Metabolic adaptation towards glycolysis supports resistance to neoadjuvant chemotherapy in early triple negative breast cancers

Human TNBC cell lines HCC38 (#CRL2314), HCC1143 (#CRL2321) and HCC1937 (#CRL2336) were purchased from ATCC. Cell lines were stored according to the supplier’s instructions and used within 6 months after resuscitation of frozen aliquots. All cells were routinely cultured in RPMI 1640 medium (#61870-010, Thermo Fisher Scientific) containing 11.1 mM d-glucose and 2 mM GlutaMAX™, and supplemented with 10% heat-inactivated fetal bovine serum (FBS; #F7524, Sigma-Aldrich) and 1% penicillin/streptomycin (#15140-122, Thermo Fisher Scientific), and maintained in exponential growth in 5% CO₂/95% air in a humidified incubator at 37 °C. All cell lines were tested for potential mycoplasma contamination with the PCR-based MycoplasmaCheck service from Eurofins Genomics. For drug testing, cells (10,000 cells/well) were seeded in 96-well plates and treatment was performed, 24 h later, in a full culture medium with 2-deoxy-glucose (2-DG; #D8375, Sigma-Aldrich), paclitaxel or epirubicin at different concentrations, as indicated in the figure legends. Chemotherapeutic agents were obtained, as ready-to-use solutions, from the central pharmacy of Cliniques universitaires Saint Luc (CUSL, Brussels, Belgium). After 72 h, cell growth was assessed by using the Presto Blue reagent (#A13262, Thermo Fisher Scientific) according to manufacturer’s instructions. To assess the capacity of TNBC cells to resume growth upon NAC treatment, Presto Blue reagent was also used at different timings: baseline (no NAC treatment), after a 24 h-treatment with paclitaxel or epirubicin, and two days after drug withdrawal.

For 3D spheroid initiation, HCC38 and HCC1937 cells (5000 cells/well) were seeded in ultra-low attachment round-bottom 96-well plates (Corning) before centrifugation [500 rpm, 5 min at room temperature]. Drug testing was carried out 96 h post-initiation by removing 100 µL of initial medium and adding 100 µL of new medium containing 2-DG alone or in combination with chemotherapeutic drugs at different concentrations, as indicated in the figure legends. Cell viability was assessed 72 h later using CellTiter-Glo 3D cell viability assay (#G9681 Promega) according to manufacturer’s instructions.

BCO17 and IDC031 organoid models were established from freshly resected tumor tissues obtained from treatment-naive TNBC patients, at CUSL or Antoni van Leeuwenhoek Hospital, respectively. The study was approved by the hospital-faculty ethical committee (UCL-ONCO2015-02—2015/13AOU/445) and by the institutional review board (NKI-B17PRE) and all subjects provided informed consent. BC tissues were obtained from surgical pieces (mastectomy) and kept on ice in a transport medium (DMEM/F12 with HEPES, no phenol red, 1% penicillin–streptomycin). After washing (3 times with ice-cold PBS), tissues were minced with a scalpel and enzymatically digested using the tumor dissociation kit (#130-095-929, Miltenyi Biotec) in the gentleMACS™ octo dissociator with heaters (Miltenyi Biotec). After dissociation, samples were filtered through a 70-µm cell strainer and rinsed with DMEM/F12 containing 10% FBS, before centrifugation (1500 rpm, 5 min at 4 °C). Cell pellets were then resuspended in Cultrex reduced growth factor Basement Membrane Extract (BME) type 2 (#3532-005-02, R&D Systems) and plated as 50 µL droplets in pre-heated 24-well plates. Plates were incubated at 37 °C (5% CO₂) for 30 min to allow Cultrex BME solidification before adding BC organoid medium (500 µL/well) (see Additional file 1: Table S1 for medium formulation).

