Adipocyte-derived β-hydroxybutyrate confers metabolic resilience in multiple myeloma

Abstract

Multiple myeloma (MM) remains an incurable blood cancer. Obesity is a known risk factor, but how adipocytes promote MM progression is not fully understood. Here, we uncover a metabolic crosstalk between adipocytes and MM cells that promotes MM cell survival under glucose deprivation. We show that glucose restriction activates AMPK, disrupting HSP90-IRF4 binding and rendering IRF4 susceptible to TRIM21-mediated proteasomal degradation. Paradoxically, the same stress stimulates adipocytes to produce β-hydroxybutyrate (β-OHB). MM cells utilize β-OHB through OXCT1-mediated ketolysis, fueling NAT10-dependent acetylation of IRF4 at K87, which restores IRF4-HSP90 binding and sustains tumor cell survival. Genetic ablation of the rate-limiting ketogenic enzyme Hmgcs2 in adipocytes abrogates this protective effect. Importantly, combining an AMPK activator (metformin) with an OXCT1 inhibitor (pimozide) or a NAT10 inhibitor (remodelin) shows synergistic anti-tumor activity in vivo. Our findings position adipocyte-derived β-OHB as a critical metabolic adaptor and highlight a potential combination therapy for MM.

Multiple cell populations within the bone tumour microenvironment have been shown to support multiple myeloma (MM). Here, the authors show that under glucose deprived conditions, adipocytes produce β-hydroxybutyrate, which promotes MM cell survival.

Introduction

Multiple myeloma (MM) is an incurable hematologic malignancy characterized by the clonal proliferation of plasma cells, representing the second most common hematologic cancer after chronic lymphocytic leukemia^1,2. Despite advancements in therapeutic strategies, the median survival for myeloma patients has stagnated at 5 to 7 years, and a complete cure remains a formidable challenge. As a disease primarily affecting middle-aged and elderly individuals, MM has been strongly linked to obesity in epidemiological studies^3–5. Notably, both aging and obesity, key risk factors for MM, are concomitant with impaired bone marrow osteogenic regeneration and adipose tissue expansion in the bone marrow^6,7, making adipocytes one of the most abundant stromal components in the myeloma microenvironment.

We previously demonstrated that myeloma cells epigenetically reprogram adipocytes, prompting them to secrete adipokines that activate osteoclastogenesis and suppress osteoblastogenesis, thereby contributing to myeloma-induced bone disease⁸. Moreover, under obese conditions, adipocyte-derived angiotensin II (Ang II) impinges on acetyl-CoA synthetase 2 (ACSS2) in myeloma cells, accelerating disease progression⁹. Despite being recognized as a critical accomplice in multiple myeloma pathogenesis¹⁰, the precise roles of adipocytes at different disease stages and their therapeutic implications remain incompletely understood.

Metabolic reprogramming is one of the hallmarks of cancer and has emerged as an attractive therapeutic target for cancer treatment¹¹. A fundamental question in cancer metabolism is how cancer cells adapt to and survive metabolically adverse conditions such as nutrient deficiency, and whether these adaptive mechanisms can be exploited as vulnerabilities for therapeutic intervention^12,13. Nutrient deficiency, caused by excessive tumor growth, low glycaemic diet, ketogenic diet or intermittent fasting, is often presumed to restrict tumor growth and survival by reducing glucose availability in the tumor microenvironment^14,15. However, the potential contribution of adipocytes to the metabolic stress resilience of myeloma cells remains largely unknown.

To investigate the role of adipocytes under glucose metabolic stress, we employed 2-deoxy-D-glucose (2-DG), a glucose analog that disrupts glycolysis, to model glucose deprivation (GD) in vivo^16,17. Here, we report that adipocytes significantly enhance myeloma cell resistance to 2-DG-induced metabolic stress across multiple mouse models. This effect is mediated by adipocyte-derived β-hydroxybutyrate (β-OHB). Mechanistically, myeloma cells utilize β-OHB to activate the 3-oxoacid CoA-transferase 1 (OXCT1) and N-acetyltransferase 10 (NAT10) axis, which promotes interferon regulatory factor 4 (IRF4) acetylation and prevents its tripartite motif-containing protein 21 (TRIM21)-mediated degradation under GD conditions. Notably, combining ketolysis inhibition with metabolic stress induction synergistically suppresses tumor growth both in vitro and in vivo. Our study thus reveals an adipocyte-myeloma metabolic crosstalk mediated by β-OHB, uncovering a druggable pathway to overcome metabolic resilience in multiple myeloma.

Results

Adipocytes confer resistance to 2-DG-induced metabolic stress in myeloma cells in vivo

To explore the impact of adipocytes on myeloma progression under metabolic stress, we used a high-fat diet-induced obesity (DIO) model in male C57BL/6J mice. The DIO mice were fed with a 60% kcal high-fat diet for 20 weeks to induce obesity, while control mice received a standard 10% kcal normal diet (ND) (Fig. 1a, b). Myeloma was established by inoculating murine Vk12598 cells into the femurs of both DIO and ND mice. To model glucose deprivation (GD) in vivo, 2-DG was administered to mice one-week post-tumor inoculation. Compared to ND controls, DIO mice developed a significantly higher tumor burden, as indicated by elevated serum M-protein levels (Fig. 1c), increased infiltration of CD138⁺ myeloma cells in the bone marrow and spleen (Fig. 1d–g and Supplementary Fig. 1e), and pronounced splenomegaly (Fig. 1f). While 2-DG treatment markedly reduced tumor burden in ND mice, this effect was significantly attenuated in DIO mice (Fig. 1c–g), suggesting that obesity confers resistance to 2-DG-induced metabolic stress in myeloma.

Fig. 1. Adipocytes promote myeloma cell resistance to 2-DG-induced metabolic stress in vivo.

a–g Male C57BL/6J mice were fed high-fat diet (DIO) or normal diet (ND) for 20 weeks, injected with Vk12598 cells, then treated with 2-DG for 4 weeks (n = 5 mice per group). a Experimental timeline. b Body weight changes. c Serum IgG2b levels. d Representative flow cytometry plots of bone marrow CD138⁺ myeloma cells. e Quantification of bone marrow CD138⁺ cells (n = 3 mice per group). f Representative spleen images. g Quantification of splenic CD138⁺ cells (n = 3 mice per group). h–l ARP1 cells alone or co-injected with human adipocytes (ADs) were subcutaneously implanted into male NCG mice, followed by 2-DG treatment. h Tumor growth curves (n = 5 mice per group). i Representative images of excised tumors. j Tumor weights (n = 5 tumors per group). k Immunohistochemical staining of tumor tissues for Ki67 (proliferation marker), CD138 (marking myeloma cells), perilipin (adipocyte marker), and p-AMPK. Scale bars: 100 µm. l Quantification of Ki67 (n = 10 areas per sample) and p-AMPK (n = 5 areas per sample). Data are presented as mean ± SD. p values were determined by two-way ANOVA (b, c, e, g, h, j, l). Source data are provided as a Source Data file.

As a hallmark feature of obesity, excessive adipose tissue accumulation creates a unique tumor microenvironment⁶. To investigate the specific contribution of adipocytes to 2-DG resistance, we employed a co-implantation model in which ARP1 myeloma cells were injected subcutaneously with mature adipocytes into one flank of male NCG mice, while ARP1 cells alone were implanted contralaterally as an internal control (Fig. 1h). Following tumor engraftment, mice received daily intraperitoneal injections of either 2-DG or vehicle. 2-DG treatment significantly inhibited the progression of ARP1-alone tumors, as evidenced by reduced tumor growth kinetics (Fig. 1h), decreased final tumor sizes (Fig. 1i), and lower excised tumor weights (Fig. 1j). Notably, the anti-tumor effect of 2-DG was abolished upon co-implantation with adipocytes (Fig. 1h–j). Immunohistochemical analysis further showed that although 2-DG treatment induced metabolic stress, as evidenced by elevated p-AMPK levels, its anti-proliferative effect (reduced Ki67 staining) was reversed in the presence of adipocytes (Fig. 1k, l).

To further validate adipocyte-mediated metabolic stress resistance, we established two complementary intrafemoral models: (1) luciferase-tagged ARP1 cells co-injected with adipocytes into male NCG mice, and (2) luciferase-tagged 5TGM1 cells with adipocytes into male Tcrb^tm1Mom/J mice (Supplementary Fig. 1a, c). Compared to vehicle controls, 2-DG treatment significantly reduced bioluminescent signal at sites implanted with myeloma cells alone. In contrast, adipocyte co-implantation markedly enhanced myeloma cell resistance to 2-DG treatment (Supplementary Fig. 1a–d). Collectively, these findings indicate that an adipocyte-rich microenvironment confers intrinsic protection against 2-DG-induced metabolic stress in myeloma.

Adipocyte-derived β-hydroxybutyrate confers resistance to glucose metabolic stress in vitro and in vivo

To explore the mechanism by which adipocytes promote myeloma cell resistance to glucose metabolic stress, ARP1 or 5TGM1 cells were cocultured with adipocytes under GD conditions using a transwell system to mimic the in vivo microenvironment (Fig. 2a). In line with our in vivo findings, co-culture with adipocytes significantly improved the survival of both ARP1 and 5TGM1 cells under GD conditions (Fig. 2a). Furthermore, GD substantially reduced the protein levels of interferon regulatory factor 4 (IRF4), a key survival factor in myeloma¹⁸. Notably, this reduction in IRF4 expression was reversed by co-culture with adipocytes (Fig. 2b), suggesting IRF4 as a key mediator of adipocyte-conferred metabolic resistance.

Fig. 2. Adipocyte-derived β-OHB protects myeloma cells from glucose metabolic stress.

a Left: Schematic of transwell coculture system using human MSC-derived adipocytes (ADs) with ARP1 cells or murine OP9-derived ADs with 5TGM1 cells. Right: Time-course viability of myeloma cells with or without ADs coculture under glucose deprivation (GD) conditions (n = 4 technical replicates; experiment independently repeated three times with similar results). b Western blot of IRF4 in ARP1 and 5TGM1 cells after 16-h coculture with ADs under GD conditions. c Western blot of IRF4 in myeloma cells treated with adipocyte-conditioned medium (AD-CM) under GD conditions, with or without proteinase K or size fractionation (<3 kDa or >3 kDa). d–f Metabolomic analysis of AD-CM under GD vs. normal glucose. d Volcano plot of differentially abundant metabolites. e Heatmap of top regulated metabolites. f Absolute quantification of β-hydroxybutyrate (β-OHB) in AD-CM (n = 5 independent experiments). g Western blot of IRF4 in myeloma cells treated with β-OHB (0.5, 1.0 mM) or acetoacetate (AcAc) (1.0 mM) under GD conditions for 16 h. h Time-course viability of myeloma cells with or without β-OHB under GD conditions (n = 4 technical replicates; experiment independently repeated three times with similar results). i Western blot of IRF4 in myeloma cells after 16-h coculture with wild-type (AD^WT) or Hmgcs2-deficient (AD^KO) adipocytes under GD conditions. j, k 5TGM1 cells alone or with AD^WT or AD^KO were subcutaneously injected into nude mice, followed by 2-DG treatment. j Representative excised tumors. k Tumor weights (n = 5 mice/group). l–o Male Hmgcs2^f/f and Hmgcs2^AKO mice on normal diet (ND) or high-fat diet (DIO) were inoculated with luciferase-tagged Vk12598 cells and treated with 2-DG. l Representative bioluminescence images. m Quantification of bioluminescent signal (n = 5 mice/group). n Serum M-protein levels (n = 5 mice/group). o Gross morphological images of livers and spleens. Data are presented as mean ± SD. p values were determined by unpaired two-tailed Student’s t test (f) and two-way ANOVA (a, h, k, m, n). Images in b, c, g, i are representative of three independent experiments. Source data are provided as a Source Data file.

