Abstract
Epidemiological studies have established a positive association between obesity and the incidence of postmenopausal (PM) breast cancer. However, the molecular mechanisms underlying this correlation are not well defined. A central phenotypic characteristic of obese individuals is increased circulating and interstitial abundance of free fatty acids. Here we demonstrate that long-term exposure to palmitic acid (PA) drives cancer cell dedifferentiation towards a cancer stem-like phenotype and enhanced tumor formation capacity. We demonstrate that this process is governed epigenetically through increased chromatin occupancy of CCAAT/enhancer-binding protein beta (C/EBPB). C/EBPB regulates cancer stem-like properties by modulating the expression of key downstream regulators of the extracellular matrix (ECM) including SERPINB2 and LCN2. Collectively, our findings demonstrate that C/EBPB plays a critical role in the initiation of cancer cells in obesity.
Statement of Significance Cellular adaptation to obesity-induced palmitic acid drives tumor initiation through activation of a C/EBPB-dependent transcriptional network. This highlights a mechanistic connection between obesity and postmenopausal hormone receptor negative breast cancer.
INTRODUCTION
Breast cancer is the most frequently diagnosed cancer and the leading cause of cancer-related death amongst women. Risk factors for breast cancer include the non-modifiable factors such as age, genetics and reproductive history, as well as modifiable factors as obesity, alcohol consumption, and tobacco smoking. The majority of cancers can be attributed to a combination of several of such genetic, hormonal, and environmental factors. As an independent risk factor, PM obesity accounts for up to 20% higher risk of developing breast cancer, and every 5-unit increase in BMI is associated with a 12% increase in breast cancer risk (1). Whereas obesity in PM individuals has been consistently linked to enhanced risk of developing estrogen receptor (ER) positive breast cancer, there has been more debate on the effect in ER negative breast cancer (2). These discrepancies likely relate to challenges to align study designs across heterogeneous studies, highlighting the need for a mechanistic understanding on the interactions between the obese state and breast cancer risk. In addition to the reported effects on breast cancer incidence, several meta-analyses have shown that overweight and obesity are associated with worse overall survival and metastasis-free survival independent of their menopause or hormone receptor status (3, 4). In ER positive breast cancers, the link to obesity has been attributed to increased estrogen signaling (5, 6). However, in ER negative breast cancers the molecular mechanisms of this connection are largely unknown – particularly for obesity-induced tumor initiation. Thus far, proposed mechanisms includes obesity-induced chronic inflammation (7, 8), altered insulin signaling (9, 10), deregulation of estrogen (11), rewiring of cancer metabolism (12) and secreted adipokines (13).
Here we aim to determine the molecular mechanisms that link breast cancer and obesity. We demonstrate that obesity had adverse effects on patient survival in PM, ER/progesterone receptor (PR) negative breast cancers compared to other subtypes. Using single cell-based analysis we show that long-term exposure to high concentrations of palmitic acid (PA) led to dedifferentiation towards cancer stem cell-like properties across human and mouse cell models of ER/PR negative breast cancer cell lines as determined by higher expression of CD44, CD133 and Axl. Using patient tissue microarray (TMA) we show that the frequency of cancer cells expressing the stem cell marker CD133 were more abundant in samples from overweight and obese patients compared to normal weight ER/PR negative PM patients. ATACseq and Cut&Run coupled with transcriptomic analysis of cells adapted to high PA levels followed by loss-of-function and gain-of-function studies, identified CCAAT/enhancer-binding protein beta (C/EBPB) as a required transcriptional regulator of PA-induced cancer stem-like properties. We further demonstrate that C/EBPB induced stemness through the modulation of the ECM proteins SERPINB2 and LCN2. Taken together, our findings indicate that C/EBPB plays a critical role in the initiation of PM/ER-/PR- breast cancer cells in obesity.
RESULTS
Transcriptional changes induced by long-term culture in palmitic acid overlap with obesity-dependent transcriptional changes in hormone receptor negative patients
To set the framework for our mechanistic studies of the connection between obesity and breast cancer, we first sought to identify a group of patients affected by the obese state. To that end, we performed Cox regression survival analyses of 115 PM (defined by age of >50) breast cancer patients using BMI and hormone (estrogen and progesterone) receptor status as variables in a highly controlled in-house dataset (14). Despite equal BMI distribution (Figure S1A), overweight and obesity (BMI>25) were associated with significantly reduced survival rates in hormone receptor negative patients as compared to non-obese patients (Figure 1A) whereas no effects of BMI were observed in the hormone receptor positive patients (Figure S1B). Importantly, within the PM/ER-/PR- patient group there were no differences between the high and low BMI groups in patient age (Figure S1C), tumor size (Figure S1D) or tumor stage (all included patients were stage 3) at the time of diagnosis. Obesity leads to the production of reactive oxygen species in adipose tissue (15, 16). Given the abundant adipose tissue in the mammary gland and the association between reactive oxygen species and mutagenesis (17) we next performed high-coverage sequencing of 360 known cancer genes (18) in tumor samples collected from the PM/ER-/PR- patient group at the time of diagnosis. Based on this analysis we were not able to detect any mutations correlating to obesity across this panel (Supplementary Table S1). Combined, this suggests that in PM/ER-/PR- breast cancer patients, obesity potently modulates the tumor biology towards a more aggressive disease independent of genetic changes. The obesity-programmed environment is complex, and includes an altered immune status and deregulated abundancies of circulating hormones and metabolites. To understand how cancer cells evolving in such an environment impacts patient survival, we focused on the interaction between obesity-induced circulating levels of free fatty acids (19–21) and cancer phenotypes. In particular, we were interested in PA as this is the most abundant fatty acid in circulation, and has been reported to be epidemiologically associated with a higher risk of developing PM breast cancer (22). We exposed hormone receptor negative breast cancer lines to increasing PA concentrations over a period of 2 months to enable cellular growth in PA concentrations corresponding to serum levels in obese patients (Figure 1B). Human hormone receptor negative (MDA-MB-231 and HCC1806) and mouse (E0771 and TeLimet) breast cancer cells consistently adapted to acquire resistance to PA-induced apoptosis to enable persistent growth even in the high PA environment (Figure 1C, D). In MDA-MB-231, HCC1806, and TeLimet cells, the acquired resistance was accompanied by a reduction in growth rate, whereas E0771 cells maintained its growth rate even after adaptation to high levels of PA (Figure 1E). To ascertain how such adaptation resembles what is observed in obese breast cancer patients, we compared the transcriptional alterations observed during cellular adaptations to PA to the transcriptional changes induced by obesity in PM hormone negative breast cancers patients. To this end, we applied iPAGE, an information-theoretic framework (23), to query how genes induced or repressed in obesity were changed upon adaptation to PA in the in vitro model. For this analysis, genes were first ordered based on their expression changes between MDA-MB-231 parental and adapted cells (termed MDApar and MDAapa, respectively) and were subsequently divided into 10 equally populated bins. We then assessed the distribution of obesity-associated genes across these bins. As shown in Figure 1F, we observed a significant depletion/enrichment pattern (MI=0.006 bits, z-score=21.14). We specifically noted a significant overlap between genes that were induced by the obesogenic state in patients and those up-regulated through in vitro adaptation to PA (Figure 1F). This shared reprogramming of the gene expression landscape suggested that the in vitro long-term adaptation to high abundancies of PA provides clinically relevant information on the molecular drivers of obesity-induced hormone receptor negative breast cancers.
