Transcriptional profiles of different states of cancer stem cells in triple-negative breast cancer

To systematically characterize the transcriptional profiles of BCSCs, we isolated four cell groups from two TNBC PDXs, and performed whole-transcriptome sequencing to identify differentially expressed genes (DEGs) between four groups (Fig. 1a): (1) group A (ALDH⁺CD24⁻CD44⁺, highly purified BCSCs); (2) group B (ALDH⁺non-CD24⁻CD44⁺, enriched epithelial-like BCSCs); (3) group C (ALDH⁻CD24⁻CD44⁺, enriched mesenchymal-like BCSCs); and (4) group D (ALDH⁻non-CD24⁻CD44⁺, differentiated tumor cells). The tumorigenicity of each cell population was analyzed in vivo, and the result demonstrated that groups A and B had significantly higher tumor-initiating capacity and CSC frequency than groups C and D (Fig. 1b), with the highest tumorigenicity for group A. Moreover, the size of tumors in group A was significantly larger than that in group B. The transcriptomic data is shown in Additional file 1: Table S1. The expression of BCSC biomarkers ALDH and CD24/CD44 were as expected [7]: CD24: A < B, C < D; CD44: A > B, C > D; ALDH: A/B > C/D (Fig. 1c). We systematically performed pair-comparisons between three subsets of BCSCs and differentiated tumor cells (Additional file 1: Figure S1) with fold change set at 1.2 based on the standard of our previous study [7]. The DEGs in A/D, B/D and C/D pair-comparisons were 3223, 3387 and 3065, respectively (Additional file 1: Figure S1a). For all states of BCSCs in common, there were 391 DEGs in the intersection set (Additional file 1: Figure S1b). The Gene Ontology (GO) analysis based on biological process indicated that the 391 DEGs involved in cellular response to hypoxia, cell adhesion, extracellular matrix organization, cell cycle, etc. (Additional file 2: Table S2). To characterize the exclusively transcriptional features of each state of BCSCs, we overlapped the DEGs of three pair-comparisons (Additional file 1: Figure S1b), and found that each state has its own unique DEGs (Additional file 1: Figure S1b), of which the altered GO terms were different identified by DAVID 6.8 and Gene Set Enrichment Analysis (GSEA) (Additional file 1: Figure S2, Additional file 3: Table S3), suggesting that three populations of the ALDH⁺CD24⁻CD44⁺, the ALDH⁺non-CD24⁻CD44⁺ and the ALDH⁻CD24⁻CD44⁺ were different states of BCSCs. In addition, we also found that the epithelial markers, CDH3, CLDN3, CLDN4, CLDN7 and MKI67, were highly expressed in the ALDH⁺non-CD24⁻CD44⁻ BCSCs, while the mesenchymal markers, CDH2, FOXC2, MMP2, SNAI2 and TWIST1, were highly expressed in the ALDH⁻CD24⁻CD44⁺ BCSCs (Additional file 1: Figure S3).

To identify the DEGs in ALDH⁺CD24⁻CD44⁺ BCSCs, we compared group A with the other three groups with fold change set at 1.2 in analyzed PDXs (Fig. 2a). The numbers of intersected A/X (X stands for groups B, C or D) DEGs overlapped in analyzed PDXs were 3505 and 2360, respectively (Fig. 2a). We performed principal component analysis to further distinguish group A from the other three groups in each PDX, trimming DEGs to 3105 and 1851 for PDX1 and PDX2, respectively (Fig. 2b, c). Then we overlapped the trimmed DEGs of analyzed PDXs and identified 513 DEGs in the intersection set (Fig.2c, d). After analyzing the 513 DEGs by GO analysis and KEGG pathway analysis, we found that ALDH⁺CD24⁻CD44⁺ BCSCs differed from the other populations in p53 signaling pathway, signaling pathways regulating pluripotency of stem cells, and central carbon metabolism in cancer, etc. (Fig. 2e, f, Additional file 4: Table S4). GSEA of the 513 DEGs also showed that the process of differentiation and development in ALDH⁺CD24⁻CD44⁺ BCSCs was significantly downregulated (Additional file 1: Figure S2, Additional file 3: Table S3).

