Genetic rofile of DRGs in HCC
To explore the role of DRGs in hepatocellular carcinoma, we described their genetic map. Among the 364 patients with HCC in the TCGA cohort, 26(7.14%) occurred to be regulated by disulfidptosis gene-related mutations(Fig. 1a). The mutation frequencies of FLNB and TLN1 were higher, while SLC7A11 and other mutations did not occur in the samples. The CNV amplification frequency of 9 genes was higher than the deletion frequency, and the CNV deletion frequency of 6 genes was higher than the amplification frequency(Fig. 1b). The location of 15 DRGs on chromosomes was demonstrated in Fig. 1c. Compared to normal tissues, all DRGs except MYH10 and IQGAP1 were up-regulated in HCC(Fig. 1d). Our study also showed that the up-regulated expression of 13 DRGs suggested a poor prognosis of this HCC(Supplementary Fig. 1). There was no difference between MYH10 and IQGAP1 in survival difference. This suggests a role for DRGs in the prognosis of HCC patients.
Identification Of A Classification Pattern Of Hcc Based On The Phenotype Of Disulfidptosis
To explore the expression characteristics of DRGs in HCC, 602 study samples from the TCCA-LIHC and ICGC-LIRI-JP cohorts were clustered using a consensus clustering algorithm based on the expression data of 15DRGs. Two disulfidptosis clusters A(253 samples) and B(349 samples) were obtained by cluster analysis(Fig. 2a-c). PCA, tSNE and UAMP reduced-dimension analysis showed that the samples of An and B types were significantly clustered(Fig. 2d-f). Patients in group B had better survival in the Kaplan-Meier curve than in group A(Fig. 2g).The heatmap showed that DRGs were more abundantly expressed in subtype A than subtype B. The heatmap showed that DRGs were more abundantly expressed in subtype A than subtype B(Fig. 2h).
Tme Characteristics And Immune Infiltration Between Two Disulfidptosis Subtypes
To clarify the characteristics of the two disulfidptosis subtypes in TME, immune scores and gene expression between the two clusters were calculated in this study. In contrast to previous perceptions, cluster A was higher than cluster B in immune score, stromal score and ESTIMATE score(Fig. 3a). In addition, most immune cells were more enriched in group A than in group B(Fig. 3b-c). Among them, CD8 + Tcells, Macrophages, NK cells, Treg, APC co-inhibition, aDCs and other immunostimulatory factors were significantly expressed in patients with subtype A.This suggests that both co-stimulators and co-inhibitors may play their respective roles in cluster A. T cells in immune response are mainly activated by two signaling modes: binding of MHC to T cell receptor, co-stimulation and co-inhibition of molecular signals.Furthermore, we analyzed the expression differences of MHC-related genes, co-stimulatory/co-inhibitory factors and immune-related genes among genotypes.The expression of most HLA family genes was significantly up-regulated in cluster A(Fig. 3d). In recent years, the newly discovered expressions of MICA and MICB are also consistent. Co-stimulatory and co-repressor factors also showed expression predominance in cluster A as seen in Fig. 3e-f. Therefore, we analyze the combined effect of high immune infiltration and immunosuppression in the tumor microenvironment of HCC.
