mTOR-driven autophagy suppression defines metabolic vulnerability in CDK4/6 inhibitor-resistant HR⁺/HER2⁻ breast cancer

Abstract

Breast cancer (BC) is the most prevalent malignancy in women, with hormone receptor-positive, HER2-negative (HR⁺/HER2⁻) tumors representing ~70% of cases. While CDK4/6 inhibitors (CDK4/6i) combined with endocrine therapy have transformed treatment for metastatic HR⁺/HER2⁻ BC, acquired resistance remains a major obstacle. Using HR⁺/HER2⁻ BC models with acquired resistance to the CDK4/6 inhibitors Palbociclib or Ribociclib, we uncovered a metabolic vulnerability in highly resistant clones, mediated by mTORC1 hyperactivation and autophagy suppression. Gene expression profiling revealed enrichment of glycolysis and mTORC1 pathways in CDK4/6i-resistant cells, which manifested as heightened sensitivity to the metabolic inhibitors Metformin and Dichloroacetate (DCA). Mechanistically, mTORC1 overactivation impaired autophagy via ULK1-Ser757 phosphorylation, as confirmed by LC3 flux assays, leaving resistant cells unable to adapt to energy stress. Treatment with metabolic drugs triggered AMPK activation, ACC inhibition, and PARP cleavage, culminating in apoptosis. Clinically, immunohistochemical analysis of a BC cohort revealed a significant correlation between mTORC1 activity (p4E-BP1^T37/46) and autophagy suppression (p62 accumulation), supporting the translational relevance of this axis. Our findings propose mTORC1-mediated autophagy defects as a biomarker for metabolic vulnerability in CDK4/6i-resistant BC, offering a rationale for targeting these tumors with metabolic therapies to overcome resistance.

Introduction

Breast cancer (BC) is the most commonly diagnosed cancer and the leading cause of cancer-related mortality among women worldwide, with incidences continuing to rise [1–3]. Based on molecular characteristics, BC is divided into four molecular subtypes that help predict clinical outcomes and determine treatment approaches [4, 5]. Among these subtypes, the most prevalent is hormone receptor-positive human epidermal growth factor receptor 2-negative (HR⁺/HER2⁻), accounting for approximately 70% of all BC cases [6].

Alterations in cell-cycle regulatory pathways, such as CCND1 (Cyclin D1) amplification, are frequently observed in HR⁺ BC and associated with poorer prognosis, highlighting the cell cycle as a promising therapeutic target in this subtype [6–8]. Cell cycle progression is tightly controlled by cyclin-dependent kinases (CDKs), a family of serine/threonine kinases that interact with cyclins to phosphorylate target proteins. A critical step is the G1-to-S phase transition, governed by phosphorylation of the retinoblastoma protein (Rb1) via the CDK4/6-cyclin D complex [7]. Based on this rationale, dual CDK4/6 inhibitors (CDK4/6i) have been developed and, in combination with endocrine therapy, have become first-line treatment for advanced and metastatic HR⁺/HER2⁻ BC [8–14].

Despite the transformative clinical success of CDK4/6 inhibitors in the treatment of HR⁺/HER2⁻ BC patients, acquired resistance has emerged as a major challenge, prompting intensive research into underlying molecular mechanisms and potential strategies to overcome it [15]. Resistance to CDK4/6i has been associated with loss or inactivation of RB1, mutations in TP53, FAT1, and ESR1, as well as amplification of CDKN2A (p16), CDK6, CDK2, CCNE1 (Cyclin E), c-MYC, and MDM2. Additionally, hyperactivation of alternative signaling pathways, like HER2, FGFR1/2, and the RAS/MEK/ERK and PI3K/AKT/mTOR cascades, has also been implicated in resistance [15–17].

The serine/threonine kinase mTOR (mechanistic target of rapamycin) acts in two distinct complexes, mTORC1 and mTORC2, to regulate cell growth and proliferation in response to growth factors, nutrient availability, and oxygen levels. Under nutrient-rich and growth-stimulating conditions, mTORC1 promotes anabolic processes including protein, lipid and nucleotide synthesis while simultaneously suppressing catabolic processes such as macroautophagy (hereafter referred to as autophagy) [11, 18, 19]. mTORC1 inhibits autophagy primarily through phosphorylation of Unc-51 Like Autophagy Activating Kinase 1 (ULK1) at Ser757, a modification that blocks autophagy initiation [20]. Autophagy is a tightly regulated degradative process in which double membraned vesicles (autophagosomes) sequester cellular components and fuse with lysosomes to enable their degradation and recycling, thereby maintaining cellular homeostasis [21]. Under conditions of starvation or energy stress, autophagy becomes essential for sustaining cellular metabolism by supplying amino acids and lipids for protein synthesis and gluconeogenesis. Catabolized metabolites can also be routed through the TCA cycle to support glycolysis, thereby preserving ATP levels and preventing apoptosis [22, 23].

It is well established that tumor cells undergo metabolic reprogramming, favoring anabolic pathways to support unrestricted growth and proliferation, thereby increasing their reliance on substrate availability [23, 24]. This is characterized by an elevated glycolytic flux, altered TCA cycle activity and a decoupling of these processes, allowing diversion of metabolic intermediates into biosynthetic pathways while preserving NADPH generation [24].

These metabolic dependencies present potential therapeutic vulnerabilities. Agents such as dichloroacetate (DCA), an inhibitor of pyruvate dehydrogenase kinase (PDK), can restore coupling of glycolysis to the TCA cycle, while biguanides like metformin inhibit the mitochondrial electron transport chain, thereby reducing ATP and NAD⁺ production [25–28]. However, the clinical application of metabolic drugs remains limited, largely due to the lack of well-defined molecular alterations that create or predict metabolically targetable phenotypes [29].