Long-term adaptation to palmitic acid in culture facilitates cellular dedifferentiation towards cancer stem cell-like properties
Cellular adaptation to new extracellular environments is a central mechanism underlying disease progression and therapy resistance (24). To understand the cellular phenotypes enriched in breast cancer cells subjected to long-term exposure to PA, we performed a single cell mass cytometry analysis using an antibody panel targeting 27 markers of cellular differentiation states and signaling pathways (Figure S2A). The distribution of cellular subpopulations during adaptation to the PA-rich environments was analyzed using the X-shift algorithm (25) for both MDA-MB-231 (Figure 2A,B) and HCC1806 cells (Figure 2D,E). This analysis revealed an enhanced expression of cancer stem cell markers CD44 (26, 27), CD133 (28) and Axl (29) in the PA-induced cell clusters (CL1059 for MDAapa and CL731 for HCCC1806apa) of the adapted cells (Figure 2C, F). This suggested that long-term adaptation to PA induced a consistent cellular dedifferentiation towards a cancer stem-cell like phenotype in MDA-MB-231 and HCC1806 cells. Increased frequency of CD133+ cell populations were validated using flow cytometry (Figure 2G). To test if the tumor stem cell-like phenotype was associated with tumor formation capacity in vitro, we subjected parental and adapted cells to a tumor spheroid formation assay. All four adapted cell lines were consistently able to form spheroids at significantly higher frequencies compared to their parental counterparts (Figure 2H). Combined, these findings suggested that cellular adaptation to long-term exposure of fatty acids induces a stem cell-like cancer phenotype.
Obesity is associated with increased frequency of stem cell-like cancer cells in PM/ER-/PR- breast cancer patients and mouse models of breast cancer
Having shown that the adapted cell lines were associated with enhanced tumor spheroid formation in vitro, we then wondered if this was translatable to in the clinical setting. To that end, we obtained tumor tissue microarrays (TMA) from the PM/ER-/PR- patients used in the initial survival analysis (Figure 1A) and immunostained the cores for CD133. The image analysis platform QuPath (30) was used to segment the images, differentiate between stromal and cancer cells and to quantify CD133+ cell frequencies. Consistent with our in vitro observations, PM/ER-/PR- breast cancer patients with a BMI above 25 displayed higher CD133+ cancer cell frequencies as compared to the normal BMI patients (Figure 3A). To confirm this obesity-induced cancer stem cell-like phenotype in vivo, we orthotopically implanted E0771 and TeLi cells at limiting dilutions in a C57BL/6J diet-induced obesity model and measured tumor formation. Following 10 weeks of high-fat diet (HFD) feeding the mice gained more weight (Figure 3B) and displayed multiple hallmarks of obesity-induced comorbidities such as liver steatosis, hyperinsulinemia, hyperglycemia and reduced glucose clearance compared to the regular chow fed mice, suggesting that the diet-induced obesity recapitulated the systemic obese environment found in humans (Figure 3C, D, E). Following mammary gland implantation of limiting number of E0771 and TeLi cells we found that the high-fat environment consistently promoted tumor formation with a 6 to 10-fold enrichment in stem cell frequencies (Figure 3F). To test the robustness of the induced stem cell-like phenotype in vivo, we next dissociated tumors formed in the obesogenic and non-obesogenic environments, allowed them to grow in standard 2D culture for 4 days and subjected them to an ex vivo spheroid formation assay. Interestingly, both E0771 and TeLi cancer cells adapted to the in vivo obesogenic environment maintained their enhanced ability to form spheroids (Figure 3G). Consistent with the overlap between transcriptional signatures between obese patients and the PA-adapted cell lines, E0771apa cells formed tumors earlier than E0771par cells (Figure 3H). Combined, these findings suggested that an obesogenic environment promoted a cancer stem-cell phenotype in breast cancer cells in patients, in in vivo obese breast cancer mouse models and in ex vivo cellular models upon long-term adaptation to PA.
Adaptation to a PA-rich environment induces metabolic reprogramming towards fatty acid oxidation
Metabolic reprogramming has been linked to stem-cell behavior in breast cancer (31, 32). To understand the mechanism(s) by which PA adaptation drives differentiation towards a stem-cell like phenotype, we first metabolically characterized parental and adapted MDA-MB-231 and HCC1806 cells using radiolabeled [1-14C]PA and D-[14C(U)]glucose as tracers. After adaptation to a PA-rich environment, both MDA-MB-231 and HCC1806 cells displayed higher reliance on free fatty acid oxidation (Figure 4A and B). In contrast, glucose oxidation was unchanged in MDA-MB-231 cells and slightly decreased in adapted HCC-1806 cells (Figure 4A and B). Consistent with the in vitro system, E0771 cells isolated from tumors grown in obese and non-obese mice also displayed increased PA oxidation and reduced glucose oxidation ex vivo (Figure 4C). In addition to deregulated PA oxidation, we also found increased lipid droplet formation in adapted cells as assessed by incorporation of fluorescent dye BODIPY, suggesting that adaptation to PA enhanced increased fatty acid oxidation and storage (Figure 4D). To functionally test if such metabolic reprogramming towards fatty acid oxidation might be linked to the observed stem cell-like properties, we treated cells with etomoxir (an inhibitor of fatty acid oxidation by inhibiting free fatty acid uptake into the mitochondria) and subjected them to the spheroid formation assay. However, at concentrations that robustly reduced cellular respiration (Figure 4E), we were unable to detect any effects on spheroid formation capacity in either HCC1806apa or E0771apa cells, suggesting that the shift in substrate oxidation was a correlative event and not the main driver of the PA-induced stem cell-like phenotype (Figure 4F).
Adaptation to a PA-rich environment induces open chromatin linked with C/EBPB occupancy
Deregulation of metabolic intermediates has recently been tightly linked to epigenetic remodeling and cell fates (33). We therefore next assessed chromatin accessibility by ATAC sequencing (ATACseq) of parental and PA-adapted cells. Interestingly, adaptation to PA led to a substantial gain and loss in chromatin accessibility in 42336 and 48991 regions across the genome, respectively (Figure 5A and Figure S5A, S5B). As expected, chromatin accessibility changes in promoter regions were correlated with concordant changes in transcription of the downstream genes as determined by RNA sequencing (RNAseq) (Figures 5B, C and Figure S5C, S5D). To identify potential regulators associated with the observed chromatin alterations upon PA-adaptation, we next aggregated changes in chromatin accessibility near putative binding motifs to infer differential motif activity and occupancy of transcription factors (34). This analysis identified the C/EBPB transcription factor as the topmost hit associated with the more accessible chromatin in the PA-adapted cells (Figure 5D, E). This suggested that C/EBPB acts as a transcriptional regulator of obesity-induced tumor initiation capacity in obese breast cancer patients. Functional depletion of C/EBPB by independent short hairpin RNAs (shRNAs) in MDAapa, HCC1806apa and E0771apa cells (Figure S5E-G) led to a significant reduction in spheroid formation capacity in vitro (Figure 5F). In contrast, depletion of RUNX1 and C/EBPBA did not affect spheroid formation capacity (Figure S5H-K). Further, upon transplantation into the mammary fat pad, depletion of C/EBPB significantly delayed tumor formation in the diet-induced obese setting, while the knockdown had no effect in the non-obese setting (Figure 5G). All together, these experiments support a model wherein C/EBPB is associated with transcriptionally active chromatin and is required for the cancer stem-like phenotype in obesity.
C/EBPB is encoded by an intron-less gene that is expressed in three isoforms; LAP1, LAP2 and LIP by alternative use of transcription start sites (35, 36). Both LAP1 and LAP2 isoforms contain a dimerization and a transcriptional regulation domain and functions as dimers (35). LIP lacks the DNA binding domain and has been suggested to function as a competitive inhibitor of LAP1/2 (35). We next asked whether C/EBPB expression itself was sufficient to confer stem-like properties. To this end, we overexpressed either LAP2 or LIP in MDA-MB-231 cells (Figure S5L). The LAP2 construct contains a conservative mutation ATG to ATC (Met to Ile) that removes the LIP translational start site and thereby preventing the co-expression of LIP (35). MDApar cells fail to form spheroids under normal culture conditions, however ectopic overexpression of LAP2, but not the LIP isoform, enabled spheroid formation in all biological replicates (Figure 5H). Interestingly, overexpression of LAP2 in MDAapa cells further enhanced spheroid formation after 1 day of culture as opposed to 5 days in control conditions (Figure 5I). In contrast and in line with its suggested dominant negative effect, LIP overexpression in the MDAapa cells abrogated spheroid formation (Figure 5I). Moreover, LAP2 overexpression increased the fraction of CD133+ cells, and this effect was more pronounced in adapted than in parental cells (Figure 5J), consistent with the higher chromatin accessibility to genomic regions containing C/EBPB binding motifs in the adapted cells. These findings collectively suggested that the C/EBPB isoform LAP2 is a key regulator of cancer stem stem-like properties.
