The autophagic tumor stroma model of cancer

Abstract

A loss of stromal caveolin-1 (Cav-1) in the tumor fibroblast compartment is associated with early tumor recurrence, lymphnode metastasis and tamoxifen-resistance, resulting in poor clinical outcome in breast cancer patients. Here, we have used Cav-1 (−/−) null mice as a pre-clinical model for this “lethal tumor micro-environment”. Metabolic profiling of Cav-1 (−/−) mammary fat pads revealed the upregulation of numerous metabolites (nearly 100), indicative of a major catabolic phenotype. Our results are consistent with the induction of oxidative stress, mitochondrial dysfunction and autophagy/mitophagy. The two most prominent metabolites that emerged from this analysis were ADMA (asymmetric dimethyl arginine) and BHB (beta-hydroxybutyrate; a ketone body), which are markers of oxidative stress and mitochondrial dysfunction, respectively. Transcriptional profiling of Cav-1 (−/−) stromal cells and human tumor stroma from breast cancer patients directly supported an association with oxidative stress, mitochondrial dysfunction and autophagy/mitophagy, as well as ADMA and ketone production. MicroRNA profiling of Cav-1 (−/−) stromal cells revealed the upregulation of two key cancer-related miR's, namely miR-31 and miR-34c. Consistent with our metabolic findings, these miR's are associated with oxidative stress (miR-34c) or activation of the hypoxic response/HIF1α (miR-31), which is sufficient to drive authophagy/mitophagy. Thus, via an unbiased comprehensive analysis of a lethal tumor micro-environment, we have identified a number of candidate biomarkers (ADMA, ketones and miR-31/34c) that could be used to identify high-risk cancer patients at diagnosis, for treatment stratification and/or for evaluating therapeutic efficacy during anti-cancer therapy. We propose that the levels of these key biomarkers (ADMA, ketones/BHB, miR-31 and miR-34c) could be (1) assayed using serum or plasma from cancer patients or (2) performed directly on excised tumor tissue. Importantly, induction of oxidative stress and autophagy/mitophagy in the tumor stromal compartment provides a means by which epithelial cancer cells can directly “feed off” of stromal-derived essential nutrients, chemical building blocks (amino acids, nucleotides) and energy-rich metabolites (glutamine, pyruvate, ketones/BHB), driving tumor progression and metastasis. Essentially, aggressive cancer cells are “eating” the cancer-associated fibroblasts via autophagy/mitophagy in the tumor micro-environment. Lastly, we discuss that this “autophagic tumor stroma model of cancer metabolism” provides a viable solution to the “autophagy paradox” in cancer etiology and chemotherapy.

Introduction

It is now well-recognized that the tumor micro-environment plays a primary role in determining tumor progression and metastasis in many different types of epithelial cancers.^1–4 In this regard, “activated or myofibroblastic” cancer-associated fibroblasts^5,6 have emerged as one of the most prominent cell types in the tumor stroma that may determine clinical outcome in breast and prostate cancers.

We have recently identified a loss of stromal Cav-1 in the tumor associated fibroblast compartment as a critical event in the progression of human breast cancers.^7–9 More specifically, a loss of stromal Cav-1 is associated with early tumor recurrence, lymphnode metastasis and tamoxifen-resistance, resulting in poor clinical outcome.⁹ Similar results were obtained with DCIS¹⁰ and prostate cancer patients,¹¹ indicating that a loss of stromal Cav-1 in cancer-associated fibroblasts is tightly associated with tumor progression and metastasis. These findings have now been replicated in several independent patient cohorts,^12,13 and also extended to other more lethal forms of breast cancer, such as the triple-negative and basal-like breast cancer sub-types.¹⁴ For example, in triple-negative breast cancers, patients with high stromal Cav-1 have a 75.5% survival rate at 12 years, while patients with an absence of stromal Cav-1 have a survival rate of less than 10% at 5 years post-diagnosis.¹⁴ Thus, it is now imperative that we dissect the molecular basis of this phenomenon, to design better therapeutics targeting this high-risk patient population.

To mechanistically understand the lethality of a Cav-1-deficient stromal micro-environment, we next chose to use Cav-1 (−/−) null mice as a model system.¹⁵ Using this approach, we isolated bone-marrow stromal cells (thought to be the precursors of cancer-associated fibroblasts⁶) and subjected them to unbiased proteomics and genome-wide transcriptional profiling.¹⁵ As a result, via our proteomics analysis, we observed that Cav-1 (−/−) null stromal cells show the upregulation of eight myo-fibroblast markers (including vimentin, calponin and collagen I), eight glycolytic enzymes [such as pyruvate kinase (PKM2) and lactate dehydrogenase (LDHA)] and two markers of oxidative stress (namely catalase and peroxiredoxin 1).¹⁵ These results were also independently supported by our data from transcriptional profiling. An informatics analysis of these findings suggested that a loss of stromal Cav-1 results in oxidative stress, driving aerobic glycolysis (a.k.a., the Warburg effect) in cancer-associated fibroblasts.^16,17 This would then provide a feed-forward mechanism by which Cav-1-deficient cancer-associated fibroblasts could literally “feed” adjacent cancer cells by providing lactate and pyruvate in a paracrine fashion.¹⁵ We have termed this novel idea the “reverse Warburg effect,” as the “classical” Warburg effect was originally thought to occur only in epithelial cancer cells and not in cancer-associated fibroblasts.¹⁵

Here, to better understand the nature and extent of these proposed metabolic changes, we have now performed an unbiased metabolomics analysis on the mammary fad pads derived from Cav-1 (−/−) null mice. With this approach, we expected to simply validate the existence of the “reverse Warburg effect”. However, what we observed was far more complex and extensive, and was characteristic of a profound catabolic phenotype. Our results are consistent with oxidative stress, mitochondrial dysfunction and autophagy/mitophagy—which would also induce aerobic glycolysis in the tumor stroma (the “reverse Warburg effect”).

Interestingly, autophagy, mitophagy and aerobic glycolysis are all induced by oxidative stress and are all controlled by the same key transcription factor, namely hypoxia-inducible factor-1alpha (HIF-1α). In this regard, we directly show that a loss of stromal Cav-1 leads to the upreglulation of miR-31, which is a known activator of HIF1α transcriptional activity.¹⁸ Thus, the lethality of a Cav-1-deficient tumor micro-environment could be explained by an autophagic/catabolic tumor stroma, which would then provide both nutrients and energy to epithelial cancer cells in a paracrine fashion. We propose to call this idea the “autophagic tumor stroma model of cancer.” This represents a unique therapeutic opportunity, as blocking autophagy in the tumor stroma should halt cancer growth, while an induction of autophagy in the epithelial cancer cells should have the same effect, thereby halting tumor growth. This new model of compartmentalized autophagy could clarify and explain the controversial role of autophagy in tumor pathogenesis and facilitate the design of novel anti-cancer therapies.

Results

Metabolomic analysis of cav-1 (−/−) null tissues: evidence for oxidative stress, mitochondrial dysfunction and autophagy.

Mammary fat pads were harvested from age-matched female WT and Cav-1 (−/−) null mice (n = 6 for each genotype) and subjected to an unbiased metabolomic analysis. Over 200 known compounds were identified by mass spectrometry analysis and their levels were quantitated. Interestingly, a large number of compounds were significantly changed in Cav-1 (−/−) mammary fat pads (n = 103; 92 UP; 11 DOWN), consistent with a severe metabolic phenotype (Table 1).

Table 1.

Metabolomic analysis of mammary fat pads from Cav-1 (−/−) deficient mice

Amino Acids:

Alanine and aspartate metabolism: alanine (1.7); asparagine (2.3); aspartate (2.7)

Cysteine, methionine, SAM, taurine metabolism: cysteine (1.6); hypotaurine (1.8); methionine (2.2); N-acetylmethionine (2.7); S-adenosylhomocysteine (SAH) (0.7)

Glutamate metabolism: glutamate (1.9); glutamine (1.6); N-acetyl-aspartyl-glutamate (NAAG) (1.6)

Glutathione metabolism: 5-oxoproline (1.5); cysteine-glutathione disulfide (1.4); glutathione, oxidized (GSSG) (1.5); glutathione, reduced (GSH) (1.7)

Glycine, serine and threonine metabolism: glycine (2.3); serine (2.5); threonine (2.0)

Histidine metabolism:histamine (2.5); histidine (2.4); urocanate (3.0)

Lysine metabolism: lysine (1.7); pipecolate (1.9)

Phenylalanine & tyrosine metabolism: phenylalanine (2.3); tyrosine (2.5)

Tryptophan metabolism:5-hydroxyindoleacetate (2.7); C-glycosyl-tryptophan (1.7); tryptophan (2.3)

Urea cycle; arginine-, proline-, metabolism: arginine (1.9); assymetric dimethylarginine (ADMA) (3.3); proline (2.3); trans-4-hydroxyproline (2.0); urea (2.5)

Valine, leucine and isoleucine metabolism:isoleucine (2.5); leucine (2.3); valine (2.0)

Peptides:

Dipeptide:glycylproline (2.8); proline-hydroxy-proline (1.8)

Gamma-glutamyl: gamma-glutamylglutamate (1.7); gamma-glutamylisoleucine (2.0); gamma-glutamylleucine (1.9); gamma-glutamylphenylalanine (1.8); gamma-glutamyltryptophan (2.1); gamma-glutamyltyrosine (1.9); gamma-glutamylvaline (1.7)

Carbohydrates:

Aminosugars metabolism: N-acetylglucosamine 6-phosphate (1.6)

Fructose, mannose, galactose, starch and sucrose metabolism: erythrose (2.0); fructose (2.0); mannose-6-phosphate (1.8)

Glycolysis, gluconeogenesis, pyruvate metabolism: fructose-6-phosphate (1.9); glucose (1.5); glucose-6-phosphate (G6p) (1.9); pyruvate (1.4)*

Nucleotide sugars, pentose metabolism: ribose (1.9); sedoheptulose-7-phosphate (1.7)

Sugar alcohol: myo-inositol (1.4)

Cofactors and vitamins:

Ascorbate and aldarate metabolism:ascorbate (Vitamin C) (11.2); threonate (1.4)

Nicotinate and nicotinamide metabolism: nicotinamide (1.3)

Pantothenate and CoA metabolism: pantothenate (1.4)

Riboflavin metabolism:riboflavin (Vitamin B2) (2.6)

Tocopherol metabolism:alpha-tocopherol (2.7)

Vitamin B6 metabolism: pyridoxate (1.8)

Energy:

Krebs cycle: fumarate (1.6); malate (1.4)

Oxidative phosphorylation: methylphosphate (1.6); phosphate (1.3)

Lipids:

Carnitine metabolism: 3-dehydrocarnitine (1.3); carnitine (0.9); malonylcarnitine (0.6)

