Full Text

Turn on search term navigation

ARTICLE

Received 21 Jan 2016 | Accepted 13 Apr 2016 | Published 19 May 2016

DOI: 10.1038/ncomms11612 OPEN

Inhibition of the hexosamine biosynthetic pathway promotes castration-resistant prostate cancer

Akash K. Kaushik1,2,*, Ali Shojaie3,*, Katrin Panzitt1,w,*, Rajni Sonavane1, Harene Venghatakrishnan4,5,Mohan Manikkam1, Alexander Zaslavsky4,5, Vasanta Putluri1, Vihas T. Vasu6, Yiqing Zhang1, Ayesha S. Khan7, Stacy Lloyd1, Adam T. Szafran1, Subhamoy Dasgupta1, David A. Bader1, Fabio Stossi1, Hangwen Li4,5,Susmita Samanta1, Xuhong Cao4,5,8, Efrosini Tsouko7, Shixia Huang1,9, Daniel E. Frigo7,10, Lawrence Chan1,2, Dean P. Edwards1,9, Benny A. Kaipparettu11, Nicholas Mitsiades1, Nancy L. Weigel1, Michael Mancini1,Sean E. McGuire1, Rohit Mehra4,8, Michael M. Ittmann12, Arul M. Chinnaiyan4,5,8,13, Nagireddy Putluri1, Ganesh S. Palapattu4,5, George Michailidis14,w & Arun Sreekumar1,2,9

The precise molecular alterations driving castration-resistant prostate cancer (CRPC) are not clearly understood. Using a novel network-based integrative approach, here, we show distinct alterations in the hexosamine biosynthetic pathway (HBP) to be critical for CRPC. Expression of HBP enzyme glucosamine-phosphate N-acetyltransferase 1 (GNPNAT1) is found to be signicantly decreased in CRPC compared with localized prostate cancer (PCa). Genetic loss-of-function of GNPNAT1 in CRPC-like cells increases proliferation and aggressiveness, in vitro and in vivo. This is mediated by either activation of the PI3K-AKT pathway in cells expressing full-length androgen receptor (AR) or by specic protein 1 (SP1)-regulated expression of carbohydrate response element-binding protein (ChREBP) in cells containing AR-V7 variant. Strikingly, addition of the HBP metabolite UDP-N-acetylglucosamine (UDP-GlcNAc) to CRPC-like cells signicantly decreases cell proliferation, both in-vitro and in animal studies, while also demonstrates additive efcacy when combined with enzalutamide in-vitro. These observations demonstrate the therapeutic value of targeting HBP in CRPC.

1 Department of Molecular and Cellular Biology and Alkek Center for Molecular Discovery, Baylor College of Medicine, Houston, Texas 77030, USA. 2 Verna and Marrs McLean Department of Biochemistry and Molecular Biology, Baylor College of Medicine, Houston, Texas 77030, USA. 3 Department of Biostatistics, University of Washington, Seattle, Washington 98195, USA. 4 Comprehensive Cancer Center, University of Michigan, Ann Arbor, Michigan 48109, USA. 5 Department of Urology, University of Michigan, Ann Arbor, Michigan 48109, USA. 6 Department of Zoology, Maharaja Sayajirao University of Baroda, Vadodara 390002, India. 7 Center for Nuclear Receptors and Cell Signaling, Department of Biology and Biochemistry, University of Houston, Houston, Texas 77204, USA. 8 Department of Pathology University of Michigan, Ann Arbor, Michigan 48109, USA. 9 Dan L Duncan Cancer Center, Baylor College of Medicine, Houston, Texas 77030, USA. 10 Genomic Medicine Program, Houston Methodist Research Institute, Houston, Texas 77030, USA. 11 Department of Molecular and Human Genetics, Baylor College of Medicine, Houston, Texas 77030, USA. 12 Department of Pathology and Immunology, Baylor College of Medicine, Houston, Texas 77030, USA. 13 Howard Hughes Medical Institute, University of Michigan, Ann Arbor, Michigan 48109, USA. 14 Department of Statistics, University of Michigan, Ann Arbor, Michigan 48109, USA. * These authors contributed equally to this work. w Present address: Department of

Hepatology and Gastroenterology, Medical University of Graz, Graz, 8036 Austria. (K.P); Department of Statistics and Computer and Information Sciences, University of Florida, Gainesville, Florida 32611-5500, USA (G.M.). Correspondence and requests for materials should be addressed to G.S.P.(email: mailto:[email protected]

Web End [email protected] ) or to G.M. (email: mailto:[email protected]

Web End [email protected] ) or to A.S.K. (email: mailto:[email protected]

Web End [email protected] ).

NATURE COMMUNICATIONS | 7:11612 | DOI: 10.1038/ncomms11612 | http://www.nature.com/naturecommunications

Web End =www.nature.com/naturecommunications 1

ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/ncomms11612

Recent scientic and technological advances have enabled biomedical researchers to collect high-throughput genomic, proteomic and metabolomic data. Functional interactions

of these biomolecules have been obtained and catalogued in carefully curated databases, often in the form of networks1,2. A growing body of evidence suggests that combining information from such high-throughput data in the context of their molecular interactions may shed light on the biology of complex diseases, including cancer36.

The development and progression of prostate cancer (PCa) from the androgen-dependent (AD) to castration-resistant state is thought to be driven by multiple events69. The initial treatment of choice for advanced disease is androgen ablation, which while initially effective often leads to therapy resistance (that is, the castration resistant state). Better and more effective therapies are needed for CRPC, a lethal condition. Our previous studies10 identied specic metabolic alterations associated with localized PCa (treatment naive) and CRPC tumours. These ndings are particularly important as they provide new insights into PCa progression and suggest metabolic targets for therapeutic intervention.

Building on this work, we employ a novel network-based integrative approach that combines patient-derived PCa gene expression with metabolomic data and identify a vital role for hexosamine biosynthetic pathway (HBP) in the castration-resistant state. Furthermore, downregulation of HBP enhances tumorigenicity of CRPC-like cells via activation of cell cycle genes regulated by either the PI3K-AKT or the SP1-ChREBP axis in cells containing either AR-full length (AR-FL) or AR-V7, respectively. Remarkably, treatment with UDP-N-acetylglucosa-mine (UDP-GlcNAc) signicantly decreases proliferation and tumour growth of CRPC-like cells while also increases efcacy of the anti-androgen enzalutamide in in-vitro experiments. These ndings are especially signicant given that the CRPC-like cells tested, inclusive of those containing AR-V7, are inherently resistant to enzalutamide. In aggregate, our data provide novel perspectives on the inuence of HBP on the castration-resistant state and supply rationale for targeting the HBP in CRPC.

Results and DiscussionNovel integrative analysis uncovers a role of HBP in PCa. To dene key biochemical pathways altered in PCa, we used meta-bolomic and transcriptomic proles from our previous study containing 12 treatment-naive localized PCa specimens and 16 benign adjacent prostate tissues (Ben)10 (Supplementary Fig. 1A, clinical information in Supplementary Table 1) and integrated these using a novel pathway-centric analytical framework (Fig. 1a). This approach combines two rankings for each pathway calculated from gene expression and metabolic data, while adjusting for variations in each of the two data sets. In particular, we used methods of Gene Set Analysis (GSA11,12) and a modied version of Network-Based Gene Set Analysis (NetGSA)13 to obtain rankings of each pathway based on genetic and metabolic data, respectively (overview in Fig. 1a). The modied NetGSA framework, unlike GSA-type methods, incorporates reactome-derived interactions and associated stoichiometry between metabolites allowing for adequate statistical power13.

The rankings of biochemical pathways enriched independently by gene expression and metabolomic proles were calculated (Supplementary Fig. 1B) and combined into a single signicance score (Fig. 1b) using a resampling procedure (see the Methods). Importantly, the individual rankings for the biochemical pathways derived from gene-expression and metabolomics data showed B65% concordance (as determined by calculating the

number of pathways within one standard deviation of all pathway ranking differences shown in Supplementary Fig. 1C). The top ve biochemical pathways enriched using the combined signicance score are: riboavin metabolism, biotin metabolism, amino sugar metabolism (also known as the Hexosamine Biosynthetic Pathway; HBP), valine-leucine and isoleucine biosynthesis and cysteine metabolism (Fig. 1b, P-values in Supplementary Table 2).

Next, to further select a pathway from those obtained using our integrative approach for follow-up functional studies, we employed a network-based permutation test using transcriptomics data to identify pathways that have the highest likelihood of playing a key role in coordinated biochemical pathway activity. Accordingly, we carried out an enrichment analysis using transcriptome data focusing on pathways that interacted with our primary list of enriched pathways (Supplementary Fig. 1D). The corresponding enrichment score was used as a measure of the likelihood of coordinated activity among the interacting pathways. In this way, the primary pathway exhibiting the highest likelihood was nominated for validation studies. Results of this network-based permutation test, which assesses the likelihood of observing particular connectivity patterns in a given network, are shown in Fig. 1b. HBP exhibited the smallest P-value (P 0.0060)

among the top ve pathways interrogated (Fig. 1b,c and Supplementary Table 3). This suggests that signicant alterations in HBP are associated with PCa. Further, these results also imply that pathways interacting with HBP, namely glycolysis/gluconeo-genesis, pentose and glucuronate interconversions, fructose and mannose metabolism, ascorbate and aldarate metabolism and glutamine and glutamate metabolism (Fig. 1c), are also important in PCa.

HBP activity is downregulated in CRPC compared with AD PCa. HBP, a nutrient sensing pathway14, generates amino sugars by combining elemental carbon and nitrogen from glucose and glutamine (Fig. 1d). Glutamine-Fructose-6-phosphate transaminase 1 (GFPT1) has been described as the rate-limiting enzyme for this pathway15,16. The acetylation step represents a key entry point for carbon and nitrogen into the HBP and is catalysed by glucosamine-phosphate N-acetyltransferase 1 (GNPNAT1, Fig. 1d). UDP-N acetyl glucosamine pyrophosphorylase 1 (UAP1) catalyses the nal enzymatic reaction of HBP, resulting in the synthesis of UDP-GlcNAc, that is required for N-linked and O-linked glycosylation reactions17,18.