Unexpectedly, we found that the protein levels of C/EBPB isoforms did not differ between adapted and parental cells (Figure S5M-O). Neither did we detect any differences in C/EBPB nuclear localization (Figure S5P). The lack of difference in protein levels or localization suggested that the activity of C/EBPB might be regulated post-translationally, and that its effect on cancer stemness could be associated with epigenetically determined increased occupancy in chromatin regions with gains in target accessibility.
Differential C/EBPB occupancy regulates the expression of extracellular matrix proteins
Having shown that C/EBPB is required and sufficient for spheroid formation capacity, we next applied Cut&Run to confirm its genome-wide occupancy and to identify its putative downstream transcriptional targets. Cut&Run uses micrococcal nuclease tethered to DNA-bound proteins to generate short DNA cleavage fragments and thus enables base-resolution digital footprints that reflect precise protein-DNA binding sites (37). We enumerated the ends of every Cut&Run fragment (≤ 120 bp) for each base of the genome and detected significant footprints de novo based on the footprint occupancy score (38). As expected, motif enrichment analysis identified C/EBPB as the topmost enriched motif in significant Cut&Run footprints, confirming successful genome-wide profiling of C/EBPB under PA adaptation (Figure 6A, Figure S6A and S6B). Similar results in motif enrichment were confirmed using a peak-based approach (Figure S6C).
In line with the ATACseq data, Cut&Run confirmed increased C/EBPB occupancy, in the same chromatin regions which had increased accessibility in the adapted cells as compared with the parental cells (Figure 6B). We linked distal and proximal gains in C/EBPB occupancy and chromatin accessibility in PA adaptation to genes whose expression correspondingly increased, and/or based on high-confidence enhancer-gene associations identified cross-platform in GeneHancer (39) (e.g. LCN2; Figure 6C). Pathway analysis of these regions revealed a significant enrichment in processes involved in extracellular matrix (Figure 6D), suggesting a potential link between ECM remodeling and cancer stemness.
To translate our data derived from the in vitro PA adaptation system into the in vivo and clinical settings, we subsequently focused on the subset of the putative C/EBPB target genes whose expression was significantly elevated in the obese as compared to the lean PM/ER-/PR- patients. This analysis identified nine genes, namely, SERPINB2, LCN2, SERPINB7, NELL2, MMP9, CLDN1, LYPD6B, CRISPLD1 and CHST4 (Figure 6E). Interestingly, all of these nine genes had elevated expression in E0771 cells analyzed ex vivo after having been grown in obese as compared with non-obese mice; whereas the expression of C/EBPB was unchanged as expected (Figure 6F). In short, these data supported a model wherein obesity induced C/EBPB chromatin binding, activating a transcriptional network involved in ECM processes.
C/EBPB target genes SERPINB2 and LCN2 are required for cancer stem cell-like capabilities
To determine the functional importance of the nine genes in CEPBP-dependent cancer stemness, we next assessed the levels of the nine genes in cells where C/EBPB was overexpressed. Ectopic overexpression of the LAP2 isoform of C/EBPB in MDAapa cells led to the induction of five of these nine genes (Figure 7A); whereas ectopic expression of LIP did not affect the expression level of the genes (Figure 7A). Interestingly, LAP2 overexpression particularly augmented the expression of SERPINB2, LCN2 and CLDN1, which paralleled the differential expression patterns observed in cells adapted to obese and non-obese environment (Figure 6E). We therefore functionally tested the role of SERPINB2, LCN2 and CLDN1 in epistatic spheroid formation assay and found that SERPINB2 and LCN2 were required for LAP2 induced spheroid formation capacity (Figure 7B and Figure S7A-C). The combined depletion of SERPINB2, LCN2 and CLDN1 also prevented LAP2-induced spheroid formation (Figure 7B). Further, individual depletion of SERPINB2 and LCN2 (Figure S7D, E) phenocopied C/EBPB knockdown and significantly reduced tumor spheroid formation capacity (Figure 7C, D). This suggested that SERPINB2 and LCN2 were the main downstream mediators of C/EBPB. To assess the clinical impact of these findings, we stratified PM/ER-/PR- patients according to their SERPINB2 and LCN2 expression and determined survival outcomes in our in-house as well as an independent dataset (GSE25066). Consistent with C/EBPB driving a more aggressive cancer phenotype in PM/ER-/PR- breast cancer patients, these survival analyses demonstrated that high expression of SERPINB2 and LCN2 were associated with worse survival outcomes (Figure 7E-F). Combined, these findings support a model wherein the obese environment epigenetically activates C/EBPB transcriptional activity that is required and sufficient for tumor formation capacity through the regulation of its target genes SERPINB2 and LCN2 (Figure 7G).
DISCUSSION
Obesity is a complex pathological condition that conceivably affects the formation and development of cancers through multiple avenues. Here we have demonstrated that cancer cell adapted to high levels of PA is one such potent mechanism through which obesity drives enhanced tumor formation capacity in PM/ER-/PR- breast cancer. We find that adaptation to PA governed dedifferentiation of cancer cells towards a tumor stem cell-like phenotype leading to augmented tumor formation capacity. Clinically this manifest in a higher cancer cell frequency of CD133+ cancer stem cells and shorter disease-specific survival in obese and overweight PM/ER-/PR- breast cancer patient compared to normal weight patients. This is corroborated epidemiologically by the association of obesity with higher cancer risk (40) and poor prognosis (4) of PM/ER-/PR- breast cancer patients. Our findings thus provide a cancer cell autonomous mechanism for the increased appreciation that obese environments lead to enhanced tumor formation capacity in breast cancer (7,11,41–43). Our findings further expand on the molecular mechanisms of PA-induced stemness, by demonstrating that that obesity-induced stemness is mediated through the epigenetic activation of a C/EBPB dependent transcriptional network. At the genome-wide level we demonstrate that the obese environment facilitates a widespread deregulation of chromatin accessibility. Chromatin regions with increased accessibility in cells adapted to high levels of PA were enriched for C/EBPB binding motifs. Through complementary sets of in vitro and in vivo experiments, we show that C/EBPB is required for obesity-induced tumor formation. Conversely, ectopic overexpression of C/EBPB enhanced the frequency of cancer stem cells. Previous reports observed that C/EBPB is required for stem cell maintenance in the developing breast (44) and that expression of the LAP2 isoform of C/EBPB can transform a non-cancerous cell line MCF10A (45), lending further support to the functional role for C/EBPB-dependent cancer stem cell-like properties.
Our unbiased Cut&Run analysis of direct C/EBPB target genes suggested that C/EBPB regulates stemness features through regulation of the surrounding ECM. Cancer cell-autonomous regulation of the ECM is intrinsically linked to cancer stemness through manipulation of mechanical properties and signaling molecules (46, 47). Consistent with our findings, obesity-induced alterations in the ECM mechanics has been reported to support tumorigenesis (48). Interestingly, a total of 9 C/EBPB target genes were also induced in obese PM/ER-/PR- breast cancer patients. Of these 9 genes, depletion of ECM proteins SERPINB2 and LCN2 phenocopied C/EBPB knockdown and were epistatically required for C/EBPB induced spheroid formation capacity suggesting that these engender the downstream effects of C/EBPB. Both of these factors have previously been implicated in the regulation of cancer stem cell-like properties. SERPINB2, also known as plasminogen activator inhibitor type 2 (PAI-2), is widely described as an extracellular urokinase inhibitor that is upregulated in many inflammatory states. In cancer biology, SERPINB2 was observed to be induced in brain metastatic breast cancer cell subpopulations (49) and was recently suggested to be broad marker for cancer stemness in multiple cell culture models (50). Although the mechanism linking SERPINB2 to stemness, particularly in the context of an obese environment, is currently unknown, our work suggests a link to extracellular plasmin homeostasis. Interestingly, our unbiased analysis of C/EBPB dependent drivers of obesity-induced stemness in breast cancer additionally highlighted a functional role of the small extracellular protein LCN2. LCN2 is induced in adipose tissue of obese individuals (51) and were previously described to reduce inflammation and fibrosis and in an obesity-driven pancreatic ductal adenocarcinoma model (52). In breast cancer, LCN2 has been linked to cellular differentiation through modulation of the epithelial to mesenchymal transition (53). While SERPINB2 and LCN2 factors have been suggested to be involved cancer stemness, future work is needed to establish the mechanistic basis of their actions – especially in the context of obese environments.
