Interception Targets of Angelica Gigas Nakai Root Extract versus Pyranocoumarins in Prostate Early Lesions and Neuroendocrine Carcinomas in TRAMP Mice

Because of its slow growth, prostate cancer is considered highly amenable to chemoprevention and interception with botanical and herbal natural products. Angelica gigas Nakai (AGN) root is used in traditional herbal medicine in Korea, China, and other Asian countries (1). We and others have shown that the major pyranocoumarin compounds in AGN root alcoholic extract decursin (D) and its structural isomer decursinol angelate (DA) are rapidly metabolized to decursinol in laboratory rodent models (2, 3) and in humans (4). We and others have reported anticancer activities of AGN extract and D/DA in animal models against a number of cancers, especially prostate cancer (5–7).

The TRAMP is a prostate carcinogenesis model generated by utilizing a rat probasin (Pbsn) promoter to drive prostate expression of simian virus 40 (SV40) T/t antigens (8). Being the first transgenic animal model of prostate carcinogenesis, the TRAMP mice have been the most utilized model animals in prostate cancer chemoprevention research. Previous studies by us and others have supported the paradigm of two distinct lineages of malignancies in TRAMP: that is, androgen receptor (AR)-positive glandular epithelial lesions termed atypical hyperplasia of T-Ag akin to high-grade prostate intraepithelial neoplasia (PIN), which mostly arise from the dorsal-lateral prostate (DLP) lobes and do not progress to adenocarcinoma; and AR negative poorly differentiated neuroendocrine carcinomas, which arise mostly from the ventral prostate lobes and not from trans-differentiation of the PIN-like lesions (9, 10). Therefore, the “adenocarcinoma” in TRAMP was a misnomer. We have reported that daily gavage treatment from 8 WOA with AGN extract inhibited neuroendocrine carcinoma growth (6) and that AGN or an equimolar D/DA (isolated from AGN as a mixture) suppressed the two lineages of carcinogenesis in the TRAMP mice with differential efficacies (7). In this report, we took advantage of the banked frozen prostate and tumor tissues from the last experiment (7) and quantified their mRNA levels using Illumina Mouse Ref-8 BeadChip expression microarrays to explore potential in vivo targets of AGN extract and equimolar D/DA mixture in the precancerous DLP lesions and in the neuroendocrine carcinomas.

Frozen prostate tissues and tumors (confirmed as neuroendocrine carcinoma by hematoxylin and eosin histology and IHC) for the transcriptomic analyses were from an animal experiment reported previously (7). The animal study had been conducted in accordance with, and with the approval of, the Institutional Animal Care and Use Committee (IACUC) of Texas Tech University Health Sciences Center (Houston, TX). Briefly, groups of male C57BL/6 TRAMP mice (RRID:IMSR_JAX:003135) at 8 WOA were each gavage-treated with 0.15 ml excipient vehicle, AGN (5 mg per mouse) or D/DA (3 mg per mouse, equimolar to that in AGN) once daily, 5 days per week till either 16 or 28 WOA for dissection and collection of prostate. Large palpable prostate tumors in some TRAMP mice necessitated earlier sacrifice than 28 WOA. The tumors were dissected and a portion of each was frozen at −80°C until used for RNA extraction (Veh neuroendocrine carcinomas; AGN neuroendocrine carcinomas). Wild-type (WT) littermate mice were treated by gavage with vehicle only and sacrificed at 16 or 28 WOA for prostate tissue collection as respective baseline reference controls.

