Ascites-Derived ALDH+CD44+ Tumour Cell Subsets Endow Stemness, Metastasis and Metabolic Switch Via PDK4-Mediated STAT3/AKT/NF-Κb/IL-8 Signalling in Ovarian Cancer

www.nature.com/bjc

ARTICLE Cellular and Molecular Biology Ascites-derived ALDH+CD44+ tumour cell subsets endow stemness, metastasis and metabolic switch via PDK4-mediated STAT3/AKT/NF-κB/IL-8 signalling in ovarian cancer

Yu-Xin Jiang1, Michelle Kwan-Yee Siu1, Jing-Jing Wang1, Xue-Tang Mo1, Thomas Ho-Yin Leung1, David Wai Chan1, Annie Nga-Yin Cheung2, Hextan Yuen-Sheung Ngan1 and Karen Kar-Loen Chan1

BACKGROUND: Ovarian cancer is characterised by frequent recurrence due to persistent presence of residual cancer stem cells (CSCs). Here, we identify and characterise tumour subsets from ascites-derived tumour cells with stemness, metastasis and metabolic switch properties and to delineate the involvement of pyruvate dehydrogenase kinase 4 (PDK4) in such process. METHODS: Ovarian cancer cells/cell lines derived from ascites were used for tumourspheres/ALDH+CD44+ subset isolation. The functional roles and downstream signalling of PDK4 were explored. Its association with clinical outcome of ovarian cancer was analysed. RESULTS: We demonstrated enhanced CSC characteristics of tumour cells derived from ovarian cancer ascites, concomitant with ALDH and CD44 subset enrichment and high PDK4 expression, compared to primary tumours. We further showed tumourspheres/ ALDH+CD44+ subsets from ascites-derived tumour cells/cell lines with CSC properties and enhanced glycolysis. Clinically, PDK4 expression was correlated with aggressive features. Notably, blockade of PDK4 in tumourspheres/ALDH+CD44+ subsets led to inhibition of CSC characteristics, glycolysis and activation of STAT3/AKT/NF-κB/IL-8 (signal transducer and activator of transcription 3/protein kinases B/nuclear factor-κB/interleukin-8) signalling. Conversely, overexpression of PDK4 in ALDH−CD44– subsets exerted the opposite effects. CONCLUSION: Ascites-derived ALDH+CD44+ tumour cell subsets endow stemness, metastatic and metabolic switch properties via PDK4-mediated STAT3/AKT/NF-κB/IL-8 signalling, suggesting PDK4 as a viable therapeutic molecular target for ovarian cancer management.

British Journal of Cancer (2020) 123:275–287; https://doi.org/10.1038/s41416-020-0865-z

BACKGROUND produced over successive generations, consistent with the known Ovarian cancer is one of the most lethal gynaecological sphere-forming and self-renewal characteristics of stem cells.4 malignancies worldwide. Cancer stem-like cells (CSCs) represent These spheroids display upregulation of CSC markers and similar a small subpopulation of tumour cells with stem-like properties properties to CSC, such as assisting metastasis development and that are responsible for tumour growth, metastasis, chemoresis- tumour escape from chemotherapy, implicating their enrichment tance and recurrence,1 leading to poor prognosis of ovarian in cells with CSC characteristics.4 Increasing evidence has been cancer patients. Effective targeting of this subset of cells may obtained on subpopulations of ascites-derived cells with CSC therefore eventually aid in control of the disease. Accumulating markers and properties, including sphere formation, chemoresis- studies have identified and characterised a self-renewing sub- tance, tumour initiation and serial tumour propagation.2,5 How- population of CSCs displaying stem-like properties in ovarian ever, the underlying mechanisms are yet to be established. cancer,2 although the associated regulatory mechanisms remain High lactate production and low glucose oxidation, regardless unknown at present. of oxygen availability, a phenomenon known as aerobic glycolysis Owing to their location as intra-abdominal tumours, ovarian (Warburg effect), are commonly observed in various cancers.6 cancer cells exfoliated from the primary ovarian tumour floating in Recent studies have shown that CSC metabolism is similar to that peritoneal fluid are prone to forming three-dimensional (3D) of healthy tissue stem cells, which prefer to rely on aerobic multicellular aggregates (spheroids), which serve as a vehicle for glycolysis for energy supply with reduced mitochondrial activity, tumour cell dissemination within the peritoneal cavity.3 Patients compared to differentiated cells.7 CSCs favour glycolysis rather with advanced ovarian cancer are often diagnosed with peritoneal than oxidative phosphorylation (OXPHOS) in multiple cancer fluid accumulation (also known as ascites fluid). Ovarian cancer types, including liver, breast and colorectal tumours.8 Exploitation spheroids derived from ascites and ovarian cancer cell lines are of the metabolic characteristics of CSCs, especially the properties

1Department of Obstetrics and Gynaecology, LKS Faculty of Medicine, The University of Hong Kong, Pokfulam, Hong Kong and 2Department of Pathology, LKS Faculty of Medicine, The University of Hong Kong, Pokfulam, Hong Kong Correspondence: Karen Kar-Loen Chan ([email protected]) Received: 2 May 2019 Revised: 27 January 2020 Accepted: 7 April 2020 Published online: 11 May 2020

© The Author(s), under exclusive licence to Cancer Research UK 2020 Ascites-derived ALDH+CD44+ tumour cell subsets endow stemness... Y.-X. Jiang et al. 276 that are not expressed by differentiated cancer cells, may facilitate centrifugation at 300 × g for 10 min, supernatant fractions were the development of novel therapeutic strategies for ovarian removed. Flow Buffer was added to resuspend cell pellets along cancer. with FcR Blocking Reagent and APC-conjugated CD44 antibody. Pyruvate dehydrogenase kinase 4 (PDK4) isoform is a gate- The mixture was incubated for 30 min at 4 °C in the dark. Sorted keeping enzyme that negatively phosphorylates pyruvate dehy- cells were collected in complete medium with 2% P/S and drogenase, leading to conversion of pyruvate to lactate in the incubated at 4 °C before seeding into culture plates. cytoplasm instead of further oxidation in mitochondria (i.e. OXPHOS).9 PDK4 is mostly detected in the skeletal muscle and siRNA transfection heart,10 and its overexpression has been reported in glioblastoma, PDK4 and control siRNA were commercially obtained (Life breast11 and colon12 cancers. A recent study revealed that PDK4 Technologies, Canada) and introduced into cancer cells using induces tumour growth through regulating the KRAS pathway in SilentFect (Bio-Rad, Canada). At 48 h after transfection, cells were lung and colorectal tumours.13 collected for use in subsequent assays. Here, we compared the CSC properties of primary ovarian tumour and ascites-derived tumour cells, and subsequently Transient overexpression of PDK4 analysed CSC properties, metabolic switches and PDK4 expression PDK4 or control vector plasmids (GFP-tagged; OriGene, MD, USA) in spheroid cells (tumourspheres) and in the ALDH+CD44+ subset were transfected into cancer cells using Lipofectamine 3000 derived from ascites and ascites-originating ovarian cancer cell (Invitrogen, Canada) for 48 h and the cells plated for subsequent lines versus monolayer cells and the ALDH−CD44− subset. The assays. clinical signiﬁcance of PDK4 in ovarian cancer was further determined. Functional characterisation was performed and the Treatment with cisplatin, DCA, stattic and DMAPT downstream signalling pathway of PDK4 potentially contributing For cisplatin experiments, ALDH+CD44+ and ALDH−CD44− SKOV3 to CSC properties in tumourspheres/ALDH+CD44+ subsets cells were plated 12 h before treatment with cisplatin (0 and 20 μM; derived from ascites or ascites-originating cell lines was deli- Sigma-Aldrich) or control vehicle (ddH2O) for 48 h. For DCA or stattic neated. Ascites-derived tumour cells exhibited enhanced CSC experiments, ALDH+CD44+ SKOV3 cells were plated 12 h before characteristics and PDK4 overexpression. Moreover, in vitro and treatment with DCA (0, 5 and 10 mM; Sigma-Aldrich), stattic (0, 7.5 in vivo experiments showed increased CSC properties and and 10 μM; Sigma-Aldrich) or control vehicles (ddH2O for DCA and enhanced glycolysis of cell subpopulations isolated based on dimethyl sulfoxide (DMSO) for stattic) for 48 h. For co-treatment