Here, we demonstrate that mTOR activity determines metabolic susceptibility in HR⁺/HER2⁻ BC resistant to CDK4/6 inhibitors. Mechanistically, this phenotype is driven by an mTORC1-mediated autophagy defect, a hallmark of highly resistant tumor cells. Immunohistochemical analysis of a BC cohort revealed a significant inverse correlation between autophagy levels and mTORC1 activity, supporting the clinical relevance of this concept. Collectively, these findings identify a potential molecularly defined subgroup of BC patients who may derive therapeutic benefit from metabolic drugs.

Results

Generation of CDK4/6i-resistant breast cancer clones

In order to identify novel therapeutic options for breast cancer patients who develop resistance to CDK4/6 inhibitors, we aimed to characterize the vulnerabilities of CDK4/6i-resistant breast cancer cells. We employed two well-established HR⁺/HER2⁻ BC cell lines, T47D and MCF7, and generated resistant derivatives through a gradual dose escalation protocol using the selective CDK4/6 inhibitors Ribociclib and Palbociclib. This approach yielded multiple resistant subclones for each drug–cell line combination. To assess the degree of resistance, we performed clonogenic growth assays across all adapted clones (Fig. 1a, Supplementary Figs. 1 and 2) [30]. Quantitative analysis revealed a spectrum of CDK4/6i responsiveness, with parental cells showing the highest sensitivity (Fig. 1b).

Fig. 1. Generation of CDK4/6i-resistant breast cancer clones.

a T47D and MCF7 BC cells were treated with increasing doses of Ribociclib or Palbociclib to obtain CDK4/6i-resistant subclones. b Quantification of clonogenic growth assays of Supplementary Figure 1. Clonogenicity is shown relative to that of untreated cells as mean of biological replicates ± SD, n = 3. c Clonogenic growth assay of the selected Ribociclib and Palbociclib resistant clones treated with 1 µM of Palbo- and Ribociclib. d Cell viability of parental, strong and partial CDK4/6i-resitant subclones after 5-day treatment with Palbociclib or Ribociclib. Shown is the mean of biological replicates ± SD, n = 3.

From each CDK4/6i – cell line combination, we selected one partially resistant clone (T47D-Ribociclib α6, T47D-Palbociclib d5; MCF7-Ribociclib b4, MCF7-Palbociclib a5) and one strongly resistant clone (T47D-Ribociclib e6, T47D-Palbociclib b3; MCF7-Ribociclib b5, MCF7-Palbociclib b2) for mechanistic investigation (Fig. 1c). To validate the degree of resistance, we performed dose–response analyses in comparison to the respective parental cell lines (Fig. 1d). CDK4/6i-adapted clones displayed markedly reduced sensitivity to both Ribociclib and Palbociclib, consistent with their classification into partially and strongly resistant phenotypes. These findings confirm the successful establishment of a robust in vitro model for studying CDK4/6i-resistance in HR⁺/HER2⁻ breast cancer (Fig. 1d).

CDK4/6i-resistance correlates with glycolysis dependency

The heterogeneous CDK4/6i-resistance observed in our models suggested the involvement of diverse adaptive mechanisms. To investigate the underlying molecular alterations, we employed HTG EdgeSeq targeted RNA-sequencing (covering 2549 genes) to compare gene expression profiles between parental T47D cells and their strongly Ribociclib- or Palbociclib-resistant derivatives. The observed enrichment of glycolysis-related genes in this tumor cell type suggested a metabolic reliance on aerobic glycolysis (Fig. 2b). Given that Metformin exerts antitumor effects, at least in part, by disrupting mitochondrial complex I and thereby perturbing cellular energy homeostasis, we hypothesized that highly glycolytic cells— which function near the upper limit of glycolytic ATP production—would be particularly susceptible to further energetic stress. This putative metabolic vulnerability led us to evaluate Metformin, a clinically approved drug, in this context, along with dichloroacetate (DCA), a pyruvate dehydrogenase kinase (PDK) inhibitor that impairs glycolysis by redirecting glycolytic intermediates into the tricarboxylic acid (TCA) cycle [26, 31].

Fig. 2. Strongly CDK4/6i-resistant BC clones are sensitive to metabolic drugs.

a Therapy-naïve parental and strongly Ribociclib or Palbociclib resistant T47D clones were analyzed using HTG EdgeSeq to perform GSEA analyses. b Enrichment plots from GSEA analyses comparing the expression of the hallmark genes involved in glycolysis in strong Ribociclib- or Palbociclib resistant clones versus parental T47D cells. c–d, Quantification of clonogenic growth assays. Clonogenic growth of parental and CDK4/6i-resistant cells treated with indicated drugs for 10 days. Clonogenicity is normalized to parental untreated cells and presented as the mean of biological replicates ± SD, n = 3. e–f, Clonogenic growth of parental and selected partially or strongly CDK4/6i-resistant clones treated with Metformin or DCA at the indicated concentrations.

In clonogenic growth assays, resistant clones displayed a broad spectrum of responses: while some resembled parental cells in their insensitivity, others — particularly those with strong CDK4/6i- resistance — exhibited pronounced susceptibility to metabolic inhibition (Fig. 2c–f, Supplementary Figs. 1–2). To systematically assess this relationship, we correlated the degree of CDK4/6i-resistance (Fig. 1) with sensitivity to metabolic drugs (Fig. 2). A significant positive correlation was observed across the panel of adapted clones (Fig. 3a, b), indicating that strong CDK4/6i-resistance is associated with a metabolic vulnerability, which may represent a therapeutically exploitable weakness in this subset of HR⁺/HER2⁻ breast cancers.

