Multiple machine learning algorithms identified SLC6A8 as a diagnostic biomarker of the late stage of Hepatocellular carcinoma

Multiple machine learning algorithms identified SLC6A8 as a diagnostic biomarker of the late stage of Hepatocellular carcinoma | 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 Multiple machine learning algorithms identified SLC6A8 as a diagnostic biomarker of the late stage of Hepatocellular carcinoma Linlin Song, Hongli Zhang, Wang Yang This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-4570554/v1 This work is licensed under a CC BY 4.0 License Status: Under Review Version 1 posted 9 You are reading this latest preprint version Abstract Hepatocellular carcinoma (HCC) is a chronic liver disease characterized by persistent tumor growth, contributing significantly to mortality rates worldwide and presenting a growing global health concern. Consequently, there is an urgent need to develop effective diagnostic and treatment strategies for HCC. This study aims to identify crucial genes for early HCC diagnosis to mitigate disease progression and to investigate differences in immune cell infiltration between early-stage and late-stage HCC. We integrated two published datasets for a comprehensive analysis, identifying 575 DEGs subjected to GSEA to reveal pathways distinguishing early-stage from late-stage HCC. Notably, the gene SLC6A8 emerged as a potential diagnostic biomarker for late-stage HCC through LASSO, SVM-RFE and RF-Boruta analyses. ROC curves for SLC6A8 were utilized to evaluate diagnostic accuracy. The ImmuCellAI algorithm assessed immune cell composition differences between early and late-stage HCC, revealing that SLC6A8 expression positively correlates with resting Tfh cells and Th2, while negatively correlating with B cells, indicating its association with immune cell infiltration patterns. To strengthen our results, we further analyzed SLC6A8 expression using single-cell transcriptome data, confirming significant upregulation in late-stage HCC, particularly in key liver cell types such as Hepatocyte cells. Overall, our study identifies SLC6A8 as a potential marker gene that enhances understanding of HCC diagnosis and therapeutic strategies. Hepatocellular carcinoma Multiple machine learning algorithms SLC6A8 Diagnostic biomarker Bioinformatics Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Figure 6 Figure 7 Introduction Hepatocellular carcinoma (HCC), a fourth leading cause of death over the world in 2018, has been a globally severe disease [ 1 ]. Most patients are diagnosed at a late stage, and the only potentially curative approaches for treating are hepatic resection and liver transplantation, which have low treatment rate [ 15 ]. So, it is critical to find the diagnostic biomarker of HCC as early as possible. However, rather than focusing on the severity of HCC, many researchers have tried to build a classification to distinguish between normal and HCC patients [9; 25; 35]. In Addition, tumor immune microenvironment plays an important role in tumor progression and treatment outcome. In recent years, immunotherapy has become a hopeful approach for treating cancer. However, response rates among HCC patients have been limited. Therefore, conducting a comprehensive analysis of different types of immune cells within tumors could aid in identifying new biomarkers for prognosis and therapeutic efficacy in HCC patients. In recent decades, HCC has been a major focus of research interest. For instance, specific genes like CLTA , TALDO1 and CSTB , identified as gene signature through single cell RNA sequencing, have been associated with survival outcome and immunotherapy response [ 16 ]. Hang D et al. conducted mass spectrometry analysis for metabolomics to identify new predictive biomarkers and pathways in HCC [ 7 ]. Genes such as Receptor Activity Modifying Protein 3 ( RAMP3 ) and CD68 Molecule ( CD68 ) have shown significantly expressed in HCC based on machine-learning algorithms [ 13 ]. Overall, these genes are considered vital candidates for potential diagnostic markers in HCC clinical settings. Recently, the rapid development of next-generation sequencing technologies provides amount of RNA sequencing data of HCC. However, RNA sequencing is typically conducted in "bulk," capturing average gene expression patterns from a multitude of cells [ 20 ]. Notably, single-cell RNA sequencing single-cell RNA sequencing is an innovative sequencing technique that offers valuable insights into the characterization of individual immune cells or tumor cells [ 30 ]. At the same time, bioinformatic analysis and machine learning have emerged as increasingly promising strategies for comprehensive and in-depth analysis of large datasets, such as transcriptome sequences, and interdisciplinary collaborations have been instrumental in advancing clinical therapeutic methods [ 2 ]. The advent of modern computer-assisted medical science has provided significant guidance and hope for previously untreatable diseases, such as utilizing the XGBoost algorithm for HCC diagnosis [ 13 ]. To meet the demand for early diagnosis, numerous efforts have focused on developing new methods based on deep learning analysis [ 33 ]. Clearly, achieving accurate clinical diagnosis of COPD remains a critical and imperative pursuit. In this study, we utilized established bioinformatic tools to screen for potential biomarkers indicating the late stage of HCC. Two transcriptome datasets were selected from the published database (TCGA, ICGC) for analysis. After identifying DEGs (Differential expressed genes) using limma, we conducted GO and GSEA analyses on these DEGs. Notably, we identified a candidate diagnostic biomarker, SLC6A8 , which was the intersection of genes selected by LASSO, SVM-REF and RF-Boruta methods and validated it using another dataset. Additionally, we investigated changes in immune cell composition between early-stage and late-stage HCC samples using ImmuCellAI (Immune Cell Abundance Identifier) to analyze immune cell infiltration. Furthermore, we explored the relationship between immune cell infiltration and the potential diagnostic biomarker of HCC, validating our findings with single-cell transcriptomic data and in vitro trial. Overall, we identified a previously unrecognized gene with the potential to guide future clinical treatment and diagnosis of HCC patients. Materials and methods Data collection and download Standardized RNA-Seq reads (Release 28) of LIHC-US and LIRI-JP projects were obtained from The Cancer Genome Atlas (TCGA, https://cancergenome.nih.gov/ ) and International Cancer Genome Collaboration (ICGC, https://dcc.icgc.org/ ). GSE14520, the validation dataset with chip data was downloaded from GEO database ( https://www.ncbi.nlm.nih.gov/geo/ ). Only patients with tumor stage and samples from the primary site were included in the analysis. For LIHC-US (TCGA), 231 patients were retained. For LIRI-JP (ICGC), 344 patients were met the criterion. Finally, we extracted 218 patients from the GSE14520. The HCC scRNA-seq dataset GSE149614 was load from the GEO database ( https://www.ncbi.nlm.nih.gov/geo/ ) and included 10 patients, only the primary tumor samples including 3 TNM stage I, 1 TNM stage II, 2 TNM stage III and 4 TNM stage IV samples will be taken for downstream analysis. According to the TNM, HCC disease was divided into four stages: stage I, stage II, stage III, and stage IV. In the following analysis, we defined stage I and stage II as HCC early-stage, stage III and stage IV as HCC late-stage. A merge data cohort comprising of LIHC-US and LIRI-JP datasets through “limma” and “snm” R packages, and the GSE14520 was as the validation dataset to confirm the analysis results. The GSE149614, an HCC scRNA-seq dataset, comprising of 4 HCC early-stage and 6 HCC early-stage patients was also taken as the validation dataset to investigate the results at single-cell level. The clinical information of these cohorts involved in this study had presented in Supplemental table 1 . Data Integration A merged dataset cohort comprising LIHC-US and LIRI-JP dataset through R package “limma” (version 3.58.11) and “SNM” (version 1.50.0), correcting the batch effect by VOOM and SNM function in Supplemental Fig. 1D [18; 27]. What’s more, the results of other approaches used to remove the batch effect in the different dataset had showed in Supplemental Fig. 1A-C . Principal component analysis (PCA) shows the samples distribution before and after batch correction. Differential Expression Genes DEGs form the merged dataset were with the cutoff criteria of |logFC| > 0.5, adjusted-pvalue < 0.05 by using “limma” (version 3.58.11) R package. A linear model was then fitted with lmFit and differential expression was assessed using the eBayes function. Functional enrichment analysis Gene ontology (GO) enrichment analysis was performed to investigate the DEG biological significant by using the clusterProfiler (version 4.10.1) R package [ 34 ]. The Molecular Signatures Database (MSigDB) Hallmark Gene Sets and Kyoto Encyclopedia of Genes and Genomes (KEGG) Pathway were used for pathway enrichment analysis by using msigdbr (version 7.5.1) and massdatabase (version 1.0.10) R package. GSVA analysis was performed using the GSVA (version 1.50.1) R package [ 8 ]. Both GSEA and GSVA were performed with the criteria of P - value < 0.05. Immune cell infiltration The ImmuCellAI, a gene set signature-based method, is a deconvolution algorithm, which can estimate the abundance of 24 immune cell types [ 19 ]. A single sample gene set enrichment analysis (ssGSEA) algorithm was applied to deconvolve bulk gene expression matrix. Immune cell infiltration (24 various cell types including 18 T-cell subsets) was precisely quantified in each gene expression profile. The detailed process was carried out by using ImmuneCellAI (version 0.1.0) R package. Potential biomarkers selection To find the potential prognostic gene biomarkers from the DEGs, we screened them by three machine-learning algorithms. 1) least absolute shrinkage and selection operator with logistic regression (LASSO-LR), 2) support vector machine-recursive feature elimination (SVM-RFE) algorithms, 3) Boruta with random forest (Brouta-RF). LASSO can identify genes significantly associated with different samples, which represents a regression analysis algorithm that applies regularization for variable selection using “glmnet” (version 4.1.7) R package [ 4 ]. In this study, LASSO with logistic regression first reduced the dimensionality successfully (425 original genes to 20 meaningful genes). Specifically, LASSO method with crossing validation using the mean of square error as cost function was performed, then they shrank into a few more important features according to mean of square error. SVM-RFE represents a widely used supervised machine-learning protocol for classification and regression, used to find the best variables by deleting SVM-generated eigenvectors. The “Caret” (version 6.0.94) R package via grid search method is employed to select hyperparameters for all classifiers using 10-fold cross-validation for the training dataset. SVM-RFE can identify the diagnostic value of biomarkers with higher discriminative power using the “e1071” (version 1.7.13) R package. The Boruta algorithm is a wrapper built around the random forest classification algorithm. The Boruta (version 8.0.0) R package is to capture all the important, interesting features you might have in the dataset with respect to an outcome variable. The “Caret” (version 6.0.94) R package via grid search method is employed to select hyperparameters of random forest for all classifiers using 10-fold cross-validation for the training dataset. Brouta-RF can identify the diagnostic value of biomarkers with higher discriminative power using the “randomforest” (version 4.7–1.1) R package. Ultimately, we combined the overlapping genes among LASSO-LR, SVM-RFE and Brouta-RF algorithm for further analysis. A two-sided P value < 0.05 between the early stage and the late stage of HCC was statistically significant on these genes. Then, we validated their expression level to estimate their ability to be candidate diagnostic biomarkers in the GSE14520 data. ROC of diagnostic biomarker The nonparametric Wilcoxon rank-sum test was used to perform inter-group comparisons of continuous variables. The degree of efficacy of each diagnostic biomarker was assessed using receiver operating characteristic (ROC) curves by “pROC” (version 1.18.5) and “multipleROC” (version 0.1.1) R package, which is the gold standard to prove the diagnostic accuracy and test the efficacy of diagnostic biomarkers in the GSE14520 cohort. Single cell transcriptome data processing and analyzing GSE149614 