Circumventing senescence is associated with stem cell properties and metformin sensitivity

Abstract Most cancers arise in old individuals, which also accumulate senescent cells. Cellular senescence can be experimentally induced by expression of oncogenes or telomere shortening during serial passage in culture. In vivo, precursor lesions of several cancer types accumulate senescent cells, which are thought to represent a barrier to malignant progression and a response to the aberrant activation of growth signaling pathways by oncogenes (oncogene toxicity). Here, we sought to define gene expression changes associated with cells that bypass senescence induced by oncogenic RAS. In the context of pancreatic ductal adenocarcinoma (PDAC), oncogenic KRAS induces benign pancreatic intraepithelial neoplasias (PanINs), which exhibit features of oncogene‐induced senescence. We found that the bypass of senescence in PanINs leads to malignant PDAC cells characterized by gene signatures of epithelial‐mesenchymal transition, stem cells, and mitochondria. Stem cell properties were similarly acquired in PanIN cells treated with LPS, and in primary fibroblasts and mammary epithelial cells that bypassed Ras‐induced senescence after reduction of ERK signaling. Intriguingly, maintenance of cells that circumvented senescence and acquired stem cell properties was blocked by metformin, an inhibitor of complex I of the electron transport chain or depletion of STAT3, a protein required for mitochondrial functions and stemness. Thus, our studies link bypass of senescence in premalignant lesions to loss of differentiation, acquisition of stemness features, and increased reliance on mitochondrial functions.

To investigate the molecular changes associated with the transition from premalignant senescent lesions to malignant tumors, we took advantage of genetically engineered mouse models (GEMMs) of pancreatic ductal adenocarcinoma (PDAC) that mimic the progression of the human disease. Activating Kras mutations in the pancreas lead to premalignant lesions known as pancreatic intraepithelial neoplasias (PanINs), which are largely nonproliferative and contain cells with markers of cellular senescence (Caldwell et al., 2012). We thus compared the transcriptome and biological properties of  Figure 1j). Since these cells still grow in culture, they likely represent a mixture of senescent and pre-senescence cells (Itahana et al., 2003). We also compared expression of several pro-inflammatory genes between PanIN cells and PDAC cells. We found that the expression of Angptl2, Ccl5, was increased in PanIN cells, while the expression of Ccl7 was decreased (Supporting Information Figure S1b,c). These genes are often upregulated in senescent cells as part of the senescence-associated secretory phenotype (SASP; Coppe et al., 2008). Further evidence for a senescence gene expression pattern of PanIN cells was obtained using novel senescent biomarkers common to multiple cell types (Hernandez-Segura et al., 2017). Again, PanIN cells F I G U R E 1 Establishment of an in vitro model of pancreatic cancer progression. (a) Illustration of stages in pancreatic cancer progression after oncogenic ras activation (Kras * ) in vivo. Adapted from Wilentz et al. (2000). Mouse pancreatic ductal cell lines were established from the indicated lesions of Pdx1-Cre;LSL-Kras G12D mice. (b) Proliferation of the indicated mouse cell lines measured by MTT. The relative proliferation represents the fold of OD at 500 nm over the indicated period of time. Each point represents the mean of triplicates ± SD. (c) Cell lines from pancreatic ductal adenocarcinoma (PDAC) form colonies in soft agar, but not cell lines from ADM/PanIN1 lesions. Scale bar = 400 μm. (d) Quantification of proliferation in soft agar over a period of 7 days for the indicated cell lines. Results were obtained using the CyQuant GR dye and are expressed as relative fluorescence unit (RFU) at 520 nm. Mean of triplicates ± SD, **p < 0.01. (e) Tumor volume and weight 15 days after subcutaneous injection of 5 × 10 5 1,499 or AH375 cells into SCID mice. Only AH375 cells form tumors. (f) Phenotype and histology of subcutaneous tumors formed by AH375 cells. H&E, hematoxylin and eosin. Black arrow, ductal histology (g) Phenotype and histology of tumors formed following orthotopic injection of AH375 cells into the pancreas of SCID mice. H&E, hematoxylin and eosin; AC, normal acinar cells; S, spleen; T, tumor. show gene expression changes corresponding to the senescent phenotype (Supporting Information Figure S1d (Yang et al., 2010), and WNT signaling (Supporting Information Figure S2c) (Clevers & Nusse, 2012). Intriguingly, while 1,499 cells expressed many genes regulated by NF-κB, AH375 cells poorly express genes downregulated by the same transcription factor (Supporting Information Figure S2c) suggesting a reprogramming of the NF-κB pathway and a role for NF-κB-mediated gene repression in pancreatic cancer. In line with this result, it was reported that NF-κB-mediated gene repression was important to protect cancer from cytotoxic stimuli (Campbell, Rocha, & Perkins, 2004). Next, we confirmed that c-Myc and Stat3 protein