Aberrant lipid metabolism is a hallmark of deregulated cancer metabolism (54). It has been widely reported that cancer cells augment their de novo lipid biosynthesis for energy production, synthesis of new membranes, to regulate membrane structures that coordinate signal transduction, and for the biosynthesis of lipid signaling molecules such as phosphatidylinositol-3,4,5-trisphosphate (55). In addition, cancer cells can stimulate the release of fatty acids from surrounding adipocytes to provide energy for tumor growth (56). Here we demonstrate that an obese environment governs a metabolic switch towards higher fatty acid oxidation. However, this reprogramming was not causally linked to obesity-induced stemness. In support of a link between fatty acids and stemness, is the observation that slow-cycling metastasis-initiating cells are dependent on the lipid uptake protein CD36 (57). While we did not observe any direct involvement of CD36 in our studies of obesity-induced breast cancer, both studies describe a critical role for fatty acid metabolism in cancer stemness.
Our findings furthermore identify a critical link between adaptation to obese environments and genome-wide changes in chromatin accessibility. This is analogous to recently observations that high fat feeding leads to alterations in chromatin interactions to drive adaptive networks (58). These interactions likely reflect diet-induced alterations in metabolic intermediates that are intimately connected to epigenetic control of gene transcription (59, 60). Interestingly, lipid-derived acetyl-CoA has been suggested to be the source of up to 90% of acetylation modifications of certain histone lysine’s (61).
Combined, our analysis of cellular adaptations to obese environments has revealed changes of cellular phenotypes, driven by the combined modulation of C/EBPB transcriptional activity. In the context of personalized medicine, this suggest that obese cancer patients might benefit from specific targeted therapies rather than generic treatment regiments.
METHODS
Breast Cancer Patient Cohort
This study enrolled a total of 223 patients with primary stage III breast cancers. Out of these 115 patients were PM patients (defined by age > 50 years). Recruitment period was between November 24, 1997 and December 16, 2003. The median age was 51 years (range 25–70). Patient’s BMI, age, hormone status at the time of diagnosis as well as patient survival times (overall survival and disease specific survival) were documented. The study was approved by the regional committees for medical and health research of Western Norway (REK-Vest; approval number 273/96-82.96). More details about the study cohort can be found in the following report (14).
Animal Models
All animal experiments were approved by the Norwegian Animal Research Authority and conducted according to the European Convention for the Protection of Vertebrates Used for Scientific Purposes, Norway. The Animal Care and Use Programs at University of Bergen are accredited by AAALAC international. The laboratory animal facility at University of Bergen was used for the housing and care of all mice. C57BL/6J mice were obtained from Jackson Laboratories and bred on site. Female mice were kept in IVC-II cages (SealsafeÒ IVC Blue Line 1284L, Tecniplast, Buguggiate, Italy); 5-6 mice were housed together and maintained under standard housing conditions at 21°C ± 0.5°C, 55% ± 5% humidity, and 12h artificial light-dark cycle (150 lux). Mice were provided with standard rodent chow (Special Diet Services, RM1 801151, Scanbur BK, Oslo Norway) and water ab libitium.
To mimic both obese and non-obese environments, 6 weeks old female littermates were randomly assigned to chow and HFD groups and fed either standard chow diet (75% kcal from fat, 17.5% from proteins and 75% from carbohydrates, Special Diet Services RM1, 801151) or high fat containing diets (60% kcal from fat, 20% from protein and 20% from carbohydrates, Research Diets, D12492) for 10 weeks prior to tumor cell implantations. Body weight was monitored every week. The respective diets were maintained throughout the experiment.
Cell Lines and Culture
MDA-MB-231 (TNBC, human), HCC1806 (TNBC, human) and HEK293T cell lines were purchased from the American Type Culture Collection (ATCC). E0771 (TNBC, mouse) cell line was purchased from the CH3 BioSystems. TeLi (basal breast cancer, mouse) cells were originally derived from a tumor formed in MMTV-Wnt1 transgenic mouse and then propagated in vivo for four generations through mammary fat pad injections before being passaged in vitro. Tumors was dissociated using Mouse tumor dissociation kit (Miltenyi Biotec, 130-096-730) according to manufacturer’s instructions. Dissociated tumor cells were cultured in vitro for two months to obtain pure tumor cells. The in vivo passaged MMTV-Wnt cells were kindly provided by Stein-Ove Døskeland, University of Bergen. The TeLimet cell line was generated in house by dissociating lung metastasis derived from tail vein injected TeLi cells that were stably transfected with a reporter plasmid containing green fluorescence protein (GFP) and luciferase. MDA-MB-231, E0771, TeLi and TeLimet cells were cultured at 37°C, 5% CO2 in high-glucose DMEM (Sigma, D5671) supplemented with 10% FBS (Sigma, F-7524), 100U/mL penicillin and 100 μg/mL streptomycin (Sigma, P-0781) and 2 mM L-glutamine (Sigma, G-7513). HCC1806 cells were cultured in RPMI1640 (Sigma, R8758) supplemented with 10% FBS and 100U/mL penicillin/ and 100 μg/mL streptomycin.
For cell line authentication, MDA-MB-231 cells were harvested for genomic DNA extraction using Genomic DNA isolation kit (Norgen Biotek, 24700). Isolated genomic DNA was analyzed by Eurofins Genomics laboratory and the cell line authenticated based on genetic fingerprinting and short tandem repeat (STR) profiling.
Patient Tissue Microarray and Transcriptomic Analysis
Tissue Microarray
Tissue specimens were from the human breast cancer patient cohort described above (14). At the time of diagnosis, each patient from the study cohort had an incisional tumor biopsy. All tissue samples were fixed in formaldehyde for paraffin embedding, in the operating theatre immediately on removal. Paraffin embedded tissue were subject to tissue microarray (TMA) construction. From each tumor, 4 cores of 1.2 mm diameter from tumor rich areas were punched out using Manual Tissue Arrayer Punchers (MP10; Beecher Instruments). The patient cores were embedded into ten 8 x 10 array blocks plus 1 to 2 liver control cores for orientation. Microtome sectioned slides were stored at 4°C until ready for use.
Immunohistochemical staining was done as described previously (28). In short, slides were dried at 58°C over two days and deparaffinization was performed using xylene, rehydrated with ethanol and dH2O. Target retrieval was done in Tris/EDTA buffer, pH 9 (Dako, S2367) in a microwave for 25 min. Slides with buffers were cooled down at room temperature for 15 min, followed by rinsing with cold dH2O. Samples were then blocked in the Peroxidase Blocking solution (Dako REAL, S2023) for 8 min, rinsed with water and then blocked in a serum-free protein block buffer for 8 min (Dako, X0909). Primary CD133 antibody (Miltenyi Biotec, 130-090-422) was diluted 1:25 in Antibody Diluent with Background Reducing Components (Dako, S3022). 200 µl of antibody solution was put on each slide to cover all TMA specimens and incubated overnight at 4°C.