Essential fatty acid: dihomo-linolenate (20:3n3 or n6) (1.5)

Fatty acid, ester: n-Butyl oleate (1.5)

Glycerolipid metabolism: choline (1.3); ethanolamine (2.1) glycerol (1.2); phosphoethanolamine (2.4)

Inositol metabolism: inositol 1-phosphate (I1p) (2.4)

Ketone bodies:3-hydroxybutyrate (BHBA) (4.3)

Long chain fatty acids: arachidonate (20:4n6) (1.5); margarate (17:0) (1.4); myristoleate (14:1n5) (1.8)

Monoacylglycerol: 1-stearoylglycerol (1-monostearin) (1.4)

Sterol/Steroid: cholesterol (1.6)

Nucleotides:

Purine metabolism, (hypo)xanthine/inosine containing: hypoxanthine (1.6); inosine (0.7); xanthine (1.7); xanthosine (1.3)

Purine metabolism, adenine containing: adenosine 2′-monophosphate (2′-AMp) (2.4); N1-methyladenosine (2.0)

Purine metabolism, guanine containing: guanosine (0.4)

Purine metabolism, urate metabolism: urate (0.7)

Pyrimidine metabolism, cytidine containing: 2′-deoxycytidine (2.2); cytidine (2.5); cytidine 5′-monophosphate (5′-CMp) (0.8); cytidine-3′-monophosphate (3′-CMp) (1.9)

Pyrimidine metabolism, uracil containing: pseudouridine (1.7); uracil (3.8)

Xenobiotics:

Benzoate metabolism: 4-ethylphenylsulfate (2.1); catechol sulfate (2.5); hippurate (1.8)

Fold-changes (KO/Wt) are shown in parentheses for all metabolites that showed a significant change (p ≤ 0.05). In one case, an asterisk (*) indicates p ≤ 0.1. Metabolites showing an increase of 2.5 or greater are underlined. All other p values were p≤0.05.

Several observations are consistent with the presence of oxidative stress. These include: (1) an increase in the amounts of several anti-oxidants, such as ascorbic acid (∼11.2-fold), vitamin E (alpha-tocopherol; 2.7-fold), 5-hydroxyindoleacetate (2.7-fold) and hypotaurine (1.8-fold); (2) an increase in the number of amino-acid metabolites associated with the glutathione pathway, more specifically γ-glutamyl amino acids and glutathione species (GSH, GSSH, 5-oxoproline, cys-glutathione-disulfide); (3) a shift towards gluconeogenesis, and the pentose phosphate pathway, which is known to produce increased amounts of NADPH, which can then be used as reducing equivalents to maintain reduced glutathione; (4) the observed increase in ribose and nucleotides, which emanate from the pentose phosphate pathway; and (5) an increase in the amount of ADMA (asymmetric dimethyl arginine; 3.3-fold), which is both a marker of protein catabolism and oxidative stress, and can also produce more oxidative stress. ADMA acts as an eNOS uncoupler, resulting in the production of superoxide, instead of nitric oxide.¹⁹ Similarly, ADMA is also a marker of chronic hypoxia²⁰ and mitochondrial dysfunction.²¹

As oxidative stress also drives mitochondrial dysfunction, autophagy and mitophagy, we also looked for evidence of these catabolic biological processes in our metabolic data set. Consistent with a generalized catabolic phenotype, we see: (1) higher levels of numerous amino acids and their catabolites; (2) elevation of 4 markers of protein or collagen degradation (assymetric dimethylarginine, trans-4-hydroxyproline, glycylproline, proline-hydroxy-proline); (3) elevated levels of a marker of increased RNA turnover, namely pseudouridine (1.7-fold); (4) increased levels (4.3-fold) of 3-hydroxy butyrate (BHB), a ketone body, which is a well-accepted marker of mitochondrial dysfunction;^22,23 and (5) higher levels of free cholesterol (1.6-fold), which can also contribute to mitochondrial dysfunction.²⁴ A decrease in mitochondrial function is also consistent with the accumulation of certain metabolites associated with glycolysis (pyruvate; 1.4-fold), and the TCA cycle (fumarate and malate; both >1.4-fold). Interestingly, we also see increases in 5-hydroxyindole (2.7-fold), which is an anti-oxidant metabolite of tryptophan, which protects against oxidative damage and mitochondrial dysfunction, as it suppresses ROS generation, lipid peroxidation, peroxynitrite generation and glutathione depletion—thereby increasing mitochondrial membrane potential.²⁵

Finally, the increased amount of ascorbic acid (Vitamin C; 11.2-fold) may also reflect that glucose is not being efficiently metabolized via glycolysis and the TCA cycle in Cav-1 (−/−) mammary fat pads. Thus, glucose instead may be shunted towards the synthesis of ascorbic acid (an antioxidant), which is generated directly from glucose by a sequence of enzyme-driven steps and employs the essential enzyme L-gulonolactone oxidase (GULO).

We also independently compared our results from the mammary fat pad with lung tissue, as adipose tissue and lung tissue express the highest levels of Cav-1. Only concordant changes were selected and are shown in Table 2. Interestingly, ADMA, pyruvate and 3-hydroxybutyrate were significantly elevated in lung tissue, consistent with the idea that Cav-1 (−/−) null tissues are undergoing (1) oxidative stress and (2) mitochondrial dysfunction. Box plots for ADMA and BHB are shown in Figure 1, and for the antioxidant Vitamins C and E in Figure 2.

Table 2.

Metabolomic analysis of mammary fat pads and lung tissue from Cav-1 (−/−) deficient mice

Metabolites	Mammary (KO/WT)	Lung (KO/WT)
pipecolate	1.9	1.3
assymetric dimethylarginine (ADMA)	3.3	1.4
glycylproline	2.8	1.4
pyruvate	1.4*	1.9
carnitine	0.9	0.9
dihomo-linolenate (20:3n3 or n6)	1.5	1.3
3-hydroxybutyrate (BHBA)	4.3	3.5
inosine	0.7	0.7
cytidine-3′-monophosphate (3′-CMP)	1.9	1.3
4-ethylphenylsulfate	2.1	2.2

Only metabolites and fold-changes (KO/WT)showing concordant changes in both the mammary fat pad and lung tissue are shown. An asterisk (*) indicates p ≤ 0.1. All other p values were p ≤ 0.05.

Figure 1.

Evidence for oxidative stress and mitochondrial dysfunction in Cav-1 (−/−) null mouse tissues: ADMA and Ketones. Note that both 3-hydroxybutyrate (BHBA) and asymmetric dimethyl arginine (ADMA) are increased ∼3–4 fold in Cav-1 (−/−) mammary fat pads. Similar results were obtained with lung tissue harvested from Cav-1 (−/−) mice. Importantly, ADMA is a marker of endothelial dysfunction and oxidative stress; it can also drive oxidative stress, as it functions as an uncoupler of NoS family member, inhibiting the production No and producing superoxide instead. In addition, BHBA is a ketone body known to be a marker of mitochondrial dysfunction. oxidative stress induces mitochrondrial dysfunction and visa versa, driving autophagy and mitophagy.

Figure 2.

Upregulation of antioxidants in Cav-1 (−/−) mammary fat pads. One compensatory response to oxidative stress is the over-production of anti-oxidants. Note that in Cav-1 (−/−) mammary fat pads an 11-fold increase in Vitamin C (ascorbic acid) and a near three-fold increase in Vitamin E (alpha-tocopherol) were observed.

It is also known that oxidative stress is indeed sufficient to induce ketone production in an animal model of Amyotrophic Lateral Sclerosis (ALS). These mice express a mutant form of SOD1 (G86R) and show progressively increased serum levels of ketone bodies.²⁶ Furthermore, ALS patients show increased serum levels of ketone bodies, both 3-hydroxybutyrate and acetone, as documented by NMR spectroscopy.²⁷ Finally, autophagy has also been implicated in the pathogenesis of ALS, both using transgenic SOD1-mutant mouse models and human patient samples.^28,29 Thus, oxidative stress, mitochondrial dysfunction and autophagy/mitophagy are all clustered together in various neurodegenerative disorders, such as ALS and Alzheimer disease.

Other noteworthy metabolites that were increased include histamine (2.5-fold) and arachidonic acid (1.5-fold), which may directly or indirectly contribute towards an inflammatory micro-environment. As arachidonic acid is the precursor of both prostaglandins and leukotrienes, increased free arachidonic acid could drive the generation of increased inflammatory mediators. Histamine also increases the differentiation of stromal cells towards a more myo-fibroblastic phenotype,^30,31 consistent with the behavior of cancer-associated fibroblasts.

Transcriptional mRNA profiling of Cav-1 (−/−) stromal cells provides validating evidence for a stromal catabolic state.

We have previously traced the lethality of a Cav-1 negative tumor micro-environment to the stromal fibroblast or cancer-associated fibroblast compartment.^8,9 Thus, to garner validating evidence for our metabolic profiling studies, we re-interrogated our transcriptional profiling data obtained via the analysis of WT and Cav-1 (−/−) stromal cells.¹⁵

We reasoned that many of the metabolic features we have observed could be explained by oxidative stress induced autophagy and mitophagy. In direct support of this notion, Table 3 shows that many of the genes that are involved in mediating autophagy and mitophagy^32,33 are indeed upregulated in Cav-1 (−/−) null stromal cells.

Table 3.

