Abstract Background Hepatocellular carcinoma (HCC) is a major global health challenge with high aggressiveness and recurrence rates. Metabolic reprogramming is a cancer hallmark, enabling tumor cells to sustain rapid growth and evade immune surveillance. Several amino acids have been found to undergo metabolic reprogramming in tumors, and thus are potential anti-tumor targets. However, the characterization and implication of lysine metabolic reprogramming in HCC remain largely unexplored. Methods We performed multi-omics profiling, including transcriptomics, proteomics, single-cell omics, immunohistochemistry, and multiplex immunofluorescence on tumor and adjacent normal tissues obtained from 30 HCC patients. Integrative analyses and quantitative evaluations were carried out to characterize the lysine metabolism and investigate its implications for tumor progression, immune microenvironment, and immunotherapy responses. Results Our analysis observed a significant downregulation of lysine metabolism in HCC, with inter-patient heterogeneity. Patients with low lysine metabolism in tumors exhibited worse prognoses and a predominance of immunosuppressive tumor immune microenvironment (TIME), characterized by increased infiltration of myeloid-derived suppressor cells (MDSCs), regulatory T cells (Tregs), and exhausted CD8^+ T cells (TIM3^+CD8^+ and LAG3^+CD8^+). These immunosuppressive cells contribute to immunotherapeutic resistance and promote tumor progression. Notably, our conclusions were consistently supported by observations at both the bulk and single-cell resolutions, as well as T cell receptor (TCR) immune repertoire profiling, reinforcing the robustness of our findings. Conclusions This study provides comprehensive evidence that lysine metabolism plays a critical role in shaping the immunosuppressive TIME in HCC and is associated with clinical outcomes and resistance to immunotherapy, offering new insights into clinical molecular subtyping and potential therapeutic strategies. Graphical abstract [44]graphic file with name 12967_2025_7056_Figa_HTML.jpg Supplementary Information The online version contains supplementary material available at 10.1186/s12967-025-07056-3. Keywords: Hepatocellular carcinoma, Lysine metabolism, Tumor immune microenvironment, Immunotherapy Introduction Hepatocellular carcinoma (HCC) is one of the most prevalent and deadly cancers worldwide. According to the 2022 Global Cancer Statistics report, it ranked as the sixth most diagnosed cancer and the third leading cause of cancer-related deaths globally [[45]1]. HCC accounts for about 90% of liver cancer cases, with incidence and mortality rates increasing by approximately 27% and 25%, respectively, in the past decade. In regions like China, where the hepatitis B virus (HBV) is endemic, HCC represents a major health burden, contributing to 45% of global liver cancer cases and deaths. HCC is characterized by high mortality, recurrence, and poorer prognosis. Surgical treatment has long been considered the primary therapeutic approach for patients diagnosed at an early stage with better liver function. However, the significantly high postoperative recurrence rate has substantially diminished the effectiveness of surgical treatment [[46]2]. HCC is also resistant to traditional chemotherapy and radiotherapy, with most patients diagnosed at an advanced stage [[47]3]. Although immune checkpoint inhibitors (ICIs) are now used for advanced HCC, survival benefits remain limited due to the tumor's immunosuppressive microenvironment [[48]4]. This tumor immune environment (TME), influenced by myeloid-derived suppressor cells (MDSCs), tumor-associated macrophages (TAMs), and other immunosuppressive cells, plays a critical role in immune escape and therapy resistance [[49]5]. The interaction between tumor cells and immune cells through metabolic reprogramming is a key factor in reshaping tumor immune response. Previous studies have shown that activated immune cells share various metabolic pathways with tumor cells [[50]6]. The immunosuppressive microenvironment mediated by tumor metabolic reprogramming is a significant factor leading to tumor immune escape. Characteristics such as hypoxia and nutrient deprivation in the TME can lead to metabolic competition between tumor cells and immune cells. Additionally, the accumulation of metabolites may affect the expression of immune molecules, thereby influencing immune responses [[51]7–[52]9]. Tumor cells, with their strong metabolic adaptability, reshape the metabolic characteristics of the TME by competing for and consuming nutrients, thereby reducing the metabolic intensity of immune cells and promoting tumor immunosuppression and evasion [[53]10]. Among them, dysregulated glucose and amino acids metabolism in tumor cells represents a major feature of metabolic reprogramming and contributes significantly to tumor immunosuppression and evasion [[54]11]. Amino acids play pivotal roles in processes such as protein synthesis, cell growth, and proliferation. Abnormal amino acid metabolism in tumors can lead to dysfunction of immune cells. Recently, the promotion of immune cell function by amino acid metabolism has been widely studied, and targeting amino acid metabolism can regulate immune responses, thereby enhancing immune efficacy. For example, glutamine is essential for T-cell activation and proliferation [[55]12]. Previous studies have indicated that the loss of branched-chain amino acid (BCAA) degradation metabolism promotes the progression of HCC [[56]13]. However, lysine, an essential amino acid, undergoes metabolic changes in HCC that are largely unexplored [[57]14]. Recent advances in single-cell omics technologies have enabled high-resolution dissection of the TME at the single-cell level, providing new opportunities for understanding cellular heterogeneity and immune regulation in HCC [[58]15]. In this study, we performed bulk and single-cell multi-omics profiling, along with histological analysis of tumors and adjacent normal tissues from 30 HCC patients, to investigate lysine metabolism and its role in tumor progression, immune microenvironment, and immunotherapy responses. Our in-house data revealed significant downregulation of lysine metabolic genes in HCC, which was further confirmed by multi-omics datasets from the TCGA, ICGC, and CPTAC. This downregulation correlates with advanced stages and poor prognosis. ScRNA-seq analysis indicated that the lysine metabolic reprogramming in tumors is primarily attributed to hepatocellular carcinoma cells, rather than other cell types. We also constructed a lysine metabolism score (LM score) to stratify patients into high and low lysine metabolism subtypes. Both bulk and single-cell omics analyses confirmed that lower lysine metabolism was associated with poorer prognosis and survival, characterized by a predominance of immunosuppressive TME and reduced anti-tumor immune responses. Notably, we observed increased infiltration of immunosuppressive cells, including MDSCs and Tregs, as well as TIM3^+ CD8^+ and LAG3^+ CD8^+ T cells in tumor tissues, which contribute to immunotherapeutic tolerance and tumor progression. Additionally, through single-cell RNA-seq and TCR analysis, we further uncovered the antigen-driven nature of CD8^+ Tex and CD4^+ Treg cells, revealing clonally expanded populations that correlate with distinct functional phenotypes. These findings suggest that lysine metabolic reprogramming and immunosuppressive cell infiltration collaboratively promote tumor immune evasion. In summary, our results highlight the critical role of lysine metabolism in HCC, offering new insights into the clinical subtyping and potential treatment strategies. Methods HCC patients and clinical samples As the discovery cohort, a total of 30 pairs of clinical samples were collected from 30 hepatocellular carcinoma (HCC) patients at the First Affiliated Hospital of Xi’an Jiaotong University (Supplementary Tables [59]S1, [60]S2). Each pair of samples includes one tumor tissue and one histologically normal tissue adjacent to the tumor (NAT). All samples were used for Immunohistochemistry (IHC), and six samples were employed for multiplex immunohistochemistry (mIHC) analyses; three pairs were used for single-cell RNA sequencing (scRNA-seq), and an additional pair of samples underwent scRNA-seq coupled with T-cell receptor sequencing (scTCR-seq); three pairs were sent to Biomics Biotech Co., Ltd. (Beijing, China) for both transcriptome sequencing and Tandem Mass Tag (TMT) labeling followed by mass spectrometry analysis, which is detailed in subsequent sections. The collecting, profiling, and sequencing procedures on these clinical samples were conducted following the ethical standards outlined in the Declaration of Helsinki and were approved by the Ethics Committee of Xi’an Jiaotong University (approval number: 2022-490), with all informed consent documents signed by all participants included in the study. RNA extraction, sequencing, and data processing The transcriptomic profiling on the tumor and adjacent normal tissues is processed following the same procedures as below. 1. RNA extraction: Total RNA was extracted using the TRIzol method. 2. Sample detection: The purity (OD260/280), concentration, and nucleic acid absorption peak of the RNA were assessed using a NanoPhotometer® spectrophotometer (IMPLEN, CA, USA). RNA integrity was evaluated using the RNA Nano 6000 Assay Kit of the Bioanalyzer 2100 system (Agilent Technologies, CA, USA). 