Single-Cell Sequencing Analysis and Multiple Machine Learning Methods identified CDT1 as an oncogene in retinoblastoma

Single-Cell Sequencing Analysis and Multiple Machine Learning Methods identified CDT1 as an oncogene in retinoblastoma | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Research Article Single-Cell Sequencing Analysis and Multiple Machine Learning Methods identified CDT1 as an oncogene in retinoblastoma Yang Yang, Yuezhi Zhang, Weiwei Xiong, Xiaolong Yin This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-4883827/v1 This work is licensed under a CC BY 4.0 License Status: Posted Version 1 posted You are reading this latest preprint version Abstract BACKGROUND Retinoblastoma (RB) is a heterogeneous primary intraocular malignant tumor. OBJECTIVE This work attempted to reveal the significant gene-related to disulfidptosis in RB. METHODS The scRNA-seq data from RB samples and normal samples obtained from GSE159977 were analyzed to distinguish cone cells from malignant cone cells using inferCNV. Subsequently, AUCcell was used to assess the disulfidptosis levels in cone cells. Disulfidptosis-related genes in RB were analyzed through weighted gene co-expression network analysis and machine learning methods. Cell proliferation, migration, and invasion were examined through colony formation and Transwell experiments. RESULTS We obtained 7 annotated cell clusters. Among them, there was a significant increase in the proportion of cone cells in RB samples. Malignant cone cells exhibited a higher disulfidptosis score. Cell trajectory analysis indicated that the disulfidptosis process intensified as cone cells transition from normal cells to malignant cells. Through machine learning, disulfidptosis-related genes were screened, ultimately identifying CDT1 as a key gene. CDT1 was upregulated in WERI-Rb1 and Y79 cells. Silencing CDT1 significantly suppressed the proliferation, migration, and invasion of RB cells. CONCLUSIONS This work revealed that tumors contained cone cell states with distinct transcriptional programs, and provided a crucial RB-related gene CDT1 for tumor progression. Thus, this article is of great significance in formulating a molecular targeted therapy scheme to prevent the development of RB. Retinoblastoma disulfidptosis single-cell RNA sequencing CDT1 oncogene Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Figure 6 Figure 7 Figure 8 Introduction Retinoblastoma (RB) is a malignant tumor derived from photoreceptor precursor cells. RB is most commonly seen in children under 5 years of age, and is the most common primary intraocular malignant tumor in infants, accounting for 2–4% of children's malignant tumors ( 1 ). The clinical manifestations of retinoblastoma are complex, including conjunctival congestion, corneal edema, iris neovascularization, vitreous opacity, elevated intraocular pressure and strabismus ( 2 , 3 ). At the same time, RB also seriously affects the quality of life of infants. RB is prone to intracranial and distant metastasis, thereby often endangering the lives of children ( 4 ). In oncology, tumor staging is extremely important for the selection of treatment methods and evaluation of efficacy, as is the case with RB ( 5 ). Moreover, many RB patients have a high mortality rate due to failure of early diagnosis, especially in developing countries ( 6 ). Thus, the characteristics and identification of genes closely related to the progression of RB have certain reference significance for the research of RB progression and early diagnosis. Disulfidptosis, a new type of programmed cell death, is a rapid mode of death caused by disulfide stress caused by excessive accumulation of cystine in cells ( 7 ). It usually occurs under conditions of glucose starvation. In cells with high expression of SLC7A11, insufficient supply of NADPH results in limited reduction of cysteine to cysteine, leading to disulfide stress and ultimately cell death ( 8 ). NF⁃κB and c-JNK signaling pathway have also been confirmed to play important roles in the formation of disulfidptosis ( 9 ). In malignant tumors, intracellular disulfidptosis and oxidative stress play a role, and the formation of invasive pseudopodia may lead to more disulfidptosis in metastatic malignant tumor cells ( 7 ). Therefore, the study of the mechanism of disulfidptosis is an important new direction in the field of disease treatment ( 10 ). A previous study has reported that biomarkers related to disulfidptosis can be used for clinical diagnosis and prognosis of liver cancer ( 11 ). However, the role of disulfidptosis-related genes in RB has not been reported. The aim of this study was to explore the role of disulfidptosis in RB and its impact on tumor biology. Using single-cell analysis, we elucidated the complex relationship between the progression of malignant cone cells and disulfidptosis, as assessed through AUCell scores and RNA velocity. Additionally, machine learning was employed to identify the disulfidptosis-related gene, CDT1. In vitro assays revealed that CDT1 silencing inhibited the proliferation, migration, and invasion of RB cells. Therefore, this study has provided valuable new data resources for understanding the pathogenesis and progression of RB. Materials and methods Data collection Single-cell transcriptomic data were retrieved from Gene Expression Omnibus ( https://www.ncbi.nlm.nih.gov/geo/ ) database. RB samples diagnosed for 4 months or 2 years and Normal samples were obtained from GSE159977 dataset. Besides, bulk transcriptomic data were retrieved from GEO databases (GSE97508, GSE110811, GSE24673, GSE208143). GSE97508, GSE208143 and GSE24673 datasets were merged as train dataset. Batch effects from non-biological technical biases were corrected using the “ComBat” algorithm of “sva” package. GSE110811 dataset was used for validation. ScRNA-Seq Analysis The analysis of the single-cell dataset was conducted using Scanpy (version 1.9.1) within Python (version 3.8.13). To ensure data quality, rigorous quality control procedures were applied to exclude the cells with undesirable attributes while the retaining cells meeting predefined criteria. These criteria encompassed gene counts ranging from 0.02 to 0.98, and mitochondrial gene counts below 20%. Additionally, Scrublet (version 1.0) was employed to identify and eliminate doublet cells. Principal Component Analysis (PCA) was executed via Scanpy to reduce the dimensionality of the cell data. To mitigate batch effects, the "harmony_integrate" function in Scanpy was employed. Clustering and visualization were performed using Leiden and UMAP algorithms. Finally, cell clusters were annotated with known cell types based on cell type-specific markers. Disulfidptosis‑related genes (DRGs) scoring The "AUCell" R package, designed to assess the active status of gene sets in single-cell RNA data, was employed to compute activity scores for DRGs in each individual cell. Following this, cells were categorized into two distinct groups based on their DRG-AUC (Disulfide-Related Genes Area Under the Curve) values: those with high DRG-AUC and those with low DRG-AUC, with the median value being utilized as the threshold for classification. Chromosome copy number variation analysis The approach described in inferCNV and implemented as infercnvpy ( https://github.com/icbi-lab/infercnvpy ) was employed to detect potential changes in chromosomal copy numbers. Control cells were selected from healthy tissues, including cone cells and immune cells. The differentiation between normal cone cells and malignant cells was accomplished through assessment of iterative clustering impact and computation of the copy number variation score. Drug Response Prediction We utilized the scDrug Python package, built upon the CaDRReS-Sc model, to predict IC50 values and tumor cell killing ratios associated with potential drug responses. This framework incorporates pre-trained GDSC and PRISM models, which have undergone rigorous training using gene expression data to establish robust pharmacogenomic relationships, ensuring precise drug response predictions. Effective data visualization using the seaborn package allowed us to focus on IC50 values and tumor cell killing ratios across different subpopulations. Our drug selection criteria were based on minimizing IC50 values and maximizing tumor cell killing ratios. RNA velocity analysis Velocyto allows for the estimation of single-cell RNA velocity by distinguishing between unspliced and spliced reads in bam files. The process involves generating Loom files using velocyto.py and subsequently importing this data into scVelo for further analysis. The 'scv.pp.moments' function is utilized to calculate the first and second order velocity vector moments, while the 'scv.t.velocity' functions are employed to compute dynamic velocity. The RNA velocities are then embedded into UMAP using the velocity graph. Developmental trajectory inference Monocle2 ((version 2.14.0) was employed for trajectory analysis to uncover alterations in cell states. The "estimateSizeFactors" and "estimateDispersions" functions in Monocle2 were utilized to identify the "dispersion" genes, which were subsequently employed for cell ordering. Dimensionality reduction was conducted through the 'reduceDimension' function, with the 'DDRTree' reduction method being applied. Consensus cluster analysis The Disulfidptosis-related genes were uniformly clustered using the ConsensusClusterPlus R package. This clustering analysis utilized the "pam" algorithm with "euclidean" distance measurement. Subsequently, 1000 bootstraps were executed, with each bootstrap comprising 80% of the patients from the training dataset. The range of cluster numbers, denoted as "k," was set from 2 to 9. The optimal "k" was determined based on both the cumulative distribution function (CDF) and the area under the curve (AUC) analysis. Additionally, Principal Component Analysis (PCA) was conducted to assess the validity of the molecular subtype distributions. Weighted correlation network analysis We initiated the process by constructing a clustering tree for the samples to detect and eliminate any outliers. Subsequently, the gene expression data from the training dataset underwent analysis using Weighted Gene Co-expression Network Analysis (WGCNA). The adjacency matrix was transformed into the