| Mouse and human cells that circumvent OIS express stemness genes
To investigate the bypass from OIS to transformation more broadly, we utilized normal human lung fibroblast IMR90, primary human mammary epithelial cell (HMEC), and primary mouse embryonic fibroblast (MEF) models of OIS. In each model, shRNA-mediated inhibition of aberrant ERK signaling promotes bypass of senescence allowing tumor formation in vivo, consistent with prior studies (Supporting Information Figure S3) (Deschenes-Simard et al., 2013). We

| Bypass from senescence is associated with the emergence of cells with stemness properties
To determine whether our findings at the gene expression level are transposable to a phenotype of CSCs, we tested the capacity of the mouse cell lines derived from the different pancreatic lesions to form free-floating tumor spheres. It was previously shown that both normal (Reynolds & Weiss, 1996)

| Pancreatic cancer cells show increased mitochondrial machinery
Another striking feature of the transcriptome of PDAC cells in comparison with PanIN cells is an increase in the expression of mitochondrial genes (Figure 5a,b). High mitochondrial mass correlates with cancer stem cell properties (Farnie, Sotgia, & Lisanti, 2015), and in pancreatic tumors, the stem cell population has an increase in respiration and dependency on electron transport and oxidative phosphorylation (Sancho et al., 2015;Viale et al., 2014). Also, RASdependent transformation requires mitochondrial STAT3 (Gough et al., 2009) a protein that is compromised in PanIN cells and OIS cells. We found that many mitochondrial enzymes involved in amino acid metabolism were upregulated in PDAC cells in comparison with PanIN cells (Figure 5a,b). These include several enzymes in one-carbon metabolism pathways such as serine hydroxymethyltransferase (Shmt2) (Lee et al., 2014) and glycine decarboxylase (GLDC) (Hiraga & Kikuchi, 1980). Notably, GLDC is important for the growth of tumor-initiating cells in lung cancer . We con- u/software/elda/). We found that 1/2,473 cells were capable of initiating tumors (Supporting Information Figure S4a). We also identified quiescent stem-like cells in AH375 cells growing as tumor spheres using a label retention assay with CFSE. AH375 cells were stained and plated for tumor spheres. After 3 days in culture, each sphere displayed a single cell labeled with CFSE (Supporting Information Figure S4b) suggesting that putative cancer stem cells divided only once asymmetrically and then remained quiescent.
To gain more insight into the heterogeneity of AH375 PDAC cells, we sorted the cells according to mitochondrial mass using MitoTracker Green (Figure 5d). We found that cells with higher mitochondrial content formed more tumor spheres than cells with lower mitochondrial content (Figure 5e,f). Also, the mitochondrial content of the cells that formed spheres was higher than that of cells growing in adherent conditions (Figure 5g). In addition, treatment of PanIN 1,497 cells with LPS stimulated their conversion into tumor sphere-forming cells (Figure 4h,i) and increased their mitochondrial content (Figure 5h). Taken together, our results show an association between mitochondrial genes and mitochondrial mass with the transition from PanIN to PDAC, a process that is stimulated by LPS.