3. Library construction: Sequencing libraries were generated using the NEBNext® Ultra™ RNA Library Prep Kit for Illumina® (NEB, USA). mRNA was enriched using oligo(dT) magnetic beads. The fragmentation buffer was then added to break the mRNA into short fragments. Using these mRNA fragments as templates, first-strand cDNA was synthesized using random hexamer primers. Second-strand cDNA synthesis was performed using DNA Polymerase I, RNase H, dNTPs, and buffer. The cDNA fragments were purified with the AMPure XP system (Beckman Coulter, Beverly, USA) to select cDNA fragments of 150–200 bp in length, followed by polymerase chain reaction (PCR) to generate the final cDNA library. 4. Library quality control: Initial quantification of the library was performed using Qubit 2.0, and the insert size was evaluated using the Agilent 2100 Bioanalyzer. The accurate concentration of the library was determined using quantitative real-time PCR (qPCR), and the library effective concentration > 2 nM. 5. Sequencing: The libraries were sequenced on the Illumina HiSeq X Ten platform (San Diego, CA, USA). 6. Sequencing quality control: The fastq data were evaluated using FastQC. Quality control was performed using TrimGalore (v0.6.6). In this process, bases overlapping with adapters by three bases, bases with a quality score lower than 25, and reads with a length shorter than 75 bases were removed. 7. Sequence Alignment: The reference genome index was built using STAR (v2.7.4a), and paired-end clean reads were aligned to the reference genome (GRCh38.p13) using STAR (v2.7.4a) [[61]16]. 8. Expression quantification: Gene expression levels for each sample were quantified using StringTie (v2.1.6) [[62]17]. The reads count, TPMs (Transcripts per million), and FPKMs (Fragments Per Kilobase of transcript per Million mapped reads) were calculated [[63]18]. The RNA-seq data described in this article are available at GEO under accession [64]GSE272510. Protein extraction, trypsin digestion The sample was ground into powder by liquid nitrogen. After that, four volumes of 10% Trichloroacetic acid (TCA)/acetone were added to the powder. The samples were then allowed to stand at − 20℃ °C for 4 h followed by centrifuging at 4500 g for 5 min at 4 °C. The supernatant was discarded, and the precipitates were washed three times with pre-cooled acetone. After drying, the precipitates were re-solubilized with 8 M urea. The supernatant was collected and the protein concentration was determined with the BCA kit (Beyotime, Shanghai, China). Equal amounts of protein from each sample were taken for trypsin digestion, with an appropriate amount of standard protein added, and the volume was adjusted to uniformity using lysis buffer. One volume of pre-cooled acetone was added, vortexed, and mixed thoroughly, followed by the addition of four volumes of pre-cooled acetone. The samples were precipitated at − 20 °C for 2 h, centrifuged at 4500 g for 5 min, and the supernatant was discarded. The precipitate was washed twice with pre-cooled acetone. After drying, Tetraethylammonium bromide (TEAB) was added to a final concentration of 200 mM, and the sample was ultrasonicated to break up the precipitate. Trypsin was added at a ratio of 1:50 (protease: protein, m/m) for overnight digestion. DL-Dithiothreitol (DTT) was added to a final concentration of 5 mM, reduced at 56 °C for 30 min. Subsequently, iodoacetic acid (IAA) was added to a final concentration of 11 mM, and the solution was incubated for 15 min at room temperature. Tandem mass tag (TMT) labeling Desalinate the trypsin-digested peptides using a Strata X C18 column (Phenomenex), and then lyophilize them. The peptides were dissolved in 0.5 M TEAB and labeled according to the TMT kit (ThermoFisher Scientific, USA) instructions. The labeling procedure was as follows: the labeling reagent was thawed and dissolved in acetonitrile, mixed with the peptides, and incubated at room temperature for 2 h. The labeled peptides were then mixed, desalted, and lyophilized. HPLC–MS/MS High-performance liquid chromatography-tandem mass spectrometry (HPLC–MS/MS) was used to profile the proteome of the samples. The peptides were fractionated using high-pH reversed-phase HPLC on an Agilent 300Extend C18 column. The peptides were graded on a gradient of 8–32% acetonitrile, pH 9.0, and 60 fractions were separated in 60 min, and then the peptides were merged into 9 fractions. After vacuum freeze-drying, the peptides were dissolved in solvent A and separated using an EASY-nLC 1200 ultra-high-performance liquid chromatography (UHPLC) system. Mobile phase A consisted of 0.1% formic acid and 2% acetonitrile in water. Mobile phase B consisted of 0.1% formic acid and 90% acetonitrile in water. The gradient was set as follows: 0–26 min, 6–25% B; 26–34 min, 25–35% B; 34–37 min, 35–80% B; 37–40 min, 80% B, with a flow rate of 450 nL/min. After separation, peptides were ionized using a nano-electrospray ionization (NSI) source and analyzed with a Q Exactive™ Plus mass spectrometer. Proteomics data search The MS/MS data were searched using MaxQuant (v1.5.2.8). The search parameters were set as follows: the database used was Homo_sapiens_9606 (20,366 sequences), with a reverse database added to calculate the false discovery rate (FDR) caused by random matches, and a common contaminants database included to eliminate the impact of contaminant proteins on the identification results. The enzyme used for digestion was set to Trypsin/P, allowing up to two missed cleavages. The minimum peptide length was set to 7 amino acid residues. The mass tolerance for precursor ions was set as 10 ppm in the First search and 5 ppm in the Main search, and the mass tolerance for fragment ions was set to 0.02 Da. Cysteine Carbamidomethyl(C) was specified as a fixed modification. The quantification method was set to TMT-6plex, and the false discovery rates (FDRs) of protein identification and PSM identification were all set as 1%. The mass spectrometry proteomics data have been deposited in the ProteomeXchange Consortium via the PRIDE partner repository with the dataset identifier PXD053184 ([65]https://proteomecentral.proteomexchange.org/cgi/GetDataset?ID=PXD0 53184). Single-cell RNA-seq library construction and sequencing Samples from each patient were processed in a single batch for library preparation. The cells are washed and resuspended to prepare a suitable cell concentration of 700 ~ 1200 cells/μL for 10 × Genomics Chromium™. For single-cell isolation, Gel Beads in Emulsion (GEMs) were constructed based on the target cell count. Post-oil treatment, the amplified cDNA was purified using magnetic beads, followed by cDNA amplification and quality assessment. The single-cell suspension was then loaded onto a microfluidic device to construct a single-cell 3’ gene expression library. The final library was sequenced on the Illumina Nova6000 platform using 150 bp paired-end reads. The raw and processed datasets for scRNA-seq are available at GEO under accession [66]GSE290298. Single-cell RNA-seq coupled with TCR sequencing The GEMs generate full-length cDNA incorporating 10 × Barcodes. Following the manufacturer’s protocol, TCR enrichment was performed using the TCR Amplification Kit (10 × Genomics, 16rxns, PN-1000252). Full-length V(D)J segments were subsequently amplified via PCR with primers specific to the constant regions of TCR, utilizing the full-length cDNA derived from poly-adenylated mRNA. The scTCR libraries were sequenced on the Illumina Nova6000 using 150 bp paired-end reads. The raw and processed datasets for scTCR-seq are available at GEO under accession [67]GSE290298. scRNA-seq/TCR data processing Cell Ranger (v6.1.2, 10 × Genomics) was used for sequencing reads mapping against the GRCh38 (hg38) human reference genome and unique molecular identifiers (UMIs) counting. Rigorous quality control (QC) procedures were implemented to ensure data reliability. Cells were filtered based on sample-specific quantile thresholds, with retention criteria for RNA feature count (nFeature_RNA) and total RNA count (nCount_RNA) within the 10% to 90% range, and no more than 10% mitochondrial reads were generally allowed per cell. To mitigate batch effects, the Harmony algorithm (v1.1.0) was applied, enabling batch correction across different samples. Seurat (v5.0.3) was then employed for sample integration, clustering, and annotation[[68]19]. The integrated data were visualized using Uniform Manifold Approximation and Projection (UMAP). The AddModuleScore function was utilized to calculate module scores for gene expression programs associated with lysine metabolism and epithelial-mesenchymal transition (EMT) processes in each single cell. Single-cell TCR sequencing data were aligned to the human reference genome GRCh38 (hg38). TCR sequences were then assembled and annotated, and contigs were filtered using the Cell Ranger V(D)J pipeline (v6.1.2, 10 × Genomics). T cells containing matched and functional TCR α and β chains were selected. TCR Clonotypes were identified by the unique combination of complementarity determining region (CDR) amino acid sequences of α and β chains. The Python package Scirpy (v0.20.0) was used for TCR clonotype identification, clonal expansion analysis, and visualization [[69]20]. Immunohistochemistry (IHC) A total of 30 sets of tumor tissue and histologically normal tissue adjacent to the tumor (NAT) from patients with HCC who underwent partial hepatectomy at the First Affiliated Hospital of