To pinpoint the responsible factors from adipocytes, conditioned medium from adipocytes (AD-CM) was collected from adipocytes cultured under GD conditions. AD-CM was treated with proteinase K (ProK) or fractionated by molecular weight cutoff filtration and applied to ARP1 or 5TGM1 cells under GD conditions. AD-CM rescued GD-induced IRF4 reduction in myeloma cells, an effect that persisted after ProK treatment, thus ruling out proteins as the active components in AD-CM (Fig. 2c). Intriguingly, the <3 kDa fraction of AD-CM, but not the >3 kDa fraction, recapitulated the effects of unfractionated AD-CM, suggesting that small non-protein metabolites mediate the activity of AD-CM on IRF4 protein (Fig. 2c).

Metabolomic profiling of AD-CM under GD versus normal glucose conditions revealed 286 differentially regulated metabolites, comprising 167 downregulated and 119 upregulated metabolites (Fig. 2d). A heatmap of the most significantly altered metabolites is shown in Fig. 2e. Notably, β-hydroxybutyrate (β-OHB), the predominant ketone body typically synthesized in the liver during metabolic stress^19,20, was significantly elevated in AD-CM under GD conditions, with absolute concentrations reaching approximately 0.5 mM (Fig. 2f). Intriguingly, both β-OHB and acetoacetate (AcAc, another unstable ketone body) restored GD-induced IRF4 reduction in ARP1 and 5TGM1 cells, with 0.5 mM β-OHB being sufficient to confer this protection (Fig. 2g). Furthermore, β‑OHB significantly enhanced the survival of both cell lines under GD conditions (Fig. 2h), indicating that adipocyte-derived β-OHB protects myeloma cells from metabolic stress.

To examine the relationship between GD and ketone body production in adipocytes, we analyzed the expression of ketogenic enzymes in adipocytes (Supplementary Fig. 2a). mRNA levels of ACAT1, BDH1, HMGCL, and HMGCS2 were significantly upregulated upon GD or 2-DG treatments (Supplementary Fig. 2b). To extend these findings in vivo, we examined the regulation of these enzymes in adipose tissues under metabolic stress. Consistent with studies showing that caloric restriction induces browning of white adipose tissue (WAT)²¹, alternate-day fasting (ADF) in male C57BL/6J mice promoted a browning-like phenotype in both epididymal and inguinal white adipose tissue (eWAT and iWAT), as evidenced by increased Ucp-1 expression (Supplementary Fig. 2c, e, g, h). ADF also upregulated ketogenic enzyme expression in iWAT and eWAT (Supplementary Fig. 2d, f). Using perilipin-1 and Ucp-1 as markers, we found that under basal conditions WAT exhibits low Hmgcs2 and Ucp-1 levels. Upon ADF-induced metabolic stress, however, Ucp-1 and Hmgcs2 were coordinately upregulated, suggesting that ketogenic capacity is acquired upon beige adipocyte transition (Supplementary Fig. 2i).

The ketone body concentrations were also significantly elevated in iWAT, eWAT, and BAT of ADF mice, reaching approximately one-third of the hepatic levels (Supplementary Fig. 2j). In line with these observations, bioinformatic analysis of public GEO datasets (GSE199963 and GSE7623) further revealed that fasting induces upregulation of Hmgcs2 mRNA levels in adipocytes (Supplementary Fig. 2k, l). Treatment with 2-DG similarly increased Hmgcs2 protein levels in liver and adipose tissues (Supplementary Fig. 2m), confirming that glucose metabolic stress enhances ketogenic capacity in adipocytes.

To assess the functional role of adipocyte‑derived β‑OHB in metabolic stress resistance, we generated adipocyte-specific Hmgcs2 knockout (Hmgcs2^AKO) mice by crossing Hmgcs2^flox/flox mice (Hmgcs2^f/f) with Adipoq-Cre transgenic mice (Supplementary Fig. 2n). Western blot analysis confirmed Hmgcs2 deletion in eWAT, iWAT, and BAT in Hmgcs2^AKO mice (Supplementary Fig. 2o). Adipocytes isolated from both Hmgcs2^f/f and Hmgcs2^AKO mice were then co-cultured with myeloma cells under GD conditions. Wild-type adipocytes (AD^WT) significantly attenuated GD-induced IRF4 reduction in both ARP1 and 5TGM1 cells, whereas this protective effect was substantially compromised in Hmgcs2-deficient adipocytes (AD^KO) (Fig. 2i).

For in vivo validation, we first subcutaneously implanted 5TGM1 cells either alone or together with an equal number of AD^WT or AD^KO into nude mice. After tumor establishment, mice were treated with 2-DG. Co-implantation with AD^WT significantly rescued 2-DG-induced tumor growth suppression, whereas this protective effect was markedly attenuated in the AD^KO group, as demonstrated by tumor sizes (Fig. 2j) and tumor weights (Fig. 2k). We next established a DIO model using male Hmgcs2^f/f and Hmgcs2^AKO mice. Myeloma was initiated by femoral inoculation of luciferase-tagged Vk12598 cells, followed by 2-DG administration. Consistent with the previous results, Hmgcs2^f/f-DIO mice exhibited significant resistance to 2-DG treatment compared with Hmgcs2^f/f-ND mice. By contrast, this resistance was markedly attenuated in Hmgcs2^AKO-DIO mice, as evidenced by the weakened bioluminescent signal (Fig. 2l, m), reduced serum M-protein levels (Fig. 2n), and regressed hepatic and splenic sizes (Fig. 2o).

To further explore the contribution of adipocyte-derived ketones in vivo, we generated liver-specific Hmgcs2 knockout mice (Hmgcs2^LKO) (Supplementary Fig. 2p). In this model, where hepatic ketogenesis is absent, myeloma cells were co-implanted intrafemorally with either AD^WT or AD^KO (Supplementary Fig. 2q). Supporting a local, paracrine role, the protective effect of adipocytes was significantly attenuated in the AD^KO group upon 2-DG treatment (Supplementary Fig. 2r, s). Consistent with these findings, we found that ADF or 2-DG treatment triggered a significant increase in circulating ketone bodies in Hmgcs2^f/f mice, a response that was substantially blunted in Hmgcs2^AKO mice (Supplementary Fig. 2t), suggesting that adipocytes are a critical source of ketone bodies under metabolic stress. Together, these results demonstrate that glucose metabolic stress induces ketone body production in adipocytes, which in turn protects myeloma cells against metabolic stress.

Glucose deprivation induces IRF4 degradation via disruption of the IRF4-HSP90 interaction

To elucidate the mechanism whereby adipocyte-derived β-OHB protects myeloma cell survival under glucose metabolic stress, we first investigated how GD impaired IRF4 expression in myeloma cells. IRF4 mRNA and protein levels were assessed in human (ARP1, RPMI8226, ARH77) and murine (5TGM1) myeloma cells exposed to GD or 2-DG for varying durations. Both treatments significantly reduced IRF4 protein levels without affecting its mRNA levels across all tested cell lines (Supplementary Fig. 3a–d). Furthermore, glucose deficiency decreased IRF4 protein in a concentration-dependent manner, again without altering mRNA levels (Supplementary Fig. 3e, f). These findings suggest that GD and 2-DG regulate IRF4 through post-transcriptional mechanisms.

Cycloheximide (CHX)-chase assays showed that IRF4 protein degradation was accelerated in cells treated with GD or 2-DG compared with controls (Fig. 3a–d). To identify the degradation pathways involved, we treated HA-IRF4-expressing myeloma cells under GD conditions with inhibitors of specific protein degradation pathways: MG-132 and bortezomib (BTZ; proteasome inhibitors), leupeptin (a lysosomal protease inhibitor), and MDL-28170 (a calpain inhibitor). Only the proteasome inhibitors MG-132 and BTZ effectively prevented GD-induced IRF4 degradation (Fig. 3e), indicating that GD destabilizes IRF4 via the ubiquitin-proteasome pathway. Consistently, GD stress induced pronounced K48-linked polyubiquitination of IRF4 in myeloma cells (Fig. 3f), the canonical ubiquitin code that directs substrates to the 26S proteasome for degradation²².

Fig. 3. GD promotes IRF4 degradation via disruption of IRF4-HSP90 interaction.

a Time-course of IRF4 protein in ARP1 and RPMI8226 cells treated with 50 µg/mL cycloheximide (CHX) under normal glucose (Glu, 10 mM) or GD. b Quantification of IRF4 protein normalized to GAPDH (n = 3 technical replicates; experiment independently repeated three times with similar results). c IRF4 protein kinetics in myeloma cells treated with or without 2-DG (10 mM) in the presence of CHX. d Quantified IRF4 protein (n = 3 technical replicates; experiment independently repeated three times with similar results). e IRF4 protein in myeloma cells after 16-h GD treatment in the presence or absence of MG-132 (1 μM), bortezomib (BTZ, 50 nM), leupeptin (10 μM), or MDL-28170 (1 μM). f HA-IRF4-expressing myeloma cells were treated with GD for 16 h (MG-132 added for final 6 h). Lysates were immunoprecipitated using anti-HA affinity resin and immunoblotted for K48-ubiquitin and HA-IRF4. g Mass spectrometry of HA-IRF4 immunoprecipitates from HA-IRF4-expressing RPMI8226 cells under normal glucose or GD conditions. Identified interactors are listed in the table. h Co-immunoprecipitation (Co-IP) of HA-IRF4 with endogenous HSP90 in HA-IRF4-expressing ARP1 cells under GD or 2-DG treatment. i Representative confocal images of IRF4-GFP (green) and endogenous HSP90 (red) under normal glucose or GD. Nuclei stained with DAPI (blue). Scale bar: 5 μm. j Fluorescence intensity quantification of the demarcated regions (white lines) using ImageJ software. k Time-course of IRF4 protein in myeloma cells treated with CHX with or without 17-AAG. Quantification shown below (n = 3 technical replicates; experiment independently repeated three times with similar results). l HA-IRF4-expressing myeloma cells were treated with or without 17-AAG for 16 h (MG-132 added for final 6 h), followed by HA immunoprecipitation and K48-ubiquitin blot. Data are presented as mean ± SD. p values were determined by two-way ANOVA (b, d, k). Images in a, c, e, f, h, i, k, l are representative of three independent experiments. Source data are provided as a Source Data file.