Upregulation of autophagy/mitophagy related gene transcripts in Cav-1 (−/−) stromal cells

Gene	Description	Fold change (KO/WT)	Accession number	p-value
Known Genes Associated with Autophagy
1200002N14Rik	RIKEN cDNA 1200002N14 gene	1.24	NM_027878	0.01
Ambra1	autophagy/beclin 1 regulator 1	1.93	NM_172669	0.04
Atg16l1	autophagy-related 16-like 1 (yeast)	1.33	NM_029846	0.01
Atg2a	ATG2 autophagy related 2 homolog A (S. cerevisiae)	1.41	NM_194348	0.02
Atg2b	ATG2 autophagy related 2 homolog B (S. cerevisiae)	1.82	NM_029654	0.03
Atg4a	autophagy-related 4A (yeast)	1.33	NM_174875	0.02
Atg4d	autophagy-related 4D (yeast)	1.57	NM_153583	0.05
Atg9a	autophagy-related 9A (yeast)	1.73	NM_001003917	0.005
Becn1	beclin 1, autophagy related	1.28	NM_019584	0.009
Fam176a	family with sequence similarity 176, member A	1.34	NM_145570	0.02
Foxo3	forkhead box O3	2.25	NM_019740	0.005
Frap1	FK506 binding protein 12-rapamycin associated protein 1	1.83	NM_020009	0.04
Il3	interleukin 3	1.42	NM_010556	0.03
Irgm1	immunity-related GTPase family M member 1	2.04	NM_008326	0.05
Irgm2	immunity-related GTPase family M member 2	2.84	-	0.01
Map1lc3a	microtubule-associated protein 1 light chain 3alpha	2.18	NM_025735	0.01
Npc1	Niemann Pick type C1	1.88	NM_008720	0.04
Rab24	RAB24, member RAS oncogene family	2.03	NM_009000	0.02
Tsc1	tuberous sclerosis 1	1.44	NM_022887	0.04
Tsc2	tuberous sclerosis 2	1.52	NM_011647	0.02
Wipi1	WD repeat domain, phosphoinositide interacting 1	1.26	NM_145940	0.05
Known Genes Associated with Mitophagy
Bnip2	BCL2/adenovirus E1B interacting protein 2	1.51	NM_016787	0.01
Bnip3	BCL2/adenovirus E1B interacting protein 3	1.75	NM_009760	0.02
Bnipl	BCL2/adenovirus E1B 19 kD interacting protein like	1.40	NM_134253	0.01
Park7	Parkinson disease (autosomal recessive, early onset) 7	1.97	NM_020569	0.02
Pacrgl	PARK2 co-regulated-like	1.31	NM_025755	0.04
Genes Associated with Mitophagy in a New Yeast Genetic Screen
Mon1b	MON1 homolog b (yeast)	1.96	NM_173015	0.03
Mon2	MON2 homolog (yeast)	2.10	NM_153395	0.02
Vps41	vacuolar protein sorting 41 (yeast)	1.61	NM_172120	0.003
Dnm1	dynamin 1	1.60	NM_010065	0.007
Mak	male germ cell-associated kinase	1.49	NM_001145803	0.004
Rpl13a	ribosomal protein L13a	1.71	-	0.03
Abcc2	ATP-binding cassette, sub-family C (CFTR/MRP), member 2	2.05	NM_013806	0.02

Since autophagy and mitophagy are dependent on increased lysosomal degradation activity, we also assessed the transcriptional profiles of lysosomal proteases (the cathepsins) and other lysosomal associated proteins (Table 4). Interestingly, numerous cathepsin genes and lysosomal associated proteins were transcriptionally upregulated in Cav-1 (−/−) stromal cells.

Table 4.

Upregulation of gene transcripts encoding lysosomal proteins in Cav-1 (−/−) stromal cells

Gene	Description	Fold change (KO/WT)	Accession number	p-value
Lysosome Associated Proteases
Ctsa	cathepsin A	1.64	NM_008906	0.03
Ctsb	cathepsin B	1.60	NM_007798	0.03
Ctsf	cathepsin F	1.62	NM_019861	0.01
Ctsh	cathepsin H	1.46	NM_007801	0.02
Ctsj	cathepsin J	1.29	NM_012007	0.05
Ctsl	cathepsin L	1.46	NM_009984	0.01
Other Lysosomal Associated Proteins
Abca2	ATP-binding cassette, sub-family A (ABC1), member 2	1.55	NM_007379	0.02
Als2	amyotrophic lateral sclerosis 2 (juvenile) homolog (human)	1.84	NM_028717	0.02
Atp6v0a2	ATPase, H+ transporting, lysosomal V0 subunit A2	1.46	NM_011596	0.03
Atp6v0a4	ATPase, H+ transporting, lysosomal V0 subunit A4	1.78	NM_080467	0.02
Atp6v0d1	ATPase, H+ transporting, lysosomal V0 subunit D1	1.60	NM_013477	0.02
Atp6v1b1	ATPase, H+ transporting, lysosomal V1 subunit B1	1.69	NM_134157	0.04
Atp6v1d	ATPase, H+ transporting, lysosomal V1 subunit D	1.79	NM_023721	0.003
Atp6v1e1	ATPase, H+ transporting, lysosomal V1 subunit E1	2.08	NM_007510	0.03
Atp6v1g2	ATPase, H+ transporting, lysosomal V1 subunit G2	1.36	NM_023179	0.02
Atp6v1h	ATPase, H+ transporting, lysosomal V1 subunit H	1.43	NM_133826	0.001
Hps1	Hermansky-Pudlak syndrome 1 homolog (human)	1.43	NM_019424	0.02
Lamp3	lysosomal-associated membrane protein 3	1.22	NM_177356	0.01
Laptm5	lysosomal-associated protein transmembrane 5	1.93	NM_010686	0.02
Lyst	lysosomal trafficking regulator	1.25	NM_010748	0.03
Manba	mannosidase, beta A, lysosomal	1.35	NM_027288	0.03
Manbal	mannosidase, beta A, lysosomal-like	1.78	NM_026968	0.03
Npc1	Niemann Pick type C1	1.88	NM_008720	0.04
Rilpl1	Rab interacting lysosomal protein-like 1	1.34	NM_021430	0.05
Tcirg1	T-cell, immune regulator 1, ATPase, H+ transporting, lysosomal V0 protein A3	1.80	NM_016921	0.05
Use1	unconventional SNARE in the ER 1 homolog (S. cerevisiae)	1.37	NM_001145780	0.03

In further support of increased oxidative stress, we also see the transcriptional overexpression of numerous genes associated with glutathione metabolism, genes responsive to oxidative stress and hypoxia, as well as numerous anti-oxidant proteins (Table 5).

Table 5.

Upregulation of telomerase and selected redox-related gene transcripts in Cav-1 (−/−) stromal cells

Gene	Description	Fold change (KO/WT)	Accession number	p-value
Telomerase
2610028A01Rik	RIKEN cDNA 2610028A01 gene	1.36	NM_028228	0.03
Tep1	telomerase associated protein 1	1.72	NM_009351	0.02
Glutathione-Related Genes
Aldh5a1	aldhehyde dehydrogenase family 5, subfamily A1	1.35	NM_172532	0.02
G6pdx	glucose-6-phosphate dehydrogenase X-linked	2.72	NM_008062	0.05
Ggt5	gamma-glutamyltransferase 5	1.43	NM_011820	0.01
Ggt7	gamma-glutamyltransferase 7	1.71	NM_144786	0.04
Gpx1	glutathione peroxidase 1	1.82	NM_008160	0.04
Gpx2	glutathione peroxidase 2	2.04	NM_030677	0.03
Gpx3	glutathione peroxidase 3	2.09	NM_001083929	0.03
Gpx4	glutathione peroxidase 4	1.57	NM_008162	0.05
Gpx5	glutathione peroxidase 5	1.51	NM_NM_010343	0.05
Gpx7	glutathione peroxidase 7	1.35	NM_024198	0.007
Gsr	glutathione reductase	1.59	NM_010344	0.05
Gsta4	glutathione S-transferase, alpha4	1.92	NM_010357	0.01
Gstcd	glutathione S-transferase, C-terminal domain containing	1.29	NM_026231	0.003
Gstm1	glutathione S-transferase, mu1	2.53	NM_010358	0.01
Gstm2	glutathione S-transferase, mu2	1.32	NM_008183	0.02
Gstm3	glutathione S-transferase, mu3	1.20	NM_010359	0.01
Gstm4	glutathione S-transferase, mu4	1.37	NM_026764	0.004
Gstm7	glutathione S-transferase, mu7	1.59	NM_026672	0.02
Gsto1	glutathione S-transferase omega1	1.65	NM_01036	0.04
Gstp2	glutathione S-transferase, pi2	1.37	NM_181796	0.02
Gstt1	glutathione S-transferase, theta1	1.44	NM_008185	0.03
Gstt3	glutathione S-transferase, theta3	2.15	NM_133994	0.02
Gstt4	glutathione S-transferase, theta4	1.55	NM_029472	0.006
Gstz1	glutathione transferase zeta1 (maleylacetoacetate isomerase)	1.24	NM_010363	0.003
Haghl	hydroxyacylglutathione hydrolase-like	1.79	NM_026897	0.01
Idh1	isocitrate dehydrogenase 1 (NADP+), soluble	1.78	NM_001111320	0.04
Oxidative Stress
Als2	amyotrophic lateral sclerosis 2 (juvenile) homolog (human)	1.84	NM_028717	0.02
Casp3	caspase 3	1.42	NM_009810	0.02
Epas1	endothelial PAS domain protein 1	2.39	NM_010137	0.0004
Ercc2	excision repair cross-complementing rodent repair deficiency, complementation group 2	1.59	NM_007949	0.04
G6pdx	glucose-6-phosphate dehydrogenase X-linked	2.72	NM_008062	0.05
Gpx1	glutathione peroxidase 1	1.82	NM_008160	0.04
Gpx2	glutathione peroxidase 2	2.04	NM_030677	0.03
Gpx3	glutathione peroxidase 3	2.09	NM_001083929	0.03
Gpx4	glutathione peroxidase 4	1.57	NM_008162	0.05
Gpx5	glutathione peroxidase 5	1.51	NM_NM_010343	0.05
Gpx7	glutathione peroxidase 7	1.35	NM_024198	0.007
Hnf1a	HNF1 homeobox A	2.58	NM_009327	0.04
Idh1	isocitrate dehydrogenase 1 (NADP+), soluble	1.78	NM_001111320	0.04
Mpo	myeloperoxidase	1.58	NM_010824	0.02
Mtf1	metal response element binding transcription factor 1	1.65	NM_008636	0.03
Oxsr1	oxidative-stress responsive 1	1.46	NM_133985	0.02
Rrm2b	ribonucleotide reductase M2 B (TP53 inducible)	1.81	NM_199476	0.03
Srxn1	sulfiredoxin 1 homolog (S. cerevisiae)	1.51	NM_029688	0.04
Tpo	thyroid peroxidase	1.99	NM_009417	0.02
Txnip	thioredoxin interacting protein	1.70	NM_001009935	0.03
Uaca	uveal autoantigen with coiled-coil domains and ankyrin repeats	1.44	NM_028283	0.04
Xpa	xeroderma pigmentosum, complementation group A	1.71	NM_011728	0.02
Hypoxia
Arnt2	aryl hydrocarbon receptor nuclear translocator 2	1.78	NM_007488	0.02
Casp1	caspase 1	1.79	NM_009807	0.05
Cd24a	CD24a antigen	1.59	NM_009846	0.06
Epas1	endothelial PAS domain protein 1	2.39	NM_010137	0.0004
Epo	erythropoietin	1.86	NM_007942	0.0005
Hyou1	hypoxia upregulated 1	1.56	NM_021395	0.02
Itpr1	inositol 1,4,5-triphosphate receptor 1	1.55	NM_010585	0.02
Kcnma1	potassium large conductance calcium-activated channel, subfamily M, alpha member 1	1.68	NM_010610	0.007
Mecp2	methyl CpG binding protein 2	1.63	NM_010788	0.01
Mmp2	matrix metallopeptidase 2	1.86	NM_008610	0.004
P2rx2	purinergic receptor P2X, ligand-gated ion channel, 2	1.57	NM_153400	0.04
Plat	plasminogen activator, tissue	1.57	NM_008872	0.01
Pml	promyelocytic leukemia	1.45	NM_178087	0.05
Slc2a8	solute carrier family 2, (facilitated glucose transporter), member 8	1.30	NM_019488	0.02
Sod3	superoxide dismutase 3, extracellular	1.28	NM_011435	0.03
Redoxins
Fdx1	ferredoxin 1	1.28	NM_007996	0.05
Fdxr	ferredoxin reductase	1.95	NM_007997	0.02
Glrx3	glutaredoxin 3	1.26	NM_023140	0.01
Nxn	nucleoredoxin	1.23	NM_008750	0.03
Park7	Parkinson disease (autosomal recessive, early onset) 7	1.97	NM_020569	0.02
Srxn1	sulfiredoxin 1 homolog (S. cerevisiae)	1.51	NM_029688	0.04
Tmx3	thioredoxin-related transmembrane protein 3	1.30	NM_198295	0.02
Txndc11	thioredoxin domain containing 11	1.28	NM_029582	0.02
Txndc3	thioredoxin domain containing 3 (spermatozoa)	2.21	NM_181591	0.007
Txndc5	thioredoxin domain containing 5	1.47	NM_145367	0.02
Txnip	thioredoxin interacting protein	1.70	NM_001009935	0.03
Txnl4a	thioredoxin-like 4A	1.89	NM_025299	0.04
Txnrd2	thioredoxin reductase 2	1.34	NM_013711	0.03