Biological Functional Analysis Of Different Disulfidptosis Subtypes
GSVA was applied to analyze the biological behavior of the two clusters. The samples of cluster A were mainly enriched in a variety of cancer-related pathways, immune system-related pathways and cytoskeleton protein-related pathways(Fig. 3g). Cluster B was enriched in biosynthesis-related pathways(Fig. 3g). To further explore the pathway relevance of the respective gene enrichment in clusters A and B, we used GSEA analysis. Highly expressed genes in cluster A were enriched in cytokine interactions and ECM receptor interactions(Fig. 3h). Low expression group genes were in concentrated in multiple amino acid metabolism and bile acid biosynthesis(Fig. 3h). Notably, the high enriched group in the B subtype was enriched in multiple amino acid metabolism, chlorophyll metabolism and bile acid biosynthesis(Fig. 3i). The low enriched group was mostly concentrated in cytokine interactions and ECM receptor interactions(Fig. 3i). Cytokines play a bidirectional role in tumors. IL-18 promotes angiogenesis, induces cancer cell proliferation and invasion, and prevents apoptosis by activating NF-κB. IL-6 similarly promotes chronic inflammatory carcinogenesis and drives intrinsic tumor progression. High doses of recombinant IL-2 were approved by the FDA in 1992 and 1998 for the treatment of metastatic renal cell carcinoma(RCC) and melanoma, respectively. The extracellular matrix(ECM) is associated with the progression of a variety of tumors including hepatocellular carcinoma, pancreatic ductal adenocarcinoma, and breast cancer. sclerosis of the ECM promotes cell proliferation, epithelial-mesenchymal transition, cancer cell metastasis, and the development of drug resistance. Sclerosis of the ECM has been reported to promote the release of tumor exosomes and activate the NORTH pathway to promote tumor invasion.The functional enrichment of these two subtypes may be responsible for the differences in patient prognosis.
Construction And Validation Of A Prognostic Model For Disulfidptosis
The classification of disulfidptosis has great potential in the clinical prognosis of hepatocellular carcinoma(HCC). In order to better understand the characteristics of disulfidptosis, we constructed a prognostic model. Initially, we identified 681 DEGs from the clusters and identified 331 prognostic-related genes through univariate Cox analysis(Supplementary Fig. 2 and Supplementary Table S2). Subsequently, Lasso-Cox regression analysis was performed on 331 prognostic genes(Fig. 4a-b), and four key genes were identified finally(Table 1). Based on these key genes, we constructed a risk score for disulfidptosis. Risk score=(-0.1806*APOC1) + (-0.2229*IL7R) + (0.0851*SPP1) + (0.2528MYBL2). We divided all HCC samples into high-risk and low-risk groups based on risk scores. The high disulfidptosis risk group had a significantly worse prognosis compared to those with a low disulfidptosis risk(Fig. 4c, P < 0.001). This result was confirmed by Kaplan-Meier survival curves for the test and train groups(Fig. 4d-e, P < 0.001). Risk curve analysis showed that increasing risk scores increased their risk of death(Fig. 4f). Risk survival analysis showed increased mortality in the high-risk group of patients, whose survival time was relatively short(Fig. 4g). In addition, we confirmed the high accuracy of the disulfide risk score in predicting prognosis after 1, 3 and 5 years. The area under the curve (AUC) values for the overall sample were 0.766, 0.736 and 0.699 for 1, 3 and 5 years(Fig. 4h). The AUC values in the internal train set were 0.808(1 year), 0.749(3 years), and 0.757(5 years)(Fig. 4i). The AUC values in the test set were 0.716, 0.704, 0.637 for 1, 3 and 5 years(Fig. 4j).
Finally, we used GSE14520 as an external independent dataset to validate the ability of the disulfidptosis risk score. We applied a risk score based on disulfidptosis to the out validation group and divided it into high and low risk groups. It has been proved that low-risk patients have a significant survival advantage(Fig. 4k). More importantly, the ROC curves showed a significant prognostic significance for the validation set cohort, with AUC values above 0.65 at 1, 3, and 5 years(Fig. 4l). Overall, our results demonstrate the good potential of the disulfidptosis risk score in evaluating prognosis.