To further dissect the mechanism underlying GD-induced IRF4 degradation, we performed comprehensive interactome profiling using HA-IRF4-expressing RPMI8226 cells cultured under either normal glucose or GD conditions. Anti-HA immunoprecipitation coupled with quantitative mass spectrometry (IP-MS) revealed decreased HSP90 binding to IRF4 under GD, along with increased interactions with the E3 ubiquitin ligases TRIM21 and FBXW11 (Fig. 3g). Co-immunoprecipitation (Co-IP) assays in HA-IRF4-expressing ARP1 cells confirmed that, under normal glucose conditions, HA-IRF4 formed a complex with endogenous HSP90; this interaction was abolished upon GD or 2-DG treatment (Fig. 3h). Similarly, in HEK293T cells co-transfected with Flag-HSP90 and HA-IRF4, GD or 2-DG disrupted the interaction between Flag-HSP90 and HA-IRF4 (Supplementary Fig. 3g). Immunofluorescence staining further demonstrated nuclear co-localization of HSP90 with IRF4-GFP under normal glucose conditions, whereas GD stress triggered nuclear-to-perinuclear redistribution of HSP90 and disrupted its co-localization with IRF4-GFP (Fig. 3i, j).

To assess the functional role of HSP90 in maintaining IRF4 stability, we employed both pharmacological and genetic inhibition strategies. Treatment with the HSP90 inhibitor 17-AAG or shRNA-mediated HSP90 knockdown significantly reduced IRF4 protein levels across different myeloma cell lines (Supplementary Fig. 3h, i). CHX-chase assays demonstrated that HSP90 inhibition dramatically accelerated IRF4 degradation in ARP1, RPMI8226, ARH77 and 5TGM1 cells (Fig. 3k and Supplementary Fig. 3j, k). Importantly, 17-AAG treatment increased K48-linked ubiquitination of IRF4 in ARP1 and RPMI8226 cells (Fig. 3l). Collectively, these results demonstrate that HSP90-IRF4 interaction stabilizes IRF4 by protecting it from ubiquitin-proteasome-mediated degradation, and this protective interaction is disrupted under GD conditions.

Glucose deprivation impairs HSP90-IRF4 interaction via AMPK activation

Given the central role of AMP-activated protein kinase (AMPK) in coordinating cellular energy homeostasis^23,24, we investigated whether AMPK activation mediates GD-induced IRF4 degradation. Treatment of ARP1 and RPMI8226 cells with the AMPK activators A-769662 or metformin (Met) dose-dependently increased AMPK phosphorylation and reduced IRF4 protein levels (Fig. 4a). CHX-chase assays showed that A-769662 or Met accelerated IRF4 degradation (Fig. 4b, c), recapitulating the effects of GD or 2-DG treatment. Notably, the AMPK inhibitor dorsomorphin (Dorso) reversed GD-induced IRF4 loss and restored IRF4 stability under GD conditions (Fig. 4d–f), suggesting that AMPK activation results in IRF4 degradation.

Fig. 4. Glucose deprivation impairs HSP90-IRF4 interaction through AMPK activation.

a Immunoblot of IRF4, p-AMPK, and AMPK in myeloma cells treated with A-769662 or metformin (Met) for 16 h. The samples derive from the same experiment but different gels for IRF4 and GAPDH, another for p-AMPK and another for AMPK were processed in parallel. The similar strategy applies to (d). b Immunoblot of HA-IRF4 in HA-IRF4-expressing cells treated with CHX plus A-769662 or Met. c Quantification of HA-IRF4 protein normalized to GAPDH (n = 3 technical replicates; experiment independently repeated three times with similar results). d Immunoblot of IRF4, p-AMPK, and AMPK in myeloma cells under normal glucose or GD with or without dorsomorphin (Dorso) for 16 h. e Degradation kinetics of HA-IRF4 in ARP1 cells treated with Dorso (4 µM) and CHX (50 µg/mL) under GD. f Quantification of HA-IRF4 protein (n = 3 technical replicates; experiment independently repeated three times with similar results). g Co-IP of HA-IRF4 with endogenous HSP90 in cells treated with A-769662 or Met. The IP samples derive from the same experiment but different gels for Ubiquitin-K48/HA-IRF4 and HSP90 were processed in parallel. The same applies to (h). h Co-IP of HA-IRF4 with endogenous HSP90 under normal glucose or GD with or without Dorso. i Immunofluorescence of IRF4-GFP (green) and HSP90 (red) in cells treated as indicated. Nuclei were stained with DAPI (blue). Scale bar: 5 μm. j Co-IP of Strep-HSP90 with individual AMPK α, β, or γ subunits in HEK293T cells. k In vitro kinase assay using purified Flag-AMPKα/β/γ complex and Strep-HSP90 (WT or 4A mutant). Phosphorylation detected with phospho-Ser/Thr antibody. l Nucleocytoplasmic fractionation of WT, phosphomimetic (4E), and phosphorylation-deficient (4A) HSP90 mutants in ARP1 cells after GD for 12 h. m Co-IP of Strep-tagged WT, 4E and 4A HSP90 mutants with endogenous IRF4 in ARP1 cells under GD for 12 h. n Immunohistochemical staining of IRF4 and p-AMPK in myeloma patient biopsies. Scale bars: 100 µm. o Correlation between p-AMPK and IRF4 in myeloma patients (n = 26 patients). Data are presented as mean ± SD. p values were determined by two-way ANOVA (c, f) and two-sided Spearman’s correlation analysis (o). Images in a, b, d, e, g–m are representative of three independent experiments. Source data are provided as a Source Data file.

Co-IP assays in HA-IRF4-expressing myeloma cells demonstrated that AMPK activation increased K48-linked ubiquitination of IRF4 and disrupted its interaction with HSP90, whereas AMPK inhibition attenuated GD-induced IRF4 ubiquitination and restored the HSP90-IRF4 interaction (Fig. 4g, h). Immunofluorescence staining further corroborated these findings: A-769662 or Met treatment mimicked GD-induced nuclear-to-perinuclear redistribution of HSP90 and reduced HSP90-IRF4 co-localization, while AMPK inhibition largely reversed these effects (Fig. 4i). Consistent with endogenous HSP90 dynamics, GD induced nuclear exclusion of ectopically expressed Strep-tagged HSP90 in ARP1 and HEK293T cells (Supplementary Fig. 4a).

To further investigate the molecular mechanism by which AMPK regulates the subcellular redistribution of HSP90, we co-transfected HEK293T cells with Strep-tagged HSP90 and the individual AMPK α, β, or γ subunits. Co-immunoprecipitation assays showed that Strep-HSP90 interacts with all three AMPK subunits (Fig. 4j). We next treated ARP1 cells expressing Strep-HSP90 with the AMPK activator A-769662. Subsequent Strep-tag immunoprecipitation, SDS-PAGE separation of the Strep-HSP90 protein band (Supplementary Fig. 4b), and mass spectrometric analysis identified four phosphorylation sites on HSP90: Ser226, Ser255, Thr285, and Ser445 (Supplementary Fig. 4c). We generated HSP90 mutants in which these serine and threonine residues were substituted with glutamic acid to mimic a phosphorylated state (4E) or with alanine to create a phosphorylation-deficient mutant (4A). To directly establish AMPK as the upstream kinase responsible for HSP90 phosphorylation, we performed an in vitro kinase assay using purified Flag-AMPKα/β/γ protein complex and Strep-HSP90. Incubation with the active AMPK complex directly and significantly increased phosphorylation of wild-type HSP90 (WT), whereas this phosphorylation was markedly attenuated in the phosphorylation-deficient mutant (4A) (Fig. 4k). These results provide direct biochemical evidence that HSP90 is a substrate of AMPK.

Immunofluorescence and cellular fractionation assays demonstrated that under GD stress, both WT HSP90 and the phosphomimetic 4E mutant exhibited predominant cytoplasmic localization. In contrast, the phosphorylation-deficient 4A mutant was retained in the nucleus (Fig. 4l and Supplementary Fig. 4d). Corroborating these findings, Co-IP assays showed that, unlike the WT and 4E mutants, the HSP90-4A mutant preserved its binding to IRF4 (Fig. 4m). These results collectively indicate that AMPK-mediated phosphorylation of HSP90 at these specific sites triggers its subcellular relocalization, thereby disrupting the protective HSP90-IRF4 interaction.

To validate the clinical relevance of these findings, we performed IHC staining for p-AMPK and IRF4 in myeloma bone marrow biopsies. A strong negative correlation was observed between IRF4 protein levels and p-AMPK levels (Fig. 4n, o), supporting our in vitro findings that AMPK activation negatively regulates IRF4 stability in myeloma. Taken together, our results demonstrate that metabolic stress-induced AMPK activation drives HSP90 subcellular redistribution and subsequent dissociation from IRF4, establishing AMPK as a promising therapeutic target for modulating IRF4 in multiple myeloma.

TRIM21 is the E3 ligase mediating IRF4 degradation under glucose metabolic stress

Our proteomic analysis revealed GD-dependent recruitment of two E3 ubiquitin ligases to IRF4: TRIM21 and FBXW11 (Fig. 3g). To determine their roles in regulating IRF4 stability, myeloma cells were transduced with lentiviral shRNAs targeting non-targeted control (shCtrl), TRIM21 (shTRIM21), or FBXW11 (shFBXW11). FBXW11 depletion did not affect IRF4 levels under either normal glucose or GD condition (Supplementary Fig. 5a, b). In contrast, TRIM21 knockdown did not alter IRF4 levels under normal glucose but significantly attenuated GD-induced IRF4 degradation (Fig. 5a). CHX-chase assays confirmed that TRIM21 deletion markedly slowed IRF4 degradation under GD conditions (Fig. 5b, c). Furthermore, TRIM21 knockdown abolished, while its overexpression enhanced, GD-induced K48-linked ubiquitination of IRF4 (Fig. 5d, e), identifying TRIM21 as an E3 ligase responsible for IRF4 degradation. Analysis of IRF4 and TRIM21 protein levels in a panel of nine myeloma cell lines revealed a strong negative correlation (Fig. 5f, g), further supporting TRIM21’s role in negatively regulating IRF4.

Fig. 5. TRIM21 mediates IRF4 degradation under glucose metabolic stress.

a Immunoblot of IRF4 and TRIM21 in ARP1 or RPMI8226 cells expressing shCtrl, shTRIM21-#1, or shTRIM21-#2, under normal glucose or GD conditions. The samples derive from the same experiment but different gels for TRIM21/GAPDH/Vinculin and IRF4 were processed in parallel. The similar strategy applies to (d, e, f, h, i, j, l, m). b Time-course of IRF4 protein in shCtrl, shTRIM21-#1, or shTRIM21-#2-expressing cells treated with CHX (50 µg/mL). c Quantification of IRF4 protein normalized to GAPDH (n = 3 technical replicates; experiment independently repeated three times with similar results). d HA-IRF4-expressing ARP1 cells were infected with shCtrl or shTRIM21 and cultured under GD conditions for 16 h. Anti-HA immunoprecipitates were probed for K48-ubiquitin and HA-IRF4. e HA-IRF4-expressing ARP1 cells were infected with vector or Flag-TRIM21 and cultured under GD conditions for 16 h. HA-IPs were analyzed for K48-ubiquitin and HA-IRF4. f Immunoblot of IRF4 and TRIM21 in various myeloma cell lines. g Correlation of TRIM21 and IRF4 protein levels across myeloma cells (n = 9 cell lines). h Co-IP of HA-IRF4 with endogenous TRIM21 in ARP1 cells cultured under normal glucose or GD conditions for 16 h. i Co-IP of HA-IRF4 with endogenous TRIM21 in ARP1 cells treated with 17-AAG (2 μM) or A-769662 (120 μM) for 16 h. j HA-IP of HEK293T cells co-expressing HA-IRF4, Strep-TRIM21, and gradient concentrations of Flag-HSP90. k Domain structure of IRF4 truncation mutants. DBD DNA-binding domain, LKD linker domain, IAD IRF association domain, AR Association Region. Co-IP of HA-IRF4 mutants with Flag-HSP90 (l) or Strep-TRIM21 (m) in HEK293T cells under GD conditions. n Proposed Model: Metabolic stress activates AMPK, inducing HSP90 redistribution, disrupting HSP90-IRF4 interaction, and promoting TRIM21-mediated IRF4 degradation. Data are presented as mean ± SD. p values were determined by two-way ANOVA (c) and Spearman’s correlation analysis (g). Images in a, b, d, e, h–j, l, m are representative of three independent experiments. Source data are provided as a Source Data file.

