Mitochondrial and ribosomal biogenesis are new hallmarks of stemness, oncometabolism and biomass accumulation in cancer: Mito‐stemness and ribo‐stemness features

Unbiased proteomics analysis: identification of proteins up-regulated in MCF7-fibroblast co- cultures

Previously, we have demonstrated that cancer cells, grown in close proximity to fibroblasts, metabolically reprogram these fibroblasts towards a more catabolic state, via the induction of autophagy, mitophagy and glycolysis, as well as senescence [16–20]. Conversely, through this interaction, cancer cells undergo reciprocal metabolic reprogramming towards a more anabolic state, with the induction of mitochondrial biogenesis and oxidative metabolism [21,22]. This metabolic co-operation primarily benefits the cancer cells by providing nutrients to generate new biomass [17,18]. Unfortunately, most of the metabolic targets in this symbiotic process remain completely un- known.

Here, we developed a new approach to identify these potential therapeutic targets, via unbiased proteomics analysis. For this purpose, we used MCF7 cells, an ER(+) human breast cancer cell line, as a model system. These MCF7 cells were co-cultured with hTERT-BJ1 fibroblasts, in a cellular ratio of 1:1, for a short 3-day period. These MCF7-fibroblast co-cultures were then directly compared to a 1:1 protein mixture of MCF7 cells and hTERT-fibroblasts, that had not been co- cultured together, but were grown instead as homotypic mono-cultures (Figure 1).

To a first approximation, using proteomics, this approach should allow us to estimate and identify which proteins are increased during the co-culture process, relative to mono-cultures. Then, these protein candidates were compared with human breast cancer samples that had undergone laser-capture micro- dissection, to validate their relevant expression in human breast cancer cells in vivo. For this purpose, we used publicly-available transcriptional-profiling data (from N=28 breast cancer patients; See Materials and Methods). A diagram highlighting this work-flow is shown as Figure 2.

Previously, we have shown that mitochondrial biogenesis in MCF7 cells is induced when they are co- cultured with fibroblasts [21,22]. However, it remains unknown exactly which mitochondrial proteins are induced during co-culture. Therefore, we focused first on which mitochondrial proteins were increased during co-culture, based on this proteomics approach.

Supplementary Table 1 illustrates the 45 mitochondrial- related proteins that were found to be significantly up- regulated during the MCF7-fibroblast co-culture process. This list includes proteins that are involved in mitochondrial biogenesis and/or are part of the mitochondrial complexes I to V, as well as mitochondrial chaperones, such as DNAJA3, HSPD1 and HSPA9, among others. Interestingly, NDUFAF2, a MYC-induced component of complex I [23], was infinitely up-regulated during co-culture. Interestingly, we have previously shown that the co-culture of MCF7 cells with fibroblasts confers Tamoxifen-resistance, that is reversible by treatment with Metformin, a complex I inhibitor [16]. Therefore, NDUFAF2 (a.k.a, Mimitin [23]) may be functionally conferring Tamoxifen- resistance during co-culture.

Since increased mitochondrial biogenesis is a hallmark of cancer stem cells (CSCs), known as “mito-stemess” [6–8,15], we next also looked for markers of protein synthesis [7], that is another hallmark of CSCs, which we now refer to here as “ribo-stemness”. Supplementary Table 2 shows that 28 components of the large and small cellular ribosomal subunits were up-regulated during co-culture. More specifically, 18 components of the large 60S ribosome and 10 components of the small 40S ribosome were increased. Interestingly, RPL4 and RPS29 were infinitely increased during co-culture.

Consistent with these findings, protein-folding chaperones were also induced. Supplementary Table 3 lists 10 chaperones that were significantly increased during co-culture. These include members of the HSP90 and HSP70 families of chaperones, as well as others. Remarkably, HSP90AB1 was infinitely up-regulated.