Topological Overlapping Matrix (TOM). A threshold of R² = 0.85 was established as the criterion for module construction. Dynamic Tree Cut was employed with parameters set to deepSplit = 2 and minModuleSize = 100 to generate these modules. Key modules were selected based on the correlation between their members and the significance of the included genes. Selection of characteristic genes Predictive models were created using common genes identified from both scRNA and WGCNA. These models, which included Random Forest (RF), Support Vector Machine (SVM), Extreme Gradient Boosting (XGBoost), and General Linear Model (GLM), were implemented within the R software. Model performance assessment included the generation of reverse cumulative distribution curves for residuals and Receiver Operating Characteristic (ROC) curves. Model selection was based on scores derived from reverse cumulative residuals and Area Under the Curve (AUC) values obtained from the ROC curves. Cell culture Human retinal pigment epithelial cells (ARPE-19) were obtained from ATCC (Manassas, VA, USA). RB cell lines WERI-Rb1 and Y79 were purchased from Procell Life Science & Technology Co., Ltd (Wuhan, China) and National Collection of Authenticated Cell Cultures (Shanghai, China), respectively. All cells were cultured in RPMI-1640 medium (Gibco, Grand Island, NY, USA) at 37°C and 5% CO 2 atmosphere. The medium was supplemented with 1% L-glutamine (Sigma-Aldrich, St. Louis, MO, USA), 1% penicillin/streptomycin (Sangon Biotech, Shanghai, China) and 10% fetal bovine serum (FBS; Gibco). Quantitative real-time PCR (qRT-PCR) RNA extraction was carried out applying TRIzol regent (Thermo Fisher Scientific, San Jose, CA, USA) to isolate RNA from cells. PrimeScript™ RT reagent Kit (Takara, Dalian, China) was applied to synthesize cDNA with RNA as template. Genomic DNA was eliminated with gDNA Eraser. PCR reacted was conducted to examine CDT1 expression applying TB Green® Premix Ex Taq™ (Takara). Primer sequences (5’-3’) were shown as follow: CDT1: forward-GACATGATGCGTAGGCGTTTT and reverse-GAGCTGGTAATCTGACCTCCT-3; GAPDH: forward-CCACCCATGGCAAATTCCATGGCA and reverse-TCTAGACGGCAGTCAGGTCCACC. The relative expression of CDT1 was analyzed by 2 −ΔΔCT method. Colony formation assay Cells at logarithmic phase was seeded into a 6-well plate with density of 1 × 103 cells/plate. After 2 weeks of incubation at 37°C and 5% CO 2 , cells were fixed with 4% paraformaldehyde, and then stained with 0.5% crystal violet. Finally, the number of colony was counted. Transwell migration and invasion assays Transwell assay was conducted to examine migration and invasion of cells applying the 8 µm migration Transwell chambers (Corning, NY, USA). For cell migration assay, cells were added to the upper chambers at a concentration of 5× 10 4 cells/200 µL of FBS-free medium, whereas the lower chambers contained 700 µL of medium and 10% FBS. For invasion assay, cells were seeded into the upper chambers coating with 25 µL Matrigel (Sigma-Aldrich, St. Louis, MO, USA). After 24 h of culture, the migrated and invaded cells were stained with 0.5% crystal violet reagent, which were then observed under an optical microscope. Statistical analysis Each assay was performed for 3 times. Data were analyzed by SPSS 22.0 statistical software (IBM, Armonk, NY, USA) and expressed as mean ± standard deviation. Two-tailed Student’s t test and one-way ANOVA were used to analyze the statistical difference. P < 0.05 was considered as a significant difference. Results Analysis of RB single‑cell sequencing data In this investigation, we utilized the scRNA-seq expression matrix to investigate the presence of disulfidptosis in RB patients. After rigorous quality control procedures, we retained a total of 10,199 high-quality cells for subsequent analysis. Following batch correction, we integrated the single-cell transcriptomic data and performed UMAP-based clustering to identify 18 different cell subclusters. Subsequently, based on established canonical marker gene, we annotated these clusters and classified them into cone cells, endothelial cells, epithelial cells, astrocytes, microglial cells, Müller glial cells, and neurons (Fig. 1 A-B). Moreover, we quantified the tissue enrichment of all these cell populations based on Ro/e analysis and found that only cone cells were enriched in RB tissue (Fig. 1 C). The DRG score of each cell line was evaluated using the R package "AUCell". As shown in Fig. 1 D, cone cells, epithelial cells, and neurons exhibited higher DRG scores. Based on the median of DRG scores, the cells were classified into high DRG-AUC and low DRG-AUC groups. We then examined the differences in DRG scores of different cell types between normal and RB tissue. Compared with normal tissue, higher DRG scores were observed in cone cells, endothelial cells, astrocytes, and microglial cells of RB tissues, while lower DRG scores were observed in epithelial cells, Muller cells and neurons of RB tissues (Fig. 1 E). Identification of potential drug candidates for RB based on scRNA-seq analyses To determine the potential drug in RB, we inferred copy number variations (CNVs) using the Python package inferCNVpy. By generating single-cell CNV scores, we differentiated between normal cone cells and malignant cone cells (Fig. 2 A). The critical threshold was determined as 0.0267 by calculating the average CNV score of cone cells, with values greater than the threshold considered as malignant cone cells (Fig. 2 B). From the enrichment analysis, normal cone cells were mainly concentrated on phototransduction, ECM-repair interaction, aldosterone synthesis and secretion. Malignant cone cells were mainly related to cell cycle, asthma, folate biosynthesis (Fig. 2 C). The Leiden algorithm was employed to cluster malignant cells, revealing the identification of four distinct clusters with varying distributions (Fig. 2 D). To identify specific drugs for the treatment of RB, researchers utilized the scDrug to predict potential drug responses based on the malignant cells. According to these criteria, the most effective potential drug for RB was determined to be GSK1070916, a highly potent Aurora B/C kinase inhibitor. GSK1070916 demonstrated broad anti-tumor activity in tissue culture cells and human tumor xenograft models, exhibiting the highest cytotoxicity across all malignant cell clusters in scDrug predictions (Fig. 2 E). Transcriptomic dynamics and trajectory analysis of RB cone cells To further investigate the functional role of cone cells in RB, RNA velocity and pseudotime analysis were employed to deduce the developmental and exhaustion processes. scVelo was utilized for RNA velocity analysis, revealing a transition from low to high levels of disulfidptosis during the process from normal cone cells to malignant cone cells (Fig. 3 A-C). Additionally, Monocle 2 was employed for pseudotime analysis to dissect cone cell development in RB. Intriguingly, it was observed that the levels of disulfidptosis was changed during the transition from normal cone cells to malignant cone cells(Figure 3 D-F). Figure 3 G illustrated the expression profiles of 23 DRGs in cone cells. Among them, SLC7A1 exhibited overexpression in malignant cells. A previous study has indicated that overexpression of SLC7A11 in cancer inhibits ferroptosis, thereby playing a crucial role in promoting tumor growth ( 12 ). Identification of DRGs clusters in RB. Based on the expression profiles of 23 DRGs, a consensus clustering algorithm was applied to analyze the disulfidptosis expression patterns. When k = 2, the cluster numbers were most steady and the results of cluster analysis were most reliable (Fig. 4 A-C). Analysis of PCA uncovered that there were significant differences in transcriptome between the two clusters, named cluster A and cluster B (Fig. 4 D). Compared with cluster B, most DRGs were highly expressed in cluster A, including SLC7A11, GYS1, NDUFS1, NDUFA11, NCKAP1, LRPPRC, SLC3A2, RPN1, ACTN4, ACTB, CD2AP, CAPZB, DSTN, INF2, IGGA1, MYH10, MYL6, MYH9 (Fig. 4 E-F). Identification of core modules of DRG through WGCNA To construct the scale-free network, the most appropriate soft threshold of β was 14 (Fig. 5 A). Cluster diagram was constructed according to the “cutree” dynamic and module eigengenes functions, and then a total of 20 module signature genes with different colors was obtained (Fig. 5 B-C). The association between each module and RB was presented in a heatmap of module-trait relationships, suggesting that the pink module had largest correlation with RB and was negatively associated with RB (Fig. 5 C). The correlation between pink module genes and RB genes was cor = 0.66 (Fig. 5 D). Additionally, the gene module-related to disulfidptosis were analyzed by WGCNA algorithm. The scale-free network was constructed when the soft threshold was set as 16 and scale-free R2 was set as 0.9, and 16 significant modules with different colors were obtained (Fig. 6 A-B). Following the module-clinical traits (cluster A and cluster B) relationship explanation, cluster A and cluster B were highly associated with the red module (Fig. 6 C). The correlation between pink module genes and DRGs was cor = 0.94 (Fig. 6 D). Selection of DRGs using Machine learning models The overlapping expression genes were screened from DEG (scRNA), modGene (Disease WGCNA) and modGene (Cluster WGCNA), 27 overlapping expression genes were obtained (Fig. 7 A). Then, four validated machine-learning models (GLM, RF, SVM and XGB) were constructed to identify the DEGs with high diagnostic values. GLM machine learning model exhibited a relatively lower residual (Fig. 7 B-C). As shown in Fig. 7 D, the feature importance diagram of each machine learning model was depended on the root mean square error. Moreover, we calculated ROC curves of each machine learning algorithm, all machine learning model displayed 1.0 of AUC (Fig. 7 E). Two overlapping genes were obtained from the four memory models, including CDT1 and NUSAP1 (Fig. 7 F). Based on the training data, both CDT1 and NUSAP1 exhibited an AUC value of 1 (Fig. 7 G). When validated with the GSE111168 dataset, CDT1 and NUSAP1 had AUC values of 0.860 and 0.763, respectively (Fig. 7 H). Hence, it was equally well in distinguishing between normal and RB when using our diagnostic model. CDT1 silencing inhibited malignant phenotypes