Organoid passaging was performed every 1–2 weeks, depending on the proliferation rate, and medium was changed twice a week. Briefly, each dome of Cultrex BME was dissolved with ice-cold PBS, collected in a tube, and then centrifuged (1500 rpm, 5 min at 4 °C). Trypsin–EDTA 0.05% was then added to the pellet, before incubation at 37 °C for 15–30 min, with frequent pipetting, to facilitate the dissociation into single cells. After addition of DMEM/F12 supplemented with 10% FBS, and centrifugation (1500 rpm, 5 min at 4 °C), single cells were resuspended in Cultrex BME, overlaid with fresh BC organoid medium. BCO17 and IDC031 organoid lines were cryopreserved with 45% medium/45% FBS/10% DMSO.

After organoid dissociation, single cells were counted with a LUNA-II™ automated cell counter (Logos Biosystems), and then seeded (10,000 cells/well) in 96-well plates pre-coated with Cultrex BME (20 µL/well), in 100 µL of BC organoid medium supplemented with 2% Cultrex BME. After 7 days, BC organoids were treated with paclitaxel or epirubicin at different doses for 7 additional days. Next, organoid viability was assessed by using the CellTiter-Glo 3D cell viability assay (#G9681, Promega) according to manufacturer’s instructions, and luminescence was read with a GloMax microplate reader (Promega).

For colony-forming ability assessment, HCC38 and HCC1143 cells were seeded in 6-well plates (1000 cells/well) and treated with 2-DG alone or in combination with paclitaxel, for 72 h, at concentrations indicated in the figure legends. Seven days after drug withdrawal, colonies were labelled with a mix of 0.5% brilliant blue (#B0149, Sigma-Aldrich), 50% ethanol and 5% acetic acid for 1 h under gentle agitation. After rinsing with ultrapure distilled water, culture wells were scanned with an Epson V600 scanner and images were processed with ImageJ to count the number of cell colonies larger than a defined threshold (30 pixels).

Subconfluent cancer cells were washed twice with ice-cold PBS and lysed in a RIPA buffer supplemented with a protease inhibitor cocktail (#P8340, Sigma-Aldrich) and a phosphatase inhibitor cocktail (#4906837001, Roche). Cell lysates were then cleared by centrifugation (10,000 rpm, 10 min, 4 °C) and stored at − 80 °C until analysis. After determination of protein concentration using a bicinchoninic acid-based assay (#23225, Thermo Fisher Scientific), samples were denaturated (5 min, 95 °C) with Laemmli sample buffer containing 100 mM dithiothreitol. Protein samples (20 µg/well) were then separated by SDS-PAGE (8 to 15% acrylamide/bis-acrylamide gels) and transferred to PVDF membranes. Membranes were blocked with 5% bovine serum albumin in TBS-0.1% Tween 20 (TTBS) and subsequently immunoblotted overnight at 4 °C with specific primary antibodies against β-actin (1:10,000; #A5441, Sigma-Aldrich), GAPDH (1:1000; #2118, Cell Signaling Technology), HK2 (1:1000; #2867, Cell Signaling Technology), LDHA (1:1000; #3582, Cell Signaling Technology) and HSP90 (1:7,500; #610419, BD Biosciences). After several washes with TTBS, membranes were then incubated (1 h, room temperature) with horseradish peroxidase-conjugated secondary antibodies (Jackson Immunoresearch) and chemoluminescent signals were revealed by using ECL Western Blotting Detection Kit (GE Healthcare) either on X-ray films in a dark chamber, or with an Amersham Imager 600 (GE Healthcare). Films were scanned with an Epson V600 scanner and images were processed with ImageJ.

