KDM5B-driven glucose metabolic reprogramming promotes enzalutamide resistance in prostate cancer via the lactate/hnRNPA1 lactylation/AR-V7 axis

Aims

Resistance to enzalutamide (Enza) in castration-resistant prostate cancer (CRPC) is linked to poor prognosis. While KDM5B is highly expressed in Enza-resistant CRPC, the mechanisms of resistance remain poorly understood.

Methods

We applied an integrated approach to study KDM5B using bioinformatics analyses of single-cell and multi-omics data, along with in vitro and in vivo validation. We explored mechanisms through lactylation proteomics, CRISPR/Cas9 editing, ChIP, and dual-luciferase reporter assays.

Results

KDM5B induces Enza resistance by epigenetically suppressing PTEN, which in turn activates the PI3K/Akt signaling pathway to upregulate PGK1 and drive metabolic reprogramming and lactate production. Lactate acts as a substrate for p300-mediated lactylation of hnRNPA1 at lysine 179 (K179), stabilizing hnRNPA1 by blocking NEDD4L-mediated ubiquitination and promoting AR-V7 splicing. A potential positive feedback loop enhances this effect: KDM5B activates AR, and AR, in turn, increases KDM5B expression. Inhibiting KDM5B or p300 can reverse Enza resistance in vivo.

Conclusions

We identify a mechanism linking metabolism, epigenetics, and a KDM5B/AR feedback loop in drug resistance. These findings suggest that multi-target strategies may represent a promising approach to overcome Enza resistance in CRPC.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12943-026-02602-z.

Highlights

• KDM5B is a histone demethylase that is upregulated in enzalutamide resistant prostate cancer.

• KDM5B drives tumor metabolic reprogramming by activating PI3K/Akt pathway through epigenetic inhibition of PTEN, which in turn transcriptionally up-regulating PGK1.

• Lactate causes lactylation of lysine at position 179 of hnRNPA1, leading to abnormal splicing of AR and upregulation of AR-V7.

• p300/HDAC1/HDAC2 jointly regulate the lactylation process of hnRNPA1.

• The ligand-independent AR signaling pathway positively feedback promotes the upregulation of KDM5B expression.

Graphical Abstract

We demonstrate that KDM5B activates the PI3K/Akt pathway through epigenetic inhibition of PTEN, which in turn transcriptionally up-regulating PGK1, leading to metabolic reprogramming and lactate accumulation. Lactate, as a substrate for p300-mediated lactylation of hnRNPA1-K179, enhances hnRNPA1 stability by inhibiting NEDD4L-mediated ubiquitination, thereby promoting AR-V7 splicing. Furthermore, we identified a positive feedback loop where KDM5B promotes ligand-independent AR activation, while AR conversely transcritionally upregulates KDM5B. Drug inhibition of KDM5B or p300 synergistically reverses enzalutamide resistance in vivo.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12943-026-02602-z.

Introduction

Prostate cancer (PCa) remains one of the leading causes of cancer-related deaths among men worldwide [1, 2]. Its incidence varies significantly across different regions. Androgen-receptor signaling inhibitors (ARSIs), such as enzalutamide (Enza), are the primary treatment for advanced PCa. These drugs work by blocking androgen receptor (AR) signaling to inhibit tumor growth. However, most patients eventually develop castration-resistant prostate cancer (CRPC) [3–5]. Resistance to ARSIs involves complex mechanisms, with one key factor being the development of constitutively active androgen receptor splice variants (AR-Vs) [6]. The most well-known variant, AR-V7, lacks the ligand-binding domain (LBD) that Enza targets. This absence makes AR-V7 continuously active, promoting a pro-tumor transcription program even in the presence of potent ARSIs [7, 8]. While the downstream effects of AR-V7 are well characterized, the upstream regulatory networks that control its splicing are not well understood.

Metabolic reprogramming is a key feature in tumor drug resistance [9, 10]. The Warburg effect describes how cancer cells tend to use anaerobic glycolysis for energy, even in the presence of oxygen [11]. Lactate buildup is associated with chemoresistance in various cancers. Histone H3 lactylation increases the expression of transcription factors YBX1 and YY1, promoting cisplatin resistance in bladder cancer [12]. In glioblastomas, glycolytic reprogramming results in XRCC1 lactylation and chemoresistance, especially in cases overexpressing ALDH1A3 [13]. However, the role of lactate metabolism in ARSI resistance in prostate cancer remains unclear.

Lysine-specific demethylase 5B (KDM5B), or JARID1B, is a crucial epigenetic regulator. It belongs to the JmjC domain-containing histone demethylase family. KDM5B’s primary role is to remove di- and trimethylation from histone H3 lysine 4 (H3K4me2/3). Since this mark activates transcription, KDM5B mainly functions as a repressor. KDM5B is frequently overexpressed in many cancers, including prostate cancer (PCa), and its high levels are strongly associated with tumor progression, metastasis, and resistance to therapy [14, 15]. In PCa, KDM5B plays a role in progression to castration resistance and in modulating the androgen receptor (AR) signaling pathway [16, 17]. However, the precise mechanisms by which KDM5B contributes to Enza resistance, particularly in connection with metabolism and non-canonical roles, remain unclear.

In this study, we combined multi-omics analyses and functional validation in preclinical models and clinical cohorts to identify a new mechanism of Enza resistance. We demonstrate that KDM5B activates the PI3K/Akt pathway by epigenetically inhibiting PTEN. This leads to the upregulation of the glycolytic enzyme phosphoglycerate kinase 1 (PGK1), which promotes metabolic reprogramming and lactate buildup. Increased lactate then serves as a substrate for p300-mediated lactylation of the splicing factor hnRNPA1 at lysine 179. HDAC1/2 can reversibly remove this modification. Lactylated hnRNPA1 (K179la) remains stable because it avoids NEDD4L-mediated ubiquitination and degradation. This modification enhances the production of the active AR-V7 splice variant, a key factor in resistance. We also identified a potential feedback loop where KDM5B-driven Akt signaling encourages AR activation without ligand, and AR subsequently increases KDM5B transcription. This loop helps sustain resistance. Our findings suggest that multi-target interventions targeting the KDM5B-mediated axis may represent a promising strategy to overcome Enza resistance.

Methods

Clinical samples

The Ethics Committee of the First Affiliated Hospital of Chongqing Medical University (Ethics number: 2021 − 608) reviewed and approved the sample collection. All participants provided informed consent, and this study adhered to the Declaration of Helsinki. Clinical cohort 1 (see Additional file 1) included 12 patients with unclassified PCa who underwent radical prostatectomy after diagnosis, with paired cancerous and paracancerous tissue samples collected. Clinical cohort 2 (see Additional file 2) consisted of 8 patients: 3 with enzalutamide-resistant (EnzR) PCa and 5 with enzalutamide-sensitive (EnzS) PCa. Their tumor tissues were obtained via prostate puncture. Clinical cohort 3 involved a patient followed from diagnosis of prostate adenocarcinoma by prostate puncture biopsy in June 2022; the patient chose Enza-based therapy (Leuprolide + Enza) and had regular follow-ups. Imaging indicated disease progression (by RECIST criteria) in May 2024, leading to another prostate puncture for further treatment guidance. Baseline characteristics are detailed in the additional files.

Analysis of single-cell sequencing

Single-cell RNA sequencing (scRNA-seq) data were obtained from the Gene Expression Omnibus (GEO) repository (GSE168668) [18] and processed using the “Seurat” package (v4.4.0) in R for data analysis. During quality control, cells expressing fewer than 300 genes or with mitochondrial gene content exceeding 10% were excluded, resulting in 7,323 cells that met quality thresholds. The scRNA-seq data were normalized using the LogNormalize method, followed by batch effect correction with the R package “Harmony”. Dimensionality reduction was performed through principal component analysis (PCA) and Uniform Manifold Approximation and Projection (UMAP). Cell clustering was subsequently conducted using the “FindNeighbors” and “FindClusters” functions in Seurat. Cell population annotation was based on differential sensitivity to Enza treatment, categorizing cells into EnzS and EnzR groups. To quantify lactylation activity, we implemented the “AUCell” algorithm. Based on lactate-regulated genes (LRGs) [19], individual cells were assigned lactate-generating activity scores and grouped into high- and low-lactate cohorts based on the median score. Their respective activity profiles were calculated using marker genes in different clusters (RH, SH, RL, and SL) via the gene set variation analysis (GSVA) algorithm. “CellChat” was used to infer and analyze ligand-receptor interactions between cell clusters. “Monocle” was employed to explore the differentiation trajectories of EnzR phenotypes in PCa and the expression trajectories of target genes.

Bioinformatics analysis

mRNA expression data and clinical data from the Cancer Genome Atlas (TCGA) dataset were downloaded from the University of California, Santa Cruz (UCSC) database (https://xenabrowser.net/). Expression data were log2(x + 1)- transformed and RSEM-normalized. Normalized gene data for datasets GSE17951 and GSE104935 were obtained from the Gene Expression Omnibus (GEO) database (https://www.ncbi.nlm.nih.gov/geo/). Differentially expressed genes were identified using the “limma” package, with significance thresholds set at |logFC| < 0.5 and P < 0.05. Gene set enrichment analysis (GSEA) was performed to identify enriched pathways with a standard false discovery rate (FDR) < 0.25 and P < 0.05. Gene Ontology (GO) terms and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathways were analyzed using the “clusterProfiler” package. Weighted gene co-expression network analysis (WGCNA) was used to identify gene sets most strongly associated with specific phenotypes. All bioinformatics analyses were performed using R software (version 4.2.1).

Cell lines and cell cultures

RWPE-1 (RRID: SCSP-5025), LNCaP (RRID: SCSP-5021), 22Rv1 (RRID: SCSP-5022), DU145 (RRID: SCSP-5024), VCaP (RRID: SCSP-5034), and PC3 (RRID: SCSP-532) cells were purchased from the cell bank of the Chinese Academy of Sciences (Shanghai, China). C4-2 (CRL-3314) and HEK-293T (CRL_11268) cells were purchased from American Type Culture Collection (ATCC). All cell lines were cultured at 37 °C and 5% CO2 in a humidified atmosphere. Short tandem repeat (STR) DNA analysis was performed periodically on all cells to ensure they remained free of mycoplasma contamination. The detailed method for constructing the EnzR cell lines were as follows.

LNCaP cells were cultured continuously for 6 months in a medium containing 10% charcoal-stripped fetal bovine serum (CS-FBS; Cat. 12676029, Gibco, USA) to establish androgen-independent LNCaP cells (LNCaP-AI), and then EnzR cells were established by continuous culture in media with different concentrations of enzalutamide (1, 2, 5, 10, 20 µmol/L) for 6 months. C4-2 cells were cultured continuously for more than 6 months in media containing 10, 20, and 40 µmol/L enzalutamide to establish EnzR cells. Subsequently, the successfully constructed EnzR cells were continuously cultured in enzalutamide-containing media at 20 µmol/L (LNCaP-EnzR) and 40 µmol/L (C4-2-EnzR) for 6 months until both cell lines could be stably passaged in enzalutamide-containing media. Both the induction process of EnzR cell lines and their subsequent stable maintenance were performed in medium supplemented with 10% CS-FBS. RWPE-1 cells were cultured in RWPE-1 Cell Complete Medium (Cat. ZM0351, Shanghai Zhong Qiao Xin Zhou Biotechnology Co., Ltd). Wild-type LNCaP, C4-2, 22Rv1, DU145, and PC-3 cells were cultured in RPMI 1640 medium (Cat. SH30809.01, Cytiva, China) containing 10% fetal bovine serum (FBS; Cat. 16000044, Thermo Fisher Scientific, USA). LNCaP-AI cells were cultured in RPMI 1640 medium containing 10% CS-FBS. VCaP and HEK-293T cells were cultured in DMEM medium (Cat. SH30243.02, Cytiva, China) supplemented with 10% FBS (Cat. 16000044, Thermo Fisher Scientific, USA). All cell lines were cultured with 1% penicillin-streptomycin (Cat. 15140122, Gibco, USA).