of retinoblastoma cells. To verify the role of CDT1 in RB, we examined CDT1 expression in WERI-Rb1 cells with low invasiveness and Y79 cells with high invasiveness. Compared with ARPE-19 cells, the mRNA expression of CDT1 was notably increased in WERI-Rb1 and Y79 cells (Fig. 8 A). Then, CDT1 was silenced in both WERI-Rb1 and Y79 cells, and cell proliferation ability was examined by performing colony formation assay. CDT1 knockdown potently reduced proliferation ability of WERI-Rb1 and Y79 cells (Fig. 8 B). This reduction was accompanied by inhibition in migration and invasion, when WERI-Rb1 and Y79 cells were transduced with si-CDT1, as confirmed by transwell migration and invasion assays (Fig. 8 C-E). These in vitro experiments demonstrated that CDT1 silencing inhibited proliferation, migration and invasion of retinoblastoma cells. Discussion Under the effect of tumorigenic factors, tumors have deteriorated and proliferated from normal cells to cancer cells, and have accumulated mutations and gradually differentiated into different genetic lineages and subgroups during the evolution process, thereby forming heterogeneity within tumors ( 13 ). Tumor intratumoral heterogeneity is closely related to tumor invasion and metastasis, and also affects the clinical diagnosis and treatment of patients. RB is also a highly heterogeneous tumor ( 14 ). ScRNA seq technology detects the heterogeneity of tumor cells, identifies rare cells, divides cell subsets, tracks cell lineage, locates mutation genes, and discovers new tumor biomarkers through a single cell ( 15 , 16 ). In this work, we analyzed 10199 cells from RB samples and normal samples by scRNA seq technology, and obtained 7 annotated cell clusters, including cone cells, epithelial cells, endothelial cells, astrocytes, microglial cells, Muller glial cells, neuron. Among them, the fraction of cone cells was significantly increased in RB samples. Cobrinik et al. have pointed out that RB originates from cone precursor cells, and RB is formed in response to single gene mutation of RB1 ( 17 ). RB1 gene encodes a tumor suppressor protein Rb, which prevents excessive cell growth by inhibiting the process of cell cycle and preventing cells from malignant proliferation and then transforming into cancer cells ( 18 , 19 ). Thus, malignant cone cells may appear in the increased cone cells to promote RB progression. Further analysis revealed that cone cells were divided into normal cells and malignant cone cells. Compared with normal cone cells, malignant cone cells exhibited high disulfidptosis. SLC7A11, a DRG, was notably associated with malignant cone cells. Disulfidptosis is a new type of cell death that is independent of existing programmed cell death such as apoptosis, ferroptosis, and necrotic apoptosis and cuproptosis ( 20 ). It is a rapid death mode caused by disulfide stress-mediated excessive accumulation of cystine in cells, which dependents on high expression of SLC7A11. Several studies have shown that SLC7A11 is highly expressed in breast cancer, lung cancer and bladder cancer, and affects the metabolism, occurrence and development of tumors ( 21 – 23 ). In this work, we found that SLC7A11 expression was associated with malignant cone cells. SLC7A11 may regulate disulfidptosis to drive malignant progression of cone cells. Applying machine learning methods, we found 2 genes were closely associated with RB, including CDT1 and NUSAP1. These genes may participate in the progression of RB, especially CDT1. CDT1 is an important regulator of DNA replication licensing, which is essential for initiation of DNA replication ( 24 ). Previous studies have confirmed that CDT1 is a potent oncogene in several cancers. For instance, Cai et al. have found that CDT1 accelerates malignant progression of hepatocellular carcinoma, which may be associated with cell cycle, DNA repair, DNA replication and minichromosome maintenance ( 25 ). High expression of CDT1 in colorectal cancer is correlated significantly with increased origin licensing, activation of the DNA damage response, and microsatellite instability ( 26 ). CDT1 is reported to related to poor prognosis of triple-negative breast cancer patients, CDT1 promotes cellular processes of triple-negative breast cancer ( 27 ). In the present work, we first confirmed the carcinogenic effect of CDT1 on RB. Compared with ARPE-19 cells, CDT1 was up-regulated in both Y79 cells with high invasiveness and WERI-Rb1 cells with low invasiveness. CDT1 knockdown suppressed the proliferation, migration and invasion of RB cells. In conclusion, this work revealed that tumors contained cone cell states with distinct transcriptional programs, and provided a crucial RB-related gene CDT1 for tumor progression. Thus, this article is of great significance in formulating a molecular targeted therapy scheme to prevent the development of RB. Declarations Ethics approval and consent to participate N/A Conflict of Interest The authors have no conflicts of interest to declare. Declaration of competing interest The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest. Consent for publication N/A. Funding Not applicable. Author Contribution Conception: Y Y.Interpretation or analysis of data: Y Y, YZ Z.Preparation of the manuscript: Y Y, YZ Z, WW X.Revision for important intellectual content: Y Y, YZ Z.Supervision: XL Y. Acknowledgments Not applicable. Availability of data and materials All data related to the article can be obtained from the corresponding author upon reasonable request. References Rao R, Honavar SG, Retinoblastoma. Indian J Pediatr. 2017;84(12):937–44. Chung SE, Sa HS, Koo HH, Yoo KH, Sung KW, Ham DI. Clinical manifestations and treatment of retinoblastoma in Korea. Br J Ophthalmol. 2008;92(9):1180–4. Ortiz MV, Dunkel IJ, Retinoblastoma. J Child Neurol. 2016;31(2):227–36. Ting SC, Kiefer T, Ehlert K, Goericke SL, Hinze R, Ketteler P, Bechrakis NE, Schildhaus HU. Bone metastasis of retinoblastoma five years after primary treatment. Am J Ophthalmol case Rep. 2020;19:100834. Ancona-Lezama D, Dalvin LA, Shields CL. Modern treatment of retinoblastoma: A 2020 review. Indian J Ophthalmol. 2020;68(11):2356–65. Dimaras H, Dimba EA, Gallie BL. Challenging the global retinoblastoma survival disparity through a collaborative research effort. Br J Ophthalmol. 2010;94(11):1415–6. Zheng P, Zhou C, Ding Y, Duan S. Disulfidptosis: a new target for metabolic cancer therapy. J experimental Clin cancer research: CR. 2023;42(1):103. Liu X, Nie L, Zhang Y, Yan Y, Wang C, Colic M, Olszewski K, Horbath A, Chen X, Lei G, Mao C, Wu S, Zhuang L, Poyurovsky MV, James You M, Hart T, Billadeau DD, Chen J, Gan B. Actin cytoskeleton vulnerability to disulfide stress mediates disulfidptosis. Nat Cell Biol. 2023;25(3):404–14. Ji PY, Li ZY, Wang H, Dong JT, Li XJ, Yi HL. Arsenic and sulfur dioxide co-exposure induce renal injury via activation of the NF-κB and caspase signaling pathway. Chemosphere. 2019;224:280–8. Chen H, Yang W, Li Y, Ma L, Ji Z. Leveraging a disulfidptosis-based signature to improve the survival and drug sensitivity of bladder cancer patients. Front Immunol. 2023;14:1198878. Wang T, Guo K, Zhang D, Wang H, Yin J, Cui H, Wu W. Disulfidptosis classification of hepatocellular carcinoma reveals correlation with clinical prognosis and immune profile. Int Immunopharmacol. 2023;120:110368. Koppula P, Zhuang L, Gan B. Cystine transporter SLC7A11/xCT in cancer: ferroptosis, nutrient dependency, and cancer therapy. Protein Cell. 2021;12(8):599–620. Tabassum DP, Polyak K. Tumorigenesis: it takes a village. Nat Rev Cancer. 2015;15(8):473–83. Stenfelt S, Blixt MKE, All-Ericsson C, Hallböök F, Boije H. Heterogeneity in retinoblastoma: a tale of molecules and models. 2017;6(1):42. Zhao Y, Feng Y, Liu M, Chen L, Meng Q, Tang X, Wang S, Liu L, Li L, Shen W, Zhang H. Single-cell RNA sequencing analysis reveals alginate oligosaccharides preventing chemotherapy-induced mucositis. Mucosal Immunol. 2020;13(3):437–48. Ross EM, Markowetz F. OncoNEM: inferring tumor evolution from single-cell sequencing data. Genome Biol. 2016;17:69. Xu XL, Singh HP, Wang L, Qi DL, Poulos BK, Abramson DH, Jhanwar SC, Cobrinik D. Rb suppresses human cone-precursor-derived retinoblastoma tumours. Nature. 2014;514(7522):385–8. McEvoy JD, Dyer MA. Genetic and Epigenetic Discoveries in Human Retinoblastoma. Crit Rev Oncog. 2015;20(3–4):217–25. Giacinti C, Giordano A. RB and cell cycle progression. Oncogene. 2006;25(38):5220–7. Liu X, Zhuang L, Gan B. Disulfidptosis: disulfide stress-induced cell death. Trends Cell Biol. 2023. Yang J, Zhou Y, Xie S, Wang J, Li Z, Chen L, Mao M, Chen C, Huang A, Chen Y, Zhang X, Khan NUH, Wang L, Zhou J. Metformin induces Ferroptosis by inhibiting UFMylation of SLC7A11 in breast cancer. J experimental Clin cancer research: CR. 2021;40(1):206. Zhang W, Sun Y, Bai L, Zhi L, Yang Y, Zhao Q, Chen C, Qi Y, Gao W, He W, Wang L, Chen D, Fan S, Chen H, Piao HL, Qiao Q, Xu Z, Zhang J, Zhao J, Zhang S, Yin Y, Peng C, Li X, Liu Q, Liu H, Wang Y. RBMS1 regulates lung cancer ferroptosis through translational control of SLC7A11. The Journal of clinical investigation. 2021;131(22). Shen L, Zhang J, Zheng Z, Yang F, Liu S, Wu Y, Chen Y, Xu T, Mao S, Yan Y, Li W, Zhang W, Yao X. PHGDH Inhibits Ferroptosis and Promotes Malignant Progression by Upregulating SLC7A11 in Bladder Cancer. Int J Biol Sci. 2022;18(14):5459–74. Ratnayeke N, Baris Y, Chung M, Yeeles JTP, Meyer T. CDT1 inhibits CMG helicase in early S phase to separate origin licensing from DNA synthesis. Mol Cell. 2023;83(1):26–e4213. Cai C, Zhang Y, Hu X, Hu W, Yang S, Qiu H, Chu T. CDT1 Is a Novel Prognostic and Predictive Biomarkers for Hepatocellular Carcinoma. Front Oncol. 2021;11:721644. Petropoulos M, Champeris Tsaniras S, Nikou S, Maxouri S, Dionellis VS, Kalogeropoulou A, Karamichali A, Ioannidis K, Danalatos IR, Obst M, Naumann R, Delinasios GJ, Gorgoulis VG, Roukos V, Anastassiadis K, Halazonetis TD, Bravou V, Lygerou Z, Taraviras S. Cdt1 overexpression drives colorectal carcinogenesis through origin overlicensing and DNA damage. J Pathol. 2023;259(1):10–20. Maimaiti Y, Zhang N, Zhang Y, Zhou J, Song H, Wang S. CircFAM64A enhances cellular processes in triple-negative breast cancer by targeting the miR-149-5p/CDT1 axis. Environ Toxicol. 