Human HR+/HER2- BC cell lines used in this study included parental ZR-75-1 cells (ATCC cat. #CRL-1500; RRID:CVCL_0588), paclitaxel-resistant ZR-75-1 (Pac-R), epirubicin-resistant ZR-75-1 (Epi-R), parental MCF-7 cells (ATCC cat. #HTB-22; RRID:CVCL_0031), epirubicin-resistant MCF-7 (Epi-R), and docetaxel-resistant MCF-7 (Doc-R). Parental lines obtained from the American Type Culture Collection (ATCC) were authenticated by the supplier via short tandem repeat (STR) profiling prior to distribution. Parental ZR-75-1 cells and their Pac-R and Epi-R derivatives, as well as parental MCF-7 cells and their Epi-R derivatives, were kindly provided by Dr. Melanie Spears (Ontario Institute for Cancer Research, Toronto, ON, Canada). MCF-7 Doc-R cells were kindly provided by Prof. Martin Michaelis (School of Biosciences, University of Kent, Canterbury, United Kingdom). ZR-75-1 resistant derivatives were generated as previously described^[19]. The docetaxel-resistant MCF-7 model was developed through sustained exposure to incrementally increasing doses of docetaxel, according to a previously described protocol^[20]. Similarly, epirubicin-resistant sublines were established by subjecting parental populations to stepwise escalations of epirubicin, initiated at a starting concentration of 0.5 nM. Chemoresistant phenotypes were confirmed when the half-maximal inhibitory concentration (IC₅₀) significantly surpassed that of the matching parental cell line, and the cultures reached a selection plateau where they could no longer tolerate higher drug pressures^[21]. While formal RRIDs are unavailable for these secondary resistant derivatives, their identities and resistance phenotypes were functionally validated through continuous selective pressure and routine dose-response viability assays.

All cell lines were cultured in Dulbecco’s Modified Eagle Medium (DMEM), low glucose, supplemented with GlutaMAX™ and sodium pyruvate (#10567014, Gibco), 10% heat-inactivated fetal bovine serum (FBS; #F7524, Merck Life Science), and 1% penicillin-streptomycin (#15140‐122, Gibco). Cells were maintained at 37 °C in a humidified atmosphere containing 5% CO₂. In the current study, chemoresistant cell lines were continuously cultured in the presence of chemotherapeutic agents as follows: 25 nM paclitaxel for ZR-75-1 Pac-R and MCF-7 Doc-R cells, 10 nM epirubicin for ZR-75-1 Epi-R cells, and 30 nM epirubicin for MCF-7 Epi-R cells. MCF-7 Doc-R cells were originally established and maintained by the providing laboratory in 20 ng/mL docetaxel; upon receipt, cultures were routinely maintained in our laboratory under continuous exposure to 25 nM paclitaxel, which shares the same taxane class and microtubule-stabilizing mechanism as docetaxel. Paclitaxel (6 mg/mL; Fresenius Kabi, lot no. 87230210AB) and epirubicin (Farmorubicin, 2 mg/mL; Pfizer, batch no. FR32501) were obtained as clinical-grade formulations from the hospital pharmacy of Cliniques universitaires Saint-Luc (Brussels, Belgium) and diluted in culture medium immediately prior to use. Cell lines were systematically screened for mycoplasma contamination utilizing a polymerase chain reaction (PCR)-based detection platform (MycoplasmaCheck, Eurofins Genomics), confirmed to be strictly mycoplasma-negative prior to use, and maintained below 20 passages post-thawing to prevent phenotypic drift.

To assess the effects of chemotherapeutic agents on cell viability and growth capacity over time, cells were seeded in triplicate in 6-well plates at a density of 2.5 × 10⁵ cells/mL for ZR-75-1 cells and 2.0 × 10⁵ cells/mL for MCF-7 cells, and allowed to adhere for 24 h. Baseline viability was measured using the PrestoBlue™ cell viability reagent (#A13262, Thermo Fisher Scientific) according to the manufacturer’s instructions. Cells were then treated with paclitaxel (25 nM) or epirubicin (10 nM for ZR-75-1 cells; 30 nM for MCF-7 cells) for 72 h, after which cell viability was reassessed. Following drug exposure, cells were washed and cultured in drug-free medium. Long-term proliferative recovery was evaluated at 4 and 7 days post-washout using PrestoBlue™. Cell growth kinetics were independently assessed by direct cell counting. Cells were seeded in triplicate in 24-well plates (50,000 cells per well), and viable cell numbers were determined on days 1, 2, 3, 4, and 7 using Trypan Blue exclusion and a hemocytometer.

