1. Cancer Biology
Download icon

Resistance to different anthracycline chemotherapeutics elicits distinct and actionable primary metabolic dependencies in breast cancer

  1. Shawn McGuirk
  2. Yannick Audet-Delage
  3. Matthew G Annis
  4. Yibo Xue
  5. Mathieu Vernier
  6. Kaiqiong Zhao
  7. Catherine St-Louis
  8. Lucía Minarrieta
  9. David A Patten
  10. Geneviève Morin
  11. Celia MT Greenwood
  12. Vincent Giguère
  13. Sidong Huang
  14. Peter M Siegel
  15. Julie St-Pierre  Is a corresponding author
  1. Department of Biochemistry, Faculty of Medicine, McGill University, Canada
  2. Goodman Cancer Research Centre, McGill University, Canada
  3. Department of Biochemistry, Microbiology, and Immunology, University of Ottawa, Canada
  4. Ottawa Institute of Systems Biology, Canada
  5. Department of Medicine, Faculty of Medicine, McGill University, Canada
  6. Department of Epidemiology, Biostatistics and Occupational Health, McGill University, Canada
  7. Lady Davis Institute, Jewish General Hospital, Canada
  8. Department of Human Genetics, McGill University, Canada
  9. Gerald Bronfman Department of Oncology, Canada
Research Article
  • Cited 0
  • Views 764
  • Annotations
Cite this article as: eLife 2021;10:e65150 doi: 10.7554/eLife.65150

Abstract

Chemotherapy resistance is a critical barrier in cancer treatment. Metabolic adaptations have been shown to fuel therapy resistance; however, little is known regarding the generality of these changes and whether specific therapies elicit unique metabolic alterations. Using a combination of metabolomics, transcriptomics, and functional genomics, we show that two anthracyclines, doxorubicin and epirubicin, elicit distinct primary metabolic vulnerabilities in human breast cancer cells. Doxorubicin-resistant cells rely on glutamine to drive oxidative phosphorylation and de novo glutathione synthesis, while epirubicin-resistant cells display markedly increased bioenergetic capacity and mitochondrial ATP production. The dependence on these distinct metabolic adaptations is revealed by the increased sensitivity of doxorubicin-resistant cells and tumor xenografts to buthionine sulfoximine (BSO), a drug that interferes with glutathione synthesis, compared with epirubicin-resistant counterparts that are more sensitive to the biguanide phenformin. Overall, our work reveals that metabolic adaptations can vary with therapeutics and that these metabolic dependencies can be exploited as a targeted approach to treat chemotherapy-resistant breast cancer.

Introduction

Therapeutic resistance is a central problem in the clinical treatment of cancer. The incidence of breast cancer has risen to over one million new cases per year worldwide, where 20–30% of cases are diagnosed at an advanced or metastatic stage and another 30% recur or develop metastases (Siegel et al., 2018; Murray et al., 2012). While both adjuvant and neoadjuvant therapies have proven effective to improve patient outcomes, not all patients respond to the same therapeutics. Furthermore, drug resistance can manifest within months of treatment and is believed to cause treatment failure in over 90% of metastatic cancers (Garrett and Arteaga, 2011; Longley and Johnston, 2005). Consequently, due to intrinsic or acquired resistance, breast cancer patients often suffer disease progression despite drug treatment (Murphy and Seidman, 2009; Moreno-Aspitia and Perez, 2009).

In the absence of targeted therapies, chemotherapy is a standard-of-care treatment for many aggressive breast cancers (Lebert et al., 2018). While this is efficient at killing fast-growing cells, it can also select for resistant cells or elicit adaptations that confer resistance in surviving populations. These may include genetic modulation of mechanisms that decrease intracellular drug concentration, like drug export through the ATP-binding cassette (ABC) transporter family (Hembruff et al., 2008) or lysosomal clearance (Guo et al., 2016), as well as adaptations that minimize or overcome therapy-associated insults like DNA damage or reactive oxygen species (ROS) (Morandi and Indraccolo, 2017). Importantly, therapeutic agents elicit diverse resistance-conferring adaptations both across tumor subtypes and within tumors due to genetic and metabolic heterogeneity (Caro et al., 2012; Viale et al., 2015).

Several recent reviews have emphasized the importance of metabolic adaptations in driving or supporting drug resistance (Morandi and Indraccolo, 2017; Viale et al., 2015; Wolf, 2014; Bosc et al., 2017; Ashton et al., 2018). Although glycolysis is likely to remain favored in resistant cancers undergoing hypoxia or with defective mitochondria (Xu et al., 2005; Zhou et al., 2012), increased reliance on mitochondrial energy metabolism and oxidative phosphorylation has been identified as a distinctive characteristic of drug resistance (Bosc et al., 2017) being central to therapeutic resistance in ovarian (Matassa et al., 2016), pancreatic (Viale et al., 2014), colon (Vellinga et al., 2015), prostate (Ippolito et al., 2016), melanoma (Vazquez et al., 2013), and breast (Lee et al., 2017) cancers, as well as large B cell lymphoma (Caro et al., 2012) and acute (Farge et al., 2017) or chronic (Kuntz et al., 2017) myeloid leukemia.

Despite these advances in our understanding of the metabolic status of treatment-resistant cancers, little is known about the impact of different therapeutic drugs on the metabolic status of a given cancer. Addressing this knowledge gap is important, as numerous monotherapy and combination therapy regimens are often available to treat each patient. Here, we show that breast cancer cells resistant to either doxorubicin or epirubicin, two anthracycline drugs that are used interchangeably for breast cancer treatment (Mao et al., 2019), rely on distinct primary metabolic processes, and that exploiting these dependencies may impair the growth of treatment resistant breast cancers.

Results

Doxorubicin-resistant and epirubicin-resistant breast cancer cells display distinct global metabolic alterations

As experimental models, we used well-established and published models of breast cancer therapeutic resistance (Hembruff et al., 2008; Guo et al., 2016; Veitch et al., 2009; Heibein et al., 2012). Briefly, these models were generated from human MCF-7 breast cancer cells, adapted to increasing concentrations of either doxorubicin or epirubicin to a maximum tolerated dose of 98.1 nM (DoxR cells) or 852 nM (EpiR cells), respectively, in a stepwise manner and over several months (Hembruff et al., 2008; Figure 1a). Parental control cells (Control cells) were maintained in DMSO throughout the extensive selection process (Hembruff et al., 2008; Figure 1a). At these maximally tolerated doses, it has been shown that resistance is not simply linked to drug exclusion by the cells. Indeed, augmenting intracellular drug levels by inhibiting ABC transporter activity in DoxR or EpiR has little effect on cell survival, highlighting the importance of adaptation mechanisms separate from that of the ABC transporters and independent of drug concentration (Hembruff et al., 2008).

Figure 1 with 1 supplement see all
Transcriptomic and metabolomic analyses uncover distinct metabolic alterations in doxorubicin and epirubicin resistant breast cancer cells, compared to anthracycline-sensitive Control cells.

(a) Model detailing stepwise generation of Control, DoxR, and EpiR breast cancer cells, as previously described (Hembruff et al., 2008). (b) Viable cell number of anthracycline-resistant cells in the presence of drugs and anthracycline-treated Control cells compared to untreated Control (DMSO). N = 4, ***p<0.001 Control vs Control +Dox, ##p<0.01 ###p<0.001 DoxR vs Control +Dox, †††p<0.001 Control vs Control +Epi, ‡p<0.05 ‡‡‡p<0.001 EpiR vs Control +Epi (two-way ANOVA, Tukey’s multiple comparison test). (c) Viability of anthracycline-resistant cells and treated or untreated (DMSO) Control cells after 7 days growth. N = 4, ***p<0.001 (paired Student’s t-test). (d) Venn diagram of differentially expressed genes in DoxR or EpiR cells compared to Control cells. Legend: green (differentially expressed in DoxR only), red (differentially expressed in EpiR only), and white (differentially expressed in both). N = 3. (e) Principal component analysis of global gene expression profiling between Control, DoxR, and EpiR cells. N = 3. (f) Gene set enrichment analysis (GSEA) of DoxR and EpiR cells compared to Control cells (KEGG database, N = 3). Data are shown as normalized enrichment score (NES) for DoxR and EpiR, where color designates the p-value associated with each enrichment (from yellow at p=0.1 to red at p=0.0001) and where white bars designate non-significant enrichments (n/s, p>0.1). (g) Partial least squares discriminant analysis of metabolite profile data between Control, DoxR, and EpiR cells. N = 6, p=0.008 (1000 permutations). (h) Volcano plot of metabolite profile of DoxR and EpiR cells compared to Control cells (N = 6). Significant features (p<0.05, paired Student’s t-test) highlighted. (i) Integrated metabolic network analysis of DoxR (left) and EpiR (right) cells compared to Control cells. Metabolites are represented by nodes, with p-value represented by node size. Enzymes are represented by edges, with p-value of gene expression represented by edge thickness. Direction and magnitude of fold changes in gene expression and metabolite abundance are represented on a yellow (depleted in resistant) to red (enriched in resistant) color scale. Major pathways are highlighted in shaded areas. (j) Model detailing shRNA screen targeting 1215 drug target genes. Changes in shRNA barcode abundance were measured after 7 days, and viable gene targets were ranked by p-value and fold change of target shRNA abundance. Depleted shRNA were considered cytotoxic or cytostatic to resistant cells in the presence of drug, while enriched barcodes were considered to promote cell proliferation or survival. (k) (left) Top five metabolic pathways identified by Gene Set Enrichment Analysis (KEGG database) of ranked shRNA gene targets depleted in DoxR (top) and EpiR (bottom) cells. (right) GSEA plots detailing enrichment of key metabolic pathways identified in DoxR cells (Glutathione Metabolism, top) or EpiR (Oxidative Phosphorylation, bottom). All data presented as averages ± S.E.M. (b-c).

DoxR and EpiR cells maintained in culture with a stable dose of their respective drug grew slower than Control cells without treatment (Figure 1b), and acute exposure of Control cells to 98.1 nM of doxorubicin had a cytostatic effect, while treatment with 852 nM epirubicin was cytotoxic (Figure 1b,c). Finally, we verified that DoxR and EpiR cells are stably resistant, retaining their level of resistance even after a 7-week drug holiday (Figure 1—figure supplement 1a).

In line with the chemical similarity and mechanism of action of both anthracycline drugs—nucleic acid intercalation, topoisomerase II inhibition leading to double-strand DNA breaks and apoptosis, and production of ROS (McGowan et al., 2017)—there was a considerable overlap in the signature of differentially expressed genes between DoxR and EpiR cells when compared to Control cells (Figure 1d, 56% overlap). These gene expression signatures also resemble that of human breast cancer tumors after epirubicin-based therapy where biopsies were taken prior to and after treatment, consisting of four cycles of epirubicin and cyclophosphamide, followed by four cycles of docetaxel (Figure 1—figure supplement 1b,c; GSE43816) (Gruosso et al., 2016). While the surviving tumor fractions in this dataset cannot be assumed to be entirely treatment-resistant, they represent a close clinical approximate of the adaptations that may occur after any anthracycline-based therapy, or as a result of selected advantages that may promote resistance. Gene expression profiles of both DoxR and EpiR cells significantly overlapped with that of post-treatment biopsies, when compared to the overlap expected from random shuffling (Figure 1—figure supplement 1b,c). The overlap in the gene expression profiles between DoxR and EpiR cells as well as that between the respective expression profile of DoxR and EpiR cells with breast cancer patients post-treatment biopsies highlight common mechanisms of adaptation to anthracyclines as well as the clinical relevance of the DoxR and EpiR breast cancer cell models.

Despite these similarities, we noted that 44% of differentially expressed genes in DoxR and EpiR cells were distinct (Figure 1d). A principal component analysis of their global gene expression profiles accordingly produced three discrete groups, with DoxR and EpiR diverging from each other in the second principal component (Figure 1e). Gene Set Enrichment Analysis (GSEA) further revealed that, although both DoxR and EpiR display enrichment in drug clearance pathways (ABC transporters, lysosome) and depletion of pathways supporting proliferation (cell cycle, one carbon pool by folate) compared to Control cells, several metabolic pathways are specifically enriched in DoxR (pyruvate metabolism, glutathione metabolism) or EpiR (nicotinate and nicotinamide metabolism, alanine, aspartate, and glutamate metabolism) (Figure 1f, Figure 1—figure supplement 1d).

Analysis of the global metabolite profiles of Control, EpiR, and DoxR also indicated significant differences between all three models (Figure 1g,h). To visualize the distinct metabolic adaptations that occur in resistance to either doxorubicin or epirubicin, we performed an integrated transcriptional and metabolic network analysis (Sergushichev et al., 2016) of DoxR and EpiR cells compared to parental Control cells (Figure 1i). DoxR cells displayed increased expression of glutathione pathway genes as well as elevated levels of key glutathione metabolites (glutamate, cysteine, glutathione disulfide), indicating a likely role of this pathway in overcoming doxorubicin-induced oxidative stress (Figure 1iPilco-Ferreto and Calaf, 2016). EpiR cells showed alterations in pathways linked to pyruvate and glutamate metabolism, in particular through elevated transamination reactions linking glutamate, alanine, and aspartate (Figure 1i).

In parallel to these transcriptomics and metabolomics analyses, we performed pooled shRNA screens focused on 1215 druggable genes to identify primary vulnerabilities of DoxR and EpiR cells (Figure 1j). In this screen, the enrichment or depletion of shRNA barcodes in each cell system is measured over 7 days (post-integration of shRNAs); constructs whose barcodes are depleted over this span indicate gene targets whose knockdown impairs growth and/or cell viability. Analyzing depleted constructs via GSEA determined that DoxR cells were particularly sensitive to knockdown of glutathione metabolism genes (GSR and GPX family genes), while EpiR cells were vulnerable to suppression of genes involved in oxidative phosphorylation (NDUF and SDH family genes) and methionine metabolism (MAT1A, MAT2A, MAT2B, BHMT, DNMT1; Figure 1k and Figure 1—figure supplement 1e). In agreement with previous work (Veitch et al., 2009; Heibein et al., 2012), both DoxR and EpiR cells were sensitive to knockdown of aldo-keto reductase family genes, which are represented in the steroid hormone biosynthesis pathway (Figure 1k). Overall, results from the shRNA screens are consistent with our integrated transcriptional and metabolic network analysis (Figure 1i) and highlight distinct metabolic vulnerabilities supporting epirubicin and doxorubicin resistance.

Doxorubicin-resistant breast cancer cells display altered glucose and glutamine metabolism

To gain greater understanding of the reliance of anthracycline-resistant breast cancer cells on the pathways identified in the integrated analyses above, we confirmed gene expression profiles by RT-qPCR (Figure 2a) and performed stable isotope tracer analyses of [U-13C]-glucose (Figure 2b–d, Figure 2—figure supplement 1a,b) and [U-13C]-glutamine (Figure 2e,f, Figure 2—figure supplement 2a,b). The full kinetics of all stable isotope tracing experiments are shown in Figure 2—figure supplements 1 and 2, in accordance with the standard practice in the field (Buescher et al., 2015).

Figure 2 with 3 supplements see all
Doxorubicin-resistant breast cancer cells fuel anaplerotic metabolism by altering glucose and glutamine metabolism.

(a) Relative expression of pyruvate metabolism, citric acid cycle, and glutamine and glutathione metabolism genes in DoxR and EpiR compared to Control cells. N = 3–6, *p<0.05 **p<0.01 ***p<0.001 resistant versus Control cells (paired Student’s t-test). (b) Stable isotope tracing diagram for [U-13C]-Glucose through glycolysis and into the citric acid cycle via pyruvate dehydrogenase (PDH, black) and adjacent pathways (gray). (c) [U-13C]-Glucose tracing into the citric acid cycle via PDH (citrate, malate, fumarate, and aspartate m + 2, 30 min tracer) in Control, DoxR, and EpiR cells expressed as fractional enrichment. N = 4, *p<0.05 resistant versus Control cells (paired Student’s t-test). (d) [U-13C]-Glucose tracing to glutamate (m + 2) via PDH and into the citric acid cycle via PC or ME1/2 activity (citrate, malate, fumarate, and aspartate m + 3, 30 min tracer) in Control, DoxR, and EpiR cells expressed as fractional enrichment. N = 4, *p<0.05 **p<0.01 resistant versus Control cells (paired Student’s t-test). (e) Stable isotope tracing diagram for [U-13C]-Glutamine into the citric acid cycle, glutathione synthesis, and adjacent pathways. Reductive carboxylation pathway shown in gray. (f) [U-13C]-Glutamine tracing to glutamate and into the citric cycle (2KG m + 5, citrate m + 4 and m + 5, 60 min tracer) or through to glutathione (GSH m + 3,5 and GSSG m + 3,5,6,8,10, 4 hr tracer) in Control, DoxR, and EpiR cells expressed as fractional enrichment. N = 4, *p<0.05 ***p<0.001 resistant versus Control cells (paired Student’s t-test). (g) Total intracellular glutathione concentration in Control, DoxR and EpiR cells. N = 4, *p<0.05 resistant vs Control cells, #p<0.05 DoxR vs EpiR cells (paired Student’s t-test). (h) Fold change in GSH:GSSG ratio of DoxR and EpiR cells compared to Control cells. N = 4, **p<0.01 resistant vs Control, #p<0.05 DoxR vs EpiR cells (paired Student’s t-test). (i) Fold change in NADH/NAD and NADPH/NAD ratio of DoxR and EpiR cells compared to Control cells. N = 6, *p<0.05 **p<0.01 resistant versus Control cells (paired Student’s t-test). Data are shown on a log2 scale. (j) Relative expression of oxidative response genes in DoxR and EpiR cells compared to Control cells. N = 4–7, *p<0.05 **p<0.01 ***p<0.001 resistant vs Control cells (paired Student’s t-test). (k) Fold change of ROS signal in DoxR and EpiR cells compared to Control cells, detected by CM-H2DCFDA. N = 3, **p<0.01 resistant vs Control cells, #p<0.05 DoxR vs EpiR cells (paired Student’s t-test). (l) Fold change of ROS signal in DoxR and EpiR compared to Control cells, after 30-min treatment with 0.03% H2O2. N = 3, *p<0.05 **p<0.01 resistant vs Control cells (paired Student’s t-test). All data presented as averages ± S.E.M.

DoxR cells exhibited significantly increased expression of anaplerotic metabolism genes (PC, ME1, ME2), glutamine metabolism genes (SLC1A5, GLS, GLUL), and, markedly, glutathione metabolism genes (GCLC, GCLM, GSS, GSR) compared to Control cells (Figure 2a). Accordingly, kinetic tracing of glucose carbons showed that while Control and EpiR cells replenish their pools of citric acid cycle intermediates (citrate and malate m + 2) principally through pyruvate dehydrogenase (PDH), DoxR cells significantly favor anaplerotic pyruvate metabolism (citrate, malate, and fumarate m + 3) through pyruvate carboxylase (PC) and/or malic enzymes (ME1/2; Figure 2d and Figure 2—figure supplement 1a,b). Interestingly, glutamate, alanine, and serine synthesis from glucose was decreased in DoxR cells compared to Control cells (Figure 2d and Figure 2—figure supplement 1a,b). Kinetic tracing further showed that glutamine metabolism was enriched in DoxR cells compared to Control and EpiR cells, evidenced by increased labeling to glutamate, α-ketoglutarate, and citrate (Figure 2e,f and Figure 2—figure supplement 2a,b). Reductive carboxylation of glutamine was particularly increased in DoxR cells compared to Control and EpiR cells, as indicated by a significant increase in m + 5 labeling to citrate (Figre 2e,f and Figure 2—figure supplement 2a,b). More strikingly, DoxR cells largely favored the use of glutamine carbons for de novo production of glutathione, evidenced by a fourfold enrichment of labeling to GSH and GSSG compared to both Control and EpiR cells (GSH m + 5, GSSG m + 5,10; Figure 2f and Figure 2—figure supplement 2a,b). This may also be driven in part by exchange of glutamate for cystine through the glutamate/cystine antiporter system (Habib et al., 2015) as DoxR cells were found to export significantly higher levels of glutamate than both Control and EpiR cells (Figure 2—figure supplement 3a).

In agreement with these stable isotope tracing results, DoxR cells displayed a significantly higher total intracellular glutathione concentration than EpiR cells, and both had higher values than Control cells (Figure 2g). Both resistant lines had a significantly higher GSH:GSSG ratio than Control cells, with DoxR cells displaying an even greater enrichment of reduced glutathione compared to EpiR cells (Figure 2h). The elevated glutathione metabolism in DoxR cells is further supported by their decreased NADH:NAD and NADPH:NADP ratios (Figure 2i) compared to Control cells, as the reduced equivalent NADPH is required for the reduction of GSSG to GSH through glutathione reductase (GSR), whose expression was increased in DoxR but not EpiR cells compared to Control cells (Figure 2a). NADPH levels may also be depleted through the reductive carboxylation of glutamine, as this pathway relies on the NADPH-dependent isocitrate dehydrogenases (IDH1/2); the expression of IDH1 is significantly increased in DoxR cells compared to both EpiR and Control cells, in line with their increased engagement of this pathway (Figure 2a,f). Conversely, EpiR cells displayed elevated NADH:NAD and NADPH:NADP ratios compared to Control cells (Figure 2i). EpiR cells also did not display any significant changes in either anaplerotic pyruvate metabolism (Figure 2d and Figure 2—figure supplement 1a,b), reductive carboxylation of glutamine, or de novo glutathione synthesis from glutamine (Figure 2f and Figure 2—figure supplement 2a,b) when compared to Control cells, even though they showed increased expression of glutamine metabolism genes (Figure 2a).

Given that both DoxR and EpiR cells had a higher GSH:GSSG ratio compared to Control cells, we further sought to determine whether these anthracycline-resistant cells displayed markers of elevated oxidative stress response. Both resistant lines had increased expression of key antioxidant genes compared to Control cells, with DoxR cells showing a higher increase in expression of many genes (NFE2L2, NQO1, SOD2, PRDX5, CAT) compared with EpiR cells (Figure 2j). Through CM-H2DCFDA experiments, we also determined that DoxR cells have decreased ROS signals compared to EpiR cells, both at baseline (Figure 2k) and after H2O2 treatment (Figure 2l), while both resistant cells had lower ROS signals than Control cells. DoxR cells likely support this greater engagement of oxidative stress response through their increased de novo glutathione synthesis from glutamine, in agreement with findings from the shRNA screen showing that glutathione metabolism is a specific vulnerability in this model (Figure 1k).

Epirubicin-resistant breast cancer cells display increased oxidative bioenergetic capacity

Given that oxidative phosphorylation was identified as a specific vulnerability for EpiR cells in the shRNA screen (Figure 1k), we sought to determine the bioenergetic profile of these cells. Using the Seahorse platform, we found that basal and maximal oxygen consumption rates were significantly increased in EpiR cells, but not DoxR cells, compared to Control cells, whereas extracellular acidification rates were not significantly different amongst the three cell lines, albeit slightly lower in EpiR cells (Figure 3a–c). From these data, we further extrapolated the total rates of ATP production (JATP) as well as the fraction of ATP generated through glycolysis (JATPglyc) or oxidative phosphorylation (JATPox) using published assumptions and algorithms (Mookerjee et al., 2017). Aligning with their increased respiration, a greater proportion of ATP produced in EpiR cells was linked to oxidative phosphorylation — 69%, compared to 62% and 60% for Control and DoxR cells respectively (Figure 3d,e). However, the total ATP production rate in EpiR cells was not significantly different than Control cells, as their glycolyic ATP production was proportionally lowered (Figure 3d). EpiR cells also acquired a significant increase in total bioenergetic capacity (25%) compared to both Control and DoxR cells (Figure 3f,g), largely driven by increased oxidative capacity (Figure 3f) and, consequently, EpiR cells have a greater reserve capacity for generation of ATP through oxidative phosphorylation (Figure 3h). This increased capacity is further supported by elevated mitochondrial volume in EpiR cells compared to Control or DoxR cells (Figure 3i).

Figure 3 with 1 supplement see all
PGC-1α is overexpressed and elevates OXPHOS capacity in epirubicin-resistant cells but is essential for sustaining growth and survival in both doxorubicin and epirubicin resistant breast cancer cells.

(a) Analysis of relative basal, leak (oligomycin), maximum (FCCP), and non-mitochondrial (rotenone and myxothiazol) respiration, as well as after addition of monensin, in Control, DoxR, and EpiR cells. All values normalized to basal respiration in Control cells. N = 9. (b) Quantification of coupled, uncoupled, and total Oxygen Consumption Rate (OCR) in Control, DoxR, and EpiR cells. N = 9, **p<0.01 uncoupled #p<0.05 total, resistant vs Control cells (paired Student’s t-test). (c) Quantification of basal Extracellular Acidification Rate (ECAR) in Control, DoxR, and EpiR cells. N = 9. (d) Quantification of ATP production rate (JATP) from oxidative phosphorylation (JATPox) or glycolysis (JATPglyc) in Control, DoxR, and EpiR cells. N = 9, *p<0.05 JATPox #p<0.05 JATPglyc, resistant vs Control cells (paired Student’s t-test). (e) Proportion of ATP produced by OXPHOS in Control, DoxR, and EpiR cells under basal conditions and at peak bioenergetic capacity (Max). N = 9, **p<0.01 ***p<0.001 resistant vs Control cells, #p<0.05 ##p<0.01 ###p<0.001 Max vs Basal (paired Student’s t-test). (f) Bioenergetic space plots of Control, DoxR, and EpiR cells. Solid points represent basal JATPglyc and JATPox values, hollow points represent theoretical maximums for JATPox (FCCP) and JATPglyc (rotenone, myxothiazol, monensin). Dotted line arrows’ length represents flexibility of ATP production within maximum boundaries (solid lines). Area under maximum boundaries represents the bioenergetic capacity. N = 9. (g) Fold change in bioenergetic capacity of DoxR and EpiR cells compared to Control cells. N = 9, *p<0.05 resistant vs Control cells (paired Student’s t-test). (h) Fraction of bioenergetic capacity used under basal conditions in Control, DoxR, and EpiR cells. N = 9, *p<0.05 resistant vs Control cells (paired Student’s t-test). (i) Quantification of mitochondrial volume as a percentage of total cytoplasmic volume, in Control, DoxR, and EpiR cells. N = 38, **p<0.01 resistant vs Control cells (Kruskal-Wallis test and Dunn’s multiple comparisons test). Data presented as a box plot. (j) Relative expression of PPARGC1A and PPARGC1B mRNA in DoxR and EpiR cells compared to Control cells. N = 7, **p<0.01 ***p<0.001 resistant vs Control cells (paired Student’s t-test). (k) Immunoblots of PGC-1β, PGC-1α, and Actin protein expression in Control, DoxR, and EpiR cells. (l) Relative expression of PPARGC1A and PPARGC1B mRNA in DoxR and EpiR 3 days after double siRNA knockdown of PGC-1α and PGC-1β, compared to control siRNA. N = 5, *p<0.05 **p<0.01 ***p<0.001 siPGC-1α/β vs siCTL (paired Student’s t-test). (m) Growth of DoxR and EpiR in the presence of anthracyclines (dox 98.1 nM or epi 852 nM) with double siRNA knockdown of PGC-1α and PGC-1β or control siRNA. N = 5, **p<0.01 ***p<0.001 DoxR siPGC-1α/β vs DoxR siCTL, #p<0.05 ##p<0.01 EpiR siPGC-1α/β vs EpiR siCTL (paired Student’s t-test). (n) Cell viability at day 6 of the growth curve shown in d. N = 5, **p<0.01 ***p<0.001 siPGC-1α/β vs siCTL (paired Student’s t-test). (o) Relative expression of glutathione metabolism genes 3 days after double siRNA knockdown of PGC-1α and PGC-1β, compared to control siRNA. N = 5, **p<0.01 ***p<0.001 siPGC-1α/β vs siCTL (paired Student’s t-test). All data presented as averages ± S.E.M.

Given the importance of peroxisome proliferator-activated receptor gamma coactivator 1-alpha (PGC-1α) in mitochondrial biogenesis (Wu et al., 1999), OXPHOS (Mootha et al., 2003), glutamine metabolism (McGuirk et al., 2013), and glutathione synthesis (Guo et al., 2018), we hypothesized that it may play a role in the metabolic adaptations that support doxorubicin and/or epirubicin resistance. Indeed, PPARGC1A expression at the mRNA (Figure 3j) and protein (Figure 3k) levels was markedly increased in EpiR compared to Control cells, while there was a modest and non-significant increase in DoxR compared to Control cells. mRNA expression of ERRα (ESRRA), the central transcription partner of PGC-1α, was also increased in DoxR and EpiR cells compared to Control cells (Figure 3—figure supplement 1aGiguère, 2008). Furthermore, both PGC-1α and ERRα were found to be enriched at the promoters of key metabolic and resistance-associated genes whose expression is modulated in DoxR or EpiR compared to Control. Specifically, they were enriched at the promoters of genes central to glutathione metabolism (GSS, GSR), oxidative response (NQO1, CAT, NFE2L2, HMOX2), drug efflux (ABCB1, ABCC1), anabolic pyruvate metabolism (PC), as well as AKR1C3 and FTH1, two genes that were previously shown to be involved in doxorubicin resistance (Figure 3—figure supplement 1b,cHembruff et al., 2008; Veitch et al., 2009). The promoter enrichment of PGC-1α is notably higher in EpiR cells, aligning with its elevated expression level in these cells (Figure 3—figure supplement 1b,c).

To test the biological significance of PGC-1α in resistant cells, we opted to knock down both PPARGC1A and PPARGC1B in DoxR and EpiR cells (Figure 3l) to avoid potential compensation between these two transcriptional coactivators as they regulate overlapping gene expression programs (Villena, 2015; Lin et al., 2005). Knockdown of both PPARGC1A and PPARGC1B significantly abrogated the growth (Figure 3m) and survival (Figure 3n) of both DoxR and EpiR cells, compared to siCTL. Also, knockdown of PGC-1s significantly reduced the expression of key glutathione metabolism genes in DoxR (GCLM, GSS, GSR) and EpiR (GCLM) cells, compared to siCTL (Figure 3o). Collectively, these data suggest that PGC-1s play an important role in promoting resistance to doxorubicin and epirubicin by regulating the expression of genes that contribute to resistance to each drug.

Doxorubicin-resistant cells rely on glutamine for sustained ATP production

Given that glutamine is a key fuel for cancer cell growth (McGuirk et al., 2013) and that DoxR cells use glutamine for glutathione synthesis, we hypothesized that anthracycline-resistant cells may be particularly sensitive to glutamine withdrawal. The proliferation rates of all cell lines were severely affected by glutamine withdrawal. However, while Control cells could maintain a low level of proliferation, both DoxR and EpiR cells were unable to proliferate in the absence of glutamine (Figure 4a). This was accompanied by a small, but significant, increase in cell death, by 5% in DoxR and 10% in EpiR cells, compared to Control cells (Figure 4b). Comparatively, neither Control, DoxR, nor EpiR cells were significantly affected by glucose limitation (Figure 4—figure supplement 1a).

Figure 4 with 1 supplement see all
Doxorubicin-resistant breast cancer cells are reliant on glutamine to sustain their OXPHOS bioenergetic capacity.

(a) Growth of Control, DoxR, and EpiR cells in glutamine-rich (4 mM) or glutamine-deprived (0 mM) conditions. Cells were differentially seeded to reach a total of 200,000 cells at day 3 under normal growth conditions; media was then changed for glutamine-rich or glutamine-depleted media, daily (day 3–11). N = 3, *p<0.05 **p<0.01 Control cells 0 mM vs 4 mM, #p<0.05 ##p<0.01 DoxR cells 0 mM vs 4 mM, †p<0.05 ††p<0.001 EpiR cells 0 mM vs 4 mM (paired Student’s t-test). (b) Fold change in viability of Control, DoxR, and EpiR cells after 6 days glutamine withdrawal compared to glutamine-rich (4 mM) conditions. N = 3, *p<0.05 **p<0.01 0 mM vs 4 mM glutamine (paired Student’s t-test). (c) Seahorse analysis of basal, leak (oligomycin), maximum (FCCP), and non-mitochondrial (rotenone and myxothiazol) oxygen consumption rates, as well as after addition of monensin, in Control, DoxR, and EpiR cells after 4 hr glutamine withdrawal compared to glutamine-rich (4 mM) conditions, presented as relative to the basal respiration rate of Control cells in the presence of glutamine. N = 5, *p<0.05 **p<0.01 DoxR 0 mM Gln vs DoxR 4 mM Gln, #p<0.05 EpiR 0 mM Gln vs EpiR 4 mM Gln (paired Student’s t-test). (d) Fold change in ATP production by oxidative phosphorylation (JATPox) of Control, DoxR, and EpiR cells after 4 hr glutamine withdrawal compared to glutamine-rich (4 mM) conditions. N = 5, #p<0.05 0 mM vs 4 mM (paired Student’s t-test). (e) Fold change in ATP production from glycolysis (JATPglyc) of Control, DoxR, and EpiR cells after 4 hr glutamine withdrawal compared to glutamine-rich (4 mM) conditions. N = 5, ##p<0.01 0 mM vs 4 mM (paired Student’s t-test). (f) Fold change in total ATP production rate (JATP) of Control, DoxR, and EpiR cells after 4 hr glutamine withdrawal compared to glutamine-rich (4 mM) conditions. N = 5, *p<0.05 resistant vs Control cells (paired Student’s t-test). (g) Fold change in bioenergetic capacity of Control, DoxR, and EpiR cells after 4 hr glutamine withdrawal compared to glutamine-rich (4 mM) conditions. N = 5, *p<0.05 **p<0.01 resistant vs Control cells, ###p<0.001 0 mM vs 4 mM glutamine. (h-j) Bioenergetic space plots of Control (h), DoxR (i), and EpiR (j) cells after 4 hr glutamine withdrawal compared to glutamine-rich (4 mM) conditions. Solid points represent actual JATPglyc and JATPox values, hollow points represent theoretical maximums for JATPox (FCCP) and JATPglyc (rotenone, myxothiazol, monensin). Dotted line arrows’ length represents flexibility of ATP production within maximum boundaries (solid lines). Area under maximum boundaries represents the bioenergetic capacity. N = 5. All data presented as averages ± S.E.M.

To further quantify the impact of glutamine withdrawal on cellular metabolism, we assessed the bioenergetics of resistant cells in the presence or absence of glutamine, through extrapolation of JATP rates from OCR and ECAR measurements (Figure 4c, Figure 4—figure supplement 1b). After 4 hr of glutamine withdrawal, all three cell lines showed a decrease in mitochondrial ATP production, with DoxR cells being the most affected (Figure 4d). This decrease in mitochondrial ATP production was compensated for by an increase in ATP production through glycolysis in EpiR and Control cells, but not DoxR cells (Figure 4e). This compensatory increase in glycolytic ATP production was sufficient to maintain total ATP production (mitochondria and glycolysis) in Control and EpiR cells, while the lack of compensatory glycolytic ATP production in DoxR cells led to a drop in their total ATP production (Figure 4f). The fact that DoxR cells were unable to increase glycolytic ATP production to compensate for diminished mitochondrial ATP production upon glutamine withdrawal may indicate a defect in glycolytic regulation or ATP sensing (Figure 4e–f).

We also measured the maximal mitochondrial bioenergetic capacity as well as the maximal glycolytic capacity of these cells (Figure 4h–jMookerjee et al., 2017). Strikingly, glutamine starvation entirely abrogated the spare oxidative capacity of DoxR cells (Figure 4i, box height relative to basal point). This significantly reduced the total bioenergetic capacity (mitochondrial and glycolytic capacities combined, Figure 4i box area, Figure 4g) of DoxR cells and their bioenergetic flexibility—the ability to dynamically shift glycolysis and OXPHOS rates while maintaining a constant ATP production rate (Figure 4i, dotted arrow length). Control and EpiR cells showed diminished oxidative capacity (box height) and increased glycolytic capacity (box width) as well as no change in flexibility after glutamine withdrawal (Figure 4h,j). However, the decrease in oxidative capacity in EpiR cells was greater than their increase in glycolytic capacity, leading to a reduced total bioenergetic capacity in the absence of glutamine (Figure 4g,j). The total bioenergetic capacity of EpiR cells remained nevertheless greater than that of DoxR cells (Figure 4g). Taken together, these data show that glutamine is specifically important for doxorubicin resistant breast cancer cells, not only for glutathione synthesis but also for mitochondrial ATP production.

Independently-derived models of doxorubicin- and epirubicin-resistance confirm their respective dependance on glutathione metabolism and oxidative phosphorylation

To independently confirm the distinct metabolic adaptations to doxorubicin and epirubicin of the well-characterized cells in this study (Hembruff et al., 2008), we generated new resistant cell lines from drug-naive MCF-7 cells over the course of 8 months, following a similar step-wise process and up to a common end-point dose of 100 nM doxorubicin (D100 cells) or epirubicin (E100 cells; Figure 5a). Matched parental control (Ctl) cells were maintained in 0.1% DMSO through parallel passages (Figure 5a). Importantly, while we sought to confirm the distinct metabolic adaptations to each drug, it was expected that there would be some variability between these independently derived resistant cells and the cells used in the rest of study given that cancer cells may engage different adaptation strategies to develop resistance to chemotherapy (Edwardson et al., 2013).

Figure 5 with 1 supplement see all
Independently derived resistant models confirm specific metabolic adaptations to doxorubicin and epirubicin.

(a) Model detailing generation of breast cancer cells resistant to increasing concentrations of anthracyclines, to a final stable extracellular concentration of 100 nM of doxorubicin (D100) or epirubicin (E100). Control cells (Ctl) were maintained in DMSO in parallel passages. (b) Relative expression of PPARGC1A and selected metabolic, glutathione, and oxidative response genes in D100 and E100 compared to Ctl. N = 3–6. *p<0.05 resistant vs Ctl cells (paired Student’s t-test). (c) [U-13C]-Glutamine tracing to glutamate and into the citric cycle (2 KG m + 5, citrate m + 4 and m + 5, malate m + 3, 2 hr tracer) or through to glutathione (GSH m + 5 and GSSG m + 5,10, 4 hr tracer) in Ctl, D100, and E100 expressed as fractional enrichment. N = 3–5, *p<0.05 **p<0.01 ***p<0.001 resistant vs Ctl cells (paired Student’s t-test). (d) Fold change of ROS signal in D100 and E100 cells compared to Ctl cells, detected by CM-H2DCFDA staining. N = 5, **p<0.01 resistant vs Control cells, #p<0.05 D100 vs E100 (paired Student’s t-test). (e) Fold change of ROS signal in D100 and E100 compared to Ctl cells, after 30 min treatment with 0.03% H2O2. N = 5, **p<0.01 resistant vs Control cells, #p<0.05 D100 vs E100 (paired Student’s t-test). (f) Quantification of ATP production (JATP) from oxidative phosphorylation (JATPox) or glycolysis (JATPglyc) in Ctl, D100, and E100. N = 5, *p<0.05 JATPox #p<0.05 JATPglyc †p<0.05 JATP, resistant vs Ctl cells (paired Student’s t-test). (g) Quantification of ATP production by OXPHOS (JATPox) in Ctl, D100 and E100 cells under basal conditions and at peak bioenergetic capacity (Max). N = 5, *p<0.05 resistant vs Ctl cells, ##p<0.01 E100 vs D100 (paired Student’s t-test). (h) Bioenergetic space plots of Ctl, D100, and E100. Solid points represent actual JATPglyc and JATPox values, hollow points represent maximums for JATPox (FCCP) and JATPglyc (rotenone, myxothiazol, monensin). Length of dotted line arrows represents flexibility of ATP production within maximum boundaries (solid lines). Area under maximum boundaries represents the bioenergetic capacity. N = 5. (i) Fold change in bioenergetic capacity of D100 and E100 compared to Ctl. N = 5 *p<0.05 **p<0.01 resistant vs Ctl cells (paired Student’s t-test). All data presented as averages ± S.E.M.

As seen in DoxR and EpiR cells (Figure 2), key oxidative response genes (NFE2L2, NQO1) as well as GLS and IDH1 were significantly overexpressed in both D100 and E100 cells, while only D100 cells exhibited significant overexpression of glutathione metabolism genes (GCLC, GCLM, GSS, GSR) compared to parental control (Ctl) cells (Figure 5b). Accordingly, stable isotope tracing analyses confirmed that D100 have significantly increased de novo synthesis of glutathione from glutamine compared to Ctl cells, whereas it was significantly decreased in E100 cells (Figure 5c and Figure 5—figure supplement 1a). D100 cells also displayed a lower ROS signal than E100 cells, both at baseline (Figure 5d) and after H2O2 treatment (Figure 5e), while both resistant cells had lower ROS signals than Ctl cells.

Similar to EpiR cells (Figure 3), E100 cells also had significantly higher expression of PPARGC1A (Figure 5b), along with increased oxygen consumption and extracellular acidification rates, increased total ATP production rates, increased basal and maximum oxidative ATP production rates (JATPox), as well as greater bioenergetic capacity compared with Ctl cells (Figure 5—figure supplement 1b,c and Figure 5f–i). Interestingly, and in contrast to DoxR cells, D100 cells displayed a significant increase in the expression of PPARGC1A (Figure 5b), as well as elevated maximum oxidative ATP production rate and bioenergetic capacity compared to Ctl (Figure 5g–i). Despite these differences in bioenergetics between the two cell models of doxorubicin resistance, these results confirm that epirubicin resistant cells display higher oxidative capacity than doxorubicin-resistant cells (Figure 5g–h).

Overall, these independently derived cell models of doxorubicin and epirubicin resistance broadly replicate the central findings of the manuscript, namely that doxorubicin-resistant cells have an elevated usage of glutamine for glutathione synthesis and that epirubicin-resistant cells display markedly increased OXPHOS capacity. Furthermore, given that the independently derived cell lines were selected to a common end-point dose of 100 nM of doxorubicin or epirubicin, these data further demonstrate that these metabolic adaptations are specific to the drug and not the dose.