In silico analysis showed that GNPNAT1 and UAP1 were signicantly elevated in PCa compared with Ben in multiple publically available gene expression data sets including the one used here for integrative analysis (Supplementary Fig. 2AC). GNPNAT1 and UAP1 were also signicantly elevated at the transcript level (Supplementary Fig. 2D) in PCa. Further, increased activity of HBP in PCa was conrmed in 15 matched tumourbenign pairs by measuring the product to substrate ratio for the reaction carried out by GNPNAT1 (N-acetylglucosamine-6-P to glucosamine-6-P; Fig. 1e) and levels of UDP-GlcNAc (Supplementary Fig. 2E), the end product of HBP. Tissue microarray analysis further conrmed signicantly higher expression of both GNPNAT1 and UAP1 in PCa compared with Ben, whereas, interestingly, their expression was signicantly lower in sites of lymph node metastasis and CRPC tissues compared with localized PCa (Fig. 1f,g and Supplementary Fig. 3A). Consistent with the ndings in CRPC, transcript levels of HBP genes were also signicantly downregulated in CRPC tissues across multiple independent publically available microarray data sets (Supplementary Fig. 3BE). Together, these results suggest signicantly

2 NATURE COMMUNICATIONS | 7:11612 | DOI: 10.1038/ncomms11612 | http://www.nature.com/naturecommunications

Web End =www.nature.com/naturecommunications

NATURE COMMUNICATIONS | DOI: 10.1038/ncomms11612 ARTICLE

diminished activity of HBP in CRPC tumours compared with AD organ-conned PCa.

HBP modulates aggressivity in CRPC. Next, we studied the role of HBP in CRPC progression by examining the role of the proximal enzyme GNPNAT1 previously observed to be consistently downregulated in CRPC. Towards this end, we

utilized a short hairpin RNA (shRNA) strategy to knockdown (KD) GNPNAT1 expression in three CRPC-like (that is, androgen responsive, but not AD) cell lines LNCaP-ABL19 and C4-2 (ref. 20; containing AR-FL) and 22Rv1 (containing AR-FL and AR-V7) (ref. 21).

In 22Rv1 cells, GNPNAT1 KD resulted in B50% and 70% reduction in mRNA (KD1 and KD5, respectively, Fig. 2a), decrease in protein expression (Fig. 2b and Supplementary

a b

Benign

(n=16)

PCa

(n=12)

Integrative score

Aminosugar pathway (HBP)

Obtain gene and

metabolite expression Data preprocessing

(imputation, normalization) Map genes and metabolites to

KEGG and select metabolic

genes and metabolites

Riboflavin metabolismBiotin metabolismAminosugar metabolism (HBP)

Valine, leucine and isoleucine biosynthesis Cysteine metabolism

log(0.01)

Select subset of KEGG

pathways with available data

Metabolic

gene

expression

data

Metabolite

expression

data

log10 (P(enrichment))

log(0.03)

log(0.1)

Define metabolic interaction

network from chemical

reactions in KEGG

Gene set enrichment

analysis (GSA) Network based gene set

analysis (NetGSA)

Rank pathways based

on genetic profiles

log(0.3)

Rank pathways based

on metabolic profiles

Combine pathway rankings,

select top pathways Construct network of pathway

interactions in KEGG Analysis of neighbours top

pathways (flux analysis) Sensitivity analysis

1 2 5 10 20 50

log(1)

Integrated score

c d

Starch and sucrosemetabolism Galactose metabolism

Fructose and mannose metabolism

Pyruvate metabolism

Pentose and glucuronate interconversions

Peptidoglycan biosynthesis

Aminosugar metabolism (HBP)

Glycine, serine and threonine metabolism

Riboflavin metabolism

Cysteine metabolism

Taurine and hypotaurine metabolism

Purine metabolism

Ascorbate and aldarate metabolism

Glycolysis/ gluconeogenesis

Pentose phosphate pathway

Valine, leucine and isoleucine biosynthesis

Methionine metabolism

Pyrimidine metabolism

Alkaloid biosynthesis II

Glutathione metabolism

Biotin metabolism

Lysine degradation

D-Glutamine and D-glutamate metabolism

Valine, leucine and isoleucine degradation

Aminoacyl-tRNA biosynthesis

GPI Glycolysis

Pyruvate

GFPT1

GNPNAT1

PGM3

GlcNac-1-P

Glu

Glu-6-P Fru-6-P

Gln

GlcN

GlcN-6-P

Acetyl-CoA

CoA

UAP1 UDP-GlcNac

GlcNAc-6-P

Uridine

e f g

P=1104

P=0.003

P=1107

P=0.04

1: Benign

4: LN-Met

Log2 (GlcNAc-6-P/GlcN-6-P)

2: PCa

5: Met

100um

Score

3: LN-Met

6: Met

Ben (n=15)

PCa (n=15)

Ben (n=48)

PCa (n=67)

LN-Met

(n=28)

Met (n=50)

NATURE COMMUNICATIONS | 7:11612 | DOI: 10.1038/ncomms11612 | http://www.nature.com/naturecommunications

Web End =www.nature.com/naturecommunications 3

ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/ncomms11612

metastases as well as jaw and pelvic bone disease (Fig. 2l, yellow arrows).

To further verify the tumour-promoting role of HBP pathway, we knocked down GFPT1 either transiently or stably in CRPC-like cells. Consistent with our GNPNAT1 KD ndings, KD of GFPT1 using short interfering RNA (siRNA) in 22Rv1 and LNCaP-ABL cells resulted in increased cell proliferation (Supplementary Figs 8AE and 22). This was further replicated in 22Rv1 cells using shRNA-based stable KD of GFPT1 (Supplementary Fig. 8F,G). Interestingly, stable KD of GFPT1 in CRPC-like cells also resulted in a signicant (Po0.05,

Students t-test) reduction in mRNA levels of GNPNAT1 (Supplementary Fig. 8F). Similarly, KD of GNPNAT1 in CRPC-like cells also signicantly (Po0.05, Students t-test) decreased

GFPT1 expression, both at the mRNA and protein level (Supplementary Figs 8H,I and 23). These ndings conrm a role for HBP pathway in promoting CRPC growth and proliferation.

HBP affects CRPC via PI3K-AKT and SP1-ChREBP. Analysis of microarray data in LNCaP-ABL KD cells compared with control revealed a set of upregulated genes indicative of elevated expression of components of AKT signalling namely PIK3CD and PIK3AP1 (Fig. 3a; GEO ID: GSE67537). In LNCaP-ABL KD cells, increased protein expression of activated PI3K (PI3K-p85), total AKT, p-AKT (T308), p-AKT (S473) and elevated levels of its downstream effectors mTOR and p-MAPK (ERK1/2) were observed (Fig. 3b and Supplementary Fig. 17). Consistent with this, treatment of LNCaP-ABL KD and NT cells with the reversible PI3K inhibitor LY294002 (ref. 22; 50 mM) and the

AKT/PI3K inhibitor Perifosine23 (20 mM) resulted in a signicant (Po0.05, Students t-test) reduction in cell proliferation in

KD1 and modest but not signicant reduction in KD5 cells compared with NT controls. In addition, the upregulated gene expression signature for LNCaP-ABL KD cells revealed increased expression of AR, the AR co-activator NCOA3 and AR target genes (KLK3 and TMPRSS2; Fig. 3a). Elevated expression of AR at the transcript (Supplementary Fig. 9A) and protein (Fig. 3d and Supplementary Fig. 17) levels as well as KLK3 (Fig. 3e) and TMPRSS2 (Supplementary Fig. 9B) was veried in LNCaP-ABL KD cells and their corresponding xenograft tumours (Fig. 3f). A similar cascade of molecular alterations involving the PI3K-AKT axis was observed in C4-2 KD cells, which contain full-length AR (Supplementary Figs 10A,B,24 and 25).

In contrast, the upregulated genes in the microarray data set (GEO ID: GSE67537) obtained from 22Rv1 cells containing GNPNAT1 KD revealed increased expression of multiple cell cycle genes (Fig. 3g, red highlight) that together enriched multiple

Figure 1 | Integrative analysis of gene expression and metabolic data sets identies alterations in the hexosamine biosynthetic pathway in prostate cancer. (a) Overview of integrative methodology. (b) Top pathways identied after integrative analysis using combined gene/metabolite-derived enrichment scores using our previously published10 data. Black dots indicate top six pathways identied as outliers and coloured arrows indicate the top ve enriched pathways chosen for secondary analysis. (c) Network representation of pathways shown in b (solid coloured circles: enriched pathways after integrative analysis using combined gene/metabolite-derived enrichment scores; circumference is correlated to pathway connectivity). Association between interacting pathways and each of the enriched pathways (solid coloured circles) obtained after the integrative analysis is shown by coloured arrows, which also show the direction of interaction. Arrow thickness correlates with number of interacting components between two pathways. Enriched associated pathways (also termed interacting pathways) interacting with those listed in b, are shown in red rimmed circles. Thus, for example, amino sugar metabolism or HBP has eight interacting pathways, 5 of which are enriched (red rimmed circle). (d) Overview of the HBP. GNPNAT1 (red) is the most proximal consistently upregulated HBP enzyme in PCa. (e) GNPNAT1 product/substrate ratio was higher in PCa compared with matched benign-adjacent prostate tissues (n 15 pairs). P-value was calculated using Unpaired Students t-test. (f) Boxplots showing immunostaining of GNPNAT1 in primary PCa

(n 67) versus benign adjacent (Ben, n 48), PCa versus lymph node mets (LN-Met, treatment nave, n 28) and PCa versus CRPC mets (Met, n 50).

P-value was calculated using Wilcoxon-Rank Sum test. (g) Representative photomicrograph of GNPNAT1 staining in 1: Ben (black arrows) with tumour nodules (red arrow); 2: PCa; 3, 4: LN-Met and 5, 6: Mets. Representative scale bar for sections 1, 3 and 4 is 100 mm (low power) and for sections 2, 5 and 6 is 25 mm (high power). In all cases, P-value of o0.05 was considered signicant. For boxplots, the horizontal line represents median value, whereas

Whiskers represent either o25 or 475 quartile ranges.

Fig. 16), signicant (Po0.05, analysis of variance (ANOVA)) reduction in the ratio of its product to substrate (Fig. 2c) as well as signicant reduction in the levels of HBP end product UDP-GlcNAc (Supplementary Fig. 4A). Similar results were obtained in LNCaP-ABL (Fig. 2df and Supplementary Figs 4B and 16) and C4-2 cells (Supplementary Figs 4CE and 19). All these cells exhibited a signicant (Po0.05, ANOVA) increase in proliferation (Fig. 2g,h and Supplementary Fig. 4FH), associated with no evidence of apoptosis as assessed by Poly-ADP ribose polymerase cleavage (Supplementary Figs 4I and 19) compared with their respective non-target (NT) controls; this was reversed by overexpression of GNPNAT1 (Supplementary Fig. 4J,K). In contrast, in AD LNCaP cells, KD of GNPNAT1 resulted in a signicant (Po0.05, Students t-test) reduction in cell proliferation, which was partially rescued by add back of HBP metabolite UDP-GlcNAc (Supplementary Figs 5AC and 20), all of which verify our earlier observation describing elevated levels of HBP in AD PCa.