To elucidate why TRIM21-mediated IRF4 degradation occurs specifically under GD conditions, we performed Co-IP assays to assess IRF4-TRIM21 interactions. GD significantly enhanced IRF4-TRIM21 binding, an effect mimicked by AMPK activation with A-769662 or HSP90 inhibition with 17-AAG (Fig. 5h, i). Since AMPK activation disrupts the HSP90-IRF4 interaction, we hypothesized that HSP90 binding to IRF4 sterically hinders TRIM21 access. Co-transfection of HEK293T cells with fixed amounts of Strep-TRIM21 and HA-IRF4 alongside increasing amounts of Flag-HSP90 demonstrated that elevated Flag-HSP90 expression progressively weakened the Strep-TRIM21-HA-IRF4 interaction (Fig. 5j), indicating competitive binding between HSP90 and TRIM21 for IRF4.

To map the interaction regions of IRF4 with HSP90 and TRIM21, a series of truncated IRF4 mutants spanning its functional domains were generated (Fig. 5k) and co-expressed with Flag-HSP90 or Strep-TRIM21 in HEK293T cells. Co-IP assays revealed that IRF_1-420, IRF_1-238, IRF_1-139, and IRF4_139-238 interacted with Strep-TRIM21, while IRF4_1-420, IRF4_1-238, IRF4_139-238, and IRF4_238-420 bound Flag-HSP90 (Fig. 5l, m). These results pinpoint the IRF4_139-238 (LKD domain) as the shared binding site for both HSP90 and TRIM21. This conclusion was supported by AlphaFold-based structural predictions, which revealed a common binding domain for HSP90 and TRIM21 on the IRF4 LKD (Supplementary Fig. 5c). Collectively, these data support a mechanistic model: HSP90 binding to IRF4 under normal glucose conditions sterically hinders TRIM21 access, thus promoting IRF4 stability. Under GD conditions, AMPK activation disrupts the HSP90-IRF4 interaction, exposing IRF4 to TRIM21-mediated ubiquitination and degradation (Fig. 5n).

β-OHB triggers IRF4 acetylation to maintain IRF4 stability under GD conditions

Given that adipocyte-derived β-OHB rescues GD-induced IRF4 reduction, we asked whether β-OHB maintains IRF4 protein levels via the AMPK-TRIM21 axis. CHX-chase and Co-IP assays revealed that β-OHB treatment restored IRF4 stability (Fig. 6a and Supplementary Fig. 6a), enhanced the IRF4-HSP90 interaction (Fig. 6b), and attenuated GD-induced TRIM21-IRF4 binding (Fig. 6b) and IRF4 K48-linked ubiquitination (Fig. 6c). However, β-OHB did not obviously affect AMPK phosphorylation under GD conditions (Supplementary Fig. 6b), suggesting that β-OHB stabilizes IRF4 independently of AMPK signaling.

Fig. 6. β-OHB triggers IRF4 acetylation to maintain protein stability under GD conditions.

a IRF4 protein kinetics in myeloma cells treated with or without β-OHB (2 mM) in the presence of CHX under normal glucose or GD conditions. b Co-IP of HA-IRF4 with endogenous TRIM21 and HSP90 in myeloma cells treated with or without β-OHB (2 mM) under GD. Blots were probed for HA, TRIM21, or HSP90. The samples derive from the same experiment but different gels for HSP90/HA-IRF4/GAPDH and TRIM21 were processed in parallel. The similar strategy applies to (h). c IP of HA-IRF4 and immunoblotting for K48-ubiquitin and HA-IRF4 in myeloma cells treated with or without β-OHB (2 mM) under normal glucose or GD conditions for 16 h. d Coomassie staining (SDS-PAGE) of HA-IP eluates from HA-IRF4-expressing ARP1 cells treated with or without β-OHB (2 mM) under GD conditions for 16 h. e Mass spectrometry of acetylated lysine at positions 87 and 399 from the HA-IRF4 band. Kinetics of IRF4 protein (f) and cell viability (g) in myeloma cells expressing either wild-type (WT) HA-IRF4 or acetylation-mimicking mutants (K87Q, K399Q) under GD (n = 3 technical replicates; experiment independently repeated three times with similar results). h Co-IP of HA-IRF4 with endogenous TRIM21 and HSP90 in ARP1 cells expressing WT, K87Q, or acetylation-defective (K87R) HA-IRF4 under GD. i IP of HA-IRF4 and immunoblotting for K48-ubiquitin in myeloma cells expressing WT or K87Q HA-IRF4, treated with 17-AAG (2 μM), GD, or A-769662 (120 μM) for 16 h. j Specificity validation of anti-IRF4-K87ac antibody by dot blot using biotinylated peptides. k HA-IP of lysates from HA-IRF4-expressing ARP1 cells treated with or without β-OHB under GD for 16 h, followed by immunoblot for IRF4-K87ac. l Immunohistochemistry (IHC) of IRF4-K87ac in tumor tissues (from Fig. 1i). Scale bar: 100 μm. m Quantification of IRF4-K87ac staining intensity using H-score (n = 10 areas/sample). Data are presented as mean ± SD. p values were determined by two-way ANOVA (g, m). Images in a–d, f, h–k are representative of three independent experiments. Source data are provided as a Source Data file.

Since both β-OHB and AcAc protect IRF4 under GD conditions (Fig. 2g), we proposed that β-OHB exerts its effects via the ketolytic pathway, which converts β-OHB and AcAc into acetyl-CoA^25,26. To determine whether β-OHB stabilizes IRF4 via acetylation, HA-IRF4-expressing ARP1 cells were treated with or without β-OHB under GD conditions. Anti-HA immunoprecipitation and mass spectrometry identified two putative lysine acetylation sites, K87 and K399, exclusively in β-OHB-treated cells (Fig. 6d, e). We next mutated each lysine (K) to glutamine (Q), an acetyl-mimetic amino acid, and found that the K87Q mutant, but not K399Q or WT, resisted IRF4 degradation induced by GD or 17-AAG (Fig. 6f and Supplementary Fig. 6c). Moreover, myeloma cells expressing K87Q exhibited greater resistance to GD- or 17-AAG-induced cell death compared to WT or K399Q-expressing cells (Fig. 6g and Supplementary Fig. 6d). K87 of IRF4 is highly conserved across various species (Supplementary Fig. 6e), underscoring its functional importance.

To determine whether K87 acetylation of IRF4 impacts its binding to HSP90 and TRIM21, we performed Co-IP assays using anti-HA resin in ARP1 cells expressing WT, K87Q, or K87R (an acetylation-defective mutant with lysine 87 replaced by arginine). Compared to WT, the K87Q mutant exhibited enhanced binding to HSP90 and reduced interaction with TRIM21 under GD conditions, whereas the K87R mutant abolished these effects (Fig. 6h). Similar results were observed in HEK293T cells co-expressing IRF4 mutants with HSP90 or TRIM21 (Supplementary Fig. 6f, g). Notably, 17-AAG, GD or A-769662-stimulated K48-linked ubiquitination of IRF4 was substantially attenuated in K87Q mutant (Fig. 6i). These findings suggest that K87 acetylation of IRF4 enhances HSP90 binding, thereby preventing TRIM21-mediated IRF4 degradation under GD conditions.

To biochemically validate these findings, we designed biotin-conjugated peptides spanning IRF4 residues 74-101, including an acetylated K87 variant (Biotin-K87ac) and an unmodified control (Biotin-K87). Pulldown assays using lysates from HEK293T cells expressing Flag-HSP90 and Strep-TRIM21 demonstrated that Biotin-K87ac precipitated significantly more HSP90 and less TRIM21 than Biotin-K87 (Supplementary Fig. 6h), providing direct evidence that K87 acetylation promotes HSP90 binding and disrupts TRIM21 interaction.

We next generated an antibody specific for acetylated K87 and verified its specificity by dot blot assay (Fig. 6j). Using this antibody, we observed that GD treatment reduced IRF4 K87 acetylation in myeloma cells, an effect reversed by β-OHB treatment (Fig. 6k). IHC staining of tumor samples from Fig. 1i revealed reduced IRF4 K87 acetylation in 2-DG-treated tumors, whereas the presence of adipocytes increased IRF4 K87ac levels (Fig. 6l, m). Collectively, these findings suggest that under metabolic stress, myeloma cells utilize adipocyte-derived β-OHB to enhance IRF4 K87 acetylation, thereby stabilizing IRF4 and promoting cell survival.

β-OHB maintains IRF4 stability via OXCT1-mediated ketolysis

To elucidate how β-OHB stabilizes IRF4 under GD conditions, we focused on the ketolysis pathway. Ketone bodies β-OHB and AcAc are catabolized by the mitochondrial enzyme succinyl-CoA:3-ketoacid coenzyme A transferase (OXCT1)²⁶. Using the OXCT1 inhibitor pimozide to block β-OHB catabolism²⁷, we found that OXCT1 inhibition abrogated the ability of β-OHB to restore IRF4 stability and cell viability under GD conditions (Supplementary Fig. 7a–c). Immunohistochemical staining revealed a strong positive correlation between OXCT1 and IRF4 protein levels (Supplementary Fig. 7d, e), supporting a role for OXCT1 in regulating IRF4. Analysis of GEO datasets (GSE2658 and GSE5900) revealed higher OXCT1 expression in myeloma plasma cells versus normal controls (Supplementary Fig. 7f). Furthermore, analysis of the Multiple Myeloma Research Foundation (MMRF)-CoMMpass database (n = 859) using the UCSC Xena platform (http://xena.ucsc.edu/) demonstrated that high OXCT1 expression correlated with poor overall survival in myeloma patients (Supplementary Fig. 7g). Collectively, these findings indicate that OXCT1-mediated ketolysis plays a critical role in sustaining IRF4 stability and myeloma cell survival, and highlight OXCT1 as a potential therapeutic target with prognostic significance in multiple myeloma.