Because of the small amount of dissected prostate lobes from each mouse, DLP samples were pooled from the WT littermates or the TRAMP mice that were found free of prostate tumors at necropsy: WT mice treated with vehicle sacrificed at 16 and 28 WOA [Veh WT DLP 16 weeks (n = 6) or 28 weeks (n = 8)]; TRAMP mice treated with vehicle and sacrificed at 16 and 28 WOA [Veh TRAMP DLP 16 weeks (n = 13) or 28 weeks (n = 11)]; TRAMP mice treated with AGN and sacrificed at 16 and 28 WOA [AGN TRAMP DLP 16 weeks (n = 10) or 28 weeks (n = 14)]; TRAMP mice treated with D/DA and sacrificed at 16 and 28 WOA [D/DA TRAMP DLP 16 weeks (n = 11) or 28 weeks (n = 13)]. In addition, we selected 3 neuroendocrine carcinomas each from vehicle-treated mice and AGN-treated mice and 2 WT whole prostates (24–28 WOA) for RNA isolation. RNA from the 16 samples was extracted by using the AllPrep DNA/RNA/Protein Mini kit (QIAGEN Inc.) according to the manufacturer's protocol.

The global mRNA expression profiles were obtained by using the Illumina Mouse Ref-8 BeadChip expression array (Illumina, Inc.). All RNA labeling and hybridization were performed by the BioMedical Genomics Center of the University of Minnesota according to the manufacturer's protocols and as we described previously (6). The relevant microarray data sets have been deposited in Gene Expression Omnibus (GEO) repository (RRID:SCR_005012) at the National Center for Biotechnology Information (NCBI) under accession number GSE164854. The R 3.1.2 software (RRID:SCR_001905; ref. 11) was used to process and analyze the obtained microarray data. The data from different chips were normalized with rank-invariant normalization method using R package lumi 2.18.0 (12, 13) and filtered with Benjamini–Hochberg P < 0.05 using R package limma 3.1 (14). Clustered image map was plotted with McQuitty linkage method for global gene expression profiling using R package gplots 2.17.0 (15). Principal component analysis (PCA) of global gene expression was computed and visualized using R package rgl 0.95.1247 (16). In the comparison analysis, the expression fold-change >1.3 or <0.7 (treatment group/control group) were set as cut-off threshold to identify the differentially expressed genes (DEG) on microarray platform.

The DEG sets were annotated and analyzed using the web-based bioinformatics tool The Database for Annotation, Visualization and Integrated Discovery (DAVID) Bioinformatics Resources version 6.7 (SAIC-Frederick, Inc., Frederick, MD, RRID:SCR_001881) and IMPaLA (Integrated Molecular Pathway Level Analysis) version 9 (Max-Planck-Gesellschaft, München, Germany) for pathway over-representation and enrichment analysis (17, 18).

Gene Ontology (GO) enrichment analysis for biological processes of the DEG sets was performed by uploading into R package topGO 2.18.0 (RRID:SCR_014798; ref. 19). Using Kolmogorov–Smirnov test, the GO terms with enrichment P < 0.01 were considered as significant. The GO biological networks were plotted with R package topGO to exhibit the distribution of significant GO terms over the GO graphs.

Independently, Ingenuity Pathway Analysis (IPA) software (Qiagen, SCR_008653) was also used to perform the core analysis to identify overlapped canonical pathways and biological functions with the following parameters: Ingenuity Knowledge Base (genes only); direct and indirect interactions; interaction networks; experimentally observed confidence intervals; mouse as the selected species. The negative log of the p value of the enrichment score was used to compare the overlapped IPA canonical pathways among 16-week TRAMP DLP/WT, 16-week AGN-treated TRAMP DLP/16-week TRAMP, 16-week D/DA-treated TRAMP DLP/16-week TRAMP, 28-week TRAMP DLP/WT, 28-week AGN-treated TRAMP DLP/28-week TRAMP, and 28-week D/DA-treated TRAMP DLP/28-week TRAMP groups. The negative log of the P value >1.3 was considered significant. Furthermore, the activation z-scores, used as a statistical measure of the match between expected relationship direction and observed gene expression of the uploaded DEG dataset, were applied to compare the overlapped IPA Top Biological Functions among the groups. Z-scores >2 or <−2 were considered significant.

