PPARδ Activation Acts Cooperatively with 3-Phosphoinositide-Dependent Protein Kinase-1 to Enhance Mammary Tumorigenesis

Peroxisome proliferator-activated receptorδ (PPARδ) is a transcription factor that is associated with metabolic gene regulation and inflammation. It has been implicated in tumor promotion and in the regulation of 3-phosphoinositide-dependent kinase-1 (PDK1). PDK1 is a key regulator of the AGC protein kinase family, which includes the proto-oncogene AKT/PKB implicated in several malignancies, including breast cancer. To assess the role of PDK1 in mammary tumorigenesis and its interaction with PPARδ, transgenic mice were generated in which PDK1 was expressed in mammary epithelium under the control of the MMTV enhancer/promoter region. Transgene expression increased pT308AKT and pS9GSK3β, but did not alter phosphorylation of mTOR, 4EBP1, ribosomal protein S6 and PKCα. The transgenic mammary gland also expressed higher levels of PPARδ and a gene expression profile resembling wild-type mice maintained on a diet containing the PPARδ agonist, GW501516. Both wild-type and transgenic mice treated with GW501516 exhibited accelerated rates of tumor formation that were more pronounced in transgenic animals. GW501516 treatment was accompanied by a distinct metabolic gene expression and metabolomic signature that was not present in untreated animals. GW501516-treated transgenic mice expressed higher levels of fatty acid and phospholipid metabolites than treated wild-type mice, suggesting the involvement of PDK1 in enhancing PPARδ-driven energy metabolism. These results reveal that PPARδ activation elicits a distinct metabolic and metabolomic profile in tumors that is in part related to PDK1 and AKT signaling.


Introduction
3-Phosphoinositide-dependant protein kinase-1 (PDK1) regulates at least 23 members of the AGC protein kinase family, including AKT, p70 ribosomal S6 kinase (S6K) and all isotypes of the protein kinase C (PKC) family [1]. PDK1 primes AGC kinases by phosphorylation of a highly conserved sequence within the Tloop or activation loop [2,3]. Although, PDK1 is constitutively active [4,5], its activity may be further enhanced by transphosphorylation on tyrosine and serine residues by other kinases. PDK1 can also regulate cell signaling in other capacities, such as serving as a nucleo-cytoplasmic shuttling protein [6], an activator of RalGDS [7], recruiting PKCh and scaffold protein CARD11 to lipid rafts [8], and as a co-activator of peroxisome proliferatoractivated receptorc (PPARc) in adipogenesis [9]. PDK1 plays an important role in mediating the effects of insulin and growth factors that regulate cell proliferation, cell size, differentiation and survival [10]. While homozygous deletion of PDK1 is embryonic lethal [10,11], hypomorphic mice with a 90% PDK1 deficiency develop normally. In contrast, PDK1 expression and gene copy number are increased in breast cancer [12,13,14] and breast cancer cell lines [15], which is consistent with down-regulation of PDK1 inhibiting cancer cell migration and metastases [16], and over-expression of PDK1 inducing transformation, drug resistance, invasion and tumorigenicity [13,17,18]. Thus, PDK1 appears to have diverse roles in both normal and malignant cells through kinase-dependent and -independent mechanisms.
PPARs belong to the nuclear receptor superfamily of liganddependent transcription factors, which control metabolic and inflammatory signaling associated with diabetes and lipodystrophies [19], but also regulate genes associated with proliferation, survival and angiogenesis in tumor cells [20,21,22]. Among the three PPAR isotypes, PPARd is distinguished by its ability to function as a promoter of tumorigenesis in many instances [21], and is highly expressed in colon cancer [23,24], head and neck cancer [25], endometrial cancer [26] and breast cancer [21]. PPARd activation promotes breast and prostate cancer cell growth [20] and a more aggressive phenotype [27]. In lung cancer cells, PPARd activation reduces PTEN to increase PDK1 and AKT activity and co-regulate their expression in concert with proliferation [28]. PPARd agonists accelerate mammary carcinogenesis [29] and promote metastatic gastric tumorigenesis [30], whereas, disruption of PPARd blocks mammary and colon tumorigenesis [31,32], although studies to the contrary have been reported [33,34]. The tumor promoting effects of PPARd may be related in part to activation of PDK1, which is a PPARd target gene in keratinocytes [35,36]. PDK1 and PPARd co-associate in DMBAinduced mammary tumors [9,29], and PPARd activates PI3K/ PDK1 signaling in a diverse range of cell types to enhance survival and growth [36,37,38]. Thus, there is evidence to suggest that PDK1 and PPARd may act cooperatively in tumorigenesis.
The in vivo consequences of PDK1 transgene expression in the mammary gland, its effect on tumorigenesis, and its influence on the tumor promoting effects of PPARd activation have not been investigated. To address these questions, transgenic mice were generated that express PDK1 in the mammary gland under the control of the mouse mammary tumor virus (MMTV) enhancer/ promoter sequence in the long terminal repeat. Here we report that PDK1 transgene expression induced PPARd and elicited a PPARd-like gene expression profile associated with glucose and lipid metabolism. Treatment with the selective PPARd agonist GW501516 markedly accelerated mammary carcinogenesis, particularly in MMTV-PDK1 mice, and the reduction in tumor latency correlated with a metabolic gene expression and metabolomic signature that differed from wild-type animals. These results suggest that the tumor promoting effects of a PPARd agonist are associated in part with PDK1 and a distinct metabolic profile related to glycolysis utilization and lipid biosynthesis.