To evaluate sensitivity to metabolic inhibitors, ZR-75-1 cells (1.5 × 10⁴ cells/well) and MCF-7 cells (7.5 × 10³ cells/well) were seeded in 96-well plates (≥ 3 technical replicates per condition) and allowed to adhere for 24 h. Cells were then treated with antimycin A (#33357, Cell Signaling Technology), IACS-010759 (#HY-112037, MedChemExpress), mitoguazone [methylglyoxal-bis(guanylhydrazone); MGBG] (#HY-106634, MedChemExpress), difluoromethylornithine (DFMO) hydrochloride hydrate (#HY-B0744B, MedChemExpress) or AMXT-1501 tetrahydrochloride (HY-124617A, MedChemExpress) for 72 h at different concentrations, as indicated in the figure legends. Cell viability was subsequently assessed using the PrestoBlue™ assay according to the manufacturer’s instructions. All cell viability and cell counting experiments included three technical replicate wells per condition and were performed across three independent biological experiments (n = 3). Technical replicates within each run were averaged to yield a single value per biological replicate; thus, every individual data point depicted in the corresponding figures represents an independent biological replicate.

Real-time monitoring of oxygen consumption rates (OCRs) and extracellular acidification rates (ECARs) was performed using a Seahorse XFe96 Extracellular Flux Analyzer (Agilent Technologies). Cells (1 × 10⁴ cells/well) were seeded in specialized XF96 cell culture microplates and allowed to fully adhere overnight. Prior to initiating the assay, the growth medium was replaced with specialized XF assay medium containing 2 mM L-glutamine (and 10 mM glucose for XF Cell MitoStress Test), adjusted to pH 7.4, followed by a 1-hour equilibration period at 37 °C in a non-CO₂ incubator. Bioenergetic profiles were mapped via sequential automatic injections of metabolic inhibitors: 1 μM oligomycin, 2 μM carbonyl cyanide-4 (trifluoromethoxy)phenylhydrazone (FCCP) and 0.5 μM rotenone/antimycin A to assess mitochondrial function parameters; 10 mM glucose, 1 µM oligomycin and 50 mM 2-deoxyglucose (2-DG) to evaluate glycolysis activity. Specifically, glucose-dependent ECAR was calculated by comparing the values before and after the addition of the substrate. All measurements were normalized to total protein content per well and expressed in mpH/min/μg protein (ECAR) or pmoles/min/μg protein (OCR).

Cellular consumption of glucose and secretion of lactate were monitored by seeding cells at 3 × 10⁵ cells/well in 12-well plates utilizing 500 µL of routine culture media. After 24 h, monolayers were washed and replenished with 500 µL of DMEM containing 10 mM D-glucose (#G8270, Sigma-Aldrich), 2 mM L-glutamine and 10% dialyzed FBS (#F0392, Sigma-Aldrich). Baseline metabolite profiles were verified using incubated, cell-free medium controls. Following a 24-hour treatment period, the extracellular media were harvested, transferred to 10 kDa centrifugal filter tubes (VWR), and spun at 10,000 × g (15 min, 4 °C) for deproteinization. Quantitative profiling of glucose and lactate in the prepared filtrates (50 µL sample volume) was performed on an ISCUSflex automated microdialysis analyzer (Aurora Borealis) using specific enzymatic reagents (CMA Microdialysis AB). Net metabolic fluxes were computed relative to the cell-free blank controls, normalized against total cellular protein per well, and reported in units of µmol/hr/mg protein.

Isolation of total RNA was performed using the Monarch Total RNA Miniprep Kit (New England Biolabs) in strict accordance with the manufacturer’s protocol. The quantitative yield and chemical purity of the isolated RNA specimens were determined by spectrophotometry using a NanoDrop 1000 (Thermo Fisher Scientific).

