Apolipoprotein A-I anti-tumor activity targets cancer cell metabolism.

Previously, we reported apolipoprotein A-I (apoA-I), the major protein component of high-density lipoprotein (HDL), has potent anti-melanoma activity. We used DNA microarray and bioinformatics to interrogate gene expression profiles of tumors from apoA-I expressing (A-I Tg+/-) versus apoA-I-null (A-I KO) animals to gain insights into mechanisms of apoA-I tumor protection. Differential expression analyses of 11 distinct tumors per group with > 1.2-fold cut-off and a false discovery rate adjusted p < 0.05, identified 176 significant transcripts (71 upregulated and 105 downregulated in A-I Tg+/- versus A-I KO group). Bioinformatic analyses identified the mevalonate and de novo serine/glycine synthesis pathways as potential targets for apoA-I anti-tumor activity. Relative to A-I KO, day 7 B16F10L melanoma tumor homografts from A-I Tg+/- exhibited reduced expression of mevalonate-5-pyrophosphate decarboxylase (Mvd), a key enzyme targeted in cancer therapy, along with a number of key genes in the sterol synthesis arm of the mevalonate pathway. Phosphoglycerate dehydrogenase (Phgdh), the first enzyme branching off glycolysis into the de novo serine synthesis pathway, was the most repressed transcript in tumors from A-I Tg+/-. We validated our mouse tumor studies by comparing the significant transcripts with adverse tumor markers previously identified in human melanoma and found 45% concordance. Our findings suggest apoA-I targets the mevalonate and serine synthesis pathways in melanoma cells in vivo, thus providing anti-tumor metabolic effects by inhibiting the flux of biomolecular building blocks for macromolecule synthesis that drive rapid tumor growth.

INTRODUCTION

Melanoma is a cancer derived from melanocytes, the pigment-producing cells of epidermis and hair follicles. It is trending upward in incidence and mortality worldwide, and has no effective treatment post metastasis [1-3]. This complex disease is driven by both genetic and epigenetic factors, and many studies have focused on gaining a more thorough understanding of the molecular mechanisms involved in disease progression [4]. Gene expression profiling of primary [5-8] or metastatic tumor specimens, as well as cell lines [6, 9–12], have identified disease markers and gene signatures for different stages of melanoma.

High-density lipoprotein (HDL), a physiological plasma molecule long known for its atheroprotective properties [13, 14], was linked to cancer in a large meta-analysis of randomized controlled trials of lipid-altering therapies that suggested an inverse relationship between plasma HDL cholesterol (HDL-c) levels and incident development of cancer [15]. Previously, we reported a potent anti-tumorigenic activity for apolipoprotein A-I (apoA-I), the major protein component of HDL, against human melanoma A375 and the highly aggressive and metastatic mouse melanoma B16F10L [16]. Accordingly, growth of syngeneic B16F10L tumor cells was severely restricted in animals expressing human apoA-I (A-ITg+/–; high plasma HDL-c levels) relative to apoA-I null mice (A-I KO; low plasma HDL-c levels). Importantly, apoA-I therapy curbed further growth of established tumors in A-I KO and induced tumor regression, thus preventing metastases and prolonging survival [16]. Tumor inhibition by apoA-I was also observed with human melanoma A375 in nude mice. We proposed an immunomodulatory role for apoA-I in melanoma with both innate and adaptive arms of immunity mediating its anti-tumor activity [16].

Cancer cells have different metabolic requirements from normal quiescent cells, given the propensity of malignant cells to proliferate at a high rate and avoid apoptotic death signals. To fulfill these priorities, cancer cells hijack normal metabolic and signaling pathways and redirect them to meet their increased need for biomolecules to synthesize proteins, lipids, and nucleic acids [17-22]. Altered cellular metabolism is now widely considered a hallmark of cancer, and interventions to disrupt cancer metabolism are fast emerging as viable therapeutic approaches in conjunction with conventional death-inducing chemotherapies [23-27]. Statins, a family of lipid-lowering drugs that target 3-hydroxy-3-methyl-glutaryl-co-enzyme A reductase (HMG-CoA reductase), the rate-limiting enzyme at the core of the mevalonate biochemical pathway, have been a focal point of research in the cancer field because their action leads to reduced cholesterol and other key metabolic end products such as activated (prenylated) small GTPases with oncogenic activity [24, 28–31]. Statins have been associated with reduced mortality from several cancers including prostate, kidney, colorectal, breast, and lung cancer [32-39]. Meta-analyses have also suggested a positive correlation between statin use and reduced incidence of melanoma [30].

Herein, we used differential gene expression analysis of primary B16F10L melanoma homografts to investigate the role of host apoA-I in the tumor microenvironment, and identified the mevalonate and de novo serine synthesis metabolic pathways as potential targets of apoA-I anti-tumor activity.

RESULTS

Whole-genome expression profiling and hierarchical clustering discriminate apoA-I-null-mouse-derived tumors from less aggressive tumors in apoA-I-expressing mice

To gain insight into molecular mechanisms and biological pathways underlying apoA-I tumor suppressive activity, we performed whole-genome expression profiling of B16F10L homografts from A-I Tg+/–, A-I KO, and WT mice by DNA microarray technology (a total of 24,613 probes corresponding to 17,877 mouse genes were interrogated). Previously, we reported a significant difference in tumor-associated angiogenesis between A-I KO and A-I Tg+/– seven days post tumor inoculation [16]. We therefore chose day 7 as a time point for transcriptomic analyses because adequate tumor mass could be harvested at this early time point. We compared 11 distinct primary tumors per genotype to compensate for variability when analyzing a heterogeneous population of cells, namely tumor and host tumor-infiltrating cells. The dendogram in Figure 1 illustrates the correlation analysis of log2 normalized gene expression levels where distance (y-axis) between clusters denotes divergence in the global gene expression pattern. While expression profiles of tumors from all three groups were generally similar, as reflected by the close distance between the clusters, for the most part the more aggressive A-I KO tumors clustered in a group distinct from the less aggressive tumors found in apoA-I-expressing animals (WT and A-I Tg+/–). Two of the eleven tumor gene signatures recovered from A-I KO animals, A-I KO 3 & 11, were exceptions in that they clustered within the A-I Tg+/– and WT groups. One of the WT gene signatures recovered (WT 360) was an outlier in a cluster furthest in distance from all other tumors (Figure 1).

Figure 1

Dendogram of unsupervised hierarchical clustering analysis of tumor gene expression (log2 normalized) profiles discriminate A-I KO-derived tumors from less aggressive tumors of apoA-I-expressing mice (A-I Tg+/– and WT).

B16F10L melanoma homograft grown in C57BL/6 mice wild type for mouse apoA-I (WT), deficient (A-I KO) or expressing human apoA-I (A-I Tg+/–) were resected 7 days after inoculation and processed for RNA isolation and microarray analysis as described in Methods.

