Human cell lines (non-malignant cell lines: 293T, and Huvec; non-small cell lung cancer cell line: A549; BC cell lines: MCF-7, MDA-MB-453 and MDA-MB-231; gastric cancer cell lines: NuGC4, MKN-45 and HGC27; melanoma cell line: A375; hepatocellular carcinoma cell line: HepG2; pancreatic cancer cell line: Panc-1) were originally purchased from the typical cell culture collection of the Committee of the Chinese Academy of Sciences Library and were cultured in RPMI-1640 medium (Invitrogen) or DMEM (Dulbecco's Modified Eagle's Medium) (Invitrogen) supplemented with 10% FBS (Bioind) and 1% penicillin–streptomycin at 37 ℃ with 5% CO₂ according to the supplier's instructions. Glucose-free RPMI-1640 medium (Thermo Fisher Scientific, cat. No. 11879020) was added with different concentrations of glucose and supplemented with 10% dialyzed FBS (ThermoFisher Scientific, cat. No. 26400-044) for the glucose deprivation assays. All cells tested negative for mycoplasma and were authenticated by short tandem repeat DNA fingerprinting at the Key Laboratory of Pharmaceutical Biotechnology and The Comprehensive Cancer Center of Nanjing University (Nanjing, China). All cell lines were not being passaged for more than 6 months in our study after resuscitation.

ISA quantified glucose metabolic flux. 1,2-¹³C₂ glucose was purchased from Merck (cat. no. 453188) and used as the tracer as previously described²⁰. Briefly, 10⁷ cells exposed to the indicated treatments were grown for 24 h in RPMI-1640 with 10% FBS. Next, the media was exchanged for RPMI-1640 containing 10 μmol/L 1,2-¹³C₂ glucose and 10% FBS, after which the cells were incubated for 24 h and harvested. Pre-cooled 80% (v/v) methanol was added to extract metabolites, followed by centrifugation at 14,000×g for 10 min at 4 ℃. The supernatant was dried using a SpeadVac at room temperature. The dried supernatant was stored at −80 ℃ until the analysis. The extracts were reconstituted with 200 mL 50% (v/v) acetonitrile solution, vortexed for 5 min, and centrifuged at 14,000×g for 10 min at 4 ℃. The supernatants were subjected to LC/MS analysis. The M+1/M+2 ratio indicates the ratio of glucose cycled through the oxidative PPP to glucose going directly through glycolysis.

10⁶ cells were transfected with the indicated overexpression vectors or incubated with inhibitors. The culture medium was changed for glucose consumption and lactate production, and the cells were incubated for an additional 24 h. Glucose levels were measured using a Glucose Colorimetric assay kit (Merck, cat. no. MAK181-1KT). Lactate levels were determined using a Lactate Assay Kit (Merck, cat. no. MAK064-1KT). The values were normalized to the total protein amount as determined using a BCA protein assay kit (Beyotime, cat. no. P0012S).

The lentiviral expression vector pLenti-CMV-EGFP-3Flag was purchased from OBIO. Full-length human CDK1, CDK2, CDK5, and G6PD were amplified and inserted into this vector. G6PD-T91A, G6PD-T466A, G6PD-T91D, G6PD-T91E, and CDK5-D144N mutated vectors were generated using a Muta-Direct Site-Directed Mutagenesis Kit (SBS Genetech Co., Ltd., China). The primers used in this study were as follows: T91A forward, 5′-CAAGGCCGCCCCAGAGGAGAAGCTCAAGCTGG-3′; T91A reverse, 5′-CCTCTGGGGCGGCCTTGAAGAAGGGCTCACT-3′; T466A forward, 5′-TATTTTCGCCCCACTGCTGCACCAGATTGAG-3′; T91A reverse, 5′- GCAGTGGGGCGAAAATACGCCAGGCCTCAC-3′. T91D forward, 5′-CAAGGCCGACCCAGAGGAGAAGCTCAAGCTGG-3′, T91D reverse, 5′-CCTCTGGGTCGGCCTTGAAGAAGGGCTCACT-3′; T91E forward, 5′-CAAGGCCGAACCAGAGGAGAAGCTCAAGCTGG-3′, T91E reverse, 5′-CCTCTGGTTCGGCCTTGAAGAAGGGCTCACT-3′; D144N forward, 5′-GGAGCTGAAATTGGCTAATTTTGGCCTGGCTCG-3′; D144N reverse, 5′-CGAGCCAGGCCAAAATTAGCCAATTTCAGCTCC-3′. All transfection experiments were conducted with Lipofectamine 2000 (Thermo Fisher Scientific, cat. no. L3000015) as recommended by the manufacturer. MDA-MB-231 cells were transfected with lentiviruses containing CDK5, G6PD-WT, G6PD-T91A, G6PD-T91D, G6PD-T91E and G6PD-T466A vectors, and stable cell lines were obtained after treatment with 2 mg/mL puromycin for 7 days.

Online sgRNA Designer were used (http://chopchop.cbu.uib.no/). To target the common exons for all human CDK5 isoforms were synthesized as follows: 5′-GGCCTTGAACACAGTTCCGT-3′; To target the common exons for all human G6PD isoforms were synthesized as follows: 5′-GTGTGTATCCGACTGATGGA-3′. All SgRNAs were cloned to lentiCRISPRv2 (one vector system) plasmids from Zhang lab. Scramble sequence: 5′-GCGTGACAGCTCGCTTCATA-3′.