1234567890();,: tumourspheres formation/CSC markers ALDH and CD44 from experiments, ALDH+CD44+ SKOV3 cells were plated 12 h before ascites and ovarian cancer cell lines, compared to monolayer cells/ simultaneous treatment with cisplatin (0, 10 and 20 μM) and DCA ALDH−CD44− subsets. Inhibition of PDK4 via the small interfering (10 mM) or control vehicle (ddH2O) for 48 h. For dimethylamino- RNA (siRNA) approach or treatment with a pan-PDK inhibitor, parthenolide (DMAPT) experiments, (i) ALDH+CD44+ SKOV3 cells dichloroacetate (DCA), led to suppression of cancer cell stemness were plated 12 h before treatment with DMAPT (0 and 1 μM; Abcam, and tumour growth through the STAT3/AKT/NF-κB/IL-8 (signal UK) or control vehicle (DMSO) for 48 h and (ii) ALDH-CD44− SKOV3 transducer and activator of transcription 3/protein kinases B/ cells transiently transfected with control vector or PDK4 were treated nuclear factor-κB/interleukin-8) signalling pathway. Conversely, with DMAPT (0 and 1 μM) or control vehicle after 24 h of transfection overexpressed PDK4 in the ALDH−CD44− subpopulation exerted for 48 h. the opposite effects. Our data collectively supported the potential utility of PDK4 as a therapeutic target for management of ovarian Treatment with IL-8 and CXCR1 neutralising antibody cancer driven by ascites-derived CSCs. Following transfection with PDK4 or control vector plasmid for 24 h, ALDH−CD44− SKOV3 cells were treated with mouse IgG (2 μg/ mL), neutralising antibody against CXCR1 (2 μg/mL) or recombi- MATERIALS AND METHODS nant IL-8 (100 ng/mL) (R&D) for 48 h. Clinical samples and cell lines, immunohistochemistry (IHC), real- time PCR (qPCR), immunoblot analysis, enzyme-linked immuno- In vivo tumorigenicity analysis via subcutaneous implantation, sorbent assay (ELISA), transwell migration and invasion assays, limiting dilution assay and serial transplantation clonogenic assay, sphere formation assay, metabolic assays and Animal experiments were conducted following protocols the Cancer Genome Atlas Dataset (TCGA) and Gene Expression approved by the Committee of the Use of Live Animals in Profiling Interactive Analysis (GEPIA) are described in Supplemen- Teaching and Research (CULATR) and were carried out under an tary Materials and methods. approved CULATR licence (No. 4598-18). BALB/c nude mice (7–8 weeks), which are athymic with T cell deficiency were used Tumourspheres culture for this study and kept in isolated cages in a sterile area supplied Cells were seeded as single-cell suspensions in 6-well ultra-low with constant air at 25 °C by Laboratory Animal Unit (LAU) of the attachment plates at a density of 2000–5000 cells/well with University of Hong Kong. The life activity and health situation serum-free DMEM/F12 (Dulbecco’s modified Eagle’s medium/F12) of the nude mice were monitored by LAU stuff. Sorted ALDH+ (1:1) (Gibco, MD, USA) supplemented with 10 ng/mL basic CD44+ and ALDH−CD44− cells were harvested and suspended in fibroblast growth factor (Sigma-Aldrich, MO, USA), 20 ng/mL a mixture containing RPMI medium and Matrigel (High Concen- human epidermal growth factor (Sigma-Aldrich) and 5 μg/mL tration, Corning, #354248). Different dilution (3 × 105,10×103,5× insulin. Fresh medium was added to each well every 2 days 103, 2.5 × 103 cells) of ALDH+CD44+ or ALDH−CD44− cells were without withdrawing the old medium. subcutaneously injected into flanks of mice in different groups. Four to six mice were randomised to each group, and tumour Flow cytometry and fluorescence-activated cell sorting incidence and latency were recorded. The estimated frequency of Anti-CD44-APC, anti-IgG-APC (Miltenyi Biotec, Germany) and CSCs was calculated using the Extreme Limiting Dilution Analysis ALDEFLUORTM kits (StemCell Technologies, Canada) were software (http://bioinf.wehi.edu.au/software/elda/). Experimental employed for flow cytometry analysis according to the manu- mice were maintained for 2 months and were killed with 150 mg/ facturer’s instructions. ALDH and CD44 double-positive and kg pentobarbital, followed by tumour dissection and recording of double-negative cells were isolated via fluorescence-activated cell tumour sizes. To confirm the lack of tumour formation in mice sorting (FACS). Cells suspended in ALDEFLUOR™ Assay Buffer were exhibiting no tumour nodules, injection sites were opened and incubated with ALDEFLUOR™ Reagent for 30 min at 37 °C. After observed. Isolated tumours were collected and dissociated into Ascites-derived ALDH+CD44+ tumour cell subsets endow stemness... Y.-X. Jiang et al. 277 a Primary ovarian tumour Ascites b c Primary ovarian tumour Ascites 15

500 ) 150 Primary 50 150

g Ascites ** 400 A

0.37% n

N Migration

100 16.5% a 40 300

R h **

c ** m

200 ** 10 100 50

4 ALDH

l 30 100

(

0 0 3 4 5 6 3 4 5 6 P

01010 10 10 01010 10 10 n

e 20 FITC-A FITC-A o

v 5

Invasion 50 600 ** 600 t

e (as % of ascites)

r 10 33.2% 96.4% e

400 p

400 R

x Cell migration/invasion

No. of spheres (5000/well) 0 0 0 200

CD44 200 Migration Invasion Primary Ascites Primary Ascites 0 0 HOSE 96-9-18 3 4 5 6 3 4 5 6

Sphere 01010 10 10 01010 10 10

formation FITC-A FITC-A d ehf 5 8 SKOV3 12 Ascites Monolayer SKOV3 ** Monolayer Tumourspheres Tumourspheres Monolayer spheres 8 4 ** Monolayer * 6 Tumourspheres OCT4 4 3 ** 2.0 SOX2 ** ** 4 n.s. 2 ** * ** KLF4 1.5 *

(fold change) BMI1 1 1.0 2 Relative lactate level n.s. PDK4

Relative mRNA expression (fold change) 0 0.5

NANOG OCT4 SOX2 KLF4 BMI1 Relative PDK4 mRNA expression 0 Actin 25 PDK1 PDK2 PDK3 PDK4 0.0 SKOV3 Monolayer 20 ** 30 Ascites SKOV3 OVCAR3 Tumourspheres ** 15 15 Monolayer 10 10 Tumourspheres g Ascites 5 ** 4 ** Monolayer spheres 3 ** ** PDK4 2 ** 5

1 (fold change) Actin

Relative mRNA expression (fold change) 0 OVCAR3 NANOG OCT4 SOX2 KLF4 BMI1 **

Relative PDK4 mRNA expression Monolayer spheres 10 OVCAR3 * Monolayer 0 PDK4 Tumourspheres Ascites SKOV3 OVCAR3 8 * Actin 6

4 **

2 ** n.s.

Relative mRNA expression (fold change) 0 NANOG OCT4 SOX2 KLF4 BMI1 Fig. 1 Ascites-derived tumour cells and tumourspheres show CSC properties and express higher levels of PDK4. a Transwell migration/ invasion and sphere formation assays of cancer cells derived from primary ovarian tumours and ascites. (Left) Representative images of migrating or invading cells (scale bar, 100 μΜ) and sphere formation (scale bar, 50 μΜ). (Middle) Cell migration and invasion are presented as a percentage of ascites. (Right) Quantification of cells forming tumourspheres. b Flow cytometry analysis of ALDH and CD44 in single cells isolated from primary ovarian tumour or ascites. c Relative PDK4 expression in HOSE 96-9-18 and cancer cells derived from primary ovarian tumours and ascites determined via qPCR. d qPCR analysis of relative stemness genes, NANOG, OCT4, SOX2, KLF4 and BMI1, of tumourspheres formed from ascites-derived tumour cells and ovarian cancer cell lines (SKOV3 and OVCAR3), compared to monolayer cells. e (Upper) Relative mRNA expression of PDK 1–4 in tumourspheres and monolayer cells isolated from SKOV3 cells determined using qPCR. (Lower) Relative PDK4 mRNA expression in tumourspheres and monolayer cells isolated from ascites, SKOV3 and OVCAR3 cells. f Immunoblotting showing protein expression of OCT4, SOX2, KLF4, BMI1 and PDK4 of tumourspheres and monolayer cells isolated from SKOV3 cells. g Immunoblotting showing PDK4 expression in tumourspheres and monolayer cells isolated from ascites cancer cells and OVCAR3 cells. h Relative lactate production of tumourspheres and monolayer cells isolated from ascites, SKOV3 and OVCAR3 cells after 24 h incubation assessed using Lactate Colorimetric Assay Kit II (*p < 0.05, **p < 0.01, results represent means ± SD from three independent experiments; n.s., not significant). single cells. The harvested cells were counted and subsequently Plotter, a free online tool (www.kmplot.com), was utilised. Hazard injected into secondary recipient mice. ratios and log-rank p values were calculated automatically based To determine the effects of DCA on tumour initiation in vivo, on the optimal cut-off value auto-selected with the software. FACS-sorted ALDH+CD44+ SKOV3 cell subpopulations were treated with DCA (10 mM) or control vehicle (ddH2O) for 3 days and harvested. Serial dilutions (3 × 105,10×103,5×103, 2.5 × 103 RESULTS cells) of treated cells were injected into mice, and tumour Ascites-derived tumour cells exhibit CSC properties and express incidence and latency were recorded. Tumour-initiating frequency high levels of PDK4 was calculated as above and tumour sizes of 3 ×1 05 cells We isolated ovarian cancer cells from primary tumours and ascites measured every 3–5 days. samples. The epithelial nature of primary cells was confirmed based on positivity for the epithelial markers, CK7 and AE1/AE3, Statistical analysis and negativity for the fibroblast marker, CD45 (Supplementary Values are presented as the means ± SEM. The GraphPad Prism Fig. S1). Transwell migration and invasion assays revealed higher 7.0 software was applied to plot data with a two-tailed t test or cell migration and invasion capabilities of ascites-derived tumour one-way analysis of variance. P values <0.05 were regarded as cells than primary ovarian tumour cells (Fig. 1a). In the sphere statistically significant (*p < 0.05 and **p < 0.01). To determine the formation assay, larger and higher numbers of spheroids were prognostic effects of PDK4 on ovarian cancer, Kaplan–Meier observed from ascites-derived tumour cells relative to primary Ascites-derived ALDH+CD44+ tumour cell subsets endow stemness... Y.-X. Jiang et al. 278 a +DEAB/CD44– isotype control ALDH/CD44

105 Q1 Q2 105 Q1 Q2 105 0.13 0.029 11.4 27.9 ALDH+CD44+

104 104 104 5%

103 103 103

102 102 102

5% CD44 (APC) log 101 101 101 Q4 Q3 Q4 Q3 ALDH-CD44– 99.8 0.038 32.5 28.2 100 100 100 100 101 102 103 104 105 100 101 102 103 104 105 100 101 102 103 104 105 ALDH (FITC) log

b c ALDH+CD44+ ALDH-CD44-

ALDH+CD44+ ALDH–CD44– 150 ALDH+CD44+ 150 ALDH–CD44– ALDH+CD44+ ALDH–CD44–

100 SKOV3 100 SKOV3 * 50 50 Cell migration ** Migration

** Colony formation (as % of ALDH+CD44+) Colony proliferation (%) ** 0 0 OVCAR3 OVCAR3 SKOV3 OVCAR3 SKOV3 OVCAR3

150 ALDH+CD44+ ALDH+CD44+ ALDH–CD44– d 80 ALDH–CD44–

100 60 SKOV3 SKOV3

Invasion 50 Cell invasion ** 20 ** Sphere formation

(as % of ALDH+CD44+) ** ** OVCAR3 0 OVCAR3 SKOV3 OVCAR3 No. of spheres (5000/well) 0 SKOV3 OVCAR3

e f 1.5 1.5 1.5 SKOV3 SKOV3 ALDH+CD44+ OVCAR3 ALDH+CD44+ ALDH–CD44– ALDH–CD44–

1.0 1.0 1.0 * * * 0.5 * 0.5 ** ** 0.5

** Relative cell viability * ** ** Control ** Cisplatin (20 µM) Relative mRNA expression (fold change) Relative mRNA expression (fold change) 0.0 0.0 0.0 NANOG OCT4 SOX2 KLF4 BMI1 NANOG OCT4 SOX2 KLF4 BMI1 ALDH+CD44+ ALDH-CD44- ghi j k 1.2 1.5 SKOV3 1000 SKOV3 10000 SKOV3 SKOV3