Thus, the transcriptional mRNA profile of Cav-1 (−/−) stromal cells is consistent with oxidative stress-induced autophagy and mitophagy. See also Supplemental Tables 1 and 2, including additional lysosome-related and peroxisome-related gene transcripts.

Transcriptional mRNA profiling of human breast cancer stroma provides evidence for stromal autophagy and mitophagy in vivo.

To further test the possible clinical relevance of our observations regarding autophagy and mitophagy, we next analyzed the transcriptional profiles of human tumor stroma that was isolated by laser-capture micro-dissection of breast cancer tumor tissue.³⁴ The methods and origins of these samples have been previously described in detail.¹⁷

Using these raw transcriptional profiling data,³⁴ we created three related gene lists: (1) tumor stroma, (2) recurrence stroma and (3) metastasis stroma.¹⁷ The tumor stroma list contains genes that were upregulated in the stroma of primary tumors, as compared with normal mammary gland stroma. The recurrence stroma list contains stromal genes, from the primary tumor, that are upregulated in patients that underwent tumor recurrence, as compared with patients that did not recur. The metastasis stromal list contains stromal genes, from the primary tumor, that are upregulated in patients that underwent lymph node metastasis at diagnosis, as compared with patients that did not show lymph node metastasis. Thus, we analyzed these 3 complementary gene lists for evidence of autophagy and mitophagy.

Table 6 shows that many of the genes that are associated with autophagy and mitophagy^32,33 are transcriptionally upregulated in the tumor stroma of human breast cancer patients. In further support of this idea, we see the upregulation of lysosomal proteases (the cathepsins and legumain), as well as other lysosomal associated proteins (Table 7). Finally, genes associated with glutathione metabolism, oxidative and hypoxic stress, as well as anti-oxidants are all transcriptionally upregulated in the tumor stroma obtained from human breast cancer patients (Table 8). Most interestingly, many of these gene transcripts are also associated with tumor recurrence and metastasis (See Tables 6–8).

Table 6.

Upregulation of autophagy/mitophagy related gene transcripts in the tumor stroma from human breast cancer patients

Gene	Description	Tumor stroma	Recurrence-stroma	Metastasis-stroma
Known Genes Associated with Autophagy
Atg9b	autophagy related 9 homolog B (yeast)	1.97E-20	1.58E-03
Atg16l1	autophagy-related 16-like 1 (yeast)	3.75E-15	1.64E-04
Atg7	autophagy-related 7 (yeast)	7.64E-14
Atg4b	autophagy-related 4B (yeast)	3.49E-05
Becn1	beclin 1, autophagy related	7.50E-05
Atg4a	autophagy-related 4A (yeast)	6.21E-03
Atg4b	autophagy-related 4B (yeast)			2.75E-02
Atg3	autophagy-related 3 (yeast)	1.01E-02	3.08E-03
Il3	interleukin 3	2.72E-08
Tsc1	tuberous sclerosis 1	2.44E-05
Tsc2	tuberous sclerosis 2			9.19E-05
Known Genes Associated with Mitophagy
Bnip2	BCL2/adenovirus E1B interacting protein 2	1.98E-12
Bnipl	BCL2/adenovirus E1B 19 kD interacting protein like	8.29E-11		3.08E-02
Bnip3	BCL2/adenovirus E1B interacting protein 3		8.08E-03
Bnip3l	BCL2/adenovirus E1B interacting protein 3-like	2.85E-08
Pink1	PTEN induced putative kinase 1	5.28E-09		4.60E-02
Park7	Parkinson disease (autosomal recessive, early onset) 7	2.93E-17
Park2	Parkinson disease (autosomal recessive, juvenile) 2, parkin	1.19E-16
Pddc1	Parkinson disease 7 domain containing 1	1.36E-08		2.65E-02
S100a8	S100 calcium binding protein A8 (calgranulin A)	5.96E-18
S100a9	S100 calcium binding protein A9 (calgranulin B)	2.53E-06	1.03E-02
Genes Associated with Mitophagy in a New Yeast Genetic Screen
Mon1a	MON1 homolog A (yeast)	1.36E-10	3.94E-03
Stx12	syntaxin 12	5.87E-10
Vps41	vacuolar protein sorting 41 (yeast)	6.57E-03
Atp2c1	ATPase, Ca++-sequestering	6.59E-24
Dnm1	dynamin 1	1.13E-11
Gsg1	germ cell-specific gene 1	2.30E-18	7.34E-04
Mak10	MAK10 homolog, amino-acid N-acetyltransferase subunit	1.70E-07
Rpl13a	ribosomal protein L13a			3.14E-02
Abcc2	ATP-binding cassette, sub-family C (CFTR/MRP), member 2			2.47E-02

Table 7.

Upregulation of gene transcripts encoding lysosomal proteins in the tumor stroma from human breast cancer patients

Gene	Description	Tumor stroma	Recurrence-stroma	Metastasis-stroma
Stromal Expresssion of Lysosome Associated Proteases
Ctsb	cathepsin B	4.11E-36	3.27E-02	1.39E-02
Ctsz	cathepsin Z	5.62E-24
Ctsk	cathepsin K	3.28E-20
Ctso	cathepsin O	1.49E-19
Ctse	cathepsin E	1.66E-19	3.78E-03
Ctss	cathepsin S	3.70E-18
Ctsw	cathepsin W	1.94E-15
Ctsf	cathepsin F	7.03E-14
Ctsh	cathepsin H	7.21E-12
Ctsg	cathepsin G	1.32E-11	4.03E-04
Lgmn	legumain			4.24E-02
Stromal Association of Other Lysosomal Associated Proteins
Atp6ap1	AtPase, H+ transporting, lysosomal accessory protein 1	2.11E-27
Lamp2	lysosomal-associated membrane protein 2	3.75E-24
Atp6v0a2	AtPase, H+ transporting, lysosomal V0 subunit A2	1.06E-22
Lamp3	lysosomal-associated membrane protein 3	3.09E-21
Atp6v0e	AtPase, H+ transporting, lysosomal V0 subunit E	1.39E-20	4.85E-02
Atp6v0d2	AtPase, H+ transporting, lysosomal V0 subunit D2	2.00E-20
Atp6v1b1	AtPase, H+ transporting, lysosomal V1 subunit B1	8.45E-20	6.56E-03
Atp6v0d1	AtPase, H+ transporting, lysosomal V0 subunit D1	1.32E-19
Lipa	lysosomal acid lipase A	4.83E-19
Atp6v1g2	AtPase, H+ transporting, lysosomal V1 subunit G2	7.64E-18
Atp6v1a	AtPase, H+ transporting, lysosomal V1 subunit A	8.45E-17	1.51E-08
Atp6v1c2	AtPase, H+ transporting, lysosomal V1 subunit C2	7.64E-16	5.72E-03
Lamp1	lysosomal-associated membrane protein 1	1.22E-15	8.38E-04
Tcirg1	T-cell, immune regulator 1, AtPase, H+ transporting, lysosomal V0 protein A3	1.26E-14	1.96E-02
Atp6v1c1	AtPase, H+ transporting, lysosomal V1 subunit C1	1.62E-13
Laptm4a	lysosomal-associated protein transmembrane 4A	2.89E-13
Atp6v0a1	AtPase, H+ transporting, lysosomal V0 subunit A1	5.35E-12	5.93E-03
Lyst l	ysosomal trafficking regulator	1.20E-11
Atp6v1g3	AtPase, H+ transporting, lysosomal V1 subunit G3	2.43E-08
Atp6v0a4	AtPase, H+ transporting, lysosomal V0 subunit A4	4.37E-07	2.34E-03
Laptm4b	lysosomal-associated protein transmembrane 4B		3.54E-02

Table 8.

Upregulation of telomerase and selected redox-related gene transcripts in the tumor stroma from human breast cancer patients

Gene	Description	Tumor stroma	Recurrence-stroma	Metastasis-stroma
Telomerase-Related
Rtel1	regulator of telomere elongation helicase 1	6.03E-06
Terf1	telomeric repeat binding factor 1			4.49E-02
Terf2	telomeric repeat binding factor 2	3.56E-08	1.63E-02
Tert	telomerase reverse transcriptase	1.34E-11	1.23E-02	2.17E-02
Tep1	telomerase associated protein 1	2.21E-07	5.05E-05
Tinf2	Terf1 (TRF1)-interacting nuclear factor 2	8.40E-06
Glutathione-Related Genes
Gpx1	glutathione peroxidase 1	1.77E-31
Mgst3	microsomal glutathione S-transferase 3	1.29E-19
Gpx6	glutathione peroxidase 6	1.66E-19
Gsta3	glutathione S-transferase, alpha3	1.00E-15	9.54E-04
Gpx7	glutathione peroxidase 7	1.93E-15
Gstt2	glutathione S-transferase, theta2	1.07E-14
Gpx5	glutathione peroxidase 5	5.30E-14	4.33E-03
Gstcd	glutathione S-transferase, C-terminal domain containing	6.17E-11
Gstm5	glutathione S-transferase, mu5	8.78E-08	4.56E-02	7.39E-03
Gpx2	glutathione peroxidase2	9.98E-08	3.10E-02
Gstm3	glutathione S-transferase, mu3	2.87E-02
Haghl	hydroxyacylglutathione hydrolase-like	4.02E-02
Gstm4	glutathione S-transferase, mu4		1.44E-02
Mgst1	microsomal glutathione S-transferase 1			1.44E-02
Gpx3	glutathione peroxidase 3			4.32E-02
Oxidative Stress and Hypoxia
Oxsr1	oxidative-stress responsive 1	5.45E-20	1.43E-02
Hif3a	hypoxia inducible factor 3alpha subunit	2.27E-15	2.15E-04
Hyou1	hypoxia upregulated 1	3.45E-14	2.85E-04
Hif1an	hypoxia-inducible factor 1alpha subunit inhibitor	8.86E-12	1.95E-03	6.98E-03
Hif1a	hypoxia inducible factor 1alpha subunit	2.10E-06
Redoxin's
Txndc3	thioredoxin domain containing 3 (spermatozoa)	9.19E-27
Txnl4a	thioredoxin-like 4A	5.81E-22
Txndc2	thioredoxin domain containing 2 (spermatozoa)	3.65E-16
Prdx3	peroxiredoxin 3	9.85E-12	2.02E-03
Txnrd2	thioredoxin reductase 2	1.21E-11
Glrx5	glutaredoxin 5 homolog (S. cerevisiae)	2.02E-10
Nxn	nucleoredoxin	3.48E-04
Txnip	thioredoxin interacting protein	1.77E-03
Glrx	glutaredoxin	2.70E-02
Txnl1	thioredoxin-like 1	2.99E-02
Prdx2	peroxiredoxin 2			2.18E-02
Prdx6	peroxiredoxin 6			3.62E-02
Fdxr	ferredoxin reductase			4.29E-02
Mterfd2	MTERF domain containing 2	8.13E-23
Mterfd1	MTERF domain containing 1		3.81E-02	1.32E-02

It is important to note that telomerase-related genes are also transcriptionally upregulated in both Cav-1 (−/−) stromal cells and the tumor stroma of human breast cancers (See Tables 5 and 8). Thus, overexpression of telomerase activity could provide an escape mechanism to keep stromal cell cells alive for much longer periods of time under conditions of oxidative stress, autophagy and mitophagy.

To independently assess the statistical association of autophagy, lysosomes, peroxisomes and telomere-related gene transcripts with the human tumor stroma of breast cancer patients, we next used more comprehensive gene ontology lists (see Suppl. Tables 3–6) to intersect with the tumor stroma, recurrence stroma and metastasis stroma gene lists. The results of this more detailed lysosomal degradation and telomere-maintenance) may play a analysis are presented in Figures 3–5, and are represented as significant pathogeneic role in generating an activated lethal Venn diagrams. tumor stroma.

Figure 3.

Venn diagrams for the transcriptional overlap between autophagy and tumor stroma from breast cancer patients. Upper panel, overlap with tumor stroma. Note the overlap of 93 genes with a p value of 2.65 × 10⁻⁶. Middle panel, Overlap with “recurrence-prone” stroma. Note the overlap of 47 genes with a p value of 2.22 × 10⁻³. Lower panel, overlap with “metastasis-prone” stroma. Note the overlap of 17 genes with a p value of 5.32 × 10⁻².

Figure 5.

Venn diagrams for the transcriptional overlap between peroxisomes and tumor stroma from breast cancer patients. Upper panel, overlap with tumor stroma. Note the overlap of 204 genes with a p value of 4.25 × 10⁻¹². Lower panel, overlap with “recurrence-prone” stroma. Note the overlap of 101 genes with a p value of 2.76 × 10⁻⁵.

More specifically, Figure 3 shows that autophagy-related genes are significantly associated with tumor stroma, recurrence-stroma and metastasis-stroma. Similarly, lysosome-related genes were significantly associated with tumor stroma and recurrence-stroma, while telomere-related genes were only associated with metastasis-stroma (Fig. 4A and B). Finally, peroxisome-related genes were significantly associated with both tumor stroma and recurrence-stroma (Fig. 5). Thus, all of these inter-related biological processes (oxidative stress, autophagy/lysosomal degradation and telomere-maintenance) may play a significant pathogeneic role in generating an activated lethal tumor stroma.

Figure 4.

Venn diagrams for the transcriptional overlap between lysosomes and telomere-related genes, with tumor stroma from breast cancer patients. Upper panel, overlap with tumor stroma. Note the overlap of 175 genes with a p value of 1.23 × 10⁻¹⁵. Middle panel, overlap with “recurrence-prone” stroma. Note the overlap of 74 genes with a p value of 2.10 × 10⁻³. Lower panel, Overlap with “metastasis-prone” stroma. Note the overlap of 38 genes with a p value of 9.67 × 10⁻⁵.

ADMA and ketone metabolism in Cav-1 (−/−) stromal cells and human tumor stroma.

Here, we have identified ADMA and 3-hydroxybutyrate (BHB) as the two major metabolites, which increased in Cav-1 (−/−) null mammary fat pads and lung tissue, along with pyruvate to a lesser extent. We believe that these two metabolites (ADMA and BHB) are reflective of oxidative stress and mitochondrial dysfunction in Cav-1 (−/−) stromal cells.

To further validate these observations, we analyzed transcriptional profiling data for the expression of the relevant enzymes that are involved in ADMA and ketone metabolism. Both transcriptional profiles from Cav-1 (−/−) null stromal cells and human breast cancer tumor stroma were analyzed in parallel and are presented in Table 9. For this purpose, we analyzed the mRNA expression of the genes involved in ADMA production (PRMT gene family members) and degradation (DDAH1/2), as well as nitric oxide (NO) related genes, as ADMA drives NOS uncoupling and the production of superoxide, instead of NO.¹⁹ Interestingly, using this approach, we see that the genes involved in both ADMA production (PRMT2/7/8) and degradation (DDAH1/2), as well as nitric oxide production (NOS1/2/3 or NOS trafficking), are all transcriptionally overexpressed, both in human tumor stroma and in Cav-1 (−/−) stromal cells.

Table 9.

Transcriptional profiling of human breast cancer tumor stroma: ADMA and BHB metabolism

Gene symbol	Tumor stroma	Recurrence-stroma	Metastasis-stroma	Cav-1 KO MSCs
ADMA Production				Fold-change	p-value
PRMT2		1.35E-02		1.68	0.1
PRMT7			1.76E-02
PRMT8	2.41E-18		3.04E-02	1.56	0.02
ADMA Degradation
DDAH1	2.36E-10			1.90	0.006
DDAH2	7.25E-16			1.46	0.005
Ketone Production
ACLY	1.55E-02	2.50E-02		2.19	0.02
HMGCS1				1.34	0.04
HMGCS2	3.91E-19	3.30E-02		1.52	0.04
HMGCL				2.24	0.03
HMGCLL1	3.19E-13	5.95E-04
BDH1	1.04E-06	6.46E-03		1.38	0.04
BDH2			3.54E-02
Ketone Re-Utilization
ACAT1
ACAT2				1.57	0.1
OXCT1				1.25	0.04
OXCT2				1.38	0.05
Nitric Oxide Production
NOS1	2.02E-13	1.82E-04
NOS2				1.36	0.05
NOS3	1.36E-06	1.21E-02		1.65	0.05
NOSIP			3.04E-02	1.59	0.03
NOSTRIN	2.09E-04			1.21	0.02

Note that for ADMA production, only PRMT genes that were upregulated in the tumor stromal gene sets or Cav-1 KO MSCs are listed. MSCs, mesenchymal stem cells.

Next, we turned our attention to ketone metabolism (Table 9). For this purpose, we analyzed the transcriptional profiles of the genes associated with both ketone production (ACYL, HMGCS1/2, HMGCL, HMGCLL1 and BDH1/2) and ketone re-utilization (ACAT1/2 and OXCT1/2). Interestingly, only the genes associated with ketone production, but not ketone re-utilization were associated with human tumor stroma. This is exactly as would be predicted, as the epithelial cancer cells should express the genes associated with ketone re-utilization, so that they can re-use 3-hydroxybutyrate as an energy source for mitochondrial oxidative metabolism. Also, many of the stromal genes involved in ketone production are specifically associated with tumor recurrence (ACLY, HMGCS2, HMGCLL1 and BDH1) and/or metastasis (BDH2). Many of these ketone production genes are also transcriptionally overexpressed in Cav-1 (−/−) stromal cells, consistent with our current metabolic analysis.

Micro-RNA (miR) profiling provides new mechanistic insight into how loss of stromal Cav-1 drives oxidative stress, autophagy and mitochondrial dysfunction.

Since miRs have recently taken center stage in the molecular analysis of tumor progression and metastasis, we subjected Cav-1 (−/−) stromal cells to miR transcriptional profiling. Using this approach, we hoped to identify a subset of miRs that could explain the oxidative and catabolic phenotypes we observe metabolically in Cav-1 (−/−) mammary fat pads.

Table 10 shows that only a select number of miRs were transcriptionally upregulated in Cav-1 (−/−) stromal cells. For this analysis, we chose a cut-off of 1.5-fold increased (KO/WT). P-values are as shown. Note that top two miRs showed the most significant p-values. Notably, miR-31 and miR-34c were increased 4.2-fold and nearly three-fold, respectively.

Table 10.

Upregulation of miR's in Cav-1 (−/−) null stromal cells

Symbol	Fold change (KO/WT)	p-value
miR-31	4.24	0.002
miR-34c	2.95	0.01
miR-423-3p	2.18	0.02
miR-193b	2.08	0.09/ns
miR-423-5p	1.99	0.03
miR-342-5p	1.96	0.05
miR-210	1.74	0.07/ns
miR-574-3p	1.72	0.02
miR-182	1.71	0.04
miR-298	1.71	0.04
miR-28	1.70	0.1/ns
miR-744	1.68	0.1/ns
miR-20b	1.62	0.1/ns
miR-467h	1.59	0.07/ns
miR-185	1.58	0.04
miR-222	1.55	0.04
miR-125a-5p	1.53	0.06/ns

ns, not significant. Noteworthy miRs are highlighted in bold.

A large body of evidence suggests that both miR-31 and miR-34c play prominent roles in both tumorigenesis and metastasis. miR-34c is normally induced under conditions of oxidative stress, DNA damage and cellular senescence,^35–37 consistent with our metabolic and mRNA transcriptional profiling data related to oxidative stress.

miR-31, on the other hand, targets FIH (factor inhibiting HIF).¹⁸ This, in turn, leads to the loss of FIH protein expression, driving HIF1α transcriptional activity.¹⁸ Both hypoxia and HIF1α activation itself are known to be sufficient to induce autophagy and mitophagy.^38,39 Thus, loss of Cav-1 expression, driving miR-31 overexpression and HIF1α activation, may be sufficient to explain our current findings related to oxidative stress, autophagy and mitophagy. In accordance with a role for Cav-1 in hypoxia regulation and HIF1α activation, we also observed an increase in miR-210, which is known to mediate many of the effects attributed to the hypoxia genetic transcriptional program.⁴⁰ However, although miR-210 was increased nearly two-fold, it did not reach statistical significance (p = 0.07).

miR-31 has recently been shown to be increased in the serum of patients with oral squamous cell cancers, and is dramatically reduced upon tumor resection, indicating that it can function as a marker for monitoring the course of cancer therapy.⁴¹ miR-31 is also upregulated in human colon cancers.⁴² Similarly, miR-210 is increased in the serum of pancreatic cancer patients.⁴³

Overexpression of autophagy and mitophagy markers in Cav-1 (−/−) null mammary fat pads: Cathepsin B and BNIP3.

To further validate the idea that a loss of Cav-1 drives the onset of oxidative-stress induced autophagy, we next assessed the expression of established autophagy markers, namely cathepsin B⁴⁴ and BNIP3,³⁸ in Cav-1 (−/−) mammary fat pads. Cathepsin B is a well-known lysosomal cysteine protease that is upregulated in the tumor stroma of human breast cancers, and its expression is also associated with tumor recurrence and metastasis (Table 7). BNIP3 is a marker of autophagy, which mediates the autophagic destruction of mitochondria by a process called mitophagy.³⁸ BNIP3 is also upregulated by oxidative stress and/or hypoxia, and is under the direct transcriptional control of HIF1α.³⁸ The stromal expression of BNIP3 is also associated with breast cancer tumor recurrence (Table 6). Importantly, Figure 6 directly shows that both cathepsin B (the pro-enzyme and activated form) and BNIP3 are significantly overexpressed in Cav-1 (−/−) null mammary fat pads, relative to wild-type controls processed in parallel. Immunoblotting with Cav-1 and beta-actin are shown for comparison. These results are consistent with the idea that a loss of Cav-1 expression promotes the onset of autophagy in the tumor stromal compartment.

Figure 6.

Over-expression of autophagy and mitophagy markers in Cav-1 (−/−) null mammary fat pads: Cathepsin B and BNIP3. To validate the idea that a loss of Cav-1 drives the onset of autophagy, we assessed the expression of established autophagy markers, namely cathepsin B and BNIP3, in Cav-1 (−/−) mammary fat pads. Cathepsin B is a well-known lysosomal protease. BNIP3 is a marker of autophagy that mediates the autophagic destruction of mitochondria. Note that both cathepsin B (the pro-enzyme and activated form) and BNIP3 are significantly overexpressed in Cav-1 (−/−) null mammary fat pads (KO), relative wild-type controls (WT). Immuno-blotting with Cav-1 and ²-actin are shown for comparison. CTSB, cathepsin B.

Discussion

Here, we have used the mammary fat pad of Cav-1 (−/−) null mice as a pre-clinical model for a “lethal tumor-microenvironment”, i.e., the tumor stroma without the tumor. We have previously documented that a loss of stromal Cav-1 in the fibroblast compartment of human breast cancer, DCIS and prostate cancer is associated with poor clinical outcome.^9–12,14 In breast cancer, a loss of stromal Cav-1 is a single independent predictor of early tumor recurrence, lymph node metastasis and tamoxifen-resistance.^9,14 In DCIS, a loss of stromal Cav-1 predicts both early recurrence and progression to invasive breast cancer.¹⁰ Finally, in prostate cancer patients, a loss of stromal Cav-1 is associated with advanced prostate cancer and tumor progression/metastasis and high Gleason score, indicative of a poor prognosis.¹¹

Oxidative stress and autophagy/mitophagy in the tumor micro-environment.

To mechanistically understand the lethality of a loss of Cav-1 in the tumor stromal compartment, we have now taken an unbiased screening approach, by performing a metabolomics analysis on fresh tissue harvested from the mammary fat pads of Cav-1 (−/−) null mice. Based on this analysis, we provide evidence for a series of severe metabolic defects in Cav-1 deficient tissues. More specifically, we show that nearly 100 metabolites are elevated in Cav-1 (−/−) null mammary fat pads. An analysis of these data is consistent with the onset of oxidative stress phenotype, combined with mitochondrial dysfunction and autophagy. The two most significant metabolites that are elevated are ADMA and 3-hydroxybutyrate. Also, several energy-rich metabolites, such as pyruvate and metabolic components of the TCA cycle are increased. These phenotypic changes could provide a logical and intriguing explanation for the lethality of a Cav-1 deficient tumor micro-environment, as oxidative stress is known to drive both mitochondrial dysfunction and autophagy/mitophagy, and this would set-up a situation in which catabolism of the tumor stroma could be used to directly “feed” the anabolic growth of tumor epithelial cancer cells. This is an exceptionally ingenious parasitic strategy that could promote tumor progression and metastasis (summarized in Fig. 7).

Figure 7.

Understanding a lethal tumor micro-environment: Oxidative stress drives stromal autophagy/mitophagy, providing stromal-derived nutrients for epithelial cancer cells. Here, using metabolic, transcriptional mRNA and miR profiling, we have identified that loss of stromal Cav-1 induces oxidative stress, mitochondrial dysfunction and autophagy/mitophagy in the tumor micro-environment. This model would then provide recycled chemical building blocks (nutrients, amino acids, energy-rich metabolites, nucleotides) derived from stromal cells (fibroblasts) that then could be harnessed by epithelial cancer cells to promote tumor growth. Mitochondrial dysfunction and mitophagy would result in aerobic glycolysis in stromal cells, explaining our previous observations on the “reverse Warburg effect”. Many of the key components we have identified here through metabolic and micro-RNA profiling are shown in BOLD: miR-31, miR-34c, ADMA, essential amino acids (AA's), nucleotides, pyruvate, ketones (BHB) and TCA cycle intermediates.

To independently validate these assertions, we next used an informatics approach to re-interrogate our transcriptional profiling data obtained from the analysis of Cav-1 (−/−) deficient stromal cells, isolated from the bone marrow of Cav-1 knockout mice. Importantly, bone marrow mesenchymal stem cells are thought to be the precursors of cancer associated fibroblasts that are recruited by epithelial tumor cells to cancerous lesions.⁶ Based on our re-analysis of this data set, we provide evidence for the upregulation of numerous gene transcripts specifically associated with authophagy/mitophagy, lysosomal biogenesis, oxidative stress, the glutathione pathway, and the compensatory upregulation of anti-oxidant enzymes. These results provide direct independent validation of our metabolic profiling studies.

To assess the relevance of our findings for human breast cancers, we next looked for evidence of the same transcriptional profiles in the tumor stroma that was laser-capture micro-dissected from the primary human tumors of patients with breast cancer. Importantly, our re-interrogation of these data sets indicated that the following biological processes are well-represented in the tumor stroma: authophagy/mitophagy, lysosomal biogenesis, oxidative stress, the glutathione pathway, and the upregulation of anti-oxidant enzymes. Many of the transcripts associated with these processes were also related to tumor recurrence and lymph-node metastasis.

Identification of ADMA and ketones as key metabolites: implications for diagnosis and drug discovery.

Since ADMA and 3-hydroxybutyrate emerged as the two most important metabolites that were increased in our metabolomic analysis, we validated that the enzymes responsible for their production were transcriptionally increased both in Cav-1 (−/−) stromal cells and the tumor stroma isolated from human breast cancers. Thus, these new observations now provide an opportunity for both diagnostic stratification of patients and the design of new drug therapies, to both identify and combat an aggressive tumor micro-environment.

ADMA is a catabolic breakdown product released from methylated proteins after their proteolytic degradation.^19–21 It is known to be strongly associated with endothelial cell dysfunction and oxidative stress.^19–21 In addition, it also has biological activity and can enhance and propagate the effects of oxidative stress. For example, it is known to function as a natural endogenous inhibitor of nitric oxide synthase (NOS) enzymes, halting the production of nitric oxide (NO). However, it also changes the specificity of the NOS enzymes, allowing them to consitutively produce superoxide instead.^19–21 Thus, ADMA is both a marker of oxidative stress and actively generates more oxidative stress. Furthermore, ADMA changes the location of eNOS and directly targets the enzyme to mitochodria, where it produces superoxide.²¹ Thus, ADMA is a mitochondrial “time-bomb” that leads to irreversible oxidative damage within mitochondria, necessitating their destruction by mitophagy.²¹ This, in turn, provides a mechanism for turning on aerobic glycolysis, so that the stromal cells will produce energy to ensure their own survival. However, aerobic glycolysis in the stroma releases both lactate and pyruvate, which can be used by epithelial cancer cells undergoing TCA-based oxidative metabolism, thereby providing paracrine energy for tumor growth.

Stromal ketone production also likely plays a strong pathogenic role. Ketone production is a well-established marker of mitochondrial dysfunction,^22,23 consistent with our assertions regarding ADMA, oxidative stress and autophagy/mitophagy. Ketones are normally produced by the liver and virtually every other organ system in the body during periods of fasting and starvation, and they are then transferred to the brain to maintain survival of the organism. Just as pyruvate and lactate can be secreted and taken up by monocarboxylic acid transporters (MCTs), the ketones 3-hydroxybutyrate and acetoacetate both follow the same principles.^45–47 So, ketone bodies can be transferred directly from stromal cancer-associated fibroblasts to epithelial cancer cells via MCTs, without any energy expenditure. Moreover, ketones are a “super-fuel” for mitochondria, producing more energy than lactate/pyruvate, and simultaneously decreasing oxygen consumption.^45–47 In fact, because of these properties, ketones have been used to prevent ischemic tissue damage, in animal models undergoing either myocardial infarctions or stroke, leading to dramatically smaller ischemic/necrotic lesion area.^48,49 So, just as ketones are a “super-fuel” under conditions of ischemia in the heart and in the brain, they could fulfill a similar function during tumorigenesis, as the hypoxic tumor exceeds its blood supply. So, stromal ketone production could obviate the need for tumor angiogenesis. Once ketones are produced and released from stromal cells, they could then be re-utilized by epithelial cancer cells, where they could directly enter the TCA cycle, just like lactate and pyruvate. In this sense, ketones are a more powerful mitochondrial fuel, as compared with lactate and pyruvate. As a consequence, we have now expanded the “reverse Warburg effect” to include ketones as a paracrine energy source (summarized in Fig. 8). In this scheme, we propose that the production of ketone bodies results from Acetyl-CoA derived from pyruvate, via pyruvate dehydrogenase (PDH) and not from the beta-oxidation of fatty acids, because Cav-1 (−/−) null mice have a defect in the beta-oxidation of fatty acids (reviewed in ref. 16). This would mechanistically explain why lactate does not accumulate. Interestingly, ACLY (a cytosolic enzyme) may also contribute to ketone production by converting citrate (a TCA metabolite) to Acetyl-CoA.