The DEGs were introduced into InnateDB (RRID:SCR_006714) to predict potential hubs, that is, highly connected protein nodes, in the protein–protein interaction (PPI) consensus network (20) using a search algorithm of mapping initial seeds and first-order neighbors. Genes with 10 or more connectivity degrees per node were considered as hubs and their scale-free networks were visualized with Cytoscape 3.2.1 (RRID:SCR_003032; ref. 21).

As we have reported (7), RNA concentration was determined by using the NanoDrop 1000 Spectrophotometer (Thermo Fisher Scientific Inc.), and 1 μg of RNA was used for cDNA synthesis by using the iScript cDNA synthesis kit (Bio-Rad Laboratories, Inc.) according to the manufacturer's protocols. Real-time quantitative reverse transcription PCR (qRT-PCR) was performed by using the Fast Start Universal SYBR Master with ROX (Hoffmann-La Roche Ltd.) on the CFX96 Touch Real-Time PCR Detection Systems (Bio-Rad Laboratories, Inc.) according to the manufacturer's instructions. All reactions were done in duplicate, and the relative expression of target mRNA in each sample was normalized against β-actin expression level. The sequences of the primers were as listed in Supplementary Table S1.

As we reported earlier (7), gavage administration of AGN extract (5 mg per mouse) and an equimolar D/DA (3 mg per mouse) inhibited the growth of TRAMP DLP by 66% and 61%, respectively, at 16 WOA; and by 67% and 72%, respectively, at 28 WOA. In addition, AGN showed a beneficial effect on TRAMP mouse surviving with neuroendocrine carcinoma to 28 WOA but D/DA did not (7). Furthermore, qRT-PCR detection indicated that AGN and D/DA affected select cell-cycle and proliferation genes [Ccna2, Cdkn1a (P21Cip1), Hmgb2, Ink4a (P16), and Pcna] in TRAMP DLP, but AGN affected select genes in neuroendocrine carcinoma related to angiogenesis, epithelial–mesenchymal transition (EMT), invasion and metastasis, and inflammation signatures with a greater scope than equimolar D/DA (7).

Figure 1A showed visualization of the mRNA expression profiles across WT whole prostate (rows 1 and 2), WT DLP (rows 3 and 4), TRAMP DLP (rows 5–10), and neuroendocrine carcinoma (rows 11–16) using all the genes filtered with hybridization density signal detection P value (P < 0.05) after rank-invariant normalization. Not surprisingly, the neuroendocrine carcinoma samples (rows 11–16) showed patterns very different from the WT whole prostate and the DLP samples regardless of WT (rows 1–4) or TRAMP genotype status (rows 5–10), consistent with the distinct malignancy lineages of their respective cellular origins (9, 10). The WT DLP at 16 WOA (row 3) and 28 WOA (row 4) shared with whole prostate (all lobes included; rows 1 and 2) some gene clusters, and differed on others, as would be expected of part vs. the whole prostate. There were age-related differences, as expected, between DLP at 16 WOA versus 28 WOA for both WT and TRAMP mice.

Principal component analysis (PCA) of the 16 samples using the same set of genes further supported that neuroendocrine carcinoma markedly differed in expression profiles from WT prostate and TRAMP DLP epithelial lesions (Fig. 1B). As expected, WT whole prostate versus WT DLP, TRAMP DLP and neuroendocrine carcinoma samples showed substantial separation in a three-dimensional space (Fig. 1B). The PCA clustering also supported three distinct gene expression profiles among WT (whole and DLP) tissue, TRAMP DLP, and neuroendocrine carcinoma (Fig. 1C). Regarding the interception modality impacts on the DLP lesions, AGN-treated TRAMP DLP was the closest to D/DA-treated TRAMP DLP at each time point and both groups were separated from the respective vehicle-treated TRAMP DLP (Fig. 1B and C), supporting that AGN and equimolar D/DA shared many affected genes, yet with subtle differences. For neuroendocrine carcinoma, the clustering was not able to clearly segregate the AGN- versus vehicle-treated tumors (Fig. 1B and C).