For RNA sequencing, three biological replicates (n = 3) per experimental group, derived from independent cell cultures and separate RNA extractions, were processed. Ribosomal RNA (rRNA) depletion was performed using the RiboCop rRNA Depletion Kit (Lexogen). RNA-seq libraries were subsequently prepared from rRNA-depleted RNA using the NEBNext Ultra II Directional RNA Library Prep Kit for Illumina (New England Biolabs). Libraries were paired-end sequenced (2 × 100 base pairs) on an Illumina NextSeq2000 platform, achieving an average sequencing depth of approximately 20 million reads per sample. For quantitative real-time PCR, total RNA (1 µg) was reverse transcribed with the RevertAid Reverse-Transcriptase, oligo-dT and random hexamers (Thermo Fisher Scientific) before quantitative PCR amplification on a ViiA 7 real-time PCR system (Applied Biosystems) using SYBR® select master mix for cfx (Thermo Fisher Scientific) and gene-specific primer sequences [Supplementary Table 1]. Data were analyzed according to ddCt method using human GTF2B as reference gene.

Untargeted metabolomics profiling was performed using liquid chromatography-mass spectrometry (LC-MS) (see Supplementary Methods for full details). Briefly, intracellular metabolites were extracted from cell pellets using a methanol/water-based extraction buffer containing N-ethylmaleimide and isotopically labeled internal standards. Chromatographic separation was performed using hydrophilic interaction LC coupled to high-resolution MS operated in both positive and negative electrospray ionization modes. Data acquisition included full-scan MS and data-dependent MS/MS. Raw data were processed using MS-DIAL (version 4.9.221218)^[22], including peak detection, alignment, normalization, and metabolite annotation using an in-house spectral library of authentic standards. Metabolite intensities were normalized using both cell numbers and isotopically labeled internal standards. Downstream statistical analyses and visualization were performed using R and Python.

Raw transcriptomic reads underwent initial quality control evaluation using FastQC (Babraham Bioinformatics). Adapter sequences and low-quality bases were trimmed using Trimmomatic. High-quality trimmed reads were then quantified at the transcript level using Salmon (v1.10.3)^[23] by mapping to the Ensembl human reference transcriptome (GRCh38, file: Homo_sapiens.GRCh38.cdna.all.fa). Differential gene expression analysis was performed using the edgeR Bioconductor package (v.4.8.2)^[24], adjusting for multiple testing using the Benjamini-Hochberg false discovery rate (FDR) procedure. Raw metabolomic concentrations were log2-transformed to stabilize variance and approximate a normal distribution. For heatmap visualization, metabolite intensities were separately log10-transformed. Systematic technical variation between samples was corrected by median-centering normalization using the normalizeIntensity function from the qmtools Bioconductor package (v.1.14.0). Differentially abundant metabolites between experimental conditions were identified using linear modeling with empirical Bayes moderation implemented in the limma Bioconductor package (v.3.66.0)^[25]. To assess the clinical relevance of our findings, we analyzed two publicly available RNA-sequencing datasets of HR+/HER2- BC patients sampled before and after NAC (GSE191127^[26] and GSE123845^[27]). GSE191127 comprises paired pre-treatment biopsies and post-treatment residual tumors (n = 16 pairs), whereas GSE123845 includes 43 pre-treatment and 7 post-treatment samples. Differential gene expression between pre- and post-treatment tumors was analyzed in R (v4.4.1) using the lmerSeq package (v0.1.7). Statistical significance was assessed using moderated t-tests, followed by Benjamini-Hochberg FDR correction. Gene set enrichment analysis of transcriptomic and metabolomic alterations was subsequently performed against the KEGG and Reactome pathway databases using the fgsea package (v1.36.0). All transcriptomic and metabolomic visualizations, including differential expression and pathway enrichment plots, were generated in R using the ggplot2 ecosystem.