Materials and Methods
MMTV-PDK1 mice were generated by pronuclear injection of FVB mouse embryos as previously described [39]. The N-terminal Myc-tagged human PDK1 cDNA [4] was kindly provided by Dr. Dario Alessi, University of Dundee, and cloned into the EcoR1 site in the MMTV-SV40-Bssk vector provided by Dr. William Muller, McMaster University, Hamilton, Ontario, Canada. The MMTV-PDK1 construct was digested with Sal I-Spe I, purified and used for microinjection. All animal studies were conducted under protocols 07-017 and 09-061, ''Chemopreventive agents in mammary progenitor cell targeting in carcinogen-induced breast cancer'', approved by the Georgetown University Animal Care and Use Committee in accordance with NIH guidelines for the ethical treatment of animals.

Genotyping
Primers designed to detect a fragment unique to the MMTV-PDK1 transgene were: forward: 59 CGC CGC AGC CTC GGA AGA AGC GGC, reverse: 59GGG TAC CTC ACT GCA CAG CGG CGT CC. Mice were screened for transgene expression by PCR using tail DNA.

Mammary carcinogenesis
DMBA (dimethylbenz(a)anthracene, Sigma) was dissolved in cottonseed oil at a concentration of 10 mg/ml. MMTV-PDK1 mice and wild-type littermates from founder line 192 were administered medroxyprogesterone/DMBA as previously described [40]. Briefly, 5 week-old virgin female mice were injected s.c. with 15 mg of medroxyprogesterone acetate suspension (Sicor Pharmaceuticals, Inc.), and one week later were administered 4 weekly doses of 1 mg DMBA in 0.1 ml cottonseed oil by gavage. A diet supplemented with 0.005% GW501516 was started one day after the last dose of DMBA [29]. GW501516 was provided by the Chemoprevention Branch, National Cancer Institute. Mice were euthanized by carbon dioxide inhalation when tumors reached 1-1.5 cm 3 . All animal protocols were approved by the Georgetown University Animal Care and Use Committee.

Histopathology
Tumor samples were dissected free of connective tissue and fixed in formalin. Paraffin blocks were prepared for hematoxylin & eosin (H&E) staining by the Histopathology and Tissue Shared Resource, Lombardi Comprehensive Cancer Center (LCCC), Georgetown University. Tumors were classified using the histological nomenclature recommended by Cardiff et al. [41] as adenocarcinomas, including acinar and solid lobular types, adenosquamous and squamous carcinomas and myoepithelial and undifferentiated carcinomas as previously described [40].