The following day, slides were washed twice with Dako Wash Buffer (S3006). Primary antibody detection was performed using MACH3 mouse probe (Biocare Medical) followed by MACH3 HRP polymer (Biocare Medical, BC-M3M530H) and the signal was developed with diamino-benzidine DAB+ (Dako, K3468). Finally, the slides were counterstained with hematoxylin (Dako, S3301), dehydrated in alcohol solutions and xylene, and mounted in Pertex Mount Agent (Histolab, 00801).
Transcriptomics
mRNA expression levels were extracted from previously reported microarray analyses (62). In brief, these analyses were performed on a Human HT-12-v4 BeadChip (Illumina) after labeling (Ambion; Aros Applied Biotechnology). Illumina BeadArray Reader (Illumina) and the Bead Scan Software (Illumina) were used to scan BeadChips. Expression signals from the beads were normalized and further processed as previously described (63). The data set was re-annotated using illuminaHumanv4.db from AnnotationDbi package, built under Bioconductor 3.3 in R (64), to select only probes with “perfect” annotation (65). The probes represented 21043 identified and unique genes.
Sequencing of 360 cancer related genes
Targeted sequencing of 360 cancer genes, was performed and described previously (18). In brief, native, genomic DNA from tumor, was fragmented and subjected to Illumina DNA sequencing library preparation. Libraries were then hybridized to custom RNA baits according to the Agilent SureSelect protocol. Paired-end, 75bp sequence reads were generated. Sequencing coverage for the targeted regions (average per bp) within each sample was >120x for all samples (mean 439x). Supplemental Table S1 lists the included 360 genes.
Proliferation assay
Cell proliferation assay was determined by high-content imaging using the IncuCyte Zoom (Essen Bioscience) according to the manufacturer’s instructions. In all experiments, cells were seeded into a 96-well culture plate and for each well four fields were imaged under 10x magnification every 2 h. The IncuCyte Zoom (v2018A) software was used to calculate confluency values.
Glucose and insulin measurements
For glucose and insulin measurements, mice were fasted overnight (9 hours) with free access to water. Blood glucose concentrations were determined using Accu-Check Aviva glucometer (Roche). For insulin measurements, blood was collected from the tail using EDTA coated capillary tubes (Fisher Scientific, 11383994), stored on ice before centrifuged at 2000 g, 4 °C for 10 min. Plasma insulin concentrations was determined in duplicates using the Ultra Sensitive Mouse Insulin ELISA Kit (Crystal Chem, 90080) following the manufactures instructions for wide range measures.
Glucose tolerance test
For glucose tolerance test, mice fed a HFD or chow-diet for 10 weeks were fasted overnight (15 hours) with free access to water. Glucose (2.5 g/kg) was administered by gavage, and blood glucose concentrations were determined by using Accu-Check Aviva glucometer (Roche).
Mammary Fat Pad Implantations
E0771 or TeLi cells were prepared in PBS and mixed 1:1 by volume with Matrigel (Corning, 356231) and orthotopically implanted into the 4th inguinal mammary fat pad of chow and HFD fed mice in a total volume of 50μL. Tumor diameters (width and length) were measured 2-3 times per week with caliper. Tumor volumes were calculated using formula Tumor volume (mm3) = Width × Length2 x π/6. Tumors were considered established when the volumes were larger than 50mm3.
Cellular Adaptation to Palmitic Acid
Cells were seeded on 10 cm culture dishes so that the confluency at the starting day of selection was 80-90%. To start selection, all media was removed and replaced by growth media supplemented with 200 µM palmitic acid (PA) (Sigma, P5585). After cells acquired resistance to this concentration, the concentration of palmitic acid was increased to 400 µM. For E0771 and TeLi cell lines we finally increased concentration to 500 µM and 600 µM, respectively, due to high intrinsic PA resistance of these cells. For HCC1806 cells the concentration were reveresed to 200 µM due to the fragility of the cells and their inability to survive in 400 µM PA after the standard cryopreservation in FBS/10%DMSO. Parental cells were cultured in parallel using growth media supplemented with 1% fatty acid free BSA (Sigma, A7030). PA adapted cells were cultured in growth media supplemented with 1% fatty acid free BSA and indicated concentration of PA. For PA supplemented media, PA was first dissolved in absolute ethanol to obtain a 50mM stock. To prepare the working concentrations, certain volumes of PA stock were added into 1%BSA growth media and incubated at 37°C for 1hour. PA stock was stored at 4°C and used for no longer than 2 weeks.
Generation of knockdown and overexpressing cell lines
Short hairpin RNAs (shRNA) for target genes and scramble (shCtrl) were purchased from Sigma as bacterial glycerol stocks (#1864). pBabe-puro plasmids containing human C/EBPB LAP2 and LIP isoforms were from Addgene (Cat.# 15712 and 15713).
For production of virus, HEK293T cells were seeded onto 10 cm plates to reach 80% confluency on the following day. For retroviral overexpression, 12µg of Gag/Pol plasmid, 6µg of VSVG plasmid and 12 µg of pBabe-puro plasmid containing C/EBPB isoforms were respectively co-transfected into the HEK293T cells using 60 μL Lipofectamine 2000 according to manufacturer’s protocol. For lentiviral-mediated depletion of target genes, cells were transfected with 12µg Gag/Pol plasmid, 6µg envelope plasmid and 12µg shRNA containing plasmid (pLKO).
6 hours following transfection, the media was replaced with fresh media. The virus was harvested 48hours post transfection by spinning the collected culture media for 5 mins at 1200 rpm and then filtered through a 0.22 µm filter to completely remove cell debris. The virus was then stored at −20°C for several days or at −80°C for several months.
To infect target cells, 5mL of the appropriate virus was used to infect a subconfluent 10 cm cell culture dish in the presence of 10 µg/mL of polybrene overnight. 48 hours after infection, puromycin was added to select for successfully infected cells: 4 µg/mL for TeLi, 2 µg/mL for MDA-MB-231 and E0771 and 1.33 µg/mL for HCC1806 cells. Uninfected control cells were processed the same way to determine the endpoint of selection. Typically, selection took 2-3 days for all cell lines. After the end of selection cells were released from puromycin for at least 1 day before starting experiments.
Spheroid Formation Assay
Cells were harvested using Trypsin, resuspended in the corresponding medium, and seeded onto ultra-low attachment U-bottom plates (Corning® Costar® Ultra-Low Attachment Multiple Well Plate, CLS7007-24EA) at a concentration of 2000 cells/100 µL/well for HCC1806, E0771 and TeLi cells or 4000 cells/100 µL/well for MDA-MB-231 cells. After 5 days MDA-MB-231 (2 days for E0771, TeLimet and 1 day for HCC1806) spheroids were imaged using the Nikon TE2000 fluorescence microscope. Spheroids were considered formed when cells were tightly adhered to each other, hindering the recognition of individual cells and formed a round sphere structure with clear boundary. Spheroid formation was quantified independently and blinded by two investigators.
Apoptosis
Analysis of apoptosis was performed using Alexa Fluor™ 488 conjugate Annexin V (Thermo Fisher, A13201) and propidium iodide (PI) according to the manufacturer’s instructions. Shortly, cells and their culture media were harvested and washed once in cold PBS. Cells were then resuspended in Annexin binding buffer (10 mM HEPES, 140 mM NaCl, and 2.5 mM CaCl2, pH 7.4) in a concentration of 1×106 cells/mL. To each 100 μL of cell suspension 5 μL of the Annexin V and 2 μL PI (at final concentration 2 μg/mL) was added. Cells were incubated in the dark at room temperature for 15 min. After the incubation period, 400 μL of Annexin binding buffer was added and cells were analyzed by flow cytometry (BD LSR Fortessa).
Flow cytometry analysis
For immunostaining for flow cytometry, cells were collected using Accutase (Sigma, A6964) and washed once in PBS. 1×106 cells per sample were stained with 0.6 µl of APC conjugated CD133 antibodies (key resource table) in 100 µl of PBS +1%BSA solution and incubated in dark for 20 min at room temperature. After incubation, cells were washed once with 5 ml of PBS/1% BSA and analyzed on flow cytometry (BD LSR Fortessa).