Lipidomic profiling was performed using LC-MS, with full methodological details provided in the Supplementary Methods. Briefly, cellular lipids were extracted using a methanol/isopropanol/methyl tert-butyl ether-based protocol in the presence of internal standards. Lipid species were separated by reverse-phase ultra-high-performance LC and analyzed by high-resolution MS operating in both positive and negative ionization modes. Peak integration, feature alignment, and lipid annotation were performed using MS-DIAL (v4.9.221218) and validated against an in-house lipid spectral library. Lipid features were normalized using the B-MIS approach, followed by normalization to cell number and internal standards. Normalized lipid intensities were then log2-transformed prior to statistical analysis and visualization. For heatmap visualization, normalized lipid intensities were separately log10-transformed. Outlier replicates were identified using a multivariate distance-based approach, based on the Euclidean distance of each sample from the group-specific median. Samples displaying substantially increased distances were excluded as outliers. This classification was further supported by principal component analysis, in which outliers separated clearly from the main sample cluster. Based on this quality-control assessment, five biological replicates were excluded from the lipidomic analysis: MCF-7 Parental replicate 1, MCF-7 Doc-R replicate 2, ZR-75-1 Parental replicate 3, ZR-75-1 Pac-R replicate 4, and ZR-75-1 Epi-R replicate 4. Accordingly, three biological replicates were included in the final analysis for each of these groups, while all four replicates were included for MCF-7 Epi-R. Differences in lipid abundance across experimental conditions were assessed using empirical Bayes linear modeling implemented in the limma Bioconductor package (v3.66.0). Log₂ fold-change values were derived from fitted model coefficients, and P-values were calculated from moderated t-statistics. In line with the exploratory nature of the analysis, lipid features were prioritized based on raw P-values combined with an effect-size threshold of log₂FC > 0.5. All data visualizations were generated in R using the ggplot2 package.

Subconfluent cell monolayers were rinsed twice with ice-cold phosphate-buffered saline and harvested in radioimmunoprecipitation assay (RIPA) lysis buffer containing protease (Sigma-Aldrich) and phosphatase (Roche) inhibitor cocktails. The resulting lysates were cleared via centrifugation (10,000 × g, 10 min, 4 °C) and preserved at -80 °C. Total protein concentrations were quantified using a bicinchoninic acid assay (Thermo Fisher Scientific). Protein samples were mixed with Laemmli buffer containing 100 mM dithiothreitol, denatured at 95 °C for 5 min, resolved by SDS-PAGE on 8%-15% polyacrylamide gels, and electrotransferred onto PVDF membranes. Nonspecific binding sites were blocked using 5% non-fat dry milk (#T145.3, Roth) prepared in TBS-0.1% Tween 20 (TBST). Membranes were then probed overnight at 4 °C with primary antibodies targeting human β-actin (clone AC-15, #A5441, Sigma-Aldrich; 1:10,000) or ATP binding cassette subfamily B member 1 (ABCB1; clone E1Y7S, #13978S, Cell Signaling Technology; 1:1,000). Following washing in TBST, membranes were exposed to horseradish peroxidase-conjugated secondary antibodies (#115-035-003 and #111-035-003, Jackson ImmunoResearch; 1:5,000) for 1 h at room temperature. Protein bands were developed using the ECL Western Blotting Detection Kit (GE Healthcare) and captured on an Amersham Imager 600 system (GE Healthcare). Western blot analyses were performed using three independent biological replicates for each parental and chemoresistant cell line (n = 3), designated as samples #1, #2, and #3.

Graphical representations and statistical analyses were performed using GraphPad Prism (version 10.6.0). Data were analyzed using an unpaired Student’s t-test, one-way analysis of variance (ANOVA) followed by Dunnett’s multiple-comparison post-hoc test, or two-way ANOVA followed by Tukey’s or Sidák’s multiple-comparison post-hoc tests, as appropriate. Statistical significance was defined as ^*P < 0.05, ^**P < 0.01, ^***P < 0.001. No statistical methods were used to predetermine sample sizes, which were instead selected based on established standards in the field. No samples or individual data points were excluded from the analyses, except for five biological replicates excluded from the lipidomic analysis following the quality-control procedure described in the Lipidomics Analysis section.