NAT10 mediates β-OHB-induced IRF4 K87 acetylation

To identify the acetyltransferase responsible for β-OHB-induced IRF4 K87 acetylation, we screened a panel of inhibitors targeting major acetyltransferases (p300, CBP, PCAF, GCN5, KAT8, and NAT10) in myeloma cells under GD conditions with β-OHB treatment. Both the KAT2B inhibitor NSC 694623 and the NAT10 inhibitor remodelin abolished the protective effect of β-OHB on IRF4 stability (Fig. 7a) and cell viability (Fig. 7b) in myeloma cells. Co-IP assays in HEK293T cells co-expressing IRF4-GFP with either HA-KAT2B or Flag-NAT10 revealed that NAT10, but not KAT2B, physically interacted with IRF4 (Fig. 7c). Consistently, remodelin (but not NSC 694623) significantly attenuated β-OHB-induced IRF4 K87 acetylation (Fig. 7d). Conversely, overexpression of Flag-NAT10 enhanced IRF4 K87 acetylation in HEK293T cells (Fig. 7e). Furthermore, in vitro acetylation assays showed that incubation with Flag-NAT10 protein stimulated the acetylation of a non-acetylated IRF4 peptide containing K87 (Fig. 7f), establishing NAT10 as the primary acetyltransferase responsible for K87 acetylation.

Fig. 7. NAT10 mediates β-OHB-induced IRF4 K87 acetylation.

IRF4 protein (a) and cell viability (b) in myeloma cells treated with β-OHB (2 mM) together with the indicated inhibitors (10 μM each) under GD for 16 h (n = 3 technical replicates; experiment independently repeated three times with similar results). c GFP immunoprecipitation in HEK293T cells co-expressing GFP or IRF4-GFP with Flag-NAT10 and HA-KAT2B. The samples derive from the same experiment but different gels for HA-KAT2B, another for Flag-NAT10, another for GFP, and another for GAPDH were processed in parallel. d IRF4 K87ac levels in cells treated with β-OHB (2 mM) plus NSC 694623 or Remodelin (10 µM) under GD for 16 h. e HEK293T cells were co-transfected with HA-IRF4 and either GFP or Flag-NAT10. IRF4 K87ac levels were assessed by IP and immunoblot. f In vitro acetylation assay with Flag-NAT10 and non-acetylated or K87-acetylated (K87ac) IRF4 peptides (n = 3 independent experiments). g NAT10 mRNA levels in plasma cells from myeloma patients (n = 559) versus healthy donors (n = 22) in GEO datasets (GSE2658, GSE5900). Box plots show the median (center line), 25th−75th percentiles (box bounds), and minima to maxima (whiskers). For healthy and MM respectively: min = 669 vs. 644, max = 1593 vs. 4132, median = 1238 vs. 1690, box (Q1–Q3) = 234 vs. 782, whiskers = 773–1593 vs. 644–3294. h Kaplan–Meier overall survival curves of myeloma patients stratified by high (n = 429) vs. low (n = 430) NAT10 expression from the MMRF-CoMMpass database. IRF4 protein (i) and cell viability (j, k) in myeloma cells expressing shCtrl, shNAT10-#1, or shNAT10-#2, treated with or without β-OHB under normal glucose or GD (n = 3 technical replicates; experiment independently repeated three times with similar results). l Cell viability of myeloma cells expressing WT IRF4 or K87Q mutant treated with Met, remodelin, or their combination (n = 3 technical replicates; experiment independently repeated three times with similar results). Data are presented as mean ± SD. p values were determined by one-way ANOVA (f, l), two-way ANOVA (b, j, k), two-tailed unpaired Student’s t test (g), or two-sided Gehan-Breslow-Wilcoxon test (h). Images in a, c, d, e, i are representative of three independent experiments. Source data are provided as a Source Data file.

Analysis of GEO datasets (GSE2658 and GSE5900) revealed higher NAT10 expression in malignant plasma cells versus normal controls (Fig. 7g), and elevated NAT10 predicted worse clinical outcomes (Fig. 7h), positioning NAT10 as a therapeutic target and prognostic biomarker in multiple myeloma. Knockdown of NAT10 in myeloma cells showed that the protective effects of β-OHB on IRF4 protein stability and cell viability under GD conditions are NAT10-dependent (Fig. 7i–k). Notably, remodelin treatment significantly enhanced the sensitivity of ARP1 cells expressing wild-type IRF4 to metformin. This effect was markedly attenuated in cells expressing the K87Q mutant (Fig. 7l), indicating that the pro-survival function of NAT10 depends on its acetylation of IRF4 at lysine 87. Finally, supporting the functional regulation of IRF4 by NAT10 or OXCT1, we observed significant positive correlations between mRNA levels of NAT10 or OXCT1 and those of established IRF4 transcriptional targets, including E2F5, STAG2, MYC, PIM2, SCL37A4, USB1, CASP3, and CANX (Supplementary Fig. 7h). Taken together, these findings establish NAT10 as the key acetyltransferase that mediates β-OHB-induced acetylation of IRF4 at K87, thereby stabilizing IRF4 and promoting myeloma cell survival under metabolic stress.

Synergistic anti-myeloma effects of AMPK activation and NAT10/OXCT1 inhibition

Given the distinct pathways by which AMPK activation destabilizes IRF4 and β-OHB stabilizes it, we hypothesized that combining AMPK activation with inhibition of NAT10 or OXCT1 could produce synergistic anti-myeloma activity. To test this, we evaluated multiple drug combinations in ARP1 cells using an inhibitor matrix analyzed with SynergyFinder 3.0²⁸. Strong synergistic effects were observed, with ZIP synergy scores of 24.279 for metformin plus Remodelin and 22.137 for metformin plus pimozide (Fig. 8a, b).

Fig. 8. Combining AMPK activation with ketolysis inhibition exhibits synergistic anti-myeloma activity.

Bliss synergy plots for the combination of metformin with remodelin (a) or pimozide (b) in ARP1 cells. Synergy scores were calculated using SynergyFinder 3.0. A ZIP score >10 denotes strong synergy. c, d NCG mice were subcutaneously inoculated with ARP1 cells alone or with adipocytes, then treated with 2-DG, pimozide, or remodelin as monotherapies or combinations (n = 5 mice per group). c Representative excised tumors. d Tumor weights (n = 5 tumors per group). e–g Vk12598-transplanted male C57BL/6J mice received monotherapies (metformin 200 mg/kg, remodelin 10 mg/kg, or pimozide 10 mg/kg) or combination (metformin+remodelin or metformin+pimozide) every other day for 3 weeks. e Serum IgG2b (M-protein) levels (n = 5 mice per group). f Representative spleen and liver images. g Kaplan–Meier analysis of overall survival using a separate cohort of mice (n = 6 mice per group). h–k Male C57BL/6J mice received alternate-day fasting (ADF) for 1 week prior to Vk12598 injection, then pimozide (10 mg/kg) or vehicle every other day for 3 weeks (n = 8 mice per group). h Serum ketone body levels. i Serum IgG2b (M-protein) levels. j Representative spleen and liver images. k Kaplan-Meier analysis of mouse overall survival. l Proposed model: combining AMPK activation with NAT10 or OXCT1 inhibition for myeloma therapy. Data are presented as mean ± SD. p values were determined by Log-rank test (g, k) and two-way ANOVA (d, e, h, i). Images in a, b are representative of three independent experiments. Source data are provided as a Source Data file.

For in vivo validation, we employed two distinct models. First, to directly assess the ability of these combinations to overcome adipocyte-mediated protection in the tumor microenvironment, we subcutaneously inoculated male NCG mice with ARP1 cells alone or co-injected with adipocytes. Mice were then treated with 2-DG, pimozide, or remodelin as monotherapies or in combination. We found that the 2-DG-pimozide and 2-DG-remodelin combinations significantly suppressed subcutaneous tumor growth and effectively overcame the resistance conferred by adipocytes against 2-DG-induced metabolic stress (Fig. 8c, d). Second, male C57BL/6 J mice inoculated with Vk12598 cells were treated with metformin, pimozide, or remodelin as monotherapies or in combination. While single agents had modest effects, metformin-pimozide or metformin-remodelin combinations significantly suppressed myeloma progression, as shown by reduced serum M-protein, decreased tumor infiltration in spleens and livers, and prolonged survival (Fig. 8e–g). Notably, these treatments were well-tolerated, as evidenced by the absence of significant body weight changes (Supplementary Fig. 8a) and no observed drug-induced organ damage in the heart, liver, spleen, lungs, or kidneys upon histological examination (Supplementary Fig. 8b).

Intermittent fasting has emerged as a potential strategy for cancer prevention and treatment²⁹. Although it can induce tumor metabolic stress, it also elevates systemic ketone bodies. To investigate whether fasting-induced ketosis compromises myeloma treatment, we implemented an ADF regimen in male C57BL/6J mice with systemic myeloma established via tail vein injection of Vk12598 cells, thereby assessing effects independent of local bone marrow adipocytes. As expected, the ADF regimen significantly increased the concentration of β-OHB in the mouse blood (Fig. 8h), confirming the successful induction of ketosis. Notably, in tumor-bearing hosts, OXCT1 inhibition paradoxically reduced blood β-OHB (Fig. 8h). This suggests that in advanced myeloma, the tumor-host metabolic interaction can override canonical feedback loops, leading to a net suppression of ketone body production when consumption is blocked. After one week of ADF, mice were inoculated with Vk12598 cells and treated with pimozide (Supplementary Fig. 8c). Both ADF and pimozide monotherapy significantly reduced myeloma burden, as indicated by decreased serum M-protein, reduced spleen and liver sizes, and improved survival (Fig. 8i–k). Moreover, the combination of ADF with pimozide further enhanced these anti-tumor effects (Fig. 8i–k). These results suggest that OXCT1-mediated ketone body catabolism supports myeloma cell survival under metabolic stress, and that elevated ketone bodies may partially counteract the efficacy of myeloma therapy. Collectively, these findings demonstrate that ketone bodies, whether derived locally from adipocytes or systemically from liver/adipose tissue, represent a shared metabolic vulnerability in myeloma. Accordingly, dual targeting of AMPK activation together with NAT10 or OXCT1 inhibition elicits potent synergistic anti-myeloma activity (Fig. 8l).

Discussion

Our study elucidates a sophisticated metabolic crosstalk between adipocytes and myeloma cells, uncovering mechanisms of tumor cell adaptation to nutrient stress. We demonstrate that glucose deprivation activates AMPK signaling, which disrupts the protective HSP90-IRF4 interaction. This disruption is mechanistically driven by AMPK-mediated direct phosphorylation of HSP90 at specific serine/threonine residues, which triggers its subcellular relocalization. This, in turn, exposes IRF4 to TRIM21-mediated ubiquitination and subsequent proteasomal degradation. Paradoxically, metabolic stress simultaneously induces adipocyte-derived-β-OHB secretion, which undergoes OXCT1-mediated ketolysis in myeloma cells to promote NAT10-dependent acetylation of IRF4 at K87. This acetylation event restores HSP90-IRF4 binding, effectively shielding IRF4 from degradation and sustaining myeloma cell survival.

While ketone bodies have traditionally been recognized as energy substrates shuttled from the liver to peripheral tissues during fasting or exercise, emerging evidence identifies them as signaling molecules produced by extrahepatic tissues, including adipocytes during β-adrenergic stimulation or cold exposure^30–32. Notably, this capability extends to pathological contexts, as evidenced by findings that mammary adipocytes secrete β-OHB to fuel breast cancer progression within the tumor microenvironment^33,34. Beyond local production, adipose tissue also exerts systemic control over ketogenesis, as evidenced by the dependence of fasting-induced hepatic ketogenesis on signals from lipolytic adipocytes³⁵. Collectively, these studies establish adipose tissue as a physiologically significant ketogenic organ across multiple contexts.