In vivo xenograft experiments using 22Rv1 KD cells without (Fig. 2i) or with luciferase (Supplementary Figs 6A,B), demonstrated a signicantly (Po0.05, GLM or generalized linear model) higher rate of tumour growth compared with NT controls. Morphometric analysis revealed KD tumours to be larger, whereas both quantitative PCR (QPCR) and immuno-blot analysis conrmed KD of GNPNAT1 (Supplementary Figs 6CE and 21). Further, histological analysis revealed KD tumours to be slightly more pleomorphic, with loss of glandular architecture and signicantly higher number of mitotic gures compared with their NT counterparts (Supplementary Fig. 6F). Moreover, KD cells in LNCaP-ABL (Fig. 2j) also showed a signicant increase in their tumorigenic potential. Interestingly, B100 days post resection of primary LNCaP-ABL tumours (refer Fig. 2j for details), ve of six mice (two NT and three KD5) developed metastatic lesions in lymph nodes. Metastatic tumours in KD mice were at least threefold bigger (Po0.038,

Students t-test) in volume (360880 mm3) compared with the NT (15100 mm3 group; Supplementary Fig. 7A,B). Consistent with this, intra-cardiac injection of luciferase-labelled 22Rv1 cells in castrated Nod-SCID Gamma mice demonstrated signicantly higher number of metastatic lesions in the KD compared with NT mice (Po0.01, Students t-test, Fig. 2k and

Supplementary Fig. 7C, n 8 NT, n 10 KD1). Interestingly,

additional metastasis to the brain as well as bones in the jaw and pelvic regions was observed in eight out of ten KD and only one out of eight NT mice (representative pictures shown in Fig. 2k white arrows and quantied in Supplementary Fig. 7C). Magnetic resonance imaging (MRI, on n 4 from each group)

using Prohance contrast agent further conrmed brain

4 NATURE COMMUNICATIONS | 7:11612 | DOI: 10.1038/ncomms11612 | http://www.nature.com/naturecommunications

Web End =www.nature.com/naturecommunications

NATURE COMMUNICATIONS | DOI: 10.1038/ncomms11612 ARTICLE

cell cycle concepts (Supplementary Fig. 11A). Elevated expression of a subset of these cell cycle genes was veried by QPCR (Supplementary Fig. 11B) and supported our observation of increased proliferation seen in vitro and in vivo upon KD of GNPNAT1. Furthermore, unlike in LNCaP-ABL KD cells, AR protein expression (both AR-FL and AR-V7) was unchanged upon GNPNAT1 KD in 22Rv1 cells (Supplementary Figs 11C and 26). Intriguingly, however, in 22Rv1 cells, transient KD of

GFPT1 modestly increased protein expression of AR-FL and AR-V7 (Supplementary Figs 8B and 22) and its stable KD signicantly (Po0.05, Students t-test) increased the mRNA levels for AR-FL and AR-V7 (Supplementary Fig. 8F).

To obtain insight into the regulatory components driving the upregulated gene signature in GNPNAT1 KD 22Rv1 cells, promoter analysis using Pscan24 revealed signicant enrichment of SP1-binding sites (Fig. 3h and Supplementary Fig. 11D).

c d f

P=3.2e06

22R- KD5

P=0.006

P=0.001

P=7.6e06

P=3.4e06

P=0.0006

P=2.6e06

Fold change (mRNA)

1.0

22R- NT

1.0

Log2 (GlcNAc-6P/GlcN-6P)

0.9

3.5

22R- NT

0.8

0.6

Fold change (mRNA)

1.6

0.7

3.0

0.6

1.4

0.5

2.5

0.4

1.2

0.4

0.3

2.0

0.2

1.0

22R- KD1

L- NT

L- KD1 L- KD5

1.5

0.8

0.6

1.0

GNPNAT1

Actin

22R- NT

21 kDa

45 kDa

GNPNAT1

Actin

L- NT

21kDa

45kDa

22R- KD1

22R- KD5

L- NT

L- KD1

L- KD5

L- KD1

L- KD5

22R- KD1

22R- KD5

g h i j

P=5.2e12

P=1.9e21

P=2.7e09

P=1.4e21

2.3

22R- NT

1,500

22R-KD1(n=7)

22R-NT(n=8)

P=0.0002

L-KD1(n=7) L-KD5(n=7) L-NT(n=5)

1200

2.2

1.8

BrdU-assay

absorbance (450 nm)

Tumour volume (mm3 )

2.1

1,000

BrdU-assay

Tumour volume (mm3 )

1.6

2.0

1.4

800 P=0.005

1.9

1.2

1.8

500

400

1.0

1.7

1.6

0.8

0.6

22R- KD1

22R- KD5

20 25 30

15 20 25 30

L- NT

L- KD1

L- KD5

Days

k l

22R-NT-LUC

22R-KD1-LUC

22R-NT-LUC

22R-KD1-LUC

#177 #170

#171

#184

Brain Jaw Jaw

Femurs

Brain

L R

#172

#178

Brain Jaw

Brain

Femurs

#182

#180

1.0 1.5

2.0 2.5 107

Luminescence

Femurs

Jaw

Femurs

NATURE COMMUNICATIONS | 7:11612 | DOI: 10.1038/ncomms11612 | http://www.nature.com/naturecommunications

Web End =www.nature.com/naturecommunications 5

ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/ncomms11612

UDP-GlcNAc has therapeutic efcacy in CRPC. As CRPC cells innately show reduced expression and activity of HBP components, which in turn activates the cell cycle, we hypothesized that addition of a HBP downstream metabolite could slow down proliferation and reduce cell viability. Accordingly, addition of 60 mM UDP-GlcNAc for 96 h to both 22Rv1 and LNCaP-ABL cells resulted in a signicant decrease in both viability and cell numbers by MTT and Celigo-based quantication assay (Bonferroni corrected P 0.01, Fig. 4c,d and Supplementary

Fig. 14A,B), with a concomitant downregulation of predicted cell cycle genes (Supplementary Fig. 14C, in 22Rv1 cells).Mass spectrometry-based quantication conrmed internalization of UDP-GlcNAc into both LNCaP-ABL and 22Rv1 cells (Supplementary Fig. 14D). A similar signicant reduction in cell numbers was obtained when the CRPC-like cells were treated either with 10 mM glucosamine (GlcN, 96 h, Supplementary Fig. 14A,B) or with 20 mM UDP-GlcNAc (96 h, Supplementary Fig. 14A,B), whereas no effect on cell numbers was observed when exposed to 20 mM mannose, an unrelated metabolite (96 h, Supplementary Fig. 14A,B).

This motivated us to combine 60 mM UDP-GlcNAc with the anti-androgen enzalutamide (a drug commonly used to treat men with CRPC28) to test whether the metabolite could improve efcacy of anti-androgen therapy on CRPC cells. As expected, addition of 10 mM enzalutamide to LNCaP-ABL for 96 h resulted in a modest reduction in cell numbers (Celigo, Supplementary Fig. 14A) and a pronounced decrease in viable cells, both of which were signicant (MTT: Fig. 4c). Similarly, treatment of AR-V7-containing 22Rv1 cells with 10 mM enzalutamide for 96 h had negligible effect on cell numbers (Celigo: Supplementary Fig. 14B) but a statistically signicant effect on viability (MTT: Fig. 4d). This modest cytostatic effect seen in 22Rv1 cells was primarily due to inhibition of AR signalling by enzalutamide as evidenced by signicant reduction in expression of two candidate AR targets, PSA and Hexokinase 2 (HK2, Supplementary Fig. 14E). Furthermore, combination of 20 or 60 mM UDP-GlcNAc with 10 mM enzalutamide displayed a signicant additive cytostatic effect in both 22Rv1 and LNCaPABL cells (MTT assay: Fig. 4c,d and Celigo assay: Supplementary Fig. 14A,B), compared with vehicle-treated controls.

As our ndings indicated a potential therapeutic value for the metabolite alone, we tested its efcacy on CRPC-like xenograft tumours in vivo. Intra-tumoral injection of 60 mM UDP-GlcNAc followed by mass spectrometry analysis of the CRPC xenografts showed the presence of 48 mM levels of the metabolite up to 48 h post injection (Supplementary Fig. 14F). Furthermore, treatment of LNCaP-ABL xenograft tumours with 60 mM UDP-GlcNAc bi-weekly resulted in a signicant (Po0.05,

Figure 2 | Effect of GNPNAT1 knockdown in CRPC-like cells. shRNA knockdown (KD) of GNPNAT1 in 22Rv1 cells (22R-KD1/5) leads to (a) signicant reduction in mRNA (n 5, representative data), (b) loss of protein (B21 kDa, n 5, representative immunoblot) and (c) signicant decrease in product/

substrate ratio (45 biological replicates) for GNPNAT1, compared with controls (22R-NT). Similarly, KD of GNPNAT1 in LNCaP-ABL cells (L-KD1/5) leads to (d) signicant reduction in mRNA (n 3, representative data), (e) loss of protein (n 3, representative immunoblot) and (f) signicant decrease in

product/substrate ratio (5 biological replicates) for GNPNAT1, compared with controls (L-NT). (g) GNPNAT1 KD in 22Rv1 cells signicantly increases proliferation compared with NT controls (5 biological replicates, 12 technical replicates). (h) Same as in g, but for GNPNAT1 KD in LNCaP-ABL cells compared with NT controls (3 biological replicates, 20 technical replicates). (i) 22Rv1 xenograft tumours containing GNPNAT1 KD show a signicantly higher growth rate compared with controls (n 7 KD1 and n 8 NT). Y axis represents median tumour volumes in mm3 and associated median absolute

deviation (MAD) for each group. (j) Same as in i, but for LNCaP-ABL cells containing GNPNAT1 KD (L-KD1 and L-KD5, 7 mice per condition) and NT controls (L-NT, 5 mice). L-KD cells grew signicantly faster than NT controls. (k) Representative images (n 4 NT, n 5 KD1) showing intra-cardiac

injection of luciferase containing 22Rv1 cells with GNPNAT1 KD or NT controls in castrated Nod-SCID Gamma mice (n 8 NT, n 10 KD1) showed

signicantly higher soft tissue metastasis. As shown in the representative gure, 4/5 KD mice and only 1/4 NT mice demonstrated metastatic lesions to the jaw and pelvic bones (white arrows). (l) Magnetic resonance imaging of representative mice shown in k. Yellow arrows show regions of metastasis in the brain as well as bones in the jaw and femur (L left, R right) supporting the luciferase data shown in k. ANOVA model was used to compute P-values

for panels shown in a,c,d,fh. For i,j showing in vivo data, a mixed effects linear model was tted to the data as described under the Methods. For boxplots, the horizontal line represents median value, whereas Whiskers represent either o25 or 475 quartile ranges.

Corroborating this, overall protein expression of SP1 was signicantly (Po0.05, Students t-test) increased in 22Rv1 KD cells (Fig. 3i and Supplementary Fig. 18). In silico analysis conrmed higher expression of SP1 in metastatic PCa tissues compared with primary tumours and benign prostate (Supplementary Fig. 11E). These ndings are consistent with earlier reports that suggest negative regulation of SP1 activity by glycosylation25. Consistent with this, global assessment of O-glycosylation using immunoblot analysis in both 22Rv1 and LNCaP-ABL cells containing GNPNAT1 KD revealed reduced glycosylation in a subset of proteins (Supplementary Figs 12A,C,E,27 and 28). This altered glycosylation prole was reversed in a number of proteins by add back of HBP end product UDP-GlcNAc (60 mM, 72 h, Supplementary Figs 12B,D,E,27 and 28).