2022;37(5):1081–92. Additional Declarations No competing interests reported. Supplementary Files S1.tif Cite Share Download PDF Status: Posted Version 1 posted You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-4883827","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":350215948,"identity":"6317a937-7340-4fba-a74f-a61674525f78","order_by":0,"name":"Yang Yang","email":"","orcid":"","institution":"Second Affiliated Hospital of Nanchang University","correspondingAuthor":false,"prefix":"","firstName":"Yang","middleName":"","lastName":"Yang","suffix":""},{"id":350215949,"identity":"07a09a77-8304-4f98-8dc1-23f3051694b4","order_by":1,"name":"Yuezhi Zhang","email":"","orcid":"","institution":"Second Affiliated Hospital of Nanchang University","correspondingAuthor":false,"prefix":"","firstName":"Yuezhi","middleName":"","lastName":"Zhang","suffix":""},{"id":350215950,"identity":"60f6c713-4319-4179-8c3c-61a6b8d181c2","order_by":2,"name":"Weiwei Xiong","email":"","orcid":"","institution":"Second Affiliated Hospital of Nanchang University","correspondingAuthor":false,"prefix":"","firstName":"Weiwei","middleName":"","lastName":"Xiong","suffix":""},{"id":350215951,"identity":"10dcc3a4-173d-4534-86bf-d704ab4e1578","order_by":3,"name":"Xiaolong Yin","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA8klEQVRIiWNgGAWjYFCCBIYPDBU2PPwMhw9ABA4Q1sI4g+FMmpxk47EEErQwth02Njh8xoA4LebsyQebec6kJTYcO/Ptwc82Bjm+GwmMnwvwaLHseZbYzFNhk9jYc3a7YW8bg7HkjQRm6Rl4tBjcyDF/DLKlWeLsNmnGNobEDTcS2Jh58GrJ/9jM23Y4sU3+zTOQlnoitOQwgrQY8zCcYQNpSTAgqOXMM8PGOcBAlmA4ZibZc07CcOaZh83SeLUcT37Y8AYYlfYHDj+T+FFmI893PPngZ3xaQIAJSYEEEDM2ENAAVPKDoJJRMApGwSgY0QAAVZpV66qJyjUAAAAASUVORK5CYII=","orcid":"","institution":"Second Affiliated Hospital of Nanchang University","correspondingAuthor":true,"prefix":"","firstName":"Xiaolong","middleName":"","lastName":"Yin","suffix":""}],"badges":[],"createdAt":"2024-08-09 02:29:17","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-4883827/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-4883827/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":64103905,"identity":"1b15bc0e-f590-43da-8549-e2b971edf6b8","added_by":"auto","created_at":"2024-09-06 23:49:34","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":6042824,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eAnnotation of cell subtypes and AUcell Scoring in scRNA-seq data\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e(A) UMAP visualization of cell type in Normal samples and RB samples. (B) Expression of cell type marker genes. (C) Line chart was generated to depict the tissue prevalence for each cell type as estimated by the Ro/e score\u003c/p\u003e\n\u003cp\u003e（D-E）All cells were evaluated based on DRGs using AUCell scoring, and subsequently categorized into high and low groups.\u003c/p\u003e","description":"","filename":"Fig1.png","url":"https://assets-eu.researchsquare.com/files/rs-4883827/v1/d5d71ddc51fee23b1406d107.png"},{"id":64103779,"identity":"e9d8faca-f708-4e7a-ac22-0294e8ed685a","added_by":"auto","created_at":"2024-09-06 23:33:34","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":5804753,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eIdentification of RB-specific candidate drugs for therapy via scRNA-seq analysis\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e(A-B) Umap plot of cells by CNVs scores and tumor types. (C) Enrichment analysis of biological changes was conducted for both normal cone cells and malignant cone cells. (D) UMAP visualization of malignant cone cell clusters. (E) Heatmaps were generated to display the potential drugs within the PRISM database, which were predicted to inhibit cell growth using CaDRReS-Sc. Each cell within the heatmaps represents the predicted sensitivity score of the tumor cell clusters to the respective drugs.\u003c/p\u003e","description":"","filename":"Fig2.png","url":"https://assets-eu.researchsquare.com/files/rs-4883827/v1/7e0b541f73613cc441482388.png"},{"id":64103776,"identity":"d659ee61-32ca-4006-a56e-8044a8a9c243","added_by":"auto","created_at":"2024-09-06 23:33:34","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":2363399,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eThe developmental trajectory of cone cells in RB\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e(A-C) RNA velocity analysis revealed dynamic transitions between RB cone cells and malignant cells, accompanied by alterations in DRG scores and variations among distinct DRG groups. (D-F) The developmental trajectory of normal and malignant cone cells with disulfidptosis score. (G) The expression of in normal and malignant cone cells.\u003c/p\u003e","description":"","filename":"Fig3.png","url":"https://assets-eu.researchsquare.com/files/rs-4883827/v1/a6f98127297cbf0927d3249f.png"},{"id":64103777,"identity":"dcbefb68-e561-4770-8612-3a5fc0cc8833","added_by":"auto","created_at":"2024-09-06 23:33:34","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":1342860,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eIdentifcation of molecular clusters related to DRGs in RB.\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e(A) Cluster-consensus matrix when k=2. (B) CDF curves. (C) CDF delta area curves.\u003c/p\u003e\n\u003cp\u003e(D) PCA visualized the distribution of two subtypes (Cluster A and Cluster B). (E-F) The difference in DRGs between Cluster A and Cluster B.\u003c/p\u003e","description":"","filename":"Fig4.png","url":"https://assets-eu.researchsquare.com/files/rs-4883827/v1/3546e9783acb5af821e95ca4.png"},{"id":64103880,"identity":"b8373a2b-54c6-4fcf-ab54-697189121f96","added_by":"auto","created_at":"2024-09-06 23:41:34","extension":"png","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":2505541,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eCo-expression network of DRGs in RB.\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e(A) The option of soft threshold (power). (B) Clustering dendrogram of genes in co-expression modules. Different colors repensented distinct co-expression modules. (C) Correlation analysis between module eigengenes and clinical status. (D) Scatter plot between module membership in pink module and the gene significance for RB.\u003c/p\u003e","description":"","filename":"Fig5.png","url":"https://assets-eu.researchsquare.com/files/rs-4883827/v1/47cae46abf4c7c565a4432f8.png"},{"id":64103783,"identity":"1941e419-d68e-4420-af29-34085cc11bcb","added_by":"auto","created_at":"2024-09-06 23:33:34","extension":"png","order_by":6,"title":"Figure 6","display":"","copyAsset":false,"role":"figure","size":2186590,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eCo-expression network of DRGs in cluster A and cluster B.\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e(A) The option of soft threshold (power). (B) Clustering dendrogram of genes in co-expression modules. Different colors repensented distinct co-expression modules. (C) Correlation analysis between module eigengenes and clusters. (D) Scatter plot between module membership in red module and the gene significance for Cluster A and Cluster B.\u003c/p\u003e","description":"","filename":"Fig6.png","url":"https://assets-eu.researchsquare.com/files/rs-4883827/v1/555783a7f6421b73c5866bb5.png"},{"id":64103781,"identity":"1bf52385-21d7-41ba-adc6-3c95b5c90364","added_by":"auto","created_at":"2024-09-06 23:33:34","extension":"png","order_by":7,"title":"Figure 7","display":"","copyAsset":false,"role":"figure","size":1609878,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eDRG selection via diverse Machine learning algorithms.\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e(A) Venn diagram showing the overlapping gene among scRNA, Disease WGCNA and Cluster WGCNA. (B) The residuals of each machine learning model were presented in boxplots. The root mean square of residuals is shown as a red dot. (C) The reverse cumulative residual distribution of each machine learning model. (D) The important features in four machine models. (E) ROC analysis of four machine models based on fivefold cross-validation in the testing group. (F) Venn diagram showing the overlapping gene among four machine models. (G) ROC curves for CDT1 and NUSPA1 Targets in the training set. (H) ROC curve for predicted results of CDT1 and NUSPA1 in the test dataset.\u003c/p\u003e","description":"","filename":"FIg7.png","url":"https://assets-eu.researchsquare.com/files/rs-4883827/v1/c9f679b19ceb278ba0d40a7c.png"},{"id":64103778,"identity":"353cc1b7-1612-4e59-8da4-b8d923d7fcbf","added_by":"auto","created_at":"2024-09-06 23:33:34","extension":"png","order_by":8,"title":"Figure 8","display":"","copyAsset":false,"role":"figure","size":2479924,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eCDT1 knockdown inhibited proliferation, migration and invasion of WERI-Rb1 and Y79 cells.\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u0026nbsp;(A) QRT-PCR detected the expression of CDT1 in WERI-Rb1 and Y79 cells. (B) Colony formation assay examined proliferation of WERI-Rb1 and Y79 cells. (C-E) Transwell assay assessed migration and invasion of WERI-Rb1 and Y79 cells. \u003csup\u003e*\u003c/sup\u003e\u003cem\u003eP\u003c/em\u003e \u0026lt; 0.05, \u003csup\u003e**\u003c/sup\u003e\u003cem\u003eP\u003c/em\u003e \u0026lt; 0.01 vs. WERI-Rb1 group; \u003csup\u003e#\u003c/sup\u003e\u003cem\u003eP\u003c/em\u003e \u0026lt; 0.05, \u003csup\u003e##\u003c/sup\u003e\u003cem\u003eP\u003c/em\u003e \u0026lt; 0.01 vs. OE-NC group.