Fig. 3. Metabolic drug treatment induces a severe energy crisis and apoptosis in CDK4/6i-resistant BC cells.

a–b Relative clonogenicity data of parental and CDK4/6i-adapted T47D and MCF7 cells treated with CDK4/6i or metabolic drugs, as shown in Figs. 1 and 2, were correlated. Data represent the mean of biological replicates ± SD, n = 3. The curve was fitted using linear regression, and statistical significance was assessed using Pearson’s correlation coefficient (r), with p < 0.05 considered significant. c Flow cytometry analysis of apoptosis (sub-G1) following 5 days of treatment with DCA or Metformin. Biological replicates are presented as individual data points with the mean ± SD, n = 3. Statistical significance was determined using two-way ANOVA with Tukey’s multiple comparisons test, p < 0.05 *; p < 0.01 **; p < 0.001 ***; p < 0.0001 ****. d Western blot of parental or strongly CDK4/6i-resistant T47D cells treated with 40 mM DCA or 4 mM Metformin for 48 h.

Metabolic targeting triggers severe energy crisis and apoptosis in CDK4/6i-resistant BC cells

To elucidate the mechanistic basis of the observed metabolic vulnerability, we examined drug-induced apoptosis across our panel of resistant clones. Flow cytometric analysis using PI staining demonstrated that Metformin and DCA treatment specifically induced apoptosis in the strongly resistant clones (e6 and b3), while partially resistant clones (α6 and d5) and parental cells remained largely unaffected (Fig. 3c). This selective apoptotic response correlated well with the clonogenic survival patterns we observed (Fig. 2c–f). Western blot analysis revealed the molecular cascade underlying this differential response. In strongly resistant clones, both Metformin and DCA treatment triggered robust phosphorylation of AMPK at Thr172, indicative of cellular energy stress. This was accompanied by phosphorylation and inhibition of acetyl-CoA carboxylase (ACC) at Ser79, followed by cleavage of PARP, confirming the induction of apoptotic cell death (Fig. 3d). These molecular events were notably absent in parental cells.

Collectively, these findings demonstrate that high-level CDK4/6i-resistance confers a metabolic vulnerability to both Metformin and DCA, which induce an energy crisis in resistant cells culminating in apoptotic cell death.

mTOR hyperactivation marks CDK4/6i-resistant, metabolically vulnerable BC cells

Hyperactivation of the PI3K/AKT/mTOR pathway has been implicated as a mechanism of CDK4/6i resistance through its functional crosstalk with the Cyclin D1/CDK4/6/Rb axis [32–34]. To explore this connection in our metabolically vulnerable resistant clones, we analyzed mTOR pathway activity through multiple complementary approaches. Re-examination of our HTG EdgeSeq data revealed significant enrichment of the mTORC1 signaling gene set in strongly CDK4/6i-resistant cells compared to parental controls (Fig. 4a).

Fig. 4. Elevated mTOR signaling in strongly CDK4/6i-resistant and metabolically vulnerable BC cells.

a Enrichment plots from GSEA analyses (Fig. 2a) comparing the expression of hallmark genes involved in mTORC1 signaling in strongly CDK4/6i-resistant clones versus parental T47D cells. b–c Western blot analysis of parental and CDK4/6i-resistant T47D cells. Protein levels were quantified using ImageJ and normalized to Actin as the loading control. The pULK (Ser757)/ULK (total) and pp70S6K (T389)/p70S6K (total) ratios were calculated by dividing the values of phosphorylated proteins by the corresponding total protein levels.

This transcriptional signature was confirmed at the protein level through immunoblot analysis. Both strongly resistant T47D clones (b3 and e6) and MCF7 clones (b2 and b5) showed markedly elevated mTORC1 activity, as evidenced by increased phosphorylation of canonical downstream target sites pp70S6K^T389 and p4E-BP1^T37/46 (Fig. 4b–c). In contrast, partially resistant clones maintained mTORC1 activity levels comparable to therapy-naïve parental cells. Notably, the heightened mTORC1 signaling in strongly resistant clones was associated with pronounced phosphorylation of ULK1 at Ser757, a well-characterized mechanism of mTORC1-mediated autophagy suppression (Fig. 4b–c).

We conclude that elevated mTORC1 signaling is a distinguishing feature of strongly CDK4/6i-resistant cells compared to their partial and therapy-naïve counterparts, suggesting that mTORC1-mediated suppression of autophagy may contribute to the enhanced susceptibility of CDK4/6i-resistant BC cells to DCA or Metformin treatment.

mTORC1 hyperactivation suppresses autophagy in strongly CDK4/6i-resistant BC cells

As a central regulator of cellular metabolism, mTOR coordinates cell growth with environmental cues by modulating critical catabolic processes including autophagy [35]. Autophagy serves as an essential survival mechanism when cells face metabolic challenges such as treatment with energy stress-inducing drugs. Our observation of significant mTORC1 hyperactivation and consequent ULK1-Ser757 phosphorylation in strongly resistant clones (Fig. 4b, c), led us to hypothesize that these cells may exhibit impaired autophagy initiation compared to their partially resistant and parental counterparts—potentially accounting for their metabolic vulnerability. To test this hypothesis, we performed a sensitive LC3 turnover assay to quantitatively measure autophagic flux dynamics across parental, partially resistant, and strongly resistant T47D clones. All cell populations were transduced with HiBiT-tagged LC3, enabling precise quantification of LC3 protein levels through NanoLuc-mediated luminescence complementation (Fig. 5a) [36]. Recognizing that basal LC3 levels poorly reflect autophagic activity, we also evaluated LC3 accumulation following lysosomal inhibition with chloroquine (CQ). Parental cells exhibited robust LC3 accumulation, demonstrating normal autophagic flux. Partially resistant clones showed comparable LC3 turnover, while strongly resistant clones displayed complete lack of accumulation, revealing a severe autophagy defect (Fig. 5b).

