More genes were suppressed than over-expressed in CSCs, i.e., 381 versus 122 genes, in the CSC feature panel, suggesting a more plastic state that CSCs represent. Thus, CSCs are similar with stem cells in a sense that they are pluripotent, but differ from stem cells as they are chaotic and occur under pathological conditions.
Down-regulated genes were enriched in ‘tight junction’ and ‘epidermis development’ (Figures 3A, 3C), attracting CSCs in a less differentiated state. Up-regulated genes were enriched in the ‘PI3K-AKT signaling pathway’ and cell migration related GO terms (Figures 3B, 3D), implicating the more proliferative and aggressive feature of CSCs than the bulk tumor cells.
The regulatory network constructed using down-regulated CSC feature genes was centered around ESR1 and ATP6V1B1. ESR1, also known as ER, is the primary biomarker used for breast cancer subtyping and typically under-expressed in triple negative breast cancers that harbor higher cancer stemness. Several genes connected with ESR1 have been associated with breast cancer stemness. For example, we previously reported the suppressive role of FOXA1 on breast cancer stemness ; SOX9 is as a stem cell factor that drives the epithelial mesenchymal transition (EMT) in non-small cell lung cancer through Wnt pathway and maintains human breast luminal progenitor and breast cancer stem cells through SOX9 mediated signaling , and both the SOX9/FXYD3/SRC  and the SOX9/SOX2  axes are critical for breast cancer stem cell functionalities; PLA2G7 is associated with ER negativity in clinical breast cancer samples and regulates EMT in vitro. The prognostic values of several markers have been reported, including PTPN6 and KRT19 in breast cancers[23-25], EPPK1 in non-small cell lung cancers, SERPINB5 in colorectal cancers, and NCCRP1 in squamous cell carcinoma. Besides, the roles of S100A7 and SLC37A1 was implicated in breast cancers[29, 30], that of SERPINB4 was reported in squamous cell carcinoma, and that of MYH14 was lately identified in pancreatic cancers. Also identified down-regulated in the CSC feature panel include several famous genes in cancer stem cell maintenance or signaling such as CDH1 , EPCAM and CLDN7 .
While the ER centered cluster has already been well-associated with cancer stemness (i.e., 12 out of 14 densely connected genes already have well-supported evidences), relatively less has been reported on the association between the cluster centered at ATP6V1B1 and CSC features. Most genes in this cluster are involved in cell energy production and metabolic regulations, suggesting the pivotal roles of metabolic reprogramming in CSC feature maintenance. Out of the 11 genes, 5 have known roles in cancer initiation and progression, i.e., SH3YL1 together with DOCK4 regulate breast cancer cell migration , ATP6V1C2 is an early prognostic marker for colorectal cancers , PRKAR2B promotes EMT and is oncogenic in prostate cancers [37, 38], MYO5B is associated with gastric cancers[39, 40], and the physiological role of AKR1B15 and its involvement in cancer development has been characterized in . Among the rest 6 members in the cluster, FAT2 encodes a cadherin family member and is aliased as FAT tumor suppressor homolog 2, MYO5C encodes a member of the same family with MYO5B that has already been implicated in cancers, ATP6V1B1, ATP6V1C2 and ATP2C2 encode ATPase subunits, and FA2H encodes fatty acid 2-hydroxylase, where the association between FA2H and cancers has been recently reported.
Up-regulated genes are centered at PRKCA which is a key component in PI3K-AKT signaling. This is consistent with the pathway analysis and suggests the prominent roles of uncontrolled proliferation and increased migration ability in CSC feature initiation and maintenance. VIM , MMP2, FGF2, ITBG3[45, 46], ANXA6 are metastasis-associated genes. The oncogenic roles of KCNMA1 has been revealed in breast, prostate, ovarian and colorectal cancers [48-52], and RRAD is associated with glucose uptake in a human ovarian cancer model and implicated in non-small cell lung cancers.
This CSC feature panel can clearly distinguish samples with high and low stemness in GSE132083, and characterize basal-A, basal-B and luminal cell lines in the E-MTAB-181 dataset. Though cell line wise difference exists in the molecular profiling of GSE132083, it is secondary to the differences characterized by cancer stemness (Figure 2). These not only suggest the validity of identified CSC feature in differentiating breast cancer cells with different CSC percentages, but also implicate the prominent roles of CSCs in driving breast cancer heterogeneity.
We experimentally explored the functionalities of ATP6V1B1 and its association with breast cancer stemness, through which the novel role of ATP6V1B1 in suppressing breast cancer stemness has been revealed. However, in-depth investigations on the regulatory mechanism of ATP6V1B1 in breast cancer progression and its potential prognostic value on breast cancer outcome await to be conducted.