Key biological pathways are dysregulated in B16F10L tumors from apoA-I null mice (A-I KO)

The expression array datasets between the tumor groups were analyzed (see Methods) to identify a cohort of genes that were differentially expressed. Volcano plots illustrating both statistical (y-axis, -log10 (p-value) and biological (x-axis, log2 (fold-change, FC)) significance of the differential expression analyses of A-I Tg+/– versus A-I KO or WT versus A-I KO are shown in Figure 2A. We restricted our genes to those with at least 1.2-fold differential expression and a false discovery rate (FDR)-adjusted p-value < 0.05. Use of this screening criteria resulted in 71 and 325 upregulated, and 105 and 177 downregulated transcript probes comparing A-I Tg+/– or WT versus A-I KO, respectively (Figure 2B). The heat map in Figure 2C illustrates the probes from differential expression analysis of A-I Tg+/– versus A-I KO with the expression in WT, serving as reference, to the right. Importantly, the overall gene expression pattern in WT group was approximately mid-way between A-I Tg+/– and A-I KO, supporting the tumor growth phenotype previously observed for this genotype relative to A-I KO and A-I Tg+/– (Figure 1 in [16]). The transcripts representing 68 (71 probes) up- and 91 (105 probes) downregulated genes are listed in Supplementary Tables 1 and 2, respectively. Tables 1A and 1B represent the top genes up- or downregulated in tumors from A-I Tg+/– versus A-I KO, respectively. Notably, Phgdh, the gene encoding phosphoglycerate dehydrogenase, PHGDH, was the most repressed transcript in tumors from apoA-I-expressing hosts (Table 1B). PHGDH is the first enzyme of the 3-step de novo serine synthesis pathway (Figure 3), which diverts glucose-derived carbon for synthesis of biomolecules necessary for rapid proliferation and survival. In this pathway, PHGDH oxidizes the glycolytic intermediate 3-phosphoglycerate (3PG) to 3-phospho-hydroxypyruvate (P-PYR) and couples to NADH production. Subsequently, P-PYR is transaminated by phosphoserine aminotransferase (PSAT1) with glutamate (Glu) serving as a nitrogen donor to form phosphoserine (P-Serine) and alpha-ketoglutarate (aKG), a key intermediate of the TCA cycle. P-Serine is dephosphorylated by phosphoserine phosphatase (PSPH) to form serine, which is essential for synthesis of glycine, proteins, lipids, and nucleic acids (Figure 3). Serine is also a major source of one-carbon units directed to the folate pool, which supports proliferation and transformation. The transcript for phosphoserine aminotransferase 1 (Psat1), the second enzyme of the de novo serine synthesis pathway, was similarly downregulated (Figure 3), however, the FDR-adjusted p-value (0.104) did not meet significance (p < 0.05).

Figure 2

Identification of statistically significant transcripts from differential expression analyses.

B16F10L melanoma homograft grown in C57BL/6 mice wild type for mouse apoA-I (WT), deficient (A-I KO) or expressing human apoA-I (A-I Tg+/–) were resected 7 days after inoculation and processed for RNA isolation and microarray analysis as described in Methods. (A) Volcano plots of expressed probes allowing for filtering of statistically significant (False Discover Rate (FDR) adjusted p < 0.05; Y-axis, -log10 (p-value)) and biologically significant probes (fold-change (FC), X-axis, log2(fold-change)). (B) Venn diagrams depicting the number of significant probes (FDR adjusted p < 0.05) that were up (left) or down (right) regulated (FC: 1.2) in A-I Tg+/– or WT with respect to A-I KO. (C) Two-way hierarchical heat map of significant transcripts (FDR adjusted p < 0.05) with fold-change cut off > 1.2 resulting from comparison of B16F10L tumors from A-I Tg+/– versus A-I KO mice. Columns refer to individual B16F10L primary tumors from host animals shown, and rows represent individual transcripts. Numbers following designated genotypes refer to unique mouse identifiers.

Table 1

Top genes up- or downregulated in tumors from A-I Tg+/– versus A-I KO

Table 1A: Upregulated genes in A-ITg vs A-IKO mice
Probe_ID	Accession no.	Symbol	Name	FC	adj. p-value
ILMN_1251748	NM_139200.4	Cytip	cytohesin 1 interacting protein	1.83	0.04
ILMN_1253182	NM_010474.1	Hs3st1	heparan sulfate (glucosamine) 3-O-sulfotransferase 1	1.81	0.012
ILMN_2622983	NM_013642.2	Dusp1	dual specificity phosphatase 1	1.67	0.033
ILMN_3161601	NM_009221.2	Snca	synuclein, alpha (non A4 component of amyloid precursor)	1.54	0.011
ILMN_2705166	NM_145933.3	St6gal1	ST6 beta-galactosamide alpha-2,6-sialyltranferase 1	1.49	0.012
ILMN_2722732	NM_011157.2	Srgn	serglycin	1.46	0.015
ILMN_3136638	NM_009221.2	Snca	synuclein, alpha (non A4 component of amyloid precursor)	1.44	0.011
ILMN_1254031	NM_010638.4	Klf9	Kruppel-like factor 9	1.43	0.008
ILMN_3161105	NM_001033476.1	Ahnak2	AHNAK nucleoprotein 2	1.42	0.047
ILMN_2829594	NM_010479.2	Hspa1a	heat shock 70kDa protein 1A	1.4	0.006
ILMN_2754985	NM_009344.1	Phlda1	pleckstrin homology-like domain, family A, member 1	1.4	0.006
ILMN_2824971	NM_001004761.1	Gpr158	G protein-coupled receptor 158	1.39	0.011
ILMN_2896314	NM_015732.3	Axin2	axin 2	1.38	0.004
ILMN_1233064	NM_183417.2	Cdk2	cyclin-dependent kinase 2	1.37	0.009
ILMN_2892441	NM_010357.1	Gsta4	glutathione S-transferase alpha 4	1.37	0.029
ILMN_2831799	NM_019517.2	Bace2	beta-site APP-cleaving enzyme 2	1.36	0.013
ILMN_2623280	NM_011019.1	Osmr	oncostatin M receptor	1.36	0.008
ILMN_2522236	NM_011661.3	Tyr	tyrosinase	1.36	0.039
ILMN_3114585	NM_001039150.1	Cd44	CD44 molecule	1.35	0.002
ILMN_1252202	NM_009397.2	Tnfaip3	tumor necrosis factor, alpha-induced protein 3	1.34	0.03
ILMN_2612895	NM_013484.1	C2	complement component 2	1.34	0.014
ILMN_2680415	NM_172537.2	Sema6d	Semaphorin	1.34	0.028
ILMN_1246153	NM_133362.2	Erdr1	erythroid differentiation regulator 1	1.34	0.017
ILMN_1240323	NM_018808.1	Dnajb1	DnaJ (Hsp40) homolog, subfamily B, member 1	1.34	0.001
ILMN_2813484	NM_011065.2	Per1	period circadian clock 1	1.33	0.015
ILMN_2774690	XM_001004685.1	LOC677317	similar to NADP-dependent malic enzyme (NADP-ME) (Malic enzyme 1)	1.33	0.049
ILMN_2734181	NM_019811.3	Acss2	acyl-CoA synthetase short-chain family member 2	1.32	0.026
ILMN_1226157	NM_181585.5	Pik3r3	phosphoinositide-3-kinase, regulatory subunit 3 (gamma)	1.32	0.042
ILMN_2746556	NM_015814.2	Dkk3	dickkopf 3 homolog (Xenopus laevis)	1.31	0.022
ILMN_3073563	NM_001001884.1	C230021P08Rik	Nckap5l NCK-associated protein 5-like	1.31	0.015
ILMN_2776619	NM_008520.2	Ltbp3	latent transforming growth factor beta binding protein 3	1.3	0.012
ILMN_1258600	XM_001481024.1	LOC100043671	LOC100043671 hypothetical protein	1.3	0.04
ILMN_2870672	NM_010180.1	Fbln1	fibulin 1	1.3	0.015

(A) Upregulated genes. (B) Downregulated genes. Eleven separate primary tumors from eleven A-I Tg+/– or A-I KO mice were subjected to differential gene expression analysis. Probes were considered significant if they met the criteria FDR adjusted p < 0.05 and fold-change (FC) cut off > 1.2. The complete list of significant genes is shown in Supplementary Tables 1 and 2.

Figure 3

PHGDH, the first enzyme of de novo serine synthesis pathway, is the most repressed transcript in tumors from A-I Tg+/– animals.