Immunofluorescent analysis
Cells were seeded in 24-well plates on Poly-L-lysin treated cover slips at 75 000 cells per well one day before the staining. On the day of the analysis, culture media was removed and 4% paraformaldehyde (PFA) in Distilled-PBS (DPBS) was added to fix cells for 20 minutes. Then PFA was removed and cells were permeabilized in 0.4% Tween/DPBS for 10 min at RT. This was followed by 3 washes in DPBS. Blocking was performed in 3%BSA/0.2% Tween/DPBS for 90 min. Slides were shortly washed in staining media containing DPBS/0.2% Tween/1.5% BSA. Then slides were covered by 500 µl of staining media with C/EBPB antibodies (1:100 dilution) and incubated overnight at 4°C on rocking platform.
Next day, slides were washed in DPBS 3 x 5 min and incubated with secondary antibodies (1:500 dilution) for 2 hours. This was followed by 5 min wash in DPBS, then 5 min incubation with DAPI (1:500 in DPBS) and then another wash in DPBS.
Further, slides were rinsed in distilled water and mounted with ProLong™ Diamond Antifade Mountant. Slides were dried overnight and imaged using Leica SP5 with 63x magnification.
Image quantification was performed using Fiji software. The nucleus and whole cell were demarcated based on DAPI and bright field, respectively. % nuclear C/EBPB were calculated by diving the nuclear signal by whole cell signal multiplied by 100.
Fatty Acid and Glucose Oxidation Assay
Fatty acid and glucose oxidation were assessed by providing 14C-labled palmitic acid or glucose to the cells, with subsequent capture of the released 14CO2; a technique previously described (66). In brief, cells were plated into 96-well tissue culture plates (MDA-MB-231, 45000 cells/well; HCC1806, 45000 cells/well; dissociated E0771, 25000 cells/well) in corresponding growth medium and incubated overnight to allow proper attachment. Radiolabeled [1-14C]palmitic acid (1 µCi/ml) and D-[14C(U)]glucose (1 µCi/ml) were given in PBS supplemented with 10mM HEPES and 1mM L-carnitine. Respective amounts of non-radiolabeled substrate were added to obtain final concentrations of D-glucose (5 mM) and BSA-conjugated palmitic acid (100 µM). Etomoxir (40 µM) was added to certain wells during palmitic acid oxidation, to monitor the non-mitochondrial CO2 production. An UniFilter®-96w GF/B microplate was activated for capture of CO2 by the addition of 1M NaOH (25 μL/well) and sealed to the top of the 96-well tissue culture plates and incubated for the indicated period of time at 37 °C. Subsequently, 30 µL scintillation liquid (MicroScint PS PerkinElmer) was added to the filters and the filter plate was sealed with a TopSealA (PerkinElmer). Radioactivity was measured using MicroBeta2 Microplate Counter (PerkinElmer). Protein measurement was performed for data normalization. The cells were washed twice with PBS, lysed by 0.1 M NaOH, and protein was measured using Pierce® BCA Protein Assay Kit.
Staining of Lipid Droplets
Cells were seeded on tissue culture plates 24 hours before the analysis. BODIPY (Thermo Fisher, D3922) was diluted in DPBS (Gibco, 14040-133) to a final concentration of 2 µM. Cells were washed in DBPS once and stained with BODIPY/DPBS mix for 15 min at 37°C. Further, cells were harvested from tissue culture plates, washed once in PBS and analysed on Flow cytometer BD LSR Fortessa. Data analysis was performed using FlowJo software.
RNA extraction, RT-PCR, and qPCR
Total RNA was extracted with a Total RNA purification Kit (NORGEN Biotek, 37500) according to the manufacturer’s protocol. cDNA was synthesized from 1 µg total RNA template with oligo-dT primers using a SuperScript® III First-Strand Synthesis kit (ThermoFisher Scientific, 18080-051) according to the manufacturer’s protocol. qPCR was carried out in quadruplicates with a LightCycler® 480 SYBR Green I Master Mix (Roche, 04887352001) using a LightCycler® 480 Instrument II (Roche, 05015243001). The results were calculated by ΔΔCt method using human HPRT (hHPRT) for human genes and mouse actin (mActin) for mouse genes. Primer sequences are listed in materials sources table (Supplementary Table S2).
Transfection of siRNA duplexes
One day before transfection cells were plated on T25 flasks at the density 250 000 cells/flask. After overnight incubation, cells were transfected using Lipofectamine® RNAiMAX according to the manufacturer’s protocol with some modifications: we used 3 µl of Lipofectamine per flask and final concentration of siRNAs was 20 nM. After 48 hours incubation cells were harvested and seeded for the analysis on ultra-low attachment U-bottom plates. After seeding, cells were infected for the second time using 0,075µl of Lipofectamine RNAiMAX per well with 20 nM siRNA.
Western blotting
Cells were lysed in RIPA lysis buffer (Thermo Scientific, 89901) complemented with protease inhibitor cocktail (cOmplete ULTRA Tablets, MINI, EDTA-free, EASYpack, 05892 791001) and phosphatase inhibitor cocktail (PhosStop, 04 906 837 001). After quantification with a BCA protein assay kit (Pierce, 23225), equal amounts of protein (typically 20-50 µg of protein per lane) were separated by electrophoresis on a NuPAGE 10% Bis-Tris Gel (Invitrogen, NP0315BOX) in NuPAGE™ MOPS SDS Running Buffer (20X, Invitrogen, NP000102) and then transferred to an activated Immobilon-P PVDF Membrane (Merck Millipore Ltd, IPVH00010 PORE SIZE: 0,45 µm). The membranes were blocked using 5% nonfat dry milk in PBS/0.1% Tween20 for 1h at RT, incubated with indicated primary antibodies for overnight at 4°C. This step was followed by an incubation with secondary IRDye-conjugated antibodies (Leicor, P/N 925-68070, P/N 926-32213). Detection and quantification were performed on Amersham Typhoon Gel and Blot Imaging Systems. A list of antibodies is given in the key resources table.
RNA sequencing
MDA-MB-231 parental and selected cells were plated at 1×105 cells/mL into 6-well plates in corresponding medium. After three days, cells were harvested, and RNA extraction was performed according to the manufacturer’s protocol. Potential DNA contaminations were removed by applying the RNA Clean & concentrator with DnaseI kit (Zymo, R1013). RNA sequencing libraries were prepared at the Genomic Core Facility at University of Bergen using Illumina TruSeq Stranded mRNA sample preparation kit according to the manufacturer’s instructions and sequenced on the same lane on a HiSeq 4000 sequencer with pair-end 75bp reads.
ATACseq library construction
ATACseq libraries were constructed as previously described (67). In brief, 5×104 cells were washed once with ice-cold PBS and pelleted by centrifugation. Cells were lysed in 50 µl RSB buffer (10 mM Tris-HCl pH 7.4, 10 mM NaCl and 3 mM MgCl2) containing 0.1% NP-40, 0.1% Tween-20 and 0.01% digitonin, and incubated on ice for 3 minutes for permeabilization. After incubation, samples were washed in 1 mL RSB containing 0.1% Tween-20 and pelleted at 500 g for 10 minutes at 4°C. Samples were then resuspended on ice in 50 µl transposition reaction mix containing 2.5 µl Tn5 transposase, 1x TD buffer (both Illumina FC-121-1030), 1x PBS, 0.1% Tween-20 and 0.01% digitonin, and incubated at 37°C for 30 minutes with agitation. Tagmented DNA was purified using Zymo DNA Clean and Concentrator-5 kit (Zymo D4014). The resulting DNA was amplified for 12-13 cycles. The libraries were purified with AMPure XP beads (Beckman A63880), quality-checked on Bioanalyzer (Agilent) and 75 bp paired-end sequenced on Illumina Hiseq 4000 at Genomic Core Facility at University of Bergen.
