Skip to main page content
U.S. flag

An official website of the United States government

Dot gov

The .gov means it’s official.
Federal government websites often end in .gov or .mil. Before sharing sensitive information, make sure you’re on a federal government site.

Https

The site is secure.
The https:// ensures that you are connecting to the official website and that any information you provide is encrypted and transmitted securely.

Access keys NCBI Homepage MyNCBI Homepage Main Content Main Navigation
. 2023 Nov 7;21(1):789.
doi: 10.1186/s12967-023-04683-6.

Integrative multi-omics analysis unveils stemness-associated molecular subtypes in prostate cancer and pan-cancer: prognostic and therapeutic significance

Kun Zheng  1 , Youlong Hai  1 , Yue Xi  2 , Yukun Zhang  3 , Zheqi Liu  4 , Wantao Chen  5 , Xiaoyong Hu  6 , Xin Zou  7   8 , Jie Hao  9
Affiliations

Integrative multi-omics analysis unveils stemness-associated molecular subtypes in prostate cancer and pan-cancer: prognostic and therapeutic significance

Kun Zheng et al. J Transl Med. .

Abstract

Background: Prostate cancer (PCA) is the fifth leading cause of cancer-related deaths worldwide, with limited treatment options in the advanced stages. The immunosuppressive tumor microenvironment (TME) of PCA results in lower sensitivity to immunotherapy. Although molecular subtyping is expected to offer important clues for precision treatment of PCA, there is currently a shortage of dependable and effective molecular typing methods available for clinical practice. Therefore, we aim to propose a novel stemness-based classification approach to guide personalized clinical treatments, including immunotherapy.

Methods: An integrative multi-omics analysis of PCA was performed to evaluate stemness-level heterogeneities. Unsupervised hierarchical clustering was used to classify PCAs based on stemness signature genes. To make stemness-based patient classification more clinically applicable, a stemness subtype predictor was jointly developed by using four PCA datasets and 76 machine learning algorithms.

Results: We identified stemness signatures of PCA comprising 18 signaling pathways, by which we classified PCA samples into three stemness subtypes via unsupervised hierarchical clustering: low stemness (LS), medium stemness (MS), and high stemness (HS) subtypes. HS patients are sensitive to androgen deprivation therapy, taxanes, and immunotherapy and have the highest stemness, malignancy, tumor mutation load (TMB) levels, worst prognosis, and immunosuppression. LS patients are sensitive to platinum-based chemotherapy but resistant to immunotherapy and have the lowest stemness, malignancy, and TMB levels, best prognosis, and the highest immune infiltration. MS patients represent an intermediate status of stemness, malignancy, and TMB levels with a moderate prognosis. We further demonstrated that these three stemness subtypes are conserved across pan-tumor. Additionally, the 9-gene stemness subtype predictor we developed has a comparable capability to 18 signaling pathways to make tumor diagnosis and to predict tumor recurrence, metastasis, progression, prognosis, and efficacy of different treatments.

Conclusions: The three stemness subtypes we identified have the potential to be a powerful tool for clinical tumor molecular classification in PCA and pan-cancer, and to guide the selection of immunotherapy or other sensitive treatments for tumor patients.

Keywords: Immunotherapy; Machine learning; Pan‑cancer; Prostate cancer; RNA sequencing; Stemness subtype.

PubMed Disclaimer

Conflict of interest statement

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as potential conflicts of interest.