We next asked how cell cycle genes could be regulated by SP1 in 22Rv1 KD cells. Our prior integrative analysis (Fig. 1c) revealed changes in glycolysis to be associated with changes in HBP. Earlier studies have demonstrated a link between carbohydrate response element-binding protein (ChREBP) and glycolytic intermediates (for example, xylulose-5-P) (ref. 26). Using LC-MS, we detected high levels of ribulose-5-P (Supplementary Fig. 13A), a structural isomer of xylulose-5-P, whereas chromatin immunoprecipitation (ChIP)-QPCR analysis demonstrated signicantly (Po0.05, Students t-test) increased binding of SP1 on the ChREBP promoter (Fig. 3j). Consistent with all of the above, mRNA microarray (Fig. 3g) and QPCR (Fig. 3k), protein (overall, Fig. 3l and Supplementary Fig. 18 nuclear and cytoplasmic, Supplementary Figs 13B and 29), combined nuclear and cytoplasmic staining (Supplementary Fig. 13C,D) and activity (Supplementary Fig. 13E) of ChREBP were elevated in 22Rv1 KD cells. A similar increase in ChREBP expression was also observed with KD of GFPT1 in 22Rv1 cells (Supplementary Fig. 8F). These ndings suggest a role for HBP in regulating ChREBP expression in CRPC-like cells. Moreover, a subset of cell cycle-associated genes previously shown to be transcriptional targets of ChREBP27 was also identied in the upregulated gene expression signature associated with GNPNAT1 KD in 22Rv1 cells (Supplementary Fig. 13F). This suggests an important role for ChREBP in regulating proliferation in CRPC-like cells upon modulation of HBP. Consistent with this, increased expression of ChREBP was also observed in human metastatic PCa tissues (Supplementary Fig. 13G).

Collectively, the above results imply that inherently decreased HBP activity in CRPC cells modulates proliferation via PI3K-AKT (in CRPC containing full-length AR; Fig. 4a) or SP1-ChREBP (in CRPC possessing AR-V7; Fig. 4b) to drive CRPC progression.

6 NATURE COMMUNICATIONS | 7:11612 | DOI: 10.1038/ncomms11612 | http://www.nature.com/naturecommunications

Web End =www.nature.com/naturecommunications

NATURE COMMUNICATIONS | DOI: 10.1038/ncomms11612 ARTICLE

GLM) reduction in the rate of tumour growth compared with vehicle controls or relative to an unrelated metabolite, 60 mM mannose (Fig. 4e and Supplementary Fig. 15A,B).

Overall, our novel integrative analysis takes into account pathway-based evidence, obtained independently from gene

expression and metabolic compartments. This method overcomes the potential lack of agreement between data points in transcriptomics and proteomics/metabolomics data sets, which has been an inherent bottleneck for concordance-based integrative approaches29. Our method relies on networks of reactomes

a b c

PIK3AP1

KLK3

PI3K-P85

Actin

140

LY294002 (50 M)

Perifosine (20 M)

85 kDa

60 kDa

289 kDa

44,42 kDa

45 kDa

BrdU-assay abs. (450 nm)

normalized to respective

untreated control

NEK1

ATP8A1

120

ATM

PI3K-19H8

mTOR

L-NT

100

ITGB2

PIK3CD

P- AKT (T308)

INSR

INPP4B

AKT

MAPK10

IGFBP2

0 L-NT L-KD1 L-KD5

PPP1R21

NCOA3

P-MAPK (ERK 1/2)

NCOR2

TMPRSS2

ITGA6

L-KD1

L-KD5

GAPDH

110 kDa

37 kDa

81 kDa

45 kDa

ITGA2

L-NT L-KD1

PIK3IP1

L-KD5

P=0.025

KLK3

L-NT L-KD1

L-KD5

Log 2 (fold change)

(mRNA)

2 1 0 1 2

P- AKT (S473)

2.0

1.5

1.0

0.5

0.0

L-NT L-KD1

Row Z score

L-NT L-KD1

L-KD5

2 AR NCOA3

GNPNAT1

KLK3

h i

CCNE1

Enrichment P=2.5e06

Log 2 (fold change)

(mRNA)

SMC4

22R-KD1

22R-KD5

ChREBP

Bits

CEP70

SP1

Actin

22R-NT

NCOA1

0 1 2 3 4 5 6 7 8 9 10

PIK3CA

Nucleotide position

CDC25A

P=5.8e05

CENPE

P=1.6e05

160

120

ChREBP

DHFR

E2F3

BUB1

ChREBP

Fold change (mRNA)

1.6

1.5

1.4

1.3

1.2

1.1

1.0

MAD2L1

CREB1

Fold enrichment

AURKA

22R-KD5

22R-NT 22R-KD1

CDK5R1

HDAC1

22R-KD5

22R-NT 22R-KD1

TFDP2

NEK2

65 kDa

PIK3CB

ChREBP

Actin

45 kDa

22R-KD5

22R-NT 22R-KD1

22R-KD5

22R-NT 22R-KD1

2 1 0 1 2

Row Z score

NATURE COMMUNICATIONS | 7:11612 | DOI: 10.1038/ncomms11612 | http://www.nature.com/naturecommunications

Web End =www.nature.com/naturecommunications 7

ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/ncomms11612

SASI_Hs02_00333099) was used to create transient KDs according to the manufacturers protocol. Similarly, luciferase (GenTarget Inc, cat# LVP402-PBS) expression in 22Rv1 cells was achieved using a virus titre of 2 MOI and cells selected using 800 mg ml 1 Geneticin. Overexpression vector for GNPNAT1 was obtained from the Cell-Based Assay Screening Service core at BCM.

Integrative analysis and biostatistics. The data set used in this study are supplied as Supplementary Data 1. For the integrative analysis, gene expression and metabolomics data for benign adjacent prostate tissue and localized PCa were used.Gene expression and metabolic data were ltered, imputed and normalized as described32.

An integrated analysis of Kyoto Encyclopedia of Genes and Genomes (KEGG) pathways based on gene and metabolic expression data was performed13,33,34 as described below. Briey, pathway enrichment based on gene expression and metabolic data was assessed using GSA (a variant of Gene Set Enrichment Analysis (GSEA) with improved power properties33) and NetGSA13,34, respectively. NetGSA belongs to the class of network topology-based pathway enrichment methods35 and incorporates information about connections between biomolecules (genes, proteins and metabolites) in a mixed effects linear regression model13,34. This mixed effects linear regression model allows for a rigorous testing procedure for pathway enrichment. In our earlier study13,34 using NetGSA to enrich pathways using gene expression data, we used existing gene network databases, such as BioGrid, YeastNet and BioCarta, to create the interaction network. In contrast, to apply NetGSA to metabolomics data, biochemical reactions among metabolites from KEGG were used to dene their respective interaction network. In this network, the nodes are metabolites and edges between metabolites are drawn based on information from biochemical reactions. In other words, an edge is drawn between two metabolites if they constitute substrateproduct pair of at least one biochemical reaction.

Pathways were ranked separately based on enrichment results from the gene expression and metabolic data. Subsequently, a combined score was derived and normalized for the variability in rankings observed in each data set. To estimate the variability of the combined scores, a bootstrap resampling procedure36 was employed wherein data from the tissue samples (in PCa and Benign classes) were randomly sampled with replacement.

For the resampling procedure, we denote the average ranking of the pathways j 1,y87 in 1,000 bootstrap samples of the gene expression (G) and metabolic

expression (M) using the terms R(j,G) and R(j,M) and their standard deviation using the terms S(j,G) and S(j,M), respectively. The nal integrative score for the jth pathway is then dened as the mean of the normalized rankings of each pathway across 1,000 bootstrap samples computed using the equation: R*(j) [88-R(j,G)/

S(j,G)] [88-R(j,M)/S(j,M)]. It is worth noting that the procedure implicitly

self-adjusts for signal strength and noise levels in the gene expression and metabolomics data sets, thus proving more powerful and robust than concordance methods37.

Upon examination of the distribution of the combined scores (Fig. 1b), four pathways were found to have a signicantly larger integrative score (Po0.03) than the rest, based on a normal test for outliers (Supplementary Table 2). To ensure that potentially important pathways were not missed, an additional pathway namely Cysteine metabolism (P 0.2), which was a borderline outlier, was

also included along with the top four enriched pathways for secondary analysis.

Overall, we selected the top ve pathways from the integrative analysis, termed henceforth as pathways derived from the primary enrichment analysis. In the next step, KEGG pathways interacting with these top ve enriched pathways were determined using information obtained from KEGG.

Figure 3 | Elucidating the mechanism of GNPNAT1 KD driving an aggressive phenotype in CRPC-like cells. (a) Heat map showing differential (Po0.05, FDR 30%) genes in LNCaP-ABL cells containing GNPNAT1 KD (L-KD1, n 4; and L-KD5, n 2) compared with NT controls (L-NT, n 3). (b) Immunoblot

showing increased expression of phosphorylated and total PI3K (B85 kDa, n 3 for PI3K-P85 and n 2 for PI3K-19H8), AKT (B60 kDa, n 4 for AKT,

n 3 for p-AKT (T308) and n 4 for p-AKT (S473)), m-TOR (B281 kDa, n 5) and p-MAPK (B44-42 kDa, n 3) in LNCaP-ABL KD cells compared with

controls. In all cases, representative immunoblots are shown. (c) Treatment of LNCaP-ABL NT and KD cells with either 20 mM Perifosine (AKT/PI3K inhibitor, (n 3) or 50 mM of LY294002 (PI3K) inhibitor (n 3), signicantly decreased KD1 cell proliferation and modestly reduced KD5, both compared

with treated NT control. (d) Immunoblot showing increased expression of AR in GNPNAT1 KD LNCaP-ABL cells compared with control (B110 kDa, n 3,

representative immunoblot). (e) QPCR validating increased expression of AR-target gene KLK3 (PSA) in LNCaP-ABL KD cells (n 3). (f) QPCR validating

increased expression of AR and its co-activator NCOA3 and downstream target KLK3 in xenograft tumours derived from LNCaP-ABL KD cells compared with NT control (n 4). (g) Heat map showing differential (Po0.05, FDR 20%) genes in 22Rv1 cells containing GNPNAT1 KD (22R-KD1 and 22R-KD5,

n 4 each) compared with NT controls (22R-NT, n 4). (h) Promoter Scan on genes upregulated in KD cells revealed enrichment for SP1-binding sites.