Fig. 5. mTORC1 hyperactivation impairs autophagy in strongly CDK4/6i-resistant BC cells.

a Schematic representation of the LC3-HiBiT reporter assay workflow with interpretation of the shown readout to measure basal autophagy, autophagy induction, and autophagic flux. b–d, LC3 turnover assays of parental and CDK4/6i-resistant T47D cells stably transfected with LC3-HiBiT reporter. b Measurement of basal autophagy. Parental, strongly and partially CDK4/6i-resitant cells were treated with increasing doses of chloroquine (CQ), up to 100 µM, for 6 h. Data are presented as mean of biological replicates ± SD, n = 2. The curve was fitted using nonlinear regression analysis. c–d, Measurement of autophagy induction (upper panel). LC3-HiBiT expressing parental, strongly and partially CDK4/6i-resitant cells were treated with increasing doses of DCA or Metformin for 6 h. The LC3-HiBiT reporter activity was quantified by luminescence and normalized to the untreated control. Measurement of autophagic flux (lower panel). LC3-HiBiT expressing parental, strongly and partially CDK4/6i-resitant cells were pre-treated with 50 μM chloroquine (CQ) for 48 h, as indicated, and subsequently exposed to increasing concentrations of Metformin or DCA for 6 h. LC3-HiBiT reporter activity was assessed by measuring luminescence and normalized to the untreated control. Data are expressed as mean of biological replicates ± SD, n = 2, with curve fitting performed using nonlinear regression analysis.

We next examined autophagic responses to metabolic stress. Treatment with Metformin or DCA induced the expected dose-dependent LC3-HiBiT decrease in both parental and partially resistant cells, indicating active autophagic degradation (Fig. 5c, d, upper panel). Most strikingly, when we challenged cells with combined chloroquine and metabolic drug treatment, parental cells showed vigorous LC3 accumulation. While partially resistant clones maintained some autophagic capacity, strongly resistant clones were completely unable to mobilize autophagic flux under these conditions (Fig. 5c, d, lower panel).

Collectively, these results suggest that a strong CDK4/6i resistance is linked to an mTORC1-mediated autophagy defect, which renders resistant cells vulnerable to metabolic perturbation by targeted drugs.

mTORC1 activity correlates with autophagy suppression in patient tumors

To evaluate whether the concept of targeting mTOR-mediated autophagy defects with metabolic drugs could be translated to BC patients, we analyzed tissue microarrays (TMAs) from a cohort of 174 patients, obtaining valid immunohistochemical staining for 133 tumor samples (Fig. 6a). The cohort reflects a breast cancer population, predominantly comprising patients diagnosed with a T1–T2 invasive ductal carcinoma without lymph node involvement. Molecularly, most patients were classified as Luminal A with intermediate histological grade (G2) (Fig. 6b). Tumor sections were stained for phosphorylated 4E-BP1 at threonine 37/46 (p4E-BP^T37/46), a marker of mTORC1 signaling activity, and for p62, a substrate of autophagic degradation [37, 38]. To quantify expression levels, we established an immunoreactive score (IRS) system (range 0 – 12), categorizing tumor spots by percentage of positive tumor cells (0 – 4) and intensity of staining (0 – 3) (Fig. 6c, d) [39].

Fig. 6. 4E-BP1^T37/46 Phosphorylation and p62 accumulation correlate in breast cancer patient samples.

a–d, Tissue microarrays (TMAs) of paraffin-embedded breast cancer patient samples (n = 174) were immunostained for p4E-BP1^T37/46 and p62 and analyzed. a CONSORT diagram detailing the number of samples available for immunoreactivity evaluation analysis. b Composition of the analyzed breast cancer cohort (n = 133) stratified by histological and molecular subtype, grading, lymph node status and tumor stage. Invasive ductal carcinoma (IDC), Invasive lobular carcinoma (ILC), HER2-positive breast cancer (HER2 + ), Triple-negative breast cancer (TNBC). c–d Representative immunostaining patterns show the intensity of staining (IS) and the percentage of positive cells (PS), with IS graded from 0 (none) to 3 (strong), and PS ranging from <10% (1) to >80% (4). The IRS score for p4E-BP1^T37/46 and p62 was calculated by multiplying IS and PS. Scale bars: 20 μM. e The IRS-score was transformed into a simplified scoring system for subsequent analyses (Score 0 = IRS 0; Score 1 = IRS 1–3; Score 2 = IRS 4–8; Score 3 = IRS 9–12). The correlation between p4E-BP1^T37/46 and p62 is illustrated across the entire cohort (n = 133). The statistical significance was determined by Chi-square test, p < 0.05 *; p < 0.01 **; p < 0.001 ***.

Interestingly, we observed a significant correlation between p4E-BP1^T37/46 and p62 expression, indicating that elevated mTORC1 signaling is associated with impaired autophagic degradation within this institutional BC patient cohort (Fig. 6e).

These findings support the translational relevance of our model, suggesting that mTORC1-driven autophagy defects are present in clinical breast cancer and may identify a subset of patients who could benefit from metabolic therapies targeting this vulnerability.

Discussion

Disease progression under therapy with CDK4/6i presents a growing challenge, necessitating a deeper understanding of the biology of therapy-resistant breast cancer and the development of novel therapeutic strategies to overcome resistance and improve patient outcomes [15]. In our study, we observed that treatment of therapy-naïve breast cancer cells with CDK4/6-inhibitors leads to the emergence of a strongly resistant subpopulation, distinct from the bulk of partially resistant BC cells (Fig. 1a–d). Notably, only the strongly CDK4/6i-resistant cells exhibited an increased dependence on OXPHOS and glycolysis (Fig. 2a – b), a vulnerability that can be pharmacologically exploited using metabolic drugs such as DCA or Metformin (Fig. 2c –d and Fig. 3c–d). Mechanistically, this metabolic dependency is driven by constitutive upregulation of mTORC1 activity (Fig. 4a–c), resulting in an impairment of autophagy (Fig. 5b–d). Crucially, metabolic treatments fail to induce autophagy in the strongly CDK4/6i-resistant cells (Fig. 5c–d), resulting in severe energy stress and subsequent induction of apoptosis (Fig. 3c–d). This metabolically vulnerable phenotype of the strongly resistant CDK4/6i BC cells is consistent across different cell lines and CDK4/6-inhibitors and contrasts with the phenotype of the partially CDK4/6i-resistant cells which lack this metabolic vulnerability (Fig. 3a–b).

In line with our findings, increased activation of mTORC1, its upstream regulators and its downstream targets are among the most commonly discussed molecular alterations observed in CDK4/6i-resistant cells [15, 17, 32–34, 40]. This is unsurprising, as there is extensive crosstalk between mTORC1 and the CDK4/6-Cyclin D complexes. For instance, mTORC1 and AKT regulate Cyclin D levels, while CDK4/6 activates mTORC1 via inhibitory phosphorylation of TSC1/2 [41–43]. Moreover, patients with increased activation of the PI3K/AKT/mTOR signaling axis have been shown to be more likely to be non-responsive when treated with endocrine therapy plus Palbociclib [44]. Therefore, combinations of mTOR- or PI3K-inhibitors with CDK4/6-Inhibitors have been found to reverse or delay resistance to CDK4/6-inhibitors in several preclinical models and have already shown promising clinical application in phase-III studies [45–47]. In our work, we take this one step further by exploiting not the increased activation of mTOR itself but rather one of its downstream effects – specifically, the heightened dependency on nutrient availability caused by the inhibition of autophagy.

Although alterations in cellular metabolism are a hallmark of cancer, they are more frequently driven by regulatory changes in signaling networks rather than by mutations in genes encoding metabolic enzymes [24]. This complicates the identification of tumors with increased metabolic dependency that may be susceptible to metabolic agents, particularly given the dose-limiting toxicities that have restricted their clinical use [29]. Cellular energy metabolism is intricately linked to key regulators of cell cycle control, most notably CDK4, D-type cyclins, and E2F transcription factors. These connections arise both indirectly, through their regulation of the mTORC1 signaling axis, and directly via independent mechanisms. This provides a strong rationale for anticipating therapeutically relevant metabolic changes in cells that have developed resistance to CDK4/6 inhibitors [48, 49]. In esophageal squamous cell carcinoma, Palbociclib-resistant cells exhibited a metabolic vulnerability driven by a Fbxo4-Cyclin D1 axis, leading to increased glutamine dependency, which could be pharmacologically targeted with a combination of Metformin and a glutaminase 1 inhibitor [50]. Similarly, in preclinical models of head and neck squamous cell carcinoma, Metformin acted as a senostatic drug, enhancing the anticancer efficacy of CDK4/6 inhibition [51]. In breast cancer, a fasting-mimicking diet restored sensitivity to Abemaciclib in both cell culture and mouse models, highlighting the potential of metabolic perturbations as a therapeutic strategy in the context of CDK4/6 inhibitor resistance [52]. Moreover, a genome-wide CRISPR knockout screen identified a synergy between Metformin and CDK4/6 inhibition, which was validated across multiple cancer cell lines, including breast cancer. However, the same study found that Metformin as a single agent failed to induce significant apoptosis [53]. Using publicly available drug screening data, we observed no significant correlation between Metformin and CDK4/6i sensitivity, although an inverse association might have been expected (Supplementary Fig. 3a). This may be explained by the fact that the DepMap BC cell lines are CDK4/6i-naïve and biologically heterogeneous. In summary, our data suggest that impaired autophagy may serve as a key marker for identifying breast cancer cells susceptible to metabolic drugs, thereby enabling their targeted use as novel therapeutic option to overcome CDK4/6i-resistance.

The therapeutic significance of autophagy in tumor cell survival extends beyond metabolic inhibition. In the context of CDK4/6i, preclinical studies have demonstrated a synergistic effect between autophagy inhibition and CDK4/6-inhibition using Hydroxychloroquine and Palbociclib, a combination that has also shown promising efficacy in early-stage clinical trials for patients with metastatic breast cancer (mBC) who relapsed after Palbociclib treatment [54, 55]. Importantly, autophagy defects are not exclusive to post-treatment progression; monoallelic loss of BECN1, a key autophagy inducer and potential haploinsufficient tumor suppressor, is observed in approximately 40% of breast cancers. This underscores the existence of sporadic autophagy deficiencies, which, if identified, could render tumors metabolically vulnerable [56].

A key aspect of our approach is the accurate and efficient assessment of autophagy status in breast cancer patients to enable a personalized, precision medicine-based strategy for identifying individuals who may benefit from metabolic therapy [29]. However, monitoring autophagy in clinical practice remains challenging. While genomic sequencing theoretically offers insights into autophagy status, mRNA levels of autophagy-related genes often do not correlate with observed autophagic activity, making this approach unreliable [57]. In line with this, DepMap-based correlation analyses of cell lines revealed no significant association between autophagy-related gene or protein expression and sensitivity to Metformin or DCA, although a trend toward greater Metformin sensitivity was observed in cell lines with lower ATG7, FIP200, and ATG14 expression (Supplementary Fig. 3b–d) [58]. Notably, CRISPR knockout data indicate that cell lines lacking ATG7, ATG14, or FIP200 are highly sensitive to metabolic inhibitors such as DCA and Metformin highlighting the challenge of accurately assessing functional autophagy activity based solely on gene or protein expression levels [27, 57]. Immunohistochemical (IHC) analysis of autophagic markers is currently considered the most specific and reliable method for assessing functional autophagy status in tumors [37]. We evaluated tissue microarrays (TMAs) from a breast cancer patient cohort for markers of autophagy and mTORC1 activity (Fig. 6a–c). Our analysis revealed a significant correlation between these markers, indicating the presence of mTORC1-mediated autophagy defects within the cohort and supporting the clinical applicability of our therapeutic strategy (Fig. 6e). Moreover, the diverse patient population suggests that this concept may not be limited to hormone receptor-positive (HR⁺) breast cancer following relapse on CDK4/6 inhibitors but could extend to other subtypes and clinical contexts (Fig. 6d).