Xi’an Jiaotong University were subjected to IHC staining (ZSGB-Bio, PV-6002, 2012D0106). The clinicopathological and tumor-related characteristics of the study cohort are provided in Supplementary Tables [70]S1 and [71]S2. The procedure of IHC staining was conducted as follows: (1) Deparaffinization at room temperature using xylene, followed by hydration with distilled water and PBS (3 min × 3 times); (2) Inactivation of endogenous peroxidase activity by incubating with 3% hydrogen peroxide (H[2]O[2]) at room temperature for 20 min; (3) Microwave heating with ethylenediaminetetraacetic acid (EDTA) antigen retrieval solution (pH 9.0, 50 ×) for antigen retrieval; (4) Addition of goat serum blocking solution at room temperature for 15 min to block nonspecific binding sites; (5) Air-drying followed by overnight incubation with primary antibodies at room temperature; (6) After overnight air-drying, incubation with secondary antibodies at 37 °C for 30 min; (7) Addition of 3,3’-diaminobenzidine (DAB) solution for color development, observation under a microscope for 3–5 min, and termination of staining when appropriate; (8) Counterstaining with hematoxylin for 1–2 min, terminating when the cell nucleus turned blue; (9) Mounting with neutral gum. All Antibodies used in this study are listed in the Supplementary Table [72]S3. IHC semi-quantitative analysis was performed to assess the differential expression of key enzymes of lysine metabolism in the tumor and adjacent normal tissues. The IHC score was based on the staining intensity and staining extent of the samples. The cytoplasmic staining intensity score was defined as follows: 0 (negative, blue), 1 (positive, light yellow), 2 (positive, tan), 3 (positive, brown). The staining extent score was defined as 0 (0%), 1 (1–25%), 2 (26–50%), 3 (51–75%), and 4 (76–100%). Optical microscopic analysis was conducted at 400 × magnification, and the staining area of positive regions in five randomly selected fields of the same size was statistically analyzed. Multiplex immunohistochemistry (mIHC) The Absin four-color multiplex immunohistochemistry (mIHC) staining kit (abs50012, Lot#J0710l01) was utilized for mIHC experiments. Initially, deparaffinization was carried out using xylene, followed by hydration. Subsequently, microwave antigen retrieval was performed using ethylenediaminetetraacetic acid (EDTA, pH = 9.0) buffer. After natural cooling to room temperature, the specimens were blocked, and the primary antibody CD8-alpha (Abcam, ab17147, 1:100) was added and incubated at room temperature overnight. The following day, the horse radish peroxidase (HRP)-conjugated secondary antibody was added and incubated at 37 °C for 20 min. Tyramide signal amplification (TSA) dye (dilution: 1:100) was then applied and incubated at 37 °C for 20 min. Subsequently, microwave heat treatment in PBS buffer was conducted to remove non-covalently bound antibodies after blocking. Next, the second primary antibody, TIM-3(Cell Signaling, #45,208, 1:500) or LAG3(Cell Signaling, #15,372, 1:300) was added and incubated at room temperature overnight, followed by the same steps as described above. After microwave treatment in PBS buffer, 4’,6-diamidino-2-phenylindole (DAPI, dilution: 1:100 in distilled water) was added for room temperature counterstaining for 8 min. Finally, an anti-fade mounting medium and coverslips were added for slide fixation. The entire clinical tissue specimen images were scanned at 20 × magnification using the ZEISS Axioscan 7 high-performance slide scanning system. Public data sources To validate the discoveries from our in-house data, additional independent data on HCC were retrieved from public databases, including The Cancer Genome Atlas (TCGA, [73]https://portal.gdc.cancer.gov/) and the International Cancer Genome Consortium (ICGC, [74]https://dcc.icgc.org/releases/current/Projects/LIRI-JP), and the proteomic data from the Clinical Proteomic Tumor Analysis Consortium (CPTAC, [75]https://pdc.cancer.gov/pdc/) [[76]21]. Besides, we downloaded the clinical survival information for TCGA-LIHC patients from the UCSC Xena ([77]https://xenabrowser.net/datapages/) portal for the validation of survival analysis, which included Disease Specific Survival (DSS), Progression Free Interval (PFI), and Disease Free Interval (DFI). The Melanoma dataset was downloaded from the Mendeley data ([78]https://data.mendeley.com/datasets/z87v2pvc8t/1). We obtained gene expression, survival, and immunotherapy information from the IMvigor210, [79]GSE78220, and [80]GSE109211 cohorts to assess the immunotherapy response. These immunotherapy cohorts included data from 348 patients with metastatic urothelial carcinoma, 28 patients with metastatic melanoma, and 68 patients with HCCrespectively [[81]22, [82]23]. The IMvigor210 and [83]GSE78220 datasets represent cohorts of patients receiving anti-programmed death-ligand 1 (PD-L1) and anti-programmed death-1 (PD-1) therapies, respectively. Additionally, the [84]GSE104580 cohort was utilized to assess Transcatheter Arterial Chemoembolization (TACE) treatment in HCC. The IMvigor210 cohort was sourced from the R package IMvigor210CoreBiologies (1.0.0). The GSE cohorts were downloaded from the Gene Expression Omnibus database (GEO, [85]https://www.ncbi.nlm.nih.gov/geo/). Differential expression and functional enrichment analyses We performed transcriptome differential gene expression analysis using the R package DESeq2 (v1.44.0). The threshold for selecting differentially expressed genes (DEGs) was set at |log[2]FoldChange|≥ 1, with a requirement of adjusted p-value (padj) ≤ 0.05. The student’s t-test was employed for differential protein expression analysis, with a filtering threshold of |log[2]FoldChange|≥ 1 and p-value ≤ 0.05. Subsequently, we utilized the R package ClusterProfiler (v4.12.6) to conduct Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) functional enrichment analyses for both differentially expressed genes and proteins. GO terms and KEGG pathways with a p-value less than 0.05 were considered significantly enriched, demonstrating statistical significance. For single-cell TCR data, the FindMarkers function was employed to identify DEGs between tumor-enriched CD8^+ Tex clonotypes (380, 58, 162, 44, 10, 131) and normal tissue-enriched CD8^+ Tex clonotypes (6, 0, 3, 61); the filtering thresholds were set at adjusted p-value < 0.001 and |avg_log[2]FC|> 1. The DEGs were ranked by log[2] fold change (log[2]FC) and subsequently subjected to Gene Set Enrichment Analysis (GSEA), including GO-BP and KEGG pathway analyses. GO terms and KEGG pathways with an adjusted p-value < 0.05 were considered significantly enriched. Quantitative evaluation of lysine metabolism and identification of tumor subtypes To gain insight into the relationship between lysine metabolism and the prognosis of HCC patients, we quantitatively evaluated the lysine metabolism levels of HCC patients based on gene set variation analysis (GSVA). We defined the lysine metabolism gene set using the key genes for lysine metabolism, including AASS, ALDH7A1, AADAT, DHTKD1, GCDH, ECHS1, HADH, ACAAT, and ACAA2 (Table [86]S4), as identified from the REACTOME database(R-HSA-71064: REACTOME_LYSINE_CATABOLISM) and supported by relevant literature [[87]14, [88]24]. The GSVA was performed to calculate enrichment scores for each gene set within each sample by converting an expression matrix of individual genes into an expression matrix of specific gene sets. This method assessed the score of each sample to reflect the variation in gene set activity. Utilizing the GSVA algorithm, we assessed the lysine metabolism levels for each sample in both the TCGA-LIHC and the ICGC-LIRI-JP datasets. The normalized enrichment score (NES) of the lysine metabolism gene set for each tumor sample was defined as the lysine metabolism score (LM score). Based on the LM score, samples were ranked and patients were divided into different lysine metabolism subtypes (LM subtypes) by excluding the middle 20% of samples with moderate LM scores entered the medians. Patients with an LM score above the median were defined as the subtype of high lysine metabolism, while those with an LM score below the median were classified as the subtype of low lysine metabolism (Fig. [89]3A). Additionally, the R package METAFlux (v0.0.0.9000) was used to assess the lysine uptake capacity in bulk RNA-seq data [[90]25]. Fig. 3. [91]Fig. 3 [92]Open in a new tab The definitions of lysine metabolism score and subtypes of HCC patients. A Definition of lysine metabolism score (LM score) and the stratification of HCC patients based on LM score. B Density plot showing the distribution of lysine metabolism scores in tumor and normal samples from the TCGA-LIHC cohort. C-D Donut plots illustrating the outcomes of Chi-square tests for clinicopathologic factors in the high and low lysine metabolism subtypes in the TCGA-LIHC C and ICGC-LIRI-JP D cohorts. E–F Kaplan–Meier survival curves showing lysine metabolism as a predictor of overall patient survival in TCGA-LIHC E and ICGC-LIRI-JP F cohorts. G Comparison of epithelial-mesenchymal transition (EMT) process scores across different lysine metabolism subtypes in tumor samples from the TCGA-LIHC cohort. H Comparison of pathway activities between high and low lysine metabolism subtypes across TCGA, ICGC, and internal single-cell datasets. I Multivariate Cox proportional hazard regression analyses of clinical prognostic factors in the TCGA-LIHC cohort. **, p-value < 0.01 