Cut&Run and library construction
Cut&Run was performed as described with minor modifications (37). Briefly, 5×105 cells were washed and bound to concanavalin A-coated magnetic beads (Bangs Laboratories, BP531). The cells were then permeabilized with Wash Buffer (20 mM HEPES pH 7.5, 150 mM NaCl, 0.5 mM spermidine and 1x Roche Complete Protease Inhibitor, EDTA-free) containing 0.025% digitonin (Digitonin Buffer) and 2 mM EDTA and incubated with primary antibody (anti-C/EBPB or IgG isotype control) overnight at 4°C. The cell-bead slurry was washed twice with Digitonin Buffer and incubated with 1x Protein-A/G-MNase (pAG-MNase; Epicypher) in Digitonin Buffer for 10 minutes at room temperature. The slurry was washed twice with Digitonin Buffer and incubated in Digitonin Buffer containing 2 mM CaCl2 for 2 hours at 4°C to activate pAG-MNase digestion. The digestion was stopped by addition of 2x Stop Buffer (340 mM NaCl, 20 mM EDTA, 4 mM EGTA, 50 µg/mL RNase A, 50 µg/mL GlycoBlue and 300 pg/mL in-house MNase-digested yeast spike-in chromatin) and the sample was incubated for 10 minutes at 37°C to release chromatin to the supernatant and degrade RNA. The supernatant was recovered, and DNA was isolated through phenol-chloroform extraction and ethanol precipitation. Libraries were constructed to enrich for sub-nucleosomal fragments using the NEBNext® Ultra™ II DNA Library Prep Kit for Illumina as described (NEB, E7645S). The libraries were size-selected and purified with AMPure XP beads, quality-checked on Tapestation (Agilent) and 100 bp paired-end sequenced on MiSeq at Genomic Core Facility at University of Bergen.
Mass Cytometry
Cells were plated in 10cm plates in triplicates to reach a confluency of 80% after 48 hours. For the analysis, cells were collected using TrypLE Express (Gibco 12604-021). 1 × 106 cells per condition were included. Cells were resuspended in cell culture media and treated with 0.25 µM Cisplatin for 5 min at RT. Further, cells were fixed in 1 mL of 1.6 % PFA in PBS for 10 min at RT. Cells were pelleted by centrifugation for 5 min at 900g and the pellets were stored at −80°C until staining with CyTOF antibodies. On the day of staining, samples were thawed on ice, resuspended in 500 µl of DPBS (Gibco, 14040-133) and incubated for 10 min at RT in DNases (Sigma, DN25)/DPBS solution. Further, cells were washed in D-WASH solution (DPBS + 1% FA-free BSA + 0,02% NaN3 + DNase) and barcoded (Fluidigm, 201060) according to the manufacturer’s protocol. Cells were then washed twice in the Cell Staining Buffer (Fluidigm, 201068), all samples were combined and labeled with surface antibody cocktail (Figure S2A, extracellular) for 30 min at RT. Further, cells were pelleted by centrifugation and incubated in 4 mL of DPBS/DNase solution for 10 min at RT. After this step cells were washed in PBS-EDTA and fixed in 2% PFA/PBS (filtered through a 0.22 µm filter) for 30 min RT, followed by wash in Cell Staining Buffer and permeabilization in cold methanol (−20°C) for 10 min. After incubation, cells were washed in once PBS, once in D-WASH and labeled with intracellular antibody cocktail (Figure S2A, intracellular) for 30 min at RT. This was followed by incubation of cells in D-WASH for 10 min at RT and double wash in D-WASH. Then cells were incubated in 2% PFA/PBS with iridium cell tracker at 4°C overnight. The samples were spun down the following day and incubated in D-WASH for 10 min at RT, washed once in PBS/EDTA once, 3 times in di water (Fluidigm, 201069), resuspended in EQ beads (Fluidigm, 201078) diluted 1:9 in water, and analyzed on Helios - Mass Cytometer.
QUANTIFICATION AND STATISTICAL ANALYSIS
CyTOF data development using X-shift
Raw FCS-files were normalized, concatenated and debarcoded in R using Cytometry dATa anALYSis Tools; CATALYST (68). X-shift algorithm from VorteX clustering and visualization environment (25) was applied on CyTOF data of parental or adapted MDA-MB-231 and HCC1806 cell lines. To enable comparisons, cells from the different conditions were downsampled (85 000 and 100 000 per sample for HCC and MDA respectively) and combined for each cell line prior to clustering. All markers were used except Keratin 7 and the cell cycle marker p-HisH3. The following parameters were used in Vortex: numerical transformation: arcsinh (x/f), f = 5.0, noise threshold = 1.0, distance measure: angular distance, clustering algorithm: X-shift (gradient assignment), density estimate: N nearest neighbors, number of neighbors for mode finding (N): determine automatically. The resulting clusters yielded elbow points of k= 27 and k= 35 for HCC and MDA cell lines respectively. For the visualization, a maximum of 20 000 events extracted from each cluster were plotted in a force-directed graph layout using VorteX and Gephi Toolkit 0.8.7 (https://gephi.org/toolkit/). For the analysis of the two more representative clusters in each cell line, the log2 fold change has been calculated from the X-shift scores converted to the normalized raw intensities values.
Survival Analysis
Patients were stratified into two groups by BMI 25. Disease-Specific survival (DSS) Kaplan-Meier curves were generated using GraphPad Prism software and statistical significance was calculated using Log-rank (Mantel-Cox) test.
The combined effect of LCN2 and SERPINB2 to patient survival was analyzed on PM/ER-/PR- patients from GSE25066. Patients were stratified by median of average normalized expression of LCN2 and SERPINB2. Kaplan-Meier curves and statistics were performed in the same way.
Mutual Information
Mutual information was calculated as described in Goodarzi 2009 (23)
Tissue Microarray Analysis
The ten CD133-stained TMA slides were scanned with an Aperio Scanscope CS Slide Scanner. The breast cancer cores were 1.2 mm in diameter with up to 4 cores per patient. Full analysis was performed on valid cores for patients 50 years and older with ER and PR negative status. Cores with too few cells, poor quality, excessive tearing, or folding were not considered valid and were omitted from analysis.
QuPath Version: 0.2.0-m5 was used to dearray the TMAs, segment cells, and classify cell types. The following detection steps and parameters were applied to all TMA slides. Simple tissue detection was used to find the approximate tissue borders within each dearrayed TMA core. A threshold of 229 (default 127), requested pixel size of 1 µm (default 20 µm), and checking the box for Expand boundaries were found to be the most important parameter setting changes for accurate tissue detection.
Watershed cell detection was used to create cell masks within the detected tissue of each valid core. The watershed parameters were optimized to detect large weakly hematoxylin stained cancer cells, to minimize false positive cell detection from areas of high background signal, and to reduce the creation of cell masks that spanned multiple cells. The watershed parameter changes deemed most important for accurate cell mask creation were: nucleus background radius of 10 µm (default 8 µm), nucleus minimum area of 24 µm2 (default 10 µm2,), nucleus maximum area of 230 µm2 (default 400 µm2), intensity parameters for threshold and max background both set to 0.07, and exclusion of DAB staining (as was recommended for membrane staining markers). Additionally, the cell expansion was set to 10 µm, 5 µm larger than the default setting, in order to capture the CD133 membrane staining on the large cancer cells.
Annotation objects were drawn around easily defined areas that contained primarily cancer cells, non-cancer cells, or platelets/RBCs and labeled as the classes tumor, stroma, or ignore, respectively. Platelets/RBCs were ignored because they appeared brown even before staining and show up as falsely positive for CD133. 9039 cells from the annotation objects drawn across 5 of the 10 slides were used to train the random forest (trees) classifier in QuPath. DAB specific measurements were excluded from the classifier selected features. The intensity feature used to identify CD133 positive cells was Cell: DAB OD max at a threshold of 0.45. With these parameters, the detection classifier created 7 classification groups of cells: total (base) tumor cells, total stroma cells, CD133+ tumor cells, CD133- tumor cells, CD133+ stroma cells, CD133- stroma cells, and ignored cells. Cell masks from cores with partial low quality due to folding or poor imaging were removed to prevent false positive cells. All cores were visually inspected for false positive cancer cell masks and false positive masks were removed. Mean CD133+ cancer cell percentage was calculated for each patient for all valid tumor cores by QuPath and exported to MS Excel. Patients with greater than 2% CD133 positive cancer cells were considered to have CD133 positive tumors. Statistical analysis was performed in GraphPad Prism.