| Metformin targets reprogrammed pancreatic cancer cells
The increased mitochondrial content of pancreatic cancer cells suggested that they might be sensitized to drugs that target mitochondria such as metformin (Owen, Doran, & Halestrap, 2000). The proportion of cancer stem-like cells in the population of AH375 cells growing in 2D is likely low. However, if these cells require an increase in mitochondrial function while they are growing in 2D, a pretreatment with metformin before plating them for the tumor sphere 3D assay should reduce the number of tumor spheres. We found that a pretreatment with 1 mM metformin for three days was sufficient to reduce by almost 50% the number of tumor spheres formed by AH375 cells (Figure 6a,b) while having a minimal effect on cell viability ( Figure 6c). Together, the results suggest that metformin could be a valuable drug to target mitochondria in cancer stem cells. Metformin also inhibited AH375 cells from forming colonies in soft agar (Supporting Information Figure S5a,b) and from growing as tumor spheres (Supporting Information Figure S5a,c).
Metformin reduced sphere formation of NB508 PDAC cells but at higher concentration (Supporting Information Figure S5d). Similar results were seen in transformed MEF and IMR90 cells (Supporting Information Figure S5e,f) and in the human pancreatic cancer cell line HPAF-II (Supporting Information Figure S5g).
To get insights into the mechanisms by which metformin inhibits stemness, we measured the levels of several transcription factors required for the stem cell-like phenotype. We found that metformin reduced the levels of Nanog, c-Myc, and phosphorylated Stat3 (Figure 6d). Metformin also induced a gene expression pattern characteristic of a more epithelial state (Figure 6e). An inhibition of the NF-κB pathway was also observed at higher concentrations, as seen    (Zambirinis et al., 2013). Finally, the subpopulation of tumor-initiating cells in PDAC cells show increased mitochondrial content and are sensitized to metformin. Taken together, these observations suggest that OIS acts as an initial barrier for the proliferation of cells with oncogenic mutations but seems to prime cells to activate a stemness gene expression program that can contribute to progression into malignant tumors.