) SKOV3

1.0 e

g 800 8000 A

** n ALDH+ ALDH–

R 0.8 ** h 1.0 CD44+ CD44–

4 d 600 * 6000 l

f PDK4

D 0.6 (

400 4000 i

i *

0.4 t 0.5

a Actin

e Relative lactate level

p 200 2000 R 0.2 x

e Oxygen consumption (RFU) Extracellular acidification (RFU) 0.0 0 0 0.0

ALDH+CD44+ALDH–CD44– ALDH+CD44+ALDH–CD44– ALDH+CD44+ALDH–CD44– ALDH+CD44+ALDH–CD44– l Primary implantation L: ALDH+CD44+ R: ALDH–CD44– Tumour incidence ALDH+CD44+ Group Frequency P value 3 × 105 104 5000 2500 Primary implantation ALDH+CD44+ 4/4 (100%) 5/6 (83.3%) 3/5 (60%) 2/5 (40%) 1/5384 ALDH–CD44– 0.00143 ALDH–CD44– 4/4 (100%) 1/6 (16.7%) 1/5 (20%) 0/5 (0) 1/44205 Secondary implantation ALDH+CD44+ 4/4 (100%) ALDH–CD44– 0/4 (0) Secondary implantation

ovarian tumour cells (Fig. 1a). Additionally, higher ALDH and cells, compared to primary ovarian tumour cells and the normal CD44 activities were detected in ascites-derived tumour cells than ovarian epithelial cell line, HOSE 96-9-18 (Fig. 1c). Our findings primary ovarian tumour cells (Fig. 1b). By qPCR analyses, clearly suggested that ascites-derived tumour cells displayed CSC significantly higher PDK4 expression in ascites-derived tumour properties with increased PDK4 expression. Ascites-derived ALDH+CD44+ tumour cell subsets endow stemness... Y.-X. Jiang et al. 279 Fig. 2 ALDH+CD44+ cells derived from ovarian cancer cells shows enhanced CSC properties and PDK4 expression. a FACS-mediated isolation of ALDH+CD44+ and ALDH−CD44− SKOV3 cells. (Left) Isotype control stained with DEAB and FITC IgG. (Middle) The top 5% cells most brightly stained for ALDH+CD44+ or bottom 5% with minimal staining for ALDH−CD44− cells were collected. (Right) Purity of ALDH and CD44 was confirmed via post sorting using flow cytometry. b Transwell migration/invasion (scale bar, 100 μΜ) and c clonogenic assays of ALDH+CD44+ and ALDH−CD44− cells isolated from SKOV3 and OVCAR3 cells were imaged and presented as a percentage of ALDH−CD44− groups. d Sphere formation assay of ALDH+CD44+ and ALDH−CD44− cells isolated from SKOV3 and OVCAR3 cells, followed by imaging (scale bar, 50 μΜ) and quantification of the number of cells that formed tumourspheres. e mRNA expression of relative stemness genes, NANOG, OCT4, SOX2, KLF4, and BMI1, in ALDH+CD44+ and ALDH−CD44− cells isolated from SKOV3 and OVCAR3 cells. f Relative cell viability of ALDH+CD44+ and ALDH−CD44− SKOV3 cells after treatment with or without 20 μM cisplatin for 48 h, determined with the XTT assay. g Relative lactate production of ALDH+CD44+ and ALDH−CD44− SKOV3 cells after 24 h incubation. h Extracellular acidification and i OCR assays of ALDH+CD44+ and ALDH−CD44− SKOV3 cells after 2 h incubation measured using the glycolysis and extracellular oxygen consumption assays, respectively (RFU: relative florescence units). Relative PDK4 j mRNA and k protein expression of ALDH+CD44+ and ALDH−CD44− SKOV3 cells via qPCR and immunoblotting, respectively. l (Left) Representative images of xenograft tumours resected from mice inoculated with ALDH+CD44+ and ALDH−CD44− SKOV3 cells (n = 4). Resected tumours were re-inoculated into the flanks of the second batch of mice. (Right) Summary of tumour incidence and estimated frequency of ALDH+CD44+ and ALDH−CD44− SKOV3 cells determined via in vivo limiting dilution and self-renewal assays (*p < 0.05, **p < 0.01, results are presented as means ± SD from three independent experiments).

Tumourspheres display CSC properties and express higher PDK4 Next, we explored the differences in glycolytic metabolism levels than monolayer cells between the two subsets. Significantly higher lactate production Tumourspheres cultured in ultra-low attachment plates containing was observed in ALDH+CD44+, compared with the ALDH−CD44− CSCs provide an efficient means to enrich CSCs in ovarian SKOV3 subset (Fig. 2g). In the glycolysis assay, extracellular cancer.14 Primary tumoursphere cultures were established with acidification rate reflecting lactic acid production through the ascites-derived tumour cells and ovarian cancer cell lines, SKOV3 glycolytic pathway was consistently upregulated in ALDH+CD44+ and OVCAR3. qPCR analysis led to the identification of significantly SKOV3 cells, compared with the ALDH−CD44− subset (Fig. 2h). upregulated messenger RNA (mRNA) of stemness genes in Oxygen consumption rate (OCR), a sensitive indicator of the tumourspheres, compared with the bulk of parental cells cultured OXPHOS pathway, was lower in ALDH+CD44+ SKOV3 relative to in monolayer conditions (Fig. 1d). Consistently, immunoblotting the ALDH−CD44 subset (Fig. 2i). Moreover, PDK4 expression was revealed higher protein expressions of stemness genes in significantly higher in ALDH+CD44+ SKOV3 cells at both mRNA tumourspheres relative to monolayer cells in SKOV3 (Fig. 1f). (Fig. 2j) and protein (Fig. 2k) levels. Our data collectively indicated Among the four PDK isoforms, PDK2 and PDK4 levels were that the ALDH+CD44+ subset displayed a distinct metabolic significantly higher in tumourspheres than monolayer cells in the profile with increased glycolytic influx and decreased OXPHOS SKOV3 group, with the greatest increase observed for PDK4 activity. (Fig. 1e). Significant PDK4 mRNA expression was additionally Two significant CSC characteristics (tumour initiation and self- detected in tumourspheres cultured from ascites-derived tumour renewal) of ALDH+CD44+ cells were investigated using in vivo cells and OVCAR3 (Fig. 1e). In addition, PDK4 protein expression tumorigenesis and limiting dilution assays. In the in vivo was higher in tumourspheres from ascites-derived tumour cells tumorigenesis assay, mice in the ALDH+CD44+ group formed (Fig. 1g) and ovarian cancer cell lines (Fig. 1f, g), as determined significantly larger tumours, compared to the ALDH−CD44− from immunoblot analyses. These findings suggested that group (Fig. 2l). The in vivo limiting dilution assay involved tumourspheres derived from ascites-derived tumour cells and cell injection of a series of cell dilutions into either the left or right lines maintain CSC properties. Furthermore, enhanced lactate flanks of nude mice. ALDH+CD44+ cells displayed enhanced production in tumourspheres relative to monolayer cells was tumour-initiating activity, as evident from the increased tumour demonstrated by lactate assay (Fig. 1h). incidence and significantly higher estimated tumour-initiating cell frequency. Tumours were collected, minced and dissociated ALDH+CD44+ cells isolated from ascites-derived tumour cells for secondary implantation onto another batch of nude mice to show enhanced CSC properties and PDK4 expression assess the self-renewal capacity of ALDH+CD44+ cells. After In view of the increased ALDH and CD44 expression in ascites- 8 days, new tumours were detected in the flanks of all four derived tumour cells, FACS was employed to isolate ALDH+CD44+ mice in the ALDH+CD44+ group. However, no tumours were and ALDH−CD44− subsets of SKOV3 and OVCAR3 cell groups. The present in the secondary implantation sites of nude mice from top 5% cells showing highest staining levels for ALDH+CD44+ and the ALDH−CD44− group (Fig. 2l). Based on these findings, we bottom 5% cells with lowest staining for ALDH−CD44− were used conclude that dual positivity for ALDH and CD44 defines a in subsequent functional assays. The purity of the ALDH and CD44 compelling marker set for isolation of the CSC subpopulation of cell subsets was confirmed via post-sorting using flow cytometry ovarian cancer. (Fig. 2a). Compared with the ALDH−CD44− subset, ALDH+CD44+ SKOV3 cells showed enhanced migration and invasion abilities PDK4 is overexpressed in ovarian cancer and correlated with (Fig. 2b). The clonogenic assay was employed to examine the metastasis and poor prognosis activity of a single cell to develop into a large colony through PDK4 protein expression as determined from tissue microarray clonal expansion, a sensitive indicator for CSCs.15 Notably, ALDH+ analyses was evaluated via IHC. Owing to predominant localisation CD44+ SKOV3 cells formed a greater number and larger-sized in the cytoplasm, PDK4 staining was stronger in malignant tumour colonies than the ALDH−CD44− subset (Fig. 2c), along with an tissues relative to normal ovarian tissues/benign tumours (Fig. 3a) enhanced sphere-forming capacity (Fig. 2d). As expected, (p < 0.01). PDK4 protein expression was significantly higher in the increased mRNA expression of stemness genes was detected in advanced stages (stage 3) versus lower stages (stages 1 and 2). ALDH+CD44+ cancer cells, compared with the ALDH−CD44− Serous ovarian tumours showed significantly higher expression of subset (Fig. 2e). Accumulating evidence suggests that chemother- PDK4, compared to non-serous tumours, supporting a significant apy targets the bulk of tumour cells, whereas CSC subpopulations correlation between upregulation of PDK4 and ovarian carcinoma of tumours show a reduced response to chemotherapeutic drugs.16 progression (Fig. 3b). In addition, we found that PDK4 mRNA was Compared with the ALDH−CD44− subset, ALDH+CD44+ SKOV3 upregulated in ovarian cancer cell lines (SKOV3, OVCAR3, cells exhibited greater chemoresistance (Fig. 2f). OVCA420, TOV112D and OVTOKO), compared to the normal Ascites-derived ALDH+CD44+ tumour cell subsets endow stemness... Y.-X. Jiang et al. 280 a i ii iii iv v