Tailored metabolic adaptations underpinning resistance to doxorubicin and epirubicin lead to primary actionable vulnerabilities

Despite some similar mechanisms supporting resistance to doxorubicin and epirubicin in breast cancer cells, our results have thus far shown that drug-dependent dominant metabolic adaptations arise in resistant cells. Indeed, resistance to doxorubicin is linked to glutathione metabolism, whereas resistance to epirubicin is tied to enhanced mitochondrial bioenergetic capacity. To further demonstrate the importance of these tailored metabolic adaptations, we sought to determine whether these primary vulnerabilities are targetable.

Using the biguanide phenformin, we assessed if epirubicin-resistant cells are specifically sensitive to inhibition of OXPHOS. As expected, phenformin had a strong and dose-dependent effect on the growth of drug-naïve (Control and Ctl) breast cancer cells (Figure 6a,b). Epirubicin-resistant cells (EpiR and E100) were similarly vulnerable to phenformin in a dose-dependent manner, whereas doxorubicin-resistant (DoxR and D100) cells were only mildly responsive to the drug; both epirubicin-resistant cells were significantly more sensitive to phenformin than their doxorubicin-resistant counterparts (Figure 6a,b).

Tailored metabolic adaptations underpinning resistance to doxorubicin and epirubicin lead to primary actionable vulnerabilities in vitro and in vivo.

(a) Relative viable cell count of Control, DoxR, and EpiR cells after 3 days treatment with a combination of phenformin and their respective drug (DMSO, 98.1 nM doxorubicin, or 852 nM epirubicin). Data are shown as relative viable cell count of phenformin-treated cells compared to cells treated with vehicle (water). N = 4, *p<0.05 **p<0.01 ***p<0.001 resistant vs Control cells, #p<0.05 DoxR vs EpiR cells (paired Student’s t-test). (b) Relative viable cell count of Ctl, D100, and E100 cells after 3 days treatment with a combination of phenformin and their respective drug (DMSO, 100 nM doxorubicin, or 100 nM epirubicin). Data are shown as relative viable cell count of phenformin-treated cells compared to cells treated with vehicle (water). N = 4, **p<0.01 ***p<0.001 resistant vs Ctl cells, ***p<0.001 D100 vs E100 cells (paired Student’s t-test). (c) Relative viable cell count of Control, DoxR, and EpiR cells after 3 days treatment with a combination of buthionine sulfoximine (BSO) and their respective drug (DMSO, 98.1 nM doxorubicin, or 852 nM epirubicin) after 7 weeks of drug holiday. Data are shown as relative viable cell count of BSO-treated cells compared to cells treated with vehicle (water). N = 3, **p<0.01 ***p<0.001 resistant vs Control cells, ##p<0.01 ###p<0.001 DoxR vs EpiR cells (paired Student’s t-test). (d) Relative viable cell count of Ctl, D100, and E100 cells after 3 days treatment with a combination of BSO and their respective drug (DMSO, 100 nM doxorubicin, or 100 nM epirubicin). Data are shown as relative viable cell count of BSO-treated cells compared to cells treated with vehicle (water). N = 3, **p<0.01 resistant vs Control cells, #p<0.05 ##p<0.01 ###p<0.001 D100 vs E100 cells (paired Student’s t-test). (e) Picture of end-point DoxR and EpiR tumors after 70 days of growth in the opposing mammary fat pads of NOD Scid Gamma mice supplemented twice weekly with subcutaneous injection of 5 µg estrogen, followed by 20 days with daily intraperitoneal injection of either 450 mg/kg BSO or vehicle (PBS). (f) Volume of DoxR and EpiR tumors measured over 20 days with daily treatments of either 450 mg/kg BSO or vehicle (PBS) by intraperitoneal injection. Data are shown as tumor volume on day 0, 6, 13, and 20. N = 4–5, *p<0.05 BSO vs vehicle (two-way ANOVA, Sidak’s post-hoc test). (g) Fold change in DoxR and EpiR tumor volumes after 20 days of daily treatment with either 450 mg/kg BSO or vehicle (PBS) by intraperitoneal injection. Data are shown as fold changes for individual tumors, relative to baseline tumor volume at day 0 (dotted line). The average fold change for each condition is shown by horizontal lines. N = 4–5, *p<0.05 BSO vs vehicle (Student’s t-test). (h) Schematic of common adaptation mechanisms and distinct primary metabolic dependencies in anthracycline resistant breast cancer cells. Doxorubicin and epirubicin both induce production of reactive oxygen species (ROS) and intercalate nucleic acids and inhibit topoisomerase II, leading to double-strand DNA breaks. Both doxorubicin- and epirubicin-resistant cells engage oxidative response, drug metabolism, and drug efflux pathways to overcome these drug mechanisms, and both are dependent on expression of PGC-1α for their survival. PGC-1α-regulated pathways may further underpin distinct and context-dependent metabolic adaptations to either drug. Compared to drug-sensitive control cells, doxorubicin-resistant cells rely on glutamine for de novo glutathione (GSH) synthesis and for mitochondrial ATP production, while epirubicin-resistant cells display elevated mitochondrial content, oxygen consumption rate (OCR), and oxidative bioenergetic capacity. These distinct primary metabolic dependencies are actionable, as epiribucin-resistant cells are more sensitive to phenformin treatment than doxorubicin-resistant cells, and the latter are specifically sensitive to inhibition of glutathione synthesis by buthionine sulfoximine (BSO) both in vitro and in vivo. Unless otherwise noted, all data presented as averages ± S.E.M.

Next, given that doxorubicin-resistant cells display elevated glutathione metabolism and that they rely on glutamine to fuel glutathione synthesis, we explored this pathway for therapeutic intervention through targeted therapy with buthionine sulfoximine (BSO), an inhibitor of the catalytic subunit of glutamate-cysteine ligase (GCLC) (Drew and Miners, 1984). Both doxorubicin-resistant cell models (DoxR and D100) were acutely and specifically sensitive to BSO treatment in vitro (Figure 6c,d). BSO was highly effective in reducing proliferation of DoxR cells even at the lowest dose tested (50 µM, 60% reduction in viable cell count), while having little to no effect on Control and EpiR cells at that concentration (Figure 6c). These results were further replicated in D100 cells, which were significantly more sensitive to BSO than Ctl and E100 cells (Figure 6d).

Given the potency of BSO treatment in vitro, we explored its effectiveness in vivo by injecting DoxR and EpiR cells into opposing mammary fat pads of immunocompromised mice, supplemented with subcutaneous estrogen in order to promote tumor growth. Once DoxR tumors reached ~100 mm3, mice were divided into two groups and treated daily by intraperitoneal injection with either 450 mg/kg of BSO or with vehicle (PBS), for 20 days (Figure 6e). While all EpiR tumors grew to a larger size than DoxR tumors before the start of treatment, both tumor types similarly doubled in size over the 20 days when treated with vehicle (from 100 mm3 to 200 mm3 for DoxR, from 325 mm3 to 650 mm3 for EpiR, Figure 6f,g). Daily BSO treatment significantly reduced the growth of DoxR tumors, which only increased in size by a factor of 30%, while the growth of EpiR tumors were unaffected (Figure 6f,g). Tumor excision after 20 days of treatment further confirmed this result (Figure 6e). Collectively, these data highlight that despite common adaptations to chemotherapeutics, distinct primary metabolic vulnerabilities arise in doxorubicin and epirubicin resistance, which can be targeted through metabolic interventions to impair drug-resistant tumor growth (Figure 6h).

Discussion

Whereas most studies on therapeutic resistance have focused on single agents or multi-drug resistance, our study presents an unprecedented side-by-side comparison of the metabolic adaptations driving resistance to distinct therapeutic agents within the same drug class. Here, we show that two anthracycline drugs, doxorubicin and epirubicin, elicit different actionable primary metabolic adaptations that support therapeutic resistance and breast cancer cell survival.

Specifically, our findings indicate that doxorubicin-resistant, but not epirubicin-resistant, cells rely on elevated usage of glutamine for de novo glutathione synthesis. This metabolic dependency can be targeted, as demonstrated by the fact that doxorubicin-resistant cells and tumors are significantly more sensitive than epirubicin-resistant cells to therapeutic intervention with BSO, which interferes with glutathione synthesis. Given that side effects such as cardiotoxicity limit the use of anthracyclines in patients to a restrictive cumulative total lifetime dose, there is important clinical relevance in reducing tumor growth in anthracycline-resistant patients through a secondary treatment option such as BSO without administrating additional anthracycline chemotherapy (Barrett-Lee et al., 2009). In contrast to doxorubicin-resistant cells, epirubicin-resistant cells display a drastic increase in OXPHOS and oxidative bioenergetic capacity and were more sensitive than doxorubicin-resistant cells to treatment with phenformin. This aligns with the compendium of evidence showing that dependence on mitochondrial energy metabolism and oxidative phosphorylation is a widespread characteristic of drug resistance, across several cancer types and therapeutic interventions (Caro et al., 2012; Bosc et al., 2017; Matassa et al., 2016; Viale et al., 2014; Vellinga et al., 2015; Ippolito et al., 2016; Vazquez et al., 2013; Lee et al., 2017; Farge et al., 2017; Kuntz et al., 2017; Thompson et al., 2017).

Our findings are particularly interesting given the structural similarity of doxorubicin and epirubicin. In line with this, breast cancer cells resistant to either drug displayed well-known mechanisms of resistance to anthracyclines, including increased drug efflux, lysosomal activity, and oxidative stress response (Hembruff et al., 2008; Guo et al., 2016). Their different metabolic state may, in part, be due to minor structural differences between the two drugs, leading to distinct on- and off-target effects under sustained treatments with either drug (Salvatorelli et al., 2006). For example, while cardiotoxicity is a common side effect of sustained anthracycline treatment, epirubicin has been shown to induce less cardiotoxic effects than doxorubicin, even if both drugs display equivalent response rate to treat breast cancer (Mao et al., 2019). Given that cardiotoxicity is linked to oxidative stress, it is possible that breast cancer cells treated with doxorubicin may face a greater oxidative challenge over the course of treatment than those treated with epirubicin, which aligns with a greater dependence of doxorubicin-resistant cells on de novo glutathione synthesis (Salvatorelli et al., 2006). Accordingly, doxorubicin-resistant cells also displayed much greater engagement of oxidative stress response than epirubicin-resistant cells. Interestingly, epirubicin-resistant cells displayed an elevated level of uncoupled respiration, which may represent an alternate approach to minimizing ROS production in this model (Echtay et al., 2002; Brand, 2000). Indeed, targeting uncoupling proteins has previously been shown to sensitize multi-drug-resistant leukemia cells to both doxorubicin and epirubicin (Mailloux et al., 2010).

Crucially, our study further upholds the viability of exploiting metabolic alterations associated with resistance to chemotherapeutic drugs to increase their success rate (Zaal and Berkers, 2018). This strategy has already shown success in numerous cancers; for example, inhibition of amino acid recycling sensitized neuroblastomas to cisplatin (Gunda et al., 2020), fueling histidine catabolism via histidine supplementation increases sensitivity of leukemic xenografts to methotrexate (Kanarek et al., 2018), and the glutaminase inhibitor CB-839 synergistically enhances the cytotoxicity of carfilzomib in treatment-resistant multiple myeloma cells, notably through its inhibition of glutamine-fueled mitochondrial respiration (Thompson et al., 2017).

It is also notable that the master regulator of mitochondrial metabolism PGC-1α regulates a significant number of pathways implicated in therapy resistance, including OXPHOS (Mootha et al., 2003), oxidative stress response (St-Pierre et al., 2006), glutamine metabolism (McGuirk et al., 2013), and glutathione metabolism (Guo et al., 2018). The context-dependent roles of PGC-1α may therefore underpin specific metabolic vulnerabilities in both doxorubicin and epirubicin resistance in breast cancer. Accordingly, both doxorubicin- and epirubicin-resistant cells were sensitive to PGC-1α knockdown in our study. This aligns with the emerging role of PGC-1α in driving bioenergetic flexibility and metabolic plasticity in the face of survival challenges involved in cancer progression (Tan et al., 2016; McGuirk et al., 2020; Andrzejewski et al., 2017); advanced cancers need to be adaptable, and thereby the context-dependent adaptations conferred by PGC-1α could further contribute to the difficulty in treating advanced cancers. Indeed, similar to OXPHOS and mitochondrial energy metabolism, PGC-1α has been shown to be implicated in drug resistance across cancer types and through various mechanisms (see Supplementary file 1). While therapeutically targeting transcription factors that relay PGC-1α effects may represent an effective strategy in some cases (De Vitto et al., 2019; Deblois et al., 2016), attempts to directly target PGC-1α have unfortunately shown little success thus far.

Ultimately, targeting global regulators of metabolic plasticity like PGC-1α may be promising as a broad strategy for treatment of therapeutic-resistant cancers. However, targeted interventions exploiting the primary metabolic dependencies associated to specific resistant cancers—such as using BSO as a therapeutic intervention for doxorubicin-resistant breast cancer—may represent a more immediate and effective approach.

Materials and methods

Tissue culture and generation of stable cell lines

Request a detailed protocol

MCF-7CC, MCF-7DOX-2, and MCF-7EPI cells were obtained from Dr. Amadeo Parissenti (Hembruff et al., 2008). Briefly, MCF-7DOX-2 and MCF-7EPI were selected over 12 sequential dose increases with their respective anthracycline drug (doxorubicin, epirubicin) to maximal doses of 98.1 nM and 852 nM (Hembruff et al., 2008). MCF-7CC cells were maintained in 0.1% DMSO through parallel passages (Hembruff et al., 2008). For simplicity, MCF-7CC, MCF-7DOX-2, and MCF-7EPI cells are referred to only as Control, DoxR, and EpiR in this study. Cells were cultured in high-glucose Dulbecco's Modified Eagle's Medium (DMEM, Wisent #319–005 CL), 10% FBS, and penicillin/streptomycin at 37°C and 5% CO2. New resistant MCF-7 models were derived from MCF-7 cells obtained from the American Type Culture Collection (ATCC) and cultured in similar media under increasing doses of doxorubicin (Abmole Biosciences #M1969) or epirubicin (Sigma Aldrich #E9406) from 0.1 nM to a final dose of 100 nM, over the course of 8 months. Ctl cells were maintained in 0.1% DMSO through parallel passages. All cells were maintained in culture with a constant dose of their respective drug or DMSO control, at all times unless otherwise specified.

Proliferation and viability

Request a detailed protocol

Proliferation assays were performed by seeding 200,000 cells in 35 mm plates and growing in full media as described above. For glutamine withdrawal experiments, media was replaced on day three with glutamine-free media or glutamine-free media re-supplemented with 4 mM glutamine. To determine cell counts, cells were washed, trypsinized, and counted using a TC10 automated cell counter (Bio-Rad). Viability was determined by exclusion of trypan blue dye.

Mouse experiments

Request a detailed protocol

Four million DoxR or EpiR cells were injected into opposing mammary fat pads of NOD Scid Gamma mice supplemented twice weekly with subcutaneous injection of 5 micrograms of estrogen in corn oil. Seventy days after tumor cell injection mice were divided into two groups and treated daily by intraperitoneal injection with either 450 mg/kg of L-Buthionine-sulfoximine (Sigma Aldrich #B2515) or vehicle (PBS). Tumor volume was measured weekly using caliper measurements and the formula length*width2*π6.

Gene expression

Request a detailed protocol

Total RNA from cultured cells was extracted using the Aurum Total RNA Mini Kit (Bio-Rad, Mississauga, Canada) and was reverse transcribed with iScript cDNA Synthesis kit (Bio-Rad). mRNA expression analyses by real-time PCR were performed using iQ SYBR Green Supermix (Bio-Rad) and gene-specific primers with the MyiQ2 Real-Time Detection System (Bio-Rad). Values were normalized to TATA-binding protein (TBP) expression. All primer sequences are listed in Supplementary file 2.

Gene expression profiling, enrichment analyses, and ranked gene list comparisons

Request a detailed protocol

Gene expression profiling of Control, DoxR, and EpiR cells was performed with Genome Québec (Montreal, Canada) using the Affymetrix Human Gene 2.0 ST Array (HT) system, for which RNA was isolated as described above. The .CEL files were analyzed and pre-processed using the Affymetrix Transcriptome Analysis Console software (RRID:SCR_016519). These data have been deposited in NCBI’s Gene Expression Omnibus (RRID:SCR_005012, Edgar et al., 2002) and are accessible through GEO Series accession number GSE125187 (https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE125187). Gene Set Enrichment Analysis (GSEA, RRID:SCR_003199) was performed on ranked gene lists, where ranks were designated by the sign of the fold change multiplied by the logarithm of the p-value (Subramanian et al., 2005).

To compare with patient data, differential expression analyses of this microarray and a publicly available patient dataset (GEO accession GSE43816, Gruosso et al., 2016) were performed using the R (RRID:SCR_001905) Bioconductor (RRID:SCR_006442) package ‘LIMMA’ and lists were ranked by t-test statistics (R Development Core Team, 2019; Phipson et al., 2016). There are 34,744 and 20,474 genes in the cell line and GEO data set, respectively. 19,038 genes appeared in both data sets. To compare the observed size of overlap between two ordered gene lists to the expected overlap when two lists are independent, we followed the methods outlined by Yang et al., 2006. Specifically, we measured the expected overlap by randomly shuffling the rank order of one list and measuring the size of overlap, repeating this over 1000 permutations. The R Bioconductor (RRID:SCR_006442) package ‘OrderedList’ was used to calculate the expected overlap.

shRNA screen for drug target genes

Request a detailed protocol

A list of 1215 genes related to clinically-approved drugs was generated based on DrugBank and The NCGC Pharmaceutical Collection (Huang et al., 2011). A library with 7847 shRNAs targeting these genes (FDA library) was constructed from the arrayed and sequence-verified RNAi Consortium (TRC) human genome-wide shRNA collection, provided by The McGill Platform for Cell Perturbation (MPCP) of the Rosalind and Morris Goodman Cancer Research Centre and Biochemistry Department at McGill University. This druggable library consists of 11 plasmid pools. Lentiviral supernatants were generated as described at http://www.broadinstitute.org/rnai/public/resources/protocols. DoxR and EpiR cells were infected separately by the 11 virus pools. Cells were then pooled and plated at 500,000 cells per 15 cm dish with 1000 times of coverage in presence of doxorubicin or epirubicin (respectively), for a total of 32 dishes per cell line. Genomic DNA was extracted from the remaining cells in the original pool, as well as in a pool of all 32 dishes after 7 days of growth, and sequencing libraries were built as previously described (Huang et al., 2012). shRNA stem sequence was segregated from each sequencing reads and aligned to TRC library. The matched reads were counted, normalized, and analyzed in R (RRID:SCR_001905) using MAGeCK (v0.5.5) (Li et al., 2014). Hits were ranked by p-value from most depleted to most enriched in DoxR or EpiR after 7 days, and ranked lists were further analyzed for over-represented pathways using Gene Set Enrichment Analysis (RRID:SCR_003199, Mootha et al., 2003; Subramanian et al., 2005).

Stable isotope tracer analysis

Request a detailed protocol

Stable isotope tracer analyses (SITA) were performed in GC/MS as previously described (McGuirk et al., 2013). Briefly, cells were seeded in 6-well dishes to achieve 70–80% confluency after 48 hr. Media was then replaced by DMEM without glucose, sodium pyruvate or L-glutamine (Wisent #319–062) supplemented with 10% dialyzed FBS, 25 mM glucose, 1X sodium pyruvate, and 4 mM glutamine for 2 hr to equilibrate metabolism. Media was further changed to equivalent labeled media made with either 25 mM [U-13C]-glucose or 4 mM [U-13C]-glutamine for the indicated time points. DMSO, doxorubicin, or epirubicin were present in the media throughout. Cells were washed twice with saline at 4°C, quenched in 80% HPLC-grade methanol at −80°C, sonicated, and centrifuged. Supernatants were supplemented with internal control (750 ng myristic acid-D27) and dried in a cold trap overnight (Labconco) at −1°C. Pellets were solubilized in 10 mg/mL methoxyamine-HCl in pyridine, incubated 30 min at 70°C, and derivatized with N-tert-Butyldimethylsilyl-N-methyltrifluoroacetamide (MTBSTFA) for 1 hr at 70°C. 1 µL was injected into an Agilent 5975C GC/MS in SCAN mode and analyzed using Chemstation (Agilent Techologies, RRID:SCR_015742) and Masshunter softwares (Agilent Technologies, RRID:SCR_015040).

Tracing glutamine carbons to glutathione was done using a similar labeling method as above. Cells were washed twice with 150 mM ammonium formate buffer in HPLC water at 4°C, quenched in 50% HPLC-grade methanol at −20°C on dry ice, and phase-separated using acetonitrile, water, and dichloromethane after vigorous bead-beating and vortexing. The aqueous phase was collected and dried in a cold trap overnight at −1°C. Pellets were solubilized in HPLC water and 5 µL was injected into an Agilent 6540 UHD Accurate-Mass Q-TOF LC/MS system coupled to ultra-high pressure liquid chromatography (UHPLC, 1290 Infinity LC System) and analyzed using Masshunter software.

All isotopic corrections were performed using an in-house algorithm designed by SM as previously described (McGuirk et al., 2013).

Metabolomics

Request a detailed protocol

Steady-state metabolite abundances were determined using GC/MS and LC/MS systems, using unlabeled media. Citric acid cycle, glycolytic intermediates, and fatty acids were measured in GC/MS as described above. Amino acids were measured in a Q-TOF system as described above. Nucleotide abundances were determined by washing 70–80% confluent 10 cm plates of cells with 150 mM ammonium formate at 4°C, quenched in 80% HPLC-grade methanol at −80°C on dry ice, after which the cell slurry was quickly transferred to tubes equilibrated in liquid nitrogen. After 24 hr, these were phase-separated using water and dichloromethane after vigorous bead-beating and vortexing. The aqueous phase was collected and flash-frozen in liquid nitrogen, then dried in a cold trap at −1°C. Once dry, pellets were maintained at −80°C and solubilized in HPLC water immediately before injection into an Agilent 6430 Triple Quadrupole LC/MS system coupled to ultra-high pressure liquid chromatography (UHPLC, 1290 Infinity LC System) separation for fast targeted analysis.

Glutathione levels were quantified using an Agilent 1100 series HPLC (Mailloux et al., 2014). Three days post-seeding, cells grown in 6-well plates were washed twice with ice-cold PBS, flash-frozen on dry ice and kept at −80°C until further processing. Cells from parallel plates were counted for normalization. Cells were lysed on ice for 20 min using a mix of 125 mM sucrose, 1.5 mM EDTA, 5 mM Tris, 0.5% TFA and 0.5% MPA in 50% mobile phase (10% HPLC grade methanol, 0.09% TFA – 0.2 μm filtered). Lysates were then centrifuged for 20 min at 14,000 g, 4°C. Each sample was run in duplicate on a Pursuit5 C18 column (150 × 4.6 mm, 5 μm; Agilent Technologies, Santa Clara, CA) with a 1 mL/min flow rate and detected at 215 nm. Standards were diluted in the same buffer and interpolated between the samples. All LC/MS data were analyzed using the Masshunter software (Agilent Technologies, RRID:SCR_015040).

Media metabolite levels were determined using a BioProfile 400 Analyzer (BioNova). Briefly, 2 mL media was collected from cells after 72 hr incubation at 37°C in a CO2 incubator. These were centrifuged to remove any cell debris, and 1 mL was used to measure glucose, lactate, glutamine, glutamate, NH4+, and H+ levels. To control for natural degradation of metabolites, values were compared to that of media incubated in parallel wells which contained no cells.

Integrated metabolic network analysis

Request a detailed protocol

Integrated metabolic network analysis was performed as previously described (Vincent et al., 2015) using the Shiny GAM application (https://artyomovlab.wustl.edu/shiny/gam/; Sergushichev et al., 2016) and visualized using Cytoscape (RRID:SCR_003032, Shannon et al., 2003). FDR was set to −0.25 for metabolites and −3.9 for gene expression for comparison of DoxR and Control, and to −0.1 and −3.4 respectively for comparison of EpiR and Control. Absent metabolite score was set to −0.5 for all analyses.

ROS measurements

Request a detailed protocol

Cells were seeded in a 96-well dish for 48 hr prior to the experiment to achieve 75–80% confluence. Cells were maintained under normal drug conditions throughout. After PBS wash, cells were incubated with 20 µM CM-H2DCFDA (Thermo Fisher Scientific #C6827) in serum-free high-glucose DMEM for 30 min at 37°C, covered in foil to prevent light exposure. Control wells without CM-H2DCFDA were supplemented with equivalent volume of DMSO. After 30 min, cells were washed with PBS and incubated an additional 30 min with high-glucose DMEM supplemented with either water or 0.03% (vol/vol) H2O2. Fluorescence was then measured in an Omega plate reader (BMG LabTech) at excitation/emission wavelengths of 495/520 nm.

Immunoblots

Request a detailed protocol

Total proteins from cultured cells were extracted with lysis buffer (50 mM Tris–HCl pH 7.4, 1% Triton X-100, 0.25% sodium deoxycholate, 150 mM NaCl, 1 mM EDTA) supplemented with inhibitors (2 μg/mL pepstatin, 1 μg/mL aprotinin, 1 μg/mL leupeptin, 0.2 mM phenylmethylsulfonyl fluoride and 1 mM sodium orthovanadate) and quantified using a BCA protein assay kit (Thermo Fisher Scientific #PI123225). The blots were incubated according to the manufacturer's instructions with the following primary antibodies: PGC-1α (Calbiochem #ST1202, RRID:AB_2237237), PGC-1β (Millipore #ABC218, RRID:AB_2891214), and Actin (Santa Cruz Biotechnology #sc-1616, RRID:AB_630836) and with horseradish peroxidase-conjugated secondary antibodies (anti-mouse, KPL #KP-074–1806; anti-rabbit, KPL #KP-074–1506; anti-goat, Abcam #ab6881, RRID:AB_955236). The results were visualized using Clarity ECL (Bio-Rad #1705060).

Respirometry, bioenergetics, and JATP calculations

Request a detailed protocol

Oxygen consumption rate (OCR) and extracellular acidification rate (ECAR) were measured using a Seahorse XFe96 Analyzer (Agilent Technologies, RRID:SCR_019545). Briefly, 10,000 cells were plated in 100 µL of their standard growth media and, after overnight culture, washed twice with XF media at pH 7.4, and equilibrated in XF media supplemented with 25 mM glucose, 4 mM glutamine, and sodium pyruvate (1X) at pH 7.4 in a CO2-free 37°C incubator for 1 hr. Three sequential measurements of OCR and ECAR were taken to assay bioenergetics under basal, proton leak (1 µM oligomycin, Sigma Aldrich #O4876), maximal respiration (1 µM FCCP, Sigma Aldrich #C2920), OXPHOS inhibition (1 µM each rotenone and myxothiazol, Sigma Aldrich #R8875 and #T5580), and high glycolytic ATP demand (20 µM monensin, Sigma Aldrich #M5273) conditions. ECAR data was corrected for media buffering power as previously described (Mookerjee et al., 2016) and both OCR and ECAR were normalized on protein levels. ATP production rates (JATP), glycolytic index, bioenergetic capacity, and ATP supply flexibility were determined quantitatively as previously described (Mookerjee et al., 2017). Glutamine deprivations were performed over 4 hr in supplemented XF media as described above compared to media without supplemented glutamine, prior to measurement of OCR and ECAR as described.

Immunofluorescence and quantification of mitochondrial volume

Request a detailed protocol

Cells were seeded onto 18 mm #1.5 glass coverslips and placed in 12-well plates overnight, then fixed with 4% PFA for 15 min at 37°C. Blocking and permeabilization was carried out by incubation with PBS containing 1% BSA and 0.5% Triton X-100. Mitochondria were visualized through staining with rabbit polyclonal anti-Tom20 antibody (Proteintech #11802–1-AP, RRID:AB_2207530) and goat anti-rabbit secondary antibody conjugated to Alexa Fluor 568 (Thermo Fisher Scientific #A-11011, RRID:AB_143157). Cytoplasm was visualized using HCS CellMask Green stain (Thermo Fisher Scientific #H32714), and nuclei were stained with DAPI. Coverslips were mounted onto microscope slides using ProLong Glass Antifade Mountant (Thermo Fisher Scientific #P36982) and kept at 4°C in the dark until imaging. Images were taken with an Axio Observer Z1 epifluorescent microscope (Zeiss), using a 63x Plan-Apochromat oil objective. Deconvolution of images was carried out in Autoquant X2 software (MediaCybernetics, RRID:SCR_002465) using an adaptive PSF with 10 iterations. Segmentation and surface rendering of mitochondria, cytoplasm, and nuclei was performed in Imaris v8 (Bitplane, RRID:SCR_007370).

ChIP

Request a detailed protocol

For ChIP analyses, chromatin was prepared from Control, DoxR, and EpiR cells maintained in drug prior harvesting. Standard ChIP was performed as described previously (Deblois et al., 2016). Quantification of ChIP enrichment by real-time quantitative PCR was carried out using the LightCycler480 instrument (Roche). ChIPs are normalized against background enrichment on anti-IgG antibody ChIP control and average enrichment on two negative control unbound regions. The antibodies used are: anti-PGC1α (Santa Cruz Biotechnology #sc-13067, RRID:AB_2166218), anti-ERRα (Abcam #Ab76228, RRID:AB_1523580). The ChIP primers are listed in Supplementary file 3.

siRNA knockdowns

Request a detailed protocol

Cells were subjected to either 40 nM control siRNA (Dharmacon #D-001810–10- 05) or a combined 40 nM pool of four siRNA specifically targeting PPARGC1A (Qiagen FlexiTube-GeneSolution #GS10891) and four siRNA specifically targeting PPARGC1B (Qiagen FlexiTube-GeneSolution #GS133522). Cells were transfected using Lipofectamine RNAiMax (ThermoFisher #13778–150) and incubated for 72 hr before pursuing subsequent experiments.

Statistical analyses

Request a detailed protocol

All statistical analyses were performed using GraphPad Prism (GraphPad Software Inc, RRID:SCR_002798), Microsoft Excel (Microsoft Corporation, RRID:SCR_016137), or R (R Foundation for Statistical Computing, RRID:SCR_001905 R Development Core Team, 2019).

Appendix 1

Appendix 1—key resources table
Reagent type (species) or resourceDesignationSource or referenceIdentifiersAdditional information
Cell line (H. sapiens)MCF-7 Control, MCF-7CCObtained from Dr. Amadeo Parissenti Hembruff et al., 2008; DOI: 10.1186/1471-2407-8-318Grown in 0.1% DMSO media. Negative for mycoplasma.
Cell line (H. sapiens)MCF-7 DoxR, MCF-7DOX-2Obtained from Dr. Amadeo Parissenti Hembruff et al., 2008; DOI: 10.1186/1471-2407-8-318Resistant to 98.1 nM doxorubicin. Negative for mycoplasma.
Cell line (H. sapiens)MCF-7 EpiR, MCF-7EPIObtained from Dr. Amadeo Parissenti Hembruff et al., 2008; DOI: 10.1186/1471-2407-8-318Resistant to 852 nM epirubicin. Negative for mycoplasma.
Cell line (H. sapiens)MCF-7 CtlThis paperGrown in 0.1% DMSO media. Derived from MCF-7 cells obtained from the American Type Culture Collection (ATCC). Negative for mycoplasma.
Cell line (H. sapiens)MCF-7 D100This paperResistant to 100 nM doxorubicin. Derived from MCF-7 cells obtained from the American Type Culture Collection (ATCC). Negative for mycoplasma.
Cell line (H. sapiens)MCF-7 E100This paperResistant to 100 nM epirubicin. Derived from MCF-7 cells obtained from the American Type Culture Collection (ATCC). Negative for mycoplasma.
AntibodyHuman PGC-1α (mouse, monoclonal)CalbiochemCat #: ST1202; RRID:AB_2237237Immunoblots, (1:1000)
AntibodyHuman PGC-1β (rabbit, polyclonal)MilliporeCat #: ABC218
RRID:AB_2891214
Immunoblots, (1:1000)
AntibodyHuman Actin (goat, polyclonal)Santa Cruz BiotechnologyCat #: sc-1616
RRID:AB_630836
Immunoblots, (1:2000)
Antibodyanti-mouse (goat, polyclonal)KPLCat #: KP-074-1806
Immunoblots, (1:10000)
Antibodyanti-rabbit (goat, polyclonal)KPLCat #:KP-074-1506
Immunoblots, (1:10000)
Antibodyanti-goat (donkey, polyclonal)AbcamCat #: ab6881
RRID:AB_955236
Immunoblots, (1:10000)
AntibodyHuman Tom20 (rabbit, polyclonal)ProteintechCat #:11802-1-AP
RRID:AB_2207530
Immuno-fluorescence
Antibodyanti-rabbit conjugated to Alexa Fluor 568 (goat, polyclonal)Thermo Fisher ScientificCat #: A-11011
RRID:AB_143157
Immuno-fluorescence
AntibodyHuman PGC-1α (rabbit, polyclonal)Santa Cruz BiotechnologyCat #: sc-13067
RRID:AB_2166218
ChIP
AntibodyHuman ERRα (rabbit, monoclonal)AbcamCat #: Ab76228
RRID:AB_1523580
ChIP
transfected construct (H. sapiens)ON-TARGETplus Non-targeting Control Pool siRNADharmaconCat #: D-001810-10-05
40 nM pool of siRNA
transfected construct (H. sapiens)ON-TARGETplus Human PPARGC1A siRNAQiagenCat #: FlexiTube-GeneSolutionGS10891
Combined 40 nM pool of four siRNA
transfected construct (H. sapiens)ON-TARGETplus Human PPARGC1B siRNAQiagenCat #: FlexiTube-GeneSolutionGS133522
Combined 40 nM pool of four siRNA
transfected construct (H. sapiens)FDA shRNA libraryThe McGill Platform for Cell Perturbation (MPCP) of the Rosalind and Morris Goodman Cancer Research Centre and Biochemistry department at McGill UniversityDeveloped by YX, GM, and SH
sequence-based reagentRT-qPCR primersSee Supplementary file 2
sequence-based reagentChIP primersSee Supplementary file 3
commercial assay or kitAurum Total RNA Mini KitBio-Rad
commercial assay or kitiScript cDNA Synthesis kitBio-Rad
commercial assay or kitiQ SYBR Green SupermixBio-Rad
commercial assay or kitBCA protein assay kitThermo Fisher ScientificCat #: PI123225
commercial assay or kitSeahorse XFe96 AnalyzerAgilent TechnologiesRRID:SCR_019545
commercial assay or kitBioProfile 400 AnalyzerBioNova
chemical compound, drugDoxorubicinAbMole BiosciencesCat #: M1969
chemical compound, drugEpirubicinSigma AldrichCat #: E9406
chemical compound, drugL-buthionine-sulfoximineSigma AldrichCat #: B2515
chemical compound, drugPhenforminSigma AldrichCat #: P7045
chemical compound, drugOligomycinSigma AldrichCat #: O4876
chemical compound, drugFCCP (Carbonyl cyanide 4-(trifluoromethoxy)phenylhydrazone)Sigma AldrichCat #: C2920
chemical compound, drugRotenoneSigma AldrichCat #: R8875
chemical compound, drugMyxothiazolSigma AldrichCat #: T5580
chemical compound, drugMonensinSigma AldrichCat #: M5273
chemical compound, drugCM-H2DCFDAThermo Fisher ScientificCat #: C6827
chemical compound, drugClarity ECLBio-RadCat #: 1705060
chemical compound, drugHCS CellMask Green stainThermo Fisher ScientificCat #: H32714
chemical compound, drugLipofectamine RNAiMaxThermo Fisher ScientificCat #: 13778-50
chemical compound, drugProLong Glass Antifade MountantThermo Fisher ScientificCat #: P36982
software, algorithmAutoquant X2 softwareMediaCyberneticsRRID:SCR_002465
software, algorithmImaris v8BitplaneRRID:SCR_007370
software, algorithmGraphPad PrismGraphPad Software, IncRRID:SCR_002798
software, algorithmMicrosoft ExcelMicrosoft CorporationRRID:SCR_016137
software, algorithmR Project for Statistical ComputingR Foundation for Statistical Computing, (R Development Core Team, 2019)RRID:SCR_001905
software, algorithmR BioconductorDOI:10.1186/gb-2004-5-10-r80RRID:SCR_006442
software, algorithmShiny GAMhttps://artyomovlab.wustl.edu/shiny/gam/ Sergushichev et al., 2016; DOI:10.1093/nar/gkw266
software, algorithmCytoscapeShannon et al., 2003; DOI:10.1101/gr.1239303RRID:SCR_003032
software, algorithmIn-house algorithm for isotopic correctionsIn-house algorithm of the St-Pierre laboratory, first described in McGuirk et al., 2013; DOI:10.1186/2049-3002-1-22Developed by SM
software, algorithmMasshunter Quantitative Analysis softwareAgilent TechnologiesRRID:SCR_015040
software, algorithmChemstation softwareAgilent TechnologiesRRID:SCR_015742
software, algorithmTranscriptome Analysis ConsoleAffymetrixRRID:SCR_016519
software, algorithmGene Set Enrichment AnalysisMootha et al., 2003; Subramanian et al., 2005; DOI:10.1038/ng1180, 10.1073/pnas.0506580102RRID:SCR_003199
software, algorithmGene Expression OmnibusNCBI Edgar et al., 2002; DOI:10.1093/nar/30.1.207RRID:SCR_005012
software, algorithmMAGeCK (v0.5.5)Li et al., 2014; DOI:10.1186/s13059-014-0554-4
otherGEO patient datasetGSE43816 Gruosso et al., 2016; DOI:10.15252/emmm.201505891Gene expression of human breast cancer tumors biopsies prior to and after treatment with four cycles of epirubicin and cyclophosphamide, followed by four cycles of docetaxel
otherGEO cell line datasetGSE125187; this paperGene expression of Control, DoxR, and EpiR cells

Data availability

Microarray data have been deposited in GEO under accession code GSE125187.

The following data sets were generated
    1. McGuirk S
    2. St-Pierre J
    (2019) NCBI Gene Expression Omnibus
    ID GSE125187. Gene expression data in Control, Doxorubicin-resistant, and Epirubicin-resistant breast cancer cells.
The following previously published data sets were used
    1. Gruosso T
    2. Kieffer Y
    3. Mechta-Grigoriou F
    (2016) NCBI Gene Expression Omnibus
    ID GSE43816. Response to Neoadjuvant Chemotherapy in Triple Negative Breast tumors.

References

  1. Book
    1. Edwardson D
    2. Chewchuk S
    3. Parissenti AM
    (2013) Resistance to Anthracyclines and Taxanes in Breast Cancer
    In: Ahmad A, editors. Breast Cancer Metastasis and Drug Resistance. New York: Springer Science+Business Media. pp. 227–247.
    https://doi.org/10.1007/978-1-4614-5647-6_13
  2. Software
    1. R Development Core Team
    (2019) R: A Language and Environment for Statistical Computing
    R Foundation for Statistical Computing, Vienna, Austria.
    1. Siegel RL
    2. Miller KD
    3. Jemal A
    (2018) Cancer statistics
    CA: A Cancer Journal for Clinicians 68:7–30.
    https://doi.org/10.3322/caac.21442
    1. Xu RH
    2. Pelicano H
    3. Zhou Y
    4. Carew JS
    5. Feng L
    6. Bhalla KN
    7. Keating MJ
    8. Huang P
    (2005)
    Inhibition of glycolysis in Cancer cells: a novel strategy to overcome drug resistance associated with mitochondrial respiratory defect and hypoxia
    Cancer Research 65:613–621.
    1. Yang X
    2. Bentink S
    3. Scheid S
    4. Spang R
    (2006) Similarities of ordered gene lists
    Journal of Bioinformatics and Computational Biology 04:693–708.
    https://doi.org/10.1142/S0219720006002120

Decision letter

  1. Matthew G Vander Heiden
    Reviewing Editor; Massachusetts Institute of Technology, United States
  2. Utpal Banerjee
    Senior Editor; University of California, Los Angeles, United States

In the interests of transparency, eLife publishes the most substantive revision requests and the accompanying author responses.

Acceptance summary:

In this paper, the authors show that doxorubicin and epirubicin resistant cells have different metabolic characteristics. Specifically, they show that doxorubicin resistant cells rely on increased glutamine consumption to produce mitochondrial ATP and to synthesize glutathione whereas epirubicin resistant cells do not. The authors speculate that the unique metabolic characteristics associated with doxorubicin resistance are caused by the induction of more oxidative stress, but exactly what drives this difference will be a topic for future work.

Decision letter after peer review:

[Editors’ note: the authors submitted for reconsideration following the decision after peer review. What follows is the decision letter after the first round of review.]

Thank you for submitting your work entitled "Tailored metabolic adaptations confer resistance to chemotherapy in breast cancer" for consideration by eLife. Your article has been reviewed by 3 peer reviewers, one of whom is a member of our Board of Reviewing Editors, and the evaluation has been overseen by a Senior Editor. The reviewers have opted to remain anonymous.

Our decision has been reached after consultation between the reviewers. Based on these discussions and the individual reviews below, we regret to inform you that your work will not be considered further for publication in eLife. While all reviewers found the topic to be of interest, our concern is that the time required to further justify the conclusions was such that you may wish to seek publication elsewhere. If you feel that revisions could be done to address the concerns of the reviewers, then such a resubmission will be overseen by the same group of reviewers and editors (if possible). Please note that we cannot guarantee the outcome of a resubmission as a new manuscript, but minimally, the following points would have to be addressed:

1. Titration of drug dose to address whether the metabolic adaptation is related to the dose or the drug itself.

2. Further work to establish causality for the specific metabolic alterations in resistance to doxorubicin versus epirubicin.

3. Some additional effort to assess the relevance of these resistance mechanisms in vivo. The reviewers felt that the patient correlation included in the study is contrary to the claim that specific metabolic adaptations arise to each drug, and thus some work to show the main claims hold in some in vivo context is needed.

Reviewer #1:

The authors select for doxorubicin and epirubicin resistant MCF7 cells in culture and determine that the global gene expression profile of these cells is similar to tumors from patients exposed to anthracylins. They then characterize the resistant cells, and argue that doxorubicin or epirubicin select for different adaptations, which is surprising given the similarity in these drugs.

1. Is the difference between doxorubicin response and epirubicin response simply drug dose? The bulk of the analysis largely compares a cells from a single dose where doxorubicin is cytostatic and epirubicin is cytotoxic. If doses of either drug are titrated to show cytostasis or cytotoxicity is there still a difference in metabolic adaptation of the resistant cells? This is only partially addressed in Figure 7, and is a crucial questions that is important to judge the central claim of the paper.

2. There is a large literature arguing that response to chemotherapy in culture is not necessarily indicative of response in patients, and there is a growing literature around cancer metabolism also being different in the cell culture and in tumors, including glutamine metabolism. The bulk of the analysis is based on cell culture, and if some data were available that speak to whether these same differences or dependencies hold in vivo it would greatly improve the manuscript.

3. The authors assess labeling of metabolites at short time points (15 min), which can be more useful for assessment of flux through pathways like glycolysis. However, this data would be much more informative if there were multiple time points assessed, as this is typically need to interpret kinetic tracing data.

4. Seahorse assays were used to suggest ATP production rate, but this approach requires several assumptions that should be more explicitly discussed. I suggest showing the raw data, with or without the calculated rates, and feel this is particularly important for a broad audience who may not appreciate the caveats associated with this analysis.

5. While the fact that targeting PGC1a overcomes resistance to both doxorubicin and epirubicin is potentially relevant for clinical medicine, from a mechanistic standpoint it does not support their claim that these drugs elicit different metabolic adaptations. Is there an example of a target that separates the contexts as evidence of differential resistance mechanism?

Reviewer #2:

In their manuscript, "Tailored Metabolic Adaptations Confer Resistance to Chemotherapy in Breast Cancer", McGuirk et al. derive breast cancer cells resistant to doxorubicin and epirubicin and investigate whether the metabolic alterations that occur during resistance are dependent on the therapeutic. They suggest that doxorubicin resistant cells rely on glutamine for mitochondrial function and glutathione synthesis, while epirubicin resistant cells increase their mitochondrial capacity and ATP production. They suggest that targeting PGC1a could eradicate cells resistant to both therapeutics, as PGC1a regulates both metabolic adaptions. While the idea that different therapeutics of the same class could cause resistance due to different metabolic mechanisms, this study was limited by a lack of evidence for a causal role for the metabolic alterations identified in drug resistance.

1. It was difficult to follow the claims of the manuscript. Many statements are vague and often the figure legends and section headings/text do not match, and are sometimes contradictory in their claims.

2. The authors generate doxorubicin resistant (DoxR) and epirubicin resistant (EpiR) resistant cell lines via exposure to increasing concentrations of drug. DoxR resistant cells are cultured in a final concentration of 98.1nM drug, while epirubicin resistant cells are cultured in 852nM drug. Treatment of chemonaive cells with these concentrations results in cytostatic (Dox) or cytotoxic (Epi) effects. This approach raises several concerns.

a. First, why these concentrations? These differing concentrations are a concern for the major findings of the manuscript.

b. Second, anthracyclins have been shown to damage mitochondrial DNA, and the use of differing concentrations may result in different amounts of damage. Related to this point does compensatory mitochondrial biogenesis explain why PGC1a expression is increased, particularly in the EpiR cells, which are treated with significantly more anthracycline? What happens to mitochondrial content in these resistant cell lines? Mitochondrially encoded respiratory components? How does PGC1a/b knockdown affect mitochondrial content?

c. Third, the final figure, in which new resistant cell lines are derived and grown in the same amount of Dox or Epi, the authors show now that while DoxR cells still increase GSH synthesis from glucose, similar increases in mitochondrial ATP and bioenergetic capacity are observed between DoxR and EpiR cells. While the effects are modestly more pronounced in EpiR cells, this finding is contrary to the thesis that Dox and Epi resistance induce different metabolic alterations as it relates to the mitochondria. These cell lines should be used throughout the study to eliminate any confounding effects of differing drug concentrations.

3. Related to the point above, it is unclear how PGC1a is promoting resistance to doxorubicin and epirubicin. In Figure 6, PGC1a associates with the antioxidant response genes only in EpiR cells, but PGC1a/b knockdown has a greater effect on gene expression in the DoxR cells. What is the effect of PGC1a knockdown alone? In addition, Figure 3i shows that EpiR cells have little to no elevation of antioxidant gene expression, despite significantly increased PGC1a binding to their promoters. How are these findings reconciled? Further, siPGC1a/b affects the growth of DoxR and EpiR cells equally. Is this in the presence of drug? It is unclear from the figure legend. Finally, is PGC1a overexpression sufficient to promote resistance of drug naïve cells? This experiment is needed to show causation.

4. The authors suggest that cells with engineered resistance to Dox and Epi have similar gene expression patterns to residual tumors of breast cancer patients following anthracycline-based chemotherapy, but there are several problems with this analysis/these claims. First, it is not clear that these tumors are resistant to anthracyclines. From the description, patients received 4 cycles of anthracycline-based therapy, followed by four cycles of docetaxel. Resultant tumors may be responders, resistant to anthracyclines, resistant to docetaxel, or both. As a consequence, it is difficult to draw meaningful conclusions from the comparison. Second, the authors compare their cell line gene expression to these tumors, but tumors were treated only with epirubicin. Since the authors argue that resistance to Dox vs. Epi is via discrete mechanisms, the argument for similarity between DoxR cells and Epi "resistant" tumors is contrary to the main claims of the manuscript. Indeed, the claim that "several distinct metabolic pathways were altered by doxorubicin or epirubicin resistance (Figure 1f,h)" is problematic due to these concerns since (1) these pathways are based on overlap with gene expression from tumors from patients that may or may not be resistant and (2) these tumors were treated only with epirubicin. Therefore, the basis for distinct metabolic pathways is unsound.

5. The increase in GCLC, ME1, NQO1 etc in DoxR cells is consistent with NRF2 activation as proposed, as is the increase in GSH M+5, which suggests an increase in GSH synthesis. However, the decrease in the GSH/GSSG ratio and NADPH, and increase in NADP+ is not consistent, nor is it consistent with the protection against ROS shown in Figures 3j and 3k. How is this reconciled? What is the total pool size of GSH + GSSG? Related to this, what is the "control" for figures 3J and 3k? Parental cells?

6. The data shown in Figure 4 is not supportive of an increased dependence of DoxR cells on glutamine, as claimed in the figure legend. Glutamine is used for many purposes in cells and neither DoxR nor EpiR cells grow without glutamine in Figure 4A. Further, the idea that glutamine is driving TCA cycle is not matched by the tracing data from Figure 3. There is no increased entry of glucose or glutamine into the TCA cycle in the DoxR cells (e.g. citrate m+2 from glucose, citrate m+4 from glutamine). Related to this, the inability of DoxR cells to compensate for ATP production upon glutamine withdrawal suggests a defect in glycolytic regulation or ATP sensing, rather than glutamine metabolism. More clarity is needed for (1) what metabolic alterations occur, (2) how they are regulated and (3) how they promote anthracycline resistance.

Reviewer #3:

In "Tailored Metabolic Adaptations Confer Resistance to Chemotherapy in Breast Cancer", McGuirk et al. use MCF-7 cells in culture to study mechanisms of resistance to doxorubicin and epirubicin. Overall, I found their approach to be intriguing. It seems like a compelling approach to study drug resistance in cancer. They subjected cells to many passages in the presence of drug, which resulted in a population of resistant cells. They could then compare these resulting resistant cells to the original drug-susceptible cells. In doing so, the authors claim to have found different metabolic adaptations to each drug. Given the differences in adaptation, they suggest that it might be better to target resistant tumors with drugs that hit global regulators of metabolism, like PGC-1a.

I have two major questions.

1. How do the authors know that the genes and metabolic adaptations assumed in the resistant cells are essential to drug resistance? Might it be the case that their original control population of cells was heterogeneous and exposure of cells to drug over long periods of time simply selected those subclones that were resistant? In such a model, while one gene/metabolic phenotype may be essential to developing resistance (perhaps ABC transporters), the other genes/metabolic phenotypes could be random. Indeed, based on the limited amount of data shown from two experiments, it seems like the metabolic phenotypes of cells exposed to dox for long periods of time are different in separate experiments. For example, the isotope labeling shown in Figure 3 looks very different from the labeling shown in Figure 7. A recent study by Speirs and Price et al. (doi: 10.18632/oncotarget.26533) did a similar experiment without drugs. After selecting subclones of cells from a culture, they found significant differences in metabolic phenotypes. It would be most interesting to compare the genetic and metabolic differences of resistance cells that were selected independently by drugs in separate experiments. This could help better resolve whether the differences observed are essential.

2. The major premise of this manuscript is that the authors have selected cells resistant to dox and epi. Data shown in Figure 1B suggest that this is the case over 6 days. Based on the data shown in Figure 4A, however, I worry that the Epi-R cells are not resistant to Epi over long periods of time. For example, at day 11, the Epi-R live cell count looks more like the control cells in Figure 1 than DoxR. Thus, if the cells are not both equally resistant, maybe the metabolic adaptations are not really "distinct" but rather just reflective of different degrees of resistance.

Other points:

1. Can the authors do any experiments to explore (or possibly speculate in the discussion) why the two drugs would elicit different resistance mechanisms?

2. The increase of pyruvate carboxylase in EpiR cells is interesting. What is the phenotypic value of such a metabolic alteration in resistance?

3. The data do not seem to support this statement: "DoxR cells were unable to increase glycolytic ATP production to compensate for diminished oxidative ATP production upon glutamine withdrawal." The Seahorse plots in Figure 4 actually show that glycolytic ATP production increases in drug resistance.

4. It would be helpful to see all of the isotopologues, instead of just a select number. For example, M+3 in malate from 13C-glucose doesn't necessarily mean pyruvate carboxylase activity. It could also be indicative of two rounds of TCA cycle (the first with labeled acetyl-CoA and the second with unlabeled). It would be easier to assess these kinds of possibilities if the authors presented full isotopologue plots.

5. Many of the interesting data shown are not explored (or discussed). For example, in Figure 5C-D, the DoxR cells have uncoupled mitochondria. Why is this? Is this the source of ROS? It might be best to remove data that are not discussed, as it is distracting from the overall message.

[Editors’ note: further revisions were suggested prior to acceptance, as described below.]

Thank you for submitting your article "Resistance to different anthracyclines elicits distinct and actionable primary metabolic dependencies in breast cancer" for consideration by eLife. Your article has been reviewed by 3 peer reviewers, one of whom is a member of our Board of Reviewing Editors, and the evaluation has been overseen by Utpal Banerjee as the Senior Editor. The reviewers have opted to remain anonymous.

The reviewers have discussed the reviews with one another and the Reviewing Editor has drafted this decision to help you prepare a revised submission.

As the editors have judged that your manuscript is of interest, and much improved from the prior version, the reviewers felt that additional experiments are required before it is published, including making more extensive use of the independently derived resistant cells. We would like to draw your attention to changes in our revision policy that we have made in response to COVID-19 (https://elifesciences.org/articles/57162). First, because many researchers have temporarily lost access to the labs, we will give authors as much time as they need to submit revised manuscripts. We are also offering, if you choose, to post the manuscript to bioRxiv (if it is not already there) along with this decision letter and a formal designation that the manuscript is "in revision at eLife". Please let us know if you would like to pursue this option.

Summary:

In this paper, the authors show that doxorubicin and epirubicin resistant cells have different metabolic characteristics. Specifically, they show that doxorubicin resistant cells rely on increased glutamine consumption to produce mitochondrial ATP and to synthesize glutathione whereas epirubicin resistant cells do not. The authors speculate that the unique metabolic characteristics associated with doxorubicin resistance are caused by the induction of more oxidative stress, but exactly what drives this difference remains unknown.

Essential revisions:

1. Concern remains regarding how relevant the metabolic differences between these two cell populations are to resistance in patients given that different drug concentrations were used to derive the EpiR and DoxR cells and that the new in vivo data did not use a de novo resistance model, but rather the same cell lines used previously. The use of independently derived cell lines should be used to confirm more of the central findings of the manuscript. In particular, please address whether the different metabolic adaptations are related to the dose or the drug itself.

2. The tumor growth differences are not impressive when considering that the DoxR cells grow much more slowly than the EpiR cells. The difference in size even for the DoxR tumors does not seem that great, and is presented as fold change. Please at least acknowledge this, consider presenting the data in a more fair way with tumor size data graphed as actual measured size with the DoxR and EpiR curves on the same plot or plots with the same scales.

3. Please test (in vitro) whether the oxidative stress response in control cells treated with either doxorubicin or epirubicin is different as a potential mechanism for how treatment with these drugs differs in a way that might impact metabolism. Please also considering examining whether Epic cells are more sensitive to OXPHOS inhibitors to consider whether PGC1a promoting the OXPHOS phenotype in epi resistant cells speaks to the divergent metabolism.

4. Please address the issues with the resistance mechanisms in vivo based on the following comment from Reviewer #3. The in vivo BSO experiment is supportive of the cell culture experiment but there are multiple issues with this experiment. First, no anthracycline is included in this experiment making its relevance to cancer therapy questionable. Second, resistant lines are used from the start, rather than assaying the role of GSH synthesis in de novo resistance that arises in vivo, which would have been more relevant to the patient situation.

The full set of comments from the reviewers is provided below for your reference.

Reviewer #1:

The authors characterize the metabolism and gene expression of doxorubicin and epirubicin resistant MCF7 cells in culture and present data that differences exist in the cells that are resistant to these drugs from the same chemotherapy class. They show that independently derived resistant lines retains many of the same phenotypic differences. Finally, they show these differences can lead to different therapeutic vulnerabilities, including response to inhibitors of glutathione synthesis.

A strength is characterizing how cells can gain resistance to similar classes of drugs. The work focuses more on the differences than the similarities, and more analysis is needed to relate their findings to resistance that arises in patients following treatment. There is also suggestion that tumors derived from the different resistant cell lines grow very differently, and this may or may not be part of the differnces in sensitivity to inhibitors of glutathione synthesis.

1. The data relating the DoxR and EpiR gene signatures to gene expression data from patients receiving anthracyclines is use to argue human relevance of the models, although it is unclear to what extent the patient data can be divided into those who received doxorubicin or epirubicin. It seems this is relating the resistance to anthracyclines in general, which speaks more to the similarity in resistance mechanisms rather than the differences that are highlighted in the title and abstract.

2. The tumor growth differences are not impressive when considering that the DoxR cells grow much more slowly than the EpiR cells. One might question if a similar response to BSO would be observed in the EpiR cells if treatment was started at a size that is matched to the DoxR tumors. The difference in size even for the DoxR tumors does not seem that great, and is presented as fold change. At least acknowledging this is needed, and ideally to avoid misleading readers the tumor size data should be graphed as actual measured size with the DoxR and EpiR curves on the same plot (or plots with the same scales).

3. I do not think a difference in glucose to lactate flux is supported by the tracing data. Time to steady state labeling is the best measure of flux and that appears very similar between DoxR and EpiR cells. The differences are based on what is variation in only one time point. This is not central to the work, but should be presented accurately.

4. In this Reviewers opinion, terms like "increased bioenergetic capacity" are not very helpful in understanding what is biologically different. It is acknowledged they are used by many, so is not fair to ask these authors to not use them, but wanted to mention this nevertheless.

Reviewer #2:

The authors show that doxorubicin and epirubicin resistant cells have different metabolic characteristics. Specifically, they show that doxorubicin resistant cells rely on increased glutamine consumption to produce mitochondrial ATP and to synthesize glutathione whereas epirubicin resistant cells do not. As noted above, this is a re-submission that has been improved. The major strengths of the work are the breadth of results showing that doxorubicin and epirubicin resistant cells have different metabolic characteristics – including gene expression data, metabolomics data, isotope tracing data, and tumor data from a mouse model. The authors generally did a nice job of responding to previous comments made by reviewers and revising language to improve clarity. The weakness of the work is that the mechanism(s) underlying the observations are not well defined. The authors speculate that the unique metabolic characteristics associated with doxorubicin resistance are caused because doxorubicin's mode of action induces more oxidative stress than epirubicin, but this key assumption is not proven and only rationalized anecdotally on the basis of cardiotoxicity patient data. Before publication, I recommend an in vitro experiment to test this idea where the authors assess the oxidative stress response in control cells treated with either doxorubicin or epirubicin.

Some other comments and questions:

1. Some sentences in the abstract might be improved by making them less general.

2. In several parts of the manuscript, it is implied that glycolysis is favored in cancer cells over oxphos. This is not necessarily true for all cancer cells and tumors.

3. The authors state that EpiR cells rely on cysteine metabolism (Figure 1k). Based on their BSO data, presumably this isn't for glutathione synthesis. Why are EpiR cells sensitive to cysteine metabolism?

4. Glutathione can be oxidized/reduced to buffer oxidative stress, but it does not provide the reducing equivalents. For that, synthesis of NADPH is needed. One possibility is that glucose is re-routed to the pentose phosphate pathway in DoxR cells, thereby increasing the need to fuel mitochondrial metabolism with glutamine.

5. The authors state that IDH1 activity is increased in DoxR cells, but the evidence presented is relatively weak.

Reviewer #3:

In their manuscript, "Tailored Metabolic Adaptations Confer Resistance to Chemotherapy in Breast Cancer", McGuirk et al. present a revised version of their study that seeks to understand the divergent metabolic changes in doxorubicin and epirubicin resistant cells. Generally, the writing of the manuscript is much improved, the claims are much easier to follow, and the presentation of the data is more focused and digestible. Many of the confusing and problematic claims have been removed, and the description of the data is more accurate. However, the major issues with this study from the previous review remain.

1. Following their previous submission, it was suggested that the authors titrate the drug dose to address whether the metabolic adaptation is related to the dose or the drug itself. Because the dose of epi used to generate resistant lines was almost 10x greater than dox, and provoke different responses in naïve cells (cytotoxicity vs cytostasis) the resulting metabolic changes may simply be attributed to differences in drug dosing. However, this point remains unaddressed. Rather than expanding upon the independently derived resistant lines generated to the same final stable concentration of drug (100nM), the authors buried this data in the supplemental with insufficient discussion.

2. It was also suggested that the authors perform additional work to establish causality for the specific metabolic alterations in resistance to doxorubicin versus epirubicin. The authors have added BSO, an inhibitor of glutathione synthesis, which they show selectively impairs the viability of dox resistant cells, but not epi resistant cells, which supports that resistance-induced metabolic rewiring is targetable. However, the role of PGC1a in promoting the OXPHOS phenotype in epi resistant cells is not addressed; nor is it demonstrated that PGC1a is sufficient to drive the OXPHOS phenotype; nor is it examined whether epi resistant cells are more sensitive to OXPHOS inhibitors. Consequently, this aspect of the revised manuscript focuses primarily on dox resistance and not the divergent metabolic adaptations as intended.

3. Finally, it was suggested that the relevance of the resistance mechanisms in vivo required additional work. The in vivo BSO experiment is supportive of the cell culture experiment but there are multiple issues with this experiment. First, no anthracycline is included in this experiment making its relevance to cancer therapy questionable. Second, resistant lines are used from the start, rather than assaying the role of GSH synthesis in de novo resistance that arises in vivo, which would have been more relevant to the patient situation. Finally, the role of PGC1a/OXPHOS in epi resistance was not addressed in vivo.

https://doi.org/10.7554/eLife.65150.sa1

Author response

[Editors’ note: the authors resubmitted a revised version of the paper for consideration. What follows is the authors’ response to the first round of review.]

1. Titration of drug dose to address whether the metabolic adaptation is related to the dose or the drug itself.

We acknowledge the concerns of the reviewers and have provided clarifications regarding the cell line models. The information below supports that the metabolic adaptations reported in our manuscript are linked to the drug itself and not the drug concentration.

The DoxR and EpiR cell models were first described by the Parissenti group in a 2008 study as outlined in our Methods section (https://dx.doi.org/10.1186%2F1471-2407-8318), and have been used in several publications over the years (for example:https://dx.doi.org/10.1158%2F0008-5472.CAN-16-0774, https://dx.doi.org/10.1038%2Fs41598-018-23496-y, https://bmccancer.biomedcentral.com/articles/10.1186/s12885-016-2790-3). These models were selected over hundreds of passages over a twelve-dose process.

Doxorubicin-resistant cells (DoxR, originally named MCF-7DOX-2) were selected up to the highest concentration of drug at which the cells could survive (the maximally tolerated dose, 98.1 nM doxorubicin), whereas Epirubicin-resistant cells (EpiR, originally named MCF-7EPI) were still viable at the concentration reached at step twelve (852 nM epirubicin).

As shown by the Parissenti’s group, the initial acquisition of resistance to doxorubicin and epirubicin was linked to reduction in drug accumulation by the cells. However, at higher doses the development of resistance was not linked to drug exclusion by the cells. Indeed, Figure 2 from Hembruff et al., 2008 clearly shows that the magnitude of resistance was not linked to reduction in drug accumulation after the initial acquisition, as indicated by the plateau.

Furthermore, forcing the accumulation of doxorubicin in DoxR cells using Cyclosporin A (CsA) did not significantly impact the sensitivity of DoxR cells (solid and dotted blue lines in Figure 5A, B in Hembruff et al., 2008). These data clearly highlight the importance of adaptation mechanisms separate from drug exclusion, and that once the cells are stably resistant, increasing drug concentration has no significant impact on drug sensitivity.

These elements considered, and given the fact that DoxR cells are more resistant to doxorubicin (resistance factor, RF=27.8) than epirubicin (RF=4.79), and that EpiR cells are more resistant to epirubicin (RF=815.3) than doxorubicin (RF=203.4), we hypothesized that the mechanisms of resistance may differ for doxorubicin and epirubicin, despite the chemical similarity of the drugs.

Third and finally, we further addressed this potential impact of difference in the dose by independently deriving cells resistant to 100nM of either doxorubicin or epirubicin. With these cells, we recapitulated the main findings of DoxR and EpiR cells, upholding the conclusions of the paper, i.e. that EpiR cells display increased OXPHOS capacity and that DoxR cells have an elevated usage of glutamine for glutathione synthesis.

2. Further work to establish causality for the specific metabolic alterations in resistance to doxorubicin versus epirubicin.

This is indeed an important point, which we have addressed by targeting a distinct and actionable metabolic vulnerability in doxorubicin versus epirubicin resistance.

Doxorubicin-resistant breast cancer cells are reliant on glutamine to synthesize the antioxidant molecule glutathione de novo, and we now show that they are specifically sensitive to inhibition of this pathway. The glutathione synthesis inhibitor buthionine sulfoximine (BSO) significantly inhibited the growth of DoxR cells in vitro, while having little to no effect on Control and EpiR cells (Figure 6C). Furthermore, as described in point #3 below, DoxR tumors were acutely sensitive to BSO while EpiR tumors were unresponsive (Figure 6E, Author response image 1). These new results clearly demonstrate that doxorubicin and epirubicin resistant cells and tumors have different metabolic vulnerabilities and that it is possible to exploit these to limit the growth of therapy resistant tumors.

We have modified the flow of the manuscript to highlight these new data. In order to further address this point, we have elaborated on mechanisms that may underpin these distinct metabolic alterations in the Discussion section (pages 16-17). Briefly, we posit that breast cancer cells treated with doxorubicin may face a greater oxidative challenge than those treated with epirubicin, which aligns with a greater dependence of doxorubicin-resistant cells on de novo glutathione synthesis.

Author response image 1

3. Some additional effort to assess the relevance of these resistance mechanisms in vivo. The reviewers felt that the patient correlation included in the study is contrary to the claim that specific metabolic adaptations arise to each drug, and thus some work to show the main claims hold in some in vivo context is needed.

We agree that further in vivo work would strengthen the central claims of the paper. We now present new data fulfilling this requirement.

As mentioned above, our revised manuscript shows that DoxR cells, which are dependent on glutamine for de novo glutathione synthesis, are specifically sensitive to the inhibition of this pathway by buthionine sulfoximine (BSO). in vitro, DoxR cells were more sensitive to BSO than EpiR and Control cells, at all doses tested (Figure 6C). In vivo, daily BSO treatment effectively reduced the growth of DoxR tumors, but had no significant impact on the growth of EpiR tumors over 20 days (Figure 6E, Author response image 1). These results have been added to the manuscript and are represented in Figure 6. The structure of the Results section, as well as the Discussion section, has also been modified to reflect these new results. The significance of this finding is further highlighted in a new schematic summarizing the central claims of the paper (Figure 6H).

Furthermore, we agree with the reviewers that the patient data used were not a perfect model of resistance to anthracyclines, and have removed all pathway analyses performed on this dataset from the manuscript. We also recognize that the concept that both similarities and differences exist in the mechanisms through which DoxR and EpiR cell lines are resistant to their respective anthracycline drug (doxorubicin or epirubicin) was not presented clearly. The manuscript has therefore been significantly revised to address this point.

We now show that doxorubicin- and epirubicin-resistant cells rely on distinct metabolic adaptations, while also exhibiting a considerable level of overlap (over 55%) in their signatures of differentially expressed genes compared to parental Control cells. This overlap is in line with the chemical similarity and mechanism of action of both anthracycline drugs and accordingly, both DoxR and EpiR cells display enrichment in drug clearance pathways and depletion of pathways supporting proliferation compared to Control cells. Given this overlap, both DoxR and EpiR gene expression signatures were found to overlap with the gene expression signature of patient biopsies after treatment with a multi-drug chemotherapy regimen including epirubicin. These results highlight common mechanisms of adaptation to anthracyclines as well as the clinical relevance of the EpiR and DoxR breast cancer cell models.

Despite these similarities, 44% of the gene expression signatures of DoxR and EpiR cells were different, revealing distinct metabolic pathways enriched in either model. These distinct metabolic adaptations are the focus of our story, and are further explored through transcriptomics, metabolomics, and functional genomics analyses. It is therefore accurate to state that there are both similarities and differences in the mechanisms through which DoxR and EpiR cell lines are resistant. Taken together, our experimental evidence demonstrates that (1) in alignment with previously published results, breast cancer cells rely on similar mechanisms such as increased drug efflux, elevated lysosomal activity, and activation of the NFE2L2 pathway to develop resistance to either doxorubicin or epirubicin, and that (2) breast cancer cells can adopt distinct metabolic adaptations to support resistance to either doxorubicin or epirubicin, notably dependence on glutathione metabolism for doxorubicin resistance and mitochondrial OXPHOS capacity for epirubicin resistance. Importantly, we now demonstrate that it is possible to target these different primary metabolic dependencies to limit the growth of therapy resistant cancer, both in vitro and in vivo.

Reviewer #1:

The authors select for doxorubicin and epirubicin resistant MCF7 cells in culture and determine that the global gene expression profile of these cells is similar to tumors from patients exposed to anthracylins. They then characterize the resistant cells, and argue that doxorubicin or epirubicin select for different adaptations, which is surprising given the similarity in these drugs.

1. Is the difference between doxorubicin response and epirubicin response simply drug dose? The bulk of the analysis largely compares a cells from a single dose where doxorubicin is cytostatic and epirubicin is cytotoxic. If doses of either drug are titrated to show cytostasis or cytotoxicity is there still a difference in metabolic adaptation of the resistant cells? This is only partially addressed in Figure 7, and is a crucial questions that is important to judge the central claim of the paper.

We acknowledge the concerns of the reviewer and we now provide clarifications regarding the cell line models.

The DoxR and EpiR cell models were first described by the Parissenti group in a 2008 study as outlined in our Methods section (https://dx.doi.org/10.1186%2F1471-2407-8- 318), and have been used in several publications over the years (for example: https://dx.doi.org/10.1158%2F0008-5472.CAN-16-0774, https://dx.doi.org/10.1038%2Fs41598-018-23496-y, https://bmccancer.biomedcentral.com/articles/10.1186/s12885-016-2790-3). These models were selected over hundreds of passages over a twelve-dose process. Doxorubicin-resistant cells (DoxR, originally named MCF-7DOX-2) were selected up to the highest concentration of drug at which the cells could survive (the maximally tolerated dose, 98.1 nM doxorubicin), whereas Epirubicin-resistant cells (EpiR, originally named MCF-7EPI) were still viable at the concentration reached at step twelve (852 nM epirubicin).

As shown by the Parissenti’s group, the initial acquisition of resistance to doxorubicin and epirubicin was linked to reduction in drug accumulation by the cells. However, at higher doses the development of resistance was not linked to drug exclusion by the cells. Indeed, the Figure 2 from Hembruff et al., 2008 clearly shows that the magnitude of resistance was not linked to reduction in drug accumulation after the initial acquisition, as indicated by the plateau.

Furthermore, forcing the accumulation of doxorubicin in DoxR cells using Cyclosporin A (CsA) did not significantly impact the sensitivity of DoxR cells (solid and dotted blue lines in Figure 5A, B in Hembruff et al., 2008). These data clearly highlight the importance of adaptation mechanisms separate from that of the ABC transporters.

These elements considered, and given the fact that DoxR cells are more resistant todoxorubicin (resistance factor, RF=27.8) than epirubicin (RF=4.79), and that EpiR cells are more resistant to epirubicin (RF=815.3) than doxorubicin (RF=203.4), we

hypothesized that the mechanisms of resistance may differ for doxorubicin and epirubicin, despite the chemical similarity of the drugs.

Third and finally, we addressed this difference in the dose by independently deriving cells resistant to 100nM of either doxorubicin or epirubicin. With these cells, we recapitulated the main findings from DoxR and EpiR cells, upholding the conclusions of the paper, i.e. that there are some similar adaptations, as well as unique adaptations, such as increased OXPHOS capacity in the EpiR cells and the elevated usage of glutamine for glutathione synthesis in DoxR cells.

2. There is a large literature arguing that response to chemotherapy in culture is not necessarily indicative of response in patients, and there is a growing literature around cancer metabolism also being different in the cell culture and in tumors, including glutamine metabolism. The bulk of the analysis is based on cell culture, and if some data were available that speak to whether these same differences or dependencies hold in vivo it would greatly improve the manuscript.

We thank the reviewer for this question, we agree that cell culture models can only provide a partial picture of therapeutic response and of the metabolic state. Our study reveals unique metabolic vulnerabilities in DoxR and EpiR cells, which we have further confirmed through the use of a targeted drug both in vitro and in vivo. Indeed, we now show that DoxR cells, which are dependent on glutamine to synthesize the antioxidant molecule glutathione, are specifically sensitive to the inhibition of the γ-glutamylcysteine synthetase enzyme by buthionine sulfoximine (BSO, https://doi.org/10.1016/0006-2952(84)90598-7).

In vitro, DoxR cells were more sensitive to BSO than EpiR and Control cells, at all doses tested (Figure 6C). in vivo, daily BSO treatment effectively reduced the growth of DoxR tumors, but had no significant impact on the growth of EpiR tumors over 20 days (Figure 6E, Author response image 1). The significance of this finding is further highlighted in a new schematic summarizing the central claims of the paper (Figure 6H). These results have been added to the manuscript and are represented in Figure 6.

3. The authors assess labeling of metabolites at short time points (15 min), which can be more useful for assessment of flux through pathways like glycolysis. However, this data would be much more informative if there were multiple time points assessed, as this is typically need to interpret kinetic tracing data.

We agree with the reviewer that multiple time points should be assessed for any tracing experiment, in accordance with the standard practice in the field (A roadmap for interpreting 13C metabolite labeling patterns from cells, https://dx.doi.org/10.1016%2Fj.copbio.2015.02.003). For this reason, all tracing experiments in this study were performed at multiple time points, and complete tracing data are provided in Figure 2—figure supplements 1 and 2, for glucose and glutamine tracing.

4. Seahorse assays were used to suggest ATP production rate, but this approach requires several assumptions that should be more explicitly discussed. I suggest showing the raw data, with or without the calculated rates, and feel this is particularly important for a broad audience who may not appreciate the caveats associated with this analysis.

We agree with the reviewer that these analyses entail several assumptions; this has been made more explicit in the manuscript text (page 9, last paragraph). Raw data for these analyses are OCR and ECAR data, which are presented alongside the ATP production rate results in Figure 3 (Figure 3a-c). OCR and ECAR data for all bioenergetics experiments have now been added to the manuscript, in Figures 4c and S3f (glutamine deprivation experiments) and S5a,b (new resistant models).

5. While the fact that targeting PGC1a overcomes resistance to both doxorubicin and epirubicin is potentially relevant for clinical medicine, from a mechanistic standpoint it does not support their claim that these drugs elicit different metabolic adaptations. Is there an example of a target that separates the contexts as evidence of differential resistance mechanism?

We would like to thank the reviewer for this comment and apologize for any confusion caused by the previous flow of the manuscript. This has been updated to further emphasize the fact that there are indeed similarities between the resistant models, but that this study focused on the differences in order to detect potential vulnerabilities specific to either drug. In this revised manuscript, we further show that only doxorubicin-resistant cells are sensitive to glutathione synthesis inhibition, via the γ-glutamylcysteine inhibitor BSO (buthionine sulfoximine). Compared to both Control and EpiR cells, DoxR cells were acutely sensitive to BSO treatment in vitro (Figure 5f). The growth of DoxR tumors in vivo was significantly decreased by daily injections of BSO, compared to vehicle control, whereas BSO had little to no impact on the growth of EpiR tumors (Figure 5g-i). These results are in accordance with the integrated transcriptomics and metabolomics analyses (Figure 1i), as well as the functional broad shRNA screening (Figure 1j-k) and tracing experiments (Figure 2b-f, Figure S2), showing that glutathione metabolism and glutamine flux are crucial for resistance to doxorubicin, but not to epirubicin.

Our data support the notion that PGC1a is an important player in the resistance of both drugs. PGC-1α is a well-known regulator of mitochondrial biogenesis (Wu, 1999, https://doi.org/10.1016/s0092-8674(00)80611-x), OXPHOS (Mootha, 2003, https://doi.org/10.1038/ng1180), glutamine metabolism (McGuirk, 2013, https://doi.org/10.1186/2049-3002-1-22) and glutathione synthesis (Guo, 2018, https://doi.org/10.1016/j.nbd.2018.02.004). In line, EpiR cells had a significantly greater expression of PGC-1α (Figure 3j), commensurate with their elevated mitochondrial volume and OXPHOS rates, and knockdown of PGC-1α/β also significantly decreased the expression of glutathione synthesis genes particularly in DoxR cells (Figure 3o). Knockdown experiments shown in Figure 3l-n further present evidence that PGC-1α is required for survival of both resistant lines in the presence of their respective drug.

Reviewer #2:

In their manuscript, "Tailored Metabolic Adaptations Confer Resistance to Chemotherapy in Breast Cancer", McGuirk et al. derive breast cancer cells resistant to doxorubicin and epirubicin and investigate whether the metabolic alterations that occur during resistance are dependent on the therapeutic. They suggest that doxorubicin resistant cells rely on glutamine for mitochondrial function and glutathione synthesis, while epirubicin resistant cells increase their mitochondrial capacity and ATP production. They suggest that targeting PGC1a could eradicate cells resistant to both therapeutics, as PGC1a regulates both metabolic adaptions. While the idea that different therapeutics of the same class could cause resistance due to different metabolic mechanisms, this study was limited by a lack of evidence for a causal role for the metabolic alterations identified in drug resistance.

1. It was difficult to follow the claims of the manuscript. Many statements are vague and often the figure legends and section headings/text do not match, and are sometimes contradictory in their claims.

We have revised the text of the manuscript and figure legends to ensure proper flow and clarity.

2. The authors generate doxorubicin resistant (DoxR) and epirubicin resistant (EpiR) resistant cell lines via exposure to increasing concentrations of drug. DoxR resistant cells are cultured in a final concentration of 98.1nM drug, while epirubicin resistant cells are cultured in 852nM drug. Treatment of chemonaive cells with these concentrations results in cytostatic (Dox) or cytotoxic (Epi) effects. This approach raises several concerns.

a. First, why these concentrations? These differing concentrations are a concern for the major findings of the manuscript.

We acknowledge the concerns of the reviewer and we now provide clarifications regarding the cell line models.

The DoxR and EpiR cell models were first described by the Parissenti group in a 2008 study as outlined in our Methods section (https://dx.doi.org/10.1186%2F1471-2407-8318), and have been used in several publications over the years (for example:

https://dx.doi.org/10.1158%2F0008-5472.CAN-16-0774, https://dx.doi.org/10.1038%2Fs41598-018-23496-y,

https://bmccancer.biomedcentral.com/articles/10.1186/s12885-016-2790-3). These models were selected over hundreds of passages over a twelve-dose process. Doxorubicin-resistant cells (DoxR, originally named MCF-7DOX-2) were selected up to the highest concentration of drug at which the cells could survive (the maximally tolerated dose, 98.1 nM doxorubicin), whereas Epirubicin-resistant cells (EpiR, originally named MCF-7EPI) were still viable at the concentration reached at step twelve (852 nM epirubicin).

As shown by the Parissenti’s group, the initial acquisition of resistance to doxorubicin and epirubicin was linked to reduction in drug accumulation by the cells. However, at higher doses the development of resistance was not linked to drug exclusion by the cells. Indeed, Figure 2 from Hembruff et al., 2008 clearly shows that the magnitude of resistance was not linked to reduction in drug accumulation after the initial acquisition, as indicated by the plateau.

Furthermore, forcing the accumulation of doxorubicin in DoxR cells using Cyclosporin A (CsA) did not significantly impact the sensitivity of DoxR cells (solid and dotted blue lines in Figure 5A, B in Hembruff et al., 2008). These data clearly highlight the importance of adaptation mechanisms separate from that of the ABC transporters.

These elements considered, and given the fact that DoxR cells are more resistant to doxorubicin (resistance factor, RF=27.8) than epirubicin (RF=4.79), and that EpiR cells are more resistant to epirubicin (RF=815.3) than doxorubicin (RF=203.4), we hypothesized that the mechanisms of resistance may differ for doxorubicin and epirubicin, despite the chemical similarity of the drugs.

Third and finally, we addressed this difference in the dose by independently deriving cells resistant to 100nM of either doxorubicin or epirubicin. With these cells, we recapitulated the main findings from DoxR and EpiR cells, upholding the conclusions of the paper, i.e. that there are some similar adaptations, as well as unique adaptations, such as increased OXPHOS capacity in the EpiR cells and the elevated usage of glutamine for glutathione synthesis in DoxR cells.

b. Second, anthracyclins have been shown to damage mitochondrial DNA, and the use of differing concentrations may result in different amounts of damage. Related to this point does compensatory mitochondrial biogenesis explain why PGC1a expression is increased, particularly in the EpiR cells, which are treated with significantly more anthracycline? What happens to mitochondrial content in these resistant cell lines? Mitochondrially encoded respiratory components? How does PGC1a/b knockdown affect mitochondrial content?

We thank the reviewer for this very interesting question. As stated in (a), the intracellular concentration of drug is known to be quite lower than extracellular concentrations, and given that these are established models, we do not know whether mitochondrial damage may have occurred earlier in the temporal acquisition of resistance.

Through immunofluorescence experiments, we have confirmed that EpiR cells indeed have elevated mitochondrial content compared to Control cells, commensurate with their increased reliance on OXPHOS and their elevated levels of PGC-1α expression (see Figure 3i). The mitochondrial content of DoxR cells is not significantly different from that of Control cells.

We also agree with the reviewer that the mitochondrial content would be expected to decrease upon PGC1a knockdown. However, we did not pursue the quantification of mitochondrial content in PGC1a KD cells considering that this experiment would not further enhance the narrative of the revised paper which focuses more specifically on targeting the differential metabolic vulnerability of doxorubicin-resistant cells through inhibition of glutathione synthesis.

c. Third, the final figure, in which new resistant cell lines are derived and grown in the same amount of Dox or Epi, the authors show now that while DoxR cells still increase GSH synthesis from glucose, similar increases in mitochondrial ATP and bioenergetic capacity are observed between DoxR and EpiR cells. While the effects are modestly more pronounced in EpiR cells, this finding is contrary to the thesis that Dox and Epi resistance induce different metabolic alterations as it relates to the mitochondria. These cell lines should be used throughout the study to eliminate any confounding effects of differing drug concentrations.

Please refer to our extensive description of the DoxR and EpiR cells and the different drug concentrations in response to point #2a.

It is expected that there will be some variation in the derivation of an independent model of therapeutic resistance. However, the most important point is that this new model confirms the main findings of our paper: PGC-1α is implicated in anthracycline resistance, elevated glutamine-derived de novo glutathione is a key feature of doxorubicin-resistant models, and epirubicin-resistant models have markedly elevated mitochondrial bioenergetic capacity. The reliance of doxorubicin-resistant cells on glutathione synthesis is the particular highlight of the revised manuscript, which was further confirmed in vitro and in vivo through targeted therapy with γ-glutamylcysteine inhibitor BSO (buthionine sulfoximine). Compared to both Control and EpiR cells, DoxR cells were acutely sensitive to BSO treatment in vitro (Figure 5f). The growth of DoxR tumors in vivo was significantly decreased by daily injections of BSO, compared to vehicle control, whereas BSO had little to no impact on the growth of EpiR tumors (Figure 5g-i).

One particular difference between the experimental models of resistance is that PGC-1α expression is more elevated in the MCF-7 cells resistant to 100nM doxorubicin (D100) than it is in DoxR compared to their respective parental lines, which likely explains the elevated basal OCR in D100 cells compared with DoxR cells (Figure S4b versus Figure 3j, Figure S5a versus Figure 3b). Nevertheless, Epi100 cells still display elevated basal OCR and much greater mitochondrial bioenergetic capacity (Max JATPox, Figure S4d) compared to D100 cells analogous to the EpiR cells compared with the DoxR cells (Figure S4c-f versus Figure 3b,f,g).

The figures displaying the new resistant cell lines was moved into supplementary, given that its purpose is solely to confirm the key findings in an independent model.

3. Related to the point above, it is unclear how PGC1a is promoting resistance to doxorubicin and epirubicin. In Figure 6, PGC1a associates with the antioxidant response genes only in EpiR cells, but PGC1a/b knockdown has a greater effect on gene expression in the DoxR cells.