Enzalutamide (Cat. S1250), Apalutamide (Cat. S2840), LY294002 (Cat. S1105), CPI-455 (Cat. S8287), C646 (Cat. S7152), Trichostatin A (Cat. S1045), Nicotinamide (Cat. S1899), Sodium L-lactate (Cat. S6010), and Sodium oxamate (Cat. S6871) were purchased from Selleck Chemicals (Shanghai, China). DHT (Cat. D-073) and IGF-1 (Cat. I1146) were purchased from MilliporeSigma (GER). MG132 (M7449) and CHX (5087390001) were obtained from Sigma (USA). The relevant reagents were dissolved in DMSO (Cat. D8370, Solarbio, Beijing, China) to prepare a 100 mM stock solution, stored at −80 °C, diluted to the desired concentration, and set aside.

Western blotting

Proteins were extracted from the specified cells or tissues using a lysis buffer. Equal amounts of proteins were separated by sodium dodecyl sulfate (SDS) - polyacrylamide gel electrophoresis and transferred onto polyvinylidene fluoride (PVDF) membranes (Cat. 1620177, Bio-Rad, USA). Incubate overnight at 4 °C with primary antibodies. Proteins were detected using horseradish peroxidase-conjugated anti-rabbit (Cat. ab6721, Abcam, USA) or anti-mouse (Cat. ab6728, Abcam, USA) secondary antibodies, and visualized with chemiluminescent reagents from the ECL kit (Cat. 1705060, Bio-Rad, USA). All samples were normalized to β-actin or GAPDH and semi-quantitatively compared to the sham control. The primary antibodies used in this study are listed in Table S1 (Supporting Information).

Immunoprecipitation and ubiquitination assays

For immunoprecipitation (IP) experiments, cells were harvested and lysed on ice for 30 min in cold RIPA buffer with a protease inhibitor cocktail (Roche). Cell lysates were centrifuged at 14,000 g for 15 min at 4 °C to remove debris. The clear supernatants were collected and incubated with the indicated primary antibody or a matching IgG control at 4 °C overnight with gentle rotation, followed by incubation with Protein A/G magnetic beads (Cat. No.: HY-K0202, MCE, China) for an additional 2 h. The beads were then washed five times with wash buffer. Finally, the immunoprecipitated complexes were eluted by boiling in 2× SDS-PAGE loading buffer and analyzed by Western blotting. For Co-IP, to detect the interaction between the target protein and the effector protein, lysates were immunoprecipitated with an anti-target protein antibody, and the resulting blots were probed with an anti-effector protein antibody. For the in vivo ubiquitination assay, to evaluate the ubiquitination of hnRNPA1, cells were first co-transfected with plasmids expressing HA-tagged ubiquitin (HA-Ub) and Flag-tagged hnRNPA1 (Flag-hnRNPA1). Before harvesting, cells were treated with the proteasome inhibitor MG132 (20 µM) for 6 h. For lysis, cells were directly boiled for 10 min in a strongly denaturing buffer (1% SDS, 50 mM Tris-HCl, pH 7.4, 150 mM NaCl, 10 mM NEM), then diluted 10-fold with ice-cold RIPA buffer. Lysates were then immunoprecipitated using an anti-Flag antibody. The resulting immunoblots were probed with an anti-HA antibody to detect polyubiquitin chains and with an anti-Flag antibody to confirm immunoprecipitation efficiency.

Determination of cell viability and IC50 value

Cell proliferation and viability were evaluated using the Cell Counting Kit-8 (CCK-8) (Cat. K1018, APExBIO, USA). For proliferation curves, cells were seeded at 1 × 10³ cells per well in 96-well plates. After 24 h of adherence, the cells were treated with Enza (10 µM). At specified time points (0, 24, 48, and 72 h post-treatment), the CCK-8 reagent was added to each well following the manufacturer’s instructions, then incubated for 1–2 h at 37 °C. The absorbance at 450 nm was measured with a microplate reader. To determine the half-maximal inhibitory concentration (IC50), cells were seeded as described and treated for 72 h with a series of two-fold serial dilutions of Enza. Cell viability was measured again using the CCK-8 assay, and dose-response curves were fitted with a nonlinear regression model in GraphPad Prism (v10.1.2) to calculate IC50 values.

Colony formation

Cells were seeded into 6-well plates at a density of 1 × 10³ cells per well. After 24 h to allow for adherence, the culture medium was replaced with fresh RPMI 1640 containing specific substances for each group. The cells were cultured for 14 days, with the medium being replenished every 3–4 days. At the end of the incubation period, the colonies were washed with PBS, fixed with 4% paraformaldehyde (Beyotime, Cat. P0099) for 15 min, and stained with 0.1% crystal violet solution (Beyotime, Cat. C0121) for 20 min. After washing to remove excess stain, the plates were air-dried and photographed. The number and/or area of the colonies were quantified using ImageJ software.

Immunohistochemistry (IHC)

Detailed IHC was performed as previously described [20]. In brief, human tissues and xenograft tumor tissues were fixed in 4% paraformaldehyde (Cat. P0099, Beyotime, Shanghai, China) and embedded in paraffin for sectioning. The tissue sections underwent antigen retrieval and peroxidase removal. Next, the sections were blocked with 10% goat serum (Cat. C0265, Beyotime, Shanghai, China) and incubated overnight at 4 °C with antibodies against KDM5B (Cat. HPA027179, Sigma-Aldrich, GER). The following day, the slides were incubated with a host-specific HRP-conjugated secondary antibody. Signal detection and nuclear counterstaining were performed using the DAB substrate kit (Cat. SK-4105, Vector Laboratories, USA) and hematoxylin QS (Cat. H-3404-100, Vector Laboratories, USA), respectively. Finally, the stained slides were coverslipped with a mounting solution (Cat. H-5000-60, Vector Laboratories, USA). The IHC score was calculated based on staining intensity and the proportion of stained cells. Staining intensity scores were assigned as follows: 0 = negative, 1 = weak, 2 = moderate, 3 = strong. The proportion of stained cells was estimated as: 0 = null, 1–10% = 1, 11–50% = 2, 51–80% = 3, and > 80% = 4. The overall IHC score was obtained by multiplying the staining intensity score by the stained cell proportion score (range 0–12).

Immunofluorescence (IF) staining

This scheme was carried out as described previously [21]. In brief, tumor tissue sections were fixed with 4% paraformaldehyde (Cat. P0099, Beyotime, Shanghai, China). Permeabilize with 0.5% Triton X-100 (Cat. 9036-19−5, Sigma-Aldrich, USA). Then, the cells were sealed with 1% bovine serum albumin (Cat. ST2254, Beyotime, Shanghai, China) and incubated overnight with primary antibodies [Rabbit monoclonal anti-Akt (Cat. 9272, Cell Signaling Technology, USA), Rabbit monoclonal anti-Phospho-Akt (Cat. 9271, Cell Signaling Technology, USA), Mouse monoclonal anti-Phospho-AR (Cat. sc-377546, Santa Cruz Biotechnology, USA)] at 4 °C. Next, the cells were incubated with DAPI and secondary antibodies (Cat. A_11008 or A-11012, Alexa Fluor 488 or 594, Thermo Fisher Scientific, USA). Images were captured using a confocal microscope (Zeiss LSM980, Germany) and ZEN 2010 software.

Tunel assay

According to the manufacturer’s instructions, the slides were stained with the TUNEL Bright Red Apoptosis Detection Kit (Cat. A113-01, Vazyme Biotech, Nanjing, China). All images were captured using a laser-scanning confocal microscope (Olympus FV3000, Japan) and analyzed in ImageJ to determine the percentage of TUNEL-positive cells.

Measurement of lactate levels

According to the manufacturer’s instructions, lactate levels in serum, cancer cells, and tumor tissues were measured using the L-Lactate Assay Kit (Cat. ab65331, Abcam, UK). Serum from patients was extracted from the supernatant of clotted whole blood by centrifuging at 3000 rpm for 15 min at 4 °C. For serum analysis, add 1.5 µL of patient serum to 48.5 µL of lactate assay buffer in a 96-well plate. Mix by pipetting an equal volume (50 µL) of the reaction mixture, then incubate at room temperature for 30 min. Measure the absorbance at 450 nm. For cancer cell and tumor tissue samples, homogenize the sample in four times the volume of lactate assay buffer, then centrifuge. The soluble fraction can then be used for detection, following the same procedure as for serum samples.

Metabolite detection and analysis

To profile metabolites, tumor tissues were isolated and washed twice using precooled sodium thiosulfate. The samples were then homogenized in 500 µL of an ice-cold methanol-acetonitrile solution to extract metabolites and precipitate proteins. The resulting suspension was centrifuged at 14,000 g for 20 min at 4 °C. An aliquot of 300 µL from the supernatant was transferred to a fresh 1.5 mL EP tube and stored at −20 °C for 30 min. After a subsequent centrifugation (14,000 g, 10 min, 4 °C), 200 µL of the supernatant was passed through a protein precipitation plate. The samples were then analyzed using an untargeted metabolomic approach with an LC-ESI-MS/MS platform (ExionLC AD UHPLC) and a QTRAP 6500 + MS system, Sciex), following the methods previously described [22].

Glucose uptake assay and flow cytometry (FC)

Cancer cells were incubated in glucose-free PBS at 37 °C for 1 h to deplete intracellular glucose, and then stained with 50 µM 2-NBDG (Cat. MX4511, MKBio, China) at 37 °C for 30 min. The treated cancer cells were washed twice with cold PBS to remove unbound 2-NBDG, and then resuspended in PBS to 1 × 106 cells/mL. The fluorescence signals of 2-NBDG in the samples were detected by FC. Appropriate lasers and detection channels were set up to optimize signal acquisition, ensuring that the background signals were effectively distinguished. At least 104 events were collected to guarantee the statistical validity of the data. Data analysis was conducted using software FlowJo. Scatter plots were used to select individual cell populations, and 2-NBDG uptake in cells was presented as histograms of fluorescence intensity. Finally, draw the peak map to present the result.

Seahorse extracellular flux assay

Real-time oxygen consumption rate (OCR), extracellular acidification rate (ECAR), and proton efflux rate (PER) were measured using a Seahorse XF96 Extracellular Flux Analyzer (Agilent Technologies). Briefly, cells were seeded at a density of 2 × 10⁴ cells per well in a Seahorse XF96 cell culture microplate and incubated overnight at 37 °C in a 5% CO₂ incubator. One hour before each assay, the culture medium was replaced with the respective pre-warmed XF assay medium, and the plate was incubated at 37 °C in a CO₂-free incubator for equilibration. All compounds were injected sequentially from the XF sensor cartridge following the manufacturer’s guidelines. To measure OCR, cells were assayed in XF Base Medium supplemented with 10 mM glucose, 2 mM L-glutamine, and 1 mM pyruvate. A series of respiratory modulators was injected to reach final concentrations of 1.5 µM oligomycin, 0.5 µM FCCP, and 0.5 µM rotenone/antimycin A (Rot/AA). ECAR was measured in XF Base Medium supplemented with 2 mM L-glutamine. Substrates and inhibitors were injected to achieve final concentrations of 10 mM glucose, 1.0 µM oligomycin, and 50 mM 2-deoxy-D-glucose (2-DG). To measure PER, cells were assayed in XF Glycolytic Rate Assay Medium supplemented with 10 mM glucose, 2 mM L-glutamine, and 1 mM pyruvate. Compounds were injected to reach final concentrations of 0.5 µM Rot/AA and 50 mM 2-DG. All ECAR and OCR data were collected and analyzed using the Seahorse XFe96 Wave software.