1.2 b 1.2 1.2 1.2 ** n.s. ** n.s. ** 1.0 ** 1.0 ** ** 1.0 1.0

0.8 0.8 0.8 0.8

0.6 0.6 0.6 0.6 Positivity Positivity Positivity Positivity 0.4 0.4 0.4 0.4 Total number of pixels) Total number of pixels) Total number of pixels) Total number of pixels) 0.2 0.2 0.2 0.2 (number of positive pixels/ (number of positive pixels/ (number of positive pixels/ (number of positive pixels/ 0.0 0.0 0.0 0.0 Normal ovary or Malignant Stage 1 Stage 2 Stage 3 Grade I–II Grade II–III Serous Mucinous Endometrioid Clear cell benign tumours tumours (n =53) (n =19) (n =25) (n =31) (n =66) (n =50)(n =18) (n =28) (n =2) (n =5) (n =97)

c 1.0 1.0 HR = 1.48 (1.14–1.91) HR = 1.47 (1.13–1.91) Log-rank P = 0.0029 Log-rank P = 0.0036

0.8 0.8

0.6 0.6 Probability Probability 0.4 0.4

0.2 0.2

Expression Expression Low Low 0.0 High 0.0 High

0 50 100 150 0 50 100 150 Time (months) Time (months) Number at risk Number at risk Low 413 97 10 2 Low 386 38 2 1 High 144 23 2 0 High 136 3 0 0

d e Primary ovarian tumours Matched metastatic tumours 60

i ii 1.2 ** ** 1.0 40 **

0.8

0.6 of PDK4 20 ** Positivity 0.4 iii iv ** Total number of pixels) 0.2

(number of positive pixels/ ** Relative mRNA expression 0 0.0 Primary tumour Metastatic foci (n =16) (n =16) SKOV3 OVCAR3OVCA420TOV112DOVTOKO

HOSE 96-9-18 Fig. 3 PDK4 is overexpressed in ovarian cancer and correlates with metastasis and poor prognosis. a Representative images (scale bar, 100 μΜ) of IHC staining for PDK4 expression in (i) normal ovarian tissue, (ii) serous, (iii) mucinous, (iv) endometrioid and (v) clear cell subtypes of tumour samples on ovarian cancer TMA (OVC1021, Biomax). b Correlation of PDK4 with clinicopathologic parameters in ovarian cancer. c (Left) Overall and (right) progression-free survival rates were analysed with the log-rank test for PDKhigh/low groups of ovarian cancer patients obtained from TGCA using the Kaplan–Meier plotter. d (Left) Representative images (scale bar, 100 μΜ) of IHC staining for PDK4 expression in ovarian primary and matched metastatic tumours. (Right) Dot plot showing PDK4 positivity in primary and metastatic ovarian tumours. e PDK4 mRNA expression in normal human ovarian surface epithelial cell line (HOSE) and ovarian cancer cell lines as determined by qPCR (*p < 0.05, **p < 0.01; n.s., not signiﬁcant).

human ovarian surface epithelial cell line, HOSE 96-9-18, by qPCR PDK4 shifts ovarian CSC energy metabolism from OXPHOS to (Fig. 3e). aerobic glycolysis and is crucial for CSC maintenance in vitro and To assess the prognostic effect of PDK4 mRNA in clinical in vivo samples, we used the Kaplan–Meier Plotter with TCGA database in To ascertain the correlation between PDK4 overexpression and which 522 patients were grouped based on the optimal cut-off altered CSC properties in ovarian cancer, we generated transient value auto-selected with the software. As shown from the survival siPDK4-transfected tumourspheres/ALDH+CD44+ cellsaswellas curves (Fig. 3c), high PDK4 mRNA expression was correlated with ALDH−CD44− cells transiently overexpressing PDK4. qPCR poor overall (p = 0.0029) and progression-free (p = 0.0036) (Fig. 4a) and immunoblot analyses (Fig. 4b) were employed to survival. The data indicated that PDK4, highly expressed in ovarian confirm transfection efficiency. PDK4 knockdown in tumour- cancer tissues, was associated with poor patient outcome. spheres/ALDH+CD44+ cells led to a significant decrease in Moreover, significantly higher immunostaining of PDK4 was migratory/invasive (Fig. 4c), clonogenic (Fig. 4d) and sphere- observed in metastatic foci than in the corresponding primary forming (Fig. 4e) capacities, while transient PDK4 overexpression tumours from 16 pairs of FFPE tissue sections (Fig. 3d) (p < 0.01), exerted the opposite effects. Since cell death could affect the supporting association of PDK4 with metastasis in ovarian cancer. effects of PDK4 as presented above, we further evaluated the cell Ascites-derived ALDH+CD44+ tumour cell subsets endow stemness... Y.-X. Jiang et al. 281

a ALDH+CD44+ 1.5 15

) 1.5 Tumourspheres Control ) ALDH–CD44–

)

e Control

A siPDK4 A

n **

a a siPDK4

m 10

1.0 1.0

l 4

(

v v 0.5 5 i * i

s 0.5 s

t t **

R R **

e e ** 0.0 0.0 ** 0 Ascites SKOV3 OVCAR3 SKOV3 OVCAR3 Vector PDK4 SKOV3

b Ascites SKOV3 OVCAR3 SKOV3 Tumourspheres Tumourspheres ALDH+CD44+ Tumourspheres ALDH+CD44+ ALDH–CD44– Control siPDK4 Control siPDK4 Control siPDK4 Control siPDK4 Control siPDK4 Vector PDK4 PDK4 PDK4-tGFP Actin Actin

ALDH–CD44– c Tumourspheres ALDH+CD44+ ALDH+CD44+ Tumourspheres 150 Vector PDK4 Control siPDK4 150 Control Control siPDK4 Control siPDK4 siPDK4 100 100

* Migration SKOV3 SKOV3 50 **

50 SKOV3 Migration

Migration ** ** 0 0 Cell migration (as % of control) Invasion Cell migration (as % of control) SKOV3 OVCAR3 OVCAR3 SKOV3 OVCAR3 OVCAR3

150 ALDH+CD44+ 600 150 Tumourspheres Control Contro siPDK4 ** lsiPDK4 100 100 SKOV3 400 SKOV3 ** * SKOV3

50 Invasion 50 200

Invasion * * ** (as % of control) Control

Cell migration/invasion PDK4 OVCAR3

Cell invasion (as % of control) 0

Cell invasion (as % of control) 0 OVCAR3 0 SKOV3 OVCAR3 SKOV3 OVCAR3 Migration Invasion ALDH–CD44–

de Tumourspheres ALDH+CD44+ 150 Tumourspheres 80 Control Control siPDK4 ALDH+CD44+ Control siPDK4 siPDK4 Control 100 * 60 siPDK4

SKOV3 40

50 Ascites *

Colony proliferation (%) ** 60 Tumourspheres Control 20 0 siPDK4 SKOV3 OVCAR3 ** 0 Number of spheres (2000/well) 150 40 Sphere formation ALDH+CD44+ SKOV3 OVCAR3 OVCAR3 SKOV3 80 100 20 * ALDH–CD44– Sphere formation ALDH-CD44- 60 * ** * Vector PDK4 50 * No. of spheres (2000/well) 0 Ascites SKOV3 OVCAR3 40 OVCAR3 Colony proliferation (%) 0 20 Control siPDK4 Sphere formation SKOV3 0

Number of spheres (2000/well) Vector PDK4 SKOV3

fgControl Control n 5

n 1.5 n 1.5

o Vector

i SKOV3

o o

i i Tumourspheres siPDK4 ALDH+CD44+ siPDK4 s ALDH–CD44– **

s s

s PDK4

s s e 4 ALDH+CD44+

)

e e

) )

r r

e e

p p

g x SKOV3

g g x 1.0 x 1.0

n Control siPDK4

e e n n 3

a **

a a A ALDH-CD44-

A A

h h

N N

c c

d R R OCT4

l d d 2 Vector

l l PDK4 ** m

m m *

o o

0.5 0.5 ** f

f ** f **

e ( **

e e

( (

v v

i i t 1

t t

* a KLF4

a a l KLF4

l l

e e

R R 0.0 R 0.0 0 OCT4 KLF4 BMI1 OCT4 KLF4 BMI1 OCT4 KLF4 BMI1 BMI1 BMI1 SKOV3 SKOV3 SKOV3

hiALDH–CD44– j ALDH–CD44– 1.5 1.5 800 ALDH+CD44+ 800 10,000 8000 ALDH+CD44+ ALDH–CD44– ** ALDH+CD44+ ** 8000 ** 600 * 600 6000 ** 1.0 1.0 6000 ** 400 400 4000 4000 0.5 0.5 Extracellular 200 Extracellular 200 2000

acidification (RFU) 2000 acidification (RFU) Relative lactate level Relative lactate level Oxygen consumption (RFU) 0.0 0.0 0 0 Oxygen consumption (RFU) 0 0 Control siPDK4 Vector PDK4 Control siPDK4 Vector PDK4 Control siPDK4 Vector PDK4 SKOV3 SKOV3 SKOV3 SKOV3 SKOV3 SKOV3

viability of tumourspheres and ALDH+CD44+ cells after PDK4 the effects on cell migration and invasion (determined after 1- knockdown. PDK4 knockdown in tumourspheres and ALDH+ day incubation) would not be related to cell death. We also CD44+ cells from SKOV3 and OVCAR3 did not show alteration in found the cell viability of PDK4-depleted tumourspheres and cell viability 1 day after incubation as determined using XTT and ALDH+CD44+ cells from SKOV3 as 0.737–0.822-fold and cell counting methods (Supplementary Fig. S2), suggesting that OVCAR3 as 0.833–0.910-fold after 3 (SKOV3) and 5 (OVCAR3) Ascites-derived ALDH+CD44+ tumour cell subsets endow stemness... Y.-X. Jiang et al. 282 Fig. 4 PDK4 shifts the mode of energy metabolism and is crucial for ovarian CSC maintenance. a Relative PDK4 mRNA and b protein expression in tumourspheres and ALDH+CD44+ cells transfected with control siRNA or siPDK4 and PDK4-overexpressing ALDH−CD44− cells, determined via qPCR and immunoblotting, respectively. c Images and quantification of data from transwell migration/invasion assays of (left) tumourspheres and (middle) ALDH+CD44+ cells transfected with control siRNA or siPDK4 and (right) PDK4-overexpressing ALDH−CD44− cells. Scale bar, 100 μΜ. d Clonogenic assays of (upper) tumourspheres and (lower) ALDH+CD44+ cells transfected with control siRNA or siPDK4 are quantified as a percentage of control groups. e Sphere formation assay of (left) tumourspheres and (right upper) ALDH+CD44+ cells transfected with control siRNA or siPDK4 and (right lower) PDK4-overexpressing ALDH−CD44− cells, followed by imaging (scale bar, 50 μΜ) and quantification of the number of cells that formed tumourspheres. f Relative mRNA and g protein expression of stemness genes OCT4, KLF4 and BMI1 in tumourspheres and ALDH+CD44+ cells transfected with control siRNA or siPDK4 and PDK4-overexpressing ALDH−CD44− cells determined via qPCR and immunoblotting, respectively. h Relative lactate production, i extracellular acidification and j OCR arrays of (left) PDK4-suppressed ALDH+CD44+ and (right) PDK4-overexpressing ALDH−CD44− SKOV3 cells after 24 h incubation (*p < 0.05, **p < 0.01, results represent means ± SD from three independent experiments).

