A mammosphere formation RNAi screen reveals that ATG4A promotes a breast cancer stem-like phenotype

Breast tumours, like many other solid tumours, consist of highly heterogeneous cell populations with varying phenotypic and functional properties [1]. Similar to the normal mammary gland, these populations include cells with luminal-epithelial, basal and stem-cell-like features [2]. Based on gene expression profiles, basal-like breast cancers have been associated with the surface marker expression CD44⁺/CD24^-/low[3] while luminal epithelial cells have been associated with CD24⁺/CD44^- expression [4, 5]. Stem-like cells with tumour initiating capabilities have been found to be enriched in the CD44⁺/CD24^-/low sub-population of basal breast carcinoma cells [6]. These stem-like cells or cancer stem cells (CSCs) are held responsible for metastasis formation [7, 8] and chemoresistance [9]. Further, it was found that CD44⁺/CD24^-/low breast cancer cells exhibit epithelial-to-mesenchymal transition (EMT) features that might be responsible for their aggressive clinical behaviour [10]. EMT has long been recognised as an important programme for embryonic development [11] and has more recently been associated with breast CSC regulation [12, 13]. It is hypothesised that differentiated cancer cells can become CSCs as a consequence of EMT, allowing them to migrate, metastasize and survive chemotherapy [7, 14]. In line with these findings, CSCs have been connected to a mesenchymal phenotype [15], and it was shown that chemoresistant cells display not only CSC but also mesenchymal features [14]. Due to the aggressive nature of CSCs, the identification of genes and pathways that they depend on is an active area of research, fuelled by the promise that combination of conventional chemotherapy with specific CSC inhibitors will increase therapeutic success rates [16].

A major restraint when studying stem cells as well as CSC is their rareness. One approach to enrich breast stem cells, which has become particularly popular over the past ten years, is culturing cells as mammospheres [17]. It was shown that rare, single-founder stem cells can form multi-cellular sphere structures under serum-free suspension conditions that are enriched for stem and early progenitor cells [17]. Later, it was also found that rare cancer cells possess the ability to form mammospheres enriched for highly tumourigenic CSCs of the CD44⁺/CD24^-/low phenotype [5, 18]. Moreover it was shown that cells enriched in mammospheres had passed through EMT and were chemoresistant [19] which are two properties typically attributed to CSCs [12]. Here, we utilised the enrichment of CSCs in mammospheres and developed a high-throughput a short hairpin RNA interference (shRNAi) screening system to assay the involvement of over 5,000 genes in the maintenance of a population of cells with CSC properties. The results give an insight into molecular mechanisms underlying CSC maintenance in mammospheres and attribute a previously unrecognised role in this process to the autophagy regulator ATG4A.

It was previously shown that cancer cells with stem-like characteristics become strongly enriched in mammospheres [17]. This enrichment is a result of their ability to grow independently of anchorage, a condition under which most cancer cells undergo anoikis [26]. The resistance to anoikis is commonly attributed to cells that have undergone EMT [13, 26]. As shown in Figure 1E, flow cytometry analysis of mammosphere-derived SUM-149 cells revealed an enrichment of the CD44⁺/CD24^low/EpCAM^-/low population compared to adherent cultured cells. In accord with the enrichment of this sub-population in mammospheres, it was found that spheres express lower levels of E-cadherin and higher levels of vimentin when compared to adherent cells (Figure 1F). These data clearly confirmed that a sub-population of cells with CSC properties became enriched during mammosphere formation. Therefore, targeting the survival of these cells should lead to impaired sphere formation. Based on this hypothesis, we established a screening system for the identification of genes that are specifically involved in mammosphere formation.

SUM-149 cells were transduced with the pooled, lentiviral DECIPHER library Module 1 under conditions that ensured a maximum of one integration event per cell. The module targets 5,045 genes for knockdown by 5 to 6 dissimilar shRNA sequences per target gene, adding up to a total of 27,500 shRNA expression constructs that integrated into the genome of the host cells. In order to identify genes whose inhibition selectively impairs the formation of mammospheres, cells were sub-cultured under two distinct conditions. One fraction of cells was cultured adherently and a second fraction was cultured under mammosphere formation conditions (Figure 2A). After fourteen days in suspension culture, 1.3 percent of cells formed mammospheres (Figure 3B) with an average size of 120 μm (Figure 3C). A total of 1.8 million mammospheres or 66 spheres per shRNA were analysed in the screen. To identify shRNAs with expression that impaired sphere formation, mammospheres larger than 40 μm were collected, and smaller spheres and single cells were discarded (Figure 2A). From cells at the beginning of the screen (baseline), cells cultured adherently for fourteen days (t_adherent) and mammospheres larger 40 μm (t_sphere), barcode sequences were recovered using PCR and quantified using next-generation sequencing. Each barcode sequence stands for a particular shRNA expression construct. Figure 2B shows barcode read-count ratio distributions from sphere-cultured cells relative to the baseline. Corresponding values are given in Additional file 1. Based on those values, the impact of every single gene on adherent proliferation as well as mammosphere formation was determined and the results are shown in Additional file 2. In a first analysis step, a set of 1,015 genes was identified the inhibition of which significantly impaired the adherent survival of cells (P <0.01). Pathway enrichment analysis using the DAVID Functional Annotation Tool [21] revealed highest enrichment of identified genes in Kyoto Encyclopedia of Genes and Genomes (KEGG) pathways related to proteasomal and ribosomal function (Table 1A). Although inhibition of the majority of those genes also impaired mammosphere formation, they cannot be considered to inhibit this process selectively. Consequently, in a second analysis step, only genes that impaired mammosphere formation (P <0.01) but had no impact on adherent proliferation (P >0.1) were used for pathway enrichment analysis. Pathway analysis showed the highest enrichment of candidate genes in Janus kinase (Jak)-signal transducers and activators of transcription (STAT) and cytokine signalling followed by mTOR and several cancer-related signalling pathways (Table 1B). Genes associated with each pathway are shown in Additional file 3. As an example, five identified key regulators acting in Jak-STAT signalling are summarised in the scheme shown in Figure 2C.