Gene Microarray
Mammary gland tissue was excised and preserved in RNAlater (Ambion) at 220uC until RNA was extracted using an RNeasy Mini Kit (Qiagen) according to the manufacturer's protocol. RNA purity was assessed by an A 260 /A 280 ratio of $1.9, and RNA quality was monitored using a microfluidic nanochip (Agilent). cRNA synthesis was carried out using the Affymetrix protocol with minor modifications as previously described [40]. Biotin-labeled cRNA was fragmented at 94uC for 35 min and used for hybridization overnight to an Affymetrix mouse 430A2.0 Gene-ChipH by the Genomics and Epigenomics Shared Resource, LCCC, Georgetown University. The GeneChipH was scanned using an Agilent Gene Array scanner, and grid alignment and raw data generation performed with the Affymetrix GeneChipH Operating software 1.1. A noise value (Q) based on the variance of low-intensity probe cells was used to calculate a minimum threshold for each GeneChip. Data generated after scanning was subjected to comparison analysis to select change calls at 100% increase or decrease compared with control for each gene. Gene array analysis was further refined by evaluating differences between paired samples and ranking changes by their log 2 ratio. Differences in signal ratio .log 2 2.0 and ,log 2 22.0 were ranked, and genes with a signal of ,300 in all experimental groups were excluded from analysis. Each cRNA was prepared from equal amounts of RNA pooled from 5 mice per group. Hierarchical clustering was carried out by standardizing transformed log base 2 values from the raw intensities using the equation Z~X {m s , and multiplying by a constant to adjust all values from 23 to +3. Euclidean distances, as described in were used to compute distances between genes as well as samples separately.
Finally the average linkage as D JM~N K D JK zN L D JL N M was applied to determine the gene clusters and sample clusters. All data is MIAME compliant and the raw data has been deposited in the GEO database.

Metabolomic analysis
Metabolomic analysis was performed using ultraperformance liquid chromatography electrospray ionization time-of-flight mass spectrometry (UPLC-ESI-TOFMS) [42] by the Proteomic and Metabolomics Shared Resource, LCCC, Georgetown University. Experimental groups consisted of mammary glands from each of five control wild-type and transgenic littermates and an equal number treated with GW501516 for seven days. Tissue was snap- frozen in liquid nitrogen and stored at 280uC. Extracts were prepared in 50% methanol containing internal standards using MagNAlyser Green Beads and a MagNA Lyser agitator (Roche). Samples were clarified, and processing and multivariate analysis performed as described [42]. Each sample (5 ml) was injected onto a reverse-phase 5062.1 mm ACQUITYH 1.7-mm C18 column (Waters Corp, Milford, MA) using an ACQUITYH UPLC system (Waters) with a gradient mobile phase consisting of 2% acetonitrile in water containing 0.1% formic acid (A) and 2% water in acetonitrile containing 0.1% formic acid (B). Each sample was resolved for 10 min at a flow rate of 0.5 ml/min. The gradient consisted of 100% A for 0.5 min then a ramp of curve 6 to 100% B from 0.5 min to 10 min. The column eluent was introduced directly into the mass spectrometer by electrospray. Mass spectrometry was performed on a Q-TOF PremierH (Waters) operating in either negative-ion (ESI2) or positive-ion (ESI+) electrospray ionization mode with a capillary voltage of 3200 V and a sampling cone voltage of 20 V in negative mode and 35 V in positive mode. The desolvation gas flow was set to 800 liters/h and the temperature was set to 350uC. The cone gas flow was 25 liters/hr, and the source temperature was 120uC. Accurate mass was maintained by introduction of LockSprayH interface of sulfadimethoxine (311.0814 [M+H] + or 309.0658 [M2H] 2 ) at a concentration of 250 pg/ml in 50% aqueous acetonitrile and a rate of 150 ml/min. Data were acquired in centroid mode from 50 to 850 m/z in MS scanning. Centroided and integrated mass spectrometry data from the UPLC-TOFMS were processed to generate a multivariate data matrix using MarkerLynxH (Water) that was used for analysis by SIMCA-P+11 software (Umetrics), and classified with Random Forest. Principal components analysis (PCA) and partial least-squares discrimination analysis (PLS-DA) was performed on Pareto-scaled MarkerLynx matrices to identify candidate metabolites that distinguished WT from transgenic tissue, as well as tissues from animals treated with GW501516. Metabolites were identified using the Madison Metabolomics Database Consortium, Lipidmaps and Scripps Centre for Mass Spectrometry, and negative-ion and positive-ion electrospray mode. Metabolites were verified using tandem MS by comparison to authentic compounds.