Table 1
Multifactorial Cox regression analysis of prognosis-related DEGs.
| id | coef | HR | HR.95L | HR.95H | pvalue |
| APOC1 | -0.18059647 | 0.834772147 | 0.731772689 | 0.952269123 | 0.00719094 |
| IL7R | -0.222889579 | 0.800203204 | 0.654681576 | 0.978071159 | 0.029519535 |
| SPP1 | 0.085147213 | 1.088877352 | 1.012623953 | 1.170872843 | 0.021526195 |
| MYBL2 | 0.252818907 | 1.287650072 | 1.13364786 | 1.462572961 | 0.000100197 |
Independent Predictive Value Of Disulfidptosis Risk Prognostic Model
We constructed a disulfidptosis risk prognostic model that demonstrated differences in survival. To further investigate, we integrated clinical information from all samples and identified three common clinical features(age, sex, stage). Our analysis showed that stage and risk score were prognostic factors for HCC(Fig. 5a and 5b). The ROC curves for risk score and stage showed their high accuracy in predicting prognosis(Fig. 5c).We also found that the risk was significantly lower in cluster B than in cluster A(Fig. 5d, P < 0.001). The sankey diagram illustrated that the low-risk group primarily consisted of patients belonging to subtype B, while most of the surviving patients originated from this group. Conversely, patients who passed away had higher risk scores and were predominantly from subtype A, aligning with previous analysis(Fig. 5e).
To further validate the applicability of our disulfidptosis feature model to HCC patients, we compared it with previously reported models. In the TCGA-LIHC cohort, the Tang signature[18], Wang signature[19], Zhang signature[20], and Deng signature[21] all demonstrated significant differences in survival with P values less than 0.01(Supplementary Fig. 2a-d). The AUC values for the four HCC feature models were all greater than 0.60 at 3 and 5 years(Supplementary Fig. 2e-h). Comparing the ability of the disulfidptosis feature model and the four other models to predict survival events, we found that among the C-index values of the five feature models, the disulfidptosis feature model had the highest value of 0.708, which was closer to the perfect model, followed by Tang signature(0.663), Wang signature(0.657), Deng signature(0.659), and Zhang signature(0.604)(Fig. 5f). Additionally, the RMS value of the disulfidptosis feature model was also the highest among the five models(Fig. 5g).
Disulfidptosis Risk Score Had Considerable Potential In The Prediction Of Tumor Treatment Effect
In this study, we analyzed the proportion of immune infiltration between different HCC risk groups(Supplementary Fig. 3a) and found that plasma cells were more abundantly infiltrated in the high-risk group(Fig. 6a, P = 0.006). The correlation between immune cells is shown in Supplementary Fig. 3b. Activated dendritic cells(R = 0.33, P = 0.018) were positively correlated with risk score(Supplementary Fig. 3c). CD8T cells, neutrophils, macrophages, activated CD4 memory cells and activated dendritic cells showed significant correlations with risk genes and scores(Supplementary Fig. 3d). Immune checkpoints are important targets for immunotherapy.The expression of CTLA4, HAVCR2, ATIC and OLA1 remained consistent with the risk score(Fig. 6b). Immunotherapy for HCC has always been a focus of attention. We also analyzed the relationship between disulfidptosis risk groups and drug treatment. The suppressive immune checkpoint genes CTLA4, HAVCR2, ATIC and OLA1 were higher in the high risk group(Fig. 6c, P < 0.01). The IPS was higher in the low risk group(Fig. 6d, P < 0.01). Low risk patients have stronger immunogenicity. In the IMvigor210 immunotherapy cohort, risk values were higher in the CR/PR(Complete Response, Partial Response) group than in the SD/PD(Progressive Disease, Stable Disease)(Fig. 6e). When comparing the proportions of patients with different immune therapy outcomes, We observed a higher proportion of CR/PR patients in the high-risk group compared to the low risk group(Fig. 6f). The lower proportion of low risk patients with high immunogenicity may be due to certain immune suppression.
Furthermore, In addition, the TIDE score assessed the clinical response to immune checkpoint inhibitor therapy. A high TIDE score indicates a low response to ICB and unfavorable tumor outcomes. TIDE score, dysfunction score were lower in the high-risk group(Fig. 6g, P < 0.001). The lower composite TIDE score obtained in the high risk group may be due to the higher proportion of immunotherapy remissions.