a ALDH+CD44+ b

Control DCA (5 mM) DCA (10 mM) 60 ALDH+CD44+ ALDH+CD44+ Control DCA (5 mM) DCA (10 mM) 40 ** Migration ** 20 Sphere SKOV3 formation

0 Number of spheres (2000/well)

Invasion 0510DCA (mM) SKOV3 c 150 ALDH+CD44+ 150 ALDH+CD44+ 1.2 ALDH+CD44+ n.s. 1.0

100 100 0.8 0.6 ** * 50 * 50 0.4 ** ** Control * Relative cell viability 0.2 DCA (10 mM)

0 Cell invasion (as % of control) 0 0.0 Cell migration (as % of control) 0510DCA (mM) 0510DCA (mM) 01020Cisplatin (µM) SKOV3 SKOV3 SKOV3 de 1.5 ALDH+CD44+ Control n.s. Tumour formation time (days), n =5 n.s. Group 1.0 #1 #2 #3 #4 #5 Mean SEM Medium ** * Control 7 9 11 11 13 10.2 1.01 11 ** ** Control DCA 0.5 DCA (5 mM) DCA (10 mM) 24 28 28 25 23 25.6 1.02 25 (fold change) DCA (10 mM)

Relative mRNA expression 0.0 OCT4 KLF4 BMI1 SKOV3

f ALDH+CD44+ 150 ALDH+CD44+ 300 Control Tumour incidence

) Group Frequency P value 3 DCA (10 mM) 5 100 3 × 10 104 5000 2500 200 Control 5/5 (100%) 4/5 (80%) 4/5 (80%) 3/5 (60%) 1/3973 0.0363 DCA (10 mM) 5/5 (100%) 4/5 (80%) 1/5 (20%) 0/5 (0%) 1/12437 50 ** 100 Tumour weight (mg)

Tumour size (mm

0 0 15 20 25 30 010DCA (mM) Days post tumour inoculation SKOV3 Fig. 5 DCA hinders ovarian CSC properties in vitro and in vivo. a Transwell migration/invasion (scale bar, 100 μΜ) and b sphere formation assays (scale bar, 50 μΜ) of ALDH+CD44+ SKOV3 cells after treatment with vehicle and DCA (5 and 10 mM) for 48 h. c Relative cell viability of vehicle or 10 mM DCA-treated ALDH+CD44+ SKOV3 cells after treatment with or without 20 μM cisplatin for 48 h with the XTT assay. d qPCR analysis of mRNA expression of relative stemness genes, OCT4, KLF4 and BMI1, of ALDH+CD44+ SKOV3 cells after treatment with DCA (5 and 10 mM). Following injection of 3 × 105 ALDH+CD44+ SKOV3 cells pre-treated with control (n = 5) DCA (n = 5) into the right/left ﬂanks of mice. e (Left) Representative images of xenograft tumours and (right) time of tumour formation (days) in the xenograft model were recorded. f (Left) Tumour size curves and (middle) weights of xenografts derived from control (n = 5) and DCA (n = 5)-pre-treated ALDH+CD44+ SKOV3 cells. (Right) The tumour initiation rate was recorded based on the in vivo limiting dilution assay (*p < 0.05, **p < 0.01, results represent means ± SD of three independent experiments; n.s., not signiﬁcant).

days incubation compared to control as determined using XTT We further used GEPIA to predict the correlations between and cell counting methods (Supplementary Fig. S2), suggesting PDK4 and ﬁve stemness genes (NANOG, OCT4, SOX2, KLF4 and that the decreased sphere-forming ability would not result BMI1) that are highly expressed in tumourspheres and ALDH+ primarily from cell death. CD44+ cells. Among the ﬁve genes, KLF4 and BMI1 mRNA levels Ascites-derived ALDH+CD44+ tumour cell subsets endow stemness... Y.-X. Jiang et al. 283 a ALDH+CD44+ Control b 12,000 ALDH+CD44+ SKOV3 AKT(S473) ALDH+CD44+ ALDH–CD44– CREB(S133) 10,000 WNK1(T60) Control siPDK4 Vector PDK4 ALDH+CD44+ALDH–CD44– 8000 p-STAT3 STAT3(S727) 6000 Hck(Y411) STAT3 siPDK4 4000 p-AKT

Mean pixel density SKOV3 2000 Control AKT siPDK4 0 p-p65 p65 AKT (S473) Hck (Y411) STAT3 (S727) CREB (S133) WNK1 (T60)

cdeALDH+CD44+ ALDH+CD44+ 1.5 ALDH–CD44– ALDH–CD44– 1.5 2.5 2.5 1.5 1.5 * 2.0 2.0 ** 1.0 1.0 1 1 1.5 1.5 * * * 1.0 1.0 0.5 0.5 0.5 * 0.5 (fold change) (fold change) (fold change)

Relative IL-8 mRNA 0.5 0.5 Relative IL-8 mRNA Relative IL-8 mRNA Relative IL-8 secretion Relative IL-8 secretion Relative IL-8 secretion expression (fold change) expression (fold change) 0.0 0.0 0.0 0.0 expression (fold change) 0 0 Control siPDK4 Control siPDK4 Vector PDK4 Vector PDK4 ALDH+CD44+ ALDH–CD44– ALDH+CD44+ ALDH–CD44– SKOV3 SKOV3 SKOV3 SKOV3 SKOV3 SKOV3

fg3 ALDH–CD44– 8 ALDH–CD44– SKOV3 ALDH+CD44+ * µ * 6 ** 0 7.5 10 Stattic ( M) 2 * p-STAT3 ** ** 4 STAT3

1 p-AKT 2 AKT Relative KLF4 mRNA Relative OCT4 mRNA expression (fold change) expression (fold change) SKOV3 SKOV3 p-p65 0 0 + – + + – + IgG + – + + – + IgG p65 – + – – + – CXCR1 antibody – + – – + – CXCR1 antibody KLF4 – – + – – + Recombinant IL-8 – – + – – + Recombinant IL-8 BMI1 Vector PDK4 Vector PDK4 Actin

hjALDH+CD44+ i ALDH–CD44– Vector PDK4 Control DMAPT (1 µM) 1.5 ALDH+CD44+ Control Sphere

formation 1.0 M) ** ** ** µ 60 ALDH+CD44+ **

** Sphere formation 0.5 SKOV3

(fold change) ** Control 40 DMAPT (1 DMAPT (1 µM)

* Relative mRNA expression 0.0 80 ALDH-CD44– 20 OCT4 SOX2 KLF4 BMI1 ALDH1 CD44 ** 60

0 Number of spheres (5000/well) * 01DMAPT (µM) 40 SKOV3 SKOV3 20 Control DMAPT (1 µM) 0 Number of spheres (2000/well) Vector PDK4

k 5 ALDH–CD44– 8 ALDH–CD44– l ** IL-8 4 ** 6 3 Cell membrane CXCR1 Cytoplasm 4 * 2 * SKOV3 SKOV3 PDK4 2 1 Control Control P Relative mRNA KLF4 Relative mRNA OCT4 µ µ expression (fold change) DMAPT (1 M) expression (fold change) DMAPT (1 M) STAT3 0 0 P Vector PDK4 Vector PDK4 AKT 4 ALDH–CD44– P 3 ** p50 p65 Target genes 2 * Nucleus P regulate CSC p50 p65 properties SKOV3 e.g. IL-8, OCT4, KLF4 1

Relative mRNA IL-8 Control µ

expression (fold change) DMAPT (1 M) 0 Vector PDK4

were positively correlated with PDK4 (Supplementary Fig. S3). cancer cells transfected with siPDK4 exhibited lower expression of NANOG, OCT4, and SOX2 displayed positive associations with OCT4, KLF4, and BMI1 genes at both mRNA (Fig. 4f) and protein PDK4, but not to a significant extent. Additionally, tumourspheres levels (Fig. 4g). Consistent with these findings, introduction of with suppression of PDK4 showed downregulated transcription of PDK4 in ALDH−CD44− cells led to enhanced OCT4, KLF4, and OCT4, KLF4 and BMI1 stemness genes (Fig. 4f). ALDH+CD44+ BMI1 mRNA (Fig. 4f), and KLF4 and BMI1 protein expression Ascites-derived ALDH+CD44+ tumour cell subsets endow stemness... Y.-X. Jiang et al. 284 Fig. 6 PDK4 mediates ovarian CSC characteristics through STAT3/AKT/NF-κB/IL-8 signalling. a (Left) Immunoblotting images and (right) quantification of deregulated phospho-kinases spotted on the phosphor-kinase array. b Immunoblotting showing protein expression of p- STAT3/STAT3, p-AKT/AKT and p-p65/p65 in (left) PDK4-suppressed ALDH+CD44+ SKOV3 cells, (middle) PDK4-overexpressing ALDH−CD44− SKOV3 cells and (right) ALDH+CD44+/ALDH−CD44− SKOV3 cells. qPCR analysis of relative IL-8 mRNA expression and ELISA analysis of relative IL-8 secretion in c PDK4-suppressed ALDH+CD44+ SKOV3 cells, d PDK4-overexpressing ALDH−CD44− SKOV3 cells and e ALDH +CD44+/ALDH−CD44− SKOV3 cells. f mRNA expression of relative stemness genes OCT4 and KLF4 in PDK-overexpressing ALDH−CD44− SKOV3 cells treated with IgG (2 μg/mL), CXCR1 antibody (2 μg/mL) or recombinant IL-8 (100 ng/mL) for 48 h. g Immunoblotting showing expression of p-STAT3/STAT3, p-AKT/AKT, p-p65/p65, KLF4 and BMI1 proteins in ALDH+CD44+ cells treated with the STAT3 inhibitor stattic (0, 7.5 and 10 μΜ) for 48 h. h Sphere formation assay was performed followed by imaging (scale bar, 50 μΜ) and quantification after DMAPT (1 μΜ) treatment of ALDH+CD44+ SKOV3 cells for 48 h. i mRNA expression of relative stemness genes, OCT4, SOX2, KLF4, BMI1, ALDH1 and CD44, in ALDH+CD44+ SKOV3 cells treated with the NF-κB inhibitor, DMAPT, determined via qPCR. j The sphere formation assay was performed, followed by imaging (scale bar, 50 μΜ) and quantification after DMAPT (1 μΜ) treatment of PDK4-overexpressing ALDH−CD44− SKOV3 cells for 48 h. k Relative mRNA expression of OCT4 and KLF4 as well as IL-8 in PDK4-overexpressing ALDH−CD44− SKOV3 cells treated with DMAPT, determined via qPCR. l Proposed mechanism by which PDK4 mediates cancer cell stemness in ovarian cancer through STAT3/AKT/NF-κB/IL- 8 signalling. Whether PDK4 has direct interaction with STAT3, leading to its phosphorylation will be examined in future studies (*p < 0.05, **p < 0.01, results represent means ± SD from three independent experiments).