Figures

Fig. 1
Fig. 1
Correlation of stemness levels with clinical, pathological, and molecular features in patients with prostate cancer (PCA). a t-distributed stochastic neighbor embedding (t-SNE) plot of malignant and benign epithelial cells from GSE193337 dataset (medium), along with their corresponding stemness scores (cytoTRACE, left), and the comparison of these scores between two groups (right). b t-SNE plot of high and low grade PCA cells from GSE141445 dataset (medium), along with their corresponding cytoTRACE scores (left), and the comparison of these scores between two groups (right). c Comparison of stemness scores (mRNAsi, mDNAsi) between benign and malignant prostate samples, as well as between high (Gleason score [GS] > 7) and low (GS < 7) grade PCA samples from TCGA-PRAD. d Kaplan–Meier (K–M) analysis demonstrated a correlation between the mRNAsi scores and the prognosis of PCA patients from TCGA-PRAD. OS overall survival, PFI progression-free interval, DFI disease-free interval, DSS disease-specific survival. Dashed line: median survival time. Color range: 95% confidence interval (CI)
Fig. 2
Fig. 2
Identification of three PCA stemness subtypes based on stemness signatures. a CircosPlot shows 288 stemness marker genes obtained from scRNA-seq data that are significantly positively correlated with cytoTRACE and significantly upregulated in both malignant and high-grade PCA cells (left), and 220 stemness marker genes derived from bulk RNA-seq data that are significantly positively correlated with mRNAsi and significantly upregulated in PCA samples (right). b Unsupervised hierarchical clustering based on the activity scores of the 18 stemness signatures classified PCA patients from TCGA-PRAD into three subtypes: low stemness (LS), medium stemness (MS), and high stemness (HS) subtypes. c 3D projection of the principal components obtained through PCA analysis. d Levels and trends of stemness score (mRNAsi) within three stemness subtypes. e Three stemness subtypes of TCGA-PRAD exhibits distinct PFI outcomes. f, g Univariate (f) and multivariate (g) Cox regression analysis of the three stemness subtypes, and clinical and molecular characteristics. *p < 0.05, **p < 0.01, ***P < 0.001, ****p < 0.0001
Fig. 3
Fig. 3
Comparison of clinicopathological and molecular features among three PCA stemness subtypes. a Sankey diagram showing sample flow for stemness subtype, sample type, and GS. b Comparison of sample type, grade, pT and pN among the three stemness subtypes. c Comparison of patient weight, age at diagnosis, GS, PSA, and AR among three stemness subtypes. d Comparison of TMB, fraction genome altered, amplifications, deletions, and exon imbalance scores among three stemness subtypes. e–g Oncoplots showing the top 10 mutated genes in LS (e), MS (f) and HS (g). h Stacked histograms showing comparisons of somatic copy number alterations (SCNAs), DNA methylation clustering, TP53 mutations and CNAs, and PTEN mutations and CNAs among the three stemness subtypes. *p < 0.05, **p < 0.01, ***P < 0.001, ****p < 0.0001
Fig. 4
Fig. 4
Comparison of drug sensitivities and TIME patterns among three PCA stem subtypes. a–h Comparisons of sensitivities of three stemness subtypes to clinically preferred and recommended drugs. i Differences in TME scores among the three stemness subtypes. j Hypergeometric tests reveal an association between stemness subtypes and TIME subtypes. k Boxplots showing comparisons of immunocyte abundance among the three stemness subtypes. l Correlation heatmap showing the correlation between stemness indices and expression levels of immune checkpoint molecules. m Stacked histogram showing differences in responsiveness of the three PCA stemness subtypes to immune-checkpoint blockade (ICB) therapy. Evaluated by TIDE algorithm. n Submap analysis reflects the sensitivity of the three PCA stemness subtypes to an-PD-1, anti-PD-L1 and anti-CTLA-4 treatments. *p < 0.05, **p < 0.01, ***P < 0.001, ****p < 0.0001
Fig. 5
Fig. 5
Construction and validation of stemness subtype predictor. a Correlation analysis between module eigengenes and stemness subtypes of TCGA-PRAD (left). The highest correlation between GS and MM in the pink module. Dots within the pink rectangle were defined as HS hub genes (right). b Protein–protein interaction (PPI) network of the 40 genes of Core.Sig, and these proteins were divided into three clusters based on the MCL inflation parameter. c Venn diagram identified the nine most critical stemness subtype marker genes that were intersected by 4 datasets and 76 machine learning algorithms (MLs). d Immunohistochemistry (IHC) staining shows the protein levels of four critical stemness subtype marker genes (SKA3, DLGAP5, NCAPG, HMMR) in benign, low-grade, and high grade PCA samples. Representative images are shown. e Histogram shows the performances of the 9-gene predictor in distinguishing benign from malignant tumors and predicting androgen deprivation therapy (ADT) response, metastasis, biochemical recurrence and progression via 100 MLs. The top 10 MLs with the best performance are exhibited. f K–M analysis shows the effect of 9-gene-based stemness-related risk score (SRS) on PFI of PCA patients from TCGA-PRAD. g Multivariate COX analysis showed that SRS was the most important independent risk factor for PCA patients
Fig. 6
Fig. 6
Interactions between stemness subtypes and ICB treatment in pan-tumors. a Unsupervised hierarchical clustering based on the stemness activity scores of the 18 stemness signatures clustered the baseline pan-tumor patients treated with ICB into three subtypes. b Hypergeometric test collaborates an association between stemness subtypes of ICB pan-tumors and responsiveness of ICB therapy. c Stacked histogram showing differences in responsiveness of the three pan-tumor stemness subtypes to ICB. d Univariate logistic regression shows the effect of the three stemness subtypes on ICB efficacy. e, f Sankey diagram showing sample flow for pre-treatment and on/post-treatment of ICB. Separate presentation for responders (e) and non-responders (f). g t-SNE plot of pre-treatment and post-treatment tumor cells of head and neck squamous cell carcinoma (HNSCC, medium), along with their corresponding cytoTRACE scores (left), and the comparison of these scores between two groups (right)
See this image and copyright information in PMC

Similar articles

See all similar articles

Cited by

References

    1. Miller K, Ortiz A, Pinheiro P, Bandi P, Minihan A, Fuchs H, Martinez Tyson D, Tortolero-Luna G, Fedewa S, Jemal A, et al. Cancer statistics for the US Hispanic/Latino population, 2021. Cancer J Clin. 2021;71(6):466–487. doi: 10.3322/caac.21695. - DOI - PubMed
    1. Siegel R, Miller K, Fuchs H, Jemal A. Cancer statistics, 2022. Cancer J Clin. 2022;72(1):7–33. doi: 10.3322/caac.21708. - DOI - PubMed
    1. Bray F, Ferlay J, Soerjomataram I, Siegel R, Torre L, Jemal A. Global cancer statistics 2018: GLOBOCAN estimates of incidence and mortality worldwide for 36 cancers in 185 countries. Cancer J Clin. 2018;68(6):394–424. doi: 10.3322/caac.21492. - DOI - PubMed
    1. Chen W, Zheng R, Baade P, Zhang S, Zeng H, Bray F, Jemal A, Yu X, He J. Cancer statistics in China, 2015. Cancer J Clin. 2016;66(2):115–132. doi: 10.3322/caac.21338. - DOI - PubMed
    1. Marhold M, Kramer G, Krainer M, Le Magnen C. The prostate cancer landscape in Europe: current challenges, future opportunities. Cancer Lett. 2022;526:304–310. doi: 10.1016/j.canlet.2021.11.033. - DOI - PubMed

Publication types

Substances