Similarly, Supplementary Table 4 shows that proteins involved in mRNA translation initiation, polypeptide elongation, tRNA synthesis and amino acid uptake were all significantly up-regulated during co-culture. Overall, this includes 30 proteins in total. For example, EIF2S1, a translation initiation factor required for mRNA binding to ribosomes, was increased by nearly 300-fold.

Furthermore, other well-known markers of “stemness” and proliferation were increased, as shown in Supplementary Table 5. More specifically, MKI67 was increased by >4,000-fold, while KRT19 and PCNA were increased by nearly 6-fold and 4-fold, respectively. The profound increase in MKI67 is more consistent with increased protein synthesis, rather than increased proliferation. Interestingly, MKI67 is expressed in all cycling cells, except for resting cells in the G0-phase and is specifically associated with ribosomal RNA (rRNA) synthesis and, thus, protein synthesis. This is consistent with our results presented in Supplementary Tables 2-4.

In this context, it is interesting to note that KRT19 is a well-established epithelial CSC marker that is used clinically to identify metastatic breast cancer cells in sentinel lymph node biopsies [24,25].

A summary of the proteins that were up-regulated by >100-fold is highlighted in Table 1. In conclusion, our current observations are consistent with the idea that MCF7-fibroblast co-cultures increase their biosynthetic cellular machinery (i.e., mitochondria and ribosomes), to expand their anabolic capacity to increase their biomass.

Ingenuity pathway analysis (IPA) of MCF7- fibroblast co-cultures

Differentially expressed proteins were also independently analyzed using Ingenuity Pathway Analysis (IPA), to identify altered canonical pathways and toxicity functions.

This analysis revealed that EIF2 signaling, which is crucial for protein synthesis, is a significantly activated canonical pathway in co-cultures, as compared to mono- cultures, as measured by a z-score > 2 (Figure 3). Other altered pathways and toxicity functions identified by IPA included mitochondrial dysfunction, NRF2- mediated oxidative stress, fatty acid metabolism, changes in mitochondrial membrane potential, HIF signaling, glutathione depletion, mitochondrial biogenesis, DNA-damage and fibrosis, among others (Figure 4). The observed similarity with renal injury appears to be related to an increase in oxidative stress and the resulting anti-oxidant response.

Overall, these results are consistent with the induction of “mito-stemness” and “ribo-stemness” features, during the co-culture process.

Validating the clinical relevance of co-culture proteomics data, using patient-derived breast tumor samples

In order to directly validate the potential clinical relevance of our findings, we next used transcriptional profiling data derived from the analysis of N=28 breast cancer patients, which were previously subjected to laser-capture micro-dissection, to physically separate breast cancer cells from adjacent stromal cells [26].

Therefore, we intersected our proteomics data, with this clinical data set, and the levels of fold-upregulation observed in the epithelial cancer cell compartment (relative to the tumor stroma), and corresponding p- values derived from these clinical samples, are shown in Supplementary Tables 6 to 9.

Supplementary Table 6 shows that of the 45 mitochondrial proteins that were up-regulated during co-culture, 34 were also significantly increased by >1.75-fold in human breast cancer cells in vivo. Therefore, >75% of the mitochondrial proteins elevated during co-culture were transcriptionally increased in patient-derived breast cancer cells in vivo.

Similarly, Supplementary Table 7 highlights that of the 28 ribosomal proteins that were up-regulated during co- culture, 27 were significantly increased by >1.7-fold in human breast cancer cells in vivo. Thus, >96% of the ribosomal proteins elevated during co-culture were transcriptionally increased in human breast cancer cells in vivo.

In addition, Supplementary Table 8 illustrates that of the 10 chaperone proteins that were up-regulated during co-culture, 7 were significantly increased by >3.1-fold in human breast cancer cells in vivo. As such, 70% of the chaperone proteins increased during co-culture were also increased in human breast cancer cells.

Finally, Supplementary Table 9 shows that of the 30 proteins associated with mRNA translation initiation, polypeptide elongation, and tRNA synthesis, 19 were significantly increased by >1.7-fold in human breast cancer cells in vivo. As a result, >60% of these proteins were also increased in human breast cancer cells.