Cancer cells (2 × 10⁵ cells/well; 3 wells/condition) were seeded in 12-well plates with 2 mL of their routine culture medium, while organoid-derived single cells (5 × 10⁵ cells/wells; 6 wells/condition) were seeded in 24-well plates with 1 mL of the BCO culture medium. After 1 or 7 days of incubation for cancer cells and TNBC organoids respectively, medium was replaced by 500 µL of DMEM containing 10 mM d-glucose, 2 mM L-glutamine and supplemented with 10% dialyzed FBS (#F0392, Sigma-Aldrich). Initial concentrations of glucose and lactate in the experimental medium were also assessed by including control wells containing only culture medium (no cells) on each plate. After cell incubation for 24 to 48 h, extracellular media were collected and deproteinized by centrifugation (15 min, 10,000 rpm, 4 °C) in 10 kDa cut-off filter tubes (VWR). Glucose and lactate concentrations were measured in the samples (50 µL) by using specific enzymatic assays and an ISCUSflex microdialysis analyzer (M Dialysis). Data analysis was done by calculating the difference in glucose and lactate concentrations between the control and experimental wells. Data were then normalized by the protein content in each well and expressed in µmol/hr/mg proteins.

Oxygen consumption rate (OCR) and extracellular acidification rate (ECAR) were measured by using the Seahorse XFe96 plate reader (Agilent). For 2D cell lines, assays were carried out using a seeding density of 20,000 cells/well (pre-treated or not with NAC for 24 h or 7 days) in non-buffered DMEM, adjusted at pH 7.4 and supplemented with specified metabolic substrates. Mitochondrial respiration was assessed in DMEM medium containing 10 mM d-glucose and 2 mM l-glutamine, at baseline conditions and after sequential treatment with 1 µM oligomycin, 2 µM FCCP and 0.5 µM rotenone/antimycin A. Glycolytic function was evaluated in DMEM medium supplemented with 2 mM L-glutamine, at baseline conditions and after sequential treatment with 10 mM d-glucose, 1 µM oligomycin and 50 mM 2-DG. For organoids, 10,000 cells/well were seeded in XF96 culture plates, pre-coated with Cultrex BME (20 µL/well), and incubated for 7 days in the BC organoid culture medium, before starting the experiments. Cells were then incubated in non-buffered DMEM, adjusted at pH 7.4 and supplemented with specified metabolic substrates. Mitochondrial and glycolytic activities were assessed, as described above for 2D cell cultures, except for glucose concentration set at 20 mM for the glycolysis test. For all assays, three cycles (3 min mixing/3 min measuring) were carried out between each treatment. Glucose-dependent ECAR was determined by calculating the difference between the values before (measurements 1–3) and after (measurements 4–6) addition of the substrate. Data were normalized by the protein content in each well and expressed in mpH/min/µg proteins (ECAR) or pmoles/min/µg proteins (OCR).

Organoids were fixed in 4% paraformaldehyde (20 min at room temperature) before embedding (first in a 2% agar solution, and then in paraffin). Organoid sections (5-µm width), as well as sections from matching primary BC tissues, were prepared and stained for ER (#13258, Cell Signaling Technology), PR (#8757, Cell Signaling Technology), HER2 (#2165, Cell Signaling Technology) and Ki67 (#MA5-14520, Thermo Fisher Scientific) with a VENTANA automated staining system (Roche) at CUSL. Image acquisition was performed using an AxioImager Z1 microscope equipped with an Apotome1 (Zeiss).

Patients with early TNBC (diagnosed between 2012 and 2022), eligible for NAC, were recruited, retrospectively and prospectively, at CUSL (NCT03314870, ethics committee number 2017/25JUL/376). For the retrospective cohort, recruited from 2012 to 2017, no informed consent was needed while for the prospective cohort, recruited from 2018 to 2022, subjects gave informed consent. Formalin-fixed, paraffin-embedded (FFPE) tissue slides from diagnostic biopsies (pre-NAC) were obtained from the Department of Pathology from CUSL. Pathological and clinical information from this cohort was collected and managed using the REDCap (Research Electronic DataCapture) secured database hosted at CUSL.