To investigate metabolic alterations associated with acquired resistance to anthracyclines and taxanes, we employed a panel of parental and chemoresistant HR+/HER2- BC cell models derived from ZR-75-1 and MCF-7 cell lines [Figure 1A]. Paclitaxel-resistant ZR-75-1 cells (Pac-R) have previously been shown to exhibit a marked reduction in paclitaxel sensitivity, with an approximately 170-fold increase in IC₅₀ compared to parental cells, as well as cross-resistance to other taxanes^[19]. However, the additional chemoresistant derivatives used in the present study, including epirubicin-resistant ZR-75-1 and MCF-7 (Epi-R), and docetaxel-resistant MCF-7 cells (Doc-R), had not been systematically evaluated under the treatment and recovery conditions required for downstream multi-omics analyses. We therefore implemented a standardized drug challenge and recovery assay to validate the functional stability of all resistant cell populations [Figure 1B]. Cells were exposed to paclitaxel or epirubicin at concentrations corresponding to their respective resistance selection, and responses were assessed both at the end of treatment and following a defined drug-free recovery period.

Across both ZR-75-1 and MCF-7 models, chemoresistant derivatives consistently displayed reduced drug sensitivity and enhanced capacity to resume growth following drug withdrawal compared with parental cells [Figure 1C-J and Supplementary Figure 1A-F]. These results indicate that the selected models represent stable chemoresistant phenotypes rather than transient drug-tolerant states, supporting their use for subsequent systems-level analyses.

To characterize molecular alterations associated with acquired resistance to anthracyclines and taxanes, we performed RNA sequencing in parental and chemoresistant HR+/HER2- BC cells. Principal component analysis revealed a clear separation between parental and resistant populations in both MCF-7 and ZR-75-1 models, indicating substantial transcriptional reprogramming upon acquisition of drug resistance [Supplementary Figure 2A and B]. Differential gene expression analysis confirmed widespread transcriptomic remodeling across all resistant models [Figure 2A and B], although the magnitude of these changes varied, with MCF-7 Doc-R cells displaying the largest number of dysregulated genes. Despite this extensive reprogramming, overlap in differentially expressed genes across all resistant models remained limited [Figure 2C and D], indicating that resistance is predominantly driven by context-dependent transcriptional adaptations rather than a shared gene expression program. Consistently, only minimal overlap was observed between MCF-7 and ZR-75-1 cells resistant to the same class of chemotherapeutic agents [Supplementary Figure 2C], whereas greater similarity was observed within models derived from the same parental cell line [Supplementary Figure 2D].

Functional annotation revealed partially convergent yet drug-biased transcriptional programs. Taxane-resistant models were characterized by the upregulation of genes involved in drug efflux (e.g., ABCB1, ABCC2) and redox homeostasis (e.g., ALDH1A3, BLVRB, and CYP1B1) [Figure 3A and B]. The preferential upregulation of ABCB1 in taxane-resistant cells was further validated by qRT-PCR in both models and corroborated at the protein level by Western blot analysis [Supplementary Figure 3A-D]. In contrast, anthracycline-resistant cells preferentially exhibited alterations in lipid metabolism-related genes (e.g., APOD, DGKA, and PLIN2) alongside increased expression of ATP13A4 and spermidine/spermine N1-acetyltransferase 1 (SAT1), two genes implicated in polyamine transport and catabolism^[28]. qRT-PCR analysis confirmed increased SAT1 expression in anthracycline-resistant MCF-7 and ZR-75-1 cells, whereas ATP13A4 expression was significantly increased only in MCF-7 Epi-R cells, with a non-significant upward trend in ZR-75-1 Epi-R cells [Supplementary Figure 3E-H]. Pathway enrichment analysis further highlighted the marked heterogeneity of resistance-associated transcriptional changes. Mitochondrial translation emerged as the only pathway consistently downregulated across all chemoresistant models, irrespective of the drug used [Figure 3C-F]. Beyond this shared feature, most pathways were regulated in a cell line- and drug-dependent manner. Notably, pathways related to mitochondrial function, glucose metabolism, lipid metabolism, and polyamine homeostasis were recurrently affected suggesting potential metabolic adaptations associated with chemoresistance. However, whether these transcriptional changes translate into functional metabolic rewiring remained unclear.