Protein was extracted with cell lysis buffer for Western blotting and IP (Beyotime, cat. no. P0013). Briefly, scraped cells were collected after centrifugation at 1000×g for 3 min. The pelleted cells were lysed in cell lysis buffer containing proteinase and phosphatase inhibitors (MCE, cat. no. HY-K0010, cat. no. HY-K0021, respectively) for 15 min on ice. The supernatant was transferred to a new tube, and the protein concentration was measured using a BCA Protein Assay Kit (Beyotime, cat. no. P0012S). SDS-PAGE and immunoblotting were performed as described previously⁵³. The following antibodies were used: G6PD (Proteintech, 25413-1-AP); PARP (Cell Signaling Technology, cat. no. #9532); active caspase-3 (Cell Signaling Technology, cat. no. #9662); CDK1 (Proteintech, cat. no. 19532-1-AP); CDK2 (Proteintech, cat. no. 22060-1-AP); CDK5R1 (Proteintech, cat. no. 67503-1-lg); CDK5 (Proteintech, cat. no. 10430-1-AP); γH2AX (Proteintech, cat. no. 10856-1-AP); HA tag (Proteintech, cat. no. 66006-2-lg); Flag tag (Proteintech, cat. no. 66008-3-lg); phospho-threonine-proline mouse mAb (Cell Signaling Technology, cat. no. #9391), 6PGD (Proteintech, cat. no. 14718-1-AP), PGLS (Proteintech, cat. no. 30560-1-AP), β-actin (Kangchen, cat. no. KC-5A08), and GAPDH (Kangchen, cat. no. KC-5G4).

Human CDK5/p35 complex was purchased from Promega (cat. no. V3271), and human G6PD recombinant protein was purchased from Genetex (cat. no. GTX66999-pro). An in vitro kinase assay was performed as described before²⁸. Briefly, the in vitro kinase assay was conducted in a 1 × kinase buffer (Cell Signaling Technology, cat. no. 9820S) supplemented with 1 μmol/L CaCl₂ and 10 μmol/L ATP (Cell Signaling Technology, cat. no. 9804S). In each reaction tube, 50 ng G6PD was incubated with 10 or 50 ng CDK5/P35 proteins with or without adding GFB-12811 (10 μmol/L) or G6PDi-1 (10 μmol/L) as indicated at 30 ℃ for 2 h. The tube maintained on ice was used as a negative control. Histone H1 (Abcam, cat. no. ab198676) was loaded as a positive control. The reaction products were subjected to Western blotting for further identification.

The phosphorylated peptide identification was performed as described before²⁸. Briefly, the phosphorylated G6PD derived from the in vitro kinase assay was digested by Lys-C (1:100, w/w) at 37 ℃ for 3 h followed by trypsin (1:50, w/w) overnight at 37 ℃. The resulting peptides were dissolved in 0.2% formic acid (FA) separated using an online Nano-LC system (Microtech Scientific) and equipped with a C18 reversed-phase column. Then, they were subjected to an LTQ-Orbitrap mass spectrometer (ThermoFisher). Full-scan spectra were acquired at a target value of 4 × 10⁵ with a resolution of 60,000. Data were acquired in a data-dependent acquisition mode using the top-speed method (3 s). The peptides were isolated using a quadrupole system (the isolation window was 0.7). Raw files were analyzed via MaxQuant software. Peak lists were searched against the UniProt protein sequence database.

DSS crosslinking assay was used to detect the G6PD dimer as described before⁵⁴. Briefly, His-tag fused WT G6PD or T91A mutated G6PD expression vectors were transfected into MDA-MB-231 cells and were cultured under glucose deprivation to induce CDK5-mediated phosphorylation of G6PD. Cells were trypsinized and washed twice by cold PBS. Then, cells were suspended in conjugation buffer (20 μmol/L HEPES, pH = 8.0) with DSS (a final concentration of 1 μmol/L). After incubation at 37 ℃ for 30 min, the total protein was extracted from the cells and subjected to Western blotting for detection of G6PD monomer and dimer.

A phosphorylated peptide fragment of G6PD (SEPFFKA(pTHR)PEEKLKLEDFFC) was synthesized and fused with KLH by GenScript. The purified fusion peptide was used to immunize female BALB/c mice according to the conventional procedure. The serum containing anti-pho-G6PD at Thr-91 was purified using Protein Gaffinity chromatography. Specificity and titration of the antibody were assessed by Western blotting and IHC in cells and tissues with different levels of pho-G6PD.

Total RNA was isolated from cells or tissues by TRIzol Reagent (Invitrogen, cat. no. 15596018), and 1 mg of RNA from each sample was reverse transcribed to generate cDNA using a Prime Script RT Master Mix Kit (Takara, cat. no. RR036Q). Next, 1 mL (50 ng) cDNA was used as a template for qPCR with AceQ Universal SYBR qPCR Master Mix (Vazyme, cat. no. Q511-02) according to the manufacturer's instructions. The primers used in our study are listed in the Supporting Information Table S1.

Tissue microarray sections (Human breast cancer tissue microarray, HBreD090CS01) were purchased from Shanghai Outdo Biotech Co., Ltd. (National Human Genetic Resources Sharing Service Platform, 2005DKA21300). This study was approved by the Institutional Review Board at Nanjing University. Tissue sections were immunostained with the antibody generated in this study for the detection of CDK5 expression and G6PD phosphorylation.