To obtain unique A/X DEGs (X stands for groups B, C or D), we identified 90 out of 513 DEGs in two PDXs, the 38 upregulated (A > X) and 52 downregulated (A < X) genes in common (Additional file 1: Figure S4a). The GO analysis based on biological process identified PPIL3, P4HA2 and FKBP2 from 38 upregulated genes were involved in peptidyl-proline modification, suggesting that there might be some epigenetic modifications exclusively in BCSCs, while 52 downregulated genes were involved in regulation of cell differentiation, positive regulation of developmental process, regulation of multicellular organismal development and regulation of cell development (Additional file 4: Table S4). To search for potential prognostic markers of TNBC, we used the Kaplan-Meier plotter [10] to screen the 90 DEGs identified from ALDH⁺CD24⁻CD44⁺ BCSCs in analyzed PDXs. Among the 90 DEGs of purified BCSCs in PDXs (Additional file 1: Figure S4a), the high expression of P4HA2 (n = 255, p = 0.00057) and PTGR1 (n = 161, p = 0.001), and low expression of RAB40B (n = 255, p = 0.0069) in TNBC patients were associated with decreased RFS (Additional file 1: Figure S4b).

As assessed by quantitative real-time PCR (qRT-PCR), the relative expressions of PTGR1, P4HA2 and RAB40B was variable across different breast cancer cell lines, for instance, the expression of RAB40B was comparatively lower in TNBC cell lines, such as SUM149, SUM159 and MDAMB231, than those of the other cell lines (Fig. 3a). To further elucidate the role of these genes in TNBC, we used shRNA to knock down each gene in TNBC cell line SUM149. The expressions of PTGR1, P4HA2 and RAB40B were significantly lower after lentivirus infection confirmed by qRT-PCR (Fig. 3b). Knockdown of P4HA2 or PTGR1 downregulated CSC-related genes, such as SOX2, OCT4 and NANOG (Fig. 3b), as well as causing a significant decrease in the proportion of BCSCs as assessed by ALDEFLUOR assay (Fig. 3c) and mammosphere formation assay (Fig. 3d). However, knockdown of P4HA2 or PTGR1 had no effect on CD24⁻CD44⁺ population of SUM149, but only on ALDH⁺ population (Fig. 3c). In addition to their effect on the BCSC population, knockdown of P4HA2 or PTGR1 also inhibited cell proliferation verified by MTT assay (Fig. 3e). When we knocked down RAB40B, the CSC-related genes, SOX2 and OCT4, were upregulated (Fig. 3b). In addition to that, the amount of the mesenchymal-like BCSCs (CD24⁻CD44⁺) was increased (Fig. 3c). Interestingly, knockdown of RAB40B also prevented mammosphere formation (Fig. 3d) and cell proliferation in SUM149 (Fig. 3e). To further validate the function of RAB40B in TNBC, we used two different shRNAs (RAB40BSh-sh2 used in SUM149, and another new sequence RAN40BSh-sh3) to knockdown the expression of RAB40B in another two TNBC cell lines: SUM159 and MDA-MB-231. The shRNAs worked well as assessed by qRT-PCR (Additional file 1: Figure S5a). Knockdown of RAB40B up-regulated CSC-related genes, such as SOX2, OCT4 and NANOG (Additional file 1: Figure S5.a), consistent with the results in SUM149 (Fig. 3b). Knockdown of RAB40B had no effect on CD24⁻CD44⁺ population of SUM159 and MDA-MB-231 (Additional file 1: Figure S5b), however, knockdown of RAB40B significantly increased ALDH⁺ population (Additional file 1: Figure S5b), as well as causing a remarkable increase in mammosphere formation (Additional file 1: Figure S5c) and proliferation (Additional file 1: Figure S5d). These results seemed contradictory with the observation from SUM149, but this observation suggested RAB40B might play different roles in different cancer cells by affecting different BCSC population and also supported our previous report about the different proliferative capacity and cellular function between ALDH⁺ population and CD24⁻CD44⁺ population [7]. The functional analysis demonstrated that knockdown of the three potential prognostic markers would significantly affect the status of BCSCs and tumor growth simultaneously, indicating these genes might serve as the important prognostic markers in TNBC.