(i) Immunoblot showing signicantly (Po0.05) increased protein levels for SP1 (B81 kDa, n 5) in 22Rv1 cells containing GNPNAT1 KD (22R-KD1 and

22R-KD5, median band intensity of 0.660.23 and 0.510.08, respectively) compared with NTcontrol (median band intensity of 0.350.06). Actin was used to control for protein loading. (j) Chromatin immunoprecipitation (ChIP) assay-coupled to QPCR to verify SP1 binding to the promoter of ChREBP. SP1 binding to the promoter of DihydroFolate Reductase (DHFR) was used as a positive control (n 3, median and standard deviation shown). (k) QPCR

validation of ChREBP expression in 22Rv1 KD and NT cells (n 3). (l) Immunoblot showing elevated protein expression of ChREBP in 22Rv1 KD cells

compared with NT control (B65 kDa, n 3, representative immunoblot). P-values were computed using Unpaired Students t-test (for f) and ANOVA

model (c,e,j and k). For e and f, the fold change was log 2 transformed to get normal distribution. In all cases, P-value: *Po0.05. For boxplots, the horizontal line represents median value, whereas Whiskers represent either o25 or 475 quartile ranges. All bar plots are represented in medians.d.

that constitute pathways, as well as the interactions of these pathways with each other. Based on this approach, we identied HBP as a key biochemical mediator of CRPC progression. HBP-associated metabolic re-wiring observed in CRPC supports cell cycle via PI3K-AKT/SP1-ChREBP-axis. Further, replenishing the HBP in these cells via the addition of a downstream metabolite rescues the KD phenotype. We observed a contrasting effect of HBP on growth of AD PCa and CRPC cells further suggesting the existence of metabolic re-wiring during PCa progression. Notably, HBP has been known to be a nutrient sensor and shown by others to regulate glycolysis30. In light of this, one may speculate that reduction in HBP in CRPC could be a selective adaptation to the increased need for glycolysis required to meet the bioenergetic demands of an invasive tumour. This is supported by our recent publication that describes increased glycolysis as one of the main metabolic adaptations seen in CRPC-like cells31. Provocatively, treatment of CRPC with HBP metabolites UDP-GlcNAc and GlcN, in vitro and UDP-GlcNAc in vivo, signicantly reduced the proliferation of CRPC-like cells. The therapeutic effect in vitro was further enhanced by combining the metabolite with the anti-androgen enzalutamide. These ndings are particularly noteworthy given that CRPC cells containing the AR-V7 variant (that is, 22Rv1) are essentially resistant to enzalutamide. Although preliminary, our results introduce the innovative concept of using metabolites to overcome therapeutic resistance in advanced tumours via the exploitation of altered metabolic pathways.

Methods

Cell culture and lentiviral transduction or siRNA transfection. Human prostate cell line 22Rv1 was purchased from American Type Culture Collection, C4-2 and LNCaP-ABL cells were obtained from Drs Nancy Weigel and Nicholas Mitsiades, respectively. LNCaP cells were obtained from the Tissue Culture Core at Baylor College of Medicine (BCM). All cell lines were veried using Short-Tandem Repeat DNA ngerprinting at the MD Anderson Cancer Center and were tested negative for mycoplasma contamination using MycoAlert Detection Kit (Lonza, Cat# LT07-418). C4-2 cells were grown in T-media with 10% fetal bovine serum (FBS), whereas 22Rv1, LNCaP and LNCaP-ABL were grown in RPMI-1640 medium. Both 22Rv1 and LNCaP were grown in 10% FBS and phenol red, whereas LNCaP-ABL was grown in 10% charcoal stripped FBS without phenol red. One hundred micrograms per millilitre of Normocin was added to all the cell culture media. To generate stable GNPNAT1 and GFPT1 KD in CRPC-like cells, lentiviral transduction using shRNA (Sigma Aldrich, Product# SHCLNV-NM_198066 (GNPNAT1, sequence in Supplementary Table 6), SHCLNV-NM_002056 (GFPT1, sequence in Supplementary Table 6), two independent shRNA clones each) targeting these genes was carried out at a viral titre of 5 MOI. Cells with stable KD of the gene were cultured in respective media added with 0.5 mg ml 1 of puromycin. In addition, where indicated, GFPT1-specic siRNA from Life

Technologies (Cat numbers SASI_Hs01_00162618, SASI_Hs01_00162619 and

8 NATURE COMMUNICATIONS | 7:11612 | DOI: 10.1038/ncomms11612 | http://www.nature.com/naturecommunications

Web End =www.nature.com/naturecommunications

NATURE COMMUNICATIONS | DOI: 10.1038/ncomms11612 ARTICLE

CRPC(AR-FL) PI3K-AKT AR

Cell cycle genes

HBP

CRPC progression

CRPC(AR-V7) SP1 ChREBP

Cell cycle genes

HBP

CRPC progression

c d

MTT assay MTT assay

** **

** ** **

22Rv1 ** P

<0.01

Mean difference in

absorbance (570 nm) with CI

0.4

0.35

0.30

0.25

LNCaP-ABL ** P

<0.01

Mean difference in

absorbance (570 nm) with CI

0.32

0.28

0.24

Control

Enz (10 M)

UDP-GlcNAc

(60 mM)

Enz (10 M)+

UDP-GlcNAc

(60 mM)

Control

Enz (10 M)

UDP-GlcNAc

(60 mM)

Enz (10 M)+

UDP-GlcNAc

(60 mM)

400

PBS (n=4)

UDP-GlcNAc (

=4, 60 mM)

=3.4e10

% Change in tumour volume

relative to baseline

300

200

100

0 0 20 40 60

Days

Figure 4 | Therapeutic targeting of HBP in CRPC. (a) Decreased HBP expression in CRPC tumours containing AR-FL is associated with increased activity of PI3K-AKT that could activate AR. This activates cell cycle genes leading to increased proliferation driving CRPC progression. (b) In CRPC containing AR-V7, decreased HBP expression activates SP1-regulated ChREBP expression leading to stimulation of cell cycle genes, increased proliferation and tumour progression. (c) MTT assay results for LNCaP-ABL (n 9) cells treated with vehicle (PBS dimethylsulphoxide) or 10 mM enzalutamide (Enz) or 60 mM

UDP-N-acetylglucosamine (UDP-GlcNAc) or both enz UDP-GlcNAc for 96 h. (d) Same as in c but treatments were done on 22Rv1 cells (n 8). In both

c and d, differences in the level of mean absorbance with corresponding condence interval (CI) to control obtained for each of the conditions (x axis) is represented on the y axis. Comparisons of all treatments with control group in c,d were signicant at a P-value of 0.01 (Bonferroni corrected). Pairwise comparisons of either Enz or combination of Enz UDP-GlcNAc with other treatments were signicant at a P-valueo0.05. Treatment with Enz resulted in

a modest to pronounced decrease in proliferation of 22Rv1 and LNCaP-ABL cells, respectively. Compared with Enz, treatment with UDP-GlcNAc alone resulted in a signicant reduction in cell proliferation in both the CRPC-like cells. Furthermore, the effect of Enz UDP-GlcNAc was signicantly higher in

CRPC-like cells, compared with any single treatment. Error bars represent s.e.m. (e) Plot showing rate of growth of LNCAP-ABL xenograft tumours (set 1, see the Methods for details) treated bi-weekly with either 60 mM UDP-GlcNAc (grey lines, n 4 mice) or PBS (black lines, n 4 mice), for a period of 72

days. For each animal, the tumour volume at each time point was normalized to its baseline value obtained before starting the treatment. The percent change in tumour volume for all the animals in group was calculated along with the associated median absolute deviation (MAD) and plotted against time (in days). P-value was calculated using GLM model as described in the text. Importantly, mice treated with UDP-GlcNAc showed a signicant reduction in rate of tumour growth compared with the control-treated tumours (P-value 3.4e 10). Error bars represent MAD.

Next, we examine whether the neighbours of these ve pathways (obtained from our integrative analysis) are also enriched based solely on gene expression data (Fig. 1c). This is based on the premise that gene expression data capture the encoded functional potential of the cell and hence could reveal interacting pathways that are most likely to be activated in the near term.

Procedurally for this secondary analysis, we rst constructed the interaction network between 87 KEGG-dened pathways. The nodes of this networkare KEGG pathways and an edges was drawn between two pathways if gene/ metabolites members of one pathway interacted with gene/metabolite members of the other. We then used this information to determine the number of interacting pathways for each of the top ve pathways obtained using the primary enrichment analysis. A total of 20 interacting pathways were identied for the top ve pathways

derived from the integrative analysis. Following this, using gene expression data we looked for enrichment of these 20 interacting pathways and rank ordered them based on their enrichment score. From this list, we next selected the top ten interacting pathways (that is, pathways that were ranked within the top 50%) and asked how many of these interact with the top ve pathways derived from the primary enrichment analysis (Supplementary Fig. 1D) and used a network-based permutation test to assess their signicance. This procedure identied the HBP as the key pathway in PCa (Supplementary Table 3).

To carry out the network-based permutation test, we denoted by N(1), y, N(5) the number of enriched neighbours of each of the top ve pathways based on the gene expression data. We then generated B 10,000 random networks by

scrambling the patterns of interactions among pathways, while preserving the

NATURE COMMUNICATIONS | 7:11612 | DOI: 10.1038/ncomms11612 | http://www.nature.com/naturecommunications

Web End =www.nature.com/naturecommunications 9

ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/ncomms11612

number of interactions for each pathway as dened based on KEGG (Fig. 1c). That is to say that the randomization process maintained the number of interacting nodes (pathways) and their degree distribution. Procedurally, this was done as follows. Let N(1,b), y, N(5,b) denote the number of enriched interacting nodes for the ve pathways obtained from the integrative analysis above in the b-th generated random network. The signicance measure for association of the interacting neighbours for the top ve enriched pathways resulting from the original integrative analysis is determined using the equation P(j) 1/B P

(N(j,b)4N(j)), where j 1,y5 and B 10,000. Importantly, the proposed

network-based permutation test preserves the topology of the pathway interaction network thus preventing inated false positive errors. R software code used for integrative analysis is available as Supplementary Information.

Microarray analyses. Human Affymetrix Human Genome U219 Array Plate was used for gene expression analysis with values for each gene being calculated using robust multi-array average38. Data were tted in weighted linear models, which are designed specically for microarray analysis39. The data were further ltered using principal component analysis to identify samples that do not cluster with the rest of the samples within a group. In LNCaP-ABL cells, this resulted in removal of three samples (one NT and two KD) from further analysis. ANOVA statistics was computed for each probe set, corrected for false discovery rate (FDR40) and mapped for over-represented pathways using ConsensusPathDB (http://cpdb.molgen.mpg.de

Web End =http://cpdb.molgen.mpg.de).

Metabolomic analyses. Mass spectrometry-based multiple reaction monitoring (MRM) studies were done as described41. A list of MRM transitions is given in Supplementary Table 4.