Building on this foundation, our study reveals a role of adipocyte-derived ketone bodies in the pathophysiology of multiple myeloma. We show that metabolic stress promotes white adipose tissue browning and upregulates adipocyte ketogenic enzymes, thereby enhancing β-OHB production. Using adipocyte-specific Hmgcs2 knockout mice, we provide in vivo evidence that adipocytes serve as a critical source of β-OHB under metabolic stress and are essential for sustaining myeloma cell survival. These results identify a targetable metabolic vulnerability specific to myeloma, offering therapeutic potential in this hematologic malignancy.

Our study also suggests a context-dependent model of adipocyte ketone body function. During systemic metabolic stress (e.g., fasting or 2-DG challenge), adipose tissues, along with the liver, contribute substantially to circulating ketone levels. In contrast, within localized nutrient-deprived niches such as the tumor microenvironment, adjacent adipocytes (including those in the bone marrow) can become a major paracrine source of ketone bodies, directly fueling tumor cell adaptation. Our genetic and functional evidence supports this dual-source model, highlighting the spatially distinct yet equally critical roles of adipocyte-derived ketones in cancer metabolism. Thereby, our work refines the paradigm of adipose support in myeloma from a broad supply of lipids and adipokines to a precise, metabolite-mediated regulation of the central oncoprotein IRF4.

Intermittent fasting has emerged as a promising intervention to lower systemic glucose and impair tumor glycolytic metabolism, a well-established vulnerability in many cancers^{14,15,36–38}. Its anti-tumor effects are attributed to both systemic glucose restriction and ketone body generation. However, the role of ketone bodies is context-dependent; while they can enhance CD8⁺ T cell anti-tumor immunity³⁹ or suppress colorectal cancer progression via the Hcar2 receptor⁴⁰, they have also been linked to adverse effects like accelerated cachexia⁴¹ or exacerbated chemotherapy toxicity⁴². Our work adds another layer of complexity to this picture, revealing a paradoxical pro-tumorigenic role for β-OHB in multiple myeloma under ADF. Although ADF itself exerts potent anti-myeloma effects, the concomitant elevation of systemic ketone bodies supports tumor survival through OXCT1-dependent ketolysis. Pharmacological inhibition of OXCT1 synergized with ADF, further suppressing tumor growth and prolonging survival in myeloma-bearing mice. Given that OXCT1 is highly expressed in myeloma and correlates with poor prognosis, our findings suggest ketolysis as a critical metabolic adaptation and highlight OXCT1 as a promising therapeutic target.

IRF4, a master transcriptional regulator, orchestrates diverse gene networks essential for myeloma cell proliferation, survival, and metabolism. Genetic ablation of IRF4 is universally lethal across myeloma subtypes regardless of their genetic background, establishing it as a compelling therapeutic target¹⁸. Our previous work revealed that ACSS2 inhibition triggers p62-dependent lysosomal degradation of IRF4, suppressing myeloma growth, thereby positioning ACSS2 as a promising target for IRF4-directed therapy⁹. Here, we unveil a dual-layer ubiquitin-proteasome regulatory system governing IRF4 protein homeostasis: (1) a degradation axis wherein glucose deprivation-activated AMPK disrupts the HSP90-IRF4 complex, enabling TRIM21-mediated ubiquitination and proteasomal degradation of IRF4; and (2) a salvage axis wherein adipocyte-derived β-OHB, via OXCT1 and NAT10, acetylates IRF4 at K87 to restore HSP90 binding under metabolic stress. This dynamic interplay underscores myeloma’s metabolic adaptability and expands opportunities for IRF4-targeted therapy.

To translate these mechanistic insights into therapeutic potential, we evaluated combinations of FDA-approved agents targeting these pathways^43–45. The combination of metformin (an AMPK activator) with either remodelin (a NAT10 inhibitor) or pimozide (an OXCT1 inhibitor) exhibited potent, synergistic anti-myeloma activity and prolonged survival in preclinical models. Given the re-entry of the glycolysis inhibitor 2-DG into clinical trials, our findings provide a strong rationale for combining glycolysis inhibitors, intermittent fasting regimens, or AMPK activators with OXCT1 or NAT10 inhibitors as an anti-myeloma strategy.

In conclusion, our findings illuminate the dual roles of adipocytes in myeloma metabolic stress resistance. Metabolic stressors suppress myeloma by inducing IRF4 degradation, while concurrently stimulating adipocyte ketogenesis, which counteracts this degradation to promote survival. This revised model unveils a therapeutic paradigm: combining metabolic stress induction with ketolysis inhibition, a promising avenue for improving myeloma treatment outcomes.

Methods

Ethics

All animal experiments were approved by the Laboratory Animal Ethics Committee of Anhui Medical University (No. 20220863). All procedures conformed to the Guidelines for Ethical Conduct in the Care and Use of Animals.

Reagents

Lysosomal-protease inhibitor leupeptin (#14026), calpain inhibitor MDL-28170 (#14283), HSP90 inhibitor 17-AAG (#11039), NAT10 inhibitor remodelin (#17346), OXCT1 inhibitor pimozide (#16222), AMPK activator A-769662 (#11900) and metformin (#16921) from Cayman Chemical. All other chemicals were acquired from MedChemExpress (MCE) unless otherwise specified. Antibody against IRF4 (acetyl K87) and peptides consisting of amino acids 74-101 of IRF4 protein with or without acetylation at 87 were custom made by GL Biochem (Shanghai) Ltd. Immunohistochemical tissue microarrays (#MMP961) were commercially procured from Wuhan Tanda Biotechnology Co., Ltd.

Mice

Male NCG (NOD/ShiLtJGpt-Prkdc^em26Cd52Il2rg^em26Cd22/Gpt), C57BL/6J wild-type mice and nude (BALB/cNj-Foxn1^nu/Gpt) mice were purchased from GemPharmatech (Nanjing, China). Male B6.129P2-Tcrb^tm1Mom/J mice were purchased from Jackson Laboratory. All mice (6–8 weeks old) were housed under controlled conditions (20–25 °C, 45–64% humidity) with a 12-h light/dark cycle and ad libitum access to water (5 mice per cage).

NCG mice, nude mice and Tcrb^tm1Mom/J mice were fed ad libitum with irradiated rodent diet (Research diets, #D12450H; 10% kcal fat, 20% kcal protein, 70% kcal carbohydrate). C57BL/6J mice were fed ad libitum a high-fat diet (Research diets, #D12492; 60% kcal fat, 20% kcal protein, 20% kcal carbohydrate) for 20 weeks to establish a diet-induced obese (DIO) model⁴⁶. Control mice were fed a normal diet (Research diets, #D12450B; 10% kcal fat, 20% kcal protein, 70% kcal carbohydrate). For the ADF model (described below), mice were subjected to an alternate-day fasting regimen: fasting days (0 kcal) alternating with feeding days (ad libitum access to normal diet, #D12450B).

Adipocyte-specific Hmgcs2 knockout mice (Hmgcs2^AKO) and liver-specific Hmgcs2 knockout mice (Hmgcs2^LKO) were generated by crossing Hmgcs2^flox/flox mice (C57BL/6N-Hmgcs2^em1Cflox/Cya, Cyagen) with Adipoq-Cre mice (C57BL/6JGpt-Tg (Adipoq-iCre)166/Gpt, GemPharmatech) or Alb-Cre mice (C57BL/6JGpt-H11^{em1Cin(Alb-iCre)}/Gpt, GemPharmatech), respectively. Mice genotypes were tested by PCR amplification of DNA extracted from tail. Tissue-specific deletion of exon 2 of Hmgcs2 was verified using Hmgcs2 floxed allele primer pair (FW 5’-TTTCAAGCAGTACCACTCCCTAAC-3’; RV 5’-GTCAGGATCTCTTATTTCACCCCAG-3’) and Cre recombinase primer pair (Cre-FW 5’-ATTTGCCTGCATTACCGGTCG-3’; Cre-RV 5’-CAGCATTGCTGTCACTTGGTC-3’).

Cell cultures

All cells were maintained at 37 °C with 5% CO₂ in a cell incubator with 10% fetal bovine serum (HyClone, #SH30406.02) and 1% penicillin/streptomycin (Gibco, #15140148). ARP1 myeloma cells was kindly provided by Dr. Zhiqiang Liu of Shandong First Medical University, and the murine myeloma cell line Vk12598 was kindly provided by Dr. P. Leif Bergsagel of the Mayo Clinic. 5TGM1 cell line was kindly provided by Dr. Frederic J. Reu of Case Western Reserve University. Other cells, such as RPMI8226, ARH77, U266, H929, IM-9 and HEK293T were purchased from the American Type Culture Collection (ATCC, USA). MOPC315 (TCM-C797) was obtained from HyCyte (China), and OP9 (CVCL_4398) was acquired from Wuhan Pricella Life Technology Co., Ltd. (Pricella, China). Myeloma cells were cultured in RPMI1640 medium. HEK293T cells and OP9 cells were maintained in Dulbecco’s modified Eagle’s medium (DMEM). 5TGM1 cells were cultured in IMDM medium.

Adipocyte isolation and culture

Mouse adipose tissue-derived adipocytes were isolated from surgical specimens using collagenase digestion. Briefly, fresh adipose tissue was minced and digested with 1 mg/mL collagenase type I (Sigma, #C0130) in PBS containing 2% BSA at 37 °C for 45–60 min with gentle shaking. The digested tissue was filtered through a 100μm cell strainer and centrifuged at 500 × g for 5 min. The floating mature adipocytes were collected and washed three times with PBS⁸.

Human fetal bone marrow-derived mesenchymal stem cells (BMSCs) at passage 2 were purchased from Cyagen Biosciences (HUXMF-01001, Suzhou, China; product information available at https://oricellbio.cn/product/bone-marrow-msc-HUXMF-01001.html). The cells were isolated from bone marrow tissues of fetal limbs collected from legally terminated pregnancies, sourced from anonymous donated fetal specimens at certified tertiary hospitals. Informed consent was obtained from the donor’s legal guardian and no financial compensation or material remuneration was provided. Primary cells were expanded through standard adherent culture and routinely maintained in Cyagen complete growth medium (HUXMF-90011, Cyagen Biosciences). Cell identity was validated by flow cytometry profiling of canonical mesenchymal markers: positive for CD29, CD44 and CD90, and negative for CD34, CD45 and CD11b. For adipogenic differentiation, BMSCs were cultured for two weeks in differentiation medium consisting of DMEM supplemented with 10% FBS, 1 μM dexamethasone (MCE, #HY-14648), 0.2 mM indomethacin (MCE, #HY-14397), 10 μg/mL insulin (MCE, #HY-P0035), and 0.5 mM 3-isobutyl-1-methylxanthine (IBMX; MCE, #HY-12318). Mouse OP9 stromal cells were differentiated into adipocytes by treatment with 2 μM rosiglitazone (MCE, #HY-17386) in α-MEM for two weeks, with medium changed every 3 days. Mature adipocytes were characterized by Oil Red O staining and expression of adipocyte markers, such as perilipin⁸. Conditioned medium was collected from adipocytes (AD-CM) cultured under glucose deprivation conditions. For some experiments, AD-CM was digested with proteinase K or fractionated (>3 kDa or <3 kDa) using Centricon filters (3 kDa cutoff, Millipore). Untreated medium was collected as a control.