Statistical Analysis
Tumor survival data were assessed by Kaplan-Meier analysis by the log rank test, and tumor number incidence by the Mann-Whitney U test using GraphPad Prism version 4 (GraphPad Software). Histological analysis of tumors was performed by Fishers' Exact test. Differences were considered to be significant at P,0.05.

MMTV-PDK1 transgenic mice
MMTV-PDK1 transgenic mice were screened for transgene expression by PCR of tail DNA, and four founder lines were identified ( Figure 1A). Founder line 192 expressed 8-10-fold higher pS241PDK1 and PDK1 levels vs. wild-type littermates, compared to a 1.5-2-fold change in other founder lines (results not shown), and thus, founder 192 was used for all subsequent studies. Shown is the relative expression of metabolic genes common to GW510516-treated (GW) wild-type mice (WT), MMTV-PDK1 transgenic mice (PDK1) and GW501516treated MMTV-PDK1 mice (PDK1+GW) vs. WT. The heatmap groups refer to Figure S2A, and the gene expression profile of all groups is presented in Table S1. doi:10.1371/journal.pone.0016215.t002 The mammary gland of nulliparous transgenic mice exhibited normal glandular structure and ductal elongation and branching at 3 and 12 weeks of age (Supporting information, Figure S1A). Lactating transgenic mice exhibited strong PDK1 expression (Supporting information, Figure S1B), but no delay in involution vs. wild-type littermates (Supporting information, Figure S1C). There were no differences between transgenic and wild-type mice in their hyperplastic response to medroxyprogesterone stimulation, and transgenic mice did not present with mammary tumors over their lifespan (results not shown).

MMTV-PDK1 mice express increased levels of pT308AKT, pS9 GSK3b and PPARd
IHC analysis of the mammary gland of 7 week-old pups indicated increased pS241PDK1, pT308AKT and PPARd expression vs. wild-type littermates ( Figure 1B). Western analysis revealed that increased PDK1 transgene expression was associated with a 3-4-fold increase in pT308AKT and a 4-5.5-fold increase in PPARd ( Figure 1C), whereas, pS473AKT remained unaltered ( Figure 1C), as previously reported in PDK1 null cells [44]. The levels of pS9GSK3b were increased 2.5-3.5-fold in transgenic mice, but no changes were noted in PTEN, pmTOR, p4EBP1, pS6 and pS657PKCa ( Figure 1C). qRT-PCR analysis indicated that PDK1 and PPARd mRNA levels were increased 3-and 1.4fold, respectively (results not shown) in comparison to the large increases in protein expression, suggesting that their co-regulation is in part due to post-translational regulation.

MMTV-PDK1 mice are sensitized to PPARd agonist GW501516
Since MMTV-PDK1 mice did not present with mammary tumors over their lifespan, they were tested for their susceptibility to mammary carcinogenesis induced by progestin stimulation and DMBA [40] (Figure 2A, B). MMTV-PDK1 mice did not exhibit a statistically significant change in tumor latency, where median tumor-free survival was 89 days vs. 110 days in wild-type mice ( Figure 2A). However, maintaining animals on a diet containing the selective and potent PPARd agonist, GW501516, immediately following DMBA, resulted in a dramatic acceleration of tumor formation (Figure 2A). The tumor promoting effect of GW501516 was more pronounced in MMTV-PDK1 mice, where GW501516 treatment reduced the median tumor-free survival from 89 days to 21 days. This represented a two-fold greater reduction in survival than observed in wild-type mice treated with GW501516, where the median tumor-free survival was reduced from 110 days to 54 days ( Figure 2B). Tumor multiplicity was similar in both groups (1.88 and 1.40, respectively) and was not altered by GW501516 treatment (Table 1). Although there were no significant differences in tumor histopathology between PDK1 and wild-type mice, GW501516 treatment produced a significant increase in the percentage of adenosquamous and squamous cell carcinomas in both groups, which correlated with rapid tumor development (Table 1).