raw data was downloaded from GEO databases. In the process of single cell transcriptome data processing, we carried out normalization, scaling, clustering of cells and achieved 6 main cell types using Seurat (version 5.0.3) R package. Single cells were extracted with the criteria of nFeature_RNA > 250, percent.mt 0.8, nCount_RNA > 500 to removing doublet and dead cells by Seurat R package (version 5.0.3). Then we normalized the filtered gene-barcode matrices using “LogNormalize” method with the “NormalizeData” function. The top 2000 highly variable genes were found by the “FindVariableFeatures” function using the “vst” method which were centered and scaled using “ScaleData” before. Then we performed principal component analysis (PCA) based on these 2000 highly variable genes with the intention of dimensionality reduction, then dimensionality-reduced clusters were showed on the 2D map produced with the t-distributed t-SNE using function “FindNeighbors”, “FindClusters” and “runTSNE” from Seurat. Kruskal–Wallis test was used to estimate the difference of gene expression level. Real-time quantitative reverse transcriptase-PCR Total RNA was extracted from liver cancer tissue using Trizol (Invitrogen, Carlsbad, America) reagent before cDNA was obtained using the Trans-Script All in-one First-strand cDNA Synthesis Supermix for qPCR Kit (TransGen Biotech, Beijing, China). Real-time quantitative PCR was performed in Step One ABI real-time. PCR System through SYBR Green Master (Roche, Basel, Switzerland). GAPDH, Forward primer: 5’-CATGTTCGTCATGGGTGTGAA-3’, Reverse primer:5’-GGCATGGACTGTGGTCATGAG-3’. SLC6A8, Forward primer: 5’-GGCAGCTACAACCGCTTCAACA-3’, Reverse primer:5’-CAGGATGGAGAAGACCACGAAG-3’. Statistical analysis R (version 4.3.3) and Rstudio (version 2023.12.1 + 402) were used for statistical analysis. Wilcox-rank-sum test was carried out to analyze the significant differences of features between the early-stage and late-stage groups. Spearman correlation coefficient was used to identify the correlations between genes and immune cells. A P -value < 0.05 indicated statistical significance (* P < 0.05; ** P < 0.01; *** P < 0.001). Results Identification of DEGs in the HCC early stage and late stage We performed differential expression analysis between 397 early-stage HCC samples and 178 late-stage HCC samples in the merge cohort (LIHC-US and LIRI-JP) by utilizing the “limma” R package. Figure 1 showed the schematic workflow of this study. 137 genes were significantly enriched in the early-stage group and 291 genes significantly enriched in the late-stage group (Fig. 2A, |logFC| > 0.5 and AdjustedPvalue < 0.05). The heatmap of DEGs also had been presented in Fig. 2B . Functional analysis of DEGs by GO and KEGG enrichment analysis To investigate the potential biological significances of DEG, we used clusterProfiler and GSVA R package to characterize the GO and KEGG pathway. The GO analysis showed that EEGs involved with response to xenobiotic stimulus, steroid metabolic process, cytoplasmic vesicle lumen, high − density lipoprotein particle, endopeptidase regulator activity, steroid hydroxylase activity, and so on (Fig. 3A). Moreover, p53 signaling pathway, ECM − receptor interaction, ABC transporters, Steroid hormone biosynthesis, and PPAR signaling pathway were identified by GSEA (Fig. 3B). In particular, nitrogen metabolism, one carbon pool by folate, taurine and hypotaurine metabolism and tryptophan metabolism were mainly enriched in the HCC early stage (Fig. 3C). Inversely, cell cycle, mismatch repair, p53 signaling pathway, and sphingolipid metabolism were significantly enriched in the HCC late stage (Fig. 3D). Additionally, GSVA results showed ABC transporters was enriched in the HCC late stage ( Supplemental Fig. 2 ). Significant changes between two stages in immune cells by ImmuneCellAI With the ImmuCellAI, a latest immune cell infiltration algorithm, we obtained the immune cell expression matrix from the whole gene expression matrix of the merge dataset (Fig. 4A). Not only Tfh cells and Th2 cells of the HCC early stage were significantly higher than the HCC late stage, but also CD8 naïve cells, Monocyte and Neutrophil were enriched in the HCC early stage. In the opposite, the HCC late stage showed a higher proportion of B cells compared to HCC early stage (Fig. 4B). The heatmap of immune cells also showed the tendency between two stages in samples (Fig. 4C). Furthermore, B cells were negatively correlated with most immune cells including CD8 naïve cells, Tr1 cells and Monocyte (Fig. 4D). On the other hand, Tfh cells, Th2 cells, and Monocyte were positively correlated with CD8 naïve cells (Fig. 4D). Potential gene biomarkers identified by multiple machine learning approaches To find the potential gene biomarkers to distinguish the early- and late-stage of HCC, we utilized three machine learning algorithms to identify the diagnostic markers from the DEGs. 20 potential DEGs were identified by the least absolute shrinkage and selection operator (LASSO) logistic regression algorithm ( Supplemental Fig. 3A ). What’s more, we used the 20 biomarkers to build LASSO logistic regression diagnostic model, whose performance was pretty good (AUC = 0.732, Supplemental Fig. 3B ). Of the 20 biomarkers, 7 DEGs had positive coefficients and 13 DEGs had negative coefficients ( Supplemental Fig. 3B, Supplemental Table 2 ). The Boruta algorithm was used to classify 39 features from the DEGs ( Supplemental Fig. 3D ). Moreover, the random forest diagnostic model based on the 39 biomarkers had also well performance (AUC = 0.704, Supplemental Fig. 3E ). SLC6A8 gene had the highest importance score (Mean decrease accuracy from random forest) in Supplemental Fig. 3F ( Supplemental Table 3 ). Another machine learning algorithm named the support vector machine-recursive feature elimination (SVM-RFE) to identify 41 features ( Supplemental Fig. 3G, Supplemental Table 4 ). and the performance of the diagnostic model for distinguish the HCC early stage and the HCC late stage was also well (AUC = 0.698, Supplemental Fig. 3H ). Next, 10 intersection biomarkers including SLC6A8 , FTCD , CYP2C9 , ANGPT2 , ENO1 , CNGA1 , KCNJ15 , SLC39A4 , ETV1 , and ACSL6 were extracted from abovementioned features (Fig. 5A). Sequentially, we validated the gene expression level of the 10 biomarkers in the validation GSE14520 dataset and found only 6 genes ( SLC6A8 , FTCD , CYP2C9 , ANGPT2 , ENO1 , CNGA1 ) had the consistent results with the merge dataset ( Supplemental Fig. 4A-B ). 3 genes ( FTCD , CYP2C9 , CNGA1 ) were significantly enriched in the HCC early stage, while other 3 genes ( SLC6A8 , ANGPT2 , ENO1 ) were significantly higher in the HCC late stage in both merge dataset and validation dataset ( Supplemental Fig. 4A-B ). To find the diagnostic makers from the 6 genes, we focused on the SLC6A8 gene which had higher expression in the HCC late stage (Fig. 5B-C) and performed the ROC analysis to check whether the SLC6A8 gene had well performance of diagnostic validity in both merge dataset. Finally, the AUC of ROC validation were 0.654 (Fig. 5D) and 0.701 (Fig. 5E) in the merge dataset and validation dataset, respectively, indicating the SLC6A8 gene had not bad performance as a diagnostic biomarker. Correlation analysis between SLC6A8 and immune cells We used spearman correlation to characterize the association between 6 genes ( SLC6A8 , FTCD , CYP2C9 , ANGPT2 , ENO1 , CNGA1 ) and infiltrating immune cells. The gene expression of most these genes had strong correlation with the 21 immune cells (Fig. 6A). SLC6A8 was significantly associated with 15 immune cells including DC, Exhausted, Th17 and CD8 naïve. Especially, the DC and CD8 naïve had the highest correlation coefficient with SLC6A8 (Fig. 6A). SLC6A8 had significantly positive correlation with DC cells (r = 0.284, p = 3.9e-12, Fig. 6C ) and other cells ( Supplemental Fig. 5A-I ), and significantly negative correlation with CD8 naïve (r = -0.22, p = 1e-07, Fig. 6D ) and other cells ( Supplemental Fig. 5J-L ). Expression level of SLC6A8 in single-cell transcriptomic data The cell types of GSE149614 dataset were showed by using t-SNE based on the Seurat-class object with the 2000 highly variable genes after data processing including normalization, scaling, clustering and so on (Fig. 7A). The expression of individual cells and each cell types for SLC6A8 are displayed in the Fig. 7B-C , respectively. Interesting, Hepatocyte, which was the dominant cells in liver presented the highest expression level of SLC6A8 among 6 main cell types (Fig. 7C). Following, we separated cells according to the tumor stage of patients into two groups (HCC early stage and HCC late stage) to validate whether the HCC enhanced SLC6A8 gene expression, which had been discovery in the bulk-seq RNA dataset. As expected, SLC6A8 gene had significantly higher expression ( P < 2e-16) in the HCC late stage compared to the HCC early stage in Hepatocyte (Fig. 7D) by Wilcoxon rank-sum test, which was consistent with the previous analysis. Subsequently, we detected the expression of SLC6A8 in patients with early and late liver cancer. Compared to early-stage HCC patients, the expression of SLC6A8 was higher in late-stage HCC patients (Fig. 7E). This result indicated that the expression level of SLC6A8 was related to the classification of liver cancer. Discussion HCC with high morbidity and mora rate has become a leading cause of cancer-related death worldwide [ 36 ]. Furthermore, HCC was diagnosed at the late tumor stage, which missing the appropriate treatment options [ 14 ]. Therefore, it is crucial for patient management to predict the disease at an early stage. Hence, the development of gene biomarkers that distinguish the early stage from the late stage of HCC is the primary goal of this study. To explore the potential biomarkers in the prediction of the early stage, we used multiple machine learning algorithms to compare the early stage and late stage of HCC based on the differentially expressed genes (DEGs). Apart from that, we also investigated the biological function of DEGs and immune cells between the early stage and the late stage. In this study, 137 down-regulated genes and 291 up-regulated genes were identified, and some of DEGs were enriched in p53 signaling pathway and tryptophan metabolism, which are cancer-related pathway [12; 23]. Interestingly, p53 signaling pathway with p53 protein mutations resulting in uncontrolled cell proliferation and cancer tumors [ 21 ], was enriched in the late stage of HCC, suggesting that there was more the loss of tumor-suppressing function at the late stage of HCC. Inversely, tryptophan metabolism promoting tumor cell intrinsic malignant properties as well as restricts antitumor immunity [ 26 ], was enriched in the early stage of HCC, indicating that the immune system had initially changed in the early stage. Moreover, the immune infiltration analysis also found that the immunity between two stages was significantly different, such as Tfh (T follicular helper) cells, providing essential help to B cells for effective antibody-mediated immune responses [ 6 ], and Th2, facilitating tissue repair [ 31 ] were both enriched in the early stage. Ten DEGs from the intersection of genes i.e SLC6A8 , FTCD , CYP2C9 , ANGPT2 , ENO1 , CNGA1 , KCNJ15 , SLC39A4 , ETV1 , and ACSL6 were screened by using LASSO analysis, RF-Boruta and SVM-REF. Previously, different reports uncover their involvement in a variety of malignancies [3; 10; 11; 32; 39; 40]. From comparing their expression value between two stages of HCC in the merge and validation cohort, we found SLC6A8 , ANGPT2 and ENO1 were observed as up-regulated genes in the late stage. Then, we used ROC curves to assess the efficacy of SLC6A8 on the diagnosis of HCC. The AUC of SLC6A8 were 0.654 and 0.701 in the merge cohort and the validation cohort, suggesting that SLC6A8 is a potential diagnostic biomarker to distinguish the early stage and the late stage of HCC. Especially, it has been shown that SLC6A8 is associated with the initiation and progression of human cancers [ 37 ]. Evidently, another study also demonstrated that SLC6A8 knockdown suppresses the invasion and migration of HCC [ 38 ]. There was different immune capacity between two stages of HCC, but whether the immunity associated with the SLC6A8 is still unknown. To further analyze these relationships, we used spearman correlation coefficient and found that the up-regulated gene SLC6A8 was positively associated with DC (dendritic cells), which provide antigens and co-stimulatory signals to cells of the adaptive immune system [ 28 ], and negatively associated with CD8-naïve cells (naïve CD8 + T cells), which selectively detect and eradicate cancer