Immunohistochemistry experiments were performed as previously described⁵⁵. Formalin-fixed, paraffin-embedded tumor tissue sections were prepared according to the manufacturer's instructions. Antigens were retrieved by boiling the slides in 0.01 mol/L sodium citrate (pH 6.0) in a microwave for 15 min, after which the tissue sections were washed three times in 1 × PBS/0.1% Tween-20 (PBST) solution. Consecutively, the slides were incubated for 1 h at room temperature in blocking solution (5% BSA in 1 × PBST), followed by overnight incubation at 4 ℃ in blocking solution in which primary antibodies were diluted (1:400). The sections were then washed three times using 1 × PBST solution and incubated at room temperature for 1–2 h in blocking solution in which HRP conjugated secondary antibodies were diluted.

A semiquantitative Histoscore was calculated for each specimen by multiplying the distribution area (0–100%) at each staining intensity level by the intensity (0, negative; 1, weak staining; 2, moderate staining; and 3, strong staining) as previously reported⁵⁶.

To detect cell proliferation, 5 × 10³ transgenic MDA-MB-231 cells (overexpressing CDK5, G6PD-WT, and G6PD-T91A) were seeded in a volume of 100 mL medium per well on a 96-well plate. After culturing for the indicated period, cell proliferation was measured using a Cell Counting Kit 8 (CCK-8; Beyotime, cat. no. C0037) according to the manufacturer's instructions. For the measurements of cell viability under treatment with H₂O₂ or anticancer agents, cells were seeded in 96-well plates (2 × 10⁴ cells/well) in the presence or absence of the indicated treatments. Cell viability was evaluated using the CCK-8 kit mentioned above. The optical density (OD) value was read at 450 nm. Fold changes were calculated using the baseline values of untreated cells as a reference (set to 100%). For the measurement of apoptosis induced by H₂O₂ and other treatments, 2 × 10⁶ MDA-MB-231 cells were seeded in 6-well plates and cultured under normal conditions or in glucose deprivation medium. After treatment with H₂O₂ or anticancer agents, the cells were harvested and stained with an Annexin V-FITC/PI Apoptosis Detection Kit (Vazyme, cat. no. A211-01), followed by flow cytometry measurement of fluorescence.

Apoptosis was detected in tumor tissues by TUNEL assay. Tissue sections were incubated with TdT/dUTP at 37 ℃ for 30 min, washed, and incubated at room temperature with HRP-conjugated anti-Digoxigenin for 30 min. Representative images from each stained section were captured using an Olympus BX43 microscope.

G6PD activity was assayed using the G6PD Activity Assay Kit with WST-8 (Beyotime, cat. no. S0189) according to the manufacturer's instructions. Briefly, cells were harvested and lysed after indicated treatments using G6PD Extracting Solution (Beyotime, cat. no. S0189-5). The lysates were centrifuged (12,000×g, 15 min, 4 ℃) and subjected to BCA assay for measuring protein concentration. The G6PD activity was measured from the obtained supernatants in the reaction mix containing (final concentration) 50 μmol/L KPi (potassium phosphate buffer, pH = 7.5), 0.5 μmol/L EDTA, 5 μmol/L MgCl₂, 2 μmol/L G6P, 0.2 μmol/L NADP⁺ and 10 mg of sample protein. The formed NADPH rapidly reduced WST-8 to produce a formazan product, which could be determined by monitoring absorbance in the 430–550 nm range. The G6PD activity was calculated according to the standard curve and normalized to protein concentration.

6PGD activity was evaluated by using 6PGD Activity Colorimetric Assay Kit (Solarbio, cat. no. BC2100, China). In this assay, the conversion of 6-phosphogluconate from 6PGD generated NADPH, which subsequently reduced a colorless probe to a colored product detectable by absorbance at 450 nm.

The total intracellular ROS level was determined as described previously²⁵. Briefly, cells were incubated with 10 μmol/L 2′,7′-dichlorodihydrofluorescein diacetate (H₂-DCFH-DA, Thermo Fisher Scientific, cat. no. D399) at 37 ℃ for 30 min. Next, the cells were harvested, washed twice with ice-cold PBS, resuspended in PBS, and subjected to flow cytometry. Intracellular NADPH and NADP⁺ levels were determined using a NADP⁺/NADPH Assay Kit with WST-8 (Beyotime, cat. no. S0179), and the GSH/GSSG ratio was determined using a GSH/GSSG Assay Kit (Beyotime, cat. no. S0053).

Anoikis was induced by plating cells (10⁶ cells/well) on ultralow attachment 6-well plates (Corning, cat. no. 3471). For the soft agar colony formation assay, the cell suspension was mixed with 0.7% soft agar in 2 × DMEM containing 20% FBS in an equal volume and layered in triplicate onto 1.4% solidified agar in 2 × DMEM containing 20% FBS. After 14 days of cell culture, the colonies were counted under a microscope and photographed.