The web-based literature mining and pathway enrichment analyses were performed for all DEGs with a fold change (over respective vehicle control) ≥1.3 or ≤0.7. Five top over-represented pathways were found in AGN-treated TRAMP DLP by 16 WOA: fatty acid beta oxidation, binding and uptake of ligands by scavenger receptors, glycerolipid metabolism, extracellular matrix organization, and collagen formation (Fig. 2A). By 28 WOA, AGN affected genes in TRAMP DLP shifted to pathways of metabolism, striated muscle contraction, metapathway biotransformation, the citric acid cycle (TCA) and respiratory electron transport, and extracellular matrix organization (Fig. 2A). For D/DA-treated TRAMP DLP, top over-represented pathways included triacylglyceride synthesis, striated muscle contraction, phase II conjugation, metabolism, biological oxidations, extracellular matrix organization, and metabolism of lipids and lipoproteins at 16 WOA (Fig. 2B), and they were grouped into metabolism, TCA and respiratory electron transport, oxidative phosphorylation, electron transport chain, metapathway biotransformation, and PPAR signaling pathway at 28WOA (Fig. 2B). Within “common” pathway categories, the compositions of the genes affected by AGN or D/DA showed some degree of overlap as well as unique hits to each interception modality as illustrated on heatmaps for 28WK DLP (“metabolism” clusters in Supplementary Fig. S1). However, it was counterintuitive that the purified D/DA compounds impacted more diverse pathways than the AGN extract that provided equimolar D/DA plus the other non-pyranocoumarin components, at both time points (Fig. 2B vs. A), perhaps implicating an “antagonism” by these extra components against D/DA actions.

Complementing the above analyses through online tools, IPA identified “oxidative phosphorylation” as the most affected by AGN and D/DA in the TRAMP DLP at 28 WOA (Supplementary Fig. S2), which would relate to the “metabolism” features above (Supplementary Fig. S1). IPA identified “Hepatic Fibrosis/Hepatic Stellate Cell Activation” biology in TRAMP DLP by 16 and 28 WOA and was modified by AGN and D/DA (Supplementary Fig. S2). The “muscle contraction” and “ECM organization” features affected by AGN or D/DA (Fig. 2) would correspond to this canonical pathway involving an accumulation of extracellular matrix (ECM) proteins from proinflammatory events activating hepatic stellate cells (HSC) to secrete cytokines to stimulate myofibroblasts to synthesize a large amount of collagens to form tissue fibrosis. IPA predicted induction of “deaths/mortality” and reductions of “cell survival/viability”, “motility/migration/movement,” and “invasion” in both AGN- and D/DA-TRAMP DLP, with subtle differences between AGN and D/DA (Supplementary Fig. S3).

In the neuroendocrine carcinoma, AGN mainly affected genes associated with signal transduction (in particular GPCR), immune (innate) systems, cell cycle, extracellular matrix organization, metabolism, and focal adhesion (Fig. 2C), diverging vastly from the TRAMP DLP patterns. Such gene landscape diversities in the neuroendocrine carcinoma versus DLP could result from the difference of two orders of magnitude in the lesion volumes (∼5 g vs. ∼50 mg) and the more complexity in neuroendocrine carcinoma cell type compositions (especially neuroendocrine cell origin), malignant cell proliferation rate, and many tumor microenvironment (TME) factors such as extracellular matrices, tissue architecture, immune and inflammatory cell infiltration, vascularization, oxygenation state, and innervation.