| The transition from PanIN
Two models can explain the protumorigenic actions of cellular senescence. First, senescent cells secrete a variety of inflammatory factors that have been previously linked to tumorigenesis in a noncell autonomous manner and stimulate the progression of cells that have not fully entered senescence (Coppe, Desprez, Krtolica, & Campisi, 2010). Consistent with this idea, the histone deacetylase-associated protein SIN3B is required for Kras-induced senescence, secretion of IL-1α, and progression to PDAC in the pancreas. Also, SASP factors induce stemness and proregenerative responses in the skin of mice (Ritschka et al., 2017). However, other studies show a tumor suppressor role for the SASP. For example, inactivation of RelA or the TGFβ pathway in mouse models of Kras-driven pancreatic cancer accelerated pancreatic cancer formation by inhibiting the SASP (Acosta et al., 2013;Lesina et al., 2016). In the liver, the SASP also plays a role in stimulation of antitumor immune responses (Iannello, Thompson, Ardolino, Lowe, & Raulet, 2013;Lujambio et al., 2013).
A second model proposes that some senescent cells can evolve into tumor cells in a cell autonomous manner. In this case, intrinsic properties of the cells are important for tumor progression. Peeper and colleagues showed that inhibition of IL-6 expression could reverse human fibroblasts from RAF-induced senescence (Kuilman et al., 2008). Abbadie and colleagues have described an interesting mechanism in which rare senescent epithelial cells can give rise to a progeny of tumorigenic cells via amitotic cell division similar to budding (Gosselin et al., 2009). This process happens with a very low frequency and is stimulated by single-stranded DNA breaks (Gosselin et al., 2009). Recently, Schmitt and colleagues reported that senescent cells induced by chemotherapy in hematopoietic tumor cells express stemness genes and if they manage to overcome the cell cycle arrest they progress into very aggressive tumors (Milanovic et al., 2018). Further work will be required to find whether malignant tumors arise from fully senescent cells that escape from the arrest or from cell expressing the driving oncogene that avoided senescence due to factors in the microenvironment. The two models presented above (noncell autonomous vs. cell autonomous) are not mutually exclusive since it is plausible that specific inflammatory factors can act in synergy with cell intrinsic mechanisms that promote escape from senescence.
An important question still unresolved is the identity of the proinflammatory factors that can help to circumvent senescence. In a mouse model of Kras-driven pancreatic cancer, pancreatitis was required for PanIN lesions to progress to PDAC in association with inhibition of senescence (Guerra et al., 2011). This study suggests that the factors secreted by PanIN cells are not sufficient to promote malignant progression and that other factors controlled by the microenvironment are important for tumorigenesis. The microbiota plays an important role in the origin of pancreatic cancer (Farrell et al., 2012;Pushalkar et al., 2018). Consistent with these results, we show here that LPS, a bacterial product, can stimulate the emergence of cells with stem cell properties from PanIN cells. Future investigation of the effect of LPS on PanIN cells can help to identify F I G U R E 2 Stemness gene expression pattern in pancreatic ductal adenocarcinoma cells. (a) Transcriptome analysis comparing 1,499 and AH375 cells (GEO accession number: GSE57566). Transcripts with a fold change higher or equal to 2 and a p < 0.05 according to a twosample Student's t test were analyzed with the Babelomics 4.3 platform. The number of transcripts in each category (nonmutually exclusive) is indicated. (b) Validation by qPCR of the microarray data in the indicated cell lines and for the indicated genes, which are involved in epithelialmesenchymal transition (EMT). Mean of triplicates ± SD. (c) DIRE prediction of upregulated transcription factors (TF) in AH375 cells. The percentage of target genes found in the submitted list of transcripts is shown for each potential TF (occurrence). The importance indicates the product of a TF occurrence with its weight in the database. (d) GSEA found gene expression signatures suggesting upregulation of Stat3 and c-Myc in AH375 cells. (e) Immunoblots with anti-phospho-Stat3 (Y705), anti-Stat3, and anti-c-Myc antibodies on extracts from the indicated cell lines. (f) Histology of lung tumors formed after tail vein injection of 1 × 10 6 AH375 cells into SCID mice. B, bronchiole; AV, alveolus; M and black arrows, tumors. H&E, hematoxylin and eosin (g) Indirect immunofluorescence staining with anti-Myc and anti-phospho-ERK antibody of mouse lung tissues containing tumors as in (f). White arrows, metastasis; DAPI, nuclear counter stain; scale bar = 100 μm   (Lin et al., 2016), which can synergize with NF-κB, thereby changing gene expression patterns and promoting tumorigenesis (Grivennikov & Karin, 2010). In addition, STAT3 can promote mitochondrial functions required for cancer cell stemness and metabolism (Genini et al., 2017). IL-6 activates STAT3 and promotes the development of PDAC from PanIN (Lesina et al., 2011). Although this cytokine is required to maintain OIS in fibrob-  (Chen et al., 2017). Also, metformin has been shown to target selectively breast and pancreatic cancer cells with stem cell properties (Hirsch et al., 2009;Sancho et al., 2015). These results are consistent with a requirement for mitochondria in cells capable of initiating tumor growth. Notably, cancer chemotherapy can induce tumor cell senescence with the potential risk that some cells escape from the process with stem cell properties (Milanovic et al., 2018). We suggest that metformin could be considered as a general adjuvant to prevent tumor regrowth from cells that acquired stemness induced by the treatment.   Retroviral and lentiviral-mediated gene transfers were done as described (Deschenes-Simard et al., 2013). For senescence bypass experiments, retrovirus expressing RasV12 and shERK2 were simultaneously used to infect target cell populations. Cell proliferation assays, SA-β-Gal, soft agar assays, sphere-forming assays, immunoblotting procedure, immunofluorescence staining protocol, reagents, and plasmids are described in the Data S1.

| Flow cytometry and MitoTracker Green
staining AH375 cells grown adherent or in suspension (tumor spheres) were trypsinized and stained with MitoTracker Green (Thermo Fisher) at 1/7,500 for 10 min at 37°C in PBS-EDTA 5 mM with 2% fetal calf serum. Cells were centrifuged and washed twice with 10 volumes of PBS-EDTA 5 mM with 2% fetal calf serum.
Flow cytometry analysis was performed on a Canto II flow cytometer. MitoTracker Green fluorescence was detected in the GFP channel, gated on PI (propidium iodide) negative single cells.
Cell sorting was performed at IRIC cytometry platform on a BD FACSAria high-speed cell sorter.

| Statistical analysis
Student's t tests or ANOVA were used for comparisons between groups based on an assumption of normal distribution. Significant differences were considered at *p < 0.05, **p < 0.01, ***p < 0.001.
Results are represented as means ± standard deviation (SD).

CONFLI CT OF INTEREST
None declared.