Eleven separate primary tumors from eleven A-I Tg+/– or A-I KO mice were subjected to differential gene expression analysis. Probes were considered significant if they met the criteria FDR adjusted p < 0.05 and fold-change (FC) cut off > 1.2. The complete list of significant genes is shown in Supplementary Tables 1 and 2. The transcripts for PHGDH and PSAT1 were down regulated (FDR adjusted p-value 0.027 and 0.104, respectively) by at least 1.2-fold in day 7 B16F10L tumor homograft from A-I Tg+/– relative to A-I KO mice. 3PG: 3-Phosphoglycerate; PHGDH: 3-Phospho-glycerate dehydrogenase; P-PYR: 3-Phospho-hydroxypyruvate; Gln: Glutamine; Glu: Glutamate; aKG: a-Ketoglutarate; PSAT1: phosphoserine aminotransferase 1; P-Serine: Phosphoserine; PSPH: Phosphoserine phosphatase.

To gain further insight into molecular mechanisms that might impart a growth advantage to tumors in the absence of host apoA-I, we compared the differentially-regulated genes listed in Supplementary Tables 1 and 2 to publically available, as well as proprietary annotated, melanoma databases for functional analyses (see Methods). Initially, we queried our list in a supervised fashion to an assortment of key biological processes relevant to cancer. These are shown in Figure 4, and the genes on this list that associated with these processes are listed in Table 2. Some genes, for example Dicer, Adamts1, and Cdc20, which were all downregulated in A-I Tg+/–, mapped to more than one biological process.

Figure 4

Genes from differential analysis of B16F10L tumors from A-I Tg+/– versus A-I KO mapped to key cancer related biological processes.

The significant genes from differential analysis of day 7 B16F10L tumors from A-I Tg+/– versus A-I KO mice (listed in Supplementary Tables 1 and 2) were subjected to bioinformatics analysis. The mapping of genes to key cancer-related processes was obtained from disparate functional annotation data sources mentioned in the Methods section.

Table 2

Genes from differential analysis of B16F10L tumors from A-I Tg+/– versus A-I KO mapped to key cancer related biological processes

Biological Processes	Up-regulated genes in A-ITg	Down-regulated genes in A-ITg
Proliferation	Axin2, Cd44, Cd80, Cdk2, Cxcr4, Dnajb1, Dusp1, Enpp2, Fbln1, Hspa1a, Irx3, Irx5, Klf9, Ltbp3, Mapt, Mitf, Mknk2, Nfat5, Osmr, Peli1, Per1, Phlda1, Pik3r3, Ptprs, Sort1, Srebf1, Srgn, St6gal1, Tnfaip3, Tob2, Tyr	Adamts1, Apex1, Cdc20, Cirbp, Dhcr24, Dicer1, Eif5a, Exosc4, Fadd, Gorasp2, Mbp, Mcm3, Mvd, Phgdh, Slc19a1, Tmpo, Vti1a
Apoptosis	Axin2, Cd44, Cdk2, Cxcr4, Dnajb1, Dusp1, Enpp2, Fbln1, Hspa1a, Ltbp3, Luc7l3, Map3k12, Mapt, Mitf, Nfat5, Peli1, Per1, Phlda1, Pik3r3, Ptprs, Snca, Sort1, Srgn, St6gal1, Tnfaip3, Tyr	Alg5, Apex1, Cdc20, Dhcr24, Dhx9, Dicer1, Eif5a, Fadd, H47, Lims2, Mbp, Slc19a1, Tcte3, Ubqln1
Inflammation	C2, Cd44, Cdk2, Cxcr4, Dusp1, Osmr, Per1, Srebf1, Srgn, St6gal1, Tnfaip3, Tyr	Dicer1, Fadd, H47, Mbp, Stx18
Immune response	C2, Cd44, Cd80, Cdk2, Cxcr4, Dusp1, Enpp2, Hspa1a, Klf9, Mitf, Osmr, Peli1, Per1, Snca, Srebf1, St6gal1, Tnfaip3, Tob2, Tyr	Dicer1, Fadd, H47, Mbp, Slc19a1
Angiogenesis	Robo4, Cd44, Cxcr4	Dicer1, Wars
Metastasis	C2, Cd44, Cxcr4, Dusp1, Enpp2, Fbln1, Map3k12, Mapt, Nfat5, Pik3r3, Robo4, Sema6d, Sort1, Srebf1, St6gal1, Tnfaip3	Apex1
Chemotaxis	Cd44, Cytip, Enpp2, Ptprs, Dusp1	Pcdh17
Leukocyte infiltration	Cd44, Cxcr4, Cytip, Mitf, Pik3r3, Snca	Dicer1
Cancer	Cd44, Cdk2, Cxcr4, Enpp2, Klf9, Mitf, Osmr, Pik3r3, Srgn, Tnfaip3, Tyr	Adamts1, Cdc20, Cirbp, Dicer1, Fadd, Insig1, Mcm3, Mvd, Nme4, Slc19a1, Sqle, Srm
Stress response	Axin2, C2, Cd44, Dnajb1, Hspa1a, Insig2, Luc7l3, Peli1, Snca, Srebf1, Tnfaip3	Apex1, Cirbp, Dhx9, Dicer1, Fadd, H47, Insig1, Mpv17, Rad51ap1, Sfrs5, Ubqln1, Wfdc12
DNA damage/repair	Cdk2, Dnajb1, Dusp1, Hspa1a, Map3k12, Srgn	Apex1, Cdc20, Dicer1, Mcm3, Nme4, Nup107, Slc19a1, Tmpo
Cell differentiation	Axin2, Cd44, Cxcr4, Erdr1, Irx3, Irx5, Ltbp3, Mapt, Mitf, Robo4, Sema6d, Sort1, Tob2	AI428936, Camk1, Cdc20, Dicer1, Eif5a, Insig1, Mbp, Nle1, Phgdh, Stk25
Transport	Acsl3, Axin2, Cd44, Cdk2, Cxcr4, Elovl5, Insig2, Mapt, Ptprs, Snca, Sort1, Srebf1, Srgn	Apex1, Camk1, Dhcr24, Dync1i1, Eif5a, Fadd, H47, Il1rl1l, Insig1, Mbp, Mybbp1a, Nup107, Phgdh, Preb, Rhobtb3, Slc19a1, Snf8, Srpr, Stard4, Stard5, Stx18, Tmed4, Vti1a, Yif1a, Zfpl1
Other Processes	2700097009Rik, Acss2, Ahnak2, Bace2, Clmn, Crem, Dkk3, Efcab6, Eif4a2, Fbxo30, Gcnt2, Gpr158, Gsta4, Hs3st1, Lekr1, Loc100043671, Loc100046802, Loc677317, Nckap5l, Plekha3, Rdm1, Tex2, Ube2o, Unc119b, Zfp597, Zmynd8	Get4, 1810065E05Rik, 2410002F23Rik, Fam176b, 3010003L21Rik, Trmt61a, Bud31, C85627, Cyp51a1, Dus1l, Elovl1, Fbp2, Fbxo36, Gale, Gmpr, Grwd1, Hm13, Hsd17b7, Oc100040592, Loc100043257, Mest, Mpdu1, Mrps18b, Msto1, Nif3l1, Nol12, Nop56, Mak16, Rnf144a, Rnf25, Rwdd3, Sc4mol, Smyd5, Snrpa, Sssca1, Sumo3, Tmem17, Tmem214, Ttc39a, Znf32

List of genes mapped to key cancer-related processes in Figure 4. List of significant genes mapped to processes relevant to cancer.

The Ingenuity Pathway Analysis (IPA) program was then used to generate biological networks from functional relationships between genes or proteins based on the published literature (http://www.Ingenuity.com). It identified lipid and cholesterol synthesis as top relevant pathways for genes downregulated in tumors from A-I Tg+/– mice (Supplementary Figure 1A, 1B and Supplementary Table 3A and 3B). Significantly, five of these genes (Mvd, Sqle, Cyp51, Hsd17b7, and Dhcr24) are within the core biochemical processes of the mevalonate pathway, an important target for anti-cancer therapy [24, 27]. The mevalonate pathway converts acetyl-CoA via mevalonate, the product of the tightly regulated enzyme HMG-CoA reductase and the target of statins, into a variety of end products fundamental to cellular and organismal life (Figure 5).