Pathway activity prediction by PROGENy We utilized the R package Progeny (v1.26.0) to assess the activity of 14 signaling pathways across different lysine metabolism subtypes in tumor samples [[93]26]. For bulk RNA-seq data, the “top” parameter was set to 100, while for scRNA-seq data, the “top” parameter was adjusted to 500 to account for the higher dimensionality and complexity. All other parameters were maintained at their default values. Survival analysis To further delineate the potential significance of key genes in lysine metabolism, as well as the prognostic value of the LM score and Tumor Mutation Burden (TMB) in HCC patients, we conducted Kaplan–Meier (KM) analysis, Cox proportional hazards regression analysis, and visualization of the results using the R packages survival (v3.6–4) and survminer (v0.5.0). Gene expression and median values of the LM score were utilized as grouping cutoffs. The Log-rank test was employed as the hypothesis testing method, and a Log-rank P-value less than 0.05 was considered statistically significant. Immune infiltration analysis To characterize the composition of immune cells in the immune microenvironment of HCC across different LM subtypes. We employed the single-sample gene set enrichment analysis (ssGSEA) algorithm to assess the infiltration levels of 28 immune cell types in the TCGA-LIHC cohort (n = 300) [[94]27]. Additionally, several different algorithms for immune infiltration analysis, such as CIBERSORT and EPIC from the R package IOBR, were used to further validate the effect of lysine metabolism on immune cell infiltration in HCC [[95]28]. Furthermore, we investigated the impact of lysine metabolism on CD8^+ T cell exhaustion by analyzing the expression of transcription factors, cytokines, chemokines, their receptors, and effector molecules associated with exhausted CD8^+ T cells across different LM subtypes [[96]29]. Finally, we employed the R package TCellSI (v0.1.0) to evaluate the scores of eight distinct T cell states across different groups at both the bulk transcriptomics and single-cell omics [[97]30]. Assessment of immunotherapy responses across LM subtypes Recent studies have indicated the existence of two mechanisms for tumor immune escape [[98]31]. One mechanism is characterized by high infiltration of cytotoxic T cells but with suppressed functionality. The other mechanism involves immune inhibitory pathways preventing the infiltration of T cells. The Tumor Immune Dysfunction and Exclusion (TIDE, [99]http://tide.dfci.harvard.edu/) online analysis platform predicts the response to immunotherapy by comprehensively assessing these two mechanisms of tumor immune escape [[100]32]. We performed TIDE analyses using the TCGA-LIHC and the ICGC-LIRI-JP datasets to assess the tumor immune escape for HCC patients across different LM subtypes. The mean-standardized gene expressions (transcripts per million, TPMs) were uploaded to the TIDE website for analysis. Furthermore, the IMvigor210 cohort (n = 348) and [101]GSE78220 cohort (n = 28) were used to evaluate the response to immunotherapy across different LM subtypes. The [102]GSE104580 cohort (n = 147) was used to assess the impact of lysine metabolism on TACE treatment. IMvigor210 cohort patients were categorized into stable disease (SD), progressive disease (PD), complete response (CR), and partial response (PR). Similarly, we excluded the top and bottom 10% of samples in the IMvigor210 cohort near the median values of the LM score. Patients in the [103]GSE78220 and [104]GSE104580 cohorts were categorized as responders (R) and non-responders (NR). Additionally, we incorporated tumor mutation burden (TMB) to determine whether the LM score can serve as an independent prognostic factor for HCC. The optimal cutoff point for TMB survival data was calculated using the R package survminer (v0.5.0). Statistical analysis Statistical analyses were carried out using R v4.4.0. IHC and mIHC quantitative analyses were conducted using GraphPad Prism 9.5.1 software. For the comparison of continuous variables between two groups with non-normal distribution in statistical analyses, the Wilcoxon test was employed, while the independent Student’s t-test was utilized for normally distributed variables. The Kaplan–Meier method and Log-rank test were applied for estimating and comparing the Disease-Specific Survival (DSS), Disease-Free Interval (DFI), Progression-Free Interval (PFI), and Overall Survival (OS) across different groups. Pearson correlation was employed for correlation analysis and statistical differences calculation. Cox proportional hazard regression analysis was conducted for univariate and multivariate analyses. The Chi-square test was used to compare clinical and pathological characteristics among different LM subtypes. Receiver Operating Characteristic (ROC) curves were generated to assess the accuracy of lysine metabolism scores in predicting TACE outcomes in HCC, as well as immunotherapy outcomes in other datasets A significance level of p-value < 0.05 was considered for all statistical determinations. Results Lysine metabolism is significantly downregulated in HCC Lysine metabolism in humans primarily occurs via mitochondrial ε-deamination, leading to the production of acetyl coenzyme A (Acetyl CoA) as the final metabolite [[105]14] (Fig. [106]1A). To explore lysine metabolic characteristics in hepatocellular carcinoma (HCC), we performed transcriptomic, proteomic, and single-cell omics profiling and analyses on tumor and adjacent normal tissues (NAT) from HCC patients. We focused on nine genes encoding the key lysine metabolic enzymes, including AASS, ALDH7A1, AADAT, DHTKD1, GCDH, ECHS1, HADH, ACAAT, and ACAA2 (Fig. [107]1A). Transcriptomic analysis identified 1432 upregulated and 1732 downregulated genes, with key lysine metabolic genes such as AASS, GCDH, ECHS1, AADAT, and ACAA2 showing significant downregulation in HCC (Fig. [108]1B, Supplementary Fig. [109]S1A). Corresponding proteomic analysis confirmed this trend, revealing that key lysine metabolic enzymes such as AASS, GCDH, ECHS1, HADH, and ACAA2 were also downregulated at the protein level (Fig. [110]1C, Supplementary Fig. [111]S1B). Fig. 1. [112]Fig. 1 [113]Open in a new tab The lysine metabolism is downregulated in HCC tumor samples. A The metabolic pathway diagram highlighting key genes coding lysine metabolic enzymes and mitochondrial lysine transporter. B The volcano plot showing differential gene expression from in-house RNA-seq data, with several lysine metabolic genes highlighted (|log[2]FC|> 1, adjusted p-value < 0.05). C The differential protein expression for key lysine metabolic enzymes, recapitulated by in-house proteomic data. D The UMAP plot showing single-cell populations in tumor and adjacent normal tissues of HCC. E The heatmap of key lysine metabolic gene expressions across different cell types in tumor and normal tissues. F The immunohistochemistry (IHC) staining results for AASS, GCDH, and AADAT, confirming their downregulation in HCC tissues (n = 30; × 40 magnification; scale bar = 50 μm). G The semi-quantitative analysis of IHC staining in Fig. 1F, shown as mean ± SD; paired two-tailed Student’s t-test (n = 30). H The KEGG pathways enrichment analysis of differentially expressed proteins (the numbers represent the count of genes significantly enriched in each term, p-value < 0.05). I METAFlux analysis of lysine uptake levels in tumor and normal tissues from the TCGA-LIHC dataset (negative values indicate uptake and their absolute values represent uptake magnitude). KEGG, Kyoto Encyclopedia of Genes and Genomes; ****, p-value < 0.0001; ***, p-value < 0.001; **, p-value < 0.01; *, p-value < 0.05 To further examine lysine metabolism at single-cell resolution, we performed scRNA-seq on tumor and NAT tissues. After quality control and batch effect removal, 61,396 single cells were annotated into 10 major cell types based on established markers [[114]33, [115]34] (Figs. [116]1D and [117]S1C, D). We found that lysine metabolic genes were primarily downregulated in tumor cells rather than in immune or stromal cells (Fig. [118]1E). Further, lysine metabolism scores derived from all lysine metabolic genes demonstrated significantly lower lysine metabolism in tumor tissues against NATs (Supplementary Fig. [119]S2A). Immunohistochemistry (IHC) results validated the notable downregulation of the key lysine metabolic enzymes in HCC, such as AASS, GCDH, and AADAT (Figs. [120]1F–G and [121]S2B). Notably, while lysine transporter SLC25A29 was not detected in proteomic data, transcriptomic and IHC data showed no significant difference in its expression between tumor and NATs (Supplementary Figs. [122]S1A and [123]S2B). These findings demonstrate that lysine metabolism is significantly downregulated in HCC, predominantly within tumor cells. To investigate the physiological implications of this downregulation, we conducted enrichment analyses on differentially expressed genes and proteins based on the Gene Ontology-Biological Process (GO-BP) and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathways. GO-BP analysis revealed that upregulated genes were associated with rapid tumor growth and proliferation [[124]35], while downregulated genes were primarily involved in amino acid metabolism and immune response activation (Supplementary Fig. [125]S2C). Similarly, KEGG analysis identified upregulated pathways related to cell proliferation and mismatch repair, while downregulated pathways were linked