(Fig. 4g), supporting the conclusion that endogenous PDK4 by which PDK4-regulated stemness in ovarian cancer using ALDH+ regulated CSC properties in ovarian cancer. CD44+ SKOV3 cells transiently transfected with siPDK4 or control Given that tumourspheres/ALDH+CD44+ cells appear to favour siRNA, and ImageJ analysis was applied to quantify the intensity of a metabolic switch from OXPHOS to glycolysis, we evaluated the the spots on the arrays. Five protein kinases (STAT3/S727, AKT/ role of PDK4 in the metabolic shift. ALDH+CD44+ SKOV3 cells S473, CREB/S133, Hck/Y411 and WNK/T60) showed a >1.5-fold with PDK4 knockdown showed decreased lactate production, decrease in siPDK4 treatment groups (Fig. 6a). Decreased p-STAT3 compared with the control group (Fig. 4h), in keeping with S727 and p-AKT S473 protein levels were confirmed via decreased extracellular acidification (Fig. 4i) and increased OCR immunoblotting (Fig. 6b). Phosphorylation at S727 is crucial to (Fig. 4j) exhibited by the PDK4-suppressed ALDH+CD44+ subset. maximise STAT3 transcriptional activity, which promotes tumor- Conversely, transient PDK4 overexpression led to increased lactate igenesis in prostate cancer19 and chronic lymphocytic leukaemia.20 production (Fig. 4h) and extracellular acidification (Fig. 4i), as well Because NF-κB is one of the downstream effectors of STAT3 and as decreased OCR (Fig. 4j) in the ALDH−CD44− subset. AKT,21 we additionally detected siPDK4-mediated reduction of p- p65 S536 protein expression (p-RelA) (Fig. 6b). Phosphorylation of DCA hinders ovarian CSC properties in vitro and in vivo p65 is one of the canonical NF-κB activation indicators.22 Next, we determined whether DCA, a pharmacological PDK Conversely, increased p-STAT3 S727, p-AKT S473 and p-p65 S536 inhibitor,17 could mimic the effects of PDK4 silencing and block protein expression was observed in PDK4-overexpressing ALDH− CSC properties in ovarian cancer cells. DCA is proposed to induce CD44− SKOV3 cells (Fig. 6b). Notably, ALDH+CD44+ cells cell apoptosis in ovarian cancer by inducing a shift from anaerobic displayed higher p-STAT3, p-AKT and p-p65 protein expression to aerobic metabolism.18 To determine the effects of DCA on CSC relative to ALDH−CD44− cells (Fig. 6b). subpopulations in SKOV3 cells, ALDH+CD44+ cells were treated siPDK4 also suppressed mRNA and protein secretion of IL-8, a with DCA (0, 5 and 10 mM) and subjected to the related functional well-known NF-κB downstream target,23 in ALDH+CD44+ cells assays. DCA effectively suppressed migration/invasion (Fig. 5a) (Fig. 6c) and tumourspheres (Supplementary Fig. S5), while and sphere formation (Fig. 5b) of ALDH+CD44+ cells in vitro. We overexpression of PDK4 led to the opposite effects (Fig. 6d). IL-8 also determined cell viability of ALDH+CD44+ cells after 10 mM mRNA and protein secretion was significantly higher in ALDH+ DCA treatment. By XTT assay and cell counting, DCA did not affect CD44+, compared with ALDH−CD44− SKOV3 cells (Fig. 6e). IL-8 is cell viability (Supplementary Fig. S4A, B). In combination with a highly expressed chemokine in ovarian cancers and ascites cisplatin, DCA enhanced the effects of the drug on ALDH+CD44+ associated with poor prognosis.24 Notably, IL-8 is reported to SKOV3 cells by XTT (Fig. 5c) and cell counting (Supplementary regulate breast CSC.25 To examine the effects of IL-8 on PDK4- Fig. S4C). Moreover, DCA suppressed OCT4, KLF4, and BMI1 mRNA mediated stemness, the ALDH−CD44− subset overexpressing levels in the ALDH+CD44+ subset (Fig. 5d). PDK4 was treated with an anti-CXCR1 (an IL-8 receptor) To assess the effects of DCA on tumour formation in vivo, we neutralising antibody or recombinant human IL-8 for 48 h. PDK4 inoculated control or DCA-pre-treated ALDH+CD44+ SKOV3 cells increased OCT4 and KLF4 mRNA expression in ALDH−CD44− cells, into the flanks of nude mice and recorded tumour formation similar to that observed with IL-8 treatment (Fig. 6f). Furthermore, times, sizes and weights. DCA pre-treatment of cells inhibited CXCR1 pre-treatment blocked OCT4 and KLF4 protein expression tumour growth and prolonged tumour formation time from an induced by PDK4 in ALDH−CD44− cells (Fig. 6f). average of 11 to 25 days (Fig. 5e). Additionally, tumour size and Remarkably, ALDH+CD44+ cells treated with the STAT3 inhibitor weight were significantly reduced in DCA-pre-treated mice, stattic (7.5, 10 μM) displayed reduced p-STAT3, p-AKT and p-p65 compared with the control group. To determine the effects of protein expression (Fig. 6g), suggesting that AKT and p65 are DCA on tumour-initiating properties, an in vivo limiting dilution downstream effectors of STAT3. DMAPT, a potent inhibitor of NF- assay was performed with serial dilutions of DCA-pre-treated κB, was shown to inhibit sphere formation capacity (Fig. 6h) and ALDH+CD44+ SKOV3 cells. Tumour incidence was decreased stemness gene mRNA expressions (Fig. 6i). To evaluate the effects from 80% to 20% in the 5000-cell group, and 60% to 0% in the of DMAPT on PDK4-mediated stemness, ALDH−CD44− cells 2500-cell group. In addition, significantly lower estimated tumour- transiently transfected with control or PDK4 plasmids were treated initiating cell frequency was found in DCA-pre-treated group with DMAPT for 48 h. We observed a significant reduction of sphere compared with the control mice (Fig. 5f). In view of these findings, numbers (Fig. 6j) and mRNA expression of the stemness genes we propose that DCA impairs CSC properties in ovarian cancer. (Fig. 6k) in DMAPT-treated relative to control cells in the PDK4- overexpressing ALDH−CD44− cell group, while in the ALDH− PDK4 mediates ovarian CSC characteristics through STAT3/AKT/ CD44− cell group transfected with the vector control, DMAPT had NF-κB/IL-8 signalling no significant effect on sphere formation or stemness gene Multiple pathways are associated with CSC regulation. A expression. Notably, IL-8, a well-known NF-κB downstream target, phosphokinase array was performed to elucidate the mechanism was also reduced in PDK4-overexpressing ALDH−CD44− cells after Ascites-derived ALDH+CD44+ tumour cell subsets endow stemness... Y.-X. Jiang et al. 285 treatment with DMAPT. Our collective findings support the stemness gene expression. By IHC, the presence of PDK4 in involvement of STAT3/AKT/NF-κB/IL-8 signalling in regulating whole ovarian tumour suggested the role of PDK4 in regulating PDK4-mediated ovarian CSCs. not only stemness but also cancer properties in ovarian cancer. The role of PDK4 in ovarian cancer properties will be determined in future studies. DISCUSSION Mechanistically, we have determined the signalling pathways Metastasis, chemoresistance and relapse are frequently observed by which PDK4 regulates tumorigenesis and stemness of ovarian in ovarian cancer despite concerted attempts to optimise cancer. PDK4 is reported to bind CREB in the cytoplasm, prevent treatment options. The residual CSCs subset that survives and its degradation and activate the mTORC1 signalling pathway in grows after treatment contributes to the high recurrence rate of cancer cells.34 Moreover, PDK4 enhances vascular calcification via ovarian cancer. Ascites are present in advanced stages of the SMAD1/5/8 phosphorylation in both the cytoplasm and mito- disease and enriched in free-floating spheroid cells.26 The 3D chondria.37 To clarify the mechanistic pathways by which PDK4 structure of cancer cell aggregates is believed to present an regulates ovarian CSCs, lysates were screened from ALDH+ efficient way to enrich CSC subpopulations.27 These spheroid cells CD44+ SKOV3 cells treated with or without siPDK4 for the exhibit CSC characteristics, such as enhanced migration, invasion phosphokinase array. Knockdown of PDK4 suppressed STAT3/ and sphere formation, as confirmed by our group and other AKT/NF-κΒ activation and IL-8 expression in ALDH+CD44+ researchers.28,29 Consistent with these findings, we showed that SKOV3 cells and the opposite results were obtained upon PDK4 tumourspheres formed from ascites-derived cells express higher overexpression in ALDH−CD44− cells. STAT3/NF-κΒ pathways levels of stemness genes. are shown to be activated in ovarian CSCs.38,39 Egusquiaguirre ALDH and CD44 are associated with tumorigenesis and CSC et al.40 identified a STAT3 target gene, TNFRSF1A, that controls properties in ovarian cancer when used as individual CSC NF-κΒ transcriptional activity through nuclear localisation of p65, markers.30,31 In the present study, we isolated the CSC subset of while another report showed that NF-κΒ-induced IL-6 contributes ascites-derived tumour cells and cell lines with the aid of both to STAT3 activation.41 Additionally, we observed that AKT and ALDH and CD44 markers, which are overexpressed in this tumour p65 are potential downstream effectors of STAT3 in ovarian CSCs. cell population, and validated the ALDH+CD44+ subset as an IL-8 is known to activate