\u0026nbsp;\u003c/p\u003e","description":"","filename":"Fig8.png","url":"https://assets-eu.researchsquare.com/files/rs-4883827/v1/e13b282d62c6860090a2547d.png"},{"id":64104021,"identity":"fb85fd57-f495-414c-8380-5be9afa1796a","added_by":"auto","created_at":"2024-09-07 00:06:00","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":23861196,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-4883827/v1/58eca32f-373a-4b17-8891-9596d01007f6.pdf"},{"id":64103784,"identity":"16602b11-4462-416a-8a8c-dbe08ae6bdf2","added_by":"auto","created_at":"2024-09-06 23:33:34","extension":"tif","order_by":10,"title":"","display":"","copyAsset":false,"role":"supplement","size":5364608,"visible":true,"origin":"","legend":"","description":"","filename":"S1.tif","url":"https://assets-eu.researchsquare.com/files/rs-4883827/v1/62c2563093b4afe8b530785b.tif"}],"financialInterests":"No competing interests reported.","formattedTitle":"Single-Cell Sequencing Analysis and Multiple Machine Learning Methods identified CDT1 as an oncogene in retinoblastoma","fulltext":[{"header":"Introduction","content":"\u003cp\u003eRetinoblastoma (RB) is a malignant tumor derived from photoreceptor precursor cells. RB is most commonly seen in children under 5 years of age, and is the most common primary intraocular malignant tumor in infants, accounting for 2\u0026ndash;4% of children's malignant tumors (\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e). The clinical manifestations of retinoblastoma are complex, including conjunctival congestion, corneal edema, iris neovascularization, vitreous opacity, elevated intraocular pressure and strabismus (\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e, \u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e). At the same time, RB also seriously affects the quality of life of infants. RB is prone to intracranial and distant metastasis, thereby often endangering the lives of children (\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e). In oncology, tumor staging is extremely important for the selection of treatment methods and evaluation of efficacy, as is the case with RB (\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e). Moreover, many RB patients have a high mortality rate due to failure of early diagnosis, especially in developing countries (\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e). Thus, the characteristics and identification of genes closely related to the progression of RB have certain reference significance for the research of RB progression and early diagnosis.\u003c/p\u003e \u003cp\u003eDisulfidptosis, a new type of programmed cell death, is a rapid mode of death caused by disulfide stress caused by excessive accumulation of cystine in cells (\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e). It usually occurs under conditions of glucose starvation. In cells with high expression of SLC7A11, insufficient supply of NADPH results in limited reduction of cysteine to cysteine, leading to disulfide stress and ultimately cell death (\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e). NF⁃κB and c-JNK signaling pathway have also been confirmed to play important roles in the formation of disulfidptosis (\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e). In malignant tumors, intracellular disulfidptosis and oxidative stress play a role, and the formation of invasive pseudopodia may lead to more disulfidptosis in metastatic malignant tumor cells (\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e). Therefore, the study of the mechanism of disulfidptosis is an important new direction in the field of disease treatment (\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e). A previous study has reported that biomarkers related to disulfidptosis can be used for clinical diagnosis and prognosis of liver cancer (\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e). However, the role of disulfidptosis-related genes in RB has not been reported.\u003c/p\u003e \u003cp\u003eThe aim of this study was to explore the role of disulfidptosis in RB and its impact on tumor biology. Using single-cell analysis, we elucidated the complex relationship between the progression of malignant cone cells and disulfidptosis, as assessed through AUCell scores and RNA velocity. Additionally, machine learning was employed to identify the disulfidptosis-related gene, CDT1. \u003cem\u003eIn vitro\u003c/em\u003e assays revealed that CDT1 silencing inhibited the proliferation, migration, and invasion of RB cells. Therefore, this study has provided valuable new data resources for understanding the pathogenesis and progression of RB.\u003c/p\u003e"},{"header":"Materials and methods","content":"\u003cdiv id=\"Sec3\" class=\"Section2\"\u003e \u003ch2\u003eData collection\u003c/h2\u003e \u003cp\u003eSingle-cell transcriptomic data were retrieved from Gene Expression Omnibus (\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://www.ncbi.nlm.nih.gov/geo/\u003c/span\u003e\u003cspan address=\"https://www.ncbi.nlm.nih.gov/geo/\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e) database. RB samples diagnosed for 4 months or 2 years and Normal samples were obtained from GSE159977 dataset. Besides, bulk transcriptomic data were retrieved from GEO databases (GSE97508, GSE110811, GSE24673, GSE208143). GSE97508, GSE208143 and GSE24673 datasets were merged as train dataset. Batch effects from non-biological technical biases were corrected using the \u0026ldquo;ComBat\u0026rdquo; algorithm of \u0026ldquo;sva\u0026rdquo; package. GSE110811 dataset was used for validation.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec4\" class=\"Section2\"\u003e \u003ch2\u003eScRNA-Seq Analysis\u003c/h2\u003e \u003cp\u003eThe analysis of the single-cell dataset was conducted using Scanpy (version 1.9.1) within Python (version 3.8.13). To ensure data quality, rigorous quality control procedures were applied to exclude the cells with undesirable attributes while the retaining cells meeting predefined criteria. These criteria encompassed gene counts ranging from 0.02 to 0.98, and mitochondrial gene counts below 20%. Additionally, Scrublet (version 1.0) was employed to identify and eliminate doublet cells. Principal Component Analysis (PCA) was executed via Scanpy to reduce the dimensionality of the cell data. To mitigate batch effects, the \"harmony_integrate\" function in Scanpy was employed. Clustering and visualization were performed using Leiden and UMAP algorithms. Finally, cell clusters were annotated with known cell types based on cell type-specific markers.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec5\" class=\"Section2\"\u003e \u003ch2\u003eDisulfidptosis‑related genes (DRGs) scoring\u003c/h2\u003e \u003cp\u003eThe \"AUCell\" R package, designed to assess the active status of gene sets in single-cell RNA data, was employed to compute activity scores for DRGs in each individual cell. Following this, cells were categorized into two distinct groups based on their DRG-AUC (Disulfide-Related Genes Area Under the Curve) values: those with high DRG-AUC and those with low DRG-AUC, with the median value being utilized as the threshold for classification.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec6\" class=\"Section2\"\u003e \u003ch2\u003eChromosome copy number variation analysis\u003c/h2\u003e \u003cp\u003eThe approach described in inferCNV and implemented as infercnvpy (\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://github.com/icbi-lab/infercnvpy\u003c/span\u003e\u003cspan address=\"https://github.com/icbi-lab/infercnvpy\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e) was employed to detect potential changes in chromosomal copy numbers. Control cells were selected from healthy tissues, including cone cells and immune cells. The differentiation between normal cone cells and malignant cells was accomplished through assessment of iterative clustering impact and computation of the copy number variation score.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec7\" class=\"Section2\"\u003e \u003ch2\u003eDrug Response Prediction\u003c/h2\u003e \u003cp\u003eWe utilized the scDrug Python package, built upon the CaDRReS-Sc model, to predict IC50 values and tumor cell killing ratios associated with potential drug responses. This framework incorporates pre-trained GDSC and PRISM models, which have undergone rigorous training using gene expression data to establish robust pharmacogenomic relationships, ensuring precise drug response predictions. Effective data visualization using the seaborn package allowed us to focus on IC50 values and tumor cell killing ratios across different subpopulations. Our drug selection criteria were based on minimizing IC50 values and maximizing tumor cell killing ratios.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec8\" class=\"Section2\"\u003e \u003ch2\u003eRNA velocity analysis\u003c/h2\u003e \u003cp\u003eVelocyto allows for the estimation of single-cell RNA velocity by distinguishing between unspliced and spliced reads in bam files. The process involves generating Loom files using velocyto.py and subsequently importing this data into scVelo for further analysis. The 'scv.pp.moments' function is utilized to calculate the first and second order velocity vector moments, while the 'scv.t.velocity' functions are employed to compute dynamic velocity. The RNA velocities are then embedded into UMAP using the velocity graph.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec9\" class=\"Section2\"\u003e \u003ch2\u003eDevelopmental trajectory inference\u003c/h2\u003e \u003cp\u003eMonocle2 ((version 2.14.0) was employed for trajectory analysis to uncover alterations in cell states. The \"estimateSizeFactors\" and \"estimateDispersions\" functions in Monocle2 were utilized to identify the \"dispersion\" genes, which were subsequently employed for cell ordering. Dimensionality reduction was conducted through the 'reduceDimension' function, with the 'DDRTree' reduction method being applied.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec10\" class=\"Section2\"\u003e \u003ch2\u003eConsensus cluster analysis\u003c/h2\u003e \u003cp\u003eThe Disulfidptosis-related genes were uniformly clustered using the ConsensusClusterPlus R package. This clustering analysis utilized the \"pam\" algorithm with \"euclidean\" distance measurement. Subsequently, 1000 bootstraps were executed, with each bootstrap comprising 80% of the patients from the training dataset. The range of cluster numbers, denoted as \"k,\" was set from 2 to 9. The optimal \"k\" was determined based on both the cumulative distribution function (CDF) and the area under the curve (AUC) analysis. Additionally, Principal Component Analysis (PCA) was conducted to assess the validity of the molecular subtype distributions.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec11\" class=\"Section2\"\u003e \u003ch2\u003eWeighted correlation network analysis\u003c/h2\u003e \u003cp\u003eWe initiated the process by constructing a clustering tree for the samples to detect and eliminate any outliers. Subsequently, the gene expression data from the training dataset underwent analysis using Weighted Gene Co-expression Network Analysis (WGCNA). The adjacency matrix was transformed into the Topological Overlapping Matrix (TOM). A threshold of R\u0026sup2; = 0.85 was established as the criterion for module construction. Dynamic Tree Cut was employed with parameters set to deepSplit\u0026thinsp;=\u0026thinsp;2 and minModuleSize\u0026thinsp;=\u0026thinsp;100 to generate these modules. Key modules were selected based on the correlation between their members and the significance of the included genes.