to amino acid metabolism and immune response (Supplementary Fig. [126]S2D). These results indicate disrupted amino acid metabolism and suppressed immune response in HCC. Further protein enrichment analysis showed significant downregulation of the lysine degradation pathway (Fig. [127]1H, Supplementary Fig. [128]S2E). Additionally, in the TCGA-LIHC dataset, tumor tissues exhibited lower lysine uptake scores compared to normal tissues, suggesting reduced lysine metabolism capacity (Fig. [129]1I). Overall, our multi-omics analyses revealed a significant downregulation of lysine metabolism in HCC, suggesting its critical role in hepatocarcinogenesis and progression. Downregulated lysine metabolism promotes HCC progression To further validate the lysine metabolic reprogramming observed in our in-house data and investigate its potential prognostic implications, we analyzed two public external HCC datasets with larger sample sizes, including the TCGA-LIHC (Tumor = 374, Normal = 50) and ICGC-LIRI-JP (Tumor = 240, Normal = 242). As expected, we found that except for ALDH7A1 and lysine transporter SLC3A2, all lysine metabolic genes were significantly downregulated in tumor tissues (Fig. [130]2A, B). Kaplan–Meier analysis of the TCGA-LIHC dataset indicated that lower expressions of lysine metabolic genes, including AASS (Log-rank P = 0.0284), GCDH (Log-rank P = 0.0203), and ACAT1 (Log-rank P = 0.0136), were significantly associated with poorer survival in HCC patients (Fig. [131]2C–E). These genes exhibited positive correlations with patient prognosis, while no significant survival differences were found for the other lysine metabolic genes (Supplementary Fig. [132]S3A). Additionally, expression levels of these genes were closely correlated with clinical staging in both TCGA-LIHC and ICGC-LIRI-JP cohorts (Supplementary Figs. [133]S3B-C). Specifically, lower expression levels of these genes were associated with higher clinical stages and worse prognosis (Supplementary Figs. [134]S3B, C), with GCDH and AADAT showing the most significant correlation with clinical staging (Fig. [135]2F, G). Fig. 2. [136]Fig. 2 [137]Open in a new tab Lysine metabolic reprogramming promotes tumor progression of HCC. A, B Downregulation of key lysine metabolic genes, as recapitulated in the TCGA-LIHC A and ICGC-LIRI-JP B cohorts. C-E The Kaplan–Meier survival analyses for lysine metabolic genes AASS C, GCDH D, and ACAT1 E in the TCGA-LIHC cohort, showing a significant association with overall survival (Log-rank p < 0.05). F, G The negative correlations of the GCDH and AADAT expressions with the clinical staging in TCGA-LIHC F and ICGC-LIRI-JP G cohorts. H Protein expressions of key lysine metabolic enzymes in the CPTAC HCC dataset, showing significant downregulation in tumor tissues. I The volcano plot showing the differentially expressed proteins in the CPTAC HCC dataset. J GO-BP enrichment analysis on differentially expressed proteins in the CPTAC dataset. K KEGG enrichment analysis of differentially expressed proteins in the CPTAC dataset (the numbers represent the count of genes significantly enriched in each term, p-value < 0.05). GO Gene Ontology; BP Biological Process; KEGG Kyoto Encyclopedia of Genes and Genomes; ****, p-value < 0.0001; ***, p-value < 0.001; **, p-value < 0.01; *, p-value < 0.05; ns not significant To confirm these findings at the proteomic level, we analyzed paired tumor and adjacent normal tissue (NAT) samples from the Clinical Proteomic Tumor Analysis Consortium (CPTAC) database (n = 159). We observed a notable downregulation of nine lysine metabolic enzymes in tumor tissues (Fig. [138]2H). Differential protein expression analysis identified 54 upregulated proteins and 401 downregulated proteins (F[139]ig. [140]2I). Enrichment analysis of upregulated proteins revealed significant involvement in biological processes (Fig. [141]2J) and KEGG pathways (Fig. [142]2K) related to DNA replication, cell cycle, and recombinant repair, supporting the growth and proliferation of tumor cells. Downregulated proteins were primarily related to fatty acid degradation and tryptophan metabolism. Most importantly, the lysine degradation pathway was significantly downregulated (Fig. [143]2K), corroborating our earlier findings. Collectively, these analyses of multiple independent large datasets further confirmed the downregulation of the lysine metabolism in HCC, driving a deeper exploration of how lysine metabolism reprogramming could promote HCC progression. Lysine metabolic reprogramming is a predictor of HCC prognosis To explore the relationship between lysine metabolic reprogramming and HCC prognosis, we defined a lysine metabolism score (LM score) based on nine key lysine metabolic genes using the Gene Set Variation Analysis (GSVA) (see Methods, Fig. [144]3A). The LM score was applied to both the TCGA-LIHC and the ICGC-LIRI-JP cohorts. In the TCGA-LIHC cohort, the distribution of LM scores revealed convergent and significantly higher lysine metabolism in normal tissues, while a heterogeneous and divergent distribution in tumors (Figs. [145]3B and [146]S4B). The dispersed LM scores in tumors, along with varied survival outcomes (Supplementary Figs. [147]S5A, B), suggest that lysine metabolic reprogramming is linked to HCC prognosis. Therefore, we ranked the tumor samples based on their LM scores and classified the patients into two subtypes by excluding the middle 20% of samples with moderate LM scores entered the medians (Fig. [148]3A). The high lysine metabolism subtype includes the top 40% of samples (n = 150 for TCGA-LIHC and n = 96 for ICGC-LIRI-JP), and the low lysine metabolism subtype represents the bottom 40% of samples (n = 150 for TCGA-LIHC and n = 96 for ICGC-LIRI-JP). Principal Component Analysis (PCA) confirmed the distinct clustering of LM subtypes (Supplementary Fig. [149]S4A). Furthermore, we excavated the relationship of lysine metabolism with other clinical characteristics, such as survival status, and clinical staging. For both the TCGA and ICGC cohorts, the high lysine metabolism subtype had a higher proportion of patients alive, as well as those in earlier clinical stages (Stage I and II; T1 and T2) (Figs. [150]3C, D and [151]S5A-B). These results suggested that elevated lysine metabolism correlates with a favorable prognosis. Moreover, Kaplan–Meier survival analysis confirmed this association, showing significantly better overall survival for the high lysine metabolism subtype in both cohorts (TCGA: Log-rank P = 6e-04, Fig. [152]3E; ICGC: Log-rank P = 0.0115, Fig. [153]3F). Additionally, patients with high lysine metabolism exhibited better disease-specific survival (DSS), disease-free interval (DFI), and progression-free interval (PFI) (DSS: Log-rank P = 8e-04; DFI: Log-rank P = 0.0377; PFI: Log-rank P = 0.006; Supplementary Fig. [154]S4C–E). To investigate the potential mechanistic links between lysine metabolism and HCC progression, we examined the epithelial-mesenchymal transition (EMT), a process known to drive tumor invasion, metastasis, and stem cell-like properties [[155]36]. We compared the EMT activity of different lysine metabolism subtypes at the bulk and single-cell transcriptomic levels. Tumor samples with lower lysine metabolism levels in the TCGA-LIHC (Fig. [156]3G) and ICGC-LIRI-JP (Supplementary Fig. [157]S4F) cohorts exhibited higher EMT activity. This finding was further validated by scRNA-seq data, confirming a correlation between the lower lysine metabolism level and increased EMT activity in tumor cells (Supplementary Fig. [158]S4G). These results suggest that impaired lysine metabolism may promote tumor progression by facilitating the EMT process. Besides, we performed Progeny pathway analysis on RNA-seq datasets from TCGA, ICGC, and in-house scRNA-seq datasets. In the low lysine metabolism subtype, we observed the activation of hypoxia, MAPK (mitogen-activated protein kinase) cascade signaling, NF-κB (Nuclear factor-κB) pathways, and transforming growth factor-β (TGFβ) (Fig. [159]3H). Hypoxia, a hallmark of the tumor microenvironment, promotes immune escape by impairing cytotoxic T cell (CTL) function and enhancing regulatory T cell (Treg) recruitment, thereby reducing tumor immunogenicity [[160]37]. Furthermore, hypoxia induces chronic inflammation through the activation of MAPK and NF-κB pathways. Notably, MAPK signaling enhances the stability and transcriptional activity of HIF-1α (Hypoxia-inducible factor-1α), amplifying the hypoxic response [[161]38]. Previous studies have demonstrated that hypoxia can also reduce the activity of natural killer (NK) cells, thereby suppressing NK cell-dependent anti-tumor immunity [[162]39]. Additionally, the TGFβ signaling pathway is a major driver of EMT, angiogenesis, and immune suppression, all of which are closely associated with tumor metastasis and progression [[163]40, [164]41]. Collectively, these mechanisms suppress anti-tumor immunity and drive tumor progression. To assess the LM score as an independent prognostic factor, we conducted a multivariate Cox proportional hazard regression analysis in the TCGA-LIHC cohort. The LM score was found to be an independent prognostic factor of survival (multivariate Cox: HR = 0.6, 95% CI = 0.43–0.85; p-value = 4.06e−03) (F[165]ig. [166]3I). These results emphasize that reduced lysine metabolism, coupled with the activation of pro-tumorigenic pathways like EMT, hypoxia, and TGFβ, drives HCC progression. The LM score holds the potential as a predictive biomarker