Figure 8.

Can ketones fuel tumor growth? Here, we propose that ketones produced in the tumor micro-enviroment (in cancer associated fibroblasts) could fuel the growth of adjacent epithelial cancer cells. Ketone-producing enzymes (in the fibroblasts) and ketone re-utilizing enzymes (in the epithelial cancer cells) are shown in bold. Transfer of ketones would be accomplished by monocarboxylate transporters (MCTs). Normally the same scheme is used by the liver (for ketone production) and the brain (for ketone re-utilization) during extreme fasting or starvation, to maintain neuronal function. Thus, the liver cells are the cancer fibroblasts and the epithelial cells are the neurons. Interestingly, Cav-1 (−/−) stromal cells and the tumor stroma both show a shift towards liver-specific gene and protein expression. For example, Cav-1 (−/−) stromal cells produce alpha-fetoprotein and albumin, as seen by proteomics.¹⁵ Alpha-fetoprotein expression has been been previously localized to cancer-associated fibroblasts in human breast cancers.⁸⁰ The enzymes involved in ketone metabolism are as follows: ACYL, ATP citrate lyase (cytosolic); HMGCS1/2, 3-hydroxy-3-methylglutaryl-Coenzyme A synthase 1 (cytosolic)/2 (mitochondrial); HMGCL, 3-hydroxymethyl-3-methylglutaryl-Coenzyme A lyase; HMGCLL1, 3-hydroxymethyl-3-methylglutaryl-Coenzyme A lyase-like 1; BDH1/2, 3-hydroxybutyrate dehydrogenase, type 1 (mitochondrial)/type 2 (cytosolic); ACAT1/2, acetyl-Co-enzyme A acetyltransferase 1 (mitochondrial)/2 (cytosolic); OXCT1/2, 3-oxoacid CoA transferase 1 (mitochondrial)/2 (testis-specific). We propose that the production of ketone bodies results from Acetyl-CoA derived from pyruvate, via pyruvate dehydrogenase (PDH), and not from the beta-oxidation of fatty acids, because Cav-1 (−/−) null mice have a defect in the beta-oxidation of fatty acids (reviewed in ref. 16). this would also mechanistically explain why lactate does not accumulate. Interestingly, ACLY (a cytosolic enzyme) may also contribute to ketone production by converting citrate (a TCA metabolite) to Acetyl-CoA. This also results in the production of oxaloacetate, another TCA metabolite.

Thus, ADMA and ketone bodies (3-hydroxybutyrate/acetoacetate) levels could be used as diagnostic tools to assess patient outcome. ADMA and ketone levels could either be measured in patient serum/plasma, or directly determined from homogenates of fresh tumor tissue. We predict that high ADMA and ketone levels in cancer patient serum or human tumor samples will strictly correlate with poor clinical outcome. These simple diagnostic tests could be performed rapidly and quantitatively, allowing us to identify and monitor high-risk cancer patients, both at diagnosis and during therapy. They could also be used for treatment stratification.

There is also a new opportunity here for new drug development via targeted therapies. We envision that inhibition of ADMA production or ketone production/re-utilization should halt tumor growth, leading to tumor regression. As such, the enzymes associated with (1) ADMA production (all PRMT family members), (2) ketone production (ACLY, HMGCS1/2, HMGCL, HMGCLL1 and BDH1/2) and (3) ketone re-utilization (ACAT1/2 and OXCT1/2) should now all be considered as “druggable targets” for cancer chemotherapy and prevention. In fact, a number of known anti-oxidants have already been shown to have anti-tumor activity,⁵⁰ such as N-acetyl cysteine (NAC), vitamin C, quercetin and curcumin. In this regard, NAC acts both as a free radical scavenger, and directly feeds into the glutathione pathway, increasing the amounts of cellular glutathione; NAC is the most promising anti-oxidant for inhibiting mitophagy.⁵¹ Furthermore, also anti-lysosomal drugs that inhibit autophagy, such as chloroquine, are known to have very significant anti-tumor activity.⁵² This may be due to their ability to inhibit autophagy in the fibroblastic stromal tumor compartment.^53–56

Cancer connections with systemic sclerosis, diabetes and fasting.

Interestingly, a variety of human diseases are also associated with high levels of ADMA. One such disease is systemic sclerosis (Scc; scleroderma),^57,58 and Scc patients have a higher incidence of cancer.⁵⁹ Similarly, diabetic patients show both high serum levels of ADMA and ketones.^60,61 Thus, our current observations may also explain the close and emerging association between diabetes and cancer susceptibility.⁶² A number of elegant studies have been carried out in mouse animal models to assess this association and chemical induction of diabetes in rats (with streptozocin) is sufficient to enhance tumor growth.⁶³ Similarly, acute fasting in rodent animal models is also sufficient to dramatically increase tumor growth.⁶⁴ Both of these experimental conditions (diabetes and fasting/starvation) are known to be highly autophagic and ketogenic and thus, are consistent with our current hypothesis that autophagy/ketone production fuels tumor growth and metastasis. Thus, the combination of ADMA and ketones may also play a crucial and causal role in promoting tumorigenesis, by providing oxidative stress and the simultaneous release of high-energy nutrients from the tumor micro-environment. Of course, this would be complemented by oxidative stress-induced autophagy/mitophagy in the tumor micro-environment, thus providing the necessary recycled chemical building blocks (amino acids, nucleotides, TCA cycle intermediates, etc.) in a paracrine fashion to cancer epithelial cells, to promote tumor growth.

In support of these ideas linking fasting/autophagy, with cancer susceptibility and diabetes, adipocytes from obese patients with type 2 diabetes show decreased mTOR signaling and substantially enhanced autophagy.⁶⁵ Similarly, hypoxia, inflammation and michondrial dysfunction all inactivate mTOR signaling, leading to autophagy.

Can autophagy in the tumor micro-environment substitute for angiogenesis in promoting tumor growth?

The combination of oxidative stress, mitochondrial dysfunction and autophagy/mitophagy in cancer-associated fibroblasts could reduce the dependence of tumor growth and survival on neo-angiogenesis and vascularization. This may explain why many of the new angiogenesis inhibitors have not been as promising as expected in ongoing clinical trials, as our current observations suggest that a Cav-1 negative fibroblastic tumor micro-environment could actually subsume the role of tumor angiogenesis, without the need for increased tumor vascularization. This may be particularly relevant in the case of pancreatic cancers, which are a highly fibrotic class of relatively avascular tumors, and are exceptionally lethal.

In support of these assertions, we have now used genetically modified glycolytic fibroblasts that lack Cav-1 expression to assess their affects on human xenograft tumor growth using co-injections with a breast cancer cell line, namely MDA-MB-231 cells.⁶⁶ In these xenograft models, the genetically modified fibroblasts, lacking Cav-1 expression, increased tumor weight by ∼4-fold and tumor volume by nearly 8-fold, without a measurable increase in tumor angiogenesis.⁶⁶

Micro-RNA profiling: associations with oxidative stress and autophagy/mitophagy.

Here, we also performed micro-RNA (miR) profiling on Cav-1 (−/−) deficient stromal cells to gain mechanistic insight into how a loss of Cav-1 may drive oxidative stress, mitochondrial dysfunction and autophagy/mitophagy. Using this approach, we identified two miR species that were highly overexpressed in Cav-1 null stromal cells, namely miR-31 and miR-34c.

The upregulation of miR-34c is consistent with our results from both metabolomics and transcriptional profiling, as it is normally upregulated by oxidative stress, and is also associated with DNA damage and senescence, which are known down-stream effects of oxidative stress.^35–37 Similarly, the upregulation of miR-31 provides a means for the transcriptional activation of HIF1α,¹⁸ which is known to induce both autophagy and mitophagy, and to inhibit mitochondrial biogenesis.^38,39 The transcriptional activation of HIF1α by miR-31 is indirectly mediated by FIH-1 (factor inhibiting HIF), which is the direct target of miR-31.¹⁸ Thus, overexpression of miR-31 blocks the transcriptional expression of a HIF inhibitory factor, FIH-1, leading to HIF activation.¹⁸

The autophagic tumor stroma model of cancer metabolism: compartment-specific autophagy resolves the “autophagy paradox” in cancer chemotherapy.

Based on our current observations, we would like to propose a new model for cancer pathogenesis. In this model, tumor cells would activate autophagy in the tumor stromal compartment via paracrine mechanisms. Autophagy in the tumor stroma, especially in cancer-associated fibroblasts, would then provide epithelial cancer cells with a steady stream of recycled nutrients and energy-rich metabolites, which could then be re-used by cancer cells to drive increases in tumor growth and metastasis. Additional mesenchymal stem cells from the bone marrow could be recruited to the tumor and induced to undergo autophagy, to satisfy the tumor's appetite. The extension of this scheme from a local to a systemic phenomenon, could explain the onset of anorexia, cachexia, insulin-resistance and metabolic syndrome, all features that are known to be associated with chronic malignancy and this would provide the tumor with autophagic/catabolic-based nutrients (including ketone bodies)—from distant systemic sources. Tumor cells might even metastasize to the major sites of ketone production (the liver or adipose-tissue-rich bone marrow) or ketone re-utilization (the brain) in search of energy-rich metabolites.

In direct support of this model, we have recently shown that epithelial tumor cells induce autophagy in cancer-associated fibroblasts via oxidative stress, driving the autophagic/lysosomal degradation of Cav-1.^67,68 Under these conditions, Cav-1 degradation in cancer-associated fibroblasts was inhibited by anti-oxidants (such as N-acetyl-cysteine) or autophagy/lysosomal inhibitors (such as chloroquine).^67,68 Similarly, acute knock-down of Cav-1 in fibroblasts using an siRNA approach was sufficient to induce ROS production, oxidative stress and mitochondrial dysfunction.⁶⁸ Thus, a loss of Cav-1 is both up-stream and downstream of oxidative stress and autophagy in cancer-associated fibroblasts.

This model also provides a rationale basis for designing new therapeutic intervention(s), as autophagy in the tumor stroma may be sustaining tumor growth. Thus, inhibition of autophagy in the tumor stroma would be expected to halt or reverse tumor growth. This could explain the effectiveness of known autophagy inhibitors as anti-tumor agents,⁵² such as chloroquine and 3-methyladenine. Conversel , induction of autophagy in epithelial cancer cells would also be expected to block or inhibit tumor growth. This idea would explain the anti-tumor activity of agents that activate autophagy, such as mTOR inhibitors.⁶⁵ Thus, using this model, compounds that either systemically block or activate autophagy would both have the same net effect, which is to disrupt the metabolic coupling between the epithelial cancer cells and the tumor stromal fibroblasts (Fig. 9). This model directly resolves the long-lived “autophagy paradox,”⁶⁹ that both systemic inhibition of autophagy and systemic stimulation of autophagy have the same net effect, which is to inhibit tumor growth. Of course this new model will require further experimental validation, but it does provide a new paradigm and rationale basis for drug development, driving new metabolic therapeutic interventions.