To appreciate the biological functions associated with the DEGs, we performed Gene Ontology (GO) enrichment analysis for the DEG sets from AGN and D/DA treatment for biological process and cellular component (Table 1, see graphic representations in Supplementary Fig. S4). For biological process, the GO terms associated with AGN- and D/DA-affected genes shared 5 common terms, e.g., GO:0051128 (regulation of cellular component organization), GO:0071310 (cellular response to organic substance), GO:0048523 (negative regulation of cellular process) in TRAMP DLP by 16 WOA (Table 1A and B), and 13 common GO terms such as GO:0019216 (regulation of lipid metabolic process), GO:0055114 (oxidation-reduction process), GO:0002684 (positive regulation of immune system process), GO:0045333 (cellular respiration) in TRAMP DLP by 28 WOA (Table 1C and D). By 28 WOA, the top 4 biological processes and cellular components affected by AGN versus D/DA remained remarkably congruent, varying only the statistical ranking orders between the two interception modalities. In the neuroendocrine carcinoma, the AGN-affected genes were mostly associated with neuronal and immune processes: regulation of neuron projection development, positive regulation of cell projection organization, innate immune response, and developmental growth (Table 1E).

As far as the Cell Component GO terms are concerned (Table 1), those enriched in the AGN- and D/DA-treated DLP were largely related to extracellular matrix, responsible for cell adhesion, intercellular communication, and cell differentiation, at both 16- and 28-WOA time points. Additional terms at 28-WOA included mitochondria energy-converting subcellular organelles. On the other hand, those linked to neuroendocrine carcinoma featured neuronal dendrite/axon structures in addition to extracellular matrix, consistent with their neuroendocrine origin (9, 10) and with our integrative omic profiling of neuroendocrine carcinoma from our previous study (6). These data indicated that AGN- and D/DA-regulated genes shared some biological functions in TRAMP DLP, but AGN affected vastly different biological processes and cellular components in neuroendocrine carcinoma.

Protein–protein interaction (PPI) network analysis can inform and identify biologically essential genes/proteins based on the topological placement of a gene/product, that is, more densely connected hub genes are more likely to be essential genes because the changes in key nodes will cause more impact on the network than the ones on marginal position (22). Through PPI analysis, 23 candidate hub genes were predicted in the constructed networks by mapping the DEGs (>30% over or below vehicle) of AGN- or D/DA-treated TRAMP DLP into the interaction databases retrieved from InnateDB (Table 2, see graphic representation in Supplementary Fig. S5A–S5D). Fbxo32 was added to Table 2 and tested due to putative upregulation of other Foxo family genes in neuroendocrine carcinoma microarray readout (Table 3).

At 16 WOA, 1 hub gene was predicted for AGN DLP (Cidea) and 2 in D/DA DLP (Cidea, Ryr1; Table 2). The Cidea gene was downregulated in AGN and D/DA groups at both 16 and 28 WOA, as verified by qRT-PCR (Table 2). The Ryr1 gene expression was down-regulated in D/DA DLP at 16 WOA, but was upregulated in both AGN and D/DA DLP at 28 WOA and the temporal change trends were verified by qRT-PCR (Table 2).

Of 8 predicted downregulated hub genes and verified by qRT-PCR at 28 WOA in the TRAMP DLP, seven, including Cidea, were shared by both AGN and D/DA, therefore attributable to regulation by D/DA, but Ntrk2 was altered by AGN only, likely brought about by the other non-D/DA components in AGN (Table 2). Of 16 candidate upregulated hub genes in the TRAMP DLP at 28 WOA, 4 were shared by AGN and D/DA (Ryr1, Bach1, Flii, and Zfp292) and were confirmed by qRT-PCR (Table 2), attributable to regulation by D/DA. Six of 7 predicted upregulated candidate hub genes in AGN DLP were verified by qRT-PCR (Table 2). Such hub genes could be attributable to regulation by the other non-D/DA AGN components. However, only Sin3a among 5 candidate upregulated hub genes in D/DA DLP 28WOA was verified by qRT-PCR (Table 2). This type of hub gene regulation was potentially through “antagonism” by other non-D/DA components against the D/DA action. The functions of the verified hub genes are discussed later.