cells by targeting the antigens including tumor-specific neoantigens and self-antigens from tumors [ 22 ]. In our study, other immune cells also associated with gene SLC6A8 , inferring that SLC6A8 might affect the tumor progression by regulating the immune cells. In the literature, it has revealed the positive relationship between DC cells and SLC6A8-mediated creatine transport [ 29 ]. Another recent study also demonstrated that SLC6A8 may be involved in the development of cancer by participating in the Notch signaling pathway, which playing important role in the specification of the immune cells [ 5 ] in the NSCLC [ 3 ]. We also investigated whether the expression of gene SLC6A8 at single-cell transcript level and in vitro trial was the same as bulk-RNA sequencing level. Interestingly, six cell types including Hepatocyte which is the main functional cells of the live [ 24 ] were identified in the liver tissue which was consistent with the previous studies [ 17 ]. Furthermore, we found that the most important live cell Hepatocyte cells had higher values of SLC6A8 in the late stage of HCC. At the same time, the Myeloid cell also had slightly significantly higher values of SLC6A8 in the late stage of HCC. Additionally, the in vitro trial also showed the obviously significant expression changes of SLC6A8 between two groups. With the repeated verification of discovery on the SLC6A8 , we believe the gene SLC6A8 may be the potential diagnostic marker. In conclusion, we demonstrated that gene SLC6A8 was significantly up-regulated in the late stage of HCC based on the transcriptomic data. In addition, gene SLC6A8 was associated with immune cell infiltration, which provides a potential target for more precise and personalized immunotherapy. Therefore, the gene SLC6A8 may be a potential diagnostic marker. Conclusion In summary, we identified the DEGs by comparing the early stage and the late stage of HCC and found that they were associated with the cancer-related pathway and SLC6A8 , a potential diagnostic biomarker for clinical diagnosis between the early stage and the late stage of HCC was verified not only in the single-cell transcriptomic data but also in vitro trial . Limitation of the study We have identified potential diagnostic biomarker of the early stage and the late stage of HCC based on the transcriptomic data and verified it in the single-cell transcriptomic data and in vitro trial in our study. One of the limitations is that the sample size of our merge data may be not very large. But we have integrated the two most influential research (LIHC-US and LIRI-JP) with more than 100 samples into the merge dataset. Furthermore, this study only focused on biomarker exploration in the transcriptomic levels. The future study could develop the combination biomarkers from the genomic, epigenetic and metabolomic data. Declarations Author contributions Wang Yang conducted the study design; Linlin Song conducted data collection and data analyses; Wang Yang did the Manuscript writing; Linlin Song and Wang Yang conducted a Manuscript review and revise. Statements and declarations Not applicable. Declaration of competing interest Not applicable. Consent for publication Not applicable. Data availability The datasets used and analyzed during the currentstudy are available from the corresponding authors onreasonable request. References BRAY, F., FERLAY, J., SOERJOMATARAM, I., SIEGEL, R.L., TORRE, L.A., and JEMAL, A., 2018. Global cancer statistics 2018: GLOBOCAN estimates of incidence and mortality worldwide for 36 cancers in 185 countries. Ca-a Cancer Journal for Clinicians 68 , 6 (Nov-Dec), 394-424. DOI= http://dx.doi.org/10.3322/caac.21492. FENG, S., WANG, J., WANG, L., QIU, Q., CHEN, D., SU, H., LI, X., XIAO, Y., and LIN, C., 2023. Current Status and Analysis of Machine Learning in Hepatocellular Carcinoma. J Clin Transl Hepatol 11 , 5 (Oct 28), 1184-1191. DOI= http://dx.doi.org/10.14218/jcth.2022.00077s. FENG, Y., GUO, X., and TANG, H., 2021. SLC6A8 is involved in the progression of non-small cell lung cancer through the Notch signaling pathway. Ann Transl Med 9 , 3 (Feb), 264. DOI= http://dx.doi.org/10.21037/atm-20-5984. FRIEDMAN, J., HASTIE, T., and TIBSHIRANI, R., 2010. Regularization Paths for Generalized Linear Models via Coordinate Descent. J Stat Softw 33 , 1, 1-22. GARIS, M. and GARRETT-SINHA, L.A., 2020. Notch Signaling in B Cell Immune Responses. Front Immunol 11 , 609324. DOI= http://dx.doi.org/10.3389/fimmu.2020.609324. GUTIéRREZ-MELO, N. and BAUMJOHANN, D., 2023. T follicular helper cells in cancer. Trends Cancer 9 , 4 (Apr), 309-325. DOI= http://dx.doi.org/10.1016/j.trecan.2022.12.007. HANG, D., YANG, X.L., LU, J.Y., SHEN, C., DAI, J.C., LU, X.F., JIN, G.F., HU, Z.B., GU, D.F., MA, H.X., and SHEN, H.B., 2022. Untargeted plasma metabolomics for risk prediction of hepatocellular carcinoma: A prospective study in two Chinese cohorts. International Journal of Cancer 151 , 12 (Dec 15), 2144-2154. DOI= http://dx.doi.org/10.1002/ijc.34229. HäNZELMANN, S., CASTELO, R., and GUINNEY, J., 2013. GSVA: gene set variation analysis for microarray and RNA-seq data. BMC Bioinformatics 14 (Jan 16), 7. DOI= http://dx.doi.org/10.1186/1471-2105-14-7. HOSHIDA, Y., NIJMAN, S.M., KOBAYASHI, M., CHAN, J.A., BRUNET, J.P., CHIANG, D.Y., VILLANUEVA, A., NEWELL, P., IKEDA, K., HASHIMOTO, M., WATANABE, G., GABRIEL, S., FRIEDMAN, S.L., KUMADA, H., LLOVET, J.M., and GOLUB, T.R., 2009. Integrative transcriptome analysis reveals common molecular subclasses of human hepatocellular carcinoma. Cancer Research 69 , 18 (Sep 15), 7385-7392. DOI= http://dx.doi.org/10.1158/0008-5472.Can-09-1089. HUA, H., PAN, S., DIAO, H., CAO, Y., QIAN, X., and ZHANG, J., 2024. Increased ACSL6 Expression Predicts a Favorable Prognosis in Triple-negative Breast Cancer. Curr Med Chem (Jan 17). DOI= http://dx.doi.org/10.2174/0109298673278846231222103420. HUANG, C.K., SUN, Y., LV, L., and PING, Y., 2022. ENO1 and Cancer. Mol Ther Oncolytics 24 (Mar 17), 288-298. DOI= http://dx.doi.org/10.1016/j.omto.2021.12.026. KODAMA, M., HANASAWA, K., and TANI, T., 1997. Blood purification for critical care medicine: endotoxin adsorption. Ther Apher 1 , 3 (Aug), 224-227. DOI= http://dx.doi.org/10.1111/j.1744-9987.1997.tb00142.x. LI, W.H., WANG, Q.W., LU, J.X., ZHAO, B., GENG, Y.Q., WU, X.W., and CHEN, X.L., 2023. Machine learning-based prognostic modeling of lysosome-related genes for predicting prognosis and immune status of patients with hepatocellular carcinoma. Frontiers in Immunology 14 (May 19). DOI= http://dx.doi.org/ARTN 116925610.3389/fimmu.2023.1169256. LIVER, E.A.S., 2018. EASL Clinical Practice Guidelines: Management of hepatocellular carcinoma. Journal of Hepatology 69 , 1 (Jul), 182-236. LU, C., RONG, D.W., ZHANG, B., ZHENG, W.B., WANG, X.H., CHEN, Z.Y., and TANG, W.W., 2019. Current perspectives on the immunosuppressive tumor microenvironment in hepatocellular carcinoma: challenges and opportunities. Molecular Cancer 18 , 1 (Aug 29). DOI= http://dx.doi.org/ARTN 13010.1186/s12943-019-1047-6. LU, J., CHEN, Y., ZHANG, X., GUO, J., XU, K., and LI, L., 2022. A novel prognostic model based on single-cell RNA sequencing data for hepatocellular carcinoma. Cancer Cell Int 22 , 1 (Jan 25), 38. DOI= http://dx.doi.org/10.1186/s12935-022-02469-2. LU, Y., YANG, A., QUAN, C., PAN, Y., ZHANG, H., LI, Y., GAO, C., LU, H., WANG, X., CAO, P., CHEN, H., LU, S., and ZHOU, G., 2022. A single-cell atlas of the multicellular ecosystem of primary and metastatic hepatocellular carcinoma. Nat Commun 13 , 1 (Aug 6), 4594. DOI= http://dx.doi.org/10.1038/s41467-022-32283-3. MECHAM, B.H., NELSON, P.S., and STOREY, J.D., 2010. Supervised normalization of microarrays. Bioinformatics 26 , 10 (May 15), 1308-1315. DOI= http://dx.doi.org/10.1093/bioinformatics/btq118. MIAO, Y.R., ZHANG, Q., LEI, Q., LUO, M., XIE, G.Y., WANG, H.X., and GUO, A.Y., 2020. ImmuCellAI: A Unique Method for Comprehensive T-Cell Subsets Abundance Prediction and its Application in Cancer Immunotherapy. Advanced Science 7 , 7 (Apr). DOI= http://dx.doi.org/ARTN 190288010.1002/advs.201902880. OLSEN, T.K. and BARYAWNO, N., 2018. Introduction to Single-Cell RNA Sequencing. Curr Protoc Mol Biol 122 , 1 (Apr), e57. DOI= http://dx.doi.org/10.1002/cpmb.57. PAPAGIANNOPOULOS, P., ANDRIANOPOULOS, A., SHARP, J.A., DAVIS, M.A., and HYNES, M.J., 1996. The hapC gene of Aspergillus nidulans is involved in the expression of CCAAT-containing promoters. Mol Gen Genet 251 , 4 (Jun 24), 412-421. DOI= http://dx.doi.org/10.1007/bf02172369. PHILIP, M. and SCHIETINGER, A., 2022. CD8(+) T cell differentiation and dysfunction in cancer. Nat Rev Immunol 22 , 4 (Apr), 209-223. DOI= http://dx.doi.org/10.1038/s41577-021-00574-3. PLATTEN, M., NOLLEN, E.A.A., RöHRIG, U.F., FALLARINO, F., and OPITZ, C.A., 2019. Tryptophan metabolism as a common therapeutic target in cancer, neurodegeneration and beyond. Nat Rev Drug Discov 18 , 5 (May), 379-401. DOI= http://dx.doi.org/10.1038/s41573-019-0016-5. PU, W. and ZHOU, B., 2022. Hepatocyte generation in liver homeostasis, repair, and regeneration. Cell Regen 11 , 1 (Jan 6), 2. DOI= http://dx.doi.org/10.1186/s13619-021-00101-8. REBOUISSOU, S. and NAULT, J.C., 2020. Advances in molecular classification and precision oncology in hepatocellular carcinoma. Journal of Hepatology 72 , 2 (Feb), 215-229. DOI= http://dx.doi.org/10.1016/j.jhep.2019.08.017. RIEDER, M.J., O'DROBINAK, D.M., TONELLATO, P.J., and GREENE, A.S., 1996. Mathematical analysis of type-I and type-IIb muscle fiber force generation in renal hypertension. Ann Biomed Eng 24 , 4 (Jul-Aug), 489-499. DOI= http://dx.doi.org/10.1007/bf02648111. RITCHIE, M.E., PHIPSON, B., WU, D., HU, Y.F., LAW, C.W., SHI, W., and SMYTH, G.K., 2015. powers differential expression analyses for RNA-sequencing and microarray studies. Nucleic Acids Research 43 , 7 (Apr 20). DOI= http://dx.doi.org/ARTN e4710.1093/nar/gkv007. SHORTMAN, K. and NAIK, S.H., 2007. Steady-state and inflammatory dendritic-cell development. Nat Rev Immunol 7 , 1 (Jan), 19-30. DOI= http://dx.doi.org/10.1038/nri1996. TIAN, C., YUAN, H., LU, Y., HE, H., LI, Q., LI, S., YANG, J., WANG, M., XU, R., LIU, Q., and XIANG, M., 2023. CARD9 deficiency promotes pancreatic cancer growth by blocking dendritic cell maturation via SLC6A8-mediated creatine transport. Oncoimmunology 12 , 1, 2204015. DOI= http://dx.doi.org/10.1080/2162402x.2023.2204015. TORROJA, C. and SANCHEZ-CABO, F., 2019. Digitaldlsorter: Deep-Learning on scRNA-Seq to Deconvolute Gene Expression Data. Frontiers in Genetics 10 (Oct 25). DOI= http://dx.doi.org/ARTN 97810.3389/fgene.2019.00978. WALKER, J.A. and MCKENZIE, A.N.J., 2018. T(H)2 cell development and function. Nat Rev Immunol 18 , 2 (Feb), 121-133. DOI= http://dx.doi.org/10.1038/nri.2017.118. WANG, S., ZHOU, Y., YU, R., LING, J., LI, B., YANG, C., CHENG, Z., QIAN, R., LIN, Z., YU, C., ZHENG, J., ZHENG, X., JIA, Q., WU, W., WU, Q., CHEN, M., YUAN, S., DONG, W., SHI, Y., JANSEN, R., YANG, C., HAO, Y., YAO, M., QIN, W., and JIN, H., 2023. Loss of hepatic FTCD promotes lipid accumulation and hepatocarcinogenesis by upregulating PPARγ and SREBP2. JHEP Rep 5 , 10 (Oct), 100843. DOI= http://dx.doi.org/10.1016/j.jhepr.2023.100843. WEI, Q.X., TAN, N.R., XIONG, S.Y., LUO, W.R., XIA, H.Y., and LUO, B.M., 2023. Deep Learning Methods in Medical Image-Based Hepatocellular Carcinoma Diagnosis: A Systematic Review and Meta-Analysis. Cancers 15 , 23 (Dec). DOI= http://dx.doi.org/ARTN 570110.3390/cancers15235701. WU, T., HU, E., XU, S., CHEN, M., GUO, P., DAI, Z., FENG, T., ZHOU, L., TANG, W., ZHAN, L., FU, X., LIU, S., BO, X., and YU, G., 2021. clusterProfiler 4.0: A universal enrichment tool for interpreting omics data. Innovation (Camb) 2 , 3 (Aug 28), 100141. DOI= http://dx.doi.org/10.1016/j.xinn.2021.100141. YANG, C., HUANG, X.W., LIU, Z.C., QIN, W.X., and WANG, C., 2020. Metabolism-associated molecular classification of hepatocellular carcinoma. Molecular Oncology 14 , 4 (Apr), 896-913. DOI= http://dx.doi.org/10.1002/1878-0261.12639. YANG, J.D., HAINAUT, P., GORES, G.J., AMADOU, A., PLYMOTH, A., and ROBERTS, L.R., 2019. A global view of hepatocellular carcinoma: trends, risk, prevention and management. Nature Reviews Gastroenterology & Hepatology 16 , 10 (Oct), 589-604. DOI= http://dx.doi.org/10.1038/s41575-019-0186-y. YANG, X. and LI, Q., 2022. Pan-Cancer Analysis of the Oncogenic and Immunological Role of Solute Carrier Family 6 Member 8 (SLC6A8). Front Genet 13 , 916439. DOI= http://dx.doi.org/10.3389/fgene.2022.916439. YUAN, L., WU, X.J., LI, W.C., ZHUO, C., XU, Z., TAN, C., MA, R., WANG, J., and PU, J., 2020. SLC6A8 Knockdown Suppresses the Invasion and Migration of Human Hepatocellular Carcinoma Huh-7 and Hep3B Cells. Technol Cancer Res Treat 19 (Jan-Dec), 