The figure described by the reviewer was only intended to show that PGC-1α can bind to the promoters of multiple genes (ChIP-qPCR) described in the paper, particularly the glutathione metabolism pathway genes. Given the low expression of PGC-1α in Control and DoxR cells, its enrichment at the promoters of genes in these cell lines is low. The loading was clear in EpiR cells, given the relatively high (11-fold) expression of PGC-1α in this model compared to Control and DoxR cells. As pointed out by the reviewer, the magnitude of binding does not necessarily scale with gene expression. These panels have been moved into supplemental Figure S3b,c.

PGC-1α is well known to regulate numerous metabolic programs, in a context-dependent manner. For example, PGC-1α regulates mitochondrial biogenesis (Wu, 1999, https://doi.org/10.1016/s0092-8674(00)80611-x), OXPHOS (Mootha, 2003, https://doi.org/10.1038/ng1180), glutamine metabolism (McGuirk, 2013, https://doi.org/10.1186/2049-3002-1-22) and glutathione synthesis (Guo, 2018, https://doi.org/10.1016/j.nbd.2018.02.004). Indeed, EpiR cells had a significantly greater expression of PGC-1α (Figure 3j), commensurate with their elevated mitochondrial volume and OXPHOS rates. Knockdown of PGC-1α/β also significantly decreased the expression of glutathione synthesis genes particularly in DoxR cells (Figure 3o). Data shown in Figure 3l-n further present evidence that PGC-1α is required for survival of both resistant lines in the presence of their respective drug, despite different specific metabolic adaptations.

What is the effect of PGC1a knockdown alone?

Based on our experience working with the PGC-1 family of transcriptional co-activators, we typically knockdown the two main isoforms because there are usually compensation mechanisms between the two (https://doi.org/10.1111/febs.13175) due to great overlap in their functions (https://doi.org/10.1016/j.cmet.2005.05.004).

In addition, Figure 3i shows that EpiR cells have little to no elevation of antioxidant gene expression, despite significantly increased PGC1a binding to their promoters. How are these findings reconciled?

The expression of antioxidant genes was, in fact, significantly elevated when compared to Control cells for both DoxR and EpiR, albeit more modestly in the latter (1.5 – 3 fold, see Figure 2J). Specifically, NFE2L2, NQO1, GPX3, PRDX5, and CAT all have elevated expression in EpiR compared to Control. As pointed out by this reviewer in a previous point, the magnitude of binding does not necessarily scale with gene expression. As noted above, the figure detailing promoter binding has been moved to Figure S3b-c.

Further, siPGC1a/b affects the growth of DoxR and EpiR cells equally. Is this in the presence of drug? It is unclear from the figure legend.

Indeed, this experiment was performed while cells were in presence of each respective drug (EpiR with epirubicin, DoxR with doxorubicin). We state in the Methods section that these cells are always kept in presence of the drug throughout all experiments presented in the paper; for clarity we reiterated the presence of anthracyclines in the figure legend for Figure 3m as well.

Finally, is PGC1a overexpression sufficient to promote resistance of drug naïve cells? This experiment is needed to show causation.

While this experiment is certainly interesting, we focused the revised version of the manuscript on the different metabolic adaptations of EpiR and DoxR cells with a notable emphasis on the glutathione metabolism dependence of DoxR cells.

4. The authors suggest that cells with engineered resistance to Dox and Epi have similar gene expression patterns to residual tumors of breast cancer patients following anthracycline-based chemotherapy, but there are several problems with this analysis/these claims. First, it is not clear that these tumors are resistant to anthracyclines. From the description, patients received 4 cycles of anthracycline-based therapy, followed by four cycles of docetaxel. Resultant tumors may be responders, resistant to anthracyclines, resistant to docetaxel, or both. As a consequence, it is difficult to draw meaningful conclusions from the comparison. Second, the authors compare their cell line gene expression to these tumors, but tumors were treated only with epirubicin. Since the authors argue that resistance to Dox vs. Epi is via discrete mechanisms, the argument for similarity between DoxR cells and Epi "resistant" tumors is contrary to the main claims of the manuscript.

We agree with the reviewer that the patient data used were not a perfect model of resistance to anthracyclines, and have removed all pathway analyses performed on this dataset from the manuscript. We also recognize that the concept that both similarities and differences exist in the mechanisms through which DoxR and EpiR cell lines are resistant to their respective anthracycline drug (doxorubicin or epirubicin) was not presented clearly. The manuscript has therefore been significantly revised to address this point.

We now show that doxorubicin- and epirubicin-resistant cells rely on distinct metabolic adaptations, while also exhibiting a considerable level of overlap (over 55%, as shown in Figure 1d) in their signatures of differentially expressed genes compared to parental Control cells. This overlap is in line with the chemical similarity and mechanism of action of both anthracycline drugs, and accordingly both DoxR and EpiR cells display enrichment in drug clearance pathways and depletion of pathways supporting proliferation compared to Control cells. Given this overlap, both DoxR and EpiR gene expression signatures were found to overlap with the gene expression signature of patient biopsies after treatment with a multi-drug chemotherapy regimen including epirubicin. These results highlight common mechanisms of adaptation to anthracyclines as well as the clinical relevance of the EpiR and DoxR breast cancer cell models.

Despite these similarities, 44% of the gene expression signatures of DoxR and EpiR cells were different, revealing distinct metabolic pathways enriched in either model. These distinct metabolic adaptations are the focus of our story, and are further explored through transcriptomics, metabolomics, and functional genomics analyses. It is therefore accurate to state that there are both similarities and differences in the mechanisms through which DoxR and EpiR cell lines are resistant. Taken together, our experimental evidence demonstrates that (1) in alignment with previously published results, breast cancer cells rely on similar mechanisms such as increased drug efflux, elevated lysosomal activity, and activation of the NFE2L2 pathway to develop resistance to either doxorubicin or epirubicin, and that (2) breast cancer cells can adopt distinct metabolic adaptations to support resistance to either doxorubicin or epirubicin, notably dependence on glutathione metabolism for doxorubicin resistance and mitochondrial OXPHOS capacity for epirubicin resistance. Importantly, we now demonstrate that it is possible to target these different primary metabolic dependencies to limit the growth of therapy resistant cancer, both in vitro and in vivo.

Indeed, the claim that "several distinct metabolic pathways were altered by doxorubicin or epirubicin resistance (Figure 1f,h)" is problematic due to these concerns since (1) these pathways are based on overlap with gene expression from tumors from patients that may or may not be resistant and (2) these tumors were treated only with epirubicin. Therefore, the basis for distinct metabolic pathways is unsound.

As stated above, in agreement with the reviewer, we have removed the pathway analysis from the manuscript.

5. The increase in GCLC, ME1, NQO1 etc in DoxR cells is consistent with NRF2 activation as proposed, as is the increase in GSH M+5, which suggests an increase in GSH synthesis. However, the decrease in the GSH/GSSG ratio and NADPH, and increase in NADP+ is not consistent, nor is it consistent with the protection against ROS shown in Figures 3j and 3k. How is this reconciled?

We thank the reviewer for this comment. In order to validate our results, we quantified GSH and GSSG using an alternative method that is particularly suited for metabolites that are easily oxidized like GSH. These new analyses show that the GSH:GSSG ratio in DoxR is higher than Control and EpiR cells. This novel extraction method revealed that there was oxidation of GSH in our original datasets, explaining the considerable variability in the absolute amounts of GSH and the GSH/GSSG ratios. Stable isotope tracing experiments were unaffected by this oxidation, and there was little variability in the fractional enrichment measured across all experiments. The new results have been incorporated in Figure 2H. These data align with the decrease in NADPH:NADP ratio (Figure S3g), likely due to increased Glutathione Reductase activity, and high GSH levels providing a protection against ROS.

What is the total pool size of GSH + GSSG?

The total pool of glutathione is bigger in drug-resistant cells and as expected, DoxR cells have the biggest pool (see Figure 2G).

Related to this, what is the "control" for figures 3J and 3k? Parental cells?

We apologize for the confusion. The parental cells are called Control, and we ensure this word was capitalized in the legend to indicate that this refers to the cell line. For clarity, we have changed the reference to “Control” in the legend to “Control cells”.

6. The data shown in Figure 4 is not supportive of an increased dependence of DoxR cells on glutamine, as claimed in the figure legend. Glutamine is used for many purposes in cells and neither DoxR nor EpiR cells grow without glutamine in Figure 4A. Further, the idea that glutamine is driving TCA cycle is not matched by the tracing data from Figure 3. There is no increased entry of glucose or glutamine into the TCA cycle in the DoxR cells (e.g. citrate m+2 from glucose, citrate m+4 from glutamine).

We have corrected the figure legend to clarify that DoxR cells have increased dependence on glutamine for mitochondrial ATP production. Indeed, while it is true that the proliferation of both DoxR and EpiR cells is arrested upon glutamine withdrawal (Figure 4a), only DoxR cells face a severe bioenergetic challenge as a result. Their bioenergetic capacity is drastically reduced (Figure 4g-j), and their basal mitochondrial ATP production is significantly decreased (Figure 4f).

Figure 3 shows that glutamine is utilized at similar levels in the TCA in DoxR, EpiR, and Control cells, There is indeed little evidence of increased entry of glutamine into the TCA in DoxR compared to Control cells, albeit a slight increase in glutamine flux to succinate in the mitochondria shown in time-course tracing diagrams in Figure S3c. Figure 4, on the other hand, includes a functional assay which demonstrates that while glutamine withdrawal impacts the growth of all cell types, it specifically and significantly impairs mitochondrial ATP production in DoxR cells – a bioenergetic process that is dependent on the TCA cycle.

Regarding the entry of glucose into the TCA cycle, glucose carbons are indeed diminished at m+2 citrate in DoxR cells compared to Control cells. However, there is increased m+3 labeling in citrate, possibly due to pyruvate carboxylase activity. In addition, m+3 malate is more elevated than m+2 malate is decreased, showing a small but net increase in glucose flux into the TCA cycle in DoxR cells compared to Control cells (see Figures 3c-d and S3a).

Related to this, the inability of DoxR cells to compensate for ATP production upon glutamine withdrawal suggests a defect in glycolytic regulation or ATP sensing, rather than glutamine metabolism. More clarity is needed for (1) what metabolic alterations occur, (2) how they are regulated and (3) how they promote anthracycline resistance.

We thank the reviewer for the suggestions related to glycolytic regulation and ATP sensing, and have now elaborated on these results in the Results (page 12, paragraph 2) and Discussion (page 16, paragraph 1) sections of the manuscript. We did not pursue further experiments on this aspect as the revised version of the manuscript is now centred on the dependence of DoxR cells on glutathione metabolism considering the editorial and reviewers’ recommendations.

Reviewer #3:

In "Tailored Metabolic Adaptations Confer Resistance to Chemotherapy in Breast Cancer", McGuirk et al. use MCF-7 cells in culture to study mechanisms of resistance to doxorubicin and epirubicin. Overall, I found their approach to be intriguing. It seems like a compelling approach to study drug resistance in cancer. They subjected cells to many passages in the presence of drug, which resulted in a population of resistant cells. They could then compare these resulting resistant cells to the original drug-susceptible cells. In doing so, the authors claim to have found different metabolic adaptations to each drug. Given the differences in adaptation, they suggest that it might be better to target resistant tumors with drugs that hit global regulators of metabolism, like PGC-1a.

I have two major questions.

1. How do the authors know that the genes and metabolic adaptations assumed in the resistant cells are essential to drug resistance?

We show several lines of evidence that support the role of metabolic genes in sustaining drug resistance. In our broad shRNA screen (Figure 1j), in which depleted shRNA barcodes indicate gene targets whose knockdown impairs growth and/or viability, glutathione metabolism genes were found to be particular targets in DoxR cells (Figure 1k). It was further determined through functional assays that glutamine carbons are highly utilized for glutathione synthesis (Figure 2f) and are necessary to sustain mitochondrial ATP production specifically in DoxR cells (Figure 4c,d,f,g,i). Finally, we have performed new experiments and found that only doxorubicin-resistant cells are sensitive to glutathione synthesis inhibition, via the γ-glutamylcysteine inhibitor BSO (buthionine sulfoximine). Compared to both Control and EpiR cells, DoxR cells were acutely sensitive to BSO treatment in vitro (Figure 5f). The growth of DoxR tumors in vivo was significantly decreased by daily injections of BSO, compared to vehicle control, whereas BSO had little to no impact on the growth of EpiR tumors (Figure 5g-i). The significance of this finding is further highlighted in a new schematic summarizing the central claims of the paper (Figure 6H).

On the other hand, the broad shRNA screen found that oxidative phosphorylation genes were key knockdown targets impairing the growth and/or viability of EpiR cells. We confirmed through functional assays that EpiR cells have elevated mitochondrial respiration rates compared to both Control and DoxR cells (Figure 3a,b), and that they have a markedly increased mitochondrial volume (Figure 3i) and OXPHOS capacity (Figure 3f-h).

We proposed that PGC-1α contribute to these distinct adaptations, given that it is known to regulate mitochondrial biogenesis (Wu, 1999, https://doi.org/10.1016/s00928674(00)80611-x), OXPHOS (Mootha, 2003, https://doi.org/10.1038/ng1180), and glutathione synthesis (Guo, 2018, https://doi.org/10.1016/j.nbd.2018.02.004). PGC-1α is also well known to have context-dependent roles in different tissues, cancer types, and in response to different treatments / stimuli (https://dx.doi.org/10.3389%2Ffonc.2018.00075). Indeed, EpiR cells had a significantly greater expression of PGC-1α (Figure 3j), commensurate with their elevated mitochondrial volume and OXPHOS rates, and knockdown of PGC-1α/β also significantly decreased the expression of glutathione synthesis genes particularly in DoxR cells (Figure 3o). Knockdown experiments shown in Figure 3l-n further present evidence that PGC-1α is required for survival of both resistant lines in the presence of their respective drug, despite different specific metabolic adaptations downstream.

Might it be the case that their original control population of cells was heterogeneous and exposure of cells to drug over long periods of time simply selected those subclones that were resistant? In such a model, while one gene/metabolic phenotype may be essential to developing resistance (perhaps ABC transporters), the other genes/metabolic phenotypes could be random. Indeed, based on the limited amount of data shown from two experiments, it seems like the metabolic phenotypes of cells exposed to dox for long periods of time are different in separate experiments. For example, the isotope labeling shown in Figure 3 looks very different from the labeling shown in Figure 7. A recent study by Speirs and Price et al. ( doi: 10.18632/oncotarget.26533 ) did a similar experiment without drugs. After selecting subclones of cells from a culture, they found significant differences in metabolic phenotypes. It would be most interesting to compare the genetic and metabolic differences of resistance cells that were selected independently by drugs in separate experiments. This could help better resolve whether the differences observed are essential.

We thank the reviewer for this important comment; this is indeed why we developed the secondary model cell lines used in Figure 7 (now, moved to Figures S4 and S5). The models that we used for most of the study (DoxR, and EpiR) were selected by the Parissenti group. These models were selected over hundreds of passages over a twelvedose process. Doxorubicin-resistant cells (DoxR, originally named MCF-7DOX-2) were selected up to the highest concentration of drug at which the cells could survive (the maximally tolerated dose, 98.1 nM doxorubicin), whereas Epirubicin-resistant cells (EpiR, originally named MCF-7EPI) were still viable at the concentration reached at step twelve (852 nM epirubicin). Control cells (originally named MCF-7CC) were developed in parallel, through passaging with a constant dose of DMSO (https://dx.doi.org/10.1186%2F1471-2407-8-318). A subclonal selection likely occurred over time through long-term passaging of these cells, either through adaptation of certain cells to the drug or due to inherent resistance in subclones. For this reason, we expected that there may be some differences between these cells and those we derived. Indeed, when comparing differential gene expression profiles between the two models of doxorubicin resistance, and between the two models of epirubicin resistance, there is a large 66% overlap in both cases (see Author response image 2). Nevertheless, one-third of differentially expressed genes are distinct between D100 and DoxR, and between E100 and EpiR (see Author response image 2). Importantly and despite these clonal differences, these new cells confirmed that glutamine metabolism through glutathione is specifically important in doxorubicin-resistant cells, and that epirubicin-resistant cells have especially elevated OXPHOS bioenergetic capacity. The metabolic differences in doxorubicin and epirubicin resistant cells is key for their resistance and this is well illustrated by the fact that doxorubicin resistant cells are more sensitive than epirubicin resistant cells to the drug BSO, that interferes with glutathione synthesis (see graphs in Figure 6).

Author response image 2
Venn diagrams of differential gene expression, comparing D100 vs Ctl and DoxR vs Control cells (left) and comparing E100 vs Ctl and EpiR vs Control cells (right).

2. The major premise of this manuscript is that the authors have selected cells resistant to dox and epi. Data shown in Figure 1B suggest that this is the case over 6 days. Based on the data shown in Figure 4A, however, I worry that the Epi-R cells are not resistant to Epi over long periods of time. For example, at day 11, the Epi-R live cell count looks more like the control cells in Figure 1 than DoxR. Thus, if the cells are not both equally resistant, maybe the metabolic adaptations are not really "distinct" but rather just reflective of different degrees of resistance.

We acknowledge the concerns of the reviewer and we now provide clarifications

regarding the cell line models.

The DoxR and EpiR cell models were first described by the Parissenti group in a 2008 study as outlined in our Methods section (https://dx.doi.org/10.1186%2F1471-2407-8-318), and have been used in several publications over the years (for example:

https://dx.doi.org/10.1158%2F0008-5472.CAN-16-0774, https://dx.doi.org/10.1038%2Fs41598-018-23496-y, https://bmccancer.biomedcentral.com/articles/10.1186/s12885-016-2790-3). These models were selected over hundreds of passages over a twelve-dose process. Doxorubicin-resistant cells (DoxR, originally named MCF-7DOX-2) were selected up to the highest concentration of drug at which the cells could survive (the maximally tolerated dose, 98.1 nM doxorubicin), whereas Epirubicin-resistant cells (EpiR, originally named MCF-7EPI) were still viable at the concentration reached at step twelve (852 nM epirubicin).

As shown by the Parissenti’s group, the initial acquisition of resistance to doxorubicin and epirubicin was linked to reduction in drug accumulation by the cells. However, at higher doses the development of resistance was not linked to drug exclusion by the cells. Indeed, Figure 2 from Hembruff et al., 2008 clearly shows that the magnitude of resistance was not linked to reduction in drug accumulation after the initial acquisition, as indicated by the plateau.

Furthermore, forcing the accumulation of doxorubicin in DoxR cells using Cyclosporin A (CsA) did not significantly impact the sensitivity of DoxR cells (solid and dotted blue lines in Figure 5A, B in Hembruff et al., 2008). These data clearly highlight the importance of adaptation mechanisms separate from that of the ABC transporters.

These elements considered, and given the fact that DoxR cells are more resistant to doxorubicin (resistance factor, RF=27.8) than epirubicin (RF=4.79), and that EpiR cells are more resistant to epirubicin (RF=815.3) than doxorubicin (RF=203.4), we hypothesized that the mechanisms of resistance may differ for doxorubicin and epirubicin, despite the chemical similarity of the drugs.

Third and finally, we addressed this difference in the dose by independently deriving cells resistant to 100nM of either doxorubicin or epirubicin. With these cells, we recapitulated the main findings from DoxR and EpiR cells, upholding the conclusions of the paper, i.e. that there are some similar adaptations, as well as unique adaptations, such as increased OXPHOS capacity in the EpiR cells and the elevated usage of glutamine for glutathione synthesis in DoxR cells.

To further address the reviewer’s concern and confirm that these cells are indeed stably resistant, we performed a drug holiday experiment. After 7 weeks of proliferation without drug, both DoxR and EpiR cells fully retained their level of resistance when re-exposed to 98.1 nM doxorubicin and 852 nM epirubicin, respectively – see Figure 1—figure supplement 1A in the resubmitted manuscript.

Other points:

1. Can the authors do any experiments to explore (or possibly speculate in the discussion) why the two drugs would elicit different resistance mechanisms?

We thank the reviewer for this point, and have elaborated on this further in the Discussion section of the manuscript (pages 16-17). Briefly, we describe that structural differences between these drugs may play a role, as they may lead to distinct on- and off-target effects including differential rates of drug-induced ROS production. Cardiotoxicity, a common side effect of anthracyclines, is linked to oxidative stress and epirubicin has been shown to induce less cardiotoxic effects than doxorubicin, even if both drugs display equivalent response rate to treat breast cancer (https://doi.org/10.1159/000500204). Breast cancer cells treated with doxorubicin may therefore face a greater oxidative challenge than those treated with epirubicin, which aligns with a greater dependence of doxorubicin-resistant cells on de novo glutathione synthesis.

2. The increase of pyruvate carboxylase in EpiR cells is interesting. What is the phenotypic value of such a metabolic alteration in resistance?

As shown in Figure 2A, expression of pyruvate carboxylase (PC) is increased in DoxR compared to Control (not EpiR, as the reviewer suggests). We suggest that this may contribute to resistance by serving as an alternate pathway for refueling the citric acid cycle, perhaps allowing excess glutamine / glutamate to be used instead for GSH synthesis.