Seventeen candidate hub genes meeting the microarray fold cutoff of >30% (over or below vehicle) were predicted in the consensus networks of DEGs from neuroendocrine carcinoma of TRAMP mice treated with AGN (Table 3, see graphic representation in Supplementary Fig. S5E). Among them, 14 genes were down-regulated and confirmed by qRT-PCR (Table 3). Irf8 was added to the list and tested by qRT-PCR because of its known involvement in inflammation and immunity. Of 3 predicted candidate upregulated hub genes, only Foxo1 showed a verifiable AGN-induced increase through qRT-PCR detection (Table 3). The subsequent discussion will therefore focus on the qRT-PCR verified hub genes with key citation information provided by PMID.

Consistent with two distinct prostate carcinogenesis lineages in the TRAMP model (9, 10), heatmap clustering (Fig. 1A) displayed the remarkable differences in mRNA expression patterns of WT whole prostate, WT DLP, and TRAMP DLP from those in the neuroendocrine carcinoma. Similarly, PCA (Fig. 1B and C) not only indicated the individual differences among genotypes and AGN or D/DA interception modalities and treatment duration but also highlighted some similarity between transcriptomic signatures of AGN and D/DA-treated TRAMP DLP. Bioinformatics analyses using public accessible as well as proprietary softwares for DEG sets further highlighted the different signaling pathway networks and cellular components affected by AGN and D/DA in the two lineages of malignancies (Table 1; Fig. 2; Supplementary Figs. S1–S3). The processes and cell components affected by AGN and D/DA in the TRAMP DLP chiefly involved lipid and mitochondrial energy metabolism and oxidation-reduction, while AGN affected those involved in neuronal signaling, immune systems and cell cycle in the neuroendocrine carcinoma. With distinct cell origins and unique biochemical wiring a priori between the two malignancy lineages (9, 10), their sheer lesion mass differential in favor of neuroendocrine carcinoma would be congruent with more diverse and complex cell types and TME involved and affected by AGN interception in the neuroendocrine carcinoma than in the TRAMP DLP.

The hub genes predicted by PPI network approach and verified by qRT-PCR merit elaboration (Tables 2 and 3; Supplementary Fig. S5). Consistent with the two lineages of carcinogenesis in TRAMP mice (9, 10), the DLP and neuroendocrine carcinoma hub genes shared little in their number and the directions of change. For example, Vim was decreased by AGN in both lineages; whereas LmnA was increased by AGN in DLP yet was decreased in neuroendocrine carcinoma.

The downregulated hub genes in the TRAMP DLP by both AGN and D/DA were generally related to oncogenesis or aggressiveness. The Cidea gene stood out because it was down-regulated by both AGN and D/DA treatments at both time points (Table 2). Cidea encodes a transcriptional coactivator linked to lipid metabolism and cancer cachexia (23). In cancer cachexia, increased CIDEA interacts with and inactivates AMP-activated protein kinase, leading to energy wasting through uncoordinated activation of both lipolytic and lipogenic pathways (24). Crem is critical in hepatocyte proliferation and survival, and differentiation of neural precursors (25) although its role in tumorigenesis is still unclear. Gata6 was highly expressed in colorectal and pancreatic cancer, and it regulated androgen-responsive genes (26). Gja1 was associated with the invasiveness of prostate cancer cells (27). Pparg was observed overexpressed in neuroendocrine carcinoma and other cancers and was considered a novel therapeutic target due to its role in cell growth and differentiation (28). Cidea is a target gene of Pparg and therefore their concordant downregulations might be hierarchically coordinated. Expression of Tuba1a and Vim were positively correlated to the invasion and migration potentials of cancer cells (29).

Three of the upregulated hub genes, Bach1, Flii, and Zfp292, in both AGN and D/DA DLP were known transcriptional factors/regulators. Bach1 was found overexpressed in a number of cancers and promoted invasion and metastasis (30). Flii was reported in diverse cancer cells as a transcriptional repressor for ChREBP activity necessary for AR-mediated promotion of transcription of human prostate-specific antigen (PSA) gene KLK3 (31). Zfp292 has recently been reported to be critical for the maintenