammation, and Cancer. Cell 140(6):883\u0026ndash;899\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eYu G, Wang LG, Han Y et al (2012) clusterProfiler: an R Package for Comparing Biological Themes Among Gene Clusters. OMICS 16(5):284\u0026ndash;287\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGuengerich FP (2000) Metabolism of chemical carcinogens. Carcinogenesis 21(3):345\u0026ndash;351\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eCopenhaver M, Yu CY, Hoffman RP (2019) Complement Components, C3 and C4, and the Metabolic Syndrome. Curr Diabetes Rev 15(1):44\u0026ndash;48\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMARINUS MG (2012) DNA Mismatch Repair. EcoSal Plus. ;5(1)\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZhang Y, Chen F, Chandrashekar DS et al (2022) Proteogenomic characterization of 2002 human cancers reveals pan-cancer molecular subtypes and associated pathways. Nat Commun 13(1):2669\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZhang M, Ceyhan Y, Kaftanovskaya EM et al (2021) INPP4B protects from metabolic syndrome and associated disorders. Commun Biol 4(1):1\u0026ndash;15\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePalacios EH, Weiss A (2004) Function of the Src-family kinases, Lck and Fyn, in T-cell development and activation. Oncogene 23(48):7990\u0026ndash;8000\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePorter L, Donoghue D (2003) Cyclin B1 and CDK1: Nuclear localization and upstream regulators. Prog Cell Cycle Res 5:335\u0026ndash;347\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHetman M, Cavanaugh JE, Kimelman D et al (2000) Role of Glycogen Synthase Kinase-3β in Neuronal Apoptosis Induced by Trophic Withdrawal. J Neurosci 20(7):2567\u0026ndash;2574\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHur EM, Zhou FQ (2010) GSK3 signalling in neural development. Nat Rev Neurosci 11(8):539\u0026ndash;551\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBenitez JA, Ma J, D\u0026rsquo;Antonio M et al (2017) PTEN regulates glioblastoma oncogenesis through chromatin-associated complexes of DAXX and histone H3.3. Nat Commun 8(1):15223\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBruning-Richardson A, Droop A, Tams D et al (2018) Identification of transcriptional targets of GSK3 involved in glioblastoma invasion. Neuro Oncol 20(Suppl 1):i26\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKotliarova S, Pastorino S, Kovell LC et al (2008) Glycogen Synthase Kinase 3 inhibition Induces Glioma Cell Death through c-MYC, NF-κB and Glucose Regulation. Cancer Res 68(16):6643\u0026ndash;6651\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMorris JDH, Eddleston ALWF, Crook T (1995) Viral infection and cancer. Lancet 346(8977):754\u0026ndash;758\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eRuan K, Song G, Ouyang G (2009) Role of hypoxia in the hallmarks of human cancer. J Cell Biochem 107(6):1053\u0026ndash;1062\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ede Vivar Chevez AR, Finke J, Bukowski R (2014) The role of inflammation in kidney cancer. Adv Exp Med Biol 816:197\u0026ndash;234\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHosein AN, Brekken RA, Maitra A (2020) Pancreatic cancer stroma: an update on therapeutic targeting strategies. Nat Rev Gastroenterol Hepatol 17(8):487\u0026ndash;505\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWu Z, Dong Z, Luo J et al (2024) A comprehensive comparison of molecular and phenotypic profiles between hepatitis B virus (HBV)-infected and non-HBV-infected hepatocellular carcinoma by multi-omics analysis. Genomics 116(3):110831\u003c/span\u003e\u003c/li\u003e\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":true,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"functional-and-integrative-genomics","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"fige","sideBox":"Learn more about [Functional \u0026 Integrative Genomics](http://link.springer.com/journal/10142)","snPcode":"10142","submissionUrl":"https://submission.nature.com/new-submission/10142/3","title":"Functional \u0026 Integrative Genomics","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"Springer Hybrid","inReviewEnabled":true,"inReviewRevisionsEnabled":false},"keywords":"Pan-cancer, Subtyping, Gene clusters, Pathway enrichment analysis, PathClustNet Algorithm, Tumor heterogeneity","lastPublishedDoi":"10.21203/rs.3.rs-5114169/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-5114169/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003ch2\u003eBackground\u003c/h2\u003e \u003cp\u003ePathways-based clustering methods have been proposed to explore tumor heterogeneity. However, such methods are currently disadvantageous in that specific pathways need to be explicitly claimed.\u003c/p\u003e\u003ch2\u003eMethods\u003c/h2\u003e \u003cp\u003eWe developed the PathClustNet algorithm, a pathway-based clustering method to identify cancer subtypes. This method first detects gene clusters and identifies overrepresented pathways associated with these gene clusters. Based on the enrichment scores of the pathways, reveals cancer subtypes by clustering analysis. We applied the method to TCGA pan-cancer to identify cancer subtypes.\u003c/p\u003e\u003ch2\u003eResults\u003c/h2\u003e \u003cp\u003eWe identified four pan-cancer subtypes, termed C1, C2, C3 and C4. C1 exhibited high metabolic activity, favourable survival and the lowest \u003cem\u003eTP53\u003c/em\u003e mutation rate. C2 had high immune, developmental, and stromal pathway activities, the lowest tumor purity and intratumor heterogeneity. C3 overexpressed cell cycle and DNA repair pathways were most genomically instable and had the highest \u003cem\u003eTP53\u003c/em\u003e mutation rate. C4 overrepresented neuronal pathways, with the lowest response rate to chemotherapy, and the highest tumor purity and genomic stability. Furthermore, ages showed positive correlations with most pathways but a negative correlation with neuronal pathways. Smoking, Viral infections and alcohol use variously affect neuron, cell cycle, immune, stromal, developmental, and metabolic pathway activities.\u003c/p\u003e\u003ch2\u003eConclusion\u003c/h2\u003e \u003cp\u003eThe novel PathClustNet clustering algorithm unveils a new classification of pan-cancer based on metabolic, immune, stromal, developmental, cell cycle, and neuronal pathways. These subtypes display different molecular and clinical features to warrant the investigation of precision oncology.\u003c/p\u003e","manuscriptTitle":"Gene clusters-based pathway enrichment analysis identifies four pan-cancer subtypes with distinct molecular and clinical features","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2024-11-04 13:36:14","doi":"10.21203/rs.3.rs-5114169/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"decision","content":"Revision requested","date":"2024-10-24T00:48:20+00:00","index":"","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2024-10-23T07:23:22+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"338205683534859791310304656499343750352","date":"2024-10-23T07:20:52+00:00","index":"hide","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2024-10-23T02:33:36+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"49184197898608561640215625532385822573","date":"2024-10-23T02:16:39+00:00","index":"hide","fulltext":""},{"type":"reviewersInvited","content":"","date":"2024-10-03T07:04:14+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2024-09-24T11:34:10+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2024-09-24T11:34:01+00:00","index":"","fulltext":""},{"type":"submitted","content":"Functional \u0026 Integrative Genomics","date":"2024-09-19T05:52:11+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"functional-and-integrative-genomics","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"fige","sideBox":"Learn more about [Functional \u0026 Integrative Genomics](http://link.springer.com/journal/10142)","snPcode":"10142","submissionUrl":"https://submission.nature.com/new-submission/10142/3","title":"Functional \u0026 Integrative Genomics","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"Springer Hybrid","inReviewEnabled":true,"inReviewRevisionsEnabled":false}}],"origin":"","ownerIdentity":"034dccee-8117-4c91-bff7-010ebac571a8","owner":[],"postedDate":"November 4th, 2024","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"published-in-journal","subjectAreas":[],"tags":[],"updatedAt":"2024-12-02T17:24:33+00:00","versionOfRecord":{"articleIdentity":"rs-5114169","link":"https://doi.org/10.1007/s10142-024-01501-0","journal":{"identity":"functional-and-integrative-genomics","isVorOnly":false,"title":"Functional \u0026 Integrative Genomics"},"publishedOn":"2024-11-28 15:57:17","publishedOnDateReadable":"November 28th, 2024"},"versionCreatedAt":"2024-11-04 13:36:14","video":"","vorDoi":"10.1007/s10142-024-01501-0","vorDoiUrl":"https://doi.org/10.1007/s10142-024-01501-0","workflowStages":[]},"version":"v1","identity":"rs-5114169","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-5114169","identity":"rs-5114169","version":["v1"]},"buildId":"8U1c8b4HqxoKbykW_rLl7","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}