Patients received NAC with sequential therapy of anthracyclines and taxanes, with 4 cycles of epirubicin (90 mg/m²)/cyclophosphamide (600 mg/m²) intravenously (iv)) followed by 12 cycles of weekly paclitaxel (80 mg/m² iv), and 2 patients out of the 29 TNBC had the same regimen with the addition of 3-weekly carboplatin (AUC 5, iv). Response to NAC was evaluated by histological examination of the surgical resection specimen, sliced at 5 mm intervals. Pathological complete response was defined as the absence of invasive tumor in the breast and in the axilla, with the calculation of the RCB score, as previously described [23]. The responding tumors obtained an RCB of 0 (no residual disease) or I (minimal residual disease) while the non-responding patients had an RCB of II (moderate residual disease) or III (extensive residual disease).

FFPE slides from diagnostic biopsies were cut with a depth of 5 µm. Delimitation of the tumoral zone was done by the pathologist after hematoxylin/eosin staining on one slide, and then reported on the other slides. On average, the tumoral zones of 15–20 slides were scratched. After a brief centrifugation, deparaffinization and extraction steps were carried out using the AllPrep DNA/RNA FFPE kit (#80234, Qiagen), according to manufacturer’s recommendations and DNA was removed from the samples with the TURBO DNA-free™ kit (#AM1907, Thermo Fisher Scientific), by following manufacturer’s instructions. Total RNA samples were then quantified by using Quant-it™ RiboGreen RNA Assay kit (#R11490, Thermo Fisher Scientific), before ribosomal RNA depletion with the NEBNext rRNA Depletion kit (Human/Mouse/Rat) with sample purification beads (#E6350, New England Biolabs). Libraries were prepared with the NEBNext Ultra II Directional RNA Library Prep with Beads kit (#E7765, New England Biolabs). Adaptors and primers used for the library preparation were from NEBNext Multiplex Oligos for Illumina (Dual Index Set 1), 96 rxns (#E7760, New England Biolabs). The quality of the libraries was evaluated using High Sensitivity DNA kit (chips and reagents) (#5067 and #4626, Agilent) with a 2100 Bioanalyzer system (Agilent). Sequence reads were generated on the Illumina NovaSeq 6000 sequencer, on an S4 cartridge (300 bp). RNA sequencing was performed in 3 distinct batches.

Raw sequences were filtered to remove low-quality reads. The quality of analyzed data was checked using FastQC and QualiMap, while trimming was carried out by Trimmomatic. Filtered data were then mapped by CLC Genomics v22 software (Qiagen) to the Homo sapiens genome hg38 and the RNA database v91. On average, 2 × 10⁶ reads could be mapped to the human genome. Samples with a mapping lower than 500,000 reads were excluded. The read count expression matrix was analyzed using the edger v3.40.2 Bioconductor package in order to detect differentially expressed genes (DEGs). Briefly, a filtering strategy was first applied in order to keep genes having sufficiently large counts to be retained in statistical analyses. Scaling factors were then computed with the trimmed mean of M-values method. A quasi-likelihood negative binomial generalized log-linear model was built on the resulting data. In addition to the variable of interest (e.g. RCB), a batch effect, a DNA contamination effect (estimated by the percentage of intergenic mapping), as well as the age of the patients were explicitly incorporated in the models in order to correct for the potential confounding effects of these factors. This modeling strategy was applied to detect DEGs between non-responding and responding tumors (i.e. RCB II-III vs. 0-I). Finally, based on the resulting gene expression modulations, gene set enrichment analysis (GSEA) was performed using the fgsea v1.24.0 bioconductor packages on hallmark gene sets of the MSigDB collection.