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec12\" class=\"Section2\"\u003e \u003ch2\u003eSelection of characteristic genes\u003c/h2\u003e \u003cp\u003ePredictive models were created using common genes identified from both scRNA and WGCNA. These models, which included Random Forest (RF), Support Vector Machine (SVM), Extreme Gradient Boosting (XGBoost), and General Linear Model (GLM), were implemented within the R software. Model performance assessment included the generation of reverse cumulative distribution curves for residuals and Receiver Operating Characteristic (ROC) curves. Model selection was based on scores derived from reverse cumulative residuals and Area Under the Curve (AUC) values obtained from the ROC curves.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec13\" class=\"Section2\"\u003e \u003ch2\u003eCell culture\u003c/h2\u003e \u003cp\u003eHuman retinal pigment epithelial cells (ARPE-19) were obtained from ATCC (Manassas, VA, USA). RB cell lines WERI-Rb1 and Y79 were purchased from Procell Life Science \u0026amp; Technology Co., Ltd (Wuhan, China) and National Collection of Authenticated Cell Cultures (Shanghai, China), respectively. All cells were cultured in RPMI-1640 medium (Gibco, Grand Island, NY, USA) at 37\u0026deg;C and 5% CO\u003csub\u003e2\u003c/sub\u003e atmosphere. The medium was supplemented with 1% L-glutamine (Sigma-Aldrich, St. Louis, MO, USA), 1% penicillin/streptomycin (Sangon Biotech, Shanghai, China) and 10% fetal bovine serum (FBS; Gibco).\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec14\" class=\"Section2\"\u003e \u003ch2\u003eQuantitative real-time PCR (qRT-PCR)\u003c/h2\u003e \u003cp\u003eRNA extraction was carried out applying TRIzol regent (Thermo Fisher Scientific, San Jose, CA, USA) to isolate RNA from cells. PrimeScript\u0026trade; RT reagent Kit (Takara, Dalian, China) was applied to synthesize cDNA with RNA as template. Genomic DNA was eliminated with gDNA Eraser. PCR reacted was conducted to examine CDT1 expression applying TB Green\u0026reg; Premix Ex Taq\u0026trade; (Takara). Primer sequences (5\u0026rsquo;-3\u0026rsquo;) were shown as follow: CDT1: forward-GACATGATGCGTAGGCGTTTT and reverse-GAGCTGGTAATCTGACCTCCT-3; GAPDH: forward-CCACCCATGGCAAATTCCATGGCA and reverse-TCTAGACGGCAGTCAGGTCCACC. The relative expression of CDT1 was analyzed by 2\u003csup\u003e\u0026minus;ΔΔCT\u003c/sup\u003e method.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec15\" class=\"Section2\"\u003e \u003ch2\u003eColony formation assay\u003c/h2\u003e \u003cp\u003eCells at logarithmic phase was seeded into a 6-well plate with density of 1 \u0026times; 103 cells/plate. After 2 weeks of incubation at 37\u0026deg;C and 5% CO\u003csub\u003e2\u003c/sub\u003e, cells were fixed with 4% paraformaldehyde, and then stained with 0.5% crystal violet. Finally, the number of colony was counted.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec16\" class=\"Section2\"\u003e \u003ch2\u003eTranswell migration and invasion assays\u003c/h2\u003e \u003cp\u003eTranswell assay was conducted to examine migration and invasion of cells applying the 8 \u0026micro;m migration Transwell chambers (Corning, NY, USA). For cell migration assay, cells were added to the upper chambers at a concentration of 5\u0026times; 10\u003csup\u003e4\u003c/sup\u003e cells/200 \u0026micro;L of FBS-free medium, whereas the lower chambers contained 700 \u0026micro;L of medium and 10% FBS. For invasion assay, cells were seeded into the upper chambers coating with 25 \u0026micro;L Matrigel (Sigma-Aldrich, St. Louis, MO, USA). After 24 h of culture, the migrated and invaded cells were stained with 0.5% crystal violet reagent, which were then observed under an optical microscope.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec17\" class=\"Section2\"\u003e \u003ch2\u003eStatistical analysis\u003c/h2\u003e \u003cp\u003eEach assay was performed for 3 times. Data were analyzed by SPSS 22.0 statistical software (IBM, Armonk, NY, USA) and expressed as mean\u0026thinsp;\u0026plusmn;\u0026thinsp;standard deviation. Two-tailed Student\u0026rsquo;s \u003cem\u003et\u003c/em\u003e test and one-way ANOVA were used to analyze the statistical difference. \u003cem\u003eP\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;0.05 was considered as a significant difference.\u003c/p\u003e \u003c/div\u003e"},{"header":"Results","content":"\u003cdiv id=\"Sec19\" class=\"Section2\"\u003e \u003ch2\u003eAnalysis of RB single‑cell sequencing data\u003c/h2\u003e \u003cp\u003eIn this investigation, we utilized the scRNA-seq expression matrix to investigate the presence of disulfidptosis in RB patients. After rigorous quality control procedures, we retained a total of 10,199 high-quality cells for subsequent analysis. Following batch correction, we integrated the single-cell transcriptomic data and performed UMAP-based clustering to identify 18 different cell subclusters. Subsequently, based on established canonical marker gene, we annotated these clusters and classified them into cone cells, endothelial cells, epithelial cells, astrocytes, microglial cells, M\u0026uuml;ller glial cells, and neurons (Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003eA-B). Moreover, we quantified the tissue enrichment of all these cell populations based on Ro/e analysis and found that only cone cells were enriched in RB tissue (Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003eC). The DRG score of each cell line was evaluated using the R package \"AUCell\". As shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003eD, cone cells, epithelial cells, and neurons exhibited higher DRG scores. Based on the median of DRG scores, the cells were classified into high DRG-AUC and low DRG-AUC groups. We then examined the differences in DRG scores of different cell types between normal and RB tissue. Compared with normal tissue, higher DRG scores were observed in cone cells, endothelial cells, astrocytes, and microglial cells of RB tissues, while lower DRG scores were observed in epithelial cells, Muller cells and neurons of RB tissues (Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003eE).\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec20\" class=\"Section2\"\u003e \u003ch2\u003eIdentification of potential drug candidates for RB based on scRNA-seq analyses\u003c/h2\u003e \u003cp\u003eTo determine the potential drug in RB, we inferred copy number variations (CNVs) using the Python package inferCNVpy. By generating single-cell CNV scores, we differentiated between normal cone cells and malignant cone cells (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003eA). The critical threshold was determined as 0.0267 by calculating the average CNV score of cone cells, with values greater than the threshold considered as malignant cone cells (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003eB). From the enrichment analysis, normal cone cells were mainly concentrated on phototransduction, ECM-repair interaction, aldosterone synthesis and secretion. Malignant cone cells were mainly related to cell cycle, asthma, folate biosynthesis (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003eC). The Leiden algorithm was employed to cluster malignant cells, revealing the identification of four distinct clusters with varying distributions (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003eD). To identify specific drugs for the treatment of RB, researchers utilized the scDrug to predict potential drug responses based on the malignant cells. According to these criteria, the most effective potential drug for RB was determined to be GSK1070916, a highly potent Aurora B/C kinase inhibitor. GSK1070916 demonstrated broad anti-tumor activity in tissue culture cells and human tumor xenograft models, exhibiting the highest cytotoxicity across all malignant cell clusters in scDrug predictions (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003eE).\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec21\" class=\"Section2\"\u003e \u003ch2\u003eTranscriptomic dynamics and trajectory analysis of RB cone cells\u003c/h2\u003e \u003cp\u003eTo further investigate the functional role of cone cells in RB, RNA velocity and pseudotime analysis were employed to deduce the developmental and exhaustion processes. scVelo was utilized for RNA velocity analysis, revealing a transition from low to high levels of disulfidptosis during the process from normal cone cells to malignant cone cells (Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003eA-C). Additionally, Monocle 2 was employed for pseudotime analysis to dissect cone cell development in RB. Intriguingly, it was observed that the levels of disulfidptosis was changed during the transition from normal cone cells to malignant cone cells(Figure \u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003eD-F). Figure\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003eG illustrated the expression profiles of 23 DRGs in cone cells. Among them, SLC7A1 exhibited overexpression in malignant cells. A previous study has indicated that overexpression of SLC7A11 in cancer inhibits ferroptosis, thereby playing a crucial role in promoting tumor growth (\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e).\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003e \u003cb\u003eIdentification of DRGs clusters in RB.\u003c/b\u003e \u003c/p\u003e \u003cp\u003eBased on the expression profiles of 23 DRGs, a consensus clustering algorithm was applied to analyze the disulfidptosis expression patterns. When k\u0026thinsp;=\u0026thinsp;2, the cluster numbers were most steady and the results of cluster analysis were most reliable (Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003eA-C). Analysis of PCA uncovered that there were significant differences in transcriptome between the two clusters, named cluster A and cluster B (Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003eD). Compared with cluster B, most DRGs were highly expressed in cluster A, including SLC7A11, GYS1, NDUFS1, NDUFA11, NCKAP1, LRPPRC, SLC3A2, RPN1, ACTN4, ACTB, CD2AP, CAPZB, DSTN, INF2, IGGA1, MYH10, MYL6, MYH9 (Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003eE-F).