1533033820983029. DOI= http://dx.doi.org/10.1177/1533033820983029. ZHANG, C., LI, Q., CHENG, Y., YANG, X., CHEN, W., HE, K., and CHEN, M., 2022. Variants in CYP2J2 and CYP2C9 Contribute to Susceptibility of Lung Cancer. Curr Cancer Drug Targets (Nov 14). DOI= http://dx.doi.org/10.2174/1568009623666221114115012. ZHU, J., WU, Y., YU, Y., LI, Y., SHEN, J., and ZHANG, R., 2022. MYBL1 induces transcriptional activation of ANGPT2 to promote tumor angiogenesis and confer sorafenib resistance in human hepatocellular carcinoma. Cell Death Dis 13 , 8 (Aug 20), 727. DOI= http://dx.doi.org/10.1038/s41419-022-05180-2. Additional Declarations No competing interests reported. Supplementary Files Supplementalmaterials.docx Cite Share Download PDF Status: Under Review Version 1 posted Editorial decision: Revision requested 28 Aug, 2024 Reviews received at journal 27 Aug, 2024 Reviewers agreed at journal 08 Aug, 2024 Reviews received at journal 07 Jul, 2024 Reviewers agreed at journal 02 Jul, 2024 Reviewers invited by journal 26 Jun, 2024 Editor assigned by journal 21 Jun, 2024 Submission checks completed at journal 21 Jun, 2024 First submitted to journal 12 Jun, 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-4570554","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":325474116,"identity":"f736a1f4-efaa-473e-9fd2-e6088313f4c8","order_by":0,"name":"Linlin Song","email":"","orcid":"","institution":"Second Affiliated Hospital of Harbin Medical University","correspondingAuthor":false,"prefix":"","firstName":"Linlin","middleName":"","lastName":"Song","suffix":""},{"id":325474117,"identity":"4e6180e1-55f5-48fe-a2f1-7711ab6fe5ff","order_by":1,"name":"Hongli Zhang","email":"","orcid":"","institution":"Second Affiliated Hospital of Harbin Medical University","correspondingAuthor":false,"prefix":"","firstName":"Hongli","middleName":"","lastName":"Zhang","suffix":""},{"id":325474118,"identity":"a59a0c79-651d-4e5d-aff7-96358d3eedbd","order_by":2,"name":"Wang Yang","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA7klEQVRIiWNgGAWjYDADNmbG9g8fDGzsiNfCx858jHFGQVoy8Vrk+NnSmHk+HGJsIKTS4PjZw695au7YtTHzmD22MTjAzMB++OgGvFrO5KVZ8xx7lgzUYm6cY3CHj4EnLe0GXi0HcsyMc9gOJ7Mx8xhI5xg8Y2aQ4DHDr+X8G6CWf1AtFgaHGRsIarmRY/w4t+2wHRszW5o0AzFaJG+8MWP+23c4gY2Z+bBhj0FaMhshv/CdzzH+OOPbYXv5/oOND378sbHjZz98DK8WhQMMbBJAOrEBJsKGTzkIyDcwMH8A0vaEFI6CUTAKRsEIBgBL/krjdScB/gAAAABJRU5ErkJggg==","orcid":"","institution":"Second Affiliated Hospital of Harbin Medical University","correspondingAuthor":true,"prefix":"","firstName":"Wang","middleName":"","lastName":"Yang","suffix":""}],"badges":[],"createdAt":"2024-06-12 13:23:10","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-4570554/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-4570554/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":60123669,"identity":"aa32ede0-73f5-4267-953c-9b452441bf63","added_by":"auto","created_at":"2024-07-12 05:24:23","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":73072,"visible":true,"origin":"","legend":"\u003cp\u003eSchematic workflow of this study\u003c/p\u003e","description":"","filename":"floatimage1.png","url":"https://assets-eu.researchsquare.com/files/rs-4570554/v1/0ab9ba2b436482f47d021e19.png"},{"id":60123665,"identity":"68d95de6-41fe-41ed-a1ac-df1e97dc879b","added_by":"auto","created_at":"2024-07-12 05:24:23","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":408669,"visible":true,"origin":"","legend":"\u003cp\u003eDEGs between the early-stage group and the late-stage group in HCC merge dataset (LIHC-US and LIRI-JP). \u003cstrong\u003eA\u003c/strong\u003e. Volcano of DEGs. Dark green dots represent enriched-in-early-stage genes, light purple dots represent enriched-in-late-stage genes and grey dots represent non-significant genes. \u003cstrong\u003eB\u003c/strong\u003e. Heatmap of DEGs. The colors of column annotations represent the different groups.\u003c/p\u003e","description":"","filename":"floatimage2.png","url":"https://assets-eu.researchsquare.com/files/rs-4570554/v1/4b7f347635e12416f4ba7135.png"},{"id":60123673,"identity":"586eb9cd-83bc-4a58-9807-c58c25f2d9e9","added_by":"auto","created_at":"2024-07-12 05:24:24","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":339557,"visible":true,"origin":"","legend":"\u003cp\u003eFunctional analysis of DEGs between the early-stage group and the late-stage group in HCC merge cohort. \u003cstrong\u003eA\u003c/strong\u003e. Dotplot of enriched GO terms in DEGs. \u003cstrong\u003eB\u003c/strong\u003e. Dotplot of enriched KEGG pathway in DEGs. \u003cstrong\u003eC\u003c/strong\u003e. GSEA results in HCC early-stage samples. \u003cstrong\u003eD\u003c/strong\u003e. GSEA results in HCC late-stage samples.\u003c/p\u003e","description":"","filename":"floatimage3.png","url":"https://assets-eu.researchsquare.com/files/rs-4570554/v1/3d677e4858779a9d7601dada.png"},{"id":60123666,"identity":"da22dddf-a85e-4af0-a0ee-4bb09016047b","added_by":"auto","created_at":"2024-07-12 05:24:23","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":300767,"visible":true,"origin":"","legend":"\u003cp\u003eEvaluation and visualization of immune cell infiltration between the HCC early stage and late stage in merge cohort. \u003cstrong\u003eA\u003c/strong\u003e. Stacked barplot of infiltrating immune cells in the HCC early-stage and late-stage samples. \u003cstrong\u003eB\u003c/strong\u003e. Boxplot of differential expression of 21 infiltrating immune cells between the HCC early stage and late stage in merge cohort. Wilcoxon rank-sum test was used for differential analysis. *\u003cem\u003eP\u003c/em\u003e \u0026lt; 0.05; **\u003cem\u003eP\u003c/em\u003e \u0026lt; 0.01. ns, non-significance (Dark green in the boxes represents the HCC early stage and light purple represents the HCC late stage). \u003cstrong\u003eC\u003c/strong\u003e. Heatmap of 21 infiltrating immune cells. D. Correlation matrix of 21 immune cell infiltration between the HCC early stage and late stage in merge cohort. Red and blue indicate positive and negative correlations, respectively. The darker color shows the stronger the correlation.\u003c/p\u003e","description":"","filename":"floatimage4.png","url":"https://assets-eu.researchsquare.com/files/rs-4570554/v1/45644407bded6c3a228c99dd.png"},{"id":60124340,"identity":"c20a4dbd-37ed-486b-a991-5680fa42aabd","added_by":"auto","created_at":"2024-07-12 05:32:23","extension":"png","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":221395,"visible":true,"origin":"","legend":"\u003cp\u003eValidation and ROC of SLC6A8 diagnostic marker identified from the merge datasets by multiple machine learning algorithms. \u003cstrong\u003eA\u003c/strong\u003e. The overlapping genes among three machine learning methods. \u003cstrong\u003eB-C\u003c/strong\u003e. The expression level of SLC6A8 diagnostic marker between the HCC early stage and the HCC late stage in the merge dataset and validation GSE14520 dataset, respectively. Wilcoxon rank-sum test was used for differential analysis. \u003cstrong\u003eD-E\u003c/strong\u003e. ROC validation of diagnostic validity of the \u003cem\u003eSLC6A8\u003c/em\u003e diagnostic marker in the merge dataset and validation GSE14520 dataset, respectively.\u003c/p\u003e","description":"","filename":"floatimage5.png","url":"https://assets-eu.researchsquare.com/files/rs-4570554/v1/eba6f3e3c954505322eedf6d.png"},{"id":60123670,"identity":"9095e20c-e8c7-4913-9146-ac3f50fd10c1","added_by":"auto","created_at":"2024-07-12 05:24:23","extension":"png","order_by":6,"title":"Figure 6","display":"","copyAsset":false,"role":"figure","size":454048,"visible":true,"origin":"","legend":"\u003cp\u003eCorrelation between \u003cem\u003eSLC6A8 \u003c/em\u003eand 21 immune cells. \u003cstrong\u003eA\u003c/strong\u003e. Heatmap of correlation between 6 biomarkers (\u003cem\u003eSLC6A8\u003c/em\u003e,\u003cem\u003eFTCD\u003c/em\u003e,\u003cem\u003e CYP2C9\u003c/em\u003e,\u003cem\u003e ANGPT2\u003c/em\u003e,\u003cem\u003e ENO1\u003c/em\u003e,\u003cem\u003e CNGA1\u003c/em\u003e) and 21 immune cells. Color represents the spearman correlation coefficient. *\u003cem\u003eP\u003c/em\u003e\u0026lt; 0.05; +\u003cem\u003eP \u003c/em\u003e\u0026lt; 0.01. \u003cstrong\u003eB\u003c/strong\u003e. Lollipop diagram of correlation between \u003cem\u003eSLC6A8 \u003c/em\u003eand 22 immune cell infiltration. Circle color represents the p-value (red indicates significance) and circle size represents the absolute value of coefficient. \u003cstrong\u003eC-D\u003c/strong\u003e. Scatter plots of correlation between \u003cem\u003eSLC6A8 \u003c/em\u003eand immune cells whose had the highest correlation coefficients (DC and CD8_naive). Dark green represents the HCC early-stage samples and light purple represents the HCC late-stage samples.\u003c/p\u003e","description":"","filename":"floatimage6.png","url":"https://assets-eu.researchsquare.com/files/rs-4570554/v1/98c9e4439686bca86c6a7aad.png"},{"id":60123672,"identity":"cb7ac27b-0ed9-4de1-843d-0e60807fb621","added_by":"auto","created_at":"2024-07-12 05:24:23","extension":"png","order_by":7,"title":"Figure 7","display":"","copyAsset":false,"role":"figure","size":281834,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cem\u003eSLC6A8\u003c/em\u003e expression\u003cem\u003e \u003c/em\u003ein single-cell GSE149614 transcriptomic dataset. \u003cstrong\u003eA\u003c/strong\u003e. The t-distributed stochastic neighbor embedding (t-SNE) plot of the 6 identified main cell types in GSE173896 dataset. \u003cstrong\u003eB\u003c/strong\u003e. t-SNE map highlighting the expression of \u003cem\u003eSLC6A8\u003c/em\u003egene. \u003cstrong\u003eC\u003c/strong\u003e. Bubble plot showing the expression of the \u003cem\u003eSLC6A8\u003c/em\u003e related different cell types. Dot size represents the percent expressed, and color represents average expression. \u003cstrong\u003eD\u003c/strong\u003e. Correlation of \u003cem\u003eSLC6A8\u003c/em\u003e with 6 types of main cell types between the HCC early stage and late stage in single cell transcriptome data. \u003cstrong\u003eE\u003c/strong\u003e. SLC6A8 expression of early-stage and late-stage HCC patients. Wilcoxon rank-sum test was used for differential analysis. *\u003cem\u003eP \u003c/em\u003e\u0026lt;\u003cem\u003e \u003c/em\u003e0.05.\u003c/p\u003e","description":"","filename":"floatimage7.png","url":"https://assets-eu.researchsquare.com/files/rs-4570554/v1/64b77783bc5178ef3fa15162.png"},{"id":60124991,"identity":"aa1500ff-b8bd-47dc-b96a-fa02d4725391","added_by":"auto","created_at":"2024-07-12 05:40:24","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":2214678,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-4570554/v1/3273d9d2-fe74-49f3-8b60-b42903e6290d.pdf"},{"id":60123667,"identity":"3be61347-5253-42d7-9549-ace5183e8e81","added_by":"auto","created_at":"2024-07-12 05:24:23","extension":"docx","order_by":1,"title":"","display":"","copyAsset":false,"role":"supplement","size":1304355,"visible":true,"origin":"","legend":"","description":"","filename":"Supplementalmaterials.docx","url":"https://assets-eu.researchsquare.com/files/rs-4570554/v1/aaf817a66fefc616769487bb.docx"}],"financialInterests":"No competing interests reported.","formattedTitle":"Multiple machine learning algorithms identified SLC6A8 as a diagnostic biomarker of the late stage of Hepatocellular carcinoma","fulltext":[{"header":"Introduction","content":"\u003cp\u003eHepatocellular carcinoma (HCC), a fourth leading cause of death over the world in 2018, has been a globally severe disease [\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e]. Most patients are diagnosed at a late stage, and the only potentially curative approaches for treating are hepatic resection and liver transplantation, which have low treatment rate [\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e]. So, it is critical to find the diagnostic biomarker of HCC as early as possible. However, rather than focusing on the severity of HCC, many researchers have tried to build a classification to distinguish between normal and HCC patients [9; 25; 35]. In Addition, tumor immune microenvironment plays an important role in tumor progression and treatment outcome. In recent years, immunotherapy has become a hopeful approach for treating cancer. However, response rates among HCC patients have been limited. Therefore, conducting a comprehensive analysis of different types of immune cells within tumors could aid in identifying new biomarkers for prognosis and therapeutic efficacy in HCC patients.