Such a high concordance rate, between i) in vitro proteomics data and ii) in vivo human breast cancer transcriptional mRNA data, clearly validates the translational significance of the MCF7-fibroblast co- culture system, as a model for studying human breast cancer.

Establishing new prognostic biomarkers and companion diagnostics for predicting tamoxifen- resistance, by exploiting proteomics data from MCF7-fibroblast co-cultures

We have previously shown that the co-culture of MCF7 cells with fibroblasts induces a Tamoxifen-resistance phenotype [16]. Therefore, to identify new potential biomarkers of Tamoxifen-resistance, here we intersected our proteomics data from MCF7-fibroblast co-cultures with publicly available transcriptional profiling data from the tumors of breast cancer patients that were treated with Tamoxifen, but did not receive any chemotherapy [27,28]. A schematic diagram summarizing this approach is shown in Figure 5.

For this purpose, we selected high-risk patients that were lymph-node positive at diagnosis, and we focused on the luminal A subtype, which represents the most common form of ER(+) breast cancers (N = 152 patients). The results of this analysis are shown in Table 2. In this context, seven distinct mitochondrial genes, represented by 9 different gene probes, showed significant prognostic value, and were able to predict tumor recurrence.

In order to increase the prognostic power of these individual mitochondrial biomarkers, we next selected the most promising ones and used them to create a new mitochondrial gene signature. This new Mito-Signature contains only 3 key genes (HSPD1, VDAC2, CPT1A) (Tables 3 and 4). K-M curves for this signature are shown in Figures 6 to 13.

Importantly, this Mito-Signature yielded a significantly improved hazard-ratio for tumor recurrence of 5.52 (p = 7.3e-10). It was also highly predictive for distant metastasis, in the same group of patients (HR = 5.51; p = 8.1e-07). (See Figures 6). Similar results were obtained in Luminal A LN-negative patients and Luminal B patients (Figure 7).

We also examined the prognostic value of this mitochondrial gene signature in a larger group of ER(+) patients (N = 855), that received hormonal therapy, but not chemotherapy. This group of patients was not segregated into luminal A and luminal B subtypes. Figure 8 shows the results of this K-M analysis for relapse-free survival (HR = 2.80; p = 8.7e-15). Similar results were also obtained distant metastasis-free survival (HR = 2.26; p = 3e-05; N = 548 patients). This mitochondrial signature was also effective if the ER(+) patient population was divided into LN(+) and LN(-) groups (Figure 9).

Next, we assessed the behavior of this Mito-Signature in predicting overall survival. Figure 10 shows that it was also highly predictive of overall survival during hormonal therapy (HR = 3.06; p = 7.8e-05; N = 170 patients).

Moreover, Figure 11 shows that this Mito-Signature was also effective in all ER(+) breast cancer patients in predicting tumor recurrence (N = 3,082), as well as distant metastasis (N = 1,395). Similar results were obtained using data from all breast cancer patients, for recurrence (N = 3,951) and metastasis (N =1,746), as well as for post-progression survival (Figures 12 and 13). Thus, this mitochondrial-based gene signature may represent an important new prognostic tool for predicting patient outcomes, in a wide variety of different breast cancer patients, but especially in ER(+) patients treated with hormonal therapies.

Markers of fibrosis and glycolysis are up-regulated in MCF7-fibroblast co-cultures, consistent with the onset of oxidative stress

During the co-culture of fibroblasts with cancer cells, it has been observed that their phenotype is drastically changed [17,18]. These changes are related to the induction of various biological processes related oxidative stress, and are consistent with a more myo- fibroblastic phenotype [17,18]. These phenotypic changes include increased expression of cytoskeletal elements and glycolytic enzymes, as well as the elevation of markers of autophagy and senescence [18,21,22]. Therefore, we examined our proteomics data for evidence of these biological processes.