AKT, indicating the existence of a ovarian CSC subpopulation for the first time to our knowledge. positive feedback loop. IL-8 enhanced stemness gene expression Apart from enhanced migratory/invasive, clonogenic and sphere- of ALDH−CD44− cancer cells, while CXCR1 antibody treatment forming capacities, we observed increased chemoresistance and exerted the opposite effect on PDK4-overexpressing ALDH− upregulation of stemness gene expression of this cell population CD44− cells, suggesting a potential regulatory role of IL-8 in vitro. In an in vivo xenograft model, the ALDH+CD44+ subset in PDK4-modulated stemness in ovarian cancer, consistent displayed tumour-initiating and self-renewal characteristics. with data obtained from a previous study supporting the utility CSCs favour glycolysis to OXPHOS in hepatocellular carcinoma, of IL-8 as a therapeutic target to diminish CSC activity in breast breast and colorectal cancers.8 However, limited information is cancer.42 Blockade of the NF-κΒ pathwaybyDMAPTcould available on the mechanism underlying the metabolic switch of suppress sphere-forming capacity and stemness gene expression this subpopulation in ovarian cancer. Notably, increased anaerobic in ALDH+CD44+ or PDK4-overexpressing ALDH−CD44− cells. glycolysis is reported in tumourspheres derived from ovarian These collective findings indicate the involvement of STAT3/AKT/ cancer cells.32 Here, we demonstrated that tumourspheres and the NF-κΒ/IL-8 signalling activation in regulating PDK4-mediated ALDH+CD44+ subset produce higher lactate levels and this cell ovarian CSCs (Fig. 6l). Whether PDK4 has direct interaction with subpopulation is endowed with reduced OCR. Furthermore, we STAT3 leading to its phosphorylation will be examined in future observed upregulation of the mitochondrial gatekeeping enzyme, studies. PDK4, in not only ascites-derived cancer cells relative to primary We further evaluated the therapeutic potential of targeting ovarian tumour cells but also in tumourspheres and the ALDH+ PDK4 with DCA, a pan-PDK inhibitor. The potential effects of CD44+ subset. The contribution of the Warburg effect to DCA has been examined in multiple cancer types.43–45 However, promoting anoikis resistance and tumour metastasis through studies to date have been primarily confined to the effects of overexpression of PDK4 has been highlighted in previous DCA on bulk tumours and not CSC subpopulations. DCA is studies.33 The collective findings support the metabolic switch known to redirect the metabolism of rat glioma CSCs from from OXPHOS to glycolysis in ovarian CSCs, especially in the ALDH+ glycolysis to mitochondrial respiration, with no effects on rat CD44+ subset. neural stem cells.46 Moreover, DCA suppresses angiogenesis and Clinically, a positive correlation was detected between PDK4 induces apoptosis in CSCs of glioblastoma,47 and disrupts CSC immunoreactivity and advanced tumour stages, grades and properties and mitochondrial functions of CSC subpopulation of serous histological subtypes, as well as metastasis, highlighting a medulloblastoma48 and hepatocarcinoma.49 Our studies showed potential role of PDK4 in ovarian cancer progression and that DCA impeded sphere formation, clonogenicity and migration/ aggressiveness. Moreover, higher PDK4 was correlated with invasion, as well as stemness mRNA expression of ovarian CSCs shorter overall and progression-free survival, implicating its utility without affecting cancer cell proliferation at low doses. In addition, as a novel prognostic marker for ovarian cancer. DCA inhibited tumorigenesis and tumour-initiating capacities Functionally, PDK4 induces tumour growth through regulat- in vivo. More intriguingly, DCA attenuated the effects of cisplatin ing the KRAS signalling pathway in lung and colorectal cancers13 on ALDH+CD44+ cancer cells, consistent with previous findings and promotes tumorigenesis in prostate cancer cells through that a dual DCA–platinum drug combination could effectively the activation of the CREB-mediated mTORC1 signalling path- overcome resistance to cisplatin in ovarian cancer.50 way.34 Tumour growth factor-β signalling is known to promote In conclusion, we have demonstrated enhanced CSC character- drug resistance through regulation of PDK4 expression.35 istics and PDK4 overexpression in ascites-derived tumour cells. Furthermore, knockdown of PDK4 has been shown to reduce PDK4 plays essential roles in endowing CSC properties and hypoxia-inducible factor-1α expression and subsequent vascular metabolic switch of the CSC subset isolated based on tumour- endothelial growth factor secretion in neuroblastoma cells.36 sphere formation/CSC markers ALDH and CD44 through a Our functional studies on manipulation of PDK4 expression in mechanism involving the STAT3/AKT/NF-κΒ/IL-8 signalling path- CSC and non-CSC subpopulations isolated from ascites-derived way. This study provides important insights supporting the cells demonstrated that PDK4 promoted migratory/invasive, therapeutic potential of targeting PDK4 via its inhibitor, DCA in clonogenic, and sphere-forming capacities and upregulated the CSC subset of ovarian cancer. Ascites-derived ALDH+CD44+ tumour cell subsets endow stemness... Y.-X. Jiang et al. 286 ACKNOWLEDGEMENTS 11. Walter, W., Thomalla, J., Bruhn, J., Fagan, D. H., Zehowski, C., Yee, D. et al. Altered We thank Ms. Erica Yau and Dr. Emily Pang from Core Facility, for providing and regulation of PDK4 expression promotes antiestrogen resistance in human breast maintaining the equipment needed for flow cytometry. cancer cells. Springerplus 4, 689 (2015). 12. Leclerc, D., Pham, D. N., Levesque, N., Truongcao, M., Foulkes, W. D., Sapienza, C. et al. Oncogenic role of PDK4 in human colon cancer cells. Br. J. Cancer 116, AUTHOR CONTRIBUTIONS 930–936 (2017). Y.-X.J., M.K.-Y.S. and K.K.-L.C. conceived the project. Y.-X.J., M.K.-Y.S. and J.-J.W. 13. Trinidad, A. G., Whalley, N., Rowlinson, R., Delpuech, O., Dudley P., Rooney C. et al. performed the experiments. X.-T.M., T.H.-Y.L., D.W.C. and A.N.-Y.C. contributed new Pyruvate dehydrogenase kinase 4 exhibits a novel role in the activation of reagents/analytic tools; Y.-X.J., M.K.-Y.S., H.Y.-S.N. and K.K.-L.C. analysed the data. mutant KRAS, regulating cell growth in lung and colorectal tumour cells. Onco- – Y.-X.J. and M.K.-Y.S. wrote the manuscript. All authors provided critical revision and gene 36, 6164 6176 (2017). approved the final manuscript. 14. Zhang, S., Balch, C., Chan, M. W., Lai, H. C., Matei, D., Schilder, J. M. et al. Identi- fication and characterization of ovarian cancer-initiating cells from primary human tumors. Cancer Res. 68, 4311–4320 (2008). ADDITIONAL INFORMATION 15. Rajendran, V. & Jain, M. V. In vitro tumorigenic assay: colony forming assay for cancer stem cells. Methods Mol. Biol. 1692,89–95 (2018). Ethics approval and consent to participate Sample collection was under the 16. Shaw, F. L., Harrison, H., Spence, K., Ablett, M. P., Simoes, B. M., Farnie, G. et al. A approval of Human Research Ethics Committee of the University of Hong Kong. detailed mammosphere assay protocol for the quantification of breast stem cell Consented patients were informed of the use of their clinical information before activity. J. Mammary Gland Biol. Neoplasia 17, 111–117 (2012). fi sample collection. The authors con rm that they have obtained written consent from 17. Michelakis, E. D., Webster, L. & Mackey, J. R. Dichloroacetate (DCA) as a potential each patient to publish the manuscript. The study was performed in accordance with metabolic-targeting therapy for cancer. Br. J. Cancer 99, 989–994 (2008). the Declaration of Helsinki. Animal experiments were conducted following protocols 18. Saed, G. M., Fletcher, N. M., Jiang, Z. L., Abu-Soud, H. M. & Diamond, M. P. approved by the Committee of the Use of Live Animals in Teaching and Research Dichloroacetate induces apoptosis of epithelial ovarian cancer cells through a (CULATR) and were carried out under an approved CULATR licence (No. 4598-18). mechanism involving modulation of oxidative stress. Reprod. Sci. 18, 1253–1261 (2011). Consent to publish This manuscript does not contain any individual person’s data. 19. Qin, H. R., Kim, H. J., Kim, J. Y., Hurt, E. M., Klarmann, G. J., Kawasaki, B. T. et al. Activation of signal transducer and activator of transcription 3 through a phos- Data availability The datasets generated and analysed during the current study are phomimetic serine 727 promotes prostate tumorigenesis independent of tyr- – available from the corresponding author on reasonable request. osine 705 phosphorylation. Cancer Res. 68, 7736 7741 (2008). 20. Hazan-Halevy, I., Harris, D., Liu, Z., Liu, J., Li, P., Chen, X. et al. STAT3 is con- stitutively phosphorylated on serine 727 residues, binds DNA, and activates Competing interests The authors declare no competing interests. transcription in CLL cells. Blood 115, 2852–2863 (2010). 