For the in vivo tumorigenesis experiments, all female nude mice (6–7 weeks old) were purchased from the Model Animal Research Center of Nanjing University (Nanjing, China). 1 × 10⁷ cells in 200 mL PBS with 30% Matrigel (BD, 356234) were subcutaneously injected into mice, and mice were randomly allocated to the groups. When tumor volume reached ∼100 mm³, mice were treated with indicated treatments. Tumor volumes and body weights were recorded. The animal care and experimental procedures of the animal studies were approved by the Laboratory Animal Welfare and Ethics Committee of Nanjing University and carried out at Nanjing University (Accreditation No. IACUC-2007015). All animal experiments conformed to the guidelines of the Animal Care and Use Committee of Nanjing University. All efforts to minimize suffering were made. Subcutaneous administration was performed as reported previously²⁵.

Cells were stably transfected with the firefly luciferase gene for the lung metastasis model. 5 × 10⁶ cells were resuspended in 200 mL of sterile PBS and subsequently injected into the tail veins of nude mice. Lung colonization was monitored at the indicated time point after intraperitoneal injection of D-luciferin (MCE, cat. no. HY-12591A) with a Lumina XR IVIS bioluminescent imaging system. After 28 days, mice were sacrificed via cervical dislocation, and the lungs were collected and paraffin-embedded to allow histopathological examination of metastatic nodules.

All statistical analyses were performed by GraphPad Prism 5 (GraphPad). Differences between the two groups were assessed by Student's t-test. Multiple comparisons between two populations were conducted by multiple t-test with type 1 error correction. Differences among multiple groups were assessed by one- or two-way ANOVA. Differences between tumor growth curves were determined by repeated measures of two-way ANOVA. The significance threshold was P < 0.05. For all figures, ∗P < 0.05, ∗∗P < 0.01, ∗∗∗P < 0.001. All data are shown as mean ± standard deviation (SD) unless noted in the figure legend. Generally, all in vitro experiments were carried out with n ≥ 3 biological replicates.

Our previous GSEA analysis (GEO Accession number GSE132902)¹⁶ indicated that CDK5 potentially promoted PPP in BC cells (Supporting Information Fig. S1A). To decipher the role of CDK5 in glucose metabolism, we analyzed whether CDK5 affected glucose consumption and lactate production in BC cell line MDA-MB-231. To this end, we treated MDA-MB-231 cells with either CDK5 inhibitor Roscovitine (Rosc) a commercial CDK5 inhibitor, or GFB-12811 a selective CDK5 inhibitor¹⁷. These treatments were used to suppress CDK5 activity in the cells. In MDA-MB-231 cells, our results showed that CDK5 inhibition significantly increased lactate production compared to that of the DMSO group (Fig. 1A, right). In contrast to CDK5 overexpression (Fig. S1B), which substantially decreased lactate secretion (Fig. 1B, right), both treatments had little effect on glucose uptake (Fig. 1A and B). Interestingly, overexpression of CDK1 or CDK2, the two other potential targets of Rosc, had no significant effects on glucose uptake and lactate secretion in MDA-MB-231 cells (Fig. S1C and S1D). Furthermore, CDK5 knockout via CRISPR/CAS9 gene editing in MDA-MB-231 cells (Fig. S1B) induced the same regulation of glucose metabolism as compared with CDK5 inhibitors (Fig. S1E). Similar results were observed in another BC cell line, MCF-7 (Fig. S1F). These results indicated that CDK5 had functional significance in regulating glucose metabolism in BC cells. To better understand how CDK5 affected glucose metabolism, we identified critical factors involved in glucose metabolism that could be potentially regulated by CDK5. As shown in Fig. S1G, CDK5 inhibition conspicuously affected the expression levels of multiple enzymes, TP53-induced glycolysis regulatory phosphatase (TIGAR) was the most affected. It is plausible that compensatory TIGAR expression aims to protect tumor cells by reducing glycolytic rates and diverting glycolytic intermediates into the PPP¹⁸, which was suppressed by CDK5 inhibition. Moreover, increased glucose flux through the oxidative PPP has been shown to decrease lactate excretion¹⁹. Thus, we speculated that CDK5 would bias glucose metabolic flux towards the PPP.

To determine whether CDK5 activation shifted BC cell glucose metabolism to the PPP, we performed isotopomer spectral analysis (ISA) in MDA-MB-231 cells using liquid chromatography/mass spectrometry (LC/MS) to trace ¹³C labels from 1,2-¹³C₂ glucose (Fig. 1C), as previously described²⁰. The major intermediate metabolites of glycolysis and PPP were subsequently analyzed to determine the fate of glucose carbon (Fig. 1D). In the PPP, the C1 carbon of incoming glucose is released as CO₂ during the conversion of 6-phosphogluconate to ribulose-5-phosphate (Ru5P), indicating that metabolites in the PPP carry one ¹³C-labelled carbon. In contrast, glycolytic intermediates and products carry two labels (Fig. 1C)²¹. According to our accurate quantification of isotopologue distribution in BC cells with CDK5 overexpression, we found that ¹³C-enrichment of the G6P [M+2] isotopologue was markedly downregulated in comparison with that of the control group. In contrast, the fructose-6-phosphate (F6P) [M+2] isotopologue and fructose-1,6-bisphosphate (FBP) [M+2] isotopologue showed minimal changes (Fig. 1E and F, Fig. S1H). However, we observed no significant alterations in the ¹³C-enrichment of the G6P [M+2] isotopologue in BC cells with CDK5 inhibition via either Rosc or GFB-12811 (Fig. 1E). This could be attributed to the low expression level of CDK5 in BC cells under normal conditions (Fig. S1B). Considering that G6P is metabolized by both glycolysis and PPP, we inferred that the observed reduction in the G6P [M+2] isotopologue indicated possible utilization by the PPP. As expected, CDK5 overexpression strongly promoted the enrichment of Ru5P [M+1] isotopologue derived from 1,2-¹³C₂ glucose (Fig. 1G), which is the end product of the oxidative branch of the PPP⁵. In contrast, inhibition of CDK5 reduced the abundance of the Ru5P [M+1] isotopologue in comparison with that of the control group (Fig. 1G). These results show that CDK5 enhanced the activity of the PPP oxidative branch, which competitively metabolized G6P and subsequently produced Ru5P.