Adenocarcinomas from MMTV-PDK1 mice express increased pT308AKT and PPARd
HC analysis indicated that adenocarcinomas induced in MMTV-PDK1 mice expressed elevated levels of PDK1, pT308AKT and PPARd in comparison to histologically matched tumors from wild-type mice ( Figure 2C). Western analysis confirmed that this phenotype was present in adenocarcinomas, but not in squamous cell carcinomas from transgenic mice, as might be expected from the mammary epithelium selectivity of the MMTV promoter. However, no changes were evident in other putative PDK1 downstream targets in either tumor histotype ( Figure 2D).

PPARd activation increases PDK1 and pT308AKT
To evaluate the effect of PPARd activation on PDK1 signaling, mammary tissue was analyzed from wild-type and transgenic animals maintained on the GW501516 diet ( Figure 3A). GW501516-treated wild-type mice expressed increased levels of PDK1, pT308AKT and pS9GSK3b, whereas, transgenic mice exhibited increased pT308AKT, but no additional changes in PDK1 and pS9GSK3b, a result that we attribute to their already elevated levels in the absence of GW501516 treatment. IHC analysis of wild-type mice maintained on the GW501516 diet confirmed increased levels of PDK1 and pT308AKT and PPARd, and b-catenin although elevated, was primarily membraneassociated and not nuclear ( Figure 3B).
The causal association between PDK1 over-expression and PPARd was also investigated in mouse mammary epithelial cell line Comma-1D stably expressing PDK1 [43], which expresses high bcatenin/TCF transcriptional activity [15]. Comma-1D/PDK1 cells, but not control cells exhibited increased levels of PPARd ( Figure 3C), and reporter gene assays in the presence and absence of GW501516, as well as in the presence of dominant-negative PPARd, confirmed that the receptor was transcriptionally active ( Figure 3D). Similar studies in HCT116 cells, which also express PPARd and high b-catenin/TCF transcriptional activity confirmed the dependence of both pathways on PDK1 (results not shown).

MMTV-PDK1 mice express a gene and metabolomic profile reflecting PPARd activation
To further characterize the phenotype of MMTV-PDK1 mice and the effect of GW501516 treatment, gene expression profiling was carried out with mammary tissue from transgenic and wild-type mice maintained on the GW501516 diet for 1 week (Supporting information, Figure S2, Table  S1, Table S2). Hierarchical cluster analysis revealed changes in gene expression that were specifically associated with GW501516 treatment of both wild-type and transgenic mice that were related to lipid (Acaca, Acly, Elovl6, Acss2) and glucose (Acly, PDK4, Slc2a5) metabolism (Table 2, Group A, Supporting information Figure S2A, Table S2), as well as non-metabolic functions (Asb5, Hrc, Mmd2, Smpx, Tcap, Trdn) ( Table 2, Group A9, Supporting information Figure  S2A, Table S2). There was also an additional gene signature common to MMTV-PDK1 and GW501516-treated wild-type mice that was not associated with increased tumori genesis (Cox7a1, Cpt1b, Elovl3, Fabp3, E2f5, Slcf5, Ucp1) ( Table 2, Group B, Supporting information Figure S2A, Table  S2).
Metabolomic analysis of the mammary gland revealed similarities and differences between the metabolome of GW501516treated wild-type and transgenic animals, as well as between control wild-type and transgenic mice. (Figure 4, Table 3, Supporting information Figure S4, Table S3, Table S4, Table  S5). The suggestion of increased lipid biosynthesis based on gene array data in GW501516-treated animals was consistent with increased fatty acid and phospholipid levels, and correlated with increased tumorigenesis (Table 3), despite the differences in specific metabolites between the two groups. In contrast, untreated transgenic mice exhibited a reduction in lipid metabolites vs. wildtype mice.