For the targeted metabolic measurements described in this manuscript, P-value was computed using t-test/ANOVA coupled to multiple comparison correction using FDR correction method described in FDRtool package in R software.

Reagents used for metabolite analysis. Reagents such as acetonitrile, methanol and water for high-performance liquid chromatography were purchased from Burdick and Jackson. Formic acid and internal standards, which included a mixture of Jasmonic acid, [15N]2-Tryptophan, Thymine-d4, Glutamic acid-d5, Gibberellic acid, Trans-Zeatin, [15N]-Anthranilic acid and Testosterone-d3, were purchased from Sigma-Aldrich. For UDP-N-Acetyl glucosamine quantication, [13C]6

N-acetyl glucosamine was used as an internal standard and the calibration curve for quantication was generated using the puried metabolite purchased from Sigma-Aldrich.

Quality controls for metabolic analysis. Aliquotes of isotopic labelled internal standards described above were used at a nal concentration of 0.25 mM. Instrument performance was monitored using matrix-free mixture of internal standards reconstituted in a 50:50 (v/v) mixture of methanol and water. Furthermore, for relative quantication of metabolite levels, isotopic labelled internal standards spiked into the samples were used to control for variation in metabolite extraction procedure. Along with experimental samples, we also extracted the metabolites from pooled liver samples (spiked with internal standards), which were analysed in tandem with experimental samples. This allowed us to control for variation from both extraction procedure and instrument performance. Similarly, for absolute quantication of UDP-N-acetyl glucosamine, calibraton curve was generated by spiking in [13C]6 N-acetyl glucosamine in a matrix containing dialysed tissue extract.

Metabolite extraction. The metabolite extraction protocol used in this study was described earlier32. Briey, both cell line and tissue samples were either sonicated or homogenized, respectively, in methanol/water (4:1) containing mixture of internal standards. Before this step, the cell line samples were subjected to multiple freezethaw cycles to achieve homogenous rupture of cell membranes. The cell line or tissue homogenate was then diluted with 450 ml of ice-cold chloroform and vortexed using a Multi-Tube Vortexer for 10 min. Following this, 150 ml of ice-cold water was added to the homogenate and vortexed for an additional 2 min. The homogenate was then incubated at 20 C for about 20 min to allow for aqueous

and organic phase separation. Following separation of organic and aqueous phases, each of them was dried at 37 C for 45 min using EZ-2 series Genevac Speed Vac system. Next, the dried aqueous samples were reconstituted in 50:50 mixture of methanol/water and subjected to de-proteination using a 3-kDa molecular lter at 4 C for 90 min. Following this, the ltrate was dried, reconstituted in 50:50 mixture of methanol/water containing 0.1% formic acid and separated by liquid chromatography before mass spectrometry analysis.

Liquid chromatography and mass spectrometry analysis. Separation of meta-bolites prior mass spectrometry was carried out by reverse phase chromatography on a Phenomenex Synergi Max-RP (250 4.6 mm2, 80 , Phenomenex) column

using an Agilent 1290 series HPLC system (Agilent Technologies). Mobile phase used was composed of 1 mM ammonium acetate with 0.05% ammonium hydro-xide/water (v/v; solvent A) and 1 mM ammonium acetate with 0.05% ammonium

hydroxide/acetonitrile (v/v; solvent B). Gradient parameters used for the separation were: 0 min-5% B; 1517 min-10% B, 22 min-90% B, 2328 min-5% B, followed by re-equilibration to the initial starting condition. Flow rate of 0.3 ml min 1 was optimized for the experiments.

Relative or absolute levels of metabolites were quantied using MRM (refer Supplementary Table 4 for MRM transitions). MRM-based mass spectrometry analysis was performed on 6490 QQQ-LC/MS (Agilent Technologies) using electrospray ionization. Both positive and negative ionization modes were employed using capillary voltages of 3,500 V (positive ionization) and 3,000 V

(negative ionization), respectively, a collision gas ow rate of 10 l min 1 and a nebulizer gas ow rate of 35 l min 1. Throughout these experiments, nebulizer gas temperature was maintained at 350 C. Nitrogen was used as the collision gas at a collision cell pressure of 2.39 10 5 Torr. For maximum sensitivity, parameters

for fragmentor voltage and collision energy for each of the metabolites assessed were optimized using the optimizer software (Agilent Technologies).

RNA and QPCR analysis. Standard RNA extraction (RNeasy Mini Kit from Qiagen), reverse transcription (Quanta Biosciences, cat# 95048-500) and SYBR green (Life Technologies, cat # 4385614) Q-PCR were performed41. Primers used in this study are listed in Supplementary Table 5.

Western blot. Antibodies to GNPNAT1 (Proteintech, cat # 16282-1-AP, 1:2,000), GFPT1 ( Santa Cruz Biotechnology, cat# sc-377479), SP1 (Cell Signaling, cat # 9389, 1:1000), ChREBP (Novus, cat # NB400-136V, 1:1,000), AR (Santa Cruz Biotechnology, cat # sc-816, sc-7305), AR-V7 (Precision Antibody, cat # AG10008), PI3K-p85 (BD Biosciences, Cat # 610045) and Actin (used as loading control; Sigma, 1:7500) were used. All the remaining antibodies were from Cell Signaling Technology and used at 1:1,000 dilution. Nuclear and cytoplasmic lysateswere prepared using NE-PER Nuclear and Cytoplasmic Extraction Reagents (Life Technologies, Cat # 78833) per the manufacturers instructions.

O-glycosylation immunoblots were performed on total protein lysates using O-GlcNAc antibody (mouse, Santa Cruz Biotechnology, cat# sc-59623, 1:1,000). Blots were scanned and immunoreactive bands having molecular weight greater than 50 kDa were quantied using Image J software. For each sample, the intensity of glycosylated immunoreactive band(s) was normalized rst to their respective GAPDH. Following this, the median intensity of each normalized band was calculated across multiple biological replicates (n 3 for 22Rv1 and n 2 for

LNCaP-ABL) and data represented as plots in Supplementary Fig. 12E. Original scans of all western blots are shown in Supplementary Figs 1629.

Tissue microarray. Tissue microarray was immunostained using GNPNAT1 antibody (Proteintech, 1:500 dilution) and UAP1 antibody (Abcam, 1:500 dilution) and scored by Dr Ittmann, a genito-urinary specialty trained pathologist. All tissues were collected under an institutional review board protocol and with informed consent at BCM and University of Michigan (UMICH).

BrdU and ChIP assays. BrdU assay (Calbiochem/Millipore, cat # 80002-206) was performed per the manufacturers instructions. ChIP assays were done following the manufacturers protocols (EMD Millipore Corporation, cat # 1710461). Immunoprecipitation was carried out using either 10 mg of rabbit SP1 antibody (Cell Signaling) or using rabbit non-immune IgG (Sigma), as a control. Fold change in enrichment of SP1 occupancy on gene-specic promoters was calculated relative to the promoter binding obtained using non-immune IgG (negative control).

FluoReporter assay. The assay was performed by quantication of DNA content using uoreporter assay (Themo Fisher Scientic, Cat # F-2962) as per the manufacturers protocol and as described previously42. Briey, number of cells seeded in a 96-well plate were optimized for both LNCaP-ABL and 22Rv1 cells with GFPT1 siRNA or control siRNA. For each experiment, empty wells were used as background control. Furthermore, uorescence was detected at 360 nm (excision) and 460 nm (emission) using a uorescence microplate reader. The intensity values of blank wells were substrated from experimental data and for each sample (GFPT1 targeted siRNA) data were normalized to control wells (with control siRNA).

Celigo-based cell counting. Celigo-based counting was used to determine change in cell numbers either after overexpression of GNPNAT1 or to assess the effect of metabolite treatments in CRPC-like cells. For all cell lines used for Celigo-based cell counting assay, optimal cell numbers were determined using parameters suggested by the manufacturer (Nexcelom Bioscience LLC). Approximately 2,000 and 3,000 cells, respectively, for 22Rv1 and LNCaP-ABL were seeded in replicates on day 0 in independent 96-well plates (Greiner 655090). After 24 h of plating, cells were subjected to treatment if any and scanned for their cell numbers at 0, 24, 48, 72 and 96 h post treatment or followed up to 6 days post transfection (GNPNAT1 over expression). At the end of 96 h of culture or 6 days post transfection, cells were visualized with diamidino-2-phenylindole, imaged using a blue lter and counted.

10 NATURE COMMUNICATIONS | 7:11612 | DOI: 10.1038/ncomms11612 | http://www.nature.com/naturecommunications

Web End =www.nature.com/naturecommunications

NATURE COMMUNICATIONS | DOI: 10.1038/ncomms11612 ARTICLE

For each treatment group, the median cell numbers across all replicates were calculated across the various time points. Following this, the cell numbersin the treatment group were represented as percentage of cells relative to the control group, along with their associated standard deviation. P-values to assess signicance of the effect were computed using an ANOVA test in R software.

Luciferase reporter assay. Plasmid containing ChREBP-luciferase was a kind gift from Dr Lawrence Chan, BCM. This was transfected using Lipofectamine 2000 (Life Technologies) into 22Rv1 KD and NT cells. Forty-eight hours later, luciferase signal was measured using a luminometer after adding Steady-Glo Luciferase reagent from Promega Inc. (cat # E2510) per the manufacturers protocol(data were normalized using total protein content in the cells).

High throughput microscopy and image analysis. 22Rv1 NT, KD1 and KD5 cells were seeded on optical glass-bottom 96-well plates (Greiner Bio-One), xed and labelled with ChREBP (Novus, cat # NB400-136V, 1:500) antibody following standard immunouorescence protocols. Plates were imaged using an IC-200 high-throughput microscope (Vala Sciences) with a 20/0.75 numerical aperture

objective. For each sample, a minimum of eight elds per well were imaged. Image analysis was performed for each cell using an automated mIA software platform43 and the total averaged intensity was normalized to the respective cell area. The normalized intensity was then used to calculate population marginal mean difference for expression of ChREBP in 22Rv1 NT and KD cells.