\u003c/p\u003e \u003cp\u003eIn recent decades, HCC has been a major focus of research interest. For instance, specific genes like \u003cem\u003eCLTA\u003c/em\u003e, \u003cem\u003eTALDO1\u003c/em\u003e and \u003cem\u003eCSTB\u003c/em\u003e, identified as gene signature through single cell RNA sequencing, have been associated with survival outcome and immunotherapy response [\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e]. Hang D et al. conducted mass spectrometry analysis for metabolomics to identify new predictive biomarkers and pathways in HCC [\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e]. Genes such as Receptor Activity Modifying Protein 3 (\u003cem\u003eRAMP3\u003c/em\u003e) and CD68 Molecule (\u003cem\u003eCD68\u003c/em\u003e) have shown significantly expressed in HCC based on machine-learning algorithms [\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e]. Overall, these genes are considered vital candidates for potential diagnostic markers in HCC clinical settings.\u003c/p\u003e \u003cp\u003eRecently, the rapid development of next-generation sequencing technologies provides amount of RNA sequencing data of HCC. However, RNA sequencing is typically conducted in \"bulk,\" capturing average gene expression patterns from a multitude of cells [\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e]. Notably, single-cell RNA sequencing single-cell RNA sequencing is an innovative sequencing technique that offers valuable insights into the characterization of individual immune cells or tumor cells [\u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e]. At the same time, bioinformatic analysis and machine learning have emerged as increasingly promising strategies for comprehensive and in-depth analysis of large datasets, such as transcriptome sequences, and interdisciplinary collaborations have been instrumental in advancing clinical therapeutic methods [\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e]. The advent of modern computer-assisted medical science has provided significant guidance and hope for previously untreatable diseases, such as utilizing the XGBoost algorithm for HCC diagnosis [\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e]. To meet the demand for early diagnosis, numerous efforts have focused on developing new methods based on deep learning analysis [\u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e33\u003c/span\u003e]. Clearly, achieving accurate clinical diagnosis of COPD remains a critical and imperative pursuit.\u003c/p\u003e \u003cp\u003eIn this study, we utilized established bioinformatic tools to screen for potential biomarkers indicating the late stage of HCC. Two transcriptome datasets were selected from the published database (TCGA, ICGC) for analysis. After identifying DEGs (Differential expressed genes) using limma, we conducted GO and GSEA analyses on these DEGs. Notably, we identified a candidate diagnostic biomarker, \u003cem\u003eSLC6A8\u003c/em\u003e, which was the intersection of genes selected by LASSO, SVM-REF and RF-Boruta methods and validated it using another dataset. Additionally, we investigated changes in immune cell composition between early-stage and late-stage HCC samples using ImmuCellAI (Immune Cell Abundance Identifier) to analyze immune cell infiltration. Furthermore, we explored the relationship between immune cell infiltration and the potential diagnostic biomarker of HCC, validating our findings with single-cell transcriptomic data and \u003cem\u003ein vitro\u003c/em\u003e trial. Overall, we identified a previously unrecognized gene with the potential to guide future clinical treatment and diagnosis of HCC patients.\u003c/p\u003e"},{"header":"Materials and methods","content":"\u003cp\u003eData collection and download\u003c/p\u003e \u003cp\u003eStandardized RNA-Seq reads (Release 28) of LIHC-US and LIRI-JP projects were obtained from The Cancer Genome Atlas (TCGA, \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://cancergenome.nih.gov/\u003c/span\u003e\u003cspan address=\"https://cancergenome.nih.gov/\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e) and International Cancer Genome Collaboration (ICGC, \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://dcc.icgc.org/\u003c/span\u003e\u003cspan address=\"https://dcc.icgc.org/\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e). GSE14520, the validation dataset with chip data was downloaded from GEO database (\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://www.ncbi.nlm.nih.gov/geo/\u003c/span\u003e\u003cspan address=\"https://www.ncbi.nlm.nih.gov/geo/\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e). Only patients with tumor stage and samples from the primary site were included in the analysis. For LIHC-US (TCGA), 231 patients were retained. For LIRI-JP (ICGC), 344 patients were met the criterion. Finally, we extracted 218 patients from the GSE14520. The HCC scRNA-seq dataset GSE149614 was load from the GEO database (\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://www.ncbi.nlm.nih.gov/geo/\u003c/span\u003e\u003cspan address=\"https://www.ncbi.nlm.nih.gov/geo/\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e) and included 10 patients, only the primary tumor samples including 3 TNM stage I, 1 TNM stage II, 2 TNM stage III and 4 TNM stage IV samples will be taken for downstream analysis.\u003c/p\u003e \u003cp\u003eAccording to the TNM, HCC disease was divided into four stages: stage I, stage II, stage III, and stage IV. In the following analysis, we defined stage I and stage II as HCC early-stage, stage III and stage IV as HCC late-stage. A merge data cohort comprising of LIHC-US and LIRI-JP datasets through \u0026ldquo;limma\u0026rdquo; and \u0026ldquo;snm\u0026rdquo; R packages, and the GSE14520 was as the validation dataset to confirm the analysis results. The GSE149614, an HCC scRNA-seq dataset, comprising of 4 HCC early-stage and 6 HCC early-stage patients was also taken as the validation dataset to investigate the results at single-cell level. The clinical information of these cohorts involved in this study had presented in \u003cb\u003eSupplemental table \u003cspan refid=\"MOESM1\" class=\"InternalRef\"\u003e1\u003c/span\u003e\u003c/b\u003e.\u003c/p\u003e \u003cp\u003eData Integration\u003c/p\u003e \u003cp\u003eA merged dataset cohort comprising LIHC-US and LIRI-JP dataset through R package \u0026ldquo;limma\u0026rdquo; (version 3.58.11) and \u0026ldquo;SNM\u0026rdquo; (version 1.50.0), correcting the batch effect by VOOM and SNM function in \u003cb\u003eSupplemental Fig.\u0026nbsp;1D\u003c/b\u003e [18; 27]. What\u0026rsquo;s more, the results of other approaches used to remove the batch effect in the different dataset had showed in \u003cb\u003eSupplemental Fig.\u0026nbsp;1A-C\u003c/b\u003e. Principal component analysis (PCA) shows the samples distribution before and after batch correction.\u003c/p\u003e \u003cp\u003eDifferential Expression Genes\u003c/p\u003e \u003cp\u003eDEGs form the merged dataset were with the cutoff criteria of |logFC| \u0026gt; 0.5, adjusted-pvalue\u0026thinsp;\u0026lt;\u0026thinsp;0.05 by using \u0026ldquo;limma\u0026rdquo; (version 3.58.11) R package. A linear model was then fitted with lmFit and differential expression was assessed using the eBayes function.\u003c/p\u003e \u003cp\u003eFunctional enrichment analysis\u003c/p\u003e \u003cp\u003eGene ontology (GO) enrichment analysis was performed to investigate the DEG biological significant by using the clusterProfiler (version 4.10.1) R package [\u003cspan citationid=\"CR34\" class=\"CitationRef\"\u003e34\u003c/span\u003e]. The Molecular Signatures Database (MSigDB) Hallmark Gene Sets and Kyoto Encyclopedia of Genes and Genomes (KEGG) Pathway were used for pathway enrichment analysis by using msigdbr (version 7.5.1) and massdatabase (version 1.0.10) R package. GSVA analysis was performed using the GSVA (version 1.50.1) R package [\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e]. Both GSEA and GSVA were performed with the criteria of \u003cem\u003eP\u003c/em\u003e- value\u0026thinsp;\u0026lt;\u0026thinsp;0.05.\u003c/p\u003e \u003cp\u003eImmune cell infiltration\u003c/p\u003e \u003cp\u003eThe ImmuCellAI, a gene set signature-based method, is a deconvolution algorithm, which can estimate the abundance of 24 immune cell types [\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e]. A single sample gene set enrichment analysis (ssGSEA) algorithm was applied to deconvolve bulk gene expression matrix. Immune cell infiltration (24 various cell types including 18 T-cell subsets) was precisely quantified in each gene expression profile. The detailed process was carried out by using ImmuneCellAI (version 0.1.0) R package.\u003c/p\u003e \u003cp\u003ePotential biomarkers selection\u003c/p\u003e \u003cp\u003eTo find the potential prognostic gene biomarkers from the DEGs, we screened them by three machine-learning algorithms. 1) least absolute shrinkage and selection operator with logistic regression (LASSO-LR), 2) support vector machine-recursive feature elimination (SVM-RFE) algorithms, 3) Boruta with random forest (Brouta-RF).\u003c/p\u003e \u003cp\u003eLASSO can identify genes significantly associated with different samples, which represents a regression analysis algorithm that applies regularization for variable selection using \u0026ldquo;glmnet\u0026rdquo; (version 4.1.7) R package [\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e]. In this study, LASSO with logistic regression first reduced the dimensionality successfully (425 original genes to 20 meaningful genes). Specifically, LASSO method with crossing validation using the mean of square error as cost function was performed, then they shrank into a few more important features according to mean of square error.\u003c/p\u003e \u003cp\u003eSVM-RFE represents a widely used supervised machine-learning protocol for classification and regression, used to find the best variables by deleting SVM-generated eigenvectors. The \u0026ldquo;Caret\u0026rdquo; (version 6.0.94) R package via grid search method is employed to select hyperparameters for all classifiers using 10-fold cross-validation for the training dataset. SVM-RFE can identify the diagnostic value of biomarkers with higher discriminative power using the \u0026ldquo;e1071\u0026rdquo; (version 1.7.13) R package.\u003c/p\u003e \u003cp\u003eThe Boruta algorithm is a wrapper built around the random forest classification algorithm. The Boruta (version 8.0.0) R package is to capture all the important, interesting features you might have in the dataset with respect to an outcome variable. The \u0026ldquo;Caret\u0026rdquo; (version 6.0.94) R package via grid search method is employed to select hyperparameters of random forest for all classifiers using 10-fold cross-validation for the training dataset. Brouta-RF can identify the diagnostic value of biomarkers with higher discriminative power using the \u0026ldquo;randomforest\u0026rdquo; (version 4.7\u0026ndash;1.1) R package.\u003c/p\u003e \u003cp\u003eUltimately, we combined the overlapping genes among LASSO-LR, SVM-RFE and Brouta-RF algorithm for further analysis. A two-sided \u003cem\u003eP\u003c/em\u003e value\u0026thinsp;\u0026lt;\u0026thinsp;0.05 between the early stage and the late stage of HCC was statistically significant on these genes. Then, we validated their expression level to estimate their ability to be candidate diagnostic biomarkers in the GSE14520 data.\u003c/p\u003e \u003cp\u003eROC of diagnostic biomarker\u003c/p\u003e \u003cp\u003eThe nonparametric Wilcoxon rank-sum test was used to perform inter-group comparisons of continuous variables. The degree of efficacy of each diagnostic biomarker was assessed using receiver operating characteristic (ROC) curves by \u0026ldquo;pROC\u0026rdquo; (version 1.18.5) and \u0026ldquo;multipleROC\u0026rdquo; (version 0.1.1) R package, which is the gold standard to prove the diagnostic accuracy and test the efficacy of diagnostic biomarkers in the GSE14520 cohort.\u003c/p\u003e \u003cp\u003eSingle cell transcriptome data processing and analyzing\u003c/p\u003e \u003cp\u003eGSE149614 raw data was downloaded from GEO databases. In the process of single cell transcriptome data processing, we carried out normalization, scaling, clustering of cells and achieved 6 main cell types using Seurat (version 5.0.3) R package. Single cells were extracted with the criteria of nFeature_RNA\u0026thinsp;\u0026gt;\u0026thinsp;250, percent.mt\u0026thinsp;\u0026lt;\u0026thinsp;20%, percent. Log10GenesPerUMI\u0026thinsp;\u0026gt;\u0026thinsp;0.8, nCount_RNA\u0026thinsp;\u0026gt;\u0026thinsp;500 to removing doublet and dead cells by Seurat R package (version 5.0.3). Then we normalized the filtered gene-barcode matrices using \u0026ldquo;LogNormalize\u0026rdquo; method with the \u0026ldquo;NormalizeData\u0026rdquo; function. The top 2000 highly variable genes were found by the \u0026ldquo;FindVariableFeatures\u0026rdquo; function using the \u0026ldquo;vst\u0026rdquo; method which were centered and scaled using \u0026ldquo;ScaleData\u0026rdquo; before. Then we performed principal component analysis (PCA) based on these 2000 highly variable genes with the intention of dimensionality reduction, then dimensionality-reduced clusters were showed on the 2D map produced with the t-distributed t-SNE using function \u0026ldquo;FindNeighbors\u0026rdquo;, \u0026ldquo;FindClusters\u0026rdquo; and \u0026ldquo;runTSNE\u0026rdquo; from Seurat. Kruskal\u0026ndash;Wallis test was used to estimate the difference of gene expression level.