Student’s t-test
Statistical analysis of flow cytomentry data was performed using student’s t-test on GraphPad Prism 8 software. 3 replicates per condition were performed and the experiment was performed 3 times.
Fisher’s exact test
Statistical analysis of spheroids formation experiments was performed using Fisher’s exact test.
Limiting dilution analysis
The frequency of tumor initiating cells was calculated using the Extreme Limiting Dilution Analysis (ELDA) (http://bioinf.wehi.edu.au/software/elda/index.html) (69)
RNA sequencing data analysis
Sequenced reads were quality checked with FastQC and aligned to the UCSC hg19 reference genome with Hisat2. Aligned reads were counted and summarized for the annotated genes using featureCounts. Differential gene expression analysis was performed by DESeq2. For visualization, normalized read counts were regularized log transformed (rlog).
ATACseq data analysis
ATACseq reads were quality-checked with FastQC (70) before and after adapter trimming with Trimmomatic (71). The trimmed reads were aligned to the UCSC hg19 reference genome using Bowtie2 (72) with the parameters --phred33 --end-to-end --very-sensitive -X 2000. Reads were then removed if they were mapped to the mitochondria and non-assembled contigs, had a mapping quality score below 10 and were PCR duplicates. Read start sites were adjusted for Tn5 insertion by offsetting +stand by +4 bp and -strand by -5 bp as previously described (67). For peak calling, MACS2 (73) was used with the parameters -q 0.01 --nomodel. Peaks residing in the ENCODE blacklisted regions were removed for further downstream analysis. deepTools (74) was used to generate 1x normalized bigwig files for visualization.
Analysis of differential accessible peaks was performed using DiffBind (75) with default settings, and annotated genome-wide with respect to the closest transcription start site with ChIPseeker (76). Peaks with a mean peak count ≥ 10 were kept for further analaysis. Annotated differential peaks were checked for phastCons conservation scores (77) for placental mammals against random noncoding background regions generated using bedtools shuffle. To infer differential transcription factor binding motif activity, diffTF (34) was used. Input transcription factor binding sites for 640 human transcription factors were generated as described using the HOCOMOCO database and PWMscan (cutoff p-value - 0.00001, background base composition - 0.29;0.21;0.21;0.29).
Cut&Run data analysis
Cut&Run reads were quality-checked with FastQC before and after adapter trimming with Trimmomatic. The trimmed reads were separately aligned to the UCSC hg19 and sacCer3 reference genomes using Bowtie2 with the parameters --local --very-sensitive-local --no-unal -- no-mixed --no-discordant --phred33 -I 10 -X 700 and --local --very-sensitive-local --no-unal --no- mixed --no-discordant --phred33 -I 10 -X 700 --no-overlap --no-dovetail, respectively. Reads were then removed if they were mapped to the mitochondria and non-assembled contigs and had a mapping quality score below 10. Mapped reads were converted to paired-end BED files containing coordinates for the termini of each read pair and the fragment length, and calibrated to the yeast spike-in using spike_in_calibration.csh (https://github.com/Henikoff/Cut-and-Run/) in bedgraph formats for visualization. Peaks were called with SEACR (78) with respect to the IgG control using the norm and stringent mode. Peaks overlapping with the ENCODE blacklisted regions were removed for further downstream analysis. Consensus peaksets across samples were generated using DiffBind. Raw counts of the consensus peaksets across samples were input to DESeq2 with the inverse of the spike-in calibration factors as sizeFactors to perform differential analysis. Differential peaks were annotated respect to the closest transcription start site with ChIPseeker.
For motif discovery within peaks, EChO (79) was run to identify direct binding sites in the foci mode. Sites with the mean fragment length ≤ 120 bp were retained, extended to a 100-bp window and converted to BED files. HOMER (80) was used for motif enrichment analysis using the position weight matrices (PWMs) from the HOCOMOCO database.
To identify enriched motif sequences protected by transcription factor binding independent of the peak calling algorithm, pA/G-MNase cutting footprints were detected. Ends of all CUT&RUN fragments ≤ 120 bp were enumerated to determine the precise single base pair cut sites and sorted. Footprints were detected using Footprint Occupancy Score (FOS) (38). Significant footprints with FOS ≤ 1 were analyzed for enriched motif sequences with HOMER.
Authors Contributions
Conceptualization, N.H; Methodology, N.H., X.L., A.R., S.G.M. and L.P.; Software, S.G.M., C.E.W., X.L. Validation: T.L., A.R., Formal Analysis, C.E.W., N.M., P.E.L., S.K., H.G., S.D.P., X.L., A.R., M.H.C.; Investigation, A.R., X.L.; Resources, N.H., S.D.P., S.M.G., J.L., S.K., P.E.L., S.K., A.M.; Writing – Original Draft, N.H., X.L., A.R.; Visualization, N.H., A.R., X.L., M.H.C., C.E.W.; Supervision, N.H.; Funding Acquisition, N.H.
Declaration of interests
The authors declare no competing interests.
CONTACT FOR REAGENT FOR AND RESOURCE SHARING
Further information and requests for resources and reagents should be directed to and will be fulfilled by the Lead Contact, Nils Halberg (nils.halberg{at}uib.no). This study did not generate new unique reagents.
SUPPLEMENTAL INFORMATION
Acknowledgements
We thank Erik Løkkevik, Bjørn Østenstad, Steinar Lundgren, Terje Risberg, and Ingvil Mjaaland for providing clinical samples. We thank the genomic score facility (GSF) at the University of Bergen, which is a part of the NorSeq consortium, provided services on RNAseq, ATACseq and Cut&Run. GSF is supported by grants from the Research Council of Norway (245979/F50) and the Trond Mohn Foundation (BFS2016-genom). The flow cytometry and mass cytometry were performed at the Flow Cytometry Core Facility, Department of Clinical Science, University of Bergen. Helios Mass Cytometer was supported by the Trond Mohn Foundation. We thank Ingeborg Winge from Department of Pathology, Haukeland University Hospital, Bergen, for provided training and help with TMA Immunohistochemistry. Hani Goodarzi is supported by R01CA240984 and R01GM123977. NH was funded by a Starter Grant from the Trond Mohn Foundation and the Norwegian Research Council (275250).
REFERENCES
- 1.↵
- 2.↵
- 3.↵
- 4.↵
- 5.↵
- 6.↵
- 7.↵
- 8.↵
- 9.↵
- 10.↵
- 11.↵
- 12.↵
- 13.↵
- 14.↵
- 15.↵
- 16.↵
- 17.↵
- 18.↵
- 19.↵
- 20.
- 21.↵
- 22.↵
- 23.↵
- 24.↵
- 25.↵
- 26.↵
- 27.↵
- 28.↵
- 29.↵
- 30.↵
- 31.↵
- 32.↵
- 33.↵
- 34.↵
- 35.↵
- 36.↵
- 37.↵
- 38.↵
- 39.↵
- 40.↵
- 41.↵
- 42.
- 43.↵
- 44.↵
- 45.↵
- 46.↵
- 47.↵
- 48.↵
- 49.↵
- 50.↵
- 51.↵
- 52.↵
- 53.↵
- 54.↵
- 55.↵
- 56.↵
- 57.↵
- 58.↵
- 59.↵
- 60.↵
- 61.↵
- 62.↵
- 63.↵
- 64.↵
- 65.↵
- 66.↵
- 67.↵
- 68.↵
- 69.↵
- 70.↵
- 71.↵
- 72.↵
- 73.↵
- 74.↵
- 75.↵
- 76.↵
- 77.↵
- 78.↵
- 79.↵
- 80.↵