Cell viability assay

To monitor cell viability with or without adipocytes under glucose deprivation conditions, MM cells were plated at a density of 1 × 10⁶ cells/well in 6-well plates for co-culture experiments with adipocytes (as shown in Fig. 2a). For other conditions, cells were seeded at 1 × 10⁴ cells/well in 96-well plates and exposed to glucose deprivation with or without β-OHB supplementation. In studies evaluating IRF4 variants, MM cells stably expressing HA-tagged wild-type (WT), K87Q, or K399Q IRF4 were plated at 1 × 10⁴ cells/well in 96-well plates and subjected to either glucose deprivation or 17-AAG treatment. Viability was measured using the CellTiter-Glo Luminescent Assay (Promega), following the manufacturer’s protocol.

Synergy determination with SynergyFinder

MM cells were seeded in 96-well plates (1.5 × 10⁴ cells/well), and viability was assessed using the CellTiter-Glo Luminescent Assay (Promega, #7570) following treatment with metformin (Cayman Chemical, #16921) in combination with either remodelin or pimozide. Drug concentrations were selected based on their respective IC₅₀ values.

For synergy analysis, metformin was tested at 100, 120, 140, 160, and 180 μM, while remodelin was evaluated at 1, 2, 4, 6, and 8 μM, and pimozide at 1, 2, 6, 8, 10, 12, and 16 μM. Drug interactions were analyzed using SynergyFinder (https://synergy-finder.fimmm.fi)²⁸, which calculated the survival index and zero interaction potency (ZIP) scores. A ZIP score >0 indicates synergy, and >10 denotes strong synergy. Additionally, drug combination response heatmaps were generated to evaluate therapeutic potential.

Quantitative real-time PCR

Total RNA was extracted from samples using the RNA isolater Total RNA Extraction Reagent (Vazyme Biotech, #R401-01) according to the manufacturer’s protocol. First-strand cDNA was synthesized from 1 μg of total RNA using the HiScript III-RT SuperMix (Vazyme Biotech, #R323-01). Quantitative PCR amplification was performed in triplicate using SYBR qPCR Master Mix (Vazyme Biotech, #Q711-02) on a StepOnePlus™ Real-Time PCR System (Applied Biosystems, USA) under the following cycling conditions: initial denaturation at 95 °C for 10 min; 40 cycles of 95 °C for 15 s and 60 °C for 1 min. All primer sequences used for amplification are provided in Supplementary Tables 1 and 2.

Plasmid constructs and shRNA knockdown

The following human expression constructs were subcloned into the pCDH-CMV-MCS vector (System Biosciences, #CD500B-1): HA-tagged IRF4 (full-length and truncations: aa 1-139, 1-238, 1-420, 1-451, 139-238, 238-420), GFP-tagged IRF4, HA-tagged AMPK α1/α2/β/γ, Strep/Flag-tagged HSP90, Strep/Flag-tagged TRIM21, HA-tagged KAT2B, and Flag-tagged NAT10. Mut Express II Fast Mutagenesis Kit V2 (Vazyme Biotech, #C214-01) was used to generate IRF4 mutants (K87Q, K87R, K399Q) from the wild-type template. For knockdown experiments, lentiviral pLKO.1 vectors (Addgene, #8453) encoding shRNAs targeting FBXW11, TRIM21, NAT10, OXCT1, and HSP90 were packaged using psPAX2/pMD2.G helper plasmids. The primer sequences used for shRNA constructs are listed in Supplementary Table 3. All plasmids were purified using the EndoFree QIAGEN Plasmid Maxi Kit (#12362). For transient transfection, plasmids were delivered at a 1:3 DNA: Lipofectamine 3000 (Thermo Fisher Scientific, #L3000001) ratio. Stable cell lines were generated via lentiviral transduction followed by 72-h puromycin selection. All newly generated plasmids and constructs are available from the corresponding author upon reasonable request.

Flow cytometry analysis

Bone marrow cells were harvested by flushing mouse femurs with ice-cold PBS containing 2% FBS. Splenocytes were prepared by mechanically dissociating spleens through a 70 μm cell strainer (Corning) in PBS supplemented with 2% FBS. Subsequently, cell suspensions were treated with ACK lysing buffer (Thermo Fisher Scientific, #A1049201) for 5 min at room temperature to remove red blood cells, followed by centrifugation at 300 × g for 5 min at 4 °C. For immunostaining, cells were incubated with APC-conjugated anti-mouse CD138 (Syndecan-1) antibody (1:100, BioLegend, #142505) in the dark at 4 °C for 30 min. After two washes with PBS containing 2% FBS, the stained cells were resuspended in 300 μL of PBS with 2% FBS. Flow cytometry analysis was performed using a Cytoflex S instrument (Beckman, Germany). Data were processed and analyzed using FlowJo software (version 10).

Immunofluorescence staining

IRF4-GFP-expressing ARP1 cells under various treatment conditions were fixed with 4% paraformaldehyde for 15 min at room temperature, followed by permeabilization with 0.3% Triton X-100 for 10 min. After blocking with 5% BSA for 1 h at room temperature, cells were incubated overnight at 4 °C with anti-HSP90 antibody (1:100; Cell Signaling Technology, #4877) and Strep antibody (1:100; Thermo Fisher Scientific, #MA5-17283). Following three PBS washes, cells were incubated with Alexa Fluor™ 555-conjugated Goat anti-Rabbit IgG secondary antibody (1:500; Thermo Fisher Scientific, #A-21428) for 1 h at room temperature in the dark. Nuclei were counterstained with DAPI (1 μg/mL) for 15 min. Fluorescence images were captured using a Zeiss LSM800 confocal microscope with ZEN imaging software.

Western blot assay

For adipose tissue protein extraction, freshly dissected tissues were flash-frozen in liquid nitrogen and homogenized in ice-cold RIPA buffer (CST, #9806) containing 1 mM PMSF (Sigma Aldrich, #P7626), Complete Protease Inhibitor Cocktail, and 5 mm steel beads using a MagNA Lyser homogenizer (Roche). After centrifugation (14,000 × g, 30 min, 4 °C), the lipid layer was removed and samples were recentrifuged (14,000 × g, 10 min) to obtain clear lysates. Cellular proteins were extracted using 1× RIPA buffer supplemented with protease inhibitors.

Cell lysates were subjected to SDS-PAGE, transferred to a nitrocellulose membrane, and immunoblotted with antibodies against IRF4 (1:4000; CST, #62834), IRF4(K87ac) (1:500; GL Biochem), HMGCS2 (1:2000; Abcam, #ab137043), Perilipin (1:2000; CST, #9349), UCP1(1:2000, CST, #72298), K48-Ubiquitin (1:1000; CST, #8081), HSP90 (1:2000; CST, #4877), TRIM21 (1:2000; CST, #92043), Phospho-AMPKα (Thr172) (1:1000; CST, #2535), AMPK (1:2000; CST, #2532), NAT10 (1:2000; Thermo Fisher Scientific, #PA5-113336), GAPDH (1:10,000; CST, #2118), Vinculin (1:1000; CST, #4650), GFP (1:1000; CST, #2955), HA (1:2000; CST, #3724), Flag (1:2000; CST, #14793), Phospho-Ser/Thr (1:1000; ABclonal, #AP0893) and Strep (1:2000; Thermo Fisher Scientific, #MA5-17283). The level of GAPDH or Vinculin served as protein loading control.

Immunoprecipitation (IP)

Cells were lysed in ice-cold mild lysis buffer (containing 1 mM PMSF and protease inhibitor cocktail) with constant agitation for 30 min. After clarification, lysates containing 1 mg total protein were incubated with anti-HA-agarose, anti-Flag-agarose, or anti-GFP-agarose beads with rotation for 2 h at 4 °C. The beads were then washed six times with lysis buffer, collected by centrifugation (2500 × g, 3 min), and resuspended in 20 μL 2× SDS sample buffer. Following denaturation at 95 °C for 5 min, the proteins were separated by 10% SDS-PAGE along with 2% input controls, and subsequently transferred to 0.45 μm nitrocellulose membranes for immunoblotting analysis.

Immunohistochemistry

Formalin-fixed, paraffin-embedded (FFPE) tissues from bone marrow microarrays and murine subcutaneous xenografts were sectioned at 4 μm thickness. After standard deparaffinization in xylene and rehydration through graded ethanol series, heat-induced epitope retrieval was performed in citrate buffer (pH 6.0) using microwave treatment. Tissue sections were incubated with primary antibodies against IRF4 (1:200; CST, #62834), IRF4(K87ac) (1:100; GL Biochem, Custom-made), phospho-AMPKα (Thr172) (1:100; CST, #2535), Ki-67 (1:100; CST, #9449), perilipin-1 (1:200; CST, #9349), and CD138 (1:100; Biolegend, #142501) using the EnVision FLEX detection system (Dako, K8002) according to the manufacturer’s protocol. Hematoxylin (CST, #14166) was used for nuclear counterstaining. Quantitative assessment was performed using the VECTRA 3.0 Automated Quantitative Pathology Imaging System (PerkinElmer) at ×20 magnification, with subsequent multispectral analysis and cell phenotyping conducted using inForm Advanced Image Analysis Software (v2.4.8, PerkinElmer).

Cycloheximide chase assay

Cycloheximide chase assays were performed to assess protein stability. Myeloma cells were treated with 50 μg/mL cycloheximide (CHX; MCE, #HY-12320) under various experimental conditions. A baseline sample (t = 0) was collected immediately prior to CHX treatment. Additional samples were obtained at designated time points thereafter. Cell pellets were harvested at each time point and analyzed by Western blot following SDS-PAGE.

ELISA for serum M-protein quantification

Serum M-protein levels in mice were measured using a commercial IgG2b ELISA kit (Thermo Fisher Scientific, #88-50430-88) following the manufacturer’s protocol. Briefly, whole blood was collected from euthanized mice and clotted at 25 °C for 30 min. After centrifugation (1600 × g, 15 min, 4 °C), the serum was separated and diluted 1:20,000 in sample dilution buffer. A standard curve was generated by regression analysis, and M-protein concentrations were interpolated from the curve.

Protein interaction modeling analysis

The protein structures of IRF4, HSP90, and TRIM21 were retrieved from the AlphaFold database (Version 4). Protein-protein docking was performed using the GRAMM docking web server (http://gramm.compbio.ku.edu/) with default parameters. The best docking models were selected based on the GRAMM scoring function. Molecular interactions and binding interfaces were visualized and analyzed using PyMOL (version 2.7, Schrödinger, LLC)⁴⁷.

Mass spectrometry analysis

Metabolomics of AD-CM: conditioned medium from adipocytes cultured under normal glucose (n = 5 biological replicates) or GD conditions (n = 7 biological replicates) were collected. A pooled quality control (QC) sample was prepared by mixing equal aliquots of all experimental samples and was injected every 6 runs to monitor system stability; features with a relative standard deviation (RSD) > 30% across QC samples were excluded. Metabolites were extracted with methanol/acetonitrile/water (2:2:1, v/v/v), sonicated, and centrifuged. Samples were analyzed by LC-MS using a UHPLC-ESI-Q-TOF-MS system (Agilent 1290 LC with AB Sciex TripleTOF 5600) on a BEH Amide column (2.1 mm × 100 mm, 1.7 μm) with a gradient of 95–40% acetonitrile (25 mM ammonium acetate/ammonium hydroxide) at 0.5 mL/min. MS data were acquired in positive and negative modes (m/z 60–1200). Raw data were processed using XCMS and analyzed by OPLS-DA based on variable importance in projection (VIP > 1.0) and two-tailed Student’s t test (p < 0.05).