To determine whether transcriptional alterations were associated with functional metabolic changes, we assessed cellular bioenergetics using Seahorse extracellular flux analysis. OCR measurements revealed an overall reduction in mitochondrial respiration in chemoresistant cells compared with their parental counterparts [Figure 4A and B]. While basal respiration showed modest changes in MCF-7 models, maximal respiratory capacity was significantly decreased, particularly in Epi-R cells [Figure 4C and D]. In ZR-75-1 cells, both basal and maximal respiration were markedly reduced in Pac-R and Epi-R conditions, indicating a pronounced impairment of mitochondrial function [Figure 4E and F]. These functional alterations are consistent with transcriptomic data identifying mitochondrial translation as the only pathway uniformly downregulated across all chemoresistant models [Figure 3C-F].

ECAR analysis further showed that glycolytic activity was not increased to compensate for reduced mitochondrial function. Instead, glycolysis and maximal glycolytic capacity were decreased, particularly in ZR-75-1 Epi-R cells, with more variable effects observed in MCF-7 models [Figure 4G-L]. Consistently, glucose consumption and lactate secretion were not elevated in chemoresistant cells [Supplementary Figure 4A-D].

Together, these data indicate that chemoresistant cells adopt a low-bioenergetic phenotype characterized by coordinated reductions in both mitochondrial respiration and glycolytic activity, suggesting a global suppression or reprogramming of energy metabolism as illustrated in cell energy maps [Figure 4M and N].

To directly assess metabolic alterations, we performed untargeted metabolomic analyses. Principal component analysis revealed substantial overlap between parental and chemoresistant cells in both MCF-7 and ZR-75-1 models [Supplementary Figure 5A and B], indicating that global metabolic profiles remain relatively conserved despite pronounced transcriptional divergence. Consistently, integrated pathway analysis combining transcriptomic and metabolomic data revealed limited concordance across omics layers [Figure 5A]. While purine and pyrimidine metabolism were decreased at both levels, most pathways did not show consistent regulation. Core bioenergetic pathways, including oxidative phosphorylation (OXPHOS) and fatty acid metabolism, displayed an overall decreasing trend, in agreement with functional analyses. At the metabolite level, changes were heterogeneous and not always aligned with pathway-level predictions [Figure 5B and C]. Notably, polyamine metabolism emerged as one of the most recurrently altered features across chemoresistant models. Increased levels of spermidine and spermine were consistently observed, although other intermediates (e.g., N8-acetylspermidine, N-acetylputrescine) displayed context-dependent regulation, indicating partial rather than uniform pathway activation.

To assess lipid alterations, we performed untargeted lipidomic analyses across parental and chemoresistant models. Principal component analysis revealed partial separation between parental and resistant cells in both MCF-7 and ZR-75-1 models [Supplementary Figure 6A and B], indicating modest but detectable lipidomic divergence. Volcano plot analysis demonstrated a strong bias toward lipid downregulation in resistant cells, although the extent varied depending on both drug and cell line. In MCF-7 cells, Epi-R cells exhibited a pronounced reduction in lipid species (358 downregulated vs. 25 upregulated), whereas Doc-R cells showed a more balanced distribution [Figure 6A]. In contrast, in ZR-75-1 cells, Pac-R cells displayed the most extensive lipid depletion (254 downregulated vs. 6 upregulated), while Epi-R cells showed a comparatively milder, yet still reduction profile (78 downregulated vs. 28 upregulated) [Figure 6B]. These findings indicate that the magnitude of lipid depletion is both drug- and model-dependent. Overlap between MCF-7 and ZR-75-1 cells resistant to the same class of chemotherapeutic agents remained limited [Supplementary Figure 6C], whereas models originating from the same parental cell line showed greater similarity [Supplementary Figure 6D], consistent with observations at the transcriptomic level. At the lipid class level, lipid abundance was broadly decreased, particularly in MCF-7 Epi-R and ZR-75-1 Pac-R cells. Neutral lipids were consistently reduced across resistant models, suggesting a decrease in lipid storage capacity. Phospholipids and sphingolipids were also predominantly decreased in resistant cells, although selective increases were observed in specific contexts, particularly in MCF-7 Doc-R and ZR-75-1 Epi-R models [Figure 6C and D]. Consistent with these findings, heatmap analysis of the top 50 most regulated lipid species revealed overall constrained lipid profiles in chemoresistant cells [Supplementary Figure 6E and F].