To explore how CDK5 promoted the activity of the PPP, the expression and catalytic activity levels of three rate-limiting enzymes in the PPP were examined, including G6PD, 6-phosphogluconate dehydrogenase (6PGD), and 6-phosphogluconolactonase (PGLS)³. Of note, CDK5 inhibition did not affect the expression of these genes in MDA-MB-231 cells (Supporting Information Fig. S2A and S2B). In contrast, it significantly reduced G6PD catalytic activity (Fig. 2A), without affecting 6PGD (Fig. S2C). To serve as a positive control, a G6PD-selective inhibitor, G6PDi-1, was employed (Fig. 2A). To further substantiate the role of CDK5 in G6PD activation, we aimed to determine whether CDK5 interruption disrupted redox balance. To accomplish this, redox status was assessed by measuring the antioxidant capacity of tumor cells. As shown in Fig. 2B, the presence of CDK5 inhibitors led to a significant decrease in NADPH levels and a corresponding increase in NADP⁺ levels. Similar results were observed in the G6PDi-1-treated group (Fig. 2A and B). Insufficient NADPH production limited the regeneration of GSH from its oxidized form, GSSG (Fig. 2C, Fig. S2D). Moreover, consistent with the trends shown by NADPH and GSH, the intracellular ROS level was elevated after CDK5 inhibition or G6PD inhibition (Fig. 2D). These results indicate that CDK5 inhibition was of great potential to break the redox balance in BC cells, potentially by modulating G6PD activity.

To our knowledge, dysregulation of redox balance induces intracellular ROS accumulation, which is harmful to rapidly proliferative cancer cells²². Thus, cancer cells adaptively execute several alternative pathways to overcome oxidative stress, such as mTOR, NRF2, and p53²³. In neurodegenerative diseases, CDK5 was previously found to be hyperactivated under oxidative stress²⁴, indicating that CDK5 might be a potential regulator in intracellular redox homeostasis. Indeed, in this study, we observed that CDK5 inhibition significantly enhanced intracellular ROS production (Fig. 2E) and promoted the susceptibility of BC cells to H₂O₂ treatment, resulting in remarkable apoptosis (Fig. 2F–I). In support of this finding, Western blotting revealed that exposure to H₂O₂ caused typical PARP1 proteolytic cleavage and elevated the expression of activated caspase-3 in BC cells after CDK5 inhibition compared to the control group (Fig. 2J). Moreover, 10 μmol/L H₂O₂, a nonlethal dose for MDA-MB-231 cells, significantly suppressed cell viability when combined with CDK5 inhibition or G6PD suppression (Fig. 2H and I). The forced expression of G6PD (Fig. S2E) or administration of the antioxidant N-acetyl-L-cysteine (NAC) significantly decreased intracellular ROS levels (Fig. S2F), maintained redox balance (Fig. S2G) and completely restored cell viability (Fig. 2K, Fig. S2H and S2I). More importantly, we noticed that after H₂O₂ administration in MDA-MB-231 cells, CDK5 expression was elevated in an H₂O₂ dose-dependent manner (Fig. 2J). In contrast, the G6PD protein level remained unchanged, which strengthened the hypothesis that CDK5 might be a potential regulator in intracellular redox homeostasis. Indeed, CDK5 knockout attenuated the increase in G6PD activity, as well as NADPH production in MDA-MB-231 cells with incubation of 10 μmol/L H₂O₂ (Fig. S2J and S2K), resulting in high ROS production (Fig. S2L). Similar results were also observed in MCF-7 cells (Supporting Information Fig. S3A). Collectively, these data illustrate that CDK5 promoted G6PD activation and was crucial for the maintenance of intracellular redox homeostasis in BC cells.

Like H₂O₂ treatment, glucose deficiency elicits oxidative stress and induces apoptosis of tumor cells²⁵. As described above, the link between CDK5 and redox homeostasis suggested that BC cells could be vulnerable to CDK5 inhibition under glucose deprivation. To test this hypothesis, we developed and established a cellular model of glucose deficiency by culturing cancer cells on a glucose-deficient medium. Consistent with a previous report²⁶, glucose deprivation significantly elevated the intracellular ROS level of MDA-MB-231 cells and induced potent apoptosis (Fig. 3A–D). Then, we examined the effect of CDK5 inhibition during glucose deprivation by treating MDA-MB-231 cells with Rosc and GFB-12811, respectively. As shown in Fig. 3E, G6PD activity significantly increased in MDA-MB-231 cells under glucose deprivation compared with that in a standard medium. In contrast, CDK5 inhibition greatly attenuated the increase of G6PD activity induced by glucose deprivation (Fig. 3E), which promoted ROS enrichment (Fig. 3D) and facilitated apoptosis (Fig. 3B and C). Correspondingly, NADPH and reduced GSH production were also significantly decreased after interruption of CDK5 activity, which was also observed in G6PDi-1-treated cells (Fig. 3F and G, Fig. S3B and S3C).