Discussion
The present study has examined the effect of PDK1 transgene expression on mammary carcinogenesis, and how it impacts the tumor promoting effects of PPARd activation. Although PDK1 transgenic mice did not exhibit changes in mammary gland development, function and tumorigenicity, they were markedly sensitive to GW501516 treatment, where median tumor-free survival was reduced four-fold in comparison to a two-fold reduction in GW501516-treated wild-type mice. This striking effect correlated with an increase in a specific metabolic gene signature indicative of glycolysis and fatty acid biosynthesis that was not present in either control wild-type or transgenic mice ( Figure 5, Supporting information Figure S3). It is well-known that human cancers exhibit a near ubiquitous expression of metabolic genes [45] that is widely regarded to support high rates of proliferation [46]. This phenotype is consistent with the acceleration of mammary tumorigenesis by GW501516 [29], the increase in fatty acid biosynthesis by GW501516 in muscle cells [47], and our analysis of the mammary gland metabolome in GW501516treated animals. Some of these changes may be related to increased AKT activation in GW501516-treated animals, since it is an important regulator of glucose and lipid metabolism in both normal and malignant cells, including breast cancer [48,49]. AKT phosphorylates and activates ATP:citrate lyase (Acly) to promote tumor growth [50], and loss of Acly counteracts AKT-driven tumorigenesis [51]. In addition, the PPARd target gene, Pdk4, [47,52] reduces the flux of pyruvate into the tricarboxylic acid cycle, and Acss2, which increases the flux of lactate to acetylCoA, were increased in GW501516-treated animals (Table 2, Figure 5). These changes are consistent with the greater levels of fatty acid and phospholipid metabolites in GW501516-treated transgenic mice in comparison to treated wild-type mice (Table 3), suggesting their involvement in enhanced tumorigenesis. This association is also in agreement with the increase in serum lysophospholipids in women with high grade ovarian cancer [53].
The lack of tumorigenicity of PDK1 transgene expression is in agreement with previous studies that found PDK1 over-expression per se was not oncogenic unless expressed in a heterozygous PTEN background [54] or together with a growth factor receptor with oncogenic potential, such as erbB2 [12]. The lack of change in pmTOR, p4EBP1, pS6, pPKCa ( Figure 1C) and pRSK (results not shown) in mammary tissue from PDK1 transgenic mice is also consistent with the lack of change in S6K and RSK activation following treatment of PDK1 hypomorphic mice with insulin [10]. Thus, low residual levels of PDK1 appear to be sufficient for mammary gland development and function and for downstream signaling.
One seminal finding in our study was the increase in PPARd expression in the transgenic mammary gland. PPARd expression is induced by K-Ras via ERK activation [55], and although PDK1 has been reported to regulate MEK1/2 activation [56], ERK activation remained unchanged in MMTV-PDK1 mice (results not shown). Another possible mechanism is that increased pS9GSK3b by AKT inhibits the ability of GSK3b to phosphorylate and destabilize b-catenin by proteasomal degradation, and thus results in enhanced transcription of TCF target genes, such as PPARd [57]. However, no evidence of nuclear b-catenin accumulation was noted in GW501516-treated wild-type animals despite increased pT308AKT expression (Fig. 3C), suggesting other signaling mechanisms. The changes in PPARd observed in transgenic mice may also have resulted from post-translational stabilization against Figure 5. Schematic of the regulation of glucose and fatty acid metabolism by PDK1 and PPARd. One mechanism depicted is that PPARd upregulates PDK1 expression and activates AKT to inhibit GSK3b and increase b-catenin/TCF-dependent transcription; however, it is controversial whether PPARd is a TCF target gene. Activation of PPARd by GW501516 in wild-type and MMTV-PDK1 mice upregulates the expression of genes (highlighted in yellow) associated with glucose transport (Slc2a5), acetylCoA formation