To functionally assess metabolic dependencies, parental and chemoresistant MCF-7 and ZR-75-1 cells were treated with pharmacological inhibitors targeting mitochondrial respiration and polyamine metabolism. Inhibition of electron transport chain (ETC) complex I using IACS-010759 reduced cell viability across all models [Figure 7A and B]. However, contrary to our initial hypothesis, chemoresistant cells were overall less sensitive to treatment than their matching parental counterparts. This reduced sensitivity was particularly pronounced in MCF-7 Doc-R and ZR-75-1 Epi-R cells, in line with their decreased mitochondrial respiratory activity. Similarly, treatment with antimycin A, an ETC complex III inhibitor, decreased viability in both parental and resistant cells, with chemoresistant lines again exhibiting lower overall sensitivity [Supplementary Figure 7A and B]. These findings indicate that the diminished mitochondrial function observed in resistant cells translates into a reduced dependence on OXPHOS for survival.

We next evaluated polyamine metabolism using mitoguazone (MGBG), an inhibitor of S-adenosyl-methionine decarboxylase. In MCF-7 models, both Doc-R and Epi-R cells exhibited increased sensitivity to 10 µM MGBG compared with parental cells [Figure 7C]. In contrast, ZR-75-1 chemoresistant derivatives (Pac-R and Epi-R) were less affected by MGBG treatment than their parental counterparts [Figure 7D], indicating reduced sensitivity to polyamine biosynthesis inhibition. Similar results were obtained using DFMO, an ornithine decarboxylase inhibitor. In MCF-7 models, both Doc-R and Epi-R cells exhibited increased sensitivity compared with parental cells [Supplementary Figure 7C], whereas no significant differences were observed between parental and chemoresistant ZR-75-1 cells [Supplementary Figure 7D]. Finally, inhibition of polyamine transport using AMXT-1501 did not reveal consistent differential sensitivity between parental and chemoresistant cells in either MCF-7 or ZR-75-1 models [Figure 7E and F]. While MCF-7 Epi-R and ZR-75-1 Pac-R cells showed a modest trend toward reduced sensitivity at selected concentrations, these differences were minor and inconsistent across resistant sublines. Collectively, these findings reveal marked cell line-dependent divergence in responses to polyamine pathway inhibition. While chemoresistant MCF-7 cells displayed increased sensitivity to polyamine biosynthesis inhibitors (MGBG and DFMO), ZR-75-1 resistant cells did not exhibit enhanced sensitivity to either biosynthesis or transport inhibition (AMXT-1501).

To assess the translational relevance of these metabolic alterations, we analyzed pre- and post-NAC RNA-sequencing datasets from patients with HR+/HER2- BC (GSE191127 and GSE123845). ABCB1 gene expression was increased following treatment in both clinical cohorts, whereas ATP13A4 showed a more modest elevation and SAT1 expression remained largely unchanged [Figure 7G]. Consistent with these transcriptomic shifts, gene set enrichment analysis revealed significant positive enrichment of the ABC transporter pathway in post-treatment residual tumors from GSE191127, with a similar positive trend observed in GSE123845 [Figure 7H and Supplementary Figure 7E]. In contrast, OXPHOS signature was significantly negatively enriched following chemotherapy in both clinical datasets [Figure 7I and Supplementary Figure 7F]. Glycolysis/gluconeogenesis pathway expression was likewise significantly negatively enriched in GSE191127, with GSE123845 showing a comparable but non-significant negative trend [Figure 7J and Supplementary Figure 7G]. Finally, neither the arginine-to-putrescine biosynthesis module nor the putrescine catabolism-related GABA biosynthesis module reached statistical significance in either patient cohort, although both exhibited a positive enrichment trend following treatment [Figure 7K and L, Supplementary Figure 7H and I].

Collectively, these findings demonstrate that chemoresistant HR+/HER2- BC cells adopt a metabolically constrained state characterized by reduced reliance on mitochondrial respiration and context-dependent vulnerabilities to polyamine pathway inhibition. Crucially, interrogation of clinical HR+/HER2- BC datasets corroborated the core metabolic features identified in our experimental models, including increased ABC transporter expression alongside the suppression of OXPHOS and glycolytic programs following NAC. In contrast, polyamine pathway alterations were less consistently detected in patient samples, supporting the concept that this metabolic adaptation is heterogeneous and defines context-specific rather than universal therapeutic vulnerabilities.