In vivo xenografts. All mouse experiments performed at either the BCM or the UMICH were approved following the guidelines of BCM Institutional Animal Care and Use Committee and Unit of Laboratory Animal Medicine- UMICH, respectively. Male SCID-Beige Fox Chase and Nod-SCID Gamma mice (between 6 and 8 weeks of age) were obtained, respectively, from Michael Lewis laboratory at BCM and Charles River laboratory (Frederick, MD). The mice were used to generate xenografts using either 22Rv1 (experiments conducted at both BCM and UMICH) or LNCaP-ABL (study conducted at UMICH) cells that were stably transduced with either control shRNA or shRNA targeting GNPNAT1. To generate the xenografts, subcutaneous injection of 100 ml of a mixture containing cells (50 ml of 3 106 22Rv1 or 1 106 LNCaP-ABL in 1 DPBS) and Matrigel (50 ml,

5 mg ml 1 protein concentration; Corning cat #354248) was carried out in the hind-ank area of the mice under isourane anaesthesia. Tumours were measured using calipers either twice (LNCaP-ABL) or thrice (22Rv1) per week for minimum of 8 weeks. Tumour volume was calculated using the formula TV (length

width2) 0.52 (ref. 44). In case of tumours generated using luciferase-labelled

22Rv1 cells, tumour growth was additionally monitored using bioluminescence imaging with a Xenogen IVIS 200 after intraperitoneal D-luciferin (100 ml of10 mg ml 1 solution) injection. Intra-cardiac injections (performed at UMICH)

using luciferase-labelled 22Rv1 cells used 2.5 105 cells in 100 ml of DPBS. Mice

were anaesthetized using isourane. Successful injections were assessed by measuring the initial bioluminescence at the tumour locations. Mice with high bioluminescent signal in areas around or in the heart during the rst 2 weeks were deemed unsuccessful, and removed from the study. At the end of the study, the tumours were harvested and ash frozen for further analysis. Tumours were observed in all the mice injected with LNCaP-ABL cells (both NT and KD). L-KD5 tumours reached 1,000 mm3 by day 53 at which point these animals were killed. Figure 2j shows the relative tumour growth curve across all groups up to this point. L-KD1 and L-NT mice were killed around days 5587. By the end of the experiment, the median tumour volumes for L-KD1 and L-NT were 908 and 859 mm3, respectively. All the molecular analysis was performed on tumours resected at the end of the experiment. In case of mice injected with 22Rv1 cells (Fig. 2i), 7 and 8 mice had tumours in the KD and NT groups, respectively. Mice in which tumours did not form were removed from the study. Mice injected with 22Rv1 labelled with luciferase (Supplementary Fig. 6AC) had 9 and 10 tumours, respectively, in the NT and KD groups. For the analysis of the data on tumour volume (Fig. 2i,j and Supplementary Fig. 6A), a GLM in the R software was used. Here both the treatment (that is, NT versus KD1 versus KD5 for ABL, and NT versus KD for 22Rv1) and the time of treatment were used as xed effects. The sample replicates were treated as random effects. Model parameter estimates were obtained using a maximum likelihood method and implemented in the lme4 package in the R statistical language. The corresponding P-values of the test statistic computed using GLM for the treatment effects are reported. We have not selected a specic time point to assess signicance. Rather, the GLM model uses the rate of tumour growth over the entire time period of treatment to assess the overall signicance of the effect.

Magnetic resonance imaging. MRI was used to conrm the presence of metastasis in jaw and pelvic bones in mice injected intra-cardially using 22Rv1 GNPNAT1 KD and NT control cells. MRI was carried out on a 7T Agilent Direct Drive system (Agilent, Inc.) with a quadrature mouse body volume coil (m2m Imaging Corp.). During all MRI procedures, animals were anaesthetized witha 12% isourane/air mixture while maintaining body temperature using a heated air system (Air-Therm Heater, World Precision Instruments). Anatomical MR images were acquired by a fast spin echo sequence with the following

parameters: repetition time/echo time 4,000/15 ms, eld of view 40 30,

matrix size 128 128, slice thickness 0.5 mm, echo train 8, echo

spacing 15 ms and number of slices: 25B30. Mice with tumours detected by

bioluminescence imaging were imaged with MRI to conrm the tumour presence. Contrast-enhancement was performed by i.p. administration of 50 ml of 0.5 M gadolinium-DTPA (Magnevist, Bayer Healthcare Pharmaceuticals) 5 min prior to image data acquisition. Delineation of tumour from healthy tissue was determined using a contrast-enhanced T1- or T2-weighted spin-echo images.

UDP-GlcNAc and mannose treatments in vivo. For UDP-GlcNAc and mannose treatments, the metabolites (Sigma-Aldrich) were resuspended in PBS. Intratumoral UDP-GlcNAc or mannose injections (50 ml at 60 mM) or PBS alone, were performed twice a week into castrated Nod-SCID Gamma male mice bearing xenograft tumours generated using wild-type LNCaP-ABL cells. The experiments were done in three independent sets. To generate tumours, 5 million cells were injected in set 1 and 10 million cells were used in sets 2 and 3, respectively. Set 1 contained four animals each treated with PBS (vehicle control) and 60 mM UDP-GlcNAc, after the xenografts reached an average volume of 300 mm3(Fig. 4e). Tumours were monitored for 72 days. Set 2 contained seven animals each treated with PBS (vehicle control) and 60 mM UDP-GlcNAc, after the xenografts reached an average volume of 100 mm3 (Supplementary Fig. 15A). Tumours were monitored for 38 days. Set 3 contained ve and six animals treated with mannose (60 mM, un-related metabolite control) and 60 mM UDP-GlcNAc, respectively, after the xenografts reached an average volume of 100 mm3 (SupplementaryFig. 15B). Tumours were monitored for 40 days. Tumour implantation sites were pre-treated with iodine swab stick before an injection. Measurement of tumour volume was carried out every 4 days in rst and third set of experiments and every 3 days in the second set of experiments. For each animal, the tumour volume at each time point was normalized to its baseline value obtained before starting the treatment. The percent change in the median tumour volume for all the animals in group was calculated along with the associated median absolute deviation and plotted against time (in days). P-value was calculated using GLM model as described earlier. Animals were euthanized once the tumour size reached a volume of 1,000 mm3.

Treatment of CRPC cells using metabolites or enzalutamide. Cell viability and cell numbers post treatment with metabolites or enzalutamide or both were done using MTT assay and Celigo-based cell counting (refer to section marked Celigo-based cell counting, above), respectively. For MTT assay, 10,000 LNCaP-ABLor 22Rv1 cells were plated independently in replicates (total of either 6 or 12 replicates) in 48-well plates and treated with either PBS/dimethylsulphoxide (vehicle) or 10 mM enzalutamide or 60 mM UDP-GlcNAc or a combination of10 mM enzalutamide and 60 mM UDP-GlcNAc. Ninety-six hours post treatment, cell viability was assessed using MTT assays per the manufacturers protocols. Data were analysed using an ANOVA model generated with three factors: (i) treatment,(ii) cell line and (iii) experimental replication. Subsequently, differences in the mean level of the treatments for different cell lines were compared controlling the Family Wise Error Rate (for multiple comparisons) at the a 0.01 level, using

Bonferroni method45. ANOVA model was used to compute P-values comparing either enzalutamide or combination of enzalutamide and UDP-GlcNAc with other treatments.

References

1. Zhang, F. & Drabier, R. IPAD: the integrated pathway analysis database for systematic enrichment analysis. BMC Bioinformatics 13 (Suppl 15) S7 (2012).

2. Mewes, H. W. et al. MIPS: curated databases and comprehensive secondary data resources in 2010. Nucleic Acids Res. 39, D220D224 (2011).

3. Pastrello, C. et al. Visual data mining of biological networks: one size does not t all. PLoS Comput. Biol. 9, e1002833 (2013).

4. Baca-Lopez, K., Mayorga, M., Hidalgo-Miranda, A., Gutierrez-Najera, N. & Hernandez-Lemus, E. The role of master regulators in the metabolic/ transcriptional coupling in breast carcinomas. PLoS ONE 7, e42678 (2012).

5. Hogan, L. E. et al. Integrated genomic analysis of relapsed childhood acute lymphoblastic leukemia reveals therapeutic strategies. Blood 118, 52185226 (2011).

6. Grasso, C. S. et al. The mutational landscape of lethal castration-resistant prostate cancer. Nature 487, 239243 (2012).

7. Waltering, K. K., Urbanucci, A. & Visakorpi, T. Androgen receptor (AR) aberrations in castration-resistant prostate cancer. Mol. Cell. Endocrinol. 360, 3843 (2012).

8. Shari, N. Minireview: Androgen metabolism in castration-resistant prostate cancer. Mol. Endocrinol. 27, 708714 (2013).

9. Taylor, B. S. et al. Integrative genomic proling of human prostate cancer. Cancer Cell 18, 1122 (2010).

10. Sreekumar, A. et al. Metabolomic proles delineate potential role for sarcosine in prostate cancer progression. Nature 457, 910914 (2009).

NATURE COMMUNICATIONS | 7:11612 | DOI: 10.1038/ncomms11612 | http://www.nature.com/naturecommunications

Web End =www.nature.com/naturecommunications 11

ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/ncomms11612

11. Biernacka, J. M. et al. Genome-wide gene-set analysis for identication of pathways associated with alcohol dependence. Int. J. Neuropsychopharmacol. 16, 271278 (2013).

12. Subramanian, A. et al. Gene set enrichment analysis: a knowledge-based approach for interpreting genome-wide expression proles. Proc. Natl Acad. Sci. USA 102, 1554515550 (2005).

13. Shojaie, A. & Michailidis, G. Analysis of gene sets based on the underlying regulatory network. J. Computat. Biol. 16, 407426 (2009).

14. Hanover, J. A., Krause, M. W. & Love, D. C. Bittersweet memories: linking metabolism to epigenetics through O-GlcNAcylation. Nat. Rev. Mol. Cell Biol. 13, 312321 (2012).

15. Marshall, S., Bacote, V. & Traxinger, R. R. Discovery of a metabolic pathway mediating glucose-induced desensitization of the glucose transport system. Role of hexosamine biosynthesis in the induction of insulin resistance. J. Biol. Chem. 266, 47064712 (1991).

16. Marshall, S., Garvey, W. T. & Traxinger, R. R. New insights into the metabolic regulation of insulin action and insulin resistance: role of glucose and amino acids. FASEB J. 5, 30313036 (1991).

17. Schleicher, E. D. & Weigert, C. Role of the hexosamine biosynthetic pathway in diabetic nephropathy. Kidney Int. Suppl. 77, S13S18 (2000).

18. Itkonen, H. M. et al. O-GlcNAc transferase integrates metabolic pathways to regulate the stability of c-MYC in human prostate cancer cells. Cancer Res. 73, 52775287 (2013).

19. Wang, Q. et al. Androgen receptor regulates a distinct transcription program in androgen-independent prostate cancer. Cell 138, 245256 (2009).

20. Ghosh, P. M. et al. Signal transduction pathways in androgen-dependent and -independent prostate cancer cell proliferation. Endocr. Relat. Cancer 12, 119134 (2005).

21. Chan, S. C., Li, Y. & Dehm, S. M. Androgen receptor splice variants activate androgen receptor target genes and support aberrant prostate cancer cell growth independent of canonical androgen receptor nuclear localization signal.J. Biol. Chem. 287, 1973619749 (2012).22. Jiang, H., Fan, D., Zhou, G., Li, X. & Deng, H. Phosphatidylinositol 3-kinase inhibitor(LY294002) induces apoptosis of human nasopharyngeal carcinoma in vitro and in vivo. J. Exp. Clin. Cancer Res. 29, 34 (2010).