\u003c/p\u003e \u003cp\u003eReal-time quantitative reverse transcriptase-PCR\u003c/p\u003e \u003cp\u003eTotal RNA was extracted from liver cancer tissue using Trizol (Invitrogen, Carlsbad, America) reagent before cDNA was obtained using the Trans-Script All in-one First-strand cDNA Synthesis Supermix for qPCR Kit (TransGen Biotech, Beijing, China). Real-time quantitative PCR was performed in Step One ABI real-time. PCR System through SYBR Green Master (Roche, Basel, Switzerland). GAPDH, Forward primer: 5\u0026rsquo;-CATGTTCGTCATGGGTGTGAA-3\u0026rsquo;, Reverse primer:5\u0026rsquo;-GGCATGGACTGTGGTCATGAG-3\u0026rsquo;. SLC6A8, Forward primer: 5\u0026rsquo;-GGCAGCTACAACCGCTTCAACA-3\u0026rsquo;, Reverse primer:5\u0026rsquo;-CAGGATGGAGAAGACCACGAAG-3\u0026rsquo;.\u003c/p\u003e \u003cdiv id=\"Sec3\" class=\"Section2\"\u003e \u003ch2\u003eStatistical analysis\u003c/h2\u003e \u003cp\u003eR (version 4.3.3) and Rstudio (version 2023.12.1\u0026thinsp;+\u0026thinsp;402) were used for statistical analysis. Wilcox-rank-sum test was carried out to analyze the significant differences of features between the early-stage and late-stage groups. Spearman correlation coefficient was used to identify the correlations between genes and immune cells. A \u003cem\u003eP\u003c/em\u003e-value\u0026thinsp;\u0026lt;\u0026thinsp;0.05 indicated statistical significance (*\u003cem\u003eP\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;0.05; **\u003cem\u003eP\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;0.01; ***\u003cem\u003eP\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;0.001).\u003c/p\u003e \u003c/div\u003e"},{"header":"Results","content":"\u003cp\u003eIdentification of DEGs in the HCC early stage and late stage\u003c/p\u003e\n\u003cp\u003eWe performed differential expression analysis between 397 early-stage HCC samples and 178 late-stage HCC samples in the merge cohort (LIHC-US and LIRI-JP) by utilizing the \u0026ldquo;limma\u0026rdquo; R package. Figure 1 showed the schematic workflow of this study.\u003c/p\u003e\n\u003cp\u003e137 genes were significantly enriched in the early-stage group and 291 genes significantly enriched in the late-stage group (Fig. 2A, |logFC| \u0026gt; 0.5 and AdjustedPvalue\u0026thinsp;\u0026lt;\u0026thinsp;0.05). The heatmap of DEGs also had been presented in \u003cstrong\u003eFig. 2B\u003c/strong\u003e.\u003c/p\u003e\n\u003cp\u003eFunctional analysis of DEGs by GO and KEGG enrichment analysis\u003c/p\u003e\n\u003cp\u003eTo investigate the potential biological significances of DEG, we used clusterProfiler and GSVA R package to characterize the GO and KEGG pathway. The GO analysis showed that EEGs involved with response to xenobiotic stimulus, steroid metabolic process, cytoplasmic vesicle lumen, high\u0026thinsp;\u0026minus;\u0026thinsp;density lipoprotein particle, endopeptidase regulator activity, steroid hydroxylase activity, and so on (Fig. 3A).\u003c/p\u003e\n\u003cp\u003eMoreover, p53 signaling pathway, ECM\u0026thinsp;\u0026minus;\u0026thinsp;receptor interaction, ABC transporters, Steroid hormone biosynthesis, and PPAR signaling pathway were identified by GSEA (Fig. 3B). In particular, nitrogen metabolism, one carbon pool by folate, taurine and hypotaurine metabolism and tryptophan metabolism were mainly enriched in the HCC early stage (Fig. 3C). Inversely, cell cycle, mismatch repair, p53 signaling pathway, and sphingolipid metabolism were significantly enriched in the HCC late stage (Fig. 3D). Additionally, GSVA results showed ABC transporters was enriched in the HCC late stage (\u003cstrong\u003eSupplemental Fig. 2\u003c/strong\u003e).\u003c/p\u003e\n\u003cp\u003eSignificant changes between two stages in immune cells by ImmuneCellAI\u003c/p\u003e\n\u003cp\u003eWith the ImmuCellAI, a latest immune cell infiltration algorithm, we obtained the immune cell expression matrix from the whole gene expression matrix of the merge dataset (Fig. 4A). Not only Tfh cells and Th2 cells of the HCC early stage were significantly higher than the HCC late stage, but also CD8 na\u0026iuml;ve cells, Monocyte and Neutrophil were enriched in the HCC early stage. In the opposite, the HCC late stage showed a higher proportion of B cells compared to HCC early stage (Fig. 4B). The heatmap of immune cells also showed the tendency between two stages in samples (Fig. 4C). Furthermore, B cells were negatively correlated with most immune cells including CD8 na\u0026iuml;ve cells, Tr1 cells and Monocyte (Fig. 4D). On the other hand, Tfh cells, Th2 cells, and Monocyte were positively correlated with CD8 na\u0026iuml;ve cells (Fig. 4D).\u003c/p\u003e\n\u003cp\u003ePotential gene biomarkers identified by multiple machine learning approaches\u003c/p\u003e\n\u003cp\u003eTo find the potential gene biomarkers to distinguish the early- and late-stage of HCC, we utilized three machine learning algorithms to identify the diagnostic markers from the DEGs. 20 potential DEGs were identified by the least absolute shrinkage and selection operator (LASSO) logistic regression algorithm (\u003cstrong\u003eSupplemental Fig. 3A\u003c/strong\u003e). What\u0026rsquo;s more, we used the 20 biomarkers to build LASSO logistic regression diagnostic model, whose performance was pretty good (AUC\u0026thinsp;=\u0026thinsp;0.732, \u003cstrong\u003eSupplemental Fig. 3B\u003c/strong\u003e). Of the 20 biomarkers, 7 DEGs had positive coefficients and 13 DEGs had negative coefficients (\u003cstrong\u003eSupplemental Fig. 3B, Supplemental Table 2\u003c/strong\u003e). The Boruta algorithm was used to classify 39 features from the DEGs (\u003cstrong\u003eSupplemental Fig. 3D\u003c/strong\u003e). Moreover, the random forest diagnostic model based on the 39 biomarkers had also well performance (AUC\u0026thinsp;=\u0026thinsp;0.704, \u003cstrong\u003eSupplemental Fig. 3E\u003c/strong\u003e). \u003cem\u003eSLC6A8\u003c/em\u003e gene had the highest importance score (Mean decrease accuracy from random forest) in \u003cstrong\u003eSupplemental Fig. 3F\u003c/strong\u003e (\u003cstrong\u003eSupplemental Table 3\u003c/strong\u003e). Another machine learning algorithm named the support vector machine-recursive feature elimination (SVM-RFE) to identify 41 features (\u003cstrong\u003eSupplemental Fig. 3G, Supplemental Table 4\u003c/strong\u003e). and the performance of the diagnostic model for distinguish the HCC early stage and the HCC late stage was also well (AUC\u0026thinsp;=\u0026thinsp;0.698, \u003cstrong\u003eSupplemental Fig. 3H\u003c/strong\u003e).\u003c/p\u003e\n\u003cp\u003eNext, 10 intersection biomarkers including \u003cem\u003eSLC6A8\u003c/em\u003e, \u003cem\u003eFTCD\u003c/em\u003e, \u003cem\u003eCYP2C9\u003c/em\u003e, \u003cem\u003eANGPT2\u003c/em\u003e, \u003cem\u003eENO1\u003c/em\u003e, \u003cem\u003eCNGA1\u003c/em\u003e, \u003cem\u003eKCNJ15\u003c/em\u003e, \u003cem\u003eSLC39A4\u003c/em\u003e, \u003cem\u003eETV1\u003c/em\u003e, and \u003cem\u003eACSL6\u003c/em\u003e were extracted from abovementioned features (Fig. 5A). Sequentially, we validated the gene expression level of the 10 biomarkers in the validation GSE14520 dataset and found only 6 genes (\u003cem\u003eSLC6A8\u003c/em\u003e, \u003cem\u003eFTCD\u003c/em\u003e, \u003cem\u003eCYP2C9\u003c/em\u003e, \u003cem\u003eANGPT2\u003c/em\u003e, \u003cem\u003eENO1\u003c/em\u003e, \u003cem\u003eCNGA1\u003c/em\u003e) had the consistent results with the merge dataset (\u003cstrong\u003eSupplemental Fig. 4A-B\u003c/strong\u003e). 3 genes (\u003cem\u003eFTCD\u003c/em\u003e, \u003cem\u003eCYP2C9\u003c/em\u003e, \u003cem\u003eCNGA1\u003c/em\u003e) were significantly enriched in the HCC early stage, while other 3 genes (\u003cem\u003eSLC6A8\u003c/em\u003e, \u003cem\u003eANGPT2\u003c/em\u003e, \u003cem\u003eENO1\u003c/em\u003e) were significantly higher in the HCC late stage in both merge dataset and validation dataset (\u003cstrong\u003eSupplemental Fig. 4A-B\u003c/strong\u003e).\u003c/p\u003e\n\u003cp\u003eTo find the diagnostic makers from the 6 genes, we focused on the \u003cem\u003eSLC6A8\u003c/em\u003e gene which had higher expression in the HCC late stage (Fig. 5B-C) and performed the ROC analysis to check whether the \u003cem\u003eSLC6A8\u003c/em\u003e gene had well performance of diagnostic validity in both merge dataset. Finally, the AUC of ROC validation were 0.654 (Fig. 5D) and 0.701 (Fig. 5E) in the merge dataset and validation dataset, respectively, indicating the \u003cem\u003eSLC6A8\u003c/em\u003e gene had not bad performance as a diagnostic biomarker.\u003c/p\u003e\n\u003cp\u003eCorrelation analysis between \u003cem\u003eSLC6A8\u003c/em\u003e and immune cells\u003c/p\u003e\n\u003cp\u003eWe used spearman correlation to characterize the association between 6 genes (\u003cem\u003eSLC6A8\u003c/em\u003e, \u003cem\u003eFTCD\u003c/em\u003e, \u003cem\u003eCYP2C9\u003c/em\u003e, \u003cem\u003eANGPT2\u003c/em\u003e, \u003cem\u003eENO1\u003c/em\u003e, \u003cem\u003eCNGA1\u003c/em\u003e) and infiltrating immune cells. The gene expression of most these genes had strong correlation with the 21 immune cells (Fig. 6A). \u003cem\u003eSLC6A8\u003c/em\u003e was significantly associated with 15 immune cells including DC, Exhausted, Th17 and CD8 na\u0026iuml;ve. Especially, the DC and CD8 na\u0026iuml;ve had the highest correlation coefficient with \u003cem\u003eSLC6A8\u003c/em\u003e (Fig. 6A). \u003cem\u003eSLC6A8\u003c/em\u003e had significantly positive correlation with DC cells (r\u0026thinsp;=\u0026thinsp;0.284, p\u0026thinsp;=\u0026thinsp;3.9e-12, \u003cstrong\u003eFig. 6C\u003c/strong\u003e) and other cells (\u003cstrong\u003eSupplemental Fig. 5A-I\u003c/strong\u003e), and significantly negative correlation with CD8 na\u0026iuml;ve (r = -0.22, p\u0026thinsp;=\u0026thinsp;1e-07, \u003cstrong\u003eFig. 6D\u003c/strong\u003e) and other cells (\u003cstrong\u003eSupplemental Fig. 5J-L\u003c/strong\u003e).\u003c/p\u003e\n\u003cp\u003eExpression level of \u003cem\u003eSLC6A8\u003c/em\u003e in single-cell transcriptomic data\u003c/p\u003e\n\u003cp\u003eThe cell types of GSE149614 dataset were showed by using t-SNE based on the Seurat-class object with the 2000 highly variable genes after data processing including normalization, scaling, clustering and so on (Fig. 7A). The expression of individual cells and each cell types for \u003cem\u003eSLC6A8\u003c/em\u003e are displayed in the \u003cstrong\u003eFig. 7B-C\u003c/strong\u003e, respectively. Interesting, Hepatocyte, which was the dominant cells in liver presented the highest expression level of \u003cem\u003eSLC6A8\u003c/em\u003e among 6 main cell types (Fig. 7C). Following, we separated cells according to the tumor stage of patients into two groups (HCC early stage and HCC late stage) to validate whether the HCC enhanced \u003cem\u003eSLC6A8\u003c/em\u003e gene expression, which had been discovery in the bulk-seq RNA dataset. As expected, \u003cem\u003eSLC6A8\u003c/em\u003e gene had significantly higher expression (\u003cem\u003eP\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;2e-16) in the HCC late stage compared to the HCC early stage in Hepatocyte (Fig. 7D) by Wilcoxon rank-sum test, which was consistent with the previous analysis.\u003c/p\u003e\n\u003cp\u003eSubsequently, we detected the expression of SLC6A8 in patients with early and late liver cancer. Compared to early-stage HCC patients, the expression of SLC6A8 was higher in late-stage HCC patients (Fig. 7E). This result indicated that the expression level of SLC6A8 was related to the classification of liver cancer.\u0026nbsp;\u003c/p\u003e"},{"header":"Discussion","content":"\u003cp\u003eHCC with high morbidity and mora rate has become a leading cause of cancer-related death worldwide [\u003cspan citationid=\"CR36\" class=\"CitationRef\"\u003e36\u003c/span\u003e]. Furthermore, HCC was diagnosed at the late tumor stage, which missing the appropriate treatment options [\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e]. Therefore, it is crucial for patient management to predict the disease at an early stage. Hence, the development of gene biomarkers that distinguish the early stage from the late stage of HCC is the primary goal of this study. To explore the potential biomarkers in the prediction of the early stage, we used multiple machine learning algorithms to compare the early stage and late stage of HCC based on the differentially expressed genes (DEGs). Apart from that, we also investigated the biological function of DEGs and immune cells between the early stage and the late stage.