Two publicly available transcriptomic datasets, namely GSE25066 and GSE123845, were analyzed to confirm our findings in TNBC patients. Gene expression profiles from the GSE25066 dataset were obtained using an Affymetrix Human Genome U133A Array while profiles from the GSE123845 dataset were obtained using expression profiling by high-throughput RNA sequencing. For both transcriptomic datasets, metadata were imported using the GEOquery v.2.66.0 Bioconductor package. Regarding the GSE25066 dataset, we selected 146 TNBC patients for which pathological response information (i.e. RCB class) was available from the full dataset of 508 patients. The limma Bioconductor package was employed to estimate fold-changes and p-values associated with pathological response (RCB), while adjusting for patient age. In the GSE123845 dataset, we selected 33 samples, corresponding to tumor biopsies from TNBC patients collected before treatment with NAC, from the full dataset of 227 samples for which information about pCR was available. The edgeR v.3.42.4 Bioconductor package was employed to estimate fold-changes and p-values associated with pCR, while adjusting for patient age. Subsequently, GSEA was conducted for both datasets using the fgsea v1.24.0 Bioconductor package on hallmark gene sets from the MSigDB collection.

Data are expressed as mean ± SEM of at least three independent experiments unless otherwise indicated. Statistical analyses were performed through GraphPad Prism 9 by using Student’s t test, one-way or two-way ANOVA with Tukey’s multiple comparison test when appropriate. Statistical significance is indicated in the figures as follows: *p < 0.05; **p < 0.01; ***p < 0.001; ns, not significant.

To study the metabolic adaptation of TNBC cells upon NAC treatment, we first used three human cell lines representative of chemotherapy-naive early TNBC: HCC1143, HCC1937 (both basal A), and HCC38 (basal B), all harbouring distinct TP53 gene alterations and BRCA1 mutation status (Additional file 1: Fig. S1A), commonly observed in the patient population [24]. We evaluated the growth-inhibitory effects of the standard frontline NAC, epirubicin and paclitaxel, on the TNBC cell lines and showed that HCC38 cells were the most sensitive to both chemotherapeutic drugs (Fig. 1A–C). While HCC1143 and HCC1937 cell lines exhibited a similar sensitivity to epirubicin (Fig. 1A), the latter was found to be more resistant to paclitaxel, with at least a 40-fold higher IC₅₀ value in comparison to HCC1143 and HCC38 cells (Fig. 1B, C). Importantly, while treatment for 24 h with IC₅₀ of epirubicin or paclitaxel decreased viability of the three TNBC lines, cells could resume growth (or at least stabilize it) upon drug withdrawal (Fig. 1D, E). Next, we assessed the metabolic activity in these three human TNBC cell lines, at baseline conditions (i.e. without NAC treatment). By using Seahorse-based bioenergetic analysis, we observed that mitochondrial respiration, evaluated by measurements of oxygen consumption rate (OCR), was different at both basal and maximal levels, between the TNBC cell lines (Additional file 1: Fig. S1B-D), with HCC1143 cells having a greater respiration capacity than HCC38 and HCC1937 cells (Additional file 1: Fig. S1E). Nevertheless, ATP production-linked respiration was similar in all TNBC cell lines (Additional file 1: Fig. S1F). Live-cell metabolic profiling, with the Seahorse analyzer, also revealed a high glycolytic profile for HCC1937 cells, as shown with greater glucose-dependent and maximal extracellular acidification rates (ECAR), reflecting lactate/H⁺ efflux, in these cells (Fig. 1F–H). Indeed, we observed that HCC1937 cells displayed higher glucose consumption and lactate secretion rates, when compared to HCC1143 and HCC38 cells (Fig. 1I, J). HCC1143 cells were also found to be more glycolytic than HCC38 cells (Fig. 1I, J), with a glucose-to-lactate ratio similar to HCC1937 cells (Additional file 1: Fig. S1G). This was associated with higher protein expression levels for two glycolytic enzymes, namely hexokinase 2 (HK2) and glyceraldehyde 3-phosphate dehydrogenase (GAPDH), in these cells (Fig. 1K–L and Additional file 1: Fig. S1H). On the contrary, lactate dehydrogenase A (LDHA), catalysing the last step of glycolysis (i.e. pyruvate-to-lactate conversion), was more present in HCC38 cells (Fig. 1K and Additional file 1: Fig. S1I). Finally, treatment with 2-deoxyglucose (2-DG), a glucose analog blocking HK2 activity, induced similar growth-inhibitory effects in the three TNBC cell lines, with HCC1937 being slightly more resistant than HCC1143 and HCC38 cells (Additional file 1: Fig. S1J-K), in line with the higher expression of HK2 in these cells (Fig. 1K, L).