Real-Time Quantitative PCR analysis (RT-qPCR)

According to the manufacturer’s plan, total RNA was extracted from xenograft tumor tissues and prostate cancer cell lines using TRIzol reagent (Invitrogen). The concentration and purity of the extracted RNA were measured using a NanoDrop 2000 spectrophotometer (Thermo Fisher Scientific, USA). For mRNA quantification, 1 ug of total RNA was reverse-transcribed into cDNA using the AdvanceFast 1 st Strand cDNA Synthesis Kit (Cat. 11149ES, Yeasen, China). Hieff qPCR SYBR Green Master Mix (Cat. 11202ES, Yeasen, China) was used for the subsequent reaction. The mRNA expression level was normalized to β-actin and calculated using 2^−ΔΔCt method. The primers used are listed in Table S2 (Supporting Information).

RNA Immunoprecipitation (RIP) assay

For RIP determination, we analyzed interactions among hnRNPA1, SRSF1, U2AF1, and the AR pre-mRNA using the EZ-Magna RIP Kit (Cat. 17–701, Sigma-Aldrich, GER) according to the manufacturer’s recommendations. In short, the cells were first lysed in a buffer and then incubated overnight at 4 °C with anti-hnRNPA1, anti-SRSF1, anti-U2AF1, or IgG. After washing, the RNA co-precipitated with hnRNPA1, SRSF1, and U2AF1 was extracted and analyzed using RT-qPCR. The primary antibodies used in this study are listed in Table S1 (Supporting Information).

Animal experiment

4-week-old male BALB/C nude mice purchased from the Animal Experiment Center (Super-B&K Laboratory Animal Corp. Ltd, Shanghai, China) were maintained under specific pathogen-free (SPF) conditions. In the subcutaneous tumor experiment, 2 × 10^6 PCa cells were subcutaneously injected into the right groin area first to allow them to grow in an androgen-dependent manner. One week after vaccination, when the tumor volume reached approximately 100 mm³, the tumor-bearing nude mice were castrated. After the remaining androgen was metabolized (one week after castration), the mice were randomly divided into a control group and each experimental group for tumor androgen-independent growth. Mice were treated with Enza (20 mg/kg, three times a week for 4 weeks) through oral tube feeding. During this period, drug intervention was carried out in specific subgroups. The intervention regimens included DMSO placebo, the KDM5B inhibitor CPI-455, the p300 inhibitor C646, and the PI3K inhibitor LY294002. Single- or combination-drug intervention was selected according to the specific experimental regimens (for details, see Figs. 2O and 7M; Figure S2E, Supporting information). Four weeks later, the mice were sacrificed, the tumor tissues were completely stripped off, weighed by light density, and fixed with 4% paraformaldehyde. Five biological replicates were set in each group. Furthermore, experiments such as IHC and TUNEL staining were performed on tumor tissues according to the specific situation.

Fig. 2.

KDM5B-mediated lactylation is a key mechanism of acquired Enza resistance in PCa. A RT-qPCR analysis of KDM5B mRNA expression across an immortalized prostate epithelial cell line and four PCa cell lines. B-C Determination of Enza half-maximal inhibitory concentrations (IC50​) in parental (LNCaP, C4-2) and Enza-resistant (LNCaP EnzR, C4-2 EnzR) cell lines. D Western blot analysis of KDM5B and Pan-Kla levels in parental versus Enza-resistant PCa cell lines (left). q-PCR detection of KDM5B mRNA levels in wild-type and drug-resistant LNCaP and C4-2 cell lines(right). E Western blot analysis of KDM5B and Pan-Kla levels in EnzR cells following KDM5B knockdown with specific shRNA (shKDM5B) compared to a non-targeting control (shNC). F Western blot analysis of KDM5B and Pan-Kla levels in parental PCa cells following KDM5B overexpression (oeKDM5B) compared to an empty vector (EV) control. G-H Western blot analysis of KDM5B and Pan-Kla levels in parental and EnzR cell lines treated with sodium lactate or sodium oxamate. I-J CCK-8 cell proliferation assays of KDM5B-knockdown EnzR cells (I) and KDM5B-overexpressing parental cells (J) treated with DMSO, Enza, or Enza + Sodium lactate or Enza + Sodium oxamate. K-N Colony formation assays assessing the long-term proliferative capacity of KDM5B-knockdown EnzR cells (K, M) and KDM5B-overexpressing parental cells (L, N) under DMSO, Enza, Enza + Sodium lactate, or Enza + Sodium oxamate treatment. Representative images and quantification are shown. O Schematic of the in vivo cell-derived xenograft (CDX) experimental design (Drawn using the Biorender platform). P-Q Tumor growth curves (P) and final tumor weights (Q) of CDX models derived from LNCaP EnzR cells (shNC vs. shKDM5B) treated with Enza or Enza plus lactate (n = 5 per group). R Representative images of excised tumors from all four treatment groups at the study endpoint. (S) Representative TUNEL immunofluorescence staining for apoptosis in tumor sections from the different treatment groups. Scale bar, 20 µm. ns, not significant; *p< 0.05; **p < 0.01; ***p < 0.001

Fig. 7.

p300 and HDAC1/2 respectively serve as the putatived “Writer” and “Erasers” for the lactylation of hnRNPA1-K179. A-D Identification of HDAC1 and HDAC2 as the delactylase enzymes ("erasers") for hnRNPA1. Western blot showing increased hnRNPA1-K179 lactylation upon treatment with the pan-HDAC inhibitor Trichostatin A (TSA), but not the sirtuin inhibitor nicotinamide (NAM) (A). Specific knockdown of HDAC1 (B) and HDAC2 (C), but not HDAC3 (D), increases K179 lactylation. E-I Identification of p300 as the lactyltransferase ("writer") for hnRNPA1. An shRNA screen targeting potential writers shows that only knockdown of p300 (E) significantly reduces K179 lactylation, whereas knockdown of other candidates has no effect (F-I). J-K Co-IP analyses confirming the physical interaction of hnRNPA1 with its writer, p300 (J), and its erasers, HDAC1 and HDAC2 (K), in EnzR cells. (L) RT-qPCR analysis demonstrating that pharmacological inhibition of p300 with C646 decreases AR-V7 mRNA levels, while inhibition of HDACs with TSA increases AR-V7 levels. M-Q In vivo validation of p300 inhibition as a therapeutic strategy to overcome Enza resistance. Schematic of the CDX model design (M) (Drawn using Biorender platform). Representative images of excised tumors (N), tumor growth curves (O), and final tumor weights (P) from mice treated with DMSO, Enza, the p300 inhibitor C646, the KDM5B inhibitor CPI-455, or the combination, demonstrating a synergistic anti-tumor effect. Representative TUNEL immunofluorescence staining for apoptosis in tumor sections (Q). ns, not significant; *p < 0.05; **p < 0.01; ***p < 0.001

Lentivirus infection and cell transfection

To establish stable KDM5B and AR-overexpressing cell lines, lentiviral vectors harboring the full-length coding sequences of KDM5B (NM_001314042.2) and AR (NM_000044.6) were purchased from GeneCopoeia (USA). These plasmids were co-transfected into HEK-293T cells with lentiviral packaging plasmids to produce viral particles. PCa cells were then infected with the harvested lentiviruses. Forty-eight hours post-infection, cells were selected with 2 µg/mL puromycin for 7 days to obtain stable overexpression cell lines. For gene knockdown, shRNA plasmids were purchased from Sigma-Aldrich (sequences are listed in Table S2 in the Supporting Information). The shRNA plasmids were transfected into PCa cells using Lipofectamine 3000 (Cat. L3000015, Invitrogen) according to the manufacturer’s instructions. The knockdown efficiency was subsequently verified by WB analysis to ensure successful target protein depletion before performing downstream functional experiments.

CRISPR/Cas9-Mediated hnRNPA1 KO in LNCaP/C4-2 EnzR cells

For hnRNPA1 KO, two single guide RNAs (sgRNAs) targeting the first exon of hnRNPA1 were selected according to the published sgRNA library. The specific sgRNA sequences were listed as follows: sghnRNPA1-1: 5’-CTCAGCTGTTCGGGCTCTTT-3’; sghnRNPA1-2: 5’-AAGAGCTTCCTCAGCTGTTC-3’. Synthesized oligos for two targeting sequences were annealed and ligated to the vector. The plasmids, including the respective sgRNA sequences, were then transfected into LNCaP/C4-2 EnzR cells, followed by screening with 5 µg/mL puromycin for 5 days. All stable cells were detected by immunoblotting.

Luciferase reporter assay

The promoter region of KDM5B was cloned into the firefly luciferase reporter vector pGL3-basic (Promega, USA) and co-transfected into HEK-293T cells together with the internal control plasmid pRL-TK (Promega, USA) and either pcDNA3.1-AR or empty pcDNA3.1 vector. Firefly luciferase activity was measured using the Dual-Luciferase Reporter Assay Kit (Beyotime, Cat. RG029S) according to the manufacturer’s protocol, with Renilla luciferase activity serving as a transfection control.

Chromatin Immunoprecipitation (ChIP) -qPCR

ChIP assays were performed using a ChIP kit (Cat. ab500, Abcam) according to the manufacturer’s protocol. Briefly, PCa cells were fixed with paraformaldehyde (1% final concentration) for 15 min to cross-link proteins to DNA. The nuclear extract was then separated and subjected to ultrasonic treatment to achieve chromatin fragmentation. To evaluate the binding of KDM5B and H3K4me3 to the PTEN promoter region, chromatin-fragmented DNA was incubated with anti-KDM5B, anti-H3K4me3, or anti-IgG. Similarly, to evaluate the binding of AR to the KDM5B promoter region, chromatin-fragmented DNA was immunoprecipitated with anti-AR or anti-IgG. Subsequently, the DNA was purified and analyzed by qPCR. The sequences of the primers used are listed in Table S2 (Supporting Information).

Protein purification

To obtain full-length human hnRNPA1 and p300 proteins, HEK293T cells were transiently transfected with plasmids encoding N-terminal Flag-tagged hnRNPA1 or C-terminal HA-tagged p300, respectively. Cells were harvested 48 h post-transfection. Recombinant proteins were purified using Flag (Cat. P2282, Beyotime, Chian) or HA affinity gels (Cat. P2287, Beyotime, China) and eluted competitively with Flag peptide (Cat. F3290, Sigma, USA) or HA peptide (Cat. P9808, Beyotime, China), following the manufacturer’s instructions. Protein integrity was verified by SDS-PAGE. Purified proteins were flash-frozen in liquid nitrogen and stored at −80 °C. The expression levels and purification yields of the hnRNPA1 mutants were comparable to those of the wild-type protein.

In vitro lactylation assay

In vitro lactylation assays were performed by incubating recombinant hnRNPA1 (WT or K179R, 1 µM) with p300 (50 nM) in a reaction buffer containing 50 mM HEPES (pH 7.4), 50 mM NaCl, 1 mM DTT, 0.01% Tween-20, and 200 µM Lactyl-CoA (Cat. FM120, Ikekule, China). Reactions were carried out for 30 min at 30 °C. For inhibition assays, p300 was pre-incubated with 2 µM C646 (0.1% DMSO final concentration) for 10 min at 25 °C prior to substrate addition. Reactions were terminated by the addition of 4× SDS loading buffer followed by boiling at 95 °C for 5 min.

In vitro delactylation assay

Lactylated hnRNPA1 was buffer-exchanged into delactylation buffer (50 mM HEPES pH 8.0, 50 mM NaCl, 1 mM DTT, 0.01% Tween-20) using 10 kDa molecular weight cutoff (MWCO) ultrafiltration columns (Cat, FUF510, Beyotime, China). The substrate was then incubated with 1 µM recombinant HDAC1 or HDAC2 for 1 h at 37 °C. Where indicated, the deacetylase inhibitor TSA (2 µM, 0.1% DMSO final) was added to the reaction. Assays were quenched with 4× SDS loading buffer and heated at 95 °C for 5 min. Samples were analyzed by immunoblotting using a specific anti-hnRNPA1-K179la antibody, with anti-hnRNPA1 and anti-HDAC antibodies used to detect total protein levels.

Lactylation LC-MS/MS omics

Total proteins of EnzS and EnzR PCa tissues were extracted using an 8 M urea buffer containing 100 mM Tris-HCl. For digestion, the protein solution was reduced with 5 mM dithiothreitol (DTT) for 30 min at 56 °C, followed by alkylation with 11 mM iodoacetamide (IAA) for 15 min at room temperature in the dark. The protein sample was then diluted to a urea concentration below 2 M. Finally, trypsin was added at a 1:50 trypsin-to-protein mass ratio for the first overnight digestion and a 1:100 ratio for the second 4-hour digestion.

The digested peptides were desalted using a Strata X C18 SPE column (Phenomenex) and vacuum-dried. To enrich for lactylated peptides, the sample was redissolved in NETN buffer (100 mM NaCl, 1 mM EDTA, 50 mM Tris-HCl, 0.5% NP-40, pH 8.0) and incubated with pre-washed Anti-L-Lactyllysine Antibody Conjugated Agarose Beads (PTM Bio, Cat. PTM-1404) at 4 °C overnight with gentle rotation. Following incubation, the beads were washed four times with NETN buffer and twice with deionized water to remove non-specifically bound peptides. The enriched lactylated peptides were then eluted from the beads with 0.1% trifluoroacetic acid (TFA).

The enriched peptides were analyzed using an EASY-nLC 1200 UHPLC system coupled to a Q Exactive HF-X mass spectrometer (Thermo Fisher Scientific). The resulting MS/MS data were processed using the MaxQuant search engine (v.1.6.15.0) and searched against the UniProt Homo sapiens database. Lactylation on lysine was specified as a variable modification. The false discovery rate (FDR) was adjusted to < 1% for both protein and peptide levels.

Statistical analysis

GraphPad Prism 10.1.2 and SPSS 26.0 were used for all statistical analyses. Data are presented as the mean ± SD of at least three independent experiments. A two-tailed Student’s t-test assessed statistical significance between two groups. One-way ANOVA followed by Tukey’s post hoc test evaluated significance among multiple groups. Correlation analysis was performed using Spearman’s correlation. A P-value less than 0.05 was considered statistically significant (*P < 0.05, **P < 0.01, ***P < 0.001).

Results

Identification of KDM5B in the high-lactylation metabolizing subpopulation of EnzR PCa

To explore the heterogeneity of Enza resistance, we first examined scRNA-seq data (GSE168668) from parental and EnzR LNCaP cells, identifying 16 distinct cell clusters (Fig. 1A, B). By classifying cells based on their resistance status and LRG activity (Table S3, Supporting Information), we identified four main subgroups: sensitive high-lactylation (SH), sensitive low-lactylation (SL), resistant high-lactylation (RH), and resistant low-lactylation (RL) (Fig. 1C). Notably, gene set variation analysis (GSVA) showed that the RH cluster was significantly enriched for glycolysis and resistance-related pathways [23, 24], while CellChat analysis indicated that this cluster had the most active intercellular communication network, with a key role as a sender of protumorigenic signals, such as collagen (Figure S1A-D).

Fig. 1.

Identification of KDM5B in the high-lactylation metabolizing subpopulation of Enza-resistant PCa. A Schematic illustrating the workflow for single-cell RNA sequencing (scRNA-seq) analysis. B Uniform Manifold Approximation and Projection (UMAP) plot depicting 16 distinct cell clusters identified from the scRNA-seq dataset. C Stratification of cell populations. Cells were categorized by sample origin (Enza-sensitive vs. -resistant, upper right) and by lactate-related gene (LRG) signature scores derived from the "AUCell" algorithm (upper left). Based on the median AUC score, cells were dichotomized into high- and low-lactylation groups (lower left), resulting in four subgroups: resistant-high lactylation (RH), resistant-low lactylation (RL), sensitive-high lactylation (SH), and sensitive-low lactylation (SL) (lower right). D Volcano plot of differentially expressed genes (DEGs) between Enza-sensitive and -resistant samples from the GSE104935 dataset (|logFC| < 0.5, p < 0.05). E Venn diagram showing the intersection of 1449 marker genes from the RH group and 1021 DEGs from GSE104935, yielding 91 overlapping genes. (F) Volcano plot of DEGs between paired PCa (PCa) and adjacent normal tissues from the TCGA-PRAD dataset (|logFC| < 0.5, p< 0.05). (G) Identification of KDM5B as the sole candidate gene by intersecting the 91 genes from (E) with 36 hub genes from the two most tumor-associated modules identified by Weighted Gene Co-expression Network Analysis (WGCNA). H scRNA-seq data from resistant cells stratified into KDM5B-positive (KDM5B+) and KDM5B-negative (KDM5B−) subgroups based on median KDM5B expression. I Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway enrichment analysis for genes upregulated in KDM5B+ resistant cells. J Pseudotime trajectory analysis of cells, colored by developmental time series (left), sample origin (right). K Western blot analysis of KDM5B expression in paired PCa and adjacent normal tissues from 12 patients in clinical cohort 1. L-M KDM5B expression is significantly upregulated in tumor samples compared to normal tissues in the GSE17951 (L) and TCGA-PRAD (M) public databases. N Violin plot demonstrating the positive correlation between KDM5B expression and Gleason scores in the TCGA-PRAD dataset. O Representative immunohistochemistry (IHC) of KDM5B in Enza-sensitive (EnzS) and Enza-resistant (EnzR) clinical samples (left). Quantitative real-time PCR (RT-qPCR) analysis of KDM5B mRNA levels in the same cohort (right). Scale bar, 20 µm. P Longitudinal monitoring of a PCa patient treated with Enza-based therapy. Prostate MRI images and corresponding HE and KDM5B IHC staining are shown at baseline (pre-treatment, top), during follow-up (middle), and at disease progression (bottom) (Drawn using the Biorender platform). Q The heatmap displays metabolomics analysis of EnzR and EnzS patients. R-S Quantification of lactate concentrations in serum (R) and tumor homogenates (S) from patients in clinical cohort 2. T Western blot analysis of KDM5B and Pan-Kla levels in EnzS and EnzR samples. Scale bar, 20 µm. ns, not significant; *p< 0.05; **p < 0.01; ***p < 0.001

To identify the main driver of the RH phenotype, we developed a multi-step bioinformatic filtering process. First, we found 1,021 genes upregulated in resistant LNCaP cells from a separate transcriptomic dataset (GSE104935) (Fig. 1D). Combining these genes with the marker genes of our RH cluster produced 91 candidates (Fig. 1E). Next, to link these genes to clinical tumor progression, we first performed differential gene analysis on cancer-adjacent paired samples from the TCGA-PRAD dataset. We then performed whole-genome co-expression network analysis (WGCNA) on these differentially expressed genes and identified 36 hub genes from the two modules most strongly correlated with tumor status (Fig. 1F; Figure S1E-H). This step-by-step screening ultimately pinpointed a single candidate gene, KDM5B, which is involved in both enzalutamide resistance and tumor progression (Fig. 1G). It should be noted that the ‘high-lactylation’ status of the RH cluster is a transcriptomic inference derived from gene signatures. While this provides a potential metabolic profile of the cell populations, it does not represent direct measurement of lactate turnover or protein-level modifications, and these results require validation in subsequent biochemical analyses.

KDM5B, a well-known histone demethylase that regulates gene expression by modifying chromatin accessibility, has been linked to cancer metabolism and therapy resistance [14, 15]. Given its crucial role in transcriptional reprogramming and the shift toward glycolytic metabolism observed in the RH cluster, we hypothesize that KDM5B may contribute to the development of the high-lactylation phenotype. To investigate this, we analyzed KDM5B expression in scRNA-seq data. We found that KDM5B levels were higher in the RH cluster compared to other clusters (Figure S1I), and KDM5B-high (KDM5B+) resistant cells showed enrichment of glycolysis and PI3K-Akt signaling pathways relative to KDM5B-low (KDM5B−) cells (Fig. 1H, I). Pseudotime trajectory analysis indicated that cells shift from a sensitive to a resistant state along the developmental pathway, with resistant cells occupying the terminal stage (Fig. 1J). Furthermore, when cells were divided into high and low KDM5B expression groups based on median KDM5B expression, the high KDM5B group showed a greater proportion in the later stages of cell differentiation (Figure S1J).

To clarify the relationship among KDM5B, tumor metabolism, and enzalutamide resistance, we conducted a systematic analysis of public databases and three separate in-house clinical cohorts. Detailed clinical information for cohorts 1 and 2 is available in Additional files 1 and 2 (Supporting Information), respectively. First, data from cohort 1 showed higher KDM5B protein levels in prostate cancer (PCa) tissues compared to adjacent normal tissues (Fig. 1K). Similar results were obtained from the analysis of the GSE17951 and TCGA-PRAD databases (Fig. 1L, M; Figure S1K). Next, based on an in-depth analysis of the TCGA database, we found that high KDM5B expression is positively correlated with Gleason score (Fig. 1N), suggesting a potential association with tumor malignant progression.

More importantly, cohort 2 showed that KDM5B expression was even more increased in cases of enzalutamide resistance (Fig. 1O; Figure S1L). A patient with ongoing disease in cohort 3 confirmed this finding. After developing resistance to enzalutamide, this patient had a significant rise in KDM5B levels in tumor tissue compared to before treatment (Fig. 1P). To explore changes in metabolic patterns in EnzR PCa tissues, we performed metabolomic analysis on tumor samples from five EnzS and three EnzR PCa patients in cohort 2. We saw that lactate production was higher in EnzR samples compared to EnzS samples (Fig. 1Q). This metabolomic result was confirmed by measuring serum lactate and tumor interstitial lactate levels in PCa patients (Fig. 1R, S). Since lactate is a substrate for lactylation, we then measured overall lactylation (Pan-Kla) levels in tumors. We found higher lactylation levels in EnzR samples, which showed a positive trend with KDM5B expression (Fig. 1T). These findings indicate that EnzR tissues have a hyper-lactate-producing metabolic profile, which may promote increased protein lactylation by providing more substrate. Additionally, KDM5B expression appears to be linked to this process.

KDM5B-mediated lactylation is a key mechanism of acquired Enza resistance in PCa

Based on RT-qPCR assay data, KDM5B expression levels were higher in all AR + PCa cell lines (VCaP, C4-2, 22RV1, and LNCaP) compared to normal prostate epithelial cells RWPE-1 (Fig. 2A). Therefore, LNCaP and C4-2 cells with moderate KDM5B levels were selected for subsequent studies. We then established EnzR cell lines through continuous dose-escalation culture (Fig. 2B, C). Notably, compared to their parental sensitive (EnzS) counterparts, both LNCaP EnzR and C4-2 EnzR cells showed significantly increased protein levels of KDM5B and Pan-Kla (Fig. 2D). This correlation led us to hypothesize that the upregulation of KDM5B might be an important factor contributing to acquired resistance, potentially through modulating protein lactylation.

To test this hypothesis, we then evaluated how KDM5B expression affects global protein lactylation. Knocking down KDM5B in EnzR cells or overexpressing KDM5B in EnzS cells led to a corresponding decrease or increase in Pan-Kla levels, respectively (Fig. 2E, F). Functionally, knocking down KDM5B in EnzR cells restored their sensitivity to Enza, shown by a significantly lower IC50 and reduced cell growth and survival under treatment (Figure S2A, C, E, G). Conversely, overexpressing KDM5B in EnzS cells gave them resistance to Enza, as indicated by an increased IC50 and sustained cell viability despite drug exposure (Figure S2B, D, F, H).

We next examined whether lactate itself mediates this phenotype. First, we verified that Pan-Kla levels could be directly adjusted by adding exogenous sodium lactate or sodium oxamate (Fig. 2G, H). Importantly, supplementing with sodium lactate reversed the increased sensitivity of KDM5B-knockdown EnzR cells, restoring their resistance to Enza in both proliferation and colony formation tests (Fig. 2I, K, M). Conversely, treating cells with sodium oxamate reduced the resistance caused by KDM5B overexpression in EnzS cells, re-sensitizing these cells to Enza (Fig. 2J, L, N).

Finally, to validate these findings in vivo, we established a mouse CDX model using LNCaP EnzR cells with or without KDM5B knockdown. Consistent with our in vitro data, KDM5B knockdown enhanced the therapeutic efficacy of Enza, leading to reductions in tumor volume and weight, and a corresponding increase in apoptosis (Fig. 2O-S). Importantly, administration of sodium lactate to mice attenuated the Enza therapeutic advantage (Fig. 2O-S). Taken together, these data support that KDM5B promotes acquired Enza resistance, in part, by activating a lactate-mediated signaling cascade in the studied cellular and animal models.

KDM5B reprograms glucose metabolism and confers Enza resistance by upregulating PGK1

Based on our scRNA-seq analysis, which showed an enrichment of glycolytic pathways in KDM5B+ resistant cells (Fig. 1I), combined with our previous data indicating a positive correlation between KDM5B expression and global Pan-Kla levels (Fig. 2E and F). To test this, we first confirmed the metabolic phenotype of our resistant cells. Indeed, EnzR cells displayed a hyperglycolytic state, characterized by increased glycoPER and ECAR compared to their sensitive (EnzS) counterparts (Fig. 3A, Figure S3A). Importantly, this glycolytic reprogramming was found to be closely related to the expression of KDM5B. This was demonstrated by KDM5B knockdown in EnzR cells, which reduced their glycolytic capacity, glucose uptake, and lactate production (Figs. 3B, D, F; Figure S3B). Conversely, ectopic KDM5B overexpression in EnzS cells was sufficient to increase these glycolytic measures (Figs. 3C, E, F; Figure S3C). Therefore, these data suggest that the reprogramming of glucose metabolism observed in Enza-resistant cells can be driven by KDM5B.

Fig. 3.

KDM5B reprograms glucose metabolism and confers Enza resistance by upregulating PGK1. A-C Seahorse XF metabolic flux analysis. Comparison of the Glycolytic Proton Efflux Rate (GlycoPER), Extracellular Acidification Rate (ECAR), and Oxygen Consumption Rate (OCR) between Enza-sensitive (EnzS) and EnzR-resistant (EnzR) LNCaP cells (A), in EnzR cells following KDM5B knockdown (B), and in parental cells following KDM5B overexpression (C). D-E Glucose uptake assays using the fluorescent glucose analog 2-NBDG. Quantification of glucose uptake in KDM5B-knockdown EnzR cells (D) and in KDM5B-overexpressing parental cells (E). F Lactate production assays in KDM5B-knockdown EnzR cells (left) and KDM5B-overexpressing parental cells (right). G Volcano plot of DEGs from RNA-sequencing of LNCaP cells overexpressing KDM5B versus an empty vector control (|logFC| < 1, p < 0.05), with AR and upregulated glycolysis-related genes marked. H The heatmap displays the relative expression levels of five upregulated glycolytic enzymes and KDM5B, along with three AR target genes, identified from RNA sequencing data. I Schematic diagram of the glycolysis pathway, highlighting the step catalyzed by these five upregulated glycolytic enzymes. J-K Western blot analysis validating the Phosphoglycerate Kinase 1 (PGK1) protein levels following KDM5B knockdown in EnzR cells (J) and KDM5B overexpression in parental cells (K). L-M Knockdown of PGK1 sensitizes resistant cells to Enza. Enza IC50​ determination (L) and CCK-8 proliferation assays (M) in LNCaP EnzR cells transfected with shNC or shPGK1. N-Q Rescue experiments demonstrating that PGK1 is a key downstream mediator of KDM5B. Enza IC50​ determination (N), CCK-8 proliferation assays (O), and colony formation assays (Q) were performed in KDM5B-overexpressing (oeKDM5B) LNCaP cells with or without concurrent PGK1 knockdown. Colony formation in PGK1-knockdown LNCaP EnzR cells is shown in (P) for comparison. ns, not significant; *p< 0.05; **p < 0.01; ***p < 0.001

To identify the mediator of this metabolic shift, we performed RNA sequencing on cells overexpressing KDM5B compared to a vector control (LNCaP/oeKDM5B vs. LNCaP/vector). Among several upregulated glycolytic enzymes, PGK1 showed the most significant increase (Fig. 3G-I). This was confirmed by RT-qPCR, which also showed that PGK1 was the most upregulated glycolytic enzyme in EnzR cells and that its expression was reduced after KDM5B knockdown (Figure S3D-F). Previous study transcriptome data (LNCaP-EnzR vs. LNCaP-WT) also indicated PGK1 was upregulated in drug-resistant cell lines [25]. Additionally, protein-protein interaction (PPI) analysis placed KDM5B, PGK1, and the AR within a tightly associated module (Figure S3G, H), suggesting a potential specific regulatory coordination between this epigenetic-androgen signaling axis and PGK1-mediated metabolism. Although overexpressing KDM5B did not change AR mRNA levels, it significantly increased the expression of its transcriptional targets (KLK3, FKBP5, and TMPRSS2), suggesting that KDM5B may exert its function by enhancing AR activity indirectly. This enhancement of AR function challenges the typical inhibitory role of KDM5B, suggesting the existence of an indirect regulatory mechanism (Fig. 3G, H).

We next validated the direct regulatory link by confirming that KDM5B knockdown or overexpression decreased or increased PGK1 protein levels, respectively (Fig. 3J, K). To determine if PGK1 was functionally required for the resistance phenotype, we performed rescue experiments. Knockdown of PGK1 in EnzR cells was sufficient to re-sensitize them to Enza, reducing their IC50 value, cell viability, and colony formation under drug treatment (Fig. 3L, M, P; Figure S3I, J, M). Additionally, concurrent knockdown of PGK1 attenuated the pro-resistant phenotype conferred by KDM5B overexpression in EnzS cells, partially restoring their sensitivity to Enza (Fig. 3N, O, Q; Figure S3K, L, N). To further determine whether the metabolic reprogramming driven by KDM5B is dependent on its histone demethylase activity, we constructed a catalytic-dead KDM5B mutant with a point mutation in the JmjC domain (H499A). While wild-type KDM5B significantly reduced global H3K4me3 levels, the H499A mutant failed to alter H3K4me3 status, confirming the loss of its catalytic function. Importantly, unlike wild-type KDM5B, overexpression of the H499A mutant failed to induce the expression of the key glycolytic enzyme PGK1(Figure S3O, P). These results indicate that the ability of KDM5B to promote​ glycolytic reprogramming is likely an active epigenetic process that requires its demethylase activity. KDM5B reprograms cellular metabolism and contributes to Enza resistance primarily by upregulating PGK1 transcription.

KDM5B activates the PI3K/Akt pathway by suppressing PTEN via epigenetic mechanisms, driving glycolysis, and promoting resistance to enzalutamide

To investigate the downstream pathways regulated by KDM5B, we analyzed differentially expressed genes (DEGs) from our KDM5B overexpression RNA-seq dataset. Functionally, the DEGs pointed towards two primary biological themes. First, as expected, GO terms related to KDM5B’s core enzymatic function, such as “histone lysine demethylation” and “histone demethylation”, were enriched (Fig. 4A). Second, and of greater mechanistic interest, KEGG pathway analysis identified enrichment in both “Glycolysis/Gluconeogenesis” and the “PI3K-Akt signaling pathway” (Fig. 4B). This finding was consistent with scRNA-seq analysis, which also showed enrichment of the PI3K-Akt pathway in KDM5B+ resistant cells (Fig. 1I).

Fig. 4.

KDM5B activates the PI3K/Akt pathway by suppressing PTEN via epigenetic mechanisms, driving glycolysis, and promoting resistance to enzalutamide. A-B The enriched Gene Ontology (GO) and KEGG functional pathways for RNA-seq data. C-D Western blot analysis demonstrating that KDM5B knockdown increases PTEN and decreases p-Akt levels in EnzR cells (C), whereas KDM5B overexpression has the opposite effect in parental cells (D). H3K4me3 levels are shown as a control for KDM5B demethylase activity. E Immunofluorescence staining showed the localization of Akt (green) and p-Akt (red) in xenograft tumors in mice under different treatment conditions. Images were captured by confocal laser scanning microscopy (scale: 20 µm). F-I Pharmacological inhibition of the PI3K/Akt pathway reverses the KDM5B-induced glycolytic phenotype. Seahorse XF metabolic flux analysis (F, H) and glucose uptake assays (G, I) were performed on EnzR cells treated with DMSO (vehicle) or the PI3K inhibitor LY294002 (10 µM). (J-K) Western blot analysis showed that LY294002-mediated PI3K inhibition altered KDM5B regulation of PGK1. L-Q PI3K inhibition restores sensitivity to Enza, showing a synergistic effect. CCK-8 proliferation assays (L-O) and colony formation assays (P, Q) were conducted on EnzR cells and oeKDM5B cells. Cells were treated with Enza (10 µM), LY294002 (10 µM), or a combination of both. ns, not significant; *p< 0.05; **p < 0.01; ***p < 0.001

Since KDM5B mediates transcriptional silencing and PTEN acts as the key negative regulator of the PI3K/Akt pathway, we hypothesized that KDM5B might directly regulate PTEN expression. In fact, KDM5B knockdown in EnzR cells increased PTEN protein levels and decreased Akt phosphorylation (p-Akt) (Fig. 4C). Conversely, KDM5B overexpression in EnzS cells decreased PTEN levels and boosted p-Akt expression (Fig. 4D). Consistent with its role as a demethylase, KDM5B levels were inversely related to global H3K4me3 levels (Fig. 4C, D). Importantly, we found that KDM5B directly occupied the PTEN promoter in EnzR cells, resulting in a reduction of the active H3K4me3 mark at this site (Figure S4B). Supporting this, pharmacological inhibition of KDM5B with CPI-455 led to increased PTEN transcription (Figure S4A) and decreased Akt phosphorylation and nuclear translocation (Fig. 4E).

Finally, we sought to determine whether the activated PI3K/Akt pathway is functionally required for the metabolic and resistance phenotypes. Pharmacological inhibition of PI3K with LY294002 reversed the hyperglycolytic phenotype of EnzR cells, decreasing their glycolytic capacity, glucose uptake, and lactate production (Fig. 4F-I; Figure S4C). Importantly, PI3K inhibition also suppressed PGK1 expression in EnzR cells and reduced the upregulation of PGK1 caused by KDM5B overexpression in EnzS cells (Fig. 4J, K). Notably, in the C4-2 cell line, LY294002 did not significantly reduce the PGK1 elevation triggered by KDM5B overexpression, possibly due to alternative compensatory mechanisms that regulate PGK1 independently of the PI3K/Akt pathway [26]. Functionally, treatment with LY294002 re-sensitized both EnzR cells and KDM5B-overexpressing EnzS cells to Enza, inhibiting cell growth and survival under drug pressure (Fig. 4L-Q). Collectively, these results support that KDM5B epigenetically suppresses PTEN, thereby activating the PI3K/Akt signaling pathway. This promotes the upregulation of PGK1 expression and enhances glycolytic function, contributing to drug resistance in tumor cells.

Lactylation of hnRNPA1-K179 drives AR-V7 splicing and confers Enza resistance in PCa

Given that KDM5B promotes glycolysis, we first sought to connect this metabolic trait to downstream signaling. GSEA of the TCGA-PRAD dataset showed an enrichment of the “alternative mRNA splicing via spliceosome” pathway (ES = 0.6876, NP = 0.0000) in the group with high KDM5B expression (Fig. 5A). This prompted us to hypothesize that KDM5B-driven lactate buildup might influence AR splicing to produce the resistance-associated variant, AR-V7. Supporting this, in our mouse CDX model, tumor tissues from KDM5B-knockdown cells had lower lactate levels (Fig. 5B). When tumor tissues were treated with lactate, there was a time- and dose-dependent increase in AR-V7 mRNA levels (Fig. 5C), establishing a direct connection between lactate and AR-V7 expression.

Fig. 5.

Lactylation of hnRNPA1-K179 drives AR-V7 splicing and confers Enza resistance in PCa. A Gene Set Enrichment Analysis (GSEA) shows enrichment of the "alternative mRNA splicing via spliceosome" pathway in the KDM5B-high expression group from the TCGA-PRAD database. B Schematic diagram of the CDX model (left) and quantification of lactate levels in resected tumors from different intervention groups (right, n = 5) (Drawn using Biorender platform). C RT-qPCR analysis demonstrates that lactate treatment increases AR-V7 mRNA levels in xenograft tissues in a dose- (left) and time-dependent (right) manner. D Schematic diagram of the lactylation omics analysis of CDXEnzR and CDXparental tissues (Drawn using Biorender platform). E Radar map showing the top 20 differentially lactylation proteins between CDXEnzR and CDXparental tissues; the larger purple circles represent higher log2FC values; blue and green numbers represent the quantification of protein lactylation in the CDXEnzR and CDXparental models, respectively. F-G Co-immunoprecipitation (Co-IP) analysis confirming increased lactylation of endogenous hnRNPA1 in Enza-resistant (EnzR) LNCaP (F) and C4-2 (G) cells compared to their parental counterparts. H-J K179 was precisely identified as a potential lactylation site. MS/MS spectra confirmed lactylation of hnRNPA1 at lysine 179 (K179) (H); molecular docking modeling revealed an interaction between L-lactate and the K179 residue (I); and multiple sequence alignment demonstrated cross-species conservation of K179 (J). K-M Validation of the K179 lactylation site using a knockout-reconstitution system. Schematic of the experimental workflow (K) (Drawn using Biorender platform), Western blot confirming hnRNPA1 knockout (KO) (L), and Co-IP analysis in KO cells reconstituted with wild-type (WT), K179R, or K166R mutant hnRNPA1, identifying K179 as the specific lactylation site (M). N Western blot analysis showing that lactate treatment increases, while sodium oxamate decreases, hnRNPA1 lactylation in EnzR cells. O Western blot analysis showed that KDM5B regulates hnRNPA1-K179la. P Western blot analysis demonstrating that overexpression of hnRNPA1 (WT), but not the hnRNPA1 (K179R) mutant, increases AR-V7 protein levels. Q-S Functional validation that hnRNPA1 lactylation at K179 is required for drug resistance. CCK8 proliferation assays (Q) and colony formation assays (R) were performed in hnRNPA1-KO EnzR cells reconstituted with WT, K179R, or K166R hnRNPA1 and treated with Enza (10 µM). Quantitative detection of colony formation (S). ns, not significant; *p < 0.05; **p < 0.01; ***p < 0.001

Previous studies have shown that AR-V7 expression was upregulated when splicing factors (e.g., hnRNPA1, SRSF1, U2AF1) were aberrantly active [27]. To identify the molecular link between lactate and splicing, we performed global lactylation proteomics on EnzS and EnzR tissues (Fig. 5D). 364 sites on 257 proteins showed increased lactylation levels, with hnRNPA1 being a prominently lactylated splice protein in EnzR tissues (Fig. 5E). Studies indicate that hnRNPA1 can undergo lactylation, promoting tumor progression [28, 29], but the underlying mechanism remains unclear. Validating this proteomics finding with Co-IP, endogenous hnRNPA1 was indeed over-lactylated in EnzR cells compared to their parental counterparts (Fig. 5F, G). Functionally, hnRNPA1’s binding to AR pre-mRNA was enhanced in EnzR cells. This increased binding depended on KDM5B and could be rescued by exogenous sodium lactate (Figure S4D, E).

Our lactylome analysis identified two potential modification sites on hnRNPA1: K166 and K179 (Fig. 5H; Figure S4F). Molecular docking further pinpointed K179 as a candidate residue for lactate binding (Fig. 5I). Furthermore, both the K166 and K179 sites were highly conserved across species, suggesting functional importance (Fig. 5J; Figure S4G). To further pinpoint the location of this modification, we used a series of truncation mutants. We found that lactylation of hnRNPA1 primarily occurs within its RRM2 domain (Figure S4H-J), which contains the K166 and K179 residues.

To definitively test the function of the K166 and K179 sites, we used CRISPR/Cas9 to knock out endogenous hnRNPA1 in EnzR cells and then reconstituted them with WT, K166R, or K179R mutant plasmids (Fig. 5K, L). As expected, the K179R mutation abolished the lactylation signal, which was confirmed to be modulated by sodium lactate and sodium oxamate (Fig. 5M, N). Furthermore, changes in KDM5B expression simultaneously affected hnRNPA1 K179 lactylation levels (Fig. 5O). To further confirm the critical role of upstream metabolic signaling, we knocked down PGK1 in EnzR cells. We found that the level of hnRNPA1 K179 lactylation was significantly reduced (Figure S4K), indicating that PGK1-driven glycolysis is the principal lactate source for K179 lactylation. Additionally, reconstitution with WT hnRNPA1, but not the K179R mutant, led to a significant increase in AR-V7 protein levels (Fig. 5P). Functionally, while hnRNPA1 knockout made cells more sensitive to Enza, re-expression of WT hnRNPA1, but not the K179R mutant, restored cell viability and colony formation under drug treatment (Fig. 5Q). Furthermore, we validated whether targeting hnRNPA1 K179 lactylation reduces Enza resistance in vivo. We found that Enza inhibits tumor growth in the hnRNPA1-KO EnzR model, and overexpression of hnRNPA1 (WT), rather than the K179R mutant, rescues tumor growth in vivo (Figure S4L-P). Finally, we confirmed that lactate treatment could no longer increase AR-V7 expression in cells expressing the K179R mutant (Figure S4Q, R). These results demonstrate that the lactylation of hnRNPA1 at K179 is a critical contributing factor that promotes AR-V7 splicing and confers enzalutamide resistance in the studied model system, highlighting its role as a context-dependent regulatory node.

Lactylation at K179 stabilizes hnRNPA1 by inhibiting its NEDD4L-mediated ubiquitination and proteasomal degradation

Next, we investigated how hnRNPA1 lactylation improves AR-V7 splicing. Because protein stability can influence splicing factor activity, we first conducted a cycloheximide (CHX) chase assay. This showed that the non-lactylatable K179R mutant was less stable than wild-type (WT) hnRNPA1. Additionally, lactate treatment increased the stability of WT hnRNPA1 but did not affect the K179R mutant, indicating that K179 lactylation boosts hnRNPA1 stability (Fig. 6A, B). Since protein stability is often controlled by ubiquitination, we measured hnRNPA1 ubiquitination levels. We observed that ubiquitination was reduced in EnzR cells (Fig. 6C). Consistent with this, inhibiting glycolysis increased hnRNPA1 ubiquitination, while lactate treatment had the opposite effect (Fig. 6D).

Fig. 6.

Lactylation at K179 stabilizes hnRNPA1 by inhibiting its NEDD4L-mediated ubiquitination and proteasomal degradation. A-B Protein stability assays. hnRNPA1-KO EnzR cells reconstituted with WT or K179R mutant hnRNPA1 were treated with the protein synthesis inhibitor cycloheximide (CHX) (40uM), with or without lactate (20uM). Western blots (A) and quantification of the hnRNPA1 half-life (B) show that lactate stabilizes WT but not the K179R hnRNPA1. C-D In vivo ubiquitination assays. Ubiquitination of Flag-hnRNPA1 was lower in EnzR cells than in parental cells (C). Lactate treatment (20uM) decreased, while sodium oxamate (20uM) increased, hnRNPA1 ubiquitination (D). E-G Identification of NEDD4L as the E3 ubiquitin ligase for hnRNPA1. Bioinformatic prediction of potential E3 ligases (top 20) using the UbiBrowser 3.0 platform (http://ubibrowser.ncpsb.org.cn) (E), a volcano plot of DEGs from KDM5B-overexpressing cells (|logFC| 1, p< 0.05) (F), and a Venn diagram intersecting the downregulated genes in oeKDM5B and genes predicted by bioinformatics identified NEDD4L as the sole candidate (G). H Molecular docking diagram of hnRNPA1 and NEDD4L. I EnzR cells were cotransfected with Flag-hnRNPA1, HA-Ub, and (or) His-NEDD4L plasmids; Flag-tagged BLM proteins were immunoprecipitated with anti-Flag antibody and analyzed by western blot with antibodies as indicated. J EnzR cells cotransfected with Flag-hnRNPA1 (WT), Flag-hnRNPA1 (K179R), His-NEDD4L, and HA-Ub plasmids were treated with 20 mM sodium lactate or left untreated; Flag-hnRNPA1 proteins were immunoprecipitated with an anti-Flag antibody or an IgG control and analyzed via western blotting with anti-Flag and target antibodies as indicated. ns, not significant; *p < 0.05; **p < 0.01; ***p < 0.001

To identify the E3 ubiquitin ligase responsible for hnRNPA1 degradation, we performed a bioinformatics screen. By overlapping the top 20 predicted E3 ligases from the UbiBrowser database (http://ubibrowser.ncpsb.org.cn) [30]with genes downregulated after KDM5B overexpression, we identified NEDD4L as the only candidate (Fig. 6E-G). Recent research shows that NEDD4L-mediated ubiquitination plays a key role in suppressing ferroptosis and conferring Enza resistance in CRPC [31]. Molecular docking predicted a strong binding interface between NEDD4L and hnRNPA1 (Fig. 6H). Confirming this, NEDD4L was found to interact with and promote the ubiquitination of hnRNPA1 in EnzR cells (Fig. 6I). Notably, we found that the lactylation status of K179 affects this interaction. The K179R mutation increased the binding between hnRNPA1 and NEDD4L, leading to more ubiquitination. Conversely, lactate treatment disrupted the hnRNPA1-NEDD4L interaction and the ubiquitination of the wild-type protein but had no effect on the K179R mutant (Fig. 6J). These findings suggest that lactylation at K179 acts as a protective shield by preventing the binding of the E3 ligase NEDD4L, thus inhibiting hnRNPA1’s ubiquitination and subsequent degradation.

p300 and HDAC1/2, respectively, serve as the putative “Writer” and “Erasers” for the lactylation of hnRNPA1-K179

To clarify the regulatory mechanism of hnRNPA1 lactylation, we conducted a series of knockdown experiments targeting different histone deacetylases (HDACs) and lactosyltransferases in EnzR cell lines. Tricostatin A (TSA) is a broad-spectrum inhibitor that mainly targets class I and II HDACs, while nicotinamide (NAM) inhibits class III HDACs. We observed that TSA treatment significantly increased lactylation at residue K179 of hnRNPA1. In contrast, NAM treatment did not produce statistically significant changes in hnRNPA1-K179 lactylation levels (Fig. 7A). Additionally, we targeted individual HDACs, including HDAC1, HDAC2, and HDAC3, and found that knockdown of HDAC1 and HDAC2 elevated lactylation at the K179 site of hnRNPA1 (Fig. 7B-D).

Additionally, we characterized a series of lactoyltransferases. It was found that knocking down p300 decreased hnRNPA1 lactylation (Fig. 7E), while knocking down the other lactyltransferases (GCN5, CBP, PCAF, and AARS1) did not cause significant changes in hnRNPA1 lactylation levels (Fig. 7F). hnRNPA1 was also observed to interact with p300 as well as the deacetylases HDAC1 and HDAC2 (Fig. 7J, K). Inhibiting p300 with C646 reduced AR-V7 mRNA levels, whereas inhibiting HDACs with TSA increased them (Fig. 7L). To definitively confirm the direct catalytic roles of p300 and HDAC1/2, we performed in vitro biochemical reconstitution assays using purified proteins. The results showed that p300 directly lactylated WT-hnRNPA1 in the presence of Lactyl-CoA, but not the K179R mutant, and this effect was inhibited by C646 (Figure S5A). Conversely, purified HDAC1 and HDAC2 proteins directly mediated the delactylation of hnRNPA1, a process sensitive to TSA treatment (Figure S5B). These biochemical data provide direct evidence for the p300-HDAC1/2 writer-eraser framework.

To investigate whether a multi-target combined intervention could restore and improve the sensitivity of EnzR cells to Enza, we conducted an animal experiment (Fig. 7M). The results showed that, based on the effect of Enza, the combination of CPI-455 or C646 synergistically inhibited tumor growth (Fig. 7N). Importantly, we found that the therapeutic strategy of Enza combined with CPI-455 and C646 exerted a greater inhibitory effect on tumors (Fig. 7N). These findings suggest that Enza, when combined with a multi-target synergistic approach, could be an effective strategy for overcoming enzalutamide resistance.

A potential positive feedback loop between KDM5B and ligand-independent AR signaling drives Enza resistance

Given that KDM5B expression is elevated in Enza-resistant PCa cells and considering the important role of AR in this process, we aimed to analyze the relationship between KDM5B and AR using RNA-seq profiles from the TCGA-PRAD database. GSEA showed a significant enrichment of the “Androgen Response” pathway in the KDM5B-high expression group (Fig. 8A), and a strong positive correlation was observed between KDM5B and AR mRNA levels (Fig. 8B). This correlation was further supported by our finding that AR-positive PCa cell lines (LNCaP, C4-2) express higher levels of KDM5B than AR-negative lines (DU-145, PC-3) (Fig. 8C). To investigate a potential direct regulatory relationship, we modulated AR activity. Knockdown of AR decreased KDM5B expression, while AR overexpression had the opposite effect (Fig. 8D, E; Figure S5C, D). Furthermore, stimulating AR activity with its ligand, dihydrotestosterone (DHT), or increasing AR protein levels via lentiviral delivery, both resulted in a dose-dependent increase in KDM5B expression (Figure S5E-H).

Fig. 8.

A positive feedback loop between KDM5B and ligand-independent AR signaling drives Enza resistance. A GSEA showing enrichment of the "Androgen Response" pathway in the KDM5B-high expression group from the TCGA-PRAD database. B-C Correlation analysis of KDM5B and AR mRNA expression in the TCGA-PRAD cohort (B) and across various prostate cell lines (C). D-E Western blot (D) and RT-qPCR (E) analyses showing that AR knockdown decreases KDM5B expression. F-G AR directly binds to the KDM5B promoter. Public ChIP-seq data (ChIP-Atlas [https://chip-atlas.org/]) showing AR occupancy at the KDM5B promoter (F). ChIP-qPCR confirming AR enrichment at the KDM5B promoter in PCa cells (G). H-I Validation of the AR binding site on the KDM5B promoter. Predicted AR binding motifs from the JASPAR database (H). Dual-luciferase reporter assay showing that AR enhances the activity of the wild-type (WT) but not the mutant (MUT1+2) KDM5B promoter (I). J Western blot analysis showing that KDM5B overexpression increases phosphorylation of Akt (p-Akt) and AR (p-AR), an effect abrogated by the PI3K inhibitor LY294002. K Representative immunofluorescence images showing increased nuclear p-AR (red) in KDM5B-overexpressing xenograft tumors. Scale bar, 10 µm. (L) Western blot analysis of the canonical AR target gene KLK3, showing that KDM5B overexpression increases KLK3 levels in a PI3K-dependent manner. M-N IGF-1-induced ligand-independent AR activation drives KDM5B expression. Western blot showing that IGF-1 (6nM) treatment increases p-Akt, p-AR, and KDM5B levels (M). ChIP-qPCR showing that IGF-1 treatment promotes AR binding to the KDM5B promoter (N). O Schematic model of the proposed mechanism. KDM5B enhances glycolysis, leading to lactate accumulation and hnRNPA1 lactylation, which ultimately increases AR-V7 expression. Concurrently, KDM5B activates the PI3K/Akt pathway, promoting ligand-independent AR phosphorylation. This activated AR, in turn, directly drives KDM5B transcription, creating a vicious cycle that sustains Enza resistance and underscores the potential of multi-target intervention as a therapeutic strategy (Drawn using the Biorender platform). ns, not significant; *p < 0.05; **p < 0.01; ***p < 0.001

To confirm this transcriptional regulation, we analyzed public ChIP-seq databases (ChIP-Atlas) and bioinformatic prediction tools, which consistently identified AR as a transcription factor binding to the KDM5B promoter (Fig. 8F; Figure S5I). We validated this prediction experimentally using ChIP-qPCR (Fig. 8G). Consistent with this, analysis of a dataset (GSE48403) showed that KDM5B expression was reduced in patients after ADT (Figure S5J).

Using the JASPAR database, we identified two potential AR binding sites in the KDM5B promoter. We then conducted a dual-luciferase reporter assay with constructs containing either the wild-type or mutated version of this promoter region. Co-transfection with an AR expression plasmid significantly increased the luciferase activity of the wild-type promoter, an effect that was reduced when the AR binding sites were mutated (Fig. 8H, I). Overall, these data suggest that AR can bind to the KDM5B promoter and promote its transcription.

Finally, we explored the other side of this regulatory axis. Since KDM5B activates the PI3K/Akt pathway, which is known to phosphorylate and activate AR in a ligand-independent way [32], we first aimed to confirm this link. Overexpression of KDM5B caused a significant rise in p-AR and its movement to the nucleus; importantly, this was reversed by the PI3K inhibitor LY294002 (Fig. 8J, K). This KDM5B-driven AR activation was functionally significant, as it was enough to trigger the expression of the usual AR target gene, KLK3, even without an androgenic ligand (Fig. 8L; Figure S5K). These results suggested supported the existence of a positive feedback loop, where KDM5B-driven PI3K/Akt signaling could, in turn, boost AR-mediated transcription of KDM5B. To test this, we used the PI3K/Akt activator IGF-1. As expected, IGF-1 mimicked the effect of KDM5B, raising p-AR levels (Fig. 8M). Importantly, IGF-1 also increased AR binding to the KDM5B promoter and raised its expression (Fig. 8M, N).

To evaluate the clinical relevance of the KDM5B axis, we tested its stability under pharmacological blockade. While apalutamide successfully suppressed the canonical AR-target KLK3, it failed to impede the KDM5B-mediated induction of PGK1, p-Akt, and AR-V7 (Figure S5L). These results indicate that the KDM5B-driven pathway sustains the expression of AR-V7 and its downstream effectors. Consequently, this pathway preserves a source of AR signaling that is intrinsically independent of the LBD, thereby persisting even when full-length AR is effectively occupied.

Discussion

As a second-generation ARSI, enzalutamide inhibits PCa progression through a multi-targeted mechanism of action. These include competitive antagonism of androgen binding to the AR, inhibition of AR nuclear translocation, and interference with AR-DNA interactions. The results of several large clinical trials have demonstrated that enzalutamide has a significant positive effect on metastatic castration-resistant prostate cancer (mCRPC) [33], non-metastatic castration-resistant prostate cancer (nmCRPC) [34], and metastatic hormone-sensitive prostate cancer (mHSPC) [35, 36]. However, tumor resistance still poses a major challenge in the treatment of PCa. It involves various aspects such as AR signaling escape [37], compensatory bypass activation [38, 39], epigenetic remodeling [40], and tumor microenvironment-mediated stress adaptation [41]. In this study, we systematically investigated the critical role played by KDM5B in enzalutamide resistance in PCa. First, KDM5B activates the PTEN/PI3K/Akt signaling pathway through demethylation, then upregulates PGK1, leading to metabolic reprogramming of tumor cells (Warburg Effect; enhanced glycolysis). Second, the metabolite lactate induces the lactylation of lysine at site 179 of hnRNPA1 through the regulatory network of p300, HDAC1, and HDAC2, thereby causing abnormal splicing of AR. The upregulation of AR-V7 expression leads to PCa resistance to enzalutamide. Furthermore, under the action of KDM5B, the ligand-independent AR signaling pathway was activated. The binding of AR to the promoter region of KDM5B instead promoted its transcription and expression (Fig. 8O). Our findings provide new insights into the interplay among metabolism, epigenetics, and enzalutamide resistance in PCa. Targeting the key molecules in this complex network is expected to reverse enzalutamide resistance in PCa.

Using single-cell sequencing data, we identified for the first time an abnormally high expression of KDM5B in an enzalutamide-resistant, high-lactylation PCa cluster, which was validated across multiple GSE datasets and clinical samples. This phenomenon also underscores the critical role of KDM5B in regulating tumor cell metabolic reprogramming. Previous studies have shown that KDM5B promotes AUP1 transcription, leading to lipid metabolic reprogramming in tumor cells and thereby promoting the malignant progression of cervical cancer [14], and that KDM5B enhances the proliferation and migration of breast cancer cells through AMPK-mediated lipid metabolic reprogramming [42]. Additionally, NEAT1_2 promotes aerobic glycolysis in papillary thyroid carcinoma by recruiting KDM5B to mediate H3K4me3 modification of the RRAD promoter [43]. Still, the Warburg effect mediated by KDM5B in the context of endocrine therapy for PCa is the first of its kind. Our results indicate that KDM5B can activate the PI3K/Akt signaling pathway by epigenetically silencing PTEN, which then upregulates the key glycolytic enzyme PGK1 and driving metabolic reprogramming in tumor cells. While multiple glycolytic enzymes such as HK2, LDHA, and PFKFB3 have been implicated in ARSI resistance through partially redundant mechanisms, our study identifies PGK1 as the predominant metabolic effector downstream of the KDM5B-PTEN-Akt axis. Although PI3K/Akt signaling broadly enhances glycolytic flux, the high sensitivity of PGK1 to KDM5B-mediated epigenetic regulation and its essential role in providing the lactate substrate for specific hnRNPA1 lactylation position it as a critical and non-redundant node in this specific resistance model. Notably, the conventional view often considers KDM5B as a classic transcriptional regulator that, in non-small cell lung cancer, enhances radioresistance by downregulating PTEN [44] and, in gastric cancer, promotes cisplatin resistance by demethylating H3K4 to recruit XRCC1 [45]. However, the findings of this study emphasize that KDM5B induces metabolic reprogramming to provide “energy and substrate” support PCa enzalutamide resistance, suggesting that its functions reach far beyond the traditional understanding.

Metabolic reprogramming is a prominent feature of cancer, with metabolism playing a crucial role both upstream and downstream in regulating epigenetic modifications. Intermediate metabolites influence chromatin dynamics through chemical PTMs that change chromatin structure and function [46]. Lactate has long been recognized as a waste product of glycolysis [47]. However, increasing evidence indicates that lactate also has an important role in epigenetic modifications. As a novel PTM, lactylation plays a unique role in bridging cellular metabolism and epigenetic regulation by dynamically integrating metabolic substrates and their modifying enzymes, and it can be either beneficial or detrimental in both cancerous and non-cancerous disorders [48]. Xu et al. found that PKM2-driven lactate overproduction induces endothelial-mesenchymal transition of ischemic flap by mediating TWIST1 lactylation [49]. He et al. observed that HDAC2-mediated delactylation of METTL3 promotes DNA damage repair and chemoresistance in triple-negative breast cancer [50]. Our results suggest that KDM5B upregulates PGK1 in PCa cells, leading to lactate accumulation, which serves as a substrate for lactylation of lysine at site 179 of hnRNPA1. Previous research identified lactylation of hnRNPA1 in oral squamous cell carcinoma [28]. In this study, we further identified the lactylation site of hnRNPA1 using global lactylation proteomics, CRISPR/Cas9 gene editing, Co-IP, and molecular docking. Functional validation experiments demonstrated that mutating this site significantly reduced AR-V7 expression in the presence of lactate. This indicates that lactylation at the K179 site of hnRNPA1 is closely linked to enzalutamide resistance in PCa.

Additionally, the present study found that p300 functions as a lactylation “Writer” enzyme during the lactylation of hnRNPA1, while HDAC1 and HDAC2 act as lactylation “Eraser” enzymes. Contrary to previous findings, the study by Sun et al. revealed that MeCP2 lactylation protects against ischemic brain injury by inhibiting the transcription of genes involved in apoptosis, with p300 and HDAC3 as key enzymes regulating MeCP2 lactylation [51]. Deng et al. found that histone H4K12 lactylation enhances GCLC expression, thereby promoting chemoresistance in colorectal cancer stem cells by inhibiting ferroptosis. In this process, p300 and HDAC1 are critical enzymes that regulate H4K12 lactylation [52]. It follows that p300 acts as a lactoyltransferase involved in the lactylation process across many diseases [53, 54]. Structural analysis of p300 in complex with various acyl-CoA molecules has clarified how p300 catalyzes lysine modifications [55]. A review discussed the roles of lactoyltransferases and de-lactoylases in lysine lactonization, as well as potential strategies for treating human diseases [56]. This study also found that combining a p300 inhibitor, C646, further restores PCa sensitivity to enzalutamide. Therefore, co-administration of p300 inhibitors or activation of de-lactoylases might be a promising strategy to overcome enzalutamide resistance in PCa.

Alternative splicing (AS) is one of the core mechanisms of gene expression regulation that allows a single gene to produce different protein isoforms with distinct functions through various mRNA splicing processes. This dynamic regulation relies on the synergistic interactions between RNA-binding proteins (such as the SR and hnRNP protein families) and pre-mRNA, and it is influenced by multiple factors. Dysregulation of AS is closely associated with tumor progression and drug resistance [57–59]. In our study, lactylation of hnRNPA1 led to abnormal AR splicing, which increased AR-V7 expression. This specific form of AR does not require ligand binding to activate the transcription of AR target genes [60]. Therefore, AR-V7 is considered a primary contributor to enzalutamide resistance in PCa [61]. Most previous research has focused on changes in the transcription and translation levels of hnRNPA1, resulting in abnormal AR-V7 expression. Zhang et al. found that increasing the interaction between hnRNPA1 and AR could decrease PCa’s sensitivity to enzalutamide by increasing AR protein levels [62]. Chen et al. discovered that activation of the PAK1/RELA/hnRNPA1/AR-V7 axis caused cross-resistance to enzalutamide and darotamide during PCa progression [8]. Quercetin can reverse resistance in PCa cells to enzalutamide by downregulating hnRNPA1 expression, thereby reducing AR-V7 levels [63]. However, unlike previous studies, this research identified a novel lactate-induced PTM of hnRNPA1 that enhances its stability, ultimately leading to increased AR-V7 expression. While our study highlights hnRNPA1-K179 lactylation as a pivotal mechanism driving AR-V7 expression, it is important to recognize that splicing is a highly coordinated process involving multiple redundant factors such as SRSF1 and U2AF1. Our global proteomics identified hnRNPA1 as the most significantly lactylated component in the spliceosome under enzalutamide pressure, yet it is likely that this modification functions by modulating the recruitment or stability of a larger splicing complex. Future studies utilizing cross-linking immunoprecipitation (CLIP-seq) may further elucidate how K179 lactylation alters the global interaction landscape of hnRNPA1 with other spliceosomal partners. This study offers a new perspective on enzalutamide resistance in PCa. It is important to consider not only the “quantity” of the protein but also its “structure” [64].

The ligand-independent AR signaling pathway is the molecular mechanism by which ARs are activated in the absence of traditional androgen ligands. Using public ChIP-Seq data and databases like JASPAR, we identified an AR-binding site in the KDM5B promoter. ChIP-qPCR and dual-luciferase reporter assays showed that AR binds to the KDM5B promoter and promotes its transcription. Previous studies also indicate that AR, as a transcription factor, influences the progression and drug resistance of PCa. Wu et al. found that AR directly binds to the promoter region of lncRNA FLJ, enhancing its transcription. FLJ increases PCa cell resistance to enzalutamide by inhibiting nuclear AR signaling and the degradation of cytoplasmic proteins, thereby activating the androgen-independent AR signaling pathway [65]. MYO6 promotes tumor progression and enzalutamide resistance in CRPC by activating the focal adhesion signaling pathway, while AR can directly bind to the MYO6 promoter to promote its transcription [39] This study shows that KDM5B activates the PI3K/Akt signaling pathway by epigenetically suppressing PTEN, thereby causing AR phosphorylation, and then enters the nucleus to drive KDM5B transcription. This forms a novel positive feedback loop, “KDM5B-PTEN/PI3K/Akt-AR-KDM5B,” which differs from previous studies. It also suggests that inhibiting a single point may be ineffective. A multi-target combination strategy may be necessary to overcome PCa’s resistance to enzalutamide and disrupt the circuit. Furthermore, from a clinical perspective, integrative medical approaches, including traditional Chinese medicine-based therapies like acupuncture and moxibustion, have been increasingly explored to alleviate treatment-related side effects and improve the overall quality of life for cancer patients [66, 67].

Although our study offers promising insights, several limitations require further research. This study primarily explains the specific molecular mechanisms by which KDM5B promotes glycolysis and how hnRNPA1 lactylation influences AR-V7 through alternative splicing to contribute to Enza resistance. However, the roles of other differentially lactylated proteins remain largely unclear, calling for more investigation into these mechanisms. Additionally, the scRNA-seq used in this study, along with the chosen cell lines and animal models, does not fully capture the complex microenvironment and heterogeneity seen in PCa patients. Therefore, they cannot provide a complete understanding of the tumor microenvironment changes that drive Enza resistance. Lastly, while we showed that hnRNPA1 lactylation affects its function and stability, the structural changes in proteins caused by lactylation need further clarification in future research.

Conclusion

Overall, our research has uncovered a novel drug resistance mechanism: KDM5B activates the PI3K/Akt signaling pathway by epigenetically silencing PTEN, which then upregulates PGK1, reprogramming glucose metabolism in tumor cells and promoting lactate accumulation. Lactate enhances the expression of AR-V7 through P300-mediated lactylation of hnRNPA1. Notably, there is also a closed-loop mechanism involving “KDM5B-PTEN/PI3K/Akt-AR-KDM5B.” This study offers a new perspective on lactate produced by glycolysis in the context of Enza resistance in PCa. It not only expands the complex network of metabolism, epigenetics, and drug resistance but also suggests a potential therapeutic strategy for multi-target combined intervention to reverse Enza resistance in PCa.

Abbreviations

Enza

Enzalutamide

CRPC

Castration-resistant Prostate Cancer

PCa

Prostate Cancer

LNCaP-AI

Androgen-independent LNCaP cells

ChIP

Chromatin Immunoprecipitation

CLIP

Cross-Linking Immunoprecipitation

ARSIs

Androgen-Receptor Cignaling Inhibitors

AR

Androgen Receptor

AR-Vs

Androgen Receptor Splice Variants

SH

Sensitive High-lactylation

SL

Sensitive Low-lactylation

RH

Resistant High-lactylation

RL

Resistant low-lactylation

PPI

Protein-Protein Interaction

RIP

RNA Immunoprecipitation

WT

Wild-Type

CHX

Cycloheximide

HDACs

Histone Deglactoylases

TSA

Tricostatin A

NAM

Nicotinamide

DHT

Dihydrotestosterone

LBD

Ligand-Binding Domain

EnzR

Enzalutamide-Resistant

EnzS

Enzalutamide-Sensitive

scRNA-seq

single-cell RNA sequencing

PCA

Principal Component Analysis

UMAP

Uniform Manifold Approximation and Projection

LRGs

Lactylation-Related Genes

GSVA

Gene Set Variation Analysis

TCGA

The Cancer Genome Atlas

UCSC

University of California, Santa Cruz

GEO

Gene Expression Omnibus

GSEA

Gene Set Enrichment Analysis

GO

Gene Ontology

KEGG

Kyoto Encyclopedia of Genes and Genomes

WGCNA

Weighted Gene Co-expression Network Analysis

SDS

Sodium Dodecyl Sulfate

PVDF

Polyvinylidene Fluoride

IP

Immunoprecipitation

IC50

Half-maximal Inhibitory Concentration

CCK8

Cell Counting Kit-8

OCR

Oxygen Consumption Rate

ECAR

Extracellular Acidification Rate

PER

Proton Efflux Rate

Rot/AA

Rotenone/antimycin A

2-DG

2-deoxy-D-glucose

SPF

Specific Pathogen-Free

sgRNAs

single guide RNAs

DTT

Dithiothreitol

IAA

Iodoacetamide

TFA

Trifluoroacetic Acid

SD

Standard Deviation

mCRPC

metastatic Castration-Resistant Prostate Cancer

mHSPC

metastatic Hormone-Sensitive Prostate Cancer

AS

Alternative Splicing

Authors’ contributions

**Rui Sun: ** Conceptualization, Investigation, Data Curation, Writing-Original Draft. **Yong Huang: ** Formal Analysis, Validation, Visualization, Funding Acquisition, Writing-Original Draft. **Hao He: ** Methodology, Investigation, Resources, Writing-Original Draft. **Qiuchen Li: ** Conceptualization, Data Curation, Formal Analysis, Writing-Original Draft. **Linfeng Wang: ** Validation, Investigation, Writing-Review & Editing. **Gaojie Zhang: ** Resources, Investigation, Writing-Review & Editing. **Ziling Wei: ** Formal Analysis, Visualization, Writing-Review & Editing. **Yang Cao: ** Investigation, Data Curation, Funding Acquisition, Writing-Review & Editing. **Jing Li: ** Validation, Investigation, Writing-Review & Editing. **Xianmin Wang: ** Resources, Investigation, Writing-Review & Editing. **Fan Yang: ** Resources, Investigation, Writing-Review & Editing. **Wenjun Chen: ** Project Administration, Resources, Writing-Review & Editing; **Xiang Li: ** Formal Analysis, Writing-Review & Editing. **Jiang Yu, Siyuan Liu, Congfeng Lei: ** Investigation, Data Curation, Writing-Review & Editing. **Yu Jiang: ** Resources, Project Administration, Writing-Review & Editing. **Yueqiang Peng: ** Investigation, Writing-Review & Editing. **Huiyi Su: ** Writing-Review & Editing. **Yingying Gao: ** Supervision, Writing-Review & Editing, Project Administration. **Weiyang He: ** Supervision, Conceptualization, Writing-Review & Editing. **Lei Yang: ** Supervision, Methodology, Writing-Review & Editing. **Jiayu Liu: ** Supervision, Conceptualization, Funding Acquisition, Project Administration, Writing-Review & Editing.

Funding

This work was supported by the Ph.D. Train Through Research Project of Chongqing, China (CSTB2022BSXM-JCX0061), the Chongqing Postgraduate Scientific Research Innovation Project (CYS240289), and the National Training Program of Innovation and Entrepreneurship for Undergraduates (202510631006).

Data availability

The data that support the findings of this study are available from the corresponding author upon reasonable request.

Declarations

Ethics approval and consent to participate

Written informed consent was obtained from all patients, and their personal information was kept confidential. This study was approved by the Medical Ethics Committee of the First Affiliated Hospital of Chongqing Medical University (Ethic number: 2021 − 608).

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.