3. The data do not seem to support this statement: "DoxR cells were unable to increase glycolytic ATP production to compensate for diminished oxidative ATP production upon glutamine withdrawal." The Seahorse plots in Figure 4 actually show that glycolytic ATP production increases in drug resistance.

Figure 4f shows that glycolytic ATP production (JATPglyc; panel e) remains unchanged after glutamine withdrawal in DoxR, whereas mitochondrial ATP production (JATPox; panel d) is significantly decreased. These results support that DoxR cells are not able to increase their glycolytic ATP production, resulting to a net reduction in total ATP

production (panel f).

4. It would be helpful to see all of the isotopologues, instead of just a select number. For example, M+3 in malate from 13C-glucose doesn't necessarily mean pyruvate carboxylase activity. It could also be indicative of two rounds of TCA cycle (the first with labeled acetyl-CoA and the second with unlabeled). It would be easier to assess these kinds of possibilities if the authors presented full isotopologue plots.

We agree with the reviewer. In the main figures we show only a few isotopologues for simplicity. We have now provided these data in Figures S2b and S2d (corresponding to tracing experiments in Figure 2) as well as in Figure S5c (corresponding to tracing experiments in Figure S4g).

5. Many of the interesting data shown are not explored (or discussed). For example, in Figure 5C-D, the DoxR cells have uncoupled mitochondria. Why is this? Is this the source of ROS? It might be best to remove data that are not discussed, as it is distracting from the overall message.

We agree with the reviewer that the uncoupled respiration in resistant cells is interesting and we have elaborated on this further in the Discussion section of the manuscript (pages 16-17). We describe that doxorubicin are associated with higher induction of ROS than epirubicin (https://doi.org/10.1074/jbc.m508343200) and that, accordingly, epirubicin has been shown to induce less cardiotoxic effects than doxorubicin (https://doi.org/10.1159/000500204). Breast cancer cells treated with doxorubicin may therefore face a greater oxidative challenge than those treated with epirubicin, which aligns with a greater dependence of doxorubicin-resistant cells on de novo glutathione synthesis. Accordingly, doxorubicin-resistant cells also displayed much greater engagement of oxidative stress response than epirubicin-resistant cells. Comparatively, EpiR cells (not DoxR, as the reviewer suggests) displayed an elevated level of uncoupled respiration, which may represent an alternate approach to minimizing ROS production in this model; uncoupled respiration can be induced by ROS, and can play a role in further minimizing ROS in the mitochondria (https://doi.org/10.1038/415096a). To further outline the level of uncoupled respiration in these cell lines, we provide Author response image 3 displaying the percentage of total oxygen consumption rate (OCR) attributed to uncoupled respiration.

Author response image 3

[Editors’ note: what follows is the authors’ response to the second round of review.]

Essential revisions:

1. Concern remains regarding how relevant the metabolic differences between these two cell populations are to resistance in patients given that different drug concentrations were used to derive the EpiR and DoxR cells and that the new in vivo data did not use a de novo resistance model, but rather the same cell lines used previously. The use of independently derived cell lines should be used to confirm more of the central findings of the manuscript. In particular, please address whether the different metabolic adaptations are related to the dose or the drug itself.

We would like to thank the reviewers for this constructive comment. Acknowledging these concerns, we have provided further confirmation of our main findings in our independently derived cell lines. In addition to previously corroborated findings that epirubicin-resistant cells display increased OXPHOS capacity and that doxorubicin-resistant cells have an elevated usage of glutamine for glutathione synthesis, these data are now described more prominently in the Results section of the manuscript and in the main figures (Figures 5 and 6).

Specifically, we have recapitulated in vitro that independently derived doxorubicin-resistant cells (D100) are also specifically vulnerable to inhibition of the glutathione synthesis pathway by buthionine sulfoximine (BSO), as they display a significantly greater proliferation inhibition to BSO when compared to epirubicin-resistant cells (Figure 6c,d). We also confirmed that independently derived doxorubicin-resistant cells display a lower ROS signal than epirubicin-resistant cells, both at baseline and after H2O2 treatment (Figure 5d,e). Finally, following point #3 below and in line with elevated OXPHOS activity in both epirubicin-resistant cell models (EpiR and independently derived E100) compared to drug-sensitive and doxorubicin-resistant cells, we now show that both epirubicin-resistant cell models are specifically sensitive to inhibition of OXPHOS using the biguanide phenformin (Figure 6a,b).

Thereby, we show in two independently derived models not only that distinct metabolic adaptations support resistance to doxorubicin or epirubicin, but also that these adaptations can be specifically targeted using metabolic interventions. Given that these independently derived cell lines (D100 and E100 cells) were selected to a common end-point dose of 100nM of doxorubicin or epirubicin, these data further demonstrate that these metabolic adaptations are specific to the drug, not the dose. New panels (d and e) in Figure 5.

2. The tumor growth differences are not impressive when considering that the DoxR cells grow much more slowly than the EpiR cells. The difference in size even for the DoxR tumors does not seem that great, and is presented as fold change. Please at least acknowledge this, consider presenting the data in a more fair way with tumor size data graphed as actual measured size with the DoxR and EpiR curves on the same plot or plots with the same scales.

We acknowledge that our original data presentation did not show the DoxR and EpiR tumors on the same scale, and now show this data as requested by the reviewers with the actual measured sizes of DoxR and EpiR tumors shown on the same plot (Figure 6f). Importantly, while all EpiR tumors are larger than DoxR tumors, our data clearly show that both types of tumors doubled in size over 20 days when treated with vehicle (from 100mm3 to 200mm3 for DoxR tumors, and from 325mm3 to 650mm3 for EpiR tumors, Figures 6g). Growth of DoxR tumors was significantly impaired by BSO treatment, as tumor size increased by only 30% over 20 days, whereas there was little impact of BSO on EpiR tumor growth (Figure 6f-g). Our conclusions remain the same, as daily BSO treatment effectively reduced the growth of DoxR tumors, but had no significant impact on the growth of EpiR tumors over 20 days.

3. Please test (in vitro) whether the oxidative stress response in control cells treated with either doxorubicin or epirubicin is different as a potential mechanism for how treatment with these drugs differs in a way that might impact metabolism. Please also considering examining whether Epic cells are more sensitive to OXPHOS inhibitors to consider whether PGC1a promoting the OXPHOS phenotype in epi resistant cells speaks to the divergent metabolism.

We thank the reviewers for these insightful suggestions and agree that these data would greatly complement our findings. Accordingly, we have performed new experiments showing that, in line with their elevated bioenergetic capacity and mitochondrial ATP production rates, both epirubicin-resistant cell models (EpiR and E100) are significantly more sensitive to inhibition of OXPHOS using the biguanide phenformin than doxorubicin-resistant cells (DoxR and D100 cells, Figures 6a-b).

We also performed CM-H2DCFDA experiments to test whether different anthracyclines induce different levels of oxidative stress in drug-sensitive Control cells. While there was a significant induction of oxidative stress when these cells were acutely challenged with either drug, we did not see any significant differences between doxorubicin and epirubicin treated cells. The lack of difference in terms of oxidative stress upon acute exposure to doxorubicin and epirubicin is perhaps not surprising in light of the fact that both resistant models used in our study were selected through a sustained and stepwise increase in drug dose over the course of 8-12 months. Indeed, while oxidative stress upon acute exposure (2-3 days) appears similar with both anthracycline drugs, it likely plays a more critical role under chronic treatment conditions and in the development of resistance to doxorubicin compared to epirubicin as demonstrated by our results using transcriptomics and metabolomics, and validated independently by performing a targeted shRNA screen. The Discussion section of the manuscript has been modified to address these points.

4. Please address the issues with the resistance mechanisms in vivo based on the following comment from Reviewer #3. The in vivo BSO experiment is supportive of the cell culture experiment but there are multiple issues with this experiment. First, no anthracycline is included in this experiment making its relevance to cancer therapy questionable. Second, resistant lines are used from the start, rather than assaying the role of GSH synthesis in de novo resistance that arises in vivo, which would have been more relevant to the patient situation.

The authors thank the reviewer for this comment.

The in vivo BSO experiments conducted during the first round of revisions, to address a specific comment made by the reviewers and editors, were designed to directly address whether doxorubicin-resistant tumors are specifically vulnerable to inhibition of the glutathione synthesis pathway using BSO. The experimental design we adopted enabled us to control for several potentially confounding variables, thereby increasing confidence in the results. First, we wanted to avoid the use of anthracycline drugs in mice given that the cell lines used are resistant to very high doses of anthracyclines, which would be highly toxic to mice. Hence, prior to in vivo experiments, we grew the resistant cell lines in the absence of anthracyclines in vitro, as would be the situation during the xenograft experiments in vivo, and confirmed that the cell models used (DoxR and EpiR) retain their resistance to their respective drug for prolonged periods of time, as shown in Figure S1a. Our experimental design also controlled for

variability between mice, by growing DoxR and EpiR cells in the same mouse. Furthermore, the results of this experiment highlight a potentially clinically relevant finding. Given that anthracycline treatments are linked to significant side effects in patients such as irreversible cardiotoxicity which limits their use to a restrictive cumulative total lifetime dose, there is important clinical relevance in reducing tumor growth in anthracycline-resistant patients through a secondary treatment option without administrating additional anthracycline chemotherapy.

We agree with the reviewer that assaying the role of GSH synthesis in de novo resistance that arises in vivo would provide interesting findings. However, this proposed experiment represents a very significant endeavor which the authors consider to be outside the scope of the present study.

https://doi.org/10.7554/eLife.65150.sa2

Article and author information

Author details

  1. Shawn McGuirk

    1. Department of Biochemistry, Faculty of Medicine, McGill University, Montreal, Canada
    2. Goodman Cancer Research Centre, McGill University, Montreal, Canada
    Contribution
    Conceptualization, Resources, Data curation, Formal analysis, Funding acquisition, Validation, Investigation, Visualization, Methodology, Writing - original draft, Project administration, Writing - review and editing
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0002-7183-7962
  2. Yannick Audet-Delage

    1. Department of Biochemistry, Microbiology, and Immunology, University of Ottawa, Ottawa, Canada
    2. Ottawa Institute of Systems Biology, Ottawa, Canada
    Contribution
    Formal analysis, Validation, Investigation, Visualization, Methodology, Writing - original draft, Project administration, Writing - review and editing
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0002-8467-6168
  3. Matthew G Annis

    1. Goodman Cancer Research Centre, McGill University, Montreal, Canada
    2. Department of Medicine, Faculty of Medicine, McGill University, Montreal, Canada
    Contribution
    Investigation, Methodology, Writing - review and editing
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0002-8776-004X
  4. Yibo Xue

    1. Department of Biochemistry, Faculty of Medicine, McGill University, Montreal, Canada
    2. Goodman Cancer Research Centre, McGill University, Montreal, Canada
    Contribution
    Investigation, Methodology
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0003-4252-4446
  5. Mathieu Vernier

    Goodman Cancer Research Centre, McGill University, Montreal, Canada
    Contribution
    Formal analysis, Investigation, Visualization, Methodology, Writing - review and editing
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0001-9356-7353
  6. Kaiqiong Zhao

    1. Department of Epidemiology, Biostatistics and Occupational Health, McGill University, Montreal, Canada
    2. Lady Davis Institute, Jewish General Hospital, Montreal, Canada
    Contribution
    Formal analysis, Visualization
    Competing interests
    No competing interests declared
  7. Catherine St-Louis

    1. Department of Biochemistry, Microbiology, and Immunology, University of Ottawa, Ottawa, Canada
    2. Ottawa Institute of Systems Biology, Ottawa, Canada
    Contribution
    Investigation, Writing - review and editing
    Competing interests
    No competing interests declared
  8. Lucía Minarrieta

    1. Department of Biochemistry, Microbiology, and Immunology, University of Ottawa, Ottawa, Canada
    2. Ottawa Institute of Systems Biology, Ottawa, Canada
    Contribution
    Investigation, Writing - review and editing
    Competing interests
    No competing interests declared
  9. David A Patten

    1. Department of Biochemistry, Microbiology, and Immunology, University of Ottawa, Ottawa, Canada
    2. Ottawa Institute of Systems Biology, Ottawa, Canada
    Contribution
    Formal analysis, Validation, Writing - review and editing
    Competing interests
    No competing interests declared
  10. Geneviève Morin

    1. Department of Biochemistry, Faculty of Medicine, McGill University, Montreal, Canada
    2. Goodman Cancer Research Centre, McGill University, Montreal, Canada
    Contribution
    Methodology
    Competing interests
    No competing interests declared
  11. Celia MT Greenwood

    1. Department of Epidemiology, Biostatistics and Occupational Health, McGill University, Montreal, Canada
    2. Lady Davis Institute, Jewish General Hospital, Montreal, Canada
    3. Department of Human Genetics, McGill University, Montreal, Canada
    4. Gerald Bronfman Department of Oncology, Montreal, Canada
    Contribution
    Resources, Supervision, Methodology, Writing - review and editing
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0002-2427-5696
  12. Vincent Giguère

    1. Department of Biochemistry, Faculty of Medicine, McGill University, Montreal, Canada
    2. Goodman Cancer Research Centre, McGill University, Montreal, Canada
    Contribution
    Resources, Supervision, Funding acquisition, Writing - review and editing
    Competing interests
    No competing interests declared
  13. Sidong Huang

    1. Department of Biochemistry, Faculty of Medicine, McGill University, Montreal, Canada
    2. Goodman Cancer Research Centre, McGill University, Montreal, Canada
    Contribution
    Resources, Formal analysis, Supervision, Funding acquisition, Visualization, Methodology, Writing - review and editing
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0002-2838-4726
  14. Peter M Siegel

    1. Goodman Cancer Research Centre, McGill University, Montreal, Canada
    2. Department of Medicine, Faculty of Medicine, McGill University, Montreal, Canada
    Contribution
    Resources, Supervision, Funding acquisition, Writing - review and editing
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0002-5568-6586
  15. Julie St-Pierre

    1. Department of Biochemistry, Faculty of Medicine, McGill University, Montreal, Canada
    2. Goodman Cancer Research Centre, McGill University, Montreal, Canada
    3. Department of Biochemistry, Microbiology, and Immunology, University of Ottawa, Ottawa, Canada
    4. Ottawa Institute of Systems Biology, Ottawa, Canada
    Contribution
    Conceptualization, Resources, Supervision, Funding acquisition, Visualization, Writing - original draft, Project administration, Writing - review and editing
    For correspondence
    julie.st-pierre@uottawa.ca
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0002-2815-7099

Funding

Canadian Institutes of Health Research (Vanier Scholarship)

  • Shawn McGuirk

Fonds de Recherche du Québec - Santé (Doctoral Training Award)

  • Shawn McGuirk

McGill University (Canderel Studentship Award)

  • Shawn McGuirk

Fonds de Recherche du Québec - Santé (Postdoctoral Training Award)

  • Yannick Audet-Delage

McGill University (Charlotte & Leo Karassik Foundation Oncology Fellowship)

  • Yibo Xue

McGill University (Rolande & Marcel Gosselin Graduate Studentship)

  • Yibo Xue

Fonds de Recherche du Québec - Santé (Doctoral Training Award)

  • Kaiqiong Zhao

McGill University (Gerald Clavet Award)

  • Kaiqiong Zhao

Fonds de Recherche du Québec - Santé (Salary Award)

  • Julie St-Pierre

Canada Research Chairs (Tier 1 - Cancer and Metabolism)

  • Julie St-Pierre

Canadian Institutes of Health Research (MOP-106603)

  • Julie St-Pierre

Canadian Institutes of Health Research (PJT-148650)

  • Peter M Siegel
  • Julie St-Pierre

Canadian Institutes of Health Research (MOP-130540)

  • Sidong Huang

Terry Fox Research Institute (242122)

  • Vincent Giguère
  • Peter M Siegel
  • Julie St-Pierre

Quebec Breast Cancer Foundation (Grant with TFRI)

  • Vincent Giguère
  • Peter M Siegel
  • Julie St-Pierre

The funders had no role in study design, data collection and interpretation, or the decision to submit the work for publication.

Acknowledgements

SM was recipient of a Vanier Canada Graduate Scholarship (Canadian Institutes of Health Research, CIHR), Doctoral Training Award (Fonds de Recherche du Québec – Santé, FRQS), Canderel Studentship Award (Goodman Cancer Research Centre, GCRC). YAD was supported by a Postdoctoral Training Award from FRQS. YX was supported by Rolande and Marcel Gosselin Graduate Studentship and Charlotte and Leo Karassik Foundation Oncology Fellowship. KZ was supported by a Doctoral Training Award (FRQS) and Gerald Clavet award (Faculty of Medicine, McGill University). JSP received salary support from FRQS and Canada Research Chair in Cancer and Metabolism. This work was supported by grants from CIHR (MOP-106603 to JSP; PJT-148650 to PS and JSP; MOP-130540 to SH) and Terry Fox Research Institute and Québec Breast Cancer Foundation (#242122 to JSP, PS, and VG). We acknowledge contributions from the Metabolomics Core Facility (MCF) of the GCRC, as well as technical assistance from Daina Avizonis, Mariana De Sa Tavares Russo, Gaëlle Bridon, and Luc Choinière. The MCF is funded by the John R and Clara M Fraser Memorial Trust, Terry Fox Research Institute and Québec Breast Cancer Foundation (#242122 to JSP, PS, and VG), and McGill University. The authors thank Amadeo Parissenti for providing resistant cell lines, and Simon-Pierre Gravel, Ouafa Najyb, David Papadopoli, Sylvia Andrzejewski, Valérie Chénard, Tina Gruosso, Uri David Akavia, Russell G Jones, and Nicole Beauchemin for thoughtful discussions. The authors extend special thanks to the staff, students, and fellows of McGill University and of the University of Ottawa who enabled a safe environment to complete this study during the COVID-19 pandemic.

Ethics

Animal experimentation: Mice were housed in facilities managed by the McGill University Animal Resources Centre and all animal experiments were conducted under a University approved animal use protocol (AUP2012-5129) in accordance with guidelines established by the Canadian Council on Animal Care.

Senior Editor

  1. Utpal Banerjee, University of California, Los Angeles, United States

Reviewing Editor

  1. Matthew G Vander Heiden, Massachusetts Institute of Technology, United States

Publication history

  1. Received: November 24, 2020
  2. Accepted: June 12, 2021
  3. Version of Record published: June 28, 2021 (version 1)

Copyright

© 2021, McGuirk et al.

This article is distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use and redistribution provided that the original author and source are credited.

Metrics

  • 764
    Page views
  • 86
    Downloads
  • 0
    Citations

Article citation count generated by polling the highest count across the following sources: Crossref, PubMed Central, Scopus.

Download links

A two-part list of links to download the article, or parts of the article, in various formats.

Downloads (link to download the article as PDF)

Download citations (links to download the citations from this article in formats compatible with various reference manager tools)

Open citations (links to open the citations from this article in various online reference manager services)

  1. Further reading

Further reading

    1. Cancer Biology
    2. Genetics and Genomics
    Andrew S McNeal et al.
    Research Article

    Benign melanocytic nevi frequently emerge when an acquired BRAFV600E mutation triggers unchecked proliferation and subsequent arrest in melanocytes. Recent observations have challenged the role of oncogene-induced senescence in melanocytic nevus formation, necessitating investigations into alternative mechanisms for the establishment and maintenance of proliferation arrest in nevi. We compared the transcriptomes of melanocytes from healthy human skin, nevi, and melanomas arising from nevi and identified a set of microRNAs as highly expressed nevus-enriched transcripts. Two of these microRNAs—MIR211-5p and MIR328-3p—induced mitotic failure, genome duplication, and proliferation arrest in human melanocytes through convergent targeting of AURKB. We demonstrate that BRAFV600E induces a similar proliferation arrest in primary human melanocytes that is both reversible and conditional. Specifically, BRAFV600E expression stimulates either arrest or proliferation depending on the differentiation state of the melanocyte. We report genome duplication in human melanocytic nevi, reciprocal expression of AURKB and microRNAs in nevi and melanomas, and rescue of arrested human nevus cells with AURKB expression. Taken together, our data describe an alternative molecular mechanism for melanocytic nevus formation that is congruent with both experimental and clinical observations.

    1. Cancer Biology
    2. Cell Biology
    Akikazu Harada et al.
    Research Article Updated

    Pancreatic cancer has a high mortality rate due to metastasis. Whereas KRAS is mutated in most pancreatic cancer patients, controlling KRAS or its downstream effectors has not been succeeded clinically. ARL4C is a small G protein whose expression is induced by the Wnt and EGF–RAS pathways. In the present study, we found that ARL4C is frequently overexpressed in pancreatic cancer patients and showed that its localization to invasive pseudopods is required for cancer cell invasion. IQGAP1 was identified as a novel interacting protein for ARL4C. ARL4C recruited IQGAP1 and its downstream effector, MMP14, to invasive pseudopods. Specific localization of ARL4C, IQGAP1, and MMP14 was the active site of invasion, which induced degradation of the extracellular matrix. Moreover, subcutaneously injected antisense oligonucleotide against ARL4C into tumor-bearing mice suppressed metastasis of pancreatic cancer. These results suggest that ARL4C–IQGAP1–MMP14 signaling is activated at invasive pseudopods of pancreatic cancer cells.