To further validate the findings described above, we evaluated the sensitivity of MDA-MB-231 cells to glucose deprivation by using a conditioned medium with varying concentrations of glucose. MDA-MB-231 cells were cultured in the conditioned medium as indicated (Fig. 3H). As shown in Fig. 3H, complete glucose deprivation did not alter MDA-MB-231 cell survival within 6 h. However, CDK5 inhibition (red line and green line) induced cell death when the concentration of glucose was 200 mg/L or lower (Fig. 3H). Notably, the reduction in cell viability following Rosc or GFB-12811 treatment under glucose deprivation was remarkably ameliorated by G6PD overexpression (Fig. 3H). Interestingly, during glucose deprivation, intracellular CDK5 expression (Fig. 3I) and G6PD activity (Fig. 3E) were simultaneously elevated. Moreover, CDK5 inhibition or G6PD suppression significantly sensitized BC cells to oxidative stress induced by glucose deprivation. Furthermore, transient expression of CDK5 also protected MDA-MB-231 cells against glucose deprivation-induced apoptosis (Fig. 3J and K). Similar results were observed in MCF-7 cells (Fig. S3D–S3F). Collectively, these findings illustrate that CDK5-induced G6PD activation influenced redox homeostasis and enabled BC cells to survive during glucose deprivation.

Next, we endeavored to identify the mechanism by which CDK5 facilitated G6PD activation. As mentioned above, G6PD expression remained unchanged at both the transcriptional and translational levels after CDK5 interruption (Fig. S2A–S2D, Fig. 2J). Thus, we hypothesized that CDK5 might induce post-translational modification (PTM) of G6PD, which could be able to modulate its catalytic activity²⁷. As expected, the phosphorylation level of G6PD was tightly influenced by CDK5 (Fig. 4A). These data illustrated the essential role of CDK5 in regulating G6PD phosphorylation. To provide further validation, we performed an in vitro kinase assay as previously reported²⁸. As shown in Fig. 4B, G6PD phosphorylation was elevated in the presence of CDK5/P35 complex, notably abrogated by the administration of CDK5 inhibitors or G6PD inhibitors. Moreover, silencing CDK5 kinase activity via D144N mutation (a catalytically inactive mutant)²⁹ failed to mediate G6PD phosphorylation (Supporting Information Fig. S4). Next, we determined to validate the amino acid residue that was potentially phosphorylated by CDK5. To our knowledge, CDK5 is known to be a proline-directed protein kinase that mediates phosphorylation on serine or threonine residues in substrates, specifically within the consensus motif S/TPXH/K/R¹³. By using a group-based prediction system³⁰, a comprehensive tool for predicting phosphorylation sites, two threonine residues followed by proline, Thr-91, and Thr-466, were identified as potential phosphorylation targets (Fig. 4C). For further exploration, threonine residues Thr-91 and Thr-466 were mutated to Ala to generate the T91A and T466A mutations, respectively, transfected them into MDA-MB-231 cells. The exogenous wild-type G6PD protein or mutated G6PD protein was isolated via co-immunoprecipitation and was furtherly analyzed by immunoblotting with P-Thr-Pro-101 mAb (CST, #9391), a mouse monoclonal antibody explicitly targeting phospho-threonine, not serine, only when followed by proline³¹. As a result, the T91A mutation completely abolished CDK5-mediated phosphorylation of G6PD. In contrast, the T466A mutation had little effect (Fig. 4D). For further validation, we performed an in vitro kinase assay coupled with HPLC–MS/MS analysis as described before²⁸. As shown in Supporting Information Fig. S5, G6PD was exclusively phosphorylated at the T91 site. Although T466 has been reported to be phosphorylated by ASKa in Arabidopsis³², it might have little contribution to CDK5-mediated G6PD phosphorylation in BC cells. Moreover, overexpression of CDK5 failed to enhance G6PD phosphorylation after T91A mutation (Fig. 4D). These results suggest that Thr-91 is crucial for CDK5-induced phosphorylation of G6PD. To further substantiate the findings described above, we generated a mouse polyclonal antibody against a recombinant peptide fragment of G6PD with phosphorylation at Thr-91 (98.8% protein identify, the peptide was synthesized by GenScript). The polyclonal antibody was evaluated through ELISA and immunoblotting (Fig. 4E, Supporting Information Fig. S6A and S6B). We observed that wide-type G6PD and mutated G6PD-T91E could be stained using the generated antibody, indicating that G6PD-T91E mutation exhibited the same conformation compared with phosphorylated G6PD-WT as described before³³. By employing the generated polyclonal antibody, we found that phosphorylation of G6PD at Thr-91 was promoted after CDK5 overexpression and suppressed by Rosc treatment (Fig. 4E). These results show that Thr-91 is a potential target site for CDK5-mediated G6PD phosphorylation.