CSCs are rare cells that are suspected to be responsible for tumour recurrence, formation of metastases and chemoresistance [7‐9]. The rareness of these cells makes it particularly hard for researchers to study their function. To date, the only functional possibility to enrich breast CSCs with tumour-initiating properties in vitro is to culture them as mammospheres [17, 18, 31]. We found that SUM-149 mammospheres were enriched for cells that expressed a surface marker signature typical for stem-like breast cancer cells (Figure 1E), passed through EMT (Figure 1B and 1F), were chemoresistant (Figure 1C) and more tumourigenic in NSG mice (Figure 1D). These are properties typically attributed to breast CSCs [12]. Although under adherent conditions this sub-population accounted for approximately 6% of the total population (Figure 5A), it became enriched 5-fold in mammospheres (Figure 5B).

Here, we exploited this enrichment to establish a high-throughput pooled RNAi screening system suitable to identify genes that are specifically involved in the maintenance of those rare cells with CSC properties. The system is based on the comparison of two separate RNAi screens performed under (i) adherent and (ii) mammosphere culture conditions (Figure 2A). With the first screen, genes essential for the survival of the total cell population were identified. The second screen identified genes important for sphere formation and hence, served as a surrogate screen to identify genes that are important for the maintenance of cells with CSC properties. Subtractive analysis finally revealed genes that are primarily important for the survival of cells with CSC properties. The identified genes were used for a pathway enrichment analysis, which returned a number of cancer-related pathways and regulatory processes (Table 1B). In total 22 candidate genes were found to be associated with the Jak-STAT signalling pathway (KEGG), making it the most significant pathway identified (P = 0.00176). A schematic presentation of some of the identified candidate genes acting in the Jak-STAT pathway is shown in Figure 2C. It is known that cytokine signalling via the IL6 receptor, GP130, JAK3, STAT1 and STAT3, as identified in our screen, is a regulator of breast CSC self renewal and differentiation [32, 33]. Further, activated Jak-STAT signalling is essential for the survival of CD44⁺/CD24^-/low stem-like breast cancer cells [34] and was shown to play an important role during mammosphere formation [35]. Last, STAT3 was identified by another RNAi screen to be a critical player in mammosphere formation and self-renewal of breast CSCs [36]. Taken together, these findings confirm the utility of the presented screening system to identify processes with specific relevance to the maintenance and expansion of CSCs.

Despite the advantages of a functional enrichment, culturing of cells as mammospheres also has some drawbacks when performing a high-throughput screen. For the analysis of the screen, we pooled all spheres bigger than 40 μm. Therefore, we could not distinguish between sphere size and number of spheres. Large spheres are believed to consist of more differentiated cells or early progenitors than smaller spheres. Another concern might be the formation of spheres because of cell aggregation instead of clonal growth. To overcome this hurdle, we chose an appropriate cell density to avoid any sphere fusion. Moreover, we validated our candidates in semi-solid soft agar colony formation assays that guarantee clonal sphere growth.

Besides Jak-STAT signalling, a number of previously uncharacterized candidate genes were identified in this screen, one of which was the positive regulator of autophagy, ATG4A. Autophagy is a lysosome-dependant degradation pathway allowing cells to remove macromolecules and damaged organelles in order to survive stress conditions [37, 38]. Interestingly, it was recently published that autophagy promotes the undifferentiated stem-like CD44⁺/CD24^-/low phenotype in breast cancer cells [39] and further evidence for the involvement of autophagy in cancer stem-like cell maintenance as well as their differentiation is accumulating rapidly [40‐43]. ATG4A is a redox-regulated cysteine protease [44]. ATG4A can cleave ATG8 near its C-terminus allowing the conjugation of ATG8 to phosphatidylethanolamine and subsequently to the membrane of the autophagosome [44]. In a second reaction, ATG4 can delipidate ATG8, releasing it from the autophagosomal membrane [44, 45]. This cleavage marks a final step in autophagy and allows the fusion of autophagosome and lysosome [46, 47]. The emerging role of autophagy in cancer stem cell maintenance together with the activation of lysosomal gene expression (Additional file 4) and upregulation of ATG4A (Figure 4A) in mammospheres strongly suggest an important role for autophagy, or more precisely, ATG4A in breast CSC maintenance. As it was demonstrated here, inhibition of ATG4A led to reduction of a sub-population with CSC properties (Figure 5A). Moreover, the enrichment of this sub-population during sphere formation was almost completely prevented by ATG4A inhibition (Figure 5B). Furthermore, ATG4A levels influenced size (Figure 3C) and numbers of mammospheres formed from breast cancer cell lines of the luminal and, even stronger, the basal type (Figure 3B and 3E). Last, modulation of ATG4A expression affected the tumourigenicity of SUM-149 cells under physiological conditions in the mammary fat pad of NSG mice (Figure 6A) as well as the composition of resulting tumours (Figure 6B). These results clearly demonstrate that ATG4A is involved in carcinogenesis and the maintenance of cells with a CSC phenotype.