\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec22\" class=\"Section2\"\u003e \u003ch2\u003eIdentification of core modules of DRG through WGCNA\u003c/h2\u003e \u003cp\u003eTo construct the scale-free network, the most appropriate soft threshold of β was 14 (Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003eA). Cluster diagram was constructed according to the \u0026ldquo;cutree\u0026rdquo; dynamic and module eigengenes functions, and then a total of 20 module signature genes with different colors was obtained (Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003eB-C). The association between each module and RB was presented in a heatmap of module-trait relationships, suggesting that the pink module had largest correlation with RB and was negatively associated with RB (Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003eC). The correlation between pink module genes and RB genes was cor\u0026thinsp;=\u0026thinsp;0.66 (Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003eD). Additionally, the gene module-related to disulfidptosis were analyzed by WGCNA algorithm. The scale-free network was constructed when the soft threshold was set as 16 and scale-free R2 was set as 0.9, and 16 significant modules with different colors were obtained (Fig.\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003eA-B). Following the module-clinical traits (cluster A and cluster B) relationship explanation, cluster A and cluster B were highly associated with the red module (Fig.\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003eC). The correlation between pink module genes and DRGs was cor\u0026thinsp;=\u0026thinsp;0.94 (Fig.\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003eD).\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cdiv id=\"Sec23\" class=\"Section3\"\u003e \u003ch2\u003eSelection of DRGs using Machine learning models\u003c/h2\u003e \u003cp\u003eThe overlapping expression genes were screened from DEG (scRNA), modGene (Disease WGCNA) and modGene (Cluster WGCNA), 27 overlapping expression genes were obtained (Fig.\u0026nbsp;\u003cspan refid=\"Fig7\" class=\"InternalRef\"\u003e7\u003c/span\u003eA). Then, four validated machine-learning models (GLM, RF, SVM and XGB) were constructed to identify the DEGs with high diagnostic values. GLM machine learning model exhibited a relatively lower residual (Fig.\u0026nbsp;\u003cspan refid=\"Fig7\" class=\"InternalRef\"\u003e7\u003c/span\u003eB-C). As shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig7\" class=\"InternalRef\"\u003e7\u003c/span\u003eD, the feature importance diagram of each machine learning model was depended on the root mean square error. Moreover, we calculated ROC curves of each machine learning algorithm, all machine learning model displayed 1.0 of AUC (Fig.\u0026nbsp;\u003cspan refid=\"Fig7\" class=\"InternalRef\"\u003e7\u003c/span\u003eE). Two overlapping genes were obtained from the four memory models, including CDT1 and NUSAP1 (Fig.\u0026nbsp;\u003cspan refid=\"Fig7\" class=\"InternalRef\"\u003e7\u003c/span\u003eF). Based on the training data, both CDT1 and NUSAP1 exhibited an AUC value of 1 (Fig.\u0026nbsp;\u003cspan refid=\"Fig7\" class=\"InternalRef\"\u003e7\u003c/span\u003eG). When validated with the GSE111168 dataset, CDT1 and NUSAP1 had AUC values of 0.860 and 0.763, respectively (Fig.\u0026nbsp;\u003cspan refid=\"Fig7\" class=\"InternalRef\"\u003e7\u003c/span\u003eH). Hence, it was equally well in distinguishing between normal and RB when using our diagnostic model.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003e \u003cb\u003eCDT1 silencing inhibited malignant phenotypes of retinoblastoma cells.\u003c/b\u003e \u003c/p\u003e \u003cp\u003eTo verify the role of CDT1 in RB, we examined CDT1 expression in WERI-Rb1 cells with low invasiveness and Y79 cells with high invasiveness. Compared with ARPE-19 cells, the mRNA expression of CDT1 was notably increased in WERI-Rb1 and Y79 cells (Fig.\u0026nbsp;\u003cspan refid=\"Fig8\" class=\"InternalRef\"\u003e8\u003c/span\u003eA). Then, CDT1 was silenced in both WERI-Rb1 and Y79 cells, and cell proliferation ability was examined by performing colony formation assay. CDT1 knockdown potently reduced proliferation ability of WERI-Rb1 and Y79 cells (Fig.\u0026nbsp;\u003cspan refid=\"Fig8\" class=\"InternalRef\"\u003e8\u003c/span\u003eB). This reduction was accompanied by inhibition in migration and invasion, when WERI-Rb1 and Y79 cells were transduced with si-CDT1, as confirmed by transwell migration and invasion assays (Fig.\u0026nbsp;\u003cspan refid=\"Fig8\" class=\"InternalRef\"\u003e8\u003c/span\u003eC-E). These \u003cem\u003ein vitro\u003c/em\u003e experiments demonstrated that CDT1 silencing inhibited proliferation, migration and invasion of retinoblastoma cells.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e \u003c/div\u003e"},{"header":"Discussion","content":"\u003cp\u003eUnder the effect of tumorigenic factors, tumors have deteriorated and proliferated from normal cells to cancer cells, and have accumulated mutations and gradually differentiated into different genetic lineages and subgroups during the evolution process, thereby forming heterogeneity within tumors (\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e). Tumor intratumoral heterogeneity is closely related to tumor invasion and metastasis, and also affects the clinical diagnosis and treatment of patients. RB is also a highly heterogeneous tumor (\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e). ScRNA seq technology detects the heterogeneity of tumor cells, identifies rare cells, divides cell subsets, tracks cell lineage, locates mutation genes, and discovers new tumor biomarkers through a single cell (\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e, \u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e). In this work, we analyzed 10199 cells from RB samples and normal samples by scRNA seq technology, and obtained 7 annotated cell clusters, including cone cells, epithelial cells, endothelial cells, astrocytes, microglial cells, Muller glial cells, neuron. Among them, the fraction of cone cells was significantly increased in RB samples. Cobrinik et al. have pointed out that RB originates from cone precursor cells, and RB is formed in response to single gene mutation of RB1 (\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e). RB1 gene encodes a tumor suppressor protein Rb, which prevents excessive cell growth by inhibiting the process of cell cycle and preventing cells from malignant proliferation and then transforming into cancer cells (\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e, \u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e). Thus, malignant cone cells may appear in the increased cone cells to promote RB progression.\u003c/p\u003e \u003cp\u003eFurther analysis revealed that cone cells were divided into normal cells and malignant cone cells. Compared with normal cone cells, malignant cone cells exhibited high disulfidptosis. SLC7A11, a DRG, was notably associated with malignant cone cells. Disulfidptosis is a new type of cell death that is independent of existing programmed cell death such as apoptosis, ferroptosis, and necrotic apoptosis and cuproptosis (\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e). It is a rapid death mode caused by disulfide stress-mediated excessive accumulation of cystine in cells, which dependents on high expression of SLC7A11. Several studies have shown that SLC7A11 is highly expressed in breast cancer, lung cancer and bladder cancer, and affects the metabolism, occurrence and development of tumors (\u003cspan additionalcitationids=\"CR22\" citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e). In this work, we found that SLC7A11 expression was associated with malignant cone cells. SLC7A11 may regulate disulfidptosis to drive malignant progression of cone cells.\u003c/p\u003e \u003cp\u003eApplying machine learning methods, we found 2 genes were closely associated with RB, including CDT1 and NUSAP1. These genes may participate in the progression of RB, especially CDT1. CDT1 is an important regulator of DNA replication licensing, which is essential for initiation of DNA replication (\u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e). Previous studies have confirmed that CDT1 is a potent oncogene in several cancers. For instance, Cai et al. have found that CDT1 accelerates malignant progression of hepatocellular carcinoma, which may be associated with cell cycle, DNA repair, DNA replication and minichromosome maintenance (\u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e). High expression of CDT1 in colorectal cancer is correlated significantly with increased origin licensing, activation of the DNA damage response, and microsatellite instability (\u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e). CDT1 is reported to related to poor prognosis of triple-negative breast cancer patients, CDT1 promotes cellular processes of triple-negative breast cancer (\u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e). In the present work, we first confirmed the carcinogenic effect of CDT1 on RB. Compared with ARPE-19 cells, CDT1 was up-regulated in both Y79 cells with high invasiveness and WERI-Rb1 cells with low invasiveness. CDT1 knockdown suppressed the proliferation, migration and invasion of RB cells.\u003c/p\u003e \u003cp\u003eIn conclusion, this work revealed that tumors contained cone cell states with distinct transcriptional programs, and provided a crucial RB-related gene CDT1 for tumor progression. Thus, this article is of great significance in formulating a molecular targeted therapy scheme to prevent the development of RB.\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e \u003ch2\u003eEthics approval and consent to participate\u003c/h2\u003e \u003cp\u003eN/A\u003c/p\u003e \u003c/p\u003e \u003cp\u003e \u003cstrong\u003eConflict of Interest\u003c/strong\u003e \u003cp\u003eThe authors have no conflicts of interest to declare.\u003c/p\u003e \u003c/p\u003e\u003cp\u003e \u003ch2\u003eDeclaration of competing interest\u003c/h2\u003e \u003cp\u003eThe authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.