Next, we assessed the significance of the PTM described above in G6PD activation. To accomplish this, we generated MDA-MB-231 cells stably overexpressing either wildtype or mutated G6PD, as indicated in Fig. 4D. Transgenic cells were subsequently incubated under glucose deprivation. As expected, the enzymatic colorimetric assay showed that glucose deprivation significantly elevated G6PD activity (Fig. 4F). Conversely, genetic inhibition by T91A mutation dramatically decreased G6PD activity compared to that of WT G6PD. In contrast, the T466A mutation had little effect (Fig. 4F). More importantly, silencing G6PD phosphorylation by T91A mutation had a minimal impact on the basal activity of G6PD, compared with that of WT G6PD in a normal medium (Fig. 4F), suggesting that CDK5-mediated T91 phosphorylation played a partial modulatory role in G6PD activity. Furthermore, in agreement with the inhibitory effect of the T91A mutation on G6PD activity, NADPH and GSH production decreased by this mutation (Fig. 4G and H, Supporting Information Fig. S7A). As expected, antioxidant deficiency following the G6PD T91A mutation led to profound ROS elevation in the presence of oxidative stress (Fig. 4I, Fig. S7B and S7C), which resulted in mitochondrial damage (Fig. S7D) and subsequent apoptosis (Fig. 4J). In contrast, phosphorylation at Thr-91 significantly enhanced the tolerance of MDA-MB-231 cells to oxidative stress (Fig. S7C). Considering endogenous G6PD expression, which might interfere with the G6PD overexpression cell model, we genetically knockout endogenous G6PD in the MDA-MB-231 cells using CRISPR/Cas9 gene-editing tool (Fig. 4K). Subsequently, WT or mutated G6PD mimicking the persistent dephosphorylated (G6PD-T91A) or phosphorylated status (G6PD-T91D or G6PD-T91E)³⁴ were transfected into G6PD-KO cells (Fig. 4K), to establish transgenic MDA-MB-231 cell lines with G6PD T91 site-mutation. As expected, genetically, G6PD-T91A mutation significantly decreased G6PD activity compared with G6PD-WT and G6PD-T91E, rather than G6PD-T91D, which enhanced G6PD activity (Fig. 4K). More importantly, G6PD with T91A mutation still obtained activity at a basal level (Fig. 4K). These data validated a modulatory role of CDK5-induced phosphorylation of G6PD at the T91 site in BC cells during oxidative stress. Since active G6PD exists in a dimer-tetramer equilibrium but not as a monomer³⁵, we determined whether CDK5-mediated phosphorylation affected G6PD assembly. As shown in Fig. 4L, CDK5 inhibition and G6PD suppression remarkably promoted the dissociation of active G6PD dimers into inactive monomers, which was blocked by T91E mutation (Fig. 4M). Interestingly, G6PD-T91E, rather than G6PD-T91D, maintained G6PD active dimeric form, even in the presence of CDK5 inhibitor and G6PD inhibitor. Generally, these findings imply that CDK5-mediated phosphorylation at Thr-91 of G6PD enhanced its catalytic activity, possibly by stabilizing the G6PD homodimer.

These results reveal that Thr-91 was the target for CDK5-induced G6PD phosphorylation, which was pivotal for G6PD inducible activation and redox homeostasis under oxidative stress.

Next, we investigated the influence of G6PD phosphorylation at the Thr-91 site on BC development in vivo. To accomplish this, we established a xenograft mouse model by implanting MDA-MB-231 cells that overexpressed either WT G6PD (WT-OE) or T91A-mutated G6PD (T91A-OE) in opposite flanks of the same mouse as shown in Fig. 5A. Although the growth rate of T91A-OE cells was unaltered in comparison with that of WT-OE cells in vitro (Supporting Information Fig. S8A), T91A mutation significantly suppressed tumor growth in vivo (Fig. 5B and C). These results indicate that interruption of G6PD phosphorylation exhibited potential anticancer activity.

Similar to glucose deficiency, matrix detachment elicits oxidative stress²⁵, evidenced by elevated ROS production (Fig. S8B), which results in anoikis³⁶. Developing anoikis resistance is critical for tumor cells to survive in circulation and form metastases³⁷. In vitro assays revealed that CDK5-mediated G6PD phosphorylation induced colony formation in soft agar (Fig. S8C) due partly to elevated ROS detoxification capacity (Fig. S8D). Moreover, apoptotic assays confirmed that phosphorylated G6PD at the Thr-91 site promoted resistance to anoikis (Fig. S8E). For in vivo assay, WT-OE or T91A-OE cells were injected into the tail vein of nude mice (Fig. 5D), as described²⁵. Pulmonary metastasis was assessed by in vivo quantitative bioluminescence imaging. As shown in Fig. 5D, WT-OE cells formed remarkably excessive lung metastases in comparison with T91A-OE cells. Hematoxylin and eosin staining of dissected lungs demonstrated the presence of more metastasis nodules in the WT-OE group in comparison with the T91A-OE group (Fig. 5E). Similarly, the same results were observed in tumor-bearing mice formed by transgenic MDA-MB-231 cell lines with G6PD T91 site-mutation (G6PD-WT, G6PD-T91A, G6PD-T91D and G6PD-T91E) (Fig. 5F and G, Fig. S8F and S8G). Altogether, our results demonstrate that CDK5-mediated G6PD phosphorylation at Thr-91 was critical for BC progression and metastasis.