Supplementary Table 10 illustrates that during MCF7- fibroblast co-cultures, many known markers of fibrosis and oxidative stress are actually increased. These changes include the up-regulation of 32 cytoskeletal and extracellular matrix proteins, 11 glycolytic enzymes, 4 lysosomal/autophagy markers and 2 markers of the senescence-associated secretory phenotype, known as SASP.

These findings are also consistent with our results from IPA analysis, shown in Figure 4. Therefore, our results potentially provide interesting new stromal targets for further validation in future studies.

Validating the relevance of MCF7-fibroblast co- cultures for drug development, using FDA-approved antibiotics that target mitochondria

Based on our current findings, we observed that 45 mitochondrial proteins were significantly increased during the co-culture of stromal fibroblasts, with MCF7 cancer cells (Supplementary Table 1). This is consistent with our previous studies employing the vital fluorescent dye MitoTracker, showing that mitochondrial mass is dramatically increased during MCF7-fibroblasts co-cultures [21,22]. This may also have implications for drug sensitivity to mitochondrial inhibitors, especially those targeting mitochondrial biogenesis.

To directly test this hypothesis, we developed an assay system to monitor the sensitivity of cancer cells to drug treatment, in the presence or absence of fibroblasts. For this purpose, we generated hTERT-BJ1-fibroblasts expressing RFP (red fluorescent protein) and MCF7 cells expressing GFP (green fluorescent protein). Representative images of these mono-cultures are shown in Figure 14.

As positive controls for this assay system, we employed two FDA-approved antibiotics (Doxycycline and Azithromycin), which have been shown to act as inhibitors of mitochondrial biogenesis, because of the long-standing evolutionary relationship between mitochondria and bacteria [8,9]. Doxycycline and Azithromycin both inhibit mitochondrial protein translation, as off-target “side-effects”, by preferentially affecting the mitochondrial ribosome.

As predicted, Doxycycline preferentially targeted MCF7-GFP cells, during their co-culture with fibro- blasts, as directly compared with MCF-GFP mono- cultures (Figure 15). Quantitation of cellular GFP fluorescence was performed using a plate-reader, at the appropriate wavelength (See Materials and Methods). At 500 μM Doxycycline, MCF7-GFP cells in co- culture were ~5-fold more sensitive, than those in mono-cultures. However, the increased sensitivity of MCF7-GFP cells in co-culture was first observed at 100 μM, but not at 50 μM. Doxycycline also begins to inhibit overall protein synthesis in mammalian cells, in the range of 100 μM to 1 mM [29]. However, this effect is likely secondary to mitochondrial ATP- depletion (IC-50 = 50 μM) [30]. Therefore, Doxycycline may effectively target both “mito-stemness” and “ribo-stemness” features, by inhibiting both i) mitochondrial protein synthesis and ii) overall protein synthesis.

Representative images of these MCF7-fibroblast co- cultures and their differential sensitivity to Doxycycline are shown in Figure 16. Note the progressive reductions in GFP-fluorescence.

Quantitatively similar results were obtained with Azithromycin, another well-established inhibitor of mitochondrial biogenesis (Figure 17). Azithromycin preferentially targeted MCF7-GFP cells, in co-culture with fibroblasts. At 500 μM Azithromycin, MCF7-GFP cells in co-culture were ~8-fold more sensitive, than those in mono-cultures.

These pharmacological data (representing enhanced sensitivity to mitochondrial protein translation inhibitors) are consistent with the increases in mitochondrial biogenesis that we observed by proteomics analysis in MCF7-fibroblast co-cultures.

Cell culture

Cell culture experiments were carried out using human skin fibroblasts immortalized with the human telomerase reverse transcriptase (hTERT-BJ1 cells) and human MCF7 breast cancer cells. hTERT-BJ1 fibroblasts and MCF7 cells were maintained in complete media: DMEM (D6546, Sigma) supplemented with 10% fetal bovine serum (F7524, Sigma), 100 units/ml of penicillin, 100 μg/ml, streptomycin (P0781, Sigma) and 1% Glutamax (#35050087, Life Technologies). For all experiments, cells were incubated at 37°C in a humidified atmosphere containing 5% CO2.