Analysis Of Tumor Mutations And Drug Sensitivity
Due to the increased antigenicity associated with high mutation burden, high-risk HCC tends to have a longer survival time. Tumor mutations were relatively high in high-risk groups of TCGA-LIHC(Fig. 7a, P = 0.05). Although the prognosis of the high mutation group was worse than the low mutation group(Fig. 7b), its immunotherapy effect prevailed in previous experience[22]. Subsequently, we analyzed the mutation spectrum between the high and low risk groups. The proportion of mutations observed in the 171 high risk patients was as high as 88.89%(Fig. 7c), and the mutation rate in the 190 low risk patients was 82.63%(Fig. 7d), with TP53, CTNNB1, TTN, MUC16, and PCLO genes having a high mutation frequency in both risk groups.
To identify sensitive drugs for high and low risk patients, we analyzed the sensitivity of 197 chemotherapy drugs. The sensitivity results showed that Afuresertib, KRAS(G12C), Doramapimod, Mitoxantrone, and Oxaliplatin had higher sensitivity in the high risk group(Fig. 7e-i, P < 0.001). These drugs may improve the prognosis of high risk HCC patients with disulfidptosis. Cediranib, Dasatinib, Docetaxel, Paclitaxel, and Tozasertib had lower sensitivity in the high-risk group(Fig. 7j-n, P < 0.001), while the low-risk group may benefit from treatment with these drugs.
Construct A Nomogram To Predict The Survival Of Hcc
To better predict the prognosis of HCC, we constructed a nomogram for the disulfidptosis risk score. According to the constructed nomogram, each patient can obtain a 1, 3 and 5 year percentage survival coefficient according to the sum of the corresponding scores of clinical information(Fig. 8a). The 1, 3 and 5 year survival rates of the patients with a score of 236 were 0.756, 0.496 and 0.248, respectively. The cumulative hazard was higher in the high risk group(Fig. 8b). The calibration curve shows that the prediction result of the nomogram was close to the actual probability(Fig. 8c). DCA analysis shows that the nomogram and risk values have better effectiveness(Fig. 8d). The AUC values for the nomogram at 1, 3, and 5 years are 0.762, 0.743, and 0.702, respectively(Fig. 8e). The AUC values of the internal train set are all greater than 0.75(0.804 for 1 year, 0.760 for 3 years, and 0.763 for 5 years, Supplementary Fig. 4a). The internal test set confirms the high accuracy of the overall survival for 1, 3, and 5 years of the nomogram(Supplementary Fig. 4b). Additionally, in the external validation set GSE14520, the AUC values for the nomogram predicting 1, 3, and 5 years are all greater than 0.6(Supplementary Fig. 4c). The calibration curve shows that the internal train set, test set, and external validation set are all very close to the actual accuracy values (Supplementary Fig. 4d-f). All of these indicate that the disulfidptosis-related nomogram has high value in predicting the survival time of HCC patients.
Spp1 And Mybl2 Can Predict The Prognosis Of Hcc
We also analyzed the ability of four genes to predict the prognosis of HCC. The AUC of ROC curve was APOC1(0.425), SPP1(0.651), IL7R(0.487) and MYBL2(0.636), respectively(Fig. 8f). It can be seen that SPP1 and MYBL2 play an important role in predicting the prognosis of HCC. Based on the four biomarkers in the disulfidptosis model, we explored their ability in HCC. First of all, we collected 10 pairs of HCC and corresponding adjacent normal tissues for local tissue verification, extracted and collected RNA from the tissues for reverse transcription, and then used cDNA for qPCR experiment. The results of PCR showed that the mRNA content of APOC1, SPP1 and MYBL2 in HCC tissues was higher than that in paracancerous tissues(Fig. 8g-i). The mRNA level of IL7R is relatively low in HCC organizations(Fig. 8j). The protein immunohistochemical staining results of APOC1, SPP1 and MYBL2 were obtained from the HPA. It can be seen from Fig. 8k that APOC1, SPP1 and MYBL2 are moderately or highly stained in HCC tissues and weakly stained in normal liver tissues. This is consistent with the level of mRNA.