for HCC prognosis. Lysine metabolic reprogramming reshapes the HCC tumor immune microenvironment The tumor immune microenvironment (TIME) plays a crucial role in tumor development and treatment response, with immunosuppressive signals contributing to immune evasion [[167]42, [168]43]. Previous studies have shown that metabolic abnormalities and immunosuppressive TIME are hallmarks of malignant tumors, with metabolic reprogramming closely linked to immune suppression. Specifically, alterations in glucose, amino acid, and lipid metabolism drive the immunosuppression in the TME and tumor immune evasion [[169]44]. To investigate the impact of lysine metabolic reprogramming on the immune microenvironment in HCC, we first assessed the infiltration of 28 immune cell types in the TCGA-LIHC cohort (Fig. [170]4A, top). We found that low lysine metabolism in tumor samples was associated with increased infiltration of myeloid-derived suppressor cells (MDSCs) and regulatory T cells (Tregs), highlighting the role of lysine metabolism in regulating immune cell infiltration and promoting an immunosuppressive microenvironment in HCC. Moreover, there was a notable negative correlation between the infiltration levels of immunosuppressive cells and the LM score (Fig. [171]4A, bottom), further supporting the role of lysine metabolism in regulating the immune microenvironment. Fig. 4. [172]Fig. 4 [173]Open in a new tab Lysine metabolic reprogramming reshapes the immune microenvironment of HCC. A Comparison of immune cell infiltrations between subtypes with high and low lysine metabolism (top panel) and the correlation between the LM scores and immune cell infiltrations (bottom panel). B The correlations of Tregs infiltration (left panel) and CD8 + T cell infiltration (right panel) with the expression of nine lysine metabolic genes based on various methods (The horizontal bar plots indicates the proportion of algorithms showing consistent results for each gene). C The expressions of signature genes, cytokines, effector molecules, and receptors related to exhausted CD8^+ T cells in high vs. low lysine metabolism subtypes. D Radar plot depicting T cell state scores in the TCGA-LIHC cohort. E Radar plot depicting T cell state scores in the internal scRNA-seq dataset. F The multiplex immunohistochemistry (mIHC) assays revealing the exhaustion of CD8^+ T cells in HCC tumors and adjacent normal tissues. G The density of LAG3^+CD8^+ and TIM3^+CD8^+ T cells (cells/mm.^2) in tumor vs. normal tissues (mean ± SD; paired two-tailed Student’s t-test for six samples). ****, p-value < 0.0001; ***, p-value < 0.001; **, p-value < 0.01; *, p-value < 0.05 Interestingly, despite the crucial role of CD8^+ T cells in anti-cancer immunity [[174]45], lysine metabolism disruption did not significantly affect their infiltration. To further investigate, we applied six different algorithms, such as CIBERSORT and EPIC, to explore immune cell infiltration in HCC. These analyses revealed the expression of lysine metabolic genes was negatively correlated with the infiltration of both Tregs and CD8^+ T cells, consistent with the findings from ssGSEA results. Notably, this correlation was consistently observed across multiple immune infiltration algorithms (Fig. [175]4B), suggesting strong methodological consensus regarding the overall trend. Although some variation was observed in statistical significance, such differences are likely attributable to discrepancies in marker gene set coverage and algorithmic modeling strategies. Based on these results, we speculated that the reduced activity of lysine metabolism in HCC may promote the formation of an immunosuppressive microenvironment, consequently leading to the exhaustion of CD8^+ T cell function under prolonged exposure to the immunosuppressive microenvironment [[176]46]. To confirm the hypothesis, we carried out a comparative analysis of exhausted CD8^+ T cells in different LM subtypes. The heatmap of cytokines, chemokines, and their receptors, as well as effector molecules, revealed increased expression in exhausted CD8^+ T cells in the low lysine metabolism subtype (Fig. [177]4C). Notably, The expression of key transcription factors and inhibitory receptors, such as HAVCR2 (TIM3), PDCD1, and LAG3 was significantly upregulated in the low lysine metabolism subtype (Fig. [178]4C), consistent with the hallmark features of T cell exhaustion [[179]47]. Furthermore, T-cell state analysis revealed that tumor samples exhibited lower cytotoxicity and higher exhaustion-related scores compared to normal tissues, suggesting that lysine metabolic reprogramming may impair T cell function (Fig. [180]4D, E). Experimentally, we further substantiated our hypothesis by multiplex immunohistochemistry (mIHC, see Methods), which shows a significant increase in the infiltration of TIM3^+CD8^+ and LAG3^+CD8^+ T cells in tumor tissues compared to adjacent normal tissues (Fig. [181]4F–G). These findings are consistent with a recent study by Chang et al. [[182]48], which reported that HCC tumor cells upregulate lysine transporter SLC3A2 to outcompete T cells for lysine uptake, thereby impairing T cell function and promoting immune evasion. To comprehensively dissect the heterogeneity and functional states of T cells within the TIME of HCC, we further performed subcluster annotation on 24,929 T cells, identifying 9 distinct subtypes, including CD4^+ Treg cells and CD8^+ Tex cells associated with low lysine metabolism levels observed in bulk RNA-seq data. (Figs. [183]5A, [184]B and [185]S6A, B). Notably, these two cell types were significantly enriched in tumor samples (Fig. [186]5C). ScRNA-seq data identified a cluster of CD8^+Tex cells characterized by high expression of inhibitory genes such as HAVCR2 and LAG3 (Fig. [187]5B). To further determine whether these CD8^+Tex cells are driven by antigen-specific clonal expansion, we examined the relationship between T-cell receptor (TCR) repertoire and functional status by TCR clonotype analysis. We observed a significant enrichment of TCR diversity in tumor-infiltrating T cells, suggesting robust antigenic stimulation within the TIME, which promotes clonal expansion (Fig. [188]5D, Supplementary Fig. [189]S6A). Single-cell TCR analysis further revealed that CD4^+ Tregs exhibited high TCR diversity and clonal expansion (Fig. [190]5E, F), along with higher expression of inhibitory genes such as CTLA4, FOXP3, and LAYN (Supplementary Fig. [191]S6C). In addition, CD8^+ Tex cells had lower TCR diversity, with a certain proportion of these cells exhibiting high levels of clonal expansion (Fig. [192]5E, F), with high expression of exhausted genes such as LAG3, HAVCR2, and PDCD1 (Supplementary Fig. [193]S6D). Importantly, both cell types were predominantly enriched in tumor tissues. We further stratified the top 10 dominant CD8^+ Tex clonotypes into tumor tissue-enriched and normal tissue-enriched populations and performed GSEA on differentially expressed genes (Supplementary Figs. [194]S6E–G). GO-BP analysis revealed that tumor-enriched CD8^+ Tex clonotypes exhibited significant downregulation in immune-related pathways, such as regulation of leukocyte-mediated immunity and natural killer cell activation (Fig. [195]5G). As well as the KEGG analysis revealed that the pathway natural killer cell-mediated cytotoxicity was significantly downregulated in tumor-enriched CD8^+ Tex clonotypes (Fig. [196]5H). Previous studies have confirmed that NK cell function is significantly impaired in HCC, characterized by reduced cytotoxicity, diminished cytokine secretion, and weakened immune recognition. These dysfunctions may contribute to the sustained suppression of pathways such as “natural killer cell-mediated cytotoxicity,” thereby promoting immune evasion in HCC [[197]49–[198]51]. These findings suggest that although CD8^+ Tex dominant clonotypes are present in normal tissues, their anti-tumor functions may be suppressed. Fig. 5. [199]Fig. 5 [200]Open in a new tab Single-cell profiling of the T-cell population and TCR repertoire. A The UMAP plot depicting the annotation of T cell subtypes. B The stacked bar chart illustrating the abundance of T cell subtypes in tumor and normal tissues. C Bubble heatmap showing the expression pattern of selected marker genes of different T cell subtypes (dot size represents the fraction of expressing cells, color indicates gene expression levels). D The UMAP plot of TCR distribution, where TCR represents cells with detected TCRs, and NA represents cells without detected TCRs. E The UMAP plot showing the distribution of cells with different clonal expansion types: single-clone TCR lineages (< = 1), low-level clonal expansion (< = 2), and high-level clonal expansion (> 2), nan: signifies T cells with missing or undetected TCR information. F The UMAP plot of clonal size distribution, with color intensity representing the number of cells per clonotype (clone_id_size). G The GO-BP GSEA enrichment analysis of differentially expressed genes in the top 10 dominant clonotypes of CD8^+ exhausted T cells (CD8^+ Tex) between tumor-enriched and normal tissue-enriched populations. H The KEGG GSEA enrichment analysis of differentially expressed genes in the top 10 dominant clonotypes of CD8^+ exhausted T cells (CD8^+ Tex) between tumor-enriched and normal tissue-enriched populations. UMAP, Uniform Manifold Approximation and Projection, TCR, T Cell Receptor; GO, Gene Ontology; BP, Biological Process; GO:0002250, adaptive immune response; GO: 