Interactome analysis (IP-MS): HA-IRF4-expressing RPMI8226 cells were cultured under normal glucose or GD conditions (Three biological replicates were pooled before tryptic digestion and analyzed as a single sample, n = 1 per condition). After anti-HA immunoprecipitation and tryptic digestion, peptides were analyzed using an EASY-nLC 1200 system coupled to a Q Exactive HF-X mass spectrometer (Thermo Fisher Scientific). Peptides were separated on a 75 μm × 25 cm C18 column at 300 nL/min with a 90-min gradient of 2–35% acetonitrile (0.1% formic acid). MS data were acquired in data-dependent acquisition (DDA) mode: full MS scan at 60,000 resolution (m/z 350–1650), followed by top 20 MS/MS scans at 15,000 resolution (normalized collision energy, NCE = 28). Spectra were searched against the UniProt human database (release 2021_04) using SEQUEST (BioWorks Browser v3.3.1) with a 1% false discovery rate (FDR) at the peptide level. Search parameters included: trypsin with up to 2 missed cleavages; fixed modification: carbamidomethyl (C); variable modifications: oxidation (M) and acetylation (protein N-term); precursor mass tolerance of 10 ppm; fragment ion tolerance of 0.5 Da; minimum peptide length of 6 amino acids. Protein identifications were accepted with a minimum of 1 unique peptide per protein group. Due to the exploratory nature of this discovery-oriented experiment, a single pooled sample was analyzed per condition; identified interactors were validated by Co‑IP assays.

Acetylation and phosphorylation site mapping: HA-IRF4-expressing ARP1 cells were treated with or without β-OHB (2 mM) under GD conditions (Three biological replicates were pooled before tryptic digestion and analyzed as a single sample, n = 1 per condition). For HSP90 phosphorylation mapping, ARP1 cells expressing Strep-HSP90 were treated with A-769662 (120 μM). The protein bands were excised from SDS-PAGE gels, in-gel digested with trypsin, and analyzed by LC-MS/MS as above. MS/MS spectra were processed using MaxQuant (v2.0.1.0) with the following settings: trypsin with up to 2 missed cleavages; fixed modification, carbamidomethyl (C); variable modifications, acetylation (K), phosphorylation (S/T/Y), oxidation (M); precursor mass tolerance of 4.5 ppm; fragment ion tolerance of 20 ppm; FDR = 1%. Minimum peptide length was set to 7 amino acids. Acetylation or phosphorylation sites were accepted with a localization probability ≥0.75. Due to the discovery-driven nature of post-translational modification mapping, a single pooled sample was analyzed per condition; identified sites were confirmed by site-directed mutagenesis and functional assays.

In vitro acetylation reaction and kinase analysis

Purified Flag-NAT10 protein from Flag-NAT10-expressing HEK293T cells was prepared for the assay. The acetylation reaction of Flag-NAT10 with K87 or K87ac peptide was assessed using the Acetyltransferase Activity Kit (Enzo Life Sciences) according to the manufacturer’s protocol.

For the in vitro kinase assay, purified Flag-AMPKα/β/γ protein complex was prepared from HEK293T cells overexpressing the three subunits. Strep-tagged wild-type (WT) or phosphorylation-deficient mutant (4A) HSP90 proteins were purified from HEK293T cells expressing the respective constructs. The kinase reaction was performed by incubating the active Flag-AMPK complex with either WT or 4A mutant in kinase buffer containing 200 μM ATP at 30 °C for 30 min. Phosphorylation of HSP90 was detected by immunoblotting using a phospho-Ser/Thr antibody.

In vivo mouse experiments

All animal experiments were approved by the Laboratory Animal Ethics Committee of Anhui Medical University. All mice were monitored at least every 3 days for signs of morbidity or mortality. Humane endpoints were defined as a tumor volume exceeding 2000 mm³, a body weight loss of >20% relative to initial weight, an inability to reach food or water, or moribund condition. Any mouse meeting these criteria was immediately euthanized by CO₂ inhalation followed by cervical dislocation.

For the DIO-Vk12598 model (Fig. 1a), 6-week-old male C57BL/6J, Hmgcs2^flox/flox, or Hmgcs2^AKO mice were fed either a high-fat diet (HFD; 60% kcal) or a normal diet (ND; 10% kcal) for 20 weeks, with body weight monitored throughout the study. Vk12598 cells (2 × 10⁶ cells/mouse) were injected intrafemorally into each mouse. One-week post-injection, mice were treated daily with intraperitoneal 2-DG (500 mg/kg) or an equivalent volume of vehicle control for 4 weeks. Disease progression was assessed by measuring serum M-protein levels as a surrogate for tumor burden. Additionally, spleen and bone marrow were harvested, and the proportion of CD138⁺ cells was quantified by flow cytometry. For survival analysis, mice were monitored daily and euthanized upon reaching the humane endpoints defined above. No mice reached these endpoints before the scheduled experimental completion.

To assess the functional contribution of adipocyte-derived ketone bodies in the absence of hepatic ketogenesis, Hmgcs2^LKO mice received intrafemoral co-implantation of luciferase-labeled Vk12598 myeloma cells (2 × 10⁶ cells/site) together with either AD^WT or AD^KO. One-week post-implantation, mice were treated with daily intraperitoneal 2-DG (500 mg/kg) for three weeks. Tumor burden was monitored weekly by bioluminescent imaging.

For the xenograft model, there are several settings. In Fig. 1h, ARP1 cells (5 × 10⁵ cells/site) were injected alone into the left flank of male NCG mice, while a mixture of ARP1 cells (5 × 10⁵ cells/site) and human MSC-derived adipocytes (5 × 10⁵ cells/site) was injected into the right flank. One-week post-injection, mice received daily intraperitoneal 2-DG (500 mg/kg). Tumor volumes were measured weekly using digital calipers, and tumor volume was calculated as (length × width²)/2. After 4 weeks of treatment, tumors were excised and subsequently either weighed for quantitative analysis or fixed for immunohistochemical (IHC) staining. In Supplementary Fig. 1a, c, equal numbers of luciferase-labeled ARP1 cells (5 × 10⁵ cells/site) either alone or co-injected with human MSC-derived adipocytes (5 × 10⁵ cells/site) were intrafemorally injected into the left or right femurs, respectively, of NCG mice. Similarly, luciferase-labeled 5TGM1 cells (5 × 10⁵ cells/site), alone or mixed with mouse OP9-derived adipocytes (5 × 10⁵ cells/site), were injected into the femurs of Tcrb^tm1Mom/J mice. Beginning 1-week post-injection, mice received daily intraperitoneal injections of either 500 mg/kg 2-DG or an equivalent volume of vehicle control for 4 weeks. Bioluminescent signals were imaged at 3 weeks after cell injection.

For the ADF model, 6-week-old male C57BL/6J mice were subjected to an ADF regimen consisting of fasting days (0 kcal) alternating with feeding days (20 kcal) for 1 week⁴⁸, with age-matched mice fed a normal diet serving as controls. All mice received intravenous injections of Vk12598 cells (2 × 10⁶ cells/mouse). Three days post-inoculation, treatment was initiated with either 10 mg/kg pimozide (oral gavage every 2 days) or vehicle control (2% Tween-80), continuing until experimental endpoints. Tumor progression was assessed through weekly measurements of serum IgG2b levels. For metabolic profiling, additional blood samples were obtained during fasting phases to assess glucose and ketone body levels.

To investigate the combinatorial effects of metformin with either remodelin or pimozide, male C57BL/6J mice received intravenous injections of Vk12598 cells (1 × 10⁶ cells/mouse). Treatment commenced one week post-inoculation and continued for 3 weeks with administration every 2 days, including: (1) monotherapy groups—200 mg/kg metformin (intraperitoneal injection)¹⁵, 10 mg/kg remodelin (intraperitoneal injection)⁴⁹, or 10 mg/kg pimozide (oral gavage)²⁷; (2) combination groups—metformin (200 mg/kg, i.p.) plus either remodelin (10 mg/kg, i.p.) or pimozide (10 mg/kg, oral); and (3) vehicle control groups (i.p. or oral). Tumor progression was monitored by serial serum M-protein measurements, and survival was recorded throughout the study.

Statistics and reproducibility

Statistical analysis was performed using GraphPad Prism (Version 9.0) software. Statistical data with error bars are presented as mean ± standard deviation (SD). Paired or unpaired two tailed Student’s t test, one-way analysis of variance (ANOVA), two-way ANOVA were applied to determine significant differences between groups indicated in the figure legends. Statistical differences in overall survival were determined by the Log-rank test. Spearman’s correlation analysis was performed to determine the correlation between IRF4 and TRIM21 or p-AMPK. p values less than 0.05 were considered statistically significant. All experiments were independently repeated at least three times or more than five mice per group for in vivo studies. No statistical method was used to predetermine sample size; no data were excluded from the analyses. The investigators were not blinded to group allocation during experiments and outcome assessment, except that the investigators who performed IHC staining were blinded to the in vitro results.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

Author contributions

Z.L. and H.L. designed all experiments and wrote the manuscript; C.S., P.Y., Y.L., T.X., R.L., M.G., W.L., T.J., Q.W., Y.W., and Y.Y. performed experiments and statistical analyses; R.L., Q.D., and X.B. performed bioinformatics analysis; C.L. and Q.D. provided critical suggestions. All authors have reviewed the final manuscript.

Peer review

Peer review information

Nature Communications thanks Tolga Emre, Sean Hartig, and the other, anonymous, reviewer(s) for their contribution to the peer review of this work. A peer review file is available.

Funding

Z.L. discloses support for the research of this work from the National Natural Science Foundation of China (grant Nos. 82270217 and 82573067). Q.W. and T.J. disclose support for the research of this work from the Postgraduate Innovation Research and Practice Program of Anhui Medical University (grant No. YJS20250068 to Q.W. and No. YJS20240076 to T.J.).

Data availability

Metabolomics data generated in this study have been deposited in the MetaboLights database under accession code MTBLS14445. Raw LC-MS/MS files are available through MetaboLights. Processed metabolite abundance tables are provided as Supplementary Data 1. Mass spectrometry proteomics data have been deposited via the ProteomeXchange Consortium under the following identifiers: PXD070988; PXD076575. The processed mass spectrometry data are available at Supplementary Data 2 and 3. The myeloma plasma cell dataset is available from the Gene Expression Omnibus (GEO) under accession codes: GSE5900; GSE2658. Public GEO datasets related to fasting and adipocyte biology are available under: GSE199963; GSE7623. RNA-seq profiles and clinical data for 859 multiple myeloma cases were sourced from the MMRF-CoMMpass study (IA18) via the UCSC Xena database (http://xena.ucsc.edu/). All data visualization and statistical analyses were performed using established bioinformatics tools and R software packages. Source data are provided with this paper.

Competing interests

The authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at https://doi.org/10.1038/s41467-026-76595-0.