Co-culture versus mixed cell populations

MCF7 cells and fibroblasts were co-cultured in the presence of Nu-Serum, essentially as we previously described [16–18]. Two million MCF7 cells were co-cultured in a regular 15 cm dish with two million hTERT-BJ1 fibroblasts, seeded in a 1:1 ratio. Likewise, the same numbers of either MCF7 cells or hTERT-BJ1 fibroblasts were seeded separately in 15 cm dishes as monocultures. After 72h of cell culture cells were lysed in RIPA lysis buffer (R0278, Sigma) containing proteinase inhibitors (05 892 970 001, Roche) and kept at 4° C for 20 minutes with rotation. Lysates were cleared by centrifugation for 10 minutes at 10,000 x g and supernatants were collected. The protein concentration of the lysates was determined by using the BCA protein assay kit (23225, Pierce). Briefly, four µg of MCF7- hTERT-BJ1 co-culture lysate or two µg of hTERT-BJ1 monoculture lysate mixed with 2 µg of MCF7 monoculture protein lysate (4 µg of protein in total) were submitted to the CRUK Proteomics Core Facility, for label-free proteomic analysis. Proteomics and statistical analyses were carried out on a fee-for-service basis by Dr. Duncan Smith and his colleagues, at the Proteomics Core Facility at the Cancer Research UK Manchester Institute, University of Manchester.

Label-free proteomics analysis

Validation with transcriptional profiling of breast cancer patient samples

To directly establish the clinical relevance of our findings from proteomics analysis of MCF7-fibroblast co-cultures, we re-analyzed the publically-available transcriptional profiles [26] of epithelial breast cancer cells and adjacent tumor stromal cells that were physically separated by laser-capture microdissection (from N=28 human breast cancer patients).

Validation with Kaplan-Meier (K-M) analyses of breast cancer patient samples

To perform K-M analysis of mitochondrial gene transcripts, we used an open-access online survival analysis tool to interrogate publically available microarray data from up to 3,455 breast cancer patients [27]. This allowed us to determine their prognostic value [28]. For this purpose, we primarily analyzed data from ER(+) patients that were LN(+) at diagnosis and were of the luminal A sub-type, that were primarily treated with Tamoxifen and not other chemotherapy (N=152 patients). In this group, 100% the patients received some form of hormonal therapy and ~95% of them received Tamoxifen. Biased and outlier array data were excluded from the analysis. This allowed us to identify mitochondrial gene transcripts, with significant prognostic value. Hazard-ratios were calculated, at the best auto-selected cut-off, and p- values were calculated using the logrank test and plotted in R. K-M curves were also generated online using the K-M-plotter (as high-resolution TIFF files), using univariate analysis:

http://kmplot.com/analysis/index.php?p=service&cancer=breast

This allowed us to directly perform in silico validation of these mitochondrial biomarker candidates. The multi- gene classifier function of the program was used to test the prognostic value of short mitochondrial gene signatures, using the mean expression of the selected probes.

Validation via drug treatment of fluorescently- labeled MCF7-fibroblast co-cultures

hTERT-BJ1-RFP cells and MCF7-GFP cells were first generated by stable transduction with lenti-viral vectors. Then, these cells were plated into clear flat bottom 96- well black microplates (Corning; CLS3603). Mono- cultures were generated by plating either hTERT-BJ1- RFP (8,000 cells/well) or MCF7-GFP cells (8,000 cells/well). In parallel, co-cultures were generated by co-plating hTERT-BJ1-RFP (6,000 cells/well) with MCF7-GFP (2,000 cells/well). The next day, the plates were treated with either Doxycycline or Azithromycin (or vehicle controls) and were incubated for 72 hours at 37^oC. The cell culture plates were then read with a plate-reader at 545/590 nm (for RFP) and at 490/510 nm (for GFP). Background readings were subtracted from the fluorescent signal and data were then normalized to controls.