0002697, regulation of immune effector process; GO: 0002703, regulation of leukocyte mediated immunity; GO:0007215, glutamate receptor signaling pathway; GO:0032814, regulation of natural killer cell activation; GO:0071604, transforming growth factor beta production; GO:0001909, leukocyte mediated cytotoxicity; GO:0038093, Fc receptor signaling pathway; GO:0001912, positive regulation of leukocyte mediated cytotoxicity; GO:0042269, regulation of natural killer cell-mediated cytotoxicity; KEGG the Kyoto Encyclopedia of Genes and Genomes; GSEA, Gene Set Enrichment Analysis In summary, our findings demonstrate that lysine metabolic reprogramming in HCC promotes immune evasion by increasing the infiltration of immunosuppressive cells, such as Tregs and MDSCs, while impairing the functionality of CD8^+ T cells. This results in the exhaustion of CD8^+ T cells and a weakened anti-tumor immune response. Through single-cell RNA-seq and TCR analysis, we further uncovered the antigen-driven nature of CD8^+ Tex and CD4^+ Treg cells, revealing clonally expanded populations that correlate with distinct functional phenotypes. These changes diminish the anti-tumor immune capacity, thus facilitating HCC progression. Lysine metabolic reprogramming induces immunotherapy resistance in HCC Our previous analyses indicated that lysine metabolic reprogramming might reshape the immune landscape of HCC. To further investigate whether lysine metabolism could serve as an indicator for the immune therapeutic response in HCC, we assessed the Tumor Immune Dysfunction and Exclusion (TIDE) scores. A higher TIDE score indicates a greater immune escape potential. In the TCGA-LIHC dataset, we observed a significantly higher proportion of non-responders (NR) in the low lysine metabolism subtype compared to the high lysine metabolism subtype (Fig. [201]6A). Additionally, the TIDE scores were negatively correlated with the LM scores (Fig. [202]6B). Moreover, in the low lysine metabolism subtype, both TIDE and T cell exclusion scores were significantly higher, whereas T cell dysfunction scores were lower compared to the high lysine metabolism subtype (Fig. [203]6C). Similar results were obtained from the ICGC-LIRI-JP dataset (Supplementary Figs. [204]S7A-C), suggesting that HCC patients with low lysine metabolism have a stronger immune escape capability, which predisposes them to immunotherapy resistance and poorer survival outcomes. Fig. 6. [205]Fig. 6 [206]Open in a new tab Lysine metabolism is significantly associated with the immunotherapy response in HCC. A The TIDE assessment of immunotherapy response between the subtypes with low and high lysine metabolism in the TCGA-LIHC cohort. B The scatterplot showing a significant negative correlation between the LM score and the TIDE score in the TCGA-LIHC cohort. C Comparison of TIDE, T cell dysfunction, and T cell exclusion scores between the subtypes with low and high lysine metabolism in the TCGA-LIHC cohort. D The boxplot demonstrating significantly higher LM scores in the responder (R) group compared to the non-responder (NR) group in the HCC TACE cohort ([207]GSE104580). E The stacked percentage plot indicating a higher proportion of NR patients in the low lysine metabolism subtype. F The ROC curve assessing the predictive efficacy of the LM score for TACE treatment. G Survival analysis of the [208]GSE78220 cohort showing a favorable survival outcome associated with the high lysine metabolism subtype. H The boxplot illustrating higher LM scores in the R group compared to the NR group. I The stacked percentage plot showing an increased proportion of NR patients in the low lysine metabolism subtype. TIDE Tumor Immune Dysfunction and Exclusion; LM Lysine Metabolism; R Responders; NR Non-Responders; OS Overall Survival; HR Hazard Ratio; CI Confidence Interval; ROC Receiver Operating Characteristic; AUC Area Under Curve; ****, p-value < 0.0001; **, p-value < 0.01; *, p-value < 0.05 Given that Transarterial Chemoembolization (TACE) therapy, combined with systemic chemotherapy, is a first-line treatment for HCC [[209]52], we assessed whether lysine metabolism levels influence patient response to TACE. In the [210]GSE104580 cohort, we observed that the responder (R) group exhibited a higher LM score than the NR group (Fig. [211]6D), with a lower proportion of responder patients in the low lysine metabolism subtype (Fig. [212]6E). These results suggested that patients in the low lysine metabolism subtype benefit less from TACE treatment and that the LM score provides high accuracy in predicting TACE outcomes (AUC = 0.777, Fig. [213]6F). To assess the predictive value of lysine metabolism in immunotherapy response for HCC, we stratified the [214]GSE78220 and IMvigor210 immunotherapy cohorts into different LM subtypes based on the LM score. Kaplan–Meier survival analysis showed that patients with high lysine metabolism had better survival outcomes (Figs. [215]6G and [216]S7D). We also found that patients in the NR and stable disease/progressive disease (SD/PD) groups had lower LM scores in both cohorts (Figs. [217]6H and [218]S7E), and a higher proportion of NR and SD/PD patients was observed in the low lysine metabolism subtype (Figs. [219]6I and [220]S7F). Furthermore, Tumor Mutational Burden (TMB) is a known potential biomarker for predicting immunotherapy response [[221]53, [222]54]. Given the crucial significance of TMB in immunotherapy, we attempted to explore the impact of TMB and LM scores on the prognosis of HCC. In the TCGA-LIHC cohort, survival analysis revealed that patients with low TMB had better overall survival compared to those with high TMB (Log-rank P = 2.17e–05, Supplementary Fig. [223]S7G). Interestingly, the high lysine metabolism subtype exhibited higher TMB (Supplementary Fig. [224]S7H), which may be attributed to the pronounced heterogeneity and the prevalence of low TMB characteristics in HCC [[225]55]. Stratified survival analysis of the four subgroups (High-TMB -High-LM score; High-TMB -Low-LM score; Low-TMB -High-LM score; Low-TMB -Low-LM score) showed that the Low-TMB -High-LM subgroup had the best prognosis, while the High-TMB -Low-LM subgroup had the poorest prognosis (Supplementary Fig. [226]S7I). Notably, survival differences were observed within the high and low TMB subgroups, indicating that TMB status does not affect the predictive value of the LM score. This suggests that the LM score can independently predict survival outcomes in HCC and assess responses to immunotherapy, regardless of TMB. To further elucidate the association between lysine metabolism and immunotherapy response, we analyzed the correlation of key lysine metabolic genes and the LM score with the expression of classical and putative immune checkpoint inhibitor (ICI) markers (PDCD1, CTLA4, CD274, LAG3, HAVCR2) in the TCGA-LIHC cohort [[227]4, [228]56]. Our results revealed that both metabolic gene expression and LM scores were significantly negatively correlated with the majority of ICI markers (Fig. S[229]8A–B), suggesting a potential role of lysine metabolic activity in shaping an immunosuppressive TME. Notably, the LM score exhibited a relatively weak correlation with PD-L1 (CD274) expression, which may be attributed to the low expression of PD-L1 in HCC [[230]57]. Additionally, we analyzed the predictive performance of the LM score in three public immunotherapy cohorts: [231]GSE78220 (Melanomas, anti-PD-1, Pembrolizumab), [232]GSE109211 (Hepatocellular carcinoma, Sorafenib), and IMvigor210 (Urothelial carcinoma, anti-PD-L1, Tecentriq). In all three datasets, the LM score demonstrated higher predictive accuracy for treatment response compared to ICI markers, supporting its potential as an independent biomarker for immunotherapy efficacy (Fig. [233]S8C–E). Collectively, our analyses of immunotherapy response in the TCGA, ICGC, and other immunotherapy cohorts suggest that patients with low lysine metabolism may derive fewer benefits from immunotherapies compared to those with high lysine metabolism. The low lysine metabolism subtype exhibited a greater potential for immune escape and poorer overall survival. Combined with previous findings, we propose that lysine metabolic reprogramming in HCC contributes to the development of an immunosuppressive TIME, which in turn leads to immunotherapy resistance. Taken together, our findings highlight the potential of lysine metabolism as a predictive factor for personalized immunotherapy in HCC. Despite the limitations of applying immunotherapy cohorts from other cancer types, analysis of TCGA-BLCA and melanoma cohorts revealed that the expression patterns of lysine metabolic genes and immune infiltration closely resembled those in HCC (Fig. [234]S9-10). These findings suggest that the immunosuppressive features associated with low lysine metabolism may represent a broader phenomenon, reinforcing the robustness of our conclusions. Discussion Hepatocellular carcinoma (HCC) remains a major health challenge worldwide, characterized by its high invasiveness and high recurrence, and a tendency for late-stage diagnosis [[235]58]. Due to its insensitivity to traditional chemotherapy and radiation therapy, targeted therapies represented by Sorafenib have become the mainstay of treatment for advanced or recurrent HCC, but their efficacy remains limited [[236]59]. In recent years, immune checkpoint inhibitors (ICIs) have brought new hope for advanced HCC patients, however, low response rates continue to pose a significant challenge [[237]60]. Additionally, randomized clinical trials of systemic and local therapies related to combined immunotherapy for HCC have not demonstrated survival benefits [[238]61]. Therefore, understanding the mechanism of immunotherapy resistance and identifying new therapeutic strategies is paramount for improving patient outcomes. Tumor metabolic reprogramming is crucial in driving cancer progression and reshaping the tumor immune microenvironment (TIME). Targeting metabolic pathways has been explored as a promising strategy to enhance tumor immunogenicity and improve response to immunotherapy [[239]10, [240]62, [241]63]. Amino acid metabolism, including the reprogramming of lysine metabolism, is a key component of tumor cell growth and immune modulation within the TME [[242]64, [243]65]. Lysine, an essential amino acid, undergoes metabolic alterations that not only support cancer cell proliferation but also influence immune responses in the TME [[244]66]. Acetyl-CoA, the final product of lysine metabolism, serves as a critical signaling molecule involved in histone acetylation and gene expression regulation [[245]67]. Aberrant acetylation can drive tumor progression by enhancing the transcription of genes involved in cell proliferation and immune evasion. Additionally, acetyl-CoA plays a crucial role in regulating immune cell functions, such as T cell metabolism, which may influence immune responses within the TME [[246]68]. Moreover, previous studies have indicated that depriving HCC cell lines of lysine can inhibit tumor cell growth and migration while inducing cell cycle arrest and apoptosis [[247]69]. These findings highlight the importance of lysine metabolism in HCC progression and its potential impact on immunotherapy efficacy. In this study, we conducted comprehensive multi-omics analyses, including transcriptomic, proteomic, and single-cell profiling of HCC tissues, to investigate the role of lysine metabolic reprogramming in tumor progression and immunotherapy resistance. Our results revealed significant downregulation of lysine metabolism in HCC, along with key enzymes such as AASS, GCDH, ECHS1, and ACAA2 showing reduced expression at both the transcript and protein levels. These findings were further validated in multiple datasets and immunohistochemistry (IHC). Kaplan–Meier analysis indicated that lower expression of key lysine metabolic genes, including AASS, GCDH, and ACAT1, correlated with poorer survival, suggesting that dysregulated lysine metabolism contributes to worse clinical outcomes. AASS serves as an important regulator of lysine metabolism, and its normal function is essential for maintaining intracellular lysine metabolism balance. AASS is a bifunctional mitochondrial enzyme. It is the first step in lysine metabolism and plays a critical rate-limiting role. AASS mutations can cause the accumulation of lysine, which can lead to type I hyperlysinemia in severe cases [[248]70, [249]71]. Moreover, a recent study found that glioblastoma stem cells (GSCs) can reprogram lysine metabolism by upregulating the lysine transporter SLC7A2 and the key metabolic enzyme GCDH, while downregulating ECHS1. This dysregulation leads to the accumulation of lysine and its intermediate metabolite crotonyl-CoA, which in turn promotes tumor progression, fostering an immunosuppressive tumor microenvironment, [[250]24]. Additionally, targeting specific metabolic pathways has been widely studied in promoting tumor immunotherapy. To further clarify the relationship between lysine metabolic reprogramming and the prognosis of HCC patients, we constructed a lysine metabolism score (LM score) based on the key lysine metabolic genes, which allowed us to classify HCC patients into subtypes with high or low lysine metabolism. This stratification revealed that the low lysine metabolism subtype is characterized by the activation of pro-tumorigenic pathways like EMT, hypoxia, and TGFβ signaling pathway, and ultimately results in a worse prognosis, immune escape, and a more immunosuppressive TME. Our immune infiltration analysis indicated that downregulated lysine metabolism promoted the infiltration of immunosuppressive cells, such as myeloid-derived suppressor cells (MDSCs) and regulatory T cells (Tregs), which are known to contribute to immune evasion and tumor progression in HCC. MDSCs are a class of cell populations with immunosuppressive functions, which have been shown to proliferate and accumulate abnormally in HCC. They inhibit T cell proliferation and CD8^+ T cell activity and can also promote the proliferation of Tregs through direct cell-to-cell contact, which is closely associated with an increased risk of disease progression and poor prognosis in HCC [[251]72, [252]73]. For another, Tregs secrete immunosuppressive cytokines such as transforming growth factor-beta (TGF-β), interleukin-10 (IL-10), and interleukin-35 (IL-35), thereby promoting the formation of an immunosuppressive microenvironment and inducing their own proliferation. Meanwhile, they inhibit the anti-tumor effects of effector T cells and natural killer (NK) cells [[253]74]. Research shows that tumor-associated Tregs are critical for HCC immune evasion [[254]75]. Furthermore, prolonged exposure to the immunosuppressive microenvironment may lead to gradual weakening or even loss of CD8^+ T cell responsiveness, resulting in the development of exhausted CD8^+ T cells. This phenomenon ultimately diminishes the anti-tumor immune capacity in HCC, leading to resistance to immunotherapy and tumor progression. In addition, single-cell TCR analysis shows that compared to the normal tissue-enriched CD8^+ Tex clonotypes, the tumor-enriched CD8^+ Tex clonotypes downregulate processes associated with anti-tumor immunity, such as natural killer cell activation. Studies have shown that NK cell dysfunction in HCC is characterized by receptor–ligand imbalance (e.g., decreased activating receptors such as NKG2D, and increased inhibitory receptors such as NKG2A), as well as disrupted Fas/FasL-mediated cytotoxicity, which both contribute to immune evasion, thereby promoting the HCC progression [[255]51]. Besides, to gain insight into the predictive value of the LM score for immunotherapy response and the differences in immunotherapy response among different LM subtypes, we conducted comprehensive analyses from multiple perspectives. Firstly, TIDE analysis was performed separately in the TCGA-LIHC and the ICGC-LIRI-JP datasets. We observed that compared to patients in the high lysine metabolism subtype, those in the low lysine metabolism subtype exhibited higher TIDE scores and T cell exclusion scores, indicating a stronger anti-tumor immune escape ability and lower response rates to ICI treatment [[256]32]. Additionally, in the [257]GSE78220 and IMvigor210 cohorts, the LM score showed potential in predicting patient response to immunotherapy. Furthermore, we found that patients in the low lysine metabolism subtype benefited less from existing TACE treatments. Finally, based on the prediction of TMB and LM score on the prognosis of HCC patients, we found that the LM score could serve as a potential independent prognostic indicator for HCC patients’ survival, regardless of TMB. In conclusion, our study demonstrates that lysine metabolic reprogramming plays a central role in reshaping the tumor microenvironment and promoting immune evasion in HCC. By identifying lysine metabolism as a critical factor in immunotherapy resistance, we provide new insights into the mechanisms underlying immune escape and tumor progression in HCC. Although direct mechanistic experiments were not included in the current study, emerging evidence from recent literature supports our findings. For example, recent research has shown that HCC tumor cells upregulate the lysine transporter SLC3A2 to compete with T cells for extracellular lysine, thereby impairing T cell proliferation and cytotoxicity through suppression of the STAT3 signaling pathway, ultimately facilitating immune escape [[258]48]. In this study, we observed a marked downregulation of key enzymes involved in lysine metabolism, accompanied by an upregulation of the lysine transporter SLC3A2 in tumor tissues. This study provides direct mechanistic evidence for lysine competition between tumor cells and T cells as a means of immunosuppression. Additionally, the LM score serves as a valuable tool for stratifying HCC patients and predicting their response to immunotherapy, offering potential for personalized treatment strategies. These findings may pave the way for the development of targeted therapies aimed at restoring lysine metabolic balance to enhance immune responses and improve clinical outcomes in HCC patients. Conclusions In summary, we revealed the characteristics of lysine metabolic reprogramming in HCC by integrative profiling of multiple omics and histology from clinical samples. We found that the lysine metabolism was significantly downregulated in HCC and was predictive of prognosis and progression. We demonstrated that the downregulated lysine metabolic activity in HCC reshaped the TIME characterized by immunosuppression, thereby inducing HCC immunotherapy tolerance and promoting HCC progression. These results revealed the relationship between lysine metabolic reprogramming and TIME remodeling, holding the promise of developing potential new targets for HCC molecular classification and immunotherapy. Supplementary Information [259]Supplementary file 1.^ (2.8MB, pdf) Acknowledgements