Figure 5

The mevalonate pathway.

The transcripts for enzymes in boxes were down regulated (FDR adjusted p < 0.05) by at least 1.2-fold in day 7 melanoma tumor homograft from A-I Tg+/– relative to A-I KO mice.

We next tested whether or not a dose-dependent expression correlation existed between each of Phgdh, Mvd, Sqle, Cyp51, Hsd17b7, Dhcr24 and HDL-c levels. Figure 6 shows a significant inverse association between transcript levels of all six genes and HDL-c consistent with the notion that higher circulating apoA-I (HDL-c) levels inhibit expression of these genes in the developing tumor.

Figure 6

Inverse correlation between circulating HDL-c levels in tumor-bearing mice and tumor transcript levels of Phgdh, the first enzyme of serine synthesis pathway as well as enzymes of the mevalonate pathway.

The gene chip expression signal of the indicated genes after positive background correction, log2 transformation, and quantile normalization was plotted against HDL-c levels (see Methods) in tumor bearing mice. Each data point represents the transcript signal from one tumor-bearing animal. The p-value were calculated with Spearman’s rank correlation and analyses were performed using R 3.5.3.

Gene expression profiles in tumors from A-I KO are concordant with human melanoma gene signature for adverse outcome

We previously reported that apoA-I therapy was effective against human melanoma in nude mice [16]. We next compared the gene expression data mined in this study with published human melanoma gene signatures [8] to determine whether genes known to track with poor prognosis in humans are concordant with those found in tumors in the A-I KO mice, the more aggressive tumor setting in our mouse model. Human microarray studies [8] identified the expression profile of genes associated with 4-year distant metastasis-free survival, metastasis, or death among 58 patients with 4-yr follow up. Of the 250 distinct genes identified by this human study, a total of 170 were eligible for comparison in our mouse study; the remaining 80 genes were not present on the mouse array. Of the 170 genes under consideration, 40 did not give any detectable signal on our array, leaving 130 genes for comparison with the expression dataset in this study. We interrogated the eligible genes in our day 7 gene expression dataset with respect to fold-change (A-I KO/A-I Tg+/–) in expression signal normalized to beta-2-microglobulin with a p-value cut off < 0.05 (Student’s t-test). Remarkably, 45% (58) of human genes reported to associate with poor prognosis were also concordant in tumors from A-I KO animals with only 3% (4 genes) going in the opposite direction. We observed no significant difference between A-I KO and A-I Tg+/– in the remaining 52% (68 genes). Table 3 lists the genes compared in the human and mouse study and their expression trend. Table 4 maps the genes to top functions in IPA: cell cycle, DNA replication and repair, nucleic acid metabolism, and cellular assembly and organization as well as cancer. The genes listed in Table 3 were also observed to be differentially regulated in two independent datasets, Riker and Talantov studies [7, 40] from the Oncomine datasets of cutaneous melanoma versus normal (Table 5). Forty-one (41) out of 58 concordant genes in Table 3 were observed to be differentially regulated (> 2-fold change) and statistically significant (p < 0.01) in these independent datasets. Of the genes upregulated in A-I KO (40 out of 41), 12 were concordant in both Riker and Talantov studies [7, 40] and 27 were concordant in either Riker or Talantov [7, 40] (Table 5). Only one gene, the nuclear hormone receptor-interacting protein nuclear receptor coactivator 6 (Ncoa6), was non-concordant, being upregulated in our dataset but downregulated in the Talantov study (Table 5). Gene F10, which encodes the vitamin K-dependent coagulation factor X of the blood coagulation pathway, was downregulated in our study and was also repressed in the Talantov data set (Table 5), as were the human melanoma signatures [8]. These comparisons with three distinct human melanoma studies, as well as the observed high degree of concordance of genes both upregulated or downregulated in melanomas associated with adverse outcomes with published human melanoma gene signatures, validate the mouse gene expression dataset results presented here.

Table 3

A significant number of genes associated with poor prognosis in human melanoma were also concordant in B16F10L tumors from A-I KO mice

PROBE_ID	SYMBOL	DEFINITION	FOLD CHANGE A-IKO/A-ITg	p-value	Expression associated with adverse prognosis in humans
ILMN_1231392	Melk	maternal embryonic leucine zipper kinase (Melk), mRNA.	1.5	0.002	Up regulated
ILMN_2804444	Pcdh17	protocadherin 17 (Pcdh17), mRNA.	1.5	0.001	Up regulated
ILMN_2741050	Rfc4	replication factor C (activator 1) 4 (Rfc4), mRNA.	1.4	0.013	Up regulated
ILMN_2757224	Srp19	signal recognition particle 19 (Srp19), mRNA.	1.4	0.005	Up regulated
ILMN_2711112	Shcbp1	Shc SH2-domain binding protein 1 (Shcbp1), mRNA.	1.4	0.001	Up regulated
ILMN_2878355	Kpna2	karyopherin (importin) alpha 2 (Kpna2), mRNA.	1.4	0.005	Up regulated
ILMN_2607926	Kdelr2	KDEL (Lys-Asp-Glu-Leu) endoplasmic reticulum protein retention receptor 2 (Kdelr2), mRNA.	1.4	0.002	Up regulated
ILMN_2714565	Rrm2	ribonucleotide reductase M2 polypeptide	1.4	0.002	Up regulated
ILMN_2756008	Nasp	nuclear autoantigenic sperm protein (histone-binding)	1.4	0.002	Up regulated
ILMN_2816754	Ndc80/Kntc2	NDC80 homolog, kinetochore complex component (S. cerevisiae) (Ndc80), mRNA.	1.4	0.006	Up regulated
ILMN_2971845	Dtymk	deoxythymidylate kinase (Dtymk), mRNA.	1.4	0.006	Up regulated
ILMN_2588362	Cdca8	cell division cycle associated 8 (Cdca8), mRNA.	1.4	0.003	Up regulated
ILMN_2934457	Ran	RAN, member RAS oncogene family (Ran), mRNA.	1.4	0.004	Up regulated
ILMN_2688944	Htra2/Prss25	HtrA serine peptidase 2 (Htra2), nuclear gene encoding mitochondrial protein, mRNA.	1.4	0.001	Up regulated
ILMN_1214319	Gemin6	gem (nuclear organelle) associated protein 6 (Gemin6), mRNA.	1.4	0.001	Up regulated
ILMN_2632712	Birc5	baculoviral IAP repeat-containing 5 (Birc5), transcript variant 1, mRNA.	1.4	0.006	Up regulated
ILMN_2999654	Psmc3ip	proteasome (prosome, macropain) 26S subunit, ATPase 3, interacting protein (Psmc3ip), mRNA.	1.4	0.003	Up regulated
ILMN_2780177	Mrps5	mitochondrial ribosomal protein S5 (Mrps5), nuclear gene encoding mitochondrial protein, mRNA.	1.4	0.005	Up regulated
ILMN_1244296	Cdc14b/Znf367	CDC14 cell division cycle 14 homolog B (S. cerevisiae) (Cdc14b), mRNA.	1.4	0.024	Up regulated
ILMN_1257552	Tars	threonyl-tRNA synthetase (Tars), mRNA.	1.4	0.008	Up regulated
ILMN_2621422	Kirrel	kin of IRRE like (Drosophila) (Kirrel), mRNA.	1.3	0.026	Up regulated
ILMN_2982965	Prim2	DNA primase, p58 subunit (Prim2), mRNA.	1.3	0.005	Up regulated
ILMN_2598852	Ranbp1	RAN binding protein 1 (Ranbp1), mRNA.	1.3	0.007	Up regulated
ILMN_3097131	Timeless	timeless homolog (Drosophila) (Timeless), transcript variant 2, mRNA.	1.3	0.013	Up regulated
ILMN_2595846	Surf4	surfeit gene 4 (Surf4), mRNA.	1.3	0.003	Up regulated
ILMN_2639036	Hspd1	heat shock protein 1 (chaperonin) (Hspd1), mRNA.	1.3	0.011	Up regulated
ILMN_2745005	Gpn3/MGC14560	GPN-loop GTPase 3 (Gpn3), mRNA.	1.3	0.005	Up regulated
ILMN_2619671	Mrps16	mitochondrial ribosomal protein S16 (Mrps16), nuclear gene encoding mitochondrial protein, mRNA.	1.3	0.011	Up regulated
ILMN_2787871	Cdca5	cell division cycle associated 5 (Cdca5), mRNA.	1.3	0.008	Up regulated
ILMN_2768984	Anln	anillin, actin binding protein (Anln), mRNA.	1.3	0.032	Up regulated
ILMN_2484707	Tyms	thymidylate synthase (Tyms), mRNA.	1.3	0.013	Up regulated
ILMN_2919433	Cdc45l	cell division cycle 45 homolog (S. cerevisiae)-like (Cdc45l), mRNA.	1.3	0.043	Up regulated
ILMN_1218967	Kif2c	kinesin family member 2C (Kif2c), mRNA.	1.3	0.015	Up regulated
ILMN_1233857	Mcm6	minichromosome maintenance deficient 6 (MIS5 homolog, S. pombe) (S. cerevisiae) (Mcm6), mRNA.	1.3	0.012	Up regulated
ILMN_2749937	Ncoa6	nuclear receptor coactivator 6 (Ncoa6), mRNA.	1.3	0.011	Up regulated
ILMN_2797642	Ncaph/Brrn1	non-SMC condensin I complex, subunit H (Ncaph), mRNA.	1.3	0.013	Up regulated
ILMN_2637203	Rfc5	replication factor C (activator 1) 5 (Rfc5), mRNA.	1.3	0.022	Up regulated
ILMN_2758690	Mrps17	mitochondrial ribosomal protein S17 (Mrps17), mRNA.	1.3	0.012	Up regulated
ILMN_2706882	Donson	downstream neighbor of SON (Donson), mRNA.	1.3	0.028	Up regulated
ILMN_2698282	Fusip1	FUSiinteracting protein	1.3	0.010	Up regulated
ILMN_2680648	Atad2	ATPase family, AAA domain containing 2 (Atad2), mRNA.	1.3	0.026	Up regulated
ILMN_2983686	Trub2	TruB pseudouridine (psi) synthase homolog 2 (E. coli) (Trub2), mRNA.	1.3	0.017	Up regulated
ILMN_2657844	Cdc2a	cell division cycle 2 homolog A (S. pombe) (Cdc2a), mRNA.	1.3	0.021	Up regulated
ILMN_2605890	Tk1	thymidine kinase 1 (Tk1), mRNA.	1.3	0.036	Up regulated
ILMN_2677595	Ncapg2/Mtb	non-SMC condensin II complex, subunit G2 (Ncapg2), mRNA.	1.2	0.022	Up regulated
ILMN_2608933	Tcof1	Treacher Collins Franceschetti syndrome 1, homolog (Tcof1), mRNA.	1.2	0.030	Up regulated
ILMN_1233065	Rbmx	RNA binding motif protein, X chromosome (Rbmx), mRNA.	1.2	0.053	Up regulated
ILMN_3163044	Ola1/PTD004	Obg-like ATPase 1 (Ola1), transcript variant 1, mRNA.	1.2	0.016	Up regulated
ILMN_1236574	Cenpa	centromere protein A (Cenpa), mRNA.	1.2	0.016	Up regulated
ILMN_3162184	AU014645/Ncbp1	expressed sequence AU014645 (AU014645), mRNA.	1.2	0.010	Up regulated
ILMN_1221067	Nme1	non-metastatic cells 1, protein (NM23A) expressed in (Nme1), mRNA.	1.2	0.006	Up regulated
ILMN_2830661	Top2a	topoisomerase (DNA) II alpha (Top2a), mRNA.	1.2	0.025	Up regulated
ILMN_2975640	Snrpg	small nuclear ribonucleoprotein polypeptide G (Snrpg), mRNA.	1.2	0.013	Up regulated
ILMN_1245757	Nans	N-acetylneuraminic acid synthase (sialic acid synthase) (Nans), mRNA.	1.2	0.027	Up regulated
ILMN_2717172	Dph3/Zcsl2	DPH3 homolog (KTI11, S. cerevisiae) (Dph3), transcript variant 2, mRNA.	1.2	0.043	Up regulated
ILMN_2936427	Mcm4	minichromosome maintenance deficient 4 homolog (S. cerevisiae) (Mcm4), mRNA.	1.2	0.050	Up regulated
ILMN_1242622	Crem	cAMP responsive element modulator (Crem), mRNA.	0.7	0.002	Up regulated
ILMN_2631423	H2-Ab1/HLA-DQB1	histocompatibility 2, class II antigen A, beta 1 (H2-Ab1), mRNA.	0.6	0.005	Down regulated
ILMN_2688912	F10	coagulation factor X	0.4	0.012	Down regulated
ILMN_1258462	Hoxa9	homeo box A9 (Hoxa9), mRNA.	1.5	0.031	Down regulated
ILMN_2655373	Siat7f	CMP-NeuAC:(b)-N-acetylgalactosaminide (a)2,6-sialyltransferase member VI	1.5	0.004	Down regulated
ILMN_1232182	Ctnnbip1	catenin beta interacting protein 1 (Ctnnbip1), mRNA.	1.3	0.020	Down regulated

One-hundred-thirty (130) human genes previously identified as part of a gene signature associated with poor prognosis in melanoma [8] were compared with genes identified in this study from the differential analysis of day 7 B16F10L melanoma tumors with respect to fold-change (A-I KO/A-I Tg+/–) in expression signal normalized to beta-2-microglobulin with a p-value cut off < 0.05 (Student’s t-test). The human microarray study identified the expression profile of genes associated with 4-year distant metastasis-free survival, metastasis, or death among 58 patients with 4-yr follow up [8]. Of the 250 human genes identified only 130 were eligible for comparison in our current study. Here we provide a list of genes in our study that were compared with the human microarray study. Boxes denote genes that were not concordant in the two studies. Genes with significant difference in expression between A-IKO and A-ITg.

Table 4

Mouse genes differentially regulated in B16F10L tumors from A-I KO and A-I Tg+/– mice and concordant with human genes associated with poor outcome in melanoma were mapped to top functions in IPA

Expression in A-I KO relative to A-I Tg mice	Gene Symbol	Top Functions
↑	Atad2, Birc5, Cdca8, Cdk1, Cenpa, Dtymk, Htra2, Kif2c, Mcmr, Mcm6, Melk, Nasp, Ndc80, Psmc3ip, Ran, Ranbp1, Rrm2, Srp19, Timeless, Tyms	DNA Replication, Recombination and Repair, Cellular Assembly and Organization, Nucleic Acid Metabolism
↓	F10, Hla-dqb1
↑	Anln, Atad2, Cdc45, Cdc14b, Gpn3, Kpna2, Mrps16, Mrps17, Nans, Ncapg2, Ncaph, Rfc4, Shcbp1, Tyms	Cell Cycle, DNA Replication, Recombination and Repair, Cellular Assembly and Organization
↑	Kdelr2, Loc100505793/Fusip1, Ncp1, Ncoa6, Ola1, Prim2, Rbmx, Rfc5, Srp19, Surf4, Tars, Trub2	Cell Cycle, Cancer, Infection Mechanism
↑	Cdca5, Gemin6, Hspd1, Nme1, Rfc4, Rfc5, Snrpg, Tcof1, Tk1, Top2a	Cell Cycle, DNA Replication, Recombination and Repair, Developmental Disorder

Genes whose expression in A-I KO relative to A-I Tg+/– were determined to be concordant with a subset of genes previously identified in a human study [8] to track with poor outcome in melanoma (Table 3), were mapped to top functions in Ingenuity Pathway Analysis (IPA) and are shown in Table 4. Arrows indicate the direction of expression of these genes in day 7 B16F10L tumors from A-I KO relative to A-I Tg+/– animals.

Table 5

Gene expression profiles in tumors from A-I KO are concordant with human melanoma gene signature for adverse outcome

Probe_ID	Gene Symbol	DEFINITION	A-IKO/ A-ITg (Mean)	p-value	Expression (in this study and in Oncomine’s)	Oncomine Datasets- Cutaneous Melanoma vs normal
ILMN_1231392	Melk	Mus musculus maternal embryonic leucine zipper kinase (Melk), mRNA.	1.5	0.002	↑↑	Riker, Talantov
ILMN_2878355	Kpna2	Mus musculus karyopherin (importin) alpha 2 (Kpna2), mRNA.	1.4	0.005	↑↑	Riker, Talantov
ILMN_2714565	Rrm2	Mus musculus ribonucleotide reductase M2 polypeptide	1.4	0.002	↑↑	Riker, Talantov
ILMN_1214319	Gemin6	Mus musculus gem (nuclear organelle) associated protein 6 (Gemin6), mRNA.	1.4	0.001	↑↑	Riker, Talantov
ILMN_2632712	Birc5	Mus musculus baculoviral IAP repeat- containing 5 (Birc5), transcript variant 1, mRNA.	1.4	0.006	↑↑	Riker, Talantov
ILMN_2621422	Kirrel	Mus musculus kin of IRRE like (Drosophila) (Kirrel), mRNA.	1.3	0.026	↑↑	Riker, Talantov
ILMN_2484707	Tyms	Mus musculus thymidylate synthase (Tyms), mRNA.	1.3	0.013	↑↑	Riker, Talantov
ILMN_1218967	Kif2c	Mus musculus kinesin family member 2C (Kif2c), mRNA. XM_986361	1.3	0.015	↑↑	Riker, Talantov
ILMN_1233857	Mcm6	Mus musculus minichromosome maintenance deficient 6 (MIS5 homolog, S. pombe) (S. cerevisiae) (Mcm6), mRNA.	1.3	0.012	↑↑	Riker, Talantov
ILMN_2797642	Ncaph/ Brrn1	Mus musculus non-SMC condensin I complex, subunit H (Ncaph), mRNA.	1.3	0.013	↑↑	Riker, Talantov
ILMN_2677595	Ncapg2/Mtb	Mus musculus non-SMC condensin II complex, subunit G2 (Ncapg2), mRNA.	1.2	0.022	↑↑	Riker, Talantov
ILMN_1245757	Nans	Mus musculus N-acetylneuraminic acid synthase (sialic acid synthase) (Nans), mRNA.	1.2	0.027	↑↑	Riker, Talantov
ILMN_2804444	Pcdh17	Mus musculus protocadherin 17 (Pcdh17), mRNA.	1.5	0.001	↑↑	Riker
ILMN_2688944	Htra2/ Prss25	Mus musculus HtrA serine peptidase 2 (Htra2), nuclear gene encoding mitochondrial protein, mRNA.	1.4	0.001	↑↑	Riker
ILMN_2741050	Rfc4	Mus musculus replication factor C (activator 1) 4 (Rfc4), mRNA.	1.4	0.013	↑↑	Riker
ILMN_1257552	Tars	Mus musculus threonyl-tRNA synthetase (Tars), mRNA.	1.4	0.008	↑↑	Riker
ILMN_3097131	Timeless	Mus musculus timeless homolog (Drosophila) (Timeless), transcript variant 2, mRNA.	1.3	0.013	↑↑	Riker
ILMN_2975640	Snrpg	Mus musculus small nuclear ribonucleoprotein polypeptide G (Snrpg), mRNA.	1.2	0.013	↑↑	Riker
ILMN_2711112	Shcbp1	Mus musculus Shc SH2-domain binding protein 1 (Shcbp1), mRNA.	1.4	0.001	↑↑	Talantov
ILMN_2607926	Kdelr2	Mus musculus KDEL (Lys-Asp-Glu-Leu) endoplasmic reticulum protein retention receptor 2 (Kdelr2), mRNA.	1.4	0.002	↑↑	Talantov
ILMN_2816754	Ndc80/ Kntc2	Mus musculus NDC80 homolog, kinetochore complex component (S. cerevisiae) (Ndc80), mRNA.	1.4	0.006	↑↑	Talantov
ILMN_2971845	Dtymk	Mus musculus deoxythymidylate kinase (Dtymk), mRNA.	1.4	0.006	↑↑	Talantov
ILMN_2588362	Cdca8	Mus musculus cell division cycle associated 8 (Cdca8), mRNA.	1.4	0.003	↑↑	Talantov
ILMN_2934457	Ran	Mus musculus RAN, member RAS oncogene family (Ran), mRNA.	1.4	0.004	↑↑	Talantov
ILMN_2999654	Psmc3ip	Mus musculus proteasome (prosome, macropain) 26S subunit, ATPase 3, interacting protein (Psmc3ip), mRNA.	1.4	0.003	↑↑	Talantov
ILMN_2598852	Ranbp1	Mus musculus RAN binding protein 1 (Ranbp1), mRNA.	1.3	0.007	↑↑	Talantov
ILMN_2639036	Hspd1	Mus musculus heat shock protein 1 (chaperonin) (Hspd1), mRNA.	1.3	0.011	↑↑	Talantov
ILMN_2745005	Gpn3/ MGC14 560	Mus musculus GPN-loop GTPase 3 (Gpn3), mRNA.	1.3	0.005	↑↑	Talantov
ILMN_2919433	Cdc45l/Cdc45	Mus musculus cell division cycle 45 homolog (S. cerevisiae)-like (Cdc45l), mRNA.	1.3	0.043	↑↑	Talantov
ILMN_2749937	Ncoa6	Mus musculus nuclear receptor coactivator 6 (Ncoa6), mRNA.	1.3	0.011	↑↓	Talantov
ILMN_2605890	Tk1	Mus musculus thymidine kinase 1 (Tk1), mRNA.	1.3	0.036	↑↑	Talantov
ILMN_2706882	Donson	Mus musculus downstream neighbor of SON (Donson), mRNA.	1.3	0.028	↑↑	Talantov
ILMN_2680648	Atad2	Mus musculus ATPase family, AAA domain containing 2 (Atad2), mRNA.	1.3	0.026	↑↑	Talantov
ILMN_2657844	Cdc2a/ Cdk1	Mus musculus cell division cycle 2 homolog A (S. pombe) (Cdc2a), mRNA.	1.3	0.021	↑↑	Talantov
ILMN_2608933	Tcof1	Mus musculus Treacher Collins Franceschetti syndrome 1, homolog (Tcof1), mRNA.	1.2	0.030	↑↑	Talantov
ILMN_1233065	Rbmx	Mus musculus RNA binding motif protein, X chromosome (Rbmx), mRNA.	1.2	0.053	↑↑	Talantov
ILMN_1236574	Cenpa	Mus musculus centromere protein A (Cenpa), mRNA.	1.2	0.016	↑↑	Talantov
ILMN_1221067	Nme1	Mus musculus non-metastatic cells 1, protein (NM23A) expressed in (Nme1), mRNA.	1.2	0.006	↑↑	Talantov
ILMN_2830661	Top2a	Mus musculus topoisomerase (DNA) II alpha (Top2a), mRNA.	1.2	0.025	↑↑	Talantov
ILMN_2936427	Mcm4	Mus musculus minichromosome maintenance deficient 4 homolog (S. cerevisiae) (Mcm4), mRNA.	1.2	0.050	↑↑	Talantov
ILMN_2688912	F10	Mus musculus coagulation factor X	0.4	0.012	↓↓	Talantov

Forty-one (41) out of 58 genes (from the comparison analysis of our experiment [ten A-I KO vs ten A-I Tg+/– tumors] with human biopsies studies that associated with adverse prognosis) were observed to be differentially regulated (> 2-fold) and statistically significant (p ≤ 0.01) in Cutaneous Melanoma vs Normal samples (from either both or one of the two datasets -Riker and Talantov) in Oncomine database. The Riker (Human Genome U133 Plus 2.0 Array, mRNA [40];) has 14 melanoma and 4 normal samples; Talantov (Human Genome U133A Array, mRNA [7];) has 45 melanoma and 7 normal samples.