\u003c/p\u003e \u003cp\u003eIn this study, 137 down-regulated genes and 291 up-regulated genes were identified, and some of DEGs were enriched in p53 signaling pathway and tryptophan metabolism, which are cancer-related pathway [12; 23]. Interestingly, p53 signaling pathway with p53 protein mutations resulting in uncontrolled cell proliferation and cancer tumors [\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e], was enriched in the late stage of HCC, suggesting that there was more the loss of tumor-suppressing function at the late stage of HCC. Inversely, tryptophan metabolism promoting tumor cell intrinsic malignant properties as well as restricts antitumor immunity [\u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e], was enriched in the early stage of HCC, indicating that the immune system had initially changed in the early stage. Moreover, the immune infiltration analysis also found that the immunity between two stages was significantly different, such as Tfh (T follicular helper) cells, providing essential help to B cells for effective antibody-mediated immune responses [\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e], and Th2, facilitating tissue repair [\u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e31\u003c/span\u003e] were both enriched in the early stage.\u003c/p\u003e \u003cp\u003eTen DEGs from the intersection of genes i.e \u003cem\u003eSLC6A8\u003c/em\u003e, \u003cem\u003eFTCD\u003c/em\u003e, \u003cem\u003eCYP2C9\u003c/em\u003e, \u003cem\u003eANGPT2\u003c/em\u003e, \u003cem\u003eENO1\u003c/em\u003e, \u003cem\u003eCNGA1\u003c/em\u003e, \u003cem\u003eKCNJ15\u003c/em\u003e, \u003cem\u003eSLC39A4\u003c/em\u003e, \u003cem\u003eETV1\u003c/em\u003e, and \u003cem\u003eACSL6\u003c/em\u003e were screened by using LASSO analysis, RF-Boruta and SVM-REF. Previously, different reports uncover their involvement in a variety of malignancies [3; 10; 11; 32; 39; 40]. From comparing their expression value between two stages of HCC in the merge and validation cohort, we found \u003cem\u003eSLC6A8\u003c/em\u003e, \u003cem\u003eANGPT2\u003c/em\u003e and \u003cem\u003eENO1\u003c/em\u003e were observed as up-regulated genes in the late stage. Then, we used ROC curves to assess the efficacy of \u003cem\u003eSLC6A8\u003c/em\u003e on the diagnosis of HCC. The AUC of \u003cem\u003eSLC6A8\u003c/em\u003e were 0.654 and 0.701 in the merge cohort and the validation cohort, suggesting that \u003cem\u003eSLC6A8\u003c/em\u003e is a potential diagnostic biomarker to distinguish the early stage and the late stage of HCC. Especially, it has been shown that \u003cem\u003eSLC6A8\u003c/em\u003e is associated with the initiation and progression of human cancers [\u003cspan citationid=\"CR37\" class=\"CitationRef\"\u003e37\u003c/span\u003e]. Evidently, another study also demonstrated that \u003cem\u003eSLC6A8\u003c/em\u003e knockdown suppresses the invasion and migration of HCC [\u003cspan citationid=\"CR38\" class=\"CitationRef\"\u003e38\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eThere was different immune capacity between two stages of HCC, but whether the immunity associated with the \u003cem\u003eSLC6A8\u003c/em\u003e is still unknown. To further analyze these relationships, we used spearman correlation coefficient and found that the up-regulated gene \u003cem\u003eSLC6A8\u003c/em\u003e was positively associated with DC (dendritic cells), which provide antigens and co-stimulatory signals to cells of the adaptive immune system [\u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e], and negatively associated with CD8-na\u0026iuml;ve cells (na\u0026iuml;ve CD8\u0026thinsp;+\u0026thinsp;T cells), which selectively detect and eradicate cancer cells by targeting the antigens including tumor-specific neoantigens and self-antigens from tumors [\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e]. In our study, other immune cells also associated with gene \u003cem\u003eSLC6A8\u003c/em\u003e, inferring that \u003cem\u003eSLC6A8\u003c/em\u003e might affect the tumor progression by regulating the immune cells. In the literature, it has revealed the positive relationship between DC cells and SLC6A8-mediated creatine transport [\u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e]. Another recent study also demonstrated that \u003cem\u003eSLC6A8\u003c/em\u003e may be involved in the development of cancer by participating in the Notch signaling pathway, which playing important role in the specification of the immune cells [\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e] in the NSCLC [\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eWe also investigated whether the expression of gene \u003cem\u003eSLC6A8\u003c/em\u003e at single-cell transcript level and \u003cem\u003ein vitro\u003c/em\u003e trial was the same as bulk-RNA sequencing level. Interestingly, six cell types including Hepatocyte which is the main functional cells of the live [\u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e] were identified in the liver tissue which was consistent with the previous studies [\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e]. Furthermore, we found that the most important live cell Hepatocyte cells had higher values of \u003cem\u003eSLC6A8\u003c/em\u003e in the late stage of HCC. At the same time, the Myeloid cell also had slightly significantly higher values of \u003cem\u003eSLC6A8\u003c/em\u003e in the late stage of HCC. Additionally, the \u003cem\u003ein vitro\u003c/em\u003e trial also showed the obviously significant expression changes of \u003cem\u003eSLC6A8\u003c/em\u003e between two groups. With the repeated verification of discovery on the \u003cem\u003eSLC6A8\u003c/em\u003e, we believe the gene \u003cem\u003eSLC6A8\u003c/em\u003e may be the potential diagnostic marker.\u003c/p\u003e \u003cp\u003eIn conclusion, we demonstrated that gene \u003cem\u003eSLC6A8\u003c/em\u003e was significantly up-regulated in the late stage of HCC based on the transcriptomic data. In addition, gene \u003cem\u003eSLC6A8\u003c/em\u003e was associated with immune cell infiltration, which provides a potential target for more precise and personalized immunotherapy. Therefore, the gene \u003cem\u003eSLC6A8\u003c/em\u003e may be a potential diagnostic marker.\u003c/p\u003e"},{"header":"Conclusion","content":"\u003cp\u003eIn summary, we identified the DEGs by comparing the early stage and the late stage of HCC and found that they were associated with the cancer-related pathway and \u003cem\u003eSLC6A8\u003c/em\u003e, a potential diagnostic biomarker for clinical diagnosis between the early stage and the late stage of HCC was verified not only in the single-cell transcriptomic data but also \u003cem\u003ein vitro trial\u003c/em\u003e.\u003c/p\u003e\n\u003ch3\u003eLimitation of the study\u003c/h3\u003e\n\u003cp\u003eWe have identified potential diagnostic biomarker of the early stage and the late stage of HCC based on the transcriptomic data and verified it in the single-cell transcriptomic data and \u003cem\u003ein vitro\u003c/em\u003e trial in our study. One of the limitations is that the sample size of our merge data may be not very large. But we have integrated the two most influential research (LIHC-US and LIRI-JP) with more than 100 samples into the merge dataset. Furthermore, this study only focused on biomarker exploration in the transcriptomic levels. The future study could develop the combination biomarkers from the genomic, epigenetic and metabolomic data.\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003eAuthor contributions\u003c/p\u003e\n\u003cp\u003eWang Yang conducted the study design; Linlin Song conducted data collection and data analyses; Wang Yang did the Manuscript writing; Linlin Song and Wang Yang conducted a Manuscript review and revise.\u003c/p\u003e\n\u003cp\u003eStatements and declarations\u003c/p\u003e\n\u003cp\u003eNot applicable.\u003c/p\u003e\n\u003cp\u003eDeclaration of competing interest\u003c/p\u003e\n\u003cp\u003eNot applicable.\u003c/p\u003e\n\u003cp\u003eConsent for publication\u003c/p\u003e\n\u003cp\u003eNot applicable.\u003c/p\u003e\n\u003cp\u003eData availability\u003c/p\u003e\n\u003cp\u003eThe datasets used and analyzed during the currentstudy are available from the corresponding authors onreasonable request.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n\u003cli\u003eBRAY, F., FERLAY, J., SOERJOMATARAM, I., SIEGEL, R.L., TORRE, L.A., and JEMAL, A., 2018. Global cancer statistics 2018: GLOBOCAN estimates of incidence and mortality worldwide for 36 cancers in 185 countries. \u003cem\u003eCa-a Cancer Journal for Clinicians 68\u003c/em\u003e, 6 (Nov-Dec), 394-424. DOI= http://dx.doi.org/10.3322/caac.21492.\u003c/li\u003e\n\u003cli\u003eFENG, S., WANG, J., WANG, L., QIU, Q., CHEN, D., SU, H., LI, X., XIAO, Y., and LIN, C., 2023. Current Status and Analysis of Machine Learning in Hepatocellular Carcinoma. \u003cem\u003eJ Clin Transl Hepatol 11\u003c/em\u003e, 5 (Oct 28), 1184-1191. DOI= http://dx.doi.org/10.14218/jcth.2022.00077s.\u003c/li\u003e\n\u003cli\u003eFENG, Y., GUO, X., and TANG, H., 2021. SLC6A8 is involved in the progression of non-small cell lung cancer through the Notch signaling pathway. \u003cem\u003eAnn Transl Med 9\u003c/em\u003e, 3 (Feb), 264. DOI= http://dx.doi.org/10.21037/atm-20-5984.\u003c/li\u003e\n\u003cli\u003eFRIEDMAN, J., HASTIE, T., and TIBSHIRANI, R., 2010. Regularization Paths for Generalized Linear Models via Coordinate Descent. \u003cem\u003eJ Stat Softw 33\u003c/em\u003e, 1, 1-22.\u003c/li\u003e\n\u003cli\u003eGARIS, M. and GARRETT-SINHA, L.A., 2020. Notch Signaling in B Cell Immune Responses. \u003cem\u003eFront Immunol 11\u003c/em\u003e, 609324. DOI= http://dx.doi.org/10.3389/fimmu.2020.609324.\u003c/li\u003e\n\u003cli\u003eGUTI\u0026eacute;RREZ-MELO, N. and BAUMJOHANN, D., 2023. T follicular helper cells in cancer. \u003cem\u003eTrends Cancer 9\u003c/em\u003e, 4 (Apr), 309-325. DOI= http://dx.doi.org/10.1016/j.trecan.2022.12.007.\u003c/li\u003e\n\u003cli\u003eHANG, D., YANG, X.L., LU, J.Y., SHEN, C., DAI, J.C., LU, X.F., JIN, G.F., HU, Z.B., GU, D.F., MA, H.X., and SHEN, H.B., 2022. Untargeted plasma metabolomics for risk prediction of hepatocellular carcinoma: A prospective study in two Chinese cohorts. \u003cem\u003eInternational Journal of Cancer 151\u003c/em\u003e, 12 (Dec 15), 2144-2154. DOI= http://dx.doi.org/10.1002/ijc.34229.\u003c/li\u003e\n\u003cli\u003eH\u0026auml;NZELMANN, S., CASTELO, R., and GUINNEY, J., 2013. GSVA: gene set variation analysis for microarray and RNA-seq data. \u003cem\u003eBMC Bioinformatics 14\u003c/em\u003e(Jan 16), 7. DOI= http://dx.doi.org/10.1186/1471-2105-14-7.\u003c/li\u003e\n\u003cli\u003eHOSHIDA, Y., NIJMAN, S.M., KOBAYASHI, M., CHAN, J.A., BRUNET, J.P., CHIANG, D.Y., VILLANUEVA, A., NEWELL, P., IKEDA, K., HASHIMOTO, M., WATANABE, G., GABRIEL, S., FRIEDMAN, S.L., KUMADA, H., LLOVET, J.M., and GOLUB, T.R., 2009. Integrative transcriptome analysis reveals common molecular subclasses of human hepatocellular carcinoma. \u003cem\u003eCancer Research 69\u003c/em\u003e, 18 (Sep 15), 7385-7392. DOI= http://dx.doi.org/10.1158/0008-5472.Can-09-1089.\u003c/li\u003e\n\u003cli\u003eHUA, H., PAN, S., DIAO, H., CAO, Y., QIAN, X., and ZHANG, J., 2024. Increased ACSL6 Expression Predicts a Favorable Prognosis in Triple-negative Breast Cancer. \u003cem\u003eCurr Med Chem\u003c/em\u003e(Jan 17). DOI= http://dx.doi.org/10.2174/0109298673278846231222103420.\u003c/li\u003e\n\u003cli\u003eHUANG, C.K., SUN, Y., LV, L., and PING, Y., 2022. ENO1 and Cancer. \u003cem\u003eMol Ther Oncolytics 24\u003c/em\u003e(Mar 17), 288-298. DOI= http://dx.doi.org/10.1016/j.omto.2021.12.026.\u003c/li\u003e\n\u003cli\u003eKODAMA, M., HANASAWA, K., and TANI, T., 1997. Blood purification for critical care medicine: endotoxin adsorption. \u003cem\u003eTher Apher 1\u003c/em\u003e, 3 (Aug), 224-227. DOI= http://dx.doi.org/10.1111/j.1744-9987.1997.tb00142.x.\u003c/li\u003e\n\u003cli\u003eLI, W.H., WANG, Q.W., LU, J.X., ZHAO, B., GENG, Y.Q., WU, X.W., and CHEN, X.L., 2023. Machine learning-based prognostic modeling of lysosome-related genes for predicting prognosis and immune status of patients with hepatocellular carcinoma. \u003cem\u003eFrontiers in Immunology 14\u003c/em\u003e(May 19). DOI= http://dx.doi.org/ARTN 116925610.3389/fimmu.2023.1169256.\u003c/li\u003e\n\u003cli\u003eLIVER, E.A.S., 2018. EASL Clinical Practice Guidelines: Management of hepatocellular carcinoma. \u003cem\u003eJournal of Hepatology 69\u003c/em\u003e, 1 (Jul), 182-236.