In summary, we have identified breast cancer patients who have progressed on CDK4/6 inhibitors as a group that may benefit from metabolic therapy. This metabolic vulnerability in highly resistant cells is driven by mTORC1 hyperactivation, which induces autophagy suppression. Our IHC-based correlation of p4E-BP1 and p62 staining suggests that autophagy-deficiency-dependent sensitivity to metabolic drugs could be translated into clinical practice, offering a potential biomarker for patient stratification and targeted therapy.

Methods

Cell culture

T47D and MCF7 cell lines were obtained from the American Tissue Collection Center (ATCC) and grown in high-glucose DMEM supplemented with 10% fetal bovine serum, 100 U ml⁻¹ and 100 µg ml⁻¹ Streptomycin at 37 °C with 5% CO₂. All cell lines were regularly tested and found to be negative for mycoplasma contamination. The metabolic drugs and molecular inhibitors were obtained from Sigma-Aldrich unless indicated otherwise and used at the following concentrations: Dichloroacetate (DCA) 0.078–40 mM, Metformin 7.8 µM–4 mM, Chloroquine 0.39–100 µM, Ribociclib (MedChemExpress) 0.125–32 µM, and Palbociclib (MedChemExpress) 0.125–32 µM. Standard concentrations used in clonogenic growth assays were 40 mM DCA, 4 mM Metformin, 1 µM Ribociclib, and 1 µM Palbociclib.

Generation of CDK4/6i-adapted cell clones

CDK4/6i-resistant T47D and MCF7 cells were generated using a dose escalation approach. Starting concentrations were set at 100-fold lower than the IC₅₀ values for the respective cell line-inhibitor combinations. Parental T47D cells were treated with Ribociclib at concentrations ranging from 9.1 nM to 1 µM, and Palbociclib at concentrations from 23.2 nM to 1 µM. Parental MCF7 cells were treated with Ribociclib from 0.73 nM to 1 µM, and Palbociclib from 2.8 nM to 1 µM. The resistant cell clones were selected and continuously cultured with 1 µM of the respective CDK4/6i. For the experiments, the cells were cultured in absence of the CDK4/6-inhibitors.

HTG EdgeSeq and gene set enrichment analysis

1 × 10⁶ cells were plated and cultivated for 48 h. For cell harvesting and counting, adherent cells were treated with Trypsin-EDTA solution (2x) for 5 min at 37 °C. The cells were resuspended in DMEM (with 10% FBS) and centrifuged. After washing with PBS, viable cell count was determined, and the appropriate number of cells (either 4 million or 2 million) was collected in a 1.5 mL Eppendorf tube. Gene expression analysis was performed targeting 2549 genes using the HTG Oncology Biomarker Panel (OBP) and HTG’s extraction-free nuclease protection assay (HTG Molecular Diagnostics, Tucson, USA). Cells were lysed and diluted to 3000 cells/35 µL of lysis buffer. 35 µL of the lysate was processed in the HTG EdgeSeq processor. After library preparation and pooling, samples were sequenced for 75 cycles on a NextSeq 550 Dx sequencer (Illumina, San Diego, USA). Read alignment and transcript counting were performed using HTG EdgeSeq Parser v5.4 with the Bowtie 2 algorithm. Differential expression analysis of the obtained transcript counts was performed using edgeR (version 3.40.0) [59]. Trimmed mean of M-Values (TMM) was used to calculate normalization factors to adjust library sizes. After estimating dispersion, differentially expressed genes were tested using the exact test proposed by Robinson and Smyth [60] followed by Benjamini-Hochberg correction. Adjusted library sizes were then used to calculate Counts per Million (CPM). CPM values were used as expression values for subsequent gene set enrichment analysis [61] GSEA software (version 4.2.6.1) was used to perform GSEA using the Hallmark gene sets collection (H) and the curated (C2), oncogenic (C6) and immunologic (C7) signature gene set collections from MSigDB (version 7.1) using phenotype permutations (1000 permutations) and Signal2Noise metric for gene ranking. Otherwise, default parameters were used.

Clonogenic growth assays and quantification

Cells were seeded overnight, treated, and cultivated for up to 10 days. The plates were fixed with 70% ethanol overnight and stained for 30 min with crystal violet solution (Sigma Aldrich) diluted 1:20 in 20% ethanol. Colony growth on each plate was quantified using the colony area plugin for ImageJ (version 1.53). Staining intensity was assessed by selecting three 15 × 15 pixel regions per plate, calculating the mean intensity, and normalizing it to that of an untreated plate of the same cell line after an equivalent growth period.

Cell viability assays

Cell viability in response to treatment was measured using the CellTiter-Glo assay (Promega) following the manufacturer’s protocol. Cells were plated in white-walled 96-well plates overnight, and triplicate wells were treated with the respective inhibitors diluted in 80 µl of DMEM the following day. After 7 days of treatment, 80 µl of CellTiter-Glo reagent was added, and the plates were incubated for 10 minutes. Luminescence was measured using the Orion II luminometer (Berthold). Background signals from empty wells were subtracted, and the luminescence values were normalized to those from untreated controls. Dose-response curves were fitted to the data using GraphPad Prism (version 10.2.0).