\u003c/p\u003e \u003c/p\u003e\u003cp\u003e \u003ch2\u003eConsent for publication\u003c/h2\u003e \u003cp\u003eN/A.\u003c/p\u003e \u003c/p\u003e\u003ch2\u003eFunding\u003c/h2\u003e \u003cp\u003eNot applicable.\u003c/p\u003e\u003ch2\u003eAuthor Contribution\u003c/h2\u003e\u003cp\u003eConception: Y Y.Interpretation or analysis of data: Y Y, YZ Z.Preparation of the manuscript: Y Y, YZ Z, WW X.Revision for important intellectual content: Y Y, YZ Z.Supervision: XL Y.\u003c/p\u003e\u003ch2\u003eAcknowledgments\u003c/h2\u003e \u003cp\u003eNot applicable.\u003c/p\u003e\u003ch2\u003eAvailability of data and materials\u003c/h2\u003e \u003cp\u003eAll data related to the article can be obtained from the corresponding author upon reasonable request.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\u003cli\u003e\u003cspan\u003eRao R, Honavar SG, Retinoblastoma. Indian J Pediatr. 2017;84(12):937\u0026ndash;44.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eChung SE, Sa HS, Koo HH, Yoo KH, Sung KW, Ham DI. Clinical manifestations and treatment of retinoblastoma in Korea. Br J Ophthalmol. 2008;92(9):1180\u0026ndash;4.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eOrtiz MV, Dunkel IJ, Retinoblastoma. J Child Neurol. 2016;31(2):227\u0026ndash;36.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eTing SC, Kiefer T, Ehlert K, Goericke SL, Hinze R, Ketteler P, Bechrakis NE, Schildhaus HU. Bone metastasis of retinoblastoma five years after primary treatment. Am J Ophthalmol case Rep. 2020;19:100834.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eAncona-Lezama D, Dalvin LA, Shields CL. Modern treatment of retinoblastoma: A 2020 review. Indian J Ophthalmol. 2020;68(11):2356\u0026ndash;65.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDimaras H, Dimba EA, Gallie BL. Challenging the global retinoblastoma survival disparity through a collaborative research effort. Br J Ophthalmol. 2010;94(11):1415\u0026ndash;6.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZheng P, Zhou C, Ding Y, Duan S. Disulfidptosis: a new target for metabolic cancer therapy. J experimental Clin cancer research: CR. 2023;42(1):103.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLiu X, Nie L, Zhang Y, Yan Y, Wang C, Colic M, Olszewski K, Horbath A, Chen X, Lei G, Mao C, Wu S, Zhuang L, Poyurovsky MV, James You M, Hart T, Billadeau DD, Chen J, Gan B. Actin cytoskeleton vulnerability to disulfide stress mediates disulfidptosis. Nat Cell Biol. 2023;25(3):404\u0026ndash;14.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eJi PY, Li ZY, Wang H, Dong JT, Li XJ, Yi HL. Arsenic and sulfur dioxide co-exposure induce renal injury via activation of the NF-κB and caspase signaling pathway. Chemosphere. 2019;224:280\u0026ndash;8.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eChen H, Yang W, Li Y, Ma L, Ji Z. Leveraging a disulfidptosis-based signature to improve the survival and drug sensitivity of bladder cancer patients. Front Immunol. 2023;14:1198878.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWang T, Guo K, Zhang D, Wang H, Yin J, Cui H, Wu W. Disulfidptosis classification of hepatocellular carcinoma reveals correlation with clinical prognosis and immune profile. Int Immunopharmacol. 2023;120:110368.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKoppula P, Zhuang L, Gan B. Cystine transporter SLC7A11/xCT in cancer: ferroptosis, nutrient dependency, and cancer therapy. Protein Cell. 2021;12(8):599\u0026ndash;620.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eTabassum DP, Polyak K. Tumorigenesis: it takes a village. Nat Rev Cancer. 2015;15(8):473\u0026ndash;83.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eStenfelt S, Blixt MKE, All-Ericsson C, Hallb\u0026ouml;\u0026ouml;k F, Boije H. Heterogeneity in retinoblastoma: a tale of molecules and models. 2017;6(1):42.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZhao Y, Feng Y, Liu M, Chen L, Meng Q, Tang X, Wang S, Liu L, Li L, Shen W, Zhang H. Single-cell RNA sequencing analysis reveals alginate oligosaccharides preventing chemotherapy-induced mucositis. Mucosal Immunol. 2020;13(3):437\u0026ndash;48.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eRoss EM, Markowetz F. OncoNEM: inferring tumor evolution from single-cell sequencing data. Genome Biol. 2016;17:69.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eXu XL, Singh HP, Wang L, Qi DL, Poulos BK, Abramson DH, Jhanwar SC, Cobrinik D. Rb suppresses human cone-precursor-derived retinoblastoma tumours. Nature. 2014;514(7522):385\u0026ndash;8.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMcEvoy JD, Dyer MA. Genetic and Epigenetic Discoveries in Human Retinoblastoma. Crit Rev Oncog. 2015;20(3\u0026ndash;4):217\u0026ndash;25.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGiacinti C, Giordano A. RB and cell cycle progression. Oncogene. 2006;25(38):5220\u0026ndash;7.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLiu X, Zhuang L, Gan B. Disulfidptosis: disulfide stress-induced cell death. Trends Cell Biol. 2023.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eYang J, Zhou Y, Xie S, Wang J, Li Z, Chen L, Mao M, Chen C, Huang A, Chen Y, Zhang X, Khan NUH, Wang L, Zhou J. Metformin induces Ferroptosis by inhibiting UFMylation of SLC7A11 in breast cancer. J experimental Clin cancer research: CR. 2021;40(1):206.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZhang W, Sun Y, Bai L, Zhi L, Yang Y, Zhao Q, Chen C, Qi Y, Gao W, He W, Wang L, Chen D, Fan S, Chen H, Piao HL, Qiao Q, Xu Z, Zhang J, Zhao J, Zhang S, Yin Y, Peng C, Li X, Liu Q, Liu H, Wang Y. RBMS1 regulates lung cancer ferroptosis through translational control of SLC7A11. The Journal of clinical investigation. 2021;131(22).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eShen L, Zhang J, Zheng Z, Yang F, Liu S, Wu Y, Chen Y, Xu T, Mao S, Yan Y, Li W, Zhang W, Yao X. PHGDH Inhibits Ferroptosis and Promotes Malignant Progression by Upregulating SLC7A11 in Bladder Cancer. Int J Biol Sci. 2022;18(14):5459\u0026ndash;74.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eRatnayeke N, Baris Y, Chung M, Yeeles JTP, Meyer T. CDT1 inhibits CMG helicase in early S phase to separate origin licensing from DNA synthesis. Mol Cell. 2023;83(1):26\u0026ndash;e4213.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eCai C, Zhang Y, Hu X, Hu W, Yang S, Qiu H, Chu T. CDT1 Is a Novel Prognostic and Predictive Biomarkers for Hepatocellular Carcinoma. Front Oncol. 2021;11:721644.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePetropoulos M, Champeris Tsaniras S, Nikou S, Maxouri S, Dionellis VS, Kalogeropoulou A, Karamichali A, Ioannidis K, Danalatos IR, Obst M, Naumann R, Delinasios GJ, Gorgoulis VG, Roukos V, Anastassiadis K, Halazonetis TD, Bravou V, Lygerou Z, Taraviras S. Cdt1 overexpression drives colorectal carcinogenesis through origin overlicensing and DNA damage. J Pathol. 2023;259(1):10\u0026ndash;20.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMaimaiti Y, Zhang N, Zhang Y, Zhou J, Song H, Wang S. CircFAM64A enhances cellular processes in triple-negative breast cancer by targeting the miR-149-5p/CDT1 axis. Environ Toxicol. 2022;37(5):1081\u0026ndash;92.\u003c/span\u003e\u003c/li\u003e\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":true,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true},"keywords":"Retinoblastoma, disulfidptosis, single-cell RNA sequencing, CDT1, oncogene","lastPublishedDoi":"10.21203/rs.3.rs-4883827/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-4883827/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003ch2\u003eBACKGROUND\u003c/h2\u003e \u003cp\u003eRetinoblastoma (RB) is a heterogeneous primary intraocular malignant tumor.\u003c/p\u003e\u003ch2\u003eOBJECTIVE\u003c/h2\u003e \u003cp\u003eThis work attempted to reveal the significant gene-related to disulfidptosis in RB.\u003c/p\u003e\u003ch2\u003eMETHODS\u003c/h2\u003e \u003cp\u003eThe scRNA-seq data from RB samples and normal samples obtained from GSE159977 were analyzed to distinguish cone cells from malignant cone cells using inferCNV. Subsequently, AUCcell was used to assess the disulfidptosis levels in cone cells. Disulfidptosis-related genes in RB were analyzed through weighted gene co-expression network analysis and machine learning methods. Cell proliferation, migration, and invasion were examined through colony formation and Transwell experiments.\u003c/p\u003e\u003ch2\u003eRESULTS\u003c/h2\u003e \u003cp\u003eWe obtained 7 annotated cell clusters. Among them, there was a significant increase in the proportion of cone cells in RB samples. Malignant cone cells exhibited a higher disulfidptosis score. Cell trajectory analysis indicated that the disulfidptosis process intensified as cone cells transition from normal cells to malignant cells. Through machine learning, disulfidptosis-related genes were screened, ultimately identifying CDT1 as a key gene. CDT1 was upregulated in WERI-Rb1 and Y79 cells. Silencing CDT1 significantly suppressed the proliferation, migration, and invasion of RB cells.\u003c/p\u003e\u003ch2\u003eCONCLUSIONS\u003c/h2\u003e \u003cp\u003eThis work revealed that tumors contained cone cell states with distinct transcriptional programs, and provided a crucial RB-related gene CDT1 for tumor progression. Thus, this article is of great significance in formulating a molecular targeted therapy scheme to prevent the development of RB.\u003c/p\u003e","manuscriptTitle":"Single-Cell Sequencing Analysis and Multiple Machine Learning Methods identified CDT1 as an oncogene in retinoblastoma","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2024-09-06 23:33:29","doi":"10.21203/rs.3.rs-4883827/v1","editorialEvents":[{"type":"communityComments","content":0}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"7cda3630-e0bb-4945-8e3f-8a89e6e9dd27","owner":[],"postedDate":"September 6th, 2024","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[],"tags":[],"updatedAt":"2024-09-06T23:33:32+00:00","versionOfRecord":[],"versionCreatedAt":"2024-09-06 23:33:29","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-4883827","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-4883827","identity":"rs-4883827","version":["v1"]},"buildId":"qtupq5eGEP_6zYnWcrvyt","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}