Paclitaxel and epirubicin induce opposite effects on glycolytic activity and mitochondrial oxidative metabolism in TNBC cells.

To assess the effect of NAC on the glycolytic activity in TNBC cells, we first treated the cell lines with paclitaxel and epirubicin, for 24 h at doses corresponding to the IC₅₀ values calculated for each cell line (see Fig. 1C). Paclitaxel, but not epirubicin, induced an increase in glucose consumption and lactate secretion in HCC1143 cells (Fig. 2A, B), and this was associated with higher glucose-dependent ECAR (Fig. 2C, D), thereby confirming an increased glycolytic activity. While similar effects for paclitaxel were found in HCC38 cells (Fig. 2E–H), we also registered greater glucose consumption and lactate secretion in this line upon epirubicin treatment (Fig. 2E, F), but it was not correlated with major ECAR changes (Fig. 2G, H). In HCC1937 cells that exhibited the highest glycolytic activity at the basal level, we could not observe any difference for glucose consumption and lactate secretion (Additional file 1: Fig. S2A-B), and ECAR values were even lower in these cells upon treatment with paclitaxel or epirubicin for 24 h (Additional file 1: Fig. S2C-D). Importantly, when treating TNBC cell lines with chemotherapy for 7 days in order to better mimic the selection or induction of a drug-resistant phenotype upon long-term treatment exposure, paclitaxel induced a consistent increase of glucose-dependent ECAR in the three cell lines while epirubicin had no effect (Additional file 1: Fig. S2E-J). Moreover, we found that paclitaxel treatment led to an upregulation of HK2, a protein crucial for the initiation of glycolysis, in all cell lines (Fig. 2I, J and Additional file 1: Fig. S2K-L), whereas GAPDH and LDHA expression levels were not or barely affected by the treatment. Oppositely, both chemotherapeutic drugs decreased mitochondrial oxygen consumption rates, at basal and maximal levels, in the three TNBC cell lines after 24 h (Fig. 3A–I) or 7 days of treatment (Additional file 1: Fig. S3A-I). These data suggest that treatment with paclitaxel and epirubicin, in TNBC cells, triggers common inhibitory effects on oxidative cell metabolism, whilst differently modulating the glycolytic pathway. We identify HK2 upregulation as a common metabolic trait in paclitaxel-treated TNBC cells, which may be further therapeutically exploited.

In order to translate our initial findings in more clinically relevant models, we established patient-derived organoids from treatment-naive TNBC specimens (Fig. 4A–C). Two models, namely BCO17 and IDC031, were validated by immunohistochemical staining showing the absence of signal for ER, PR and HER2, as well as the presence of the proliferation marker Ki67, as observed in the matching primary tumor tissues (Fig. 4D and Additional file 1: Fig. S4A), thereby confirming that our TNBC organoid models recapitulate the histological pattern of the tissue of origin. We evaluated the baseline glycolytic activity in these two models. IDC031 exhibited significantly higher glucose consumption and lactate secretion rates compared to BCO17 organoids (Fig. 4E, F). Similarly, glucose-dependent and maximal ECAR were found to be higher in IDC031 TNBC organoid model (vs. BCO17) (Fig. 4G, H and Additional file 1: Fig. S4B). Western blot analyses also revealed a greater expression of GAPDH and LDHA enzymes in IDC031, in comparison to BCO17 organoids (Additional file 1: Fig. S4C). Finally, we observed a higher mitochondrial respiration activity, both at basal and maximal levels, in IDC031 (Additional file 1: Fig. S4D-F), thereby suggesting that this organoid model has a higher overall metabolic activity than the BCO17 line.

Since the two TNBC organoid models were established from chemotherapy-naive tumors, we assessed their response to paclitaxel and epirubicin. BCO17 and IDC031 organoids exhibited similar sensitivity to paclitaxel and epirubicin, with dose-dependent growth-inhibitory effects after 7 days of treatment (Fig. 5A–C). As carried out with the TNBC cell lines, we evaluated the effect of chemotherapy on the glycolytic activity of TNBC organoids. While paclitaxel was found to induce a significant increase of glucose consumption only in the IDC031 model (Fig. 5D), lactate secretion was higher in both TNBC organoid models upon treatment with this drug (Fig. 5E). Epirubicin did not induce any major change in the glycolytic activity in the two models, except for an exacerbated lactate release in IDC031 (Fig. 5D–E). Furthermore, we showed that glucose-dependent ECAR was increased in paclitaxel-treated BCO17 organoids, and maximal ECAR was higher in both TNBC organoid models upon paclitaxel treatment (Fig. 5F, G). On the contrary, epirubicin did not modify these parameters in any model, confirming that the two chemotherapeutic agents have distinct effects on the glycolytic activity of TNBC cells. No consistent change of protein expression was observed in BCO17 and IDC031 organoids upon treatment with either paclitaxel or epirubicin treatment (Additional file 1: Fig. S5A). Finally, mitochondrial respiration was consistently decreased in TNBC organoids upon treatment with paclitaxel or epirubicin, with basal and maximal OCR values lower than the untreated condition (Additional file 1: Fig. S5B-G), in line with our previous results obtained with TNBC cell lines.

Our data from the in vitro models suggest that glycolytic activity may have a predictive value for NAC treatment response in TNBC. Therefore, we next addressed whether glycolytic activity could predict final NAC response in TNBC patients. To do so, we first analysed the transcriptomic profile of diagnostic biopsies obtained from human patients with early TNBC, treated with NAC (Fig. 6A). Among the 29 adult female patients included in this study, 21 were considered as responders and 8 as non-responders for NAC (Fig. 6B and Table 1). Surprisingly, RNA-sequencing analysis did not identify genes being significantly regulated between the two groups (i.e. non-responders vs responders), suggesting that the observed resistance cannot be attributed to the differential regulation of one specific gene. However, when assessing gene sets by GSEA, we identified several significantly enriched hallmarks (Fig. 6C). The glycolysis hallmark was one of the most upregulated pathways in non-responders (Fig. 6C, D), thereby suggesting a potential predictive value for glycolytic activity in the final response to NAC in TNBC patients. By analysing two other publicly available datasets (GSE123845 and GSE25066) [25, 26], containing RNA-seq data from diagnostic biopsies in TNBC patients, and for which information about the response to NAC (assessed by RCB score) was available (Fig. 6B), we validated glycolysis as one the main hallmarks being upregulated in diagnostic biopsies from non-responders (i.e. RCB II/III) (Fig. 6E, F and Additional file 1: Fig. S6A, B). Encouraged by these results, we assessed the therapeutic potential of glycolysis inhibition to improve NAC response for decreasing TNBC cell aggressiveness. We showed that a combinatory treatment with paclitaxel and 2-DG led to a reduced colony-forming ability in TNBC cells, at a greater extent than single treatments, in particular for HCC38 cells (Fig. 7A). Moreover, a dramatic synergistic inhibition of 3D growth capacity was observed for TNBC cell line-derived spheroids upon treatment with paclitaxel and 2-DG (Fig. 7B), but not with epirubicin (Fig. 7C). Importantly, similar growth-inhibitory effects were obtained upon treatment of patient-derived TNBC organoid models with NAC and 2-DG (Fig. 7D–E), thereby validating our initial findings with clinically relevant models of early TNBC. Altogether these data highlight the importance of glycolysis activity in supporting resistance to NAC in TNBC and provide further rationale for the development and use of glycolysis-related biomarkers or inhibitors to better predict or improve the response to NAC in human TNBC patients, respectively.