DISCUSSION

In this study, we used DNA microarray technology to analyze the tumor microenvironment for transcriptomic patterns that might provide insight as to why the growth and metastasis of B16F10L melanoma was attenuated in apoA-I-expressing mice compared to A-I null animals [16]. We identified 535 transcripts that were differentially expressed in A-I KO and A-I Tg+/– tumors. Of these, 176 transcripts showed statistically significant differences between the two groups (adjusted FDR p < 0.05) with at least a 1.2-fold change in expression. We compared our gene expression dataset with published melanoma gene signatures and determined that the expression of 45% of genes associated with poor prognosis in human studies were concordant in tumors from A-I KO relative to A-I Tg+/–, thus validating our dataset for mechanistic mining purposes.

Cancer cells adapt their metabolism to fuel their increased proliferation and requisite need for biomass. Figure 3 illustrates that both glucose and glutamine are required for the de novo serine/glycine synthesis pathway. Several chemotherapeutic agents target this pathway, underscoring the significance of serine, glycine and one-carbon metabolism for tumor progression [41]. In this study, we identified the de novo serine synthesis pathway as a potential target for apoA-I anti-neoplastic activity. Phgdh, the gene encoding the first enzyme of this pathway, was the most repressed transcript in tumors recovered from A-I Tg+/– mice relative to A-I KO mice (Table 1B and Supplementary Table 2). This finding suggests that within the tumor cell and tumor microenvironment, apoA-I may inhibit a major metabolic pathway branching off the glycolytic cascade in melanoma.

In humans, the chromosomal locus of Phgdh (1p12) is amplified in a subset of melanoma and breast cancers (estrogen receptor-negative), and recent reports using an in vivo shRNA library screen and metabolomic profiling support the notion that increased expression of PHGDH promotes tumor growth trending with poor outcome [42, 43]. Importantly, shRNA against Phgdh suppressed tumor growth in a human breast xenograft model [43]. Thus, inhibition of PHGDH in cancers overexpressing this key enzyme affects glutamine flux into the TCA cycle as well as serine synthesis leading to growth arrest.

Increased copy number of the Phgdh gene in mice led to increased PHGDH expression and promoted the development of melanoma and breast cancer cells [44]. PHGDH was also recently reported to have a non-metabolic role in glioma tumorigenesis via stabilization of the transcription factor FOXM1 [45]. The expression of PHGDH (mRNA and protein), which is absent in normal brain tissue, was increased in gliomas as a direct function of tumor grade, with more aggressive tumors expressing higher levels of this protein [45]. Patient survival (5-year) within grade III and IV groups was 43.3% in the low-PHGDH expression group compared with 18.5% in the high-PHGDH expression group (p < 0.001), underscoring PHGDH as a prognostic marker for glioma [45]. Recent reports also identify FOXM1 as being overexpressed in metastatic melanoma [46, 47], and its targeting led to apoptosis in animal models of melanoma and in melanoma cells in culture [48, 49]. In our current study, FOXM1 was not differentially expressed in tumor tissue from A-I KO vs A-I Tg hosts. However, this does not rule out a role for this transcription factor in concert with PHGDH to effect melanoma progression in our animal model.

In this study, we also identified the mevalonate pathway as a potential target for apoA-I anti-tumor activity (Figure 5). Transcripts for five enzymes in this pathway were downregulated in tumors from A-I Tg+/– mice (Figure 5, enzymes highlighted in box). The mevalonate pathway feeds into several growth promoting pathways including cell signaling, lipid metabolism, cell structure, and nutrient levels [50], and has also been identified as a positive regulator of YAP and TAZ proto-oncogenes, which are targets of the Hippo tumor-suppressor pathway [51, 52]. The transcriptional expression of YAP is driven by the ETS transcript factor GA-binding protein (GABP) [53]. Interestingly, GABP DNA-binding elements were significantly enriched in the promoter region of genes down regulated in this study’s dataset (data not shown). Inactivating mutations in the tumor suppressor gene p53 has also been correlated with increased expression of mevalonate pathway genes [54]. In cancers expressing mutant p53, the latter acts as a coactivator of SREBP (sterol regulatory element-binding protein), and serves as the positive master transcription factor for enzymes of the mevalonate pathway and for fatty acid synthesis [55].

The YAP/TAZ proto-oncogenes promote cancer stemness and metastatic potential [56, 57]. Inhibition of the mevalonate pathway by statins leads to transcriptional repression of the YAP-dependent pro-metastatic gene RHAMM in breast cancer cells expressing mutant p53 [58, 59]. Four (Sqle, Cyp51, Hsd17b7, and Dhcr24) of the five negatively regulated genes in this study’s dataset are in the sterol synthesis arm of mevalonate pathway (Figure 5). The fifth, Mvd, which encodes mevalonate-5-pyrophosphate decarboxylase, is upstream of the branching point (farnesyl-diphosphate; farnesyl-PP in Figure 5) in the pathway. Thus, inhibition of MVD is likely to have a larger impact, as it affects all branches of the mevalonate pathway. The anti-tumor effect of statin therapy via inhibition of the mevalonate pathway appears to be restricted to cancers expressing mutant p53, and in breast cancers is further restricted to the hormone receptor negative subtypes (ER-/PR-) [55]. B16F10, the murine melanoma cell line used in this study expresses mutant p53 [60]. Recently, the ATP-binding cassette transporter A1 (ABCA1), which facilitates the active efflux of lipids and cholesterol to apoA-I thereby initiating HDL biogenesis, was identified as a transcriptional target of wild-type p53, and mediator of p53 inhibition of the mevalonate pathway via SREBP-2 maturation, in a murine model of liver cancer [61]. Whether apoA-I mediates mevalonate pathway inhibition through the p53/SREBP arm in melanoma remains to be determined.

Supplementary Table 2 lists genes that were downregulated in tumors from A-I Tg+/– animals (relative to A-I KO) and encode proteins that included Dicer, Adamts1 and Cdc20. Dicer, an RNase III family member, processes precursor miRNAs to smaller active species. In melanoma, miRNA have been shown to be deregulated in cell lines [62-64], metastatic lesions ([65], and primary clinical samples [66]. Increased Dicer expression was correlated with progression of melanoma from common melanocytic nevus to invasive melanoma [67]. The extracellular protease Adamts1 (a disintegrin and metalloproteinase with thrombospondin motifs 1) has been implicated in promoting cancer [68], and stroma-derived Adamts1 was recently shown to drive both the growth and metastasis of melanoma [69]. ApoA-I likely inhibits expression of Adamts1 in both tumor and stromal cells, though the mechanism remains to be elucidated.

Cdc20, a key cell cycle regulator of spindle checkpoint and mitotic exit, is suppressed by p53 tumor suppressor protein and is frequently up regulated in cancers due to inactivation of p53 [70]. Recently, systemic delivery of Cdc20 siRNA by intraperitoneal injection was shown to inhibit B16F10 melanoma growth in tumor bearing mice [71, 72] and Cdc20 has been proposed as a cancer therapeutic target [72]. In the present study, two separate probes specific for Cdc20 showed statistically significant reduced expression in 11 separate tumors resected from A-I Tg+/– animals relative to A-I KO (Supplementary Table 2).