23. Richardson, P. G., Eng, C., Kolesar, J., Hideshima, T. & Anderson, K. C. Perifosine, an oral, anti-cancer agent and inhibitor of the Akt pathway: mechanistic actions, pharmacodynamics, pharmacokinetics, and clinical activity. Expert Opin. Drug Metab. Toxicol. 8, 623633 (2012).

24. Zambelli, F., Pesole, G. & Pavesi, G. Pscan: nding over-represented transcription factor binding site motifs in sequences from co-regulated or co-expressed genes. Nucleic Acids Res. 37, W247W252 (2009).

25. Yang, X. et al. O-linkage of N-acetylglucosamine to Sp1 activation domain inhibits its transcriptional capability. Proc. Natl Acad. Sci. USA 98, 66116616 (2001).

26. Iizuka, K., Wu, W., Horikawa, Y. & Takeda, J. Role of glucose-6-phosphate and xylulose-5-phosphate in the regulation of glucose-stimulated gene expression in the pancreatic beta cell line, INS-1E. Endocr. J. 60, 473482 (2013).

27. Poungvarin, N. et al. Genome-wide analysis of ChREBP binding sites on male mouse liver and white adipose chromatin. Endocrinology 156, 19821994 (2015).

28. Beer, T. M. & Tombal, B. Enzalutamide in metastatic prostate cancer before chemotherapy. N. Engl. J. Med. 371, 17551756 (2014).

29. McRedmond, J. P. et al. Integration of proteomics and genomics in platelets: a prole of platelet proteins and platelet-specic genes. Mol. Cell Proteomics 3, 133144 (2004).

30. Yi, W. et al. Phosphofructokinase 1 glycosylation regulates cell growth and metabolism. Science 337, 975980 (2012).

31. Sha, A. A. et al. Differential regulation of metabolic pathways by androgen receptor (AR) and its constitutively active splice variant, AR-V7, in prostate cancer cells. Oncotarget 6, 3199732012 (2015).

32. Putluri, N. et al. Metabolomic proling reveals a role for androgen in activating amino acid metabolism and methylation in prostate cancer cells. PLoS ONE 6, e21417 (2011).

33. Efron, B. & Tibshirani, R. On testing the signicance of sets of genes. Ann. Appl. Stat. 1, 107109 (2007).

34. Shojaie, A. & Michailidis, G. Network enrichment analysis in complex experiments. Stat. Appl. Genet. Mol. Biol. 9, Article22 (2010).

35. Khatri, P., Sirota, M. & Butte, A. J. Ten years of pathway analysis: current approaches and outstanding challenges. PLoS Comput. Biol. 8, e1002375 (2012).

36. Efron, B. A. T. & R. J. (eds). An Introduction to the Bootstrap (Chapman & Hall, 1993).

37. Imielinski, M. et al. Integrated proteomic, transcriptomic, and biological network analysis of breast carcinoma reveals molecular features of tumorigenesis and clinical relapse. Mol. Cell Proteomics 11, M111 014910 (2012).

38. Irizarry, R. A. et al. Summaries of affymetrix GeneChip probe level data. Nucleic Acids Res. 31, e15 (2003).

39. Smyth, G. K. Linear models and empirical bayes methods for assessing differential expression in microarray experiments. Stat. Appl. Genet. Mol. Biol. 3, Article3 (2004).

40. Benjamini, Y. & Hochberg, Y. Controlling the false discovery rate:a practical and powerful approach to multiple testing. J. R. Stat. Soc. B 57, 289300 (1995).

41. Putluri, N. et al. Metabolomic proling reveals potential markers and bioprocesses altered in bladder cancer progression. Cancer Res. 71, 73767386 (2011).

42. Shi, Y. et al. Androgens promote prostate cancer cell growth through induction of autophagy. Mol. Endocrinol. 27, 280295 (2013).

43. Szafran, A. T. & Mancini, M. A. The myImageAnalysis project: a web-based application for high-content screening. Assay Drug Dev. Technol. 12, 8799 (2014).

44. OReilly, M. S. et al. Endostatin: an endogenous inhibitor of angiogenesis and tumor growth. Cell 88, 277285 (1997).

45. Dunn, O. J. Multiple comparisons among means. J. Am. Stat. Assoc. 56, 52 (1961).

Acknowledgements

We acknowledge the joint participation by Diana Helis Henry Medical Research Foundation through its direct engagement in the continuous active conduct of medical research in conjunction with the Baylor College of Medicine. This research was partially supported by the following grants: NIH 1RO1CA133458-01 (A.S.K.), NIH U01 CA167234 and DOD W81XWH-12-1-0130 (A.S.K., G.S.P. and G.M.), NSF DMS-1161759 (A.S.K., G.M., A.S.), NIH RCA145444A (G.M. and A.S.K.), NSF DMS-12-28164 (G.M.), U01 CA111275 and P50CA186786 (A.M.C.), NIH 1R21CA205257-01 (S.E.M), NIH Grant 5F30CA196108-02 (D.A.B.), NIH R01 CA184208 (D.E.F.), NIH R21CA173150 and R21CA179720 (B.A.K.), ACS 127430-RSG-15-105-01-CNE (N.P.), NSF DMS-11617838 (G.M.) and RP150451 (A.S.K.), from CPRIT, as well as funds from the Alkek Center for Molecular Discovery (A.S.K.). This project was also supported by the Agilent Technologies Center of Excellence in Mass Spectrometry at Baylor College of Medicine, Integrated Microscopy Core and shared Proteomics and Metabolomics core at Baylor College of Medicine with funding from the NIH (HD007495, DK56338and P30 CA125123), CPRIT Proteomics and Metabolomics Core Facility (D.P.E.), (RP120092), the Dan L. Duncan Cancer Center and the John S. Dunn Gulf Coast Consortium for Chemical Genomics.

Author contributions

Project was designed by A.S.K., G.M., G.S.P., A.K.K., A.S. and K.P., integrative analysis was performed by A.S. and G.M., data analysis was performed by A.K.K., A.S., K.P., R.S., H.V., M.M., A.Z., V.P., A.T.S., F.S., M.M.I., N.P., G.M. and A.S.K., experiments were performed by A.K.K., K.P., R.S., H.V., M.M., A.Z., V.P., V.T.V., Y.Z., A.S. Khan, S.L., S.D., D.A.B., H.L., S.S., X.C., E.T., S.H., B.A.K., R.M., M.M.I., N.P., A.S.K., manuscript writing and editing was done by A.K.K., A.S., K.P., A.Z., N.L.W., N.P., G.S.P., G.M. and A.S.K. Critical scientic inputs were given by D.E.F., L.C., D.P.E., B.A.K., N.M., N.L.W., M.M., S.E.M., R.M., M.M.I., A.M.C., N.P., G.S.P., G.M. and A.S.K.

Additional information

Accession codes: The gene-expression data for GNPNAT1 KD 22Rv1 and LNCaP-ABL cells have been deposited in the GEO database and can be accessed using the GEO ID GSE67537.

Supplementary Information accompanies this paper at http://www.nature.com/naturecommunications

Web End =http://www.nature.com/ http://www.nature.com/naturecommunications

Web End =naturecommunications

Competing nancial interests: The authors declare no competing nancial interests.

Reprints and permission information is available online at http://npg.nature.com/reprintsandpermissions/

Web End =http://npg.nature.com/ http://npg.nature.com/reprintsandpermissions/

Web End =reprintsandpermissions/

How to cite this article: Kaushik, A. K. et al. Inhibition of the hexosamine biosynthetic pathway promotes castration-resistant prostate cancer. Nat. Commun. 7:11612doi: 10.1038/ncomms11612 (2016).

This work is licensed under a Creative Commons Attribution 4.0 International License. The images or other third party material in this article are included in the articles Creative Commons license, unless indicated otherwise in the credit line; if the material is not included under the Creative Commons license, users will need to obtain permission from the license holder to reproduce the material. To view a copy of this license, visit http://creativecommons.org/licenses/by/4.0/

Web End =http://creativecommons.org/licenses/by/4.0/

12 NATURE COMMUNICATIONS | 7:11612 | DOI: 10.1038/ncomms11612 | http://www.nature.com/naturecommunications

Web End =www.nature.com/naturecommunications

Word count: 12016

Show less

Abstract

Translate

The precise molecular alterations driving castration-resistant prostate cancer (CRPC) are not clearly understood. Using a novel network-based integrative approach, here, we show distinct alterations in the hexosamine biosynthetic pathway (HBP) to be critical for CRPC. Expression of HBP enzyme glucosamine-phosphate N-acetyltransferase 1 (GNPNAT1) is found to be significantly decreased in CRPC compared with localized prostate cancer (PCa). Genetic loss-of-function of GNPNAT1 in CRPC-like cells increases proliferation and aggressiveness, in vitro and in vivo. This is mediated by either activation of the PI3K-AKT pathway in cells expressing full-length androgen receptor (AR) or by specific protein 1 (SP1)-regulated expression of carbohydrate response element-binding protein (ChREBP) in cells containing AR-V7 variant. Strikingly, addition of the HBP metabolite UDP-N-acetylglucosamine (UDP-GlcNAc) to CRPC-like cells significantly decreases cell proliferation, both in-vitro and in animal studies, while also demonstrates additive efficacy when combined with enzalutamide in-vitro. These observations demonstrate the therapeutic value of targeting HBP in CRPC.

Details

Title

Inhibition of the hexosamine biosynthetic pathway promotes castration-resistant prostate cancer

Author

Kaushik, Akash K; Shojaie, Ali; Panzitt, Katrin; Sonavane, Rajni; Venghatakrishnan, Harene; Manikkam, Mohan; Zaslavsky, Alexander; Putluri, Vasanta; Vasu, Vihas T; Zhang, Yiqing; Khan, Ayesha S; Lloyd, Stacy; Szafran, Adam T; Dasgupta, Subhamoy; Bader, David A; Stossi, Fabio; Li, Hangwen; Samanta, Susmita; Cao, Xuhong; Tsouko, Efrosini; Huang, Shixia; Frigo, Daniel E; Chan, Lawrence; Edwards, Dean P; Kaipparettu, Benny A; Mitsiades, Nicholas; Weigel, Nancy L; Mancini, Michael; Mcguire, Sean E; Mehra, Rohit; Ittmann, Michael M; Chinnaiyan, Arul M; Putluri, Nagireddy; Palapattu, Ganesh S; Michailidis, George; Sreekumar, Arun

Pages

11612

Publication year

2016

Publication date

May 2016

Publisher

Nature Publishing Group

e-ISSN

20411723

Source type

Scholarly Journal

Language of publication

English

DOI

https://doi.org/10.1038/ncomms11612

ProQuest document ID

1789762760

Inhibition of the hexosamine biosynthetic pathway promotes castration-resistant prostate cancer

Jump to:

Full Text

Abstract

Details

Suggested sources