Single-Cell RNA-seq Reveals a Subpopulation of Prostate Cancer Cells with Enhanced Cell-Cycle-Related Transcription and Attenuated Androgen Response

Abstract

Increasing evidence suggests the presence of minor cell subpopulations in prostate cancer that are androgen independent and poised for selection as dominant clones after androgen deprivation therapy. In this study, we investigated this phenomenon by stratifying cell subpopulations based on transcriptome profiling of 144 single LNCaP prostate cancer cells treated or untreated with androgen after cell-cycle synchronization. Model-based clustering of 397 differentially expressed genes identified eight potential subpopulations of LNCaP cells, revealing a previously unappreciable level of cellular heterogeneity to androgen stimulation. One subpopulation displayed stem-like features with a slower cell doubling rate, increased sphere formation capability, and resistance to G₂-M arrest induced by a mitosis inhibitor. Advanced growth of this subpopulation was associated with enhanced expression of 10 cell-cycle-related genes (CCNB2, DLGAP5, CENPF, CENPE, MKI67, PTTG1, CDC20, PLK1, HMMR, and CCNB1) and decreased dependence upon androgen receptor signaling. In silico analysis of RNA-seq data from The Cancer Genome Atlas further demonstrated that concordant upregulation of these genes was linked to recurrent prostate cancers. Analysis of receiver operating characteristic curves implicates aberrant expression of these genes and could be useful for early identification of tumors that subsequently develop biochemical recurrence. Moreover, this single-cell approach provides a better understanding of how prostate cancer cells respond heterogeneously to androgen deprivation therapies and reveals characteristics of subpopulations resistant to this treatment.Significance: Illustrating the challenge in treating cancers with targeted drugs, which by selecting for drug resistance can drive metastatic progression, this study characterized the plasticity and heterogeneity of prostate cancer cells with regard to androgen dependence, defining the character or minor subpopulations of androgen-independent cells that are poised for clonal selection after androgen-deprivation therapy. Cancer Res; 78(4); 853-64. ©2017 AACR.

Figure 1

Single-cell analysis of 144 androgen-treated, -untreated, and control LNCaP cells reveals 397 differentially expressed genes. A, Schematic for isolating 48 single cells per treatment group in preparation for single cell transcriptome amplification. B, Number of aligned reads (millions) per single cells and bulk sample per treatment group. C, Number of significantly differentially expressed genes (adjusted P value < 0.05). D, Hierarchical clustering heatmap of gene expression (Z score of log₂ normalized RPKM) for each of the 397 genes (horizontal axis) and each of the single cells (vertical axis) within each treatment group. Of these, 125+18 genes were identified as cell cycle-dependent while 254 genes were categorized as androgen-responsive. Note: 18 genes assigned in the cell cycle-dependent group depicted here were also androgen-responsive.

Figure 2

Subpopulations of LNCaP cells exhibit distinct transcription responses to androgen stimulation. A, Principal component analysis (PCA) plot representing each cell as a dot localized based on its 397-gene expression profile. B, Alternative view of the first four principal components along with results of the mixed model-based clustering algorithm showing the likely subpopulations of cells. C, Overlapping view of the original principal component analysis plot with each of the eight subpopulations identified by specific symbols. D, Percentage of the fraction of cells from each treatment group represented in each subgroup.

Figure 3

Distinct gene signatures characterize the individual androgen-treated and -untreated LNCaP subpopulations. A, Hierarchical clustering heatmap showing the correlation (Pearson r value) of the gene expression values for each of the 397 significant genes with the average/central location for each subpopulation of cells on the PCA plot in Fig. 2A. This figure is the basis for the “positively correlated” gene expression profiles created for each subpopulation. B, Bar charts indicating the level of significance that each gene ontology term from DAVID is associated with each subpopulation based on the “positively correlated” gene expression profiles. Statistically significant gene ontology term associations are indicated by bars ≥1.3.

Figure 4

Increased expression of the subpopulation E gene profile predicts biochemical recurrent prostate cancer. A, Hierarchical clustering heatmap of prostate cancer tissue disease status and the subpopulation E gene expression profile from TCGA (Z score of log₂-normalized RPKM). B, Receiver operating characteristic (ROC) curves for genes related to subpopulation E cells. The blue line indicates the false and true positive rates for predicting biochemical recurrence at each point throughout the range of gene expression for each gene. C, Kaplan-Meier plot predicting disease-free status of patients based on ≥2-fold gene expression changes of any of the genes in the 10-gene set.

Figure 5

The HMMR^hi subpopulation of androgen-treated LNCaP cells is less responsive to androgen stimulation. A, Bar chart of RPKM values of CD81 and HMMR for each cell within each original treatment group. Bars representing gene expression for subcluster E cells are colored red. B, Flow cytometry plot of androgen-treated LNCaP cells stained with anti-HMMR and anti-CD81 antibodies conjugated to APC and PE, respectively. The HMMR^hi sorted subpopulation is LNCaP 8.1. C, Gene expression of subpopulation E-related genes in LNCaP and 8.1 cells. D, Cell viability assay of LNCaP and 8.1 cells treated with enzalutamide. Results are shown as means ± SD from triplicates and analyzed for statistical significance using ANOVA then Sidak’s multiple comparison post-hoc test to compare within cell lines. *, p<0.05; **, p<0.01; ***, p<0.001. No asterisk indicates no significant difference.

Figure 6

The HMMR^hi subpopulation of androgen-treated LNCaP cells has stem cell-like features. A, Live-cell imaging quantification showing the normalized proliferation fold changes of LNCaP and 8.1 (n=4 images per every biological triplicate). B, Live-cell imaging quantification of LNCaP and 8.1 treated with and without the PLK1 inhibitor, BI2536 (2 nM)(n=4 images per every biological triplicate). C, Cell cycle analysis of LNCaP and 8.1 in the presence and absence of BI2536 (2 nM). D, Colony formation photos of representative regions within the culture dish. Also, colony formation area analysis of paraformaldehyde fixed and crystal violet stained images of another colony formation assay of the LNCaP subpopulations performed with technical triplicates (500 cells per well in a 6-well dish cultured for 10 days before fixing). E, Single hand-picked LNCaP and 8.1 cells analyzed using a biomark gene panel and presented as a Z-score normalized heatmap. F, Violin plots quantifying the epithelial characteristics gene expression from panel E. G, Schematic illustration for the development of LNCaP subpopulations to androgen-dependence and -independence. The model depicts the majority of pre-existing cell subpopulations that are sensitive to androgen-driven proliferation. The model also shows cell-cycle progression and expansion of stem-like cell subpopulations (i.e., less sensitive to androgen) under an androgen-deprivation condition.