The emerging theme in this study is that in animals expressing apoA-I, two major metabolic pathways, namely the mevalonate and serine synthesis pathways, are inhibited within the melanoma tumor microenvironment. The products of these pathways provide molecules indispensable for synthesis of essential lipids, proteins, and nucleic acids, and inhibitors of the mevalonate pathway have been shown to suppress tumor development and metastasis in both mouse [73, 74] and human melanoma cells [75, 76]. The mechanism (s) by which increased levels of apoA-I in host animals engenders this inhibition remains to be elucidated. The apoA-I receptor ABCA1 was shown to mediate inhibition of the mevalonate pathway via SREBP-2 maturation [61]. ABCA1, ABCG1 (HDL receptor) and SRBP-1 (HDL receptor) are candidates for further investigation as potential mediators of apoA-I anti-tumor activity in B16F10 melanoma.

Finally, recent studies with statin treatment of melanoma cells concluded that inhibition of the mevalonate pathway stimulates melanoma immunogenicity [76] and leads to increased adaptive [77, 78] and innate [75] immune response against the tumor [76]. Our current study comparing A-I Tg+/– and A-I KO tumor bearing mice suggest that apoA-I may target cancer metabolism. Importantly, we observed an inverse correlation between circulating HDL-c and transcript levels of Phgdh and five enzymes of the mevalonate pathway (Mvd, Sqle, Cyp51, Hsd17b7, and Dhcr24) in tumor bed, suggestive of an inhibitory role for apoA-I/HDL in cancer cell metabolism (Figure 6). Thus, apoA-I has emerged as a viable therapeutic participant at the crossroads between cardiovascular disease and cancer, considering the growing evidence of biological and mechanistic overlap between these two main causes of mortality worldwide [79-81].

MATERIALS AND METHODS

All chemicals were from Sigma Chemical (St. Louis, MO) and all solvents were HPLC grade unless otherwise indicated.

Mice

All mouse studies were performed under approved Institutional Animal Care and Use Committee protocols at the Cleveland Clinic. C57BL/6J (wild type (WT) mice), A-I KO, A-I Tg+/+, mice were purchased from Jackson Laboratories and bred at Cleveland Clinic’s Biological Research Unit (BRU).

Plasma HDL-c levels

Blood was drawn at the time of tumor resection by cardiac puncture from anaesthetized (300 mg/kg ketamine (Ketaset (Ketamine HCl Injection, USP) Zoetis 100 mg/mL) plus 30 mg/kg xylazine (AnaSed Injection Akorn Animal Health 100 mg/mL NDC 59399-111-50) mice on day-7 post tumor inoculation. Plasma (EDTA) was separated and diluted in saline for HDL-c measurements on an Architect ci8200 (Abbott Diagnostics, Abbot Park, IL) using the Ultra HDL assay kit (3K33-20).

Tumor cell line

Mouse tumor cell lines B16F10 melanoma were obtained from American Type Culture Collection (ATTC, Bethesda, MD) and cultured in DMEM supplemented with 10% heat-inactivated fetal calf serum (FCS), 2 mM L-glutamine and antibiotic/antimycotic (Invitrogen, Grand Island, NY) at 37°C and 5% CO2 in a humidified atmosphere. B16F10L was isolated from lung metastasis in WT C57BL/6J and a variant expressing luciferase (firefly, Photinus pyralis) was engineered by cotransfecting B16F10L cells with pKCPIRlucBGH5.3 (kind gift from Yan Xu, Indiana University School of Medicine) and pcDNA3 (Invitrogen, 3:1 ratio) using Lipofectamine (Invitrogen) and selected with 1.1 mg/mL G418 (Invitrogen) over 3 wk. Pools of luciferase expressing tumor cells were used for the study.

RNA isolation from tumor tissue

Eight to ten (8–10) week male and female animals were inoculated subcutaneously on both flanks with 105 B16F10L tumor cells per site (2 sites/flank). Tumor tissue from one flank was resected on day 7 post inoculation and stored in RNA later. Tumor tissue (30 mg) was homogenized with zirconium oxide beads using the Bullet blender tissue homogenizer (Next Advance, Troy, NY, USA) and total RNA was prepared (RNeasy Mini kit, Qiagen, Valencia, CA) with on column DNase treatment (Cat# 79254, Qiagen).

Gene expression analysis

Global gene expression patterns were obtained using Illumina gene chips. Total RNA, prepared as described above, was amplified and biotinylated for hybridization to Illumina Murine Beadchips (mouseRef-8 v.2, Illumina, Inc., San Diego, CA) using Illumina TotalPrep RNA Amplification Kit (AMIL1791, Applied Biosystems, Life Technologies) according to the manufacturer’s protocol. Arrays were scanned in an Illumina Bead Station and the images were processed using Illumina Bead Studio software. The complete microarray data have been deposited in the Gene Express Omnibus (GEO) database (GEO: GSE137532). Gene expression profiles were compared using unsupervised hierarchical clustering analysis.

Validation of mouse gene expression set with studies with human melanoma

Gene expression data were analyzed in GenomeStudio. Raw data were quantile-normalized, background corrected, and filtered with a detection p < 0.05 and a 1.2-fold-change in expression between A-I KO and A-I Tg+/–. One-hundred-thirty (130) human genes previously identified as part of a gene signature associated with poor prognosis in melanoma [8] were compared with genes identified in this study from the differential analysis of day 7 B16F10L melanoma tumors with respect to fold-change (A-I KO/A-I Tg+/–) in expression signal normalized to beta-2-microglobulin with a p-value cut off < 0.05 (Student’s t-test). The human microarray study identified the expression profile of genes associated with 4-year distant metastasis-free survival, metastasis, or death among 58 patients with 4-yr follow up [8]. Of the 250 human genes identified, only 130 were eligible for comparison in our current study.

Statistical analysis of gene expression data: unbiased mining for significant genes with differential regulation in A-I Tg+/-vs A-I KO mouse tumors

Raw data were read in GenomeStudio and exported for further data processing and analysis in R software (http://www.r-project.org/). Analyses were done on the data after force-positive background correction, log2 transformation, and quantile normalization. Quality control analysis was performed. Comparisons were made to check differences between each of two groups Tg+/– vs KO and WT vs KO. The comparisons were performed using linear models. Empirical Bayes and other shrinkage methods were used to borrow information across genes. FDR (Benjamini & Hochberg correction,) adjusted p-values were calculated to account for multi-testing. R limma package was used. Genes are significantly expressed if FDR adjusted p< 0.05. Volcano plots were made to show both the statistical significance (y axis, -log10(p- value), the higher the number is, the more significant the test is) and biological significance (x axis, log2(Fold-Change), the further right from the zero point, the more upregulated, the further left, the more down regulated). Heatmap and dendrogram from unsupervised hierarchical clustering analysis were plotted.

Bioinformatics analyses

Functional annotation of statistically significant list of genes, in terms of how they relate to key processes or pathways of interest, were obtained using publicly available genome annotation tools and pathway databases such as KEGG, DAVID, GOTermMapper, GOTermFinder, Mouse Genome Informatics, and the commercial pathway analysis tools Metacore (Metacore™, https://portal.genego.com/) and Ingenuity Pathway Analysis (Ingenuity® Systems, http://www.ingenuity.com/). Commercial pathway databases were used in addition to publicly available databases because they provided more comprehensive functional information for these genes, not only in the context of physiological processes and pathway but also their molecular interaction networks, drug targets, diseases, and other vital information. For genes that did not map to any known pathways, a map of proteins interacting upstream or downstream of these genes were obtained from these commercially available pathway analysis. The data discussed in this publication have been deposited in NCBI′s Gene Expression Omnibus (Edgar et al., 2002) and are accessible through GEO Series accession number GSE137532, https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc= GSE137532).