Figure 9.

The autophagic tumor stroma model of cancer: Resolving the autophagy paradox in cancer therapy. Here, we propose that autophagy/mitophagy (AM) in the tumor stroma may be sustaining tumor growth. The large black arrow signifies energy transfer (E.T.) from the stromal cancer associated fibroblasts (CAFs) to the epithelial cancer cells, via stromal autophagy/mitophagy. Thus, inhibition of autophagy in the tumor stroma would be expected to halt or reverse tumor growth. This could explain the effectiveness of known autophagy inhibitors as anti-tumor agents,⁵² such as chloroquine and 3-methyladenine (Upper part). Conversely, induction of autophagy in epithelial cancer cells would also be expected to block or inhibit tumor growth (Lower part). This idea would also explain the anti-tumor activity of agents that activate autophagy, such as mTOR inhibitors.⁶⁵ Thus, using this model, compounds that either systemically block or activate autophagy would both have the same net effect, which is to disrupt the metabolic coupling between the epithelial cancer cells and the tumor stromal fibroblasts. This model directly resolves the long-lived “autophagy paradox”, that both systemic inhibition of autophagy and systemic stimulation of autophagy have the same net effect, which is to inhibit tumor growth. E.T., energy transfer; AM⁺, increased autophagy/mitophagy; AM⁻, decreased autophagy/mitophagy; Rx, therapy with autophagy promoters or inhibitors.

Clinical connections with malignancy: ADMA, ATG16L and SPARC.

Unfortunately, little is known about the association of ADMA and ketone production with malignancy. An extensive search of the literature revealed only one paper on ADMA and cancer. In this paper, the authors show that the serum levels of ADMA are indeed elevated in patients with hematological malignancies, as compared with normal healthy volunteers.⁷⁰ Although the authors did not link their findings to clinical outcome, their results with ADMA are indeed consistent with our current findings based on our metabolomics analysis of a mouse model of a “lethal tumor micro-environment”.

Similarly, little is known about the role of autophagy in the tumor micro-environment. We identified only one paper looking at the possible role of an autophagic marker, namely ATG16L, as a candidate cancer biomarker in oral squamous cell cancers.⁷¹ Interestingly, these authors showed that the epithelial tumor cell levels of ATG16L did not correlate with any clinico-pathological variables. However, high levels of stromal ATG16L were associated with (1) the lympho-vascular invasion of tumor cells and (2) positive lymph node status—consistent with our proposed model. Unfortunately, no data on clinical outcome were presented. We also observed that ATG16L was transcriptionally overexpressed in Cav-1 (−/−) stromal cells and the tumor stroma of human breast cancer patients, and its expression was associated with tumor recurrence (see Tables 3 and 6). Thus, future studies are warranted to determine if high expression of autophagy-associated biomarkers in the tumor stroma are a general feature of human epithelial cancers and are associated with poor clinical outcome.

SPARC is a multi-functional extracellular matrix protein that is associated with the tumor stroma. Recently, SPARC overexpression has been shown to be sufficient to induce autophagy in cells in culture via the upregulation of cathepsin B.⁴⁴ Similarly, we have previously demonstrated that Cav-1 (−/−) deficient stromal cells overexpress SPARC, as evidenced by both unbiased proteomic and genome-wide transcriptional profiling.¹⁵ Also, we have shown that the stromal expression of SPARC accurately predicts DCIS recurrence and/or progression.⁷² Taken together, these finding are consistent with the idea that a loss of stromal Cav-1 induces SPARC and autophagy in the tumor micro-environment, thereby promoting tumor progression in DCIS patients. In accordance with this idea, a loss of stromal Cav-1 is strongly associated with progression to invasive breast cancer in DCIS patients.⁹

Materials and Methods

Animals.

All animals were housed and maintained in a pathogen-free environment/barrier facility at the Kimmel Cancer Center at Thomas Jefferson University according to the guidelines of the National Institutes of Health. Mice were kept on a 12-hour light/dark cycle with ad libitum access to chow and water. Cav-1 (−/−) deficient mice were generated, as we previously described.⁷³ All mice used for metabolomics analysis were in the FVB/N genetic background. Animal protocols used for this study were pre-approved by the institutional animal care and use committee.

Metabolomics analysis and metabolite identification.

Metabolomic profiling analysis was achieved through a contractual service agreement with Metabolon, Inc., (www.metabolon.com/) using the following protocol, as described in Lawton et al. (2008).^74,75 Unbiased metabolic profiling technology based on sample extraction and mass spectrometry was applied to the (1) mammary fat pads and (2) lung tissue samples, from virgin female 5-month-old WT and Cav-1 (−/−) null mice (n = 6 tissues samples, for each organ type and for each genotype, for a total of 24 samples). To efficiently recover various metabolites, the sample preparation consisted of sequential organic and aqueous extractions. Four solvent extraction steps were used: 400 µl tridecanoic acid (2.5 mg/mL) in ethyl acetate:ethyl alcohol (1:1), 200 µl methanol, 200 µl methanol:H₂O (3:1) and 200 µl dichloromethane:methanol (1:1). The resulting pooled extract was equally divided into a liquid chromatography (LC) fraction and a gas chromatography (GC) fraction. After chromatographic separation, a full-scan mass spectra was carried out to record all detectable ions present in the samples. In both the GC and LC methods, we used internal standards to calibrate retention times of metabolites across all of the samples in the study and for quality control on each instrument run. Known chemical structure metabolites were identified by matching the ions' chromatographic retention index and mass spectra fragmentation signatures, with Metabolon's reference library entries created using Metabolon's software. Each entry identified by this software is visually inspected with VPhil Software to confirm the acceptance of that metabolite in the study. Peptides were identified using standard tandem MS sequencing techniques.⁷⁶ Missing values for a specific metabolite were assumed to have fallen below the limits of detection and were imputed with the observed minimum. Quantitative values were derived from integrated raw detector counts of the mass spectrometers. For the convenience of data visualization, the raw area counts for each biochemical were rescaled by dividing each sample's value by the median value for the specific biochemical. Statistical analysis of the data was performed using JMP (SAS, www.jmp.com), a commercial software package and “R” (cran.r-project.org/), which is a freely available, open-source software package. A log transform was applied to the observed relative concentrations for each biochemical because, in general, the variance increased as a function of a biochemical's average response. To account for multiple hypothesis testing, we estimated the false discovery rate⁷⁷ with the q-value method.⁷⁸ Data analyses were carried out by JMP (SAS Institute, Inc., Cary, NC) and R (R Foundation for Statistical Computing, Vienna, Austria). Significance was defined at p ≤ 0.05 and q ≤ 0.1.

Gene set overlap and p-value computation for venn diagrams.

We created four gene sets comprising genes related to autophagy, lysosomes, peroxisomes and telomeres. Each set was created by integrating information from two sources: (a) gene descriptions obtained from the ENSEMBL database, release 56 (www.ensembl.org) and (b) the Molecular Signatures Database (MsigDB⁷⁹). For each gene set we searched the related keyword (i.e., autophagy, lysosome, peroxisome or telomere) in both the genes' descriptions in ENSEMBL and the gene sets' descriptions in MSigDB, thus including all relevant genes according to these databases. Then, we computed all the pair-wise overlaps along with their p-values between these four gene sets and:

1. genes overexpressed in tumor versus normal stroma (p value ≤ 0.05).¹⁷

2. genes overexpressed in Recurrence + versus Recurrence-stroma (p value ≤ 0.05).¹⁷

3. genes overexpressed in LN⁺ versus LN⁻ stroma (p value ≤ 0.05).¹⁷

For each pair of gene sets X and Y, we computed the probability (p-value) that the observed overlap between sets X and Y is less than or equal to the overlap between set X and a randomly-chosen set of size equal to the size of set Y. This probability was calculated by applying the cumulative density function of the hypergeometric distribution on the size of set X, the size of set Y, the observed overlap between X and Y and the total number of available genes.

Micro RNA (miR) gene chip profiling.

The FlashTag™ Biotin RNA Labeling Kit (Genisphere, LLC, Hatfield, PA) was used, according to the manufacturer's protocol, to generate labeled miRNA molecules from 500 ng of total RNA by poly (A) tailing followed by ligation of the biotinylated signal molecule to the target RNA sample. Biotin labeled sample was hybridized to Affymetrix® GeneChip® miRNA Arrays (Affymetrix, Santa Clara, CA), Sanger miRBase V11. Hybridized chips were washed and stained using GeneChip® Fluidic Station 450 and scanned on an Affymetrix GeneChip® Scanner 3000, using Command Console Software 3.1. Affymetrix miRNA QCtool (V1.0.33.0) was used for data summarization and quality control. Background correction and normalization were done using Robust Multichip Average (RMA), baseline to median of all samples, analysis were performed with GeneSpring GX V 10.0 software (Agilent, Palo Alto, CA).

Western blot analysis.

Primary antibodies were obtained from the following commercial sources: caveolin-1 (N-20; sc-894; Santa Cruz Biotech); cathepsin B (FL-339; sc-13985; Santa Cruz Biotech); BNIP3 (ab10433; Abcam); beta-actin (clone AC15; A5441; Sigma). Mammary fat pads were surgically harvested from virgin female wild-type and Cav-1 (−/−) deficient mice. Then, the excised tissues were minced and placed in lysis buffer (10 mM Tris-HCl, pH 7.5, 150 mM NaCl, 1% Triton X-100, 60 mM octylglucoside), containing protease inhibitors (Roche Applied Science) and phosphatase inhibitors (Roche Applied Science). Samples were homogenized for 30 seconds and centrifuged at 13,000x g for 15 min at 4°C to remove insoluble debris. Protein concentrations were analyzed using the BCA reagent (Pierce). 30 µg of proteins were loaded and separated by SDS-PAGE and transferred onto a 0.2 µm nitrocellulose membrane. After blocking for 30 min in TBST (10 mM Tris-HCl pH 8.0, 150 mM NaCl, 0.05% Tween-20) with 3% BSA, the membranes were incubated with the primary antibody for 1 hour, washed and incubated for 30 min with horseradish peroxidase-conjugated secondary antibodies. The membranes were washed and incubated with an enhanced chemi-luminescence substrate (ECL; Thermo-Fisher Scientific).