\u003c/li\u003e\n\u003cli\u003eLU, C., RONG, D.W., ZHANG, B., ZHENG, W.B., WANG, X.H., CHEN, Z.Y., and TANG, W.W., 2019. Current perspectives on the immunosuppressive tumor microenvironment in hepatocellular carcinoma: challenges and opportunities. \u003cem\u003eMolecular Cancer 18\u003c/em\u003e, 1 (Aug 29). DOI= http://dx.doi.org/ARTN 13010.1186/s12943-019-1047-6.\u003c/li\u003e\n\u003cli\u003eLU, J., CHEN, Y., ZHANG, X., GUO, J., XU, K., and LI, L., 2022. A novel prognostic model based on single-cell RNA sequencing data for hepatocellular carcinoma. \u003cem\u003eCancer Cell Int 22\u003c/em\u003e, 1 (Jan 25), 38. DOI= http://dx.doi.org/10.1186/s12935-022-02469-2.\u003c/li\u003e\n\u003cli\u003eLU, Y., YANG, A., QUAN, C., PAN, Y., ZHANG, H., LI, Y., GAO, C., LU, H., WANG, X., CAO, P., CHEN, H., LU, S., and ZHOU, G., 2022. A single-cell atlas of the multicellular ecosystem of primary and metastatic hepatocellular carcinoma. \u003cem\u003eNat Commun 13\u003c/em\u003e, 1 (Aug 6), 4594. DOI= http://dx.doi.org/10.1038/s41467-022-32283-3.\u003c/li\u003e\n\u003cli\u003eMECHAM, B.H., NELSON, P.S., and STOREY, J.D., 2010. Supervised normalization of microarrays. \u003cem\u003eBioinformatics 26\u003c/em\u003e, 10 (May 15), 1308-1315. DOI= http://dx.doi.org/10.1093/bioinformatics/btq118.\u003c/li\u003e\n\u003cli\u003eMIAO, Y.R., ZHANG, Q., LEI, Q., LUO, M., XIE, G.Y., WANG, H.X., and GUO, A.Y., 2020. ImmuCellAI: A Unique Method for Comprehensive T-Cell Subsets Abundance Prediction and its Application in Cancer Immunotherapy. \u003cem\u003eAdvanced Science 7\u003c/em\u003e, 7 (Apr). DOI= http://dx.doi.org/ARTN 190288010.1002/advs.201902880.\u003c/li\u003e\n\u003cli\u003eOLSEN, T.K. and BARYAWNO, N., 2018. Introduction to Single-Cell RNA Sequencing. \u003cem\u003eCurr Protoc Mol Biol 122\u003c/em\u003e, 1 (Apr), e57. DOI= http://dx.doi.org/10.1002/cpmb.57.\u003c/li\u003e\n\u003cli\u003ePAPAGIANNOPOULOS, P., ANDRIANOPOULOS, A., SHARP, J.A., DAVIS, M.A., and HYNES, M.J., 1996. The hapC gene of Aspergillus nidulans is involved in the expression of CCAAT-containing promoters. \u003cem\u003eMol Gen Genet 251\u003c/em\u003e, 4 (Jun 24), 412-421. DOI= http://dx.doi.org/10.1007/bf02172369.\u003c/li\u003e\n\u003cli\u003ePHILIP, M. and SCHIETINGER, A., 2022. CD8(+) T cell differentiation and dysfunction in cancer. \u003cem\u003eNat Rev Immunol 22\u003c/em\u003e, 4 (Apr), 209-223. DOI= http://dx.doi.org/10.1038/s41577-021-00574-3.\u003c/li\u003e\n\u003cli\u003ePLATTEN, M., NOLLEN, E.A.A., R\u0026ouml;HRIG, U.F., FALLARINO, F., and OPITZ, C.A., 2019. Tryptophan metabolism as a common therapeutic target in cancer, neurodegeneration and beyond. \u003cem\u003eNat Rev Drug Discov 18\u003c/em\u003e, 5 (May), 379-401. DOI= http://dx.doi.org/10.1038/s41573-019-0016-5.\u003c/li\u003e\n\u003cli\u003ePU, W. and ZHOU, B., 2022. Hepatocyte generation in liver homeostasis, repair, and regeneration. \u003cem\u003eCell Regen 11\u003c/em\u003e, 1 (Jan 6), 2. DOI= http://dx.doi.org/10.1186/s13619-021-00101-8.\u003c/li\u003e\n\u003cli\u003eREBOUISSOU, S. and NAULT, J.C., 2020. Advances in molecular classification and precision oncology in hepatocellular carcinoma. \u003cem\u003eJournal of Hepatology 72\u003c/em\u003e, 2 (Feb), 215-229. DOI= http://dx.doi.org/10.1016/j.jhep.2019.08.017.\u003c/li\u003e\n\u003cli\u003eRIEDER, M.J., O\u0026apos;DROBINAK, D.M., TONELLATO, P.J., and GREENE, A.S., 1996. Mathematical analysis of type-I and type-IIb muscle fiber force generation in renal hypertension. \u003cem\u003eAnn Biomed Eng 24\u003c/em\u003e, 4 (Jul-Aug), 489-499. DOI= http://dx.doi.org/10.1007/bf02648111.\u003c/li\u003e\n\u003cli\u003eRITCHIE, M.E., PHIPSON, B., WU, D., HU, Y.F., LAW, C.W., SHI, W., and SMYTH, G.K., 2015. powers differential expression analyses for RNA-sequencing and microarray studies. \u003cem\u003eNucleic Acids Research 43\u003c/em\u003e, 7 (Apr 20). DOI= http://dx.doi.org/ARTN e4710.1093/nar/gkv007.\u003c/li\u003e\n\u003cli\u003eSHORTMAN, K. and NAIK, S.H., 2007. Steady-state and inflammatory dendritic-cell development. \u003cem\u003eNat Rev Immunol 7\u003c/em\u003e, 1 (Jan), 19-30. DOI= http://dx.doi.org/10.1038/nri1996.\u003c/li\u003e\n\u003cli\u003eTIAN, C., YUAN, H., LU, Y., HE, H., LI, Q., LI, S., YANG, J., WANG, M., XU, R., LIU, Q., and XIANG, M., 2023. CARD9 deficiency promotes pancreatic cancer growth by blocking dendritic cell maturation via SLC6A8-mediated creatine transport. \u003cem\u003eOncoimmunology 12\u003c/em\u003e, 1, 2204015. DOI= http://dx.doi.org/10.1080/2162402x.2023.2204015.\u003c/li\u003e\n\u003cli\u003eTORROJA, C. and SANCHEZ-CABO, F., 2019. Digitaldlsorter: Deep-Learning on scRNA-Seq to Deconvolute Gene Expression Data. \u003cem\u003eFrontiers in Genetics 10\u003c/em\u003e(Oct 25). DOI= http://dx.doi.org/ARTN 97810.3389/fgene.2019.00978.\u003c/li\u003e\n\u003cli\u003eWALKER, J.A. and MCKENZIE, A.N.J., 2018. T(H)2 cell development and function. \u003cem\u003eNat Rev Immunol 18\u003c/em\u003e, 2 (Feb), 121-133. DOI= http://dx.doi.org/10.1038/nri.2017.118.\u003c/li\u003e\n\u003cli\u003eWANG, S., ZHOU, Y., YU, R., LING, J., LI, B., YANG, C., CHENG, Z., QIAN, R., LIN, Z., YU, C., ZHENG, J., ZHENG, X., JIA, Q., WU, W., WU, Q., CHEN, M., YUAN, S., DONG, W., SHI, Y., JANSEN, R., YANG, C., HAO, Y., YAO, M., QIN, W., and JIN, H., 2023. Loss of hepatic FTCD promotes lipid accumulation and hepatocarcinogenesis by upregulating PPAR\u0026gamma; and SREBP2. \u003cem\u003eJHEP Rep 5\u003c/em\u003e, 10 (Oct), 100843. DOI= http://dx.doi.org/10.1016/j.jhepr.2023.100843.\u003c/li\u003e\n\u003cli\u003eWEI, Q.X., TAN, N.R., XIONG, S.Y., LUO, W.R., XIA, H.Y., and LUO, B.M., 2023. Deep Learning Methods in Medical Image-Based Hepatocellular Carcinoma Diagnosis: A Systematic Review and Meta-Analysis. \u003cem\u003eCancers 15\u003c/em\u003e, 23 (Dec). DOI= http://dx.doi.org/ARTN 570110.3390/cancers15235701.\u003c/li\u003e\n\u003cli\u003eWU, T., HU, E., XU, S., CHEN, M., GUO, P., DAI, Z., FENG, T., ZHOU, L., TANG, W., ZHAN, L., FU, X., LIU, S., BO, X., and YU, G., 2021. clusterProfiler 4.0: A universal enrichment tool for interpreting omics data. \u003cem\u003eInnovation (Camb) 2\u003c/em\u003e, 3 (Aug 28), 100141. DOI= http://dx.doi.org/10.1016/j.xinn.2021.100141.\u003c/li\u003e\n\u003cli\u003eYANG, C., HUANG, X.W., LIU, Z.C., QIN, W.X., and WANG, C., 2020. Metabolism-associated molecular classification of hepatocellular carcinoma. \u003cem\u003eMolecular Oncology 14\u003c/em\u003e, 4 (Apr), 896-913. DOI= http://dx.doi.org/10.1002/1878-0261.12639.\u003c/li\u003e\n\u003cli\u003eYANG, J.D., HAINAUT, P., GORES, G.J., AMADOU, A., PLYMOTH, A., and ROBERTS, L.R., 2019. A global view of hepatocellular carcinoma: trends, risk, prevention and management. \u003cem\u003eNature Reviews Gastroenterology \u0026amp; Hepatology 16\u003c/em\u003e, 10 (Oct), 589-604. DOI= http://dx.doi.org/10.1038/s41575-019-0186-y.\u003c/li\u003e\n\u003cli\u003eYANG, X. and LI, Q., 2022. Pan-Cancer Analysis of the Oncogenic and Immunological Role of Solute Carrier Family 6 Member 8 (SLC6A8). \u003cem\u003eFront Genet 13\u003c/em\u003e, 916439. DOI= http://dx.doi.org/10.3389/fgene.2022.916439.\u003c/li\u003e\n\u003cli\u003eYUAN, L., WU, X.J., LI, W.C., ZHUO, C., XU, Z., TAN, C., MA, R., WANG, J., and PU, J., 2020. SLC6A8 Knockdown Suppresses the Invasion and Migration of Human Hepatocellular Carcinoma Huh-7 and Hep3B Cells. \u003cem\u003eTechnol Cancer Res Treat 19\u003c/em\u003e(Jan-Dec), 1533033820983029. DOI= http://dx.doi.org/10.1177/1533033820983029.\u003c/li\u003e\n\u003cli\u003eZHANG, C., LI, Q., CHENG, Y., YANG, X., CHEN, W., HE, K., and CHEN, M., 2022. Variants in CYP2J2 and CYP2C9 Contribute to Susceptibility of Lung Cancer. \u003cem\u003eCurr Cancer Drug Targets\u003c/em\u003e(Nov 14). DOI= http://dx.doi.org/10.2174/1568009623666221114115012.\u003c/li\u003e\n\u003cli\u003eZHU, J., WU, Y., YU, Y., LI, Y., SHEN, J., and ZHANG, R., 2022. MYBL1 induces transcriptional activation of ANGPT2 to promote tumor angiogenesis and confer sorafenib resistance in human hepatocellular carcinoma. \u003cem\u003eCell Death Dis 13\u003c/em\u003e, 8 (Aug 20), 727. DOI= http://dx.doi.org/10.1038/s41419-022-05180-2.\u003c/li\u003e\n\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":"discover-oncology","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"dion","sideBox":"Learn more about [Discover Oncology](https://www.springer.com/12672)","snPcode":"","submissionUrl":"","title":"Discover Oncology","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"Discover Series","inReviewEnabled":true,"inReviewRevisionsEnabled":true},"keywords":"Hepatocellular carcinoma, Multiple machine learning algorithms, SLC6A8, Diagnostic biomarker, Bioinformatics","lastPublishedDoi":"10.21203/rs.3.rs-4570554/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-4570554/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eHepatocellular carcinoma (HCC) is a chronic liver disease characterized by persistent tumor growth, contributing significantly to mortality rates worldwide and presenting a growing global health concern. Consequently, there is an urgent need to develop effective diagnostic and treatment strategies for HCC. This study aims to identify crucial genes for early HCC diagnosis to mitigate disease progression and to investigate differences in immune cell infiltration between early-stage and late-stage HCC. We integrated two published datasets for a comprehensive analysis, identifying 575 DEGs subjected to GSEA to reveal pathways distinguishing early-stage from late-stage HCC. Notably, the gene \u003cem\u003eSLC6A8\u003c/em\u003e emerged as a potential diagnostic biomarker for late-stage HCC through LASSO, SVM-RFE and RF-Boruta analyses. ROC curves for \u003cem\u003eSLC6A8\u003c/em\u003e were utilized to evaluate diagnostic accuracy. The ImmuCellAI algorithm assessed immune cell composition differences between early and late-stage HCC, revealing that \u003cem\u003eSLC6A8\u003c/em\u003e expression positively correlates with resting Tfh cells and Th2, while negatively correlating with B cells, indicating its association with immune cell infiltration patterns. To strengthen our results, we further analyzed \u003cem\u003eSLC6A8\u003c/em\u003e expression using single-cell transcriptome data, confirming significant upregulation in late-stage HCC, particularly in key liver cell types such as Hepatocyte cells. Overall, our study identifies \u003cem\u003eSLC6A8\u003c/em\u003e as a potential marker gene that enhances understanding of HCC diagnosis and therapeutic strategies.\u003c/p\u003e","manuscriptTitle":"Multiple machine learning algorithms identified SLC6A8 as a diagnostic biomarker of the late stage of Hepatocellular carcinoma","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2024-07-12 05:24:18","doi":"10.21203/rs.3.rs-4570554/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"decision","content":"Revision requested","date":"2024-08-28T06:25:13+00:00","index":"","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2024-08-28T03:25:47+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"184883787403577480941435246220589861791","date":"2024-08-08T15:06:18+00:00","index":"hide","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2024-07-07T15:59:13+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"138423638495655993551286051484525225941","date":"2024-07-02T14:30:49+00:00","index":"hide","fulltext":""},{"type":"reviewersInvited","content":"","date":"2024-06-26T13:04:35+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2024-06-21T16:24:23+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2024-06-21T16:24:16+00:00","index":"","fulltext":""},{"type":"submitted","content":"Discover Oncology","date":"2024-06-12T13:21:27+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"discover-oncology","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"dion","sideBox":"Learn more about [Discover Oncology](https://www.springer.com/12672)","snPcode":"","submissionUrl":"","title":"Discover Oncology","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"Discover Series","inReviewEnabled":true,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"4e89092b-28d4-47ba-ae5d-7b435d5c497f","owner":[],"postedDate":"July 12th, 2024","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"under-review","subjectAreas":[],"tags":[],"updatedAt":"2025-04-09T05:08:32+00:00","versionOfRecord":[],"versionCreatedAt":"2024-07-12 05:24:18","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-4570554","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-4570554","identity":"rs-4570554","version":["v1"]},"buildId":"qtupq5eGEP_6zYnWcrvyt","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}