Western blotting

Cells were harvested and lysed in NP-40 lysis buffer (50 mM Tris-HCL, 150 mM NaCl, 5 mM ethylenediaminetetraacetic acid (EDTA), 2% NP-40, pH 8.0) and supplemented with protease inhibitor (complete ULTRA tablets EASYpack, Roche). The protein yield was determined by Bradford assay (Bio-Rad). A total of 25–35 µg protein was loaded and separated on NuPage SDS Gels (Life Technologies) and then tank blotted to PVDF membranes (Bio-Rad). The membranes were blocked in tris-buffered saline with polysorbate 20 (TBST; 5 mM Tris, 15 mM NaCl, 0.1% Tween 20, pH 7.5) with 5% nonfat dry milk and then incubated with primary antibodies diluted in TBST/5% nonfat dry milk overnight at 4°C. The primary antibodies used were: cleaved PARP (Asp214) (1:5000, #9541, Cell Signaling), phospho-ULK1 (Ser757) (1:5000, #14202, Cell Signaling), p62/SQSTM1 (1:5000, P0067, Sigma), LC3B (LC3I/II) (1:5000, ab48394, Abcam), phospho-4E-BP1 (Thr37/46) (1:5000, 236B4 #2855, Cell Signaling), phospho-AMPKα (Thr172) (40H9) (1:5000, #2535, Cell Signaling), phospho-Acetyl-CoA Carboxylase (Ser79) (1:5000, #3661, Cell Signaling), 4E-BP1 (1:1000, sc-9977, SantaCruzBiotech), phospho-p70S6Kinase (Thr389) (1:1000, 108D2 #9234, Cell Signaling), total p70S6Kinase (H9) (1:500, sc-8418, Santa Cruz Biotechnology), ULK1 (1:1000, #8054, Cell Signaling), AMPKα (1:1000, #2603, Cell Signaling), Acetyl-CoA Carboxylase (1:1000, #3676, Cell Signaling), β-Actin (AC-15) (1:10.000, ab6276, Abcam). Proteins were detected using secondary antibodies and the WesternBright ECL Substrate Sirius kit (Biozym). The secondary antibodies used were anti-mouse IgG-HRP (GE Healthcare, 1:2000) and anti-rabbit IgG-HRP (GE Healthcare, 1:5000). Quantification was performed using ImageJ, where the area of the bands was determined and normalized to the corresponding total protein as well as the β−Actin band.

Autophagy analysis

For functional autophagy measurement, the LC3-HiBiT reporter vector (Promega, GA2552) was used. The vector encodes a fusion protein consisting of human LC3B, a small N-terminal 11 amino acid HiBiT tag and a proprietary spacer region that enhances reporter specificity. It employs the constitutive HSV-TK promoter with PyF101 enhancer for low-to-moderate expression of the reporter in mammalian cells. The resistant cell lines were transfected with the vector using Lipofectamine 2000 and then selected with 800 µg/ml of Geneticin (Gibco). Transfection success was determined by comparing the HiBiT signal of the newly transfected cells to that of negative and positive controls. The stably transfected LC3-HiBiT reporter cell lines were then cultured under 40 µg/ml Geneticin. The day before an experiment, cells were plated on 96 well plates in the absence of Geneticin.

For luminescence measurement, the Nano-Glo HiBiT Lytic Detection System (Promega, N3040) and an Orion IIluminometer (Berthold) were used according to the manufacturer’s protocol. All measurements were performed in biological duplicates. To analyze basal autophagy, cells were treated with increasing concentrations of Chloroquine for 6 h before measuring the HiBiT signal. An elevated signal indicates an accumulation of LC3B, reflecting a high basal autophagy rate. Next, cells were treated with increasing concentrations of metabolic drugs for 6 h, and the HiBiT signal was measured to assess whether the autophagy cascade is activated in response. A decrease in signal correlates with LC3B degradation, indicating activation of autophagy. Finally, cells were first treated with Chloroquine for 48 h, washed, and then treated with increasing concentrations of metabolic drugs for 6 h. The HiBiT signal was measured again to assess the effect of metabolic drugs on autophagic flux. A stronger signal reflects increased LC3B accumulation, indicating heightened autophagic flux in response to metabolic drugs. After background subtraction, luminescence signals were normalized to untreated samples.

Apoptosis analysis

For each clone, 100,000 cells were seeded in triplicates overnight, treated with metabolic drugs and cultured for 5 days. On day 5, the cell culture supernatant was collected and the adherent cells were trypsinized and combined with their respective supernatant. The resulting cell pellets were washed with 3 ml washing buffer (PBS supplemented with 1% FBS and 1 mM EDTA). The cells were counted and 100,000 cells per sample were resuspended in 1 ml washing buffer and fixed by drop-wise addition of 5 ml ice-cold 70% ethanol while vortexing. Following overnight fixation, the cells were stained with 100 µl of 10 µg/ml propidium iodide solution supplemented with 100 µg/ml RNase A. The samples were then incubated at 37°C for 30 minutes and analyzed for sub-G1 content on a CytoFlex LX cytometer (Beckman Coulter). The final analysis of the data was done using FlowJo v.10 (BD Biosciences).

Immunohistochemical staining and tissue microarray analysis

Accumulation of p62 and p4E-BP1^T37/46 in patient tumor samples was evaluated using tissue microarrays (TMAs). Immunostaining was performed a using a DAKO Autostainer Link 48. Following deparaffinization, antigen retrieval was conducted by treating the TMA slides with Target Retrieval Solution Citrate (pH 6.0, S 2369 Fa. Dako) for 30 min in a pressure cooker. After blocking with Peroxidase-Blocking Solution (S2023; Dako), the samples were incubated with a 1:1000 dilution of an anti-p62 antibody (mouse mAB; clone 2C11; abcam) and a 1:200 dilution of an anti-p4E-BP1^T37/46 antibody (rabbit mAb; clone 236B4, Cell Signaling). For detection, the DAKO REAL EnVision HRP Rabbit/Mouse system (K5007; Dako) was used as the secondary antibody. The chromogen (K5007; Dako) reacted with HRP, leading to a colorimetric signal. The slides were subsequently digitized using an Aperio AT2 Slide Scanner (Leica Biosystems). Cytoplasmic expression of p62 and p4E-BP1^T37/46 was evaluated using VM Slide Explorer 2.2 (VMscope, Berlin, Germany). TMA spots were assessed for staining intensity (IS, graded 0–3) and the percentage of positive cells (PS, graded 1–4), with the IRS score for p4E-BP1^T37/46 and p62 calculated by multiplying IS and PS.