Next, we turned to human BC specimens to address the importance of our findings in clinical samples. We generated a tissue microarray (TMA) composed of 45 paired adjacent-normal breast tissues and tumor tissues at various stages (IA: 6 cases; IIA: 15 cases; IIB: 9 cases; IIIA: 7 cases; IIIC: 8 cases). Next, the TMA was subjected to IHC by applying polyclonal antibodies to detect G6PD phosphorylation and applying polyclonal antibodies to detect G6PD phosphorylation and CDK5 expression. The intensity of each sample was quantified in a blind manner based on Histoscore. As shown in Fig. 6A, the CDK5 and phosphorylated G6PD levels in cancer tissues were much higher than in adjacent noncancerous breast tissues (Fig. 6A and B). CDK5 expression was tightly correlated with G6PD phosphorylation in clinical specimens (Fig. 6C), strengthening the conclusion that CDK5-mediated G6PD phosphorylation at the T91 site. Moreover, CDK5 and the G6PD phosphorylation levels were also elevated in advanced-stage tumor tissues (Fig. 6D and E). G6PD phosphorylation was dramatically evoked in tumors with large size (T2 and T3, Fig. 6F), primary tumors with lymph node metastasis (N1, N2, and N3, Fig. 6G), as well as metastatic tumors (Fig. 6H), which was consistent with the role of G6PD phosphorylation in limiting ROS and promoting the survival and metastasis of tumor cells during tumor development, as demonstrated in our mouse model (Fig. 5). These results reveal that G6PD phosphorylation at Thr-91 site occurs primarily in tumor tissues, rather than in normal tissues and it is closely associated with human BC development. Interestingly, we observed that CDK5-induced G6PD phosphorylation widely occurred in BC cell lines (Supporting Information Fig. S9). These data highlight CDK5-induced G6PD phosphorylation as a potential and promising target for BC therapy.

Given that oxidative stress defense programs are known to mediate therapy resistance, as many anticancer treatments cause cellular redox perturbation to induce cell death³⁸, we speculated that inhibition of CDK5-mediated G6PD phosphorylation might be an effective adjuvant therapy to enhance sensitization of cancer cells to ROS-inducing anticancer therapy. We applied several ROS-based therapeutic agents for further analysis to test this hypothesis. These included 5-fluorouracil (5-Fu)³⁹, paclitaxel⁴⁰, 5-azacytosine⁴¹, olaparib⁴², and berberine⁴³. As shown in Fig. 7A, the intracellular ROS level was increased in MDA-MB-231 cells after treatment with these anticancer agents. Meanwhile, CDK5-mediated phosphorylation of G6PD at Thr-91 was induced in response to these anticancer treatments, particularly Olaparib treatment (Supporting Information Fig. S10A). These results prompted us to focus our subsequent in vivo analyses on Olaparib. Furthermore, T91A mutation significantly increased cell death in response to Olaparib and other treatments (Fig. 7B, Fig. S10B–S10D). CDK5-D144N (dead mutation) also sensitized BC cells to therapeutic agents (Fig. S10E). These observations indicated that CDK5-mediated G6PD phosphorylation was crucial for the adaptation of cancer cells to ROS-based anticancer therapies.

Next, we aimed to investigate the anti-tumoral activity of the combined therapeutic strategy by using CDK5 inhibitors and olaparib. In agreement with the biological effects triggered by T91A mutation observed previously, CDK5 inhibitors sensitized MDA-MB-231 cells to olaparib (Fig. 7C), due partly to enhanced DNA damage (Fig. 7D). As expected, G6PD phosphorylation at Thr-91 was notably abrogated in MDA-MB-231 cells after CDK5 inhibitors when combined with olaparib (Fig. 7D). However, little cytotoxicity was observed in normal HEK293 cells (Fig. S10F and S10G). The main reason might be that CDK5-induced phosphorylation of G6PD was more likely to be employed by tumor cells rather than normal cells (Fig. S9A and S9H).

Finally, we investigated whether inhibiting G6PD phosphorylation enhanced the therapeutic effects of Olaparib on BC in vivo. To this end, a xenograft mouse model was established as mentioned above, and the mice were subsequently injected with Olaparib (10 mg/kg) or CDK5 inhibitors (Rosc, 2 mg/kg; GFB-12811, 5 mg/kg), singly or in combination, on Days 7, 11, 15, 19, 23 and 27, post tumor implantation. As expected, combinatorial use of CDK5 inhibitors with Olaparib strongly inhibited tumor growth (Fig. 7E and F), and mice that received this treatment showed no alteration in body weight (Fig. S10I). For further confirmation, tumor tissues were systemically analyzed by IHC to assess proliferation, apoptosis, and DNA damage. As shown in Fig. 7G, administration of CDK5 inhibitors dramatically enhanced Olaparib-induced DNA damage, characterized by elevated gH2AX expression. Moreover, combinatorial treatments showed significant induction of persistent apoptosis, as demonstrated by active caspase 3 and TUNEL assay (Fig. 7G). Additionally, treatment with CDK5 inhibitors or olaparib, individually or in combination, had a slight inhibitory effect on the proliferation of tumor cells, defined by Ki67 expression (Fig. 7G). The anti-tumoral activity of this combined therapeutic strategy was also observed in MCF-7-bearing mice (Fig. S10J and S10K).

Taken together, these results illustrate that interrupting the CDK5/pho-G6PD axis synergistically enhanced the therapeutic effect of olaparib in HR-proficient BC cells. Furthermore, these findings demonstrate that interrupting the CDK5/pho-G6PD axis is a potential promising adjuvant for ROS-based anticancer therapy in BC patients.