21.Korkaya,H.,Liu,S.&Wicha,M.S.Regulation of cancer stem cells by cytokine Funding information The work was jointly funded by the University of Hong Kong networks: attacking cancer’sinflammatory roots. Clin. Cancer Res. 17, (201711159248) and by the Hong Kong Research Grants Council General Research 6125–6129 (2011). Fund (HKU 17101414), and the Research Fund from the Department of Obstetrics and 22. Bradford, J. W. & Baldwin, A. S. IKK/nuclear factor-kappaB and oncogenesis: roles Gynaecology. in tumor-initiating cells and in the tumor microenvironment. Adv. Cancer Res. 121, 125–145 (2014). Supplementary information is available for this paper at https://doi.org/10.1038/ 23. Wang, S., Liu, Z., Wang, L. & Zhang, X. NF-kappaB signaling pathway, inflamma- s41416-020-0865-z. tion and colorectal cancer. Cell. Mol. Immunol. 6, 327–334 (2009). 24. Merritt, W. M., Lin, Y. G., Spannuth, W. A., Fletcher, M. S., Kamat, A. A., Han, L. Y. et al. Note This work is published under the standard license to publish agreement. After Effect of interleukin-8 gene silencing with liposome-encapsulated small interfering 12 months the work will become freely available and the license terms will switch to RNA on ovarian cancer cell growth. J. Natl Cancer Inst. 100,359–372 (2008). a Creative Commons Attribution 4.0 International (CC BY 4.0). 25. Ginestier, C., Liu, S., Diebel, M. E., Korkaya, H., Luo, M., Brown, M. et al. CXCR1 blockade selectively targets human breast cancer stem cells in vitro and in Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims xenografts. J. Clin. Invest. 120, 485–497 (2010). in published maps and institutional affiliations. 26. He, M., Wang, D., Zou, D., Wang, C., Lopes-Bastos, B., Jiang, W. G. et al. Re- purposing of curcumin as an anti-metastatic agent for the treatment of epithelial ovarian cancer: in vitro model using cancer stem cell enriched ovarian cancer REFERENCES spheroids. Oncotarget 7, 86374–86387 (2016). 1. Pattabiraman, D. R. & Weinberg, R. A. Tackling the cancer stem cells—what 27. Weiswald, L. B., Bellet, D. & Dangles-Marie, V. Spherical cancer models in tumor – challenges do they pose? Nat. Rev. Drug Discov. 13, 497–512 (2014). biology. Neoplasia 17,1 15 (2015). 2. Kwon, M. J. & Shin, Y. K. Regulation of ovarian cancer stem cells or tumor- 28. Yilmazer, A. Evaluation of cancer stemness in breast cancer and glioblastoma initiating cells. Int. J. Mol. Sci. 14, 6624–6648 (2013). spheroids in vitro. 3 Biotech 8, 390 (2018). 3. Lengyel, E. Ovarian cancer development and metastasis. Am. J. Pathol. 177, 29. Leung, H. W., Ko, C. H., Yue, G. G., Herr, I. & Lau, C. B. The natural agent 4- 1053–1064 (2010). vinylphenol targets metastasis and stemness features in breast cancer stem-like – 4. Condello, S., Morgan, C. A., Nagdas, S., Cao, L., Turek, J., Hurley, T. D. et al. Beta- cells. Cancer Chemother. Pharm. 82, 185 197 (2018). catenin-regulated ALDH1A1 is a target in ovarian cancer spheroids. Oncogene 34, 30. Young, M. J., Wu, Y. H., Chiu, W. T., Weng, T. Y., Huang, Y. F. & Chou, C. Y. All-trans 2297–2308 (2015). retinoic acid downregulates ALDH1-mediated stemness and inhibits tumour – 5. Alvero, A. B., Chen, R., Fu, H. H., Montagna, M., Schwartz, P. E., Rutherford, T. et al. formation in ovarian cancer cells. Carcinogenesis 36, 498 507 (2015). Molecular phenotyping of human ovarian cancer stem cells unravels the 31. Lin, J. & Ding, D. The prognostic role of the cancer stem cell marker CD44 in mechanisms for repair and chemoresistance. Cell Cycle 8, 158–166 (2009). ovarian cancer: a meta-analysis. Cancer Cell Int. 17, 8 (2017). 6. Koppenol, W. H., Bounds, P. L. & Dang, C. V. Otto Warburg’s contributions to 32. Liao, J., Qian, F., Tchabo, N., Mhawech-Fauceglia, P., Beck, A., Qian, Z. et al. Ovarian current concepts of cancer metabolism. Nat. Rev. Cancer 11, 325–337 (2011). cancer spheroid cells with stem cell-like properties contribute to tumor gen- 7. Chen, C. T., Hsu, S. H. & Wei, Y. H. Upregulation of mitochondrial function and eration, metastasis and chemotherapy resistance through hypoxia-resistant antioxidant defense in the differentiation of stem cells. Biochim. Biophys. Acta metabolism. PLoS ONE 9, e84941 (2014). 1800, 257–263 (2010). 33. Kamarajugadda, S., Stemboroski, L., Cai, Q., Simpson, N. E., Nayak, S., Tan, M. et al. 8. Wong, T. L., Che, N. & Ma, S. Reprogramming of central carbon metabolism in Glucose oxidation modulates anoikis and tumor metastasis. Mol. Cell. Biol. 32, – cancer stem cells. Biochim. Biophys. Acta 1863, 1728–1738 (2017). 1893 1907 (2012). 9. Sutendra, G. & Michelakis, E. D. Pyruvate dehydrogenase kinase as a novel 34. Liu, Z., Chen, X., Wang, Y., Peng, H., Wang, Y., Jing, Y. et al. PDK4 protein promotes therapeutic target in oncology. Front. Oncol. 3, 38 (2013). tumorigenesis through activation of cAMP-response element-binding protein 10. Bonnet, S., Archer, S. L., Allalunis-Turner, J., Haromy, A., Beaulieu, C., Thompson, R. (CREB)-Ras homolog enriched in brain (RHEB)-mTORC1 signaling cascade. J. Biol. – et al. A mitochondria-K+ channel axis is suppressed in cancer and its normal- Chem. 289, 29739 29749 (2014). ization promotes apoptosis and inhibits cancer growth. Cancer Cell 11,37–51 35. Zhang, Y., Zhang, Y., Geng, L., Yi, H., Huo, W., Talmon, G. et al. Transforming (2007). growth factor beta mediates drug resistance by regulating the expression of Ascites-derived ALDH+CD44+ tumour cell subsets endow stemness... Y.-X. Jiang et al. 287 pyruvate dehydrogenase kinase 4 in colorectal cancer. J. Biol. Chem. 291, 43. Velpula, K. K., Bhasin, A., Asuthkar, S. & Tsung, A. J. Combined targeting of PDK1 17405–17416 (2016). and EGFR triggers regression of glioblastoma by reversing the Warburg effect. 36. Rellinger, E. J., Romain, C., Choi, S., Qiao, J. & Chung, D. H. Silencing gastrin- Cancer Res. 73, 7277–7289 (2013). releasing peptide receptor suppresses key regulators of aerobic glycolysis in 44. Lin, G., Hill, D. K., Andrejeva, G., Boult, J. K., Troy, H., Fong, A. C. et al. Dichlor- neuroblastoma cells. Pediatr. Blood Cancer 62, 581–586 (2015). oacetate induces autophagy in colorectal cancer cells and tumours. Br. J. Cancer 37. Lee, S. J., Jeong, J. Y., Oh, C. J., Park, S., Kim, J. Y., Kim, H. J. et al. Pyruvate 111, 375–385 (2014). dehydrogenase kinase 4 promotes vascular calcification via SMAD1/5/8 phos- 45. Shen, Y. C., Ou, D. L., Hsu, C., Lin, K. L., Chang, C. Y., Lin, C. Y. et al. Activating phorylation. Sci. Rep. 5, 16577 (2015). oxidative phosphorylation by a pyruvate dehydrogenase kinase inhibitor over- 38. House, C. D., Jordan, E., Hernandez, L., Ozaki, M., James, J. M., Kim, M. et al. comes sorafenib resistance of hepatocellular carcinoma. Br. J. Cancer 108,72–81 NFkappaB promotes ovarian tumorigenesis via classical pathways that support (2013). proliferative cancer cells and alternative pathways that support ALDH(+) cancer 46. Morfouace, M., Lalier, L., Bahut, M., Bonnamain, V., Naveilhan, P., Guette, C. et al. stem-like cells. Cancer Res. 77, 6927–6940 (2017). Comparison of spheroids formed by rat glioma stem cells and neural stem cells 39. Burgos-Ojeda, D., Wu, R., McLean, K., Chen, Y. C., Talpaz, M., Yoon, E. et al. CD24+ reveals differences in glucose metabolism and promising therapeutic applica- ovarian cancer cells are enriched for cancer-initiating cells and dependent on tions. J. Biol. Chem. 287, 33664–33674 (2012). JAK2 signaling for growth and metastasis. Mol. Cancer Ther. 14, 1717–1727 47. Michelakis, E. D. S. G., Dromparis, P., Webster, L., Haromy, A. & Niven, E. Metabolic (2015). modulation of glioblastoma with dichloroacetate. Sci. Transl. Med. 2, 31ra34 (2010). 40. Egusquiaguirre, S. P., Yeh, J. E., Walker, S. R., Liu, S. & Frank, D. A. The STAT3 target 48. Sun, L., Moritake, T., Ito, K., Matsumoto, Y., Yasui, H., Nakagawa, H. et al. Metabolic gene TNFRSF1A modulates the NF-kappaB pathway in breast cancer cells. Neo- analysis of radioresistant medulloblastoma stem-like clones and potential ther- plasia 20, 489–498 (2018). apeutic targets. PLoS ONE 12, e0176162 (2017). 41. McFarland, B. C., Hong, S. W., Rajbhandari, R., Twitty, G. B. Jr., Gray, G. K., Yu, H. 49. Song, K., Kwon, H., Han, C., Zhang, J., Dash, S., Lim, K. et al. Active glycolytic et al. NF-kappaB-induced IL-6 ensures STAT3 activation and tumor aggressive- metabolism in CD133(+) hepatocellular cancer stem cells: regulation by MIR-122. ness in glioblastoma. PLoS ONE 8, e78728 (2013). Oncotarget 6, 40822–40835 (2015). 42. Singh, J. K., Simoes, B. M., Howell, S. J., Farnie, G. & Clarke, R. B. Recent advances 50. Zhang, Y., Guo, G., Ma, B., Du, R., Xiao, H., Yang, X. et al. A hybrid platinum drug reveal IL-8 signaling as a potential key to targeting breast cancer stem cells. dichloroacetate-platinum(II) overcomes cisplatin drug resistance through dual Breast Cancer Res. 15, 210 (2013). organelle targeting. Anticancer Drugs 26, 698–705 (2015).