Correlation analyses using DepMap datasets

Correlation analyses were performed using publicly available datasets from the DepMap portal (https://depmap.org/portal/), including PRISM drug response profiles (PRISM Repurposing Public 24Q2), gene expression data (RNA-seq, log₂ TPM + 1; DepMap Public 25Q3 release), and proteomic data from the harmonized mass spectrometry–based CCLE proteomics dataset generated by the Gygi laboratory (DepMap Public 25Q3 release) for breast cancer cell lines [58]. Plots were generated using the DepMap Data Explorer and filtered to include up to n = 26 breast cancer cell lines. Exported data were subsequently analyzed and visualized using GraphPad Prism, version 10.4.1 (532), December 3, 2024.

Correlations were assessed between (i) Metformin and CDK4/6i (Ribociclib, Palbociclib) sensitivities (n = 26), (ii) SQSTM1 mRNA expression or relative p62 protein levels (n = 26 and n = 20, respectively) versus Metformin and DCA sensitivity, and (iii) ATG7, FIP200, and ATG14 mRNA or protein expression with Metformin sensitivity. Linear regression analyses were performed; Pearson correlation coefficients (r) and two-tailed p-values were calculated. Results were visualized as scatter plots with regression lines illustrating correlation direction and strength.

Statistical analysis

All statistical analyses and data visualizations were performed using GraphPad Prism software (version 10.4.1). Quantification of clonogenic growth assays and Western blot results was conducted using ImageJ software (version 1.53). Figures were created using Adobe Illustrator (version 28.5) and BioRender.com (Figs. 1a, 2, 5a). For in vitro experiments, no statistical methods were used to predetermine sample size. The results presented in the graphs represent the mean or median values obtained from n replicates. Error bars in the figures indicate the standard deviation (SD) unless stated otherwise. Pearson’s correlation coefficient (r) and the corresponding p-value were used to assess statistical significance of correlations (e.g., Fig. 3a). Experiments investigating the interaction of two variables (e.g. treatment, Fig. 3c) were analyzed using two-way analysis of variance (ANOVA) followed by multiple comparison testing according to Tukey. The ANOVA results and selected pairwise comparisons are reported in the figures. A p value of less than 0.05 was considered statistically significant. Where p values are represented as asterisks: * indicates p < 0.05, **p < 0.01, ***p < 0.001 and ****p < 0.0001. When results from representative experiments are shown (e.g., Western blots, clonogenic growth assays, and luciferase assays), they are based on at least two independent experiments that yielded similar results.

Author contributions

NG and TS jointly planned and supervised the study. NG generated CDK4/6 inhibitor-resistant clones. LvW performed most experiments with contributions from NG, AS and MWitt. NG, LvW and JT-S prepared and analyzed samples using HTG EdgeSeq. M.M. performed GSEA analyses. BMP and CD provided breast cancer tissue microarrays and clinical data. LvW and A-SL performed immunohistochemistry and statistical analysis of the tissue microarrays. TW, UW, TS and NG provided resources. M Wanzel, SG, TW and UW provided essential intellectual and conceptual input. LvW, NG and TS analyzed data, performed statistics and wrote the manuscript. All authors have read and approved the final manuscript.

Funding

NG was supported by the Clinician Scientist Program (SUCCESS-Program) of Philipps University Marburg and the UniversityHospital of Giessen and Marburg (UKGM), as well as the Clinical Trialist Program (SUCCESS meets MSNZ) of the Department ofMedicine. NG received research funding from the Deutsche Forschungsgemeinschaft (DFG) (GRK 2573/2 – 2024), the UniversityMedical Center Giessen and Marburg (UKGM) (Grant 3/2022 MR to NG), the von Behring-Röntgen Foundation (Grant 70_0027 toNG), the PE Kempkes Foundation (Grant 01/2021 to NG), and the Medical Foundation (Grant 04/2021 to NG). Open Access funding enabled and organized by Projekt DEAL.

Data availability

All data generated or analyzed during this study are included in this published article (and its supplementary figures). The following source western blot data are provided with this paper: uncropped images for for Figs. 3d, 4c and 4d. The GSEA data generated and analyzed for this paper are available at BioStudies with accession code E-MTAB-15265.

Competing interests

CD received personal fees from Novartis, Roche, MSD Oncology, Daiichi Sankyo, AstraZeneca, Molecular Health, and Merck, all outside the submitted work. CD is cofounder of Sividon Diagnostics. In addition, CD has a patent on VMScope digital pathology software with royalties paid; a patent WO2020109570A1—cancer immunotherapy pending; and patents WO2015114146A1 and WO2010076322A1—therapy response issued. PJ reports research grant and travel expenses from GILEAD Sciences GmbH out-side the submitted work. NH declares to be GBG Forschungs GmbH employee. GBG Forschungs GmbH reports financial funding from AstraZeneca and Myriad during the conduct of the study; received funding for research grants from Abbvie, Amgen, AstraZeneca, BMS, Daiichi-Sankyo, Gilead, Molecular Health, Stemline Menarini, Celgene/BMS, Novartis, Pfizer and Roche (paid to the institution); GBG Forschungs GmbH has licensing fees from VMscope GmbH. In addition, GBG Forschungs GmbH has a patent EP21152186.9 pending, a patent EP19808852.8 pending, and a patent EP14153692.0 pending. The other authors declare no conflicts of interest.

Ethics approval and consent to participate

Immunohistochemical staining and evaluation of breast cancer patient tissue samples were approved by the Ethics Committee of the Charité (Ethics Opinion No. EA1/139/05) and conducted in accordance with the Declaration of Helsinki. Written informed consent was obtained from all patients for the use of their tissue samples in this study.

Supplementary information

The online version contains supplementary material available at 10.1038/s41419-026-08496-5.