through acetylCoA (Acss32, Pdk4), and fatty acid biosynthesis (Acaca, Elovl6). Genes associated with fatty acid transport and transcriptional regulation (Fabp3), formation of acetylCoA from citrate (Acly), fatty acid transport (Cpt1b), CoA fatty acid esters (Slc27a2) and increased oxidative phosphorylation (Cox7a1 and Ucp1) are increased by GW501516 in wildtype mice and in untreated PDK1 transgenic mice, but do not correlate with GW501516-induced tumorigenesis. HDAC, histone deacetylase; FA, fatty acid; PDH, pyruvate dehydrogenase; inhibition. doi:10.1371/journal.pone.0016215.g005 proteasomal degradation when ligand-bound [58]. This view is supported by our results in PDK1-transduced mouse mammary epithelial cells, which expressed increased PPARd and PPARddependent reporter gene activity ( Figure 3C). It is equally plausible that endogenous PPARd ligands generated by fatty acid metabolism serve as PPARd ligands [59] to increase expression posttranslationally. Lastly, the co-association of PDK1 and PPARd noted in mammary tumors [29] may also have enhanced resistance of the receptor to ubiquitination and degradation.
In addition to increased expression of PPARd in MMTV-PDK1 mice, PDK1 levels were increased by PPARd agonist GW501516 ( Figure 3B). The mouse PDK1 locus contains a PPRE in an upstream enhancer region [36], and deletion of PPARd resulted in a 50% reduction in PDK1 mRNA and .80% decrease in PDK1 protein expression [35]. This could result in a feed forward mechanism resulting from the effect of PDK1 and PPARd on each others expression, and could explain their synergism in tumorigenesis.
Many of the genes whose expression increased more than 3-fold in MMTV-PDK1 mice were associated with muscle architecture or motor function, e.g. myosin, nebulin, troponin 1, tropomyosin and titin. A recent study profiling gene expression in the mammary gland side population also noted a muscle-specific expression pattern [60]. Although the function of these genes in non-myogenic cells is unknown, troponin 1 has been found to bind ERRa, enhance its transcriptional activity [61] and regulate the metabolic switch to oxidative phosphorylation [62]. Thus, the expression of these genes may be a factor in the increase in fatty acid transport and oxidation in muscle cells treated with GW501516 [47].
In summary, PDK1 expression in the mammary gland was not oncogenic, but accelerated tumor formation in conjunction with a PPARd agonist. GW501516-enhanced tumorigenesis was associated with a distinct gene and metabolomic signature related to glycolysis and fatty acid biosynthesis. These results suggest that PDK1 and PPARd may drive tumorigenesis by enhancing energy metabolism. Figure S1 (A) Whole mounts of the mammary gland at 3 and 12 weeks of age in founder 192. Upper panel, Magnification 56; lower panel, Magnification 206. (B) Response of PDK1 transgenic mice to lactation and involution. Western blot of PDK1 expression in non-lactating transgenic mice (+(1) and +(2)) and in lactating mice at day 1 (D1) and day 10 (D10) following forced involution by teat sealing. (C) Lactation and involution in transgenic mice. H&E stained sections were prepared on day 1 (D1), day 3 (D3), day 7 (D7) and day 10 (D10) of lactating wild-type (2) and transgenic (+) mice following forced involution. (DOC) Figure S2 Gene expression profiling of the mammary gland from MMTV-PDK1 and wild-type mice before and after treatment with GW501516. (A) Gene expression in wild-type (WT) and MMTV-PDK1 (PDK1) mice with and without GW501516 treatment. Untreated MMTV-PDK1 mice expressed a phenotype indicative of wild-type mice treated with GW501516.

Supporting Information
A list of gene expression changes is included in Table S1. (B) qRT-PCR and gene microarray analysis. Shown are the -fold changes in mammary gene expression between MMTV-PDK1 mice (PDK1), PDK1 mice treated with GW510516 (PDK1+GW), and wild-type mice treated with GW501516 (WT+GW) relative to untreated WT mice. Each experimental group is based on pooled samples from five mice. (DOC)  Table 2, genes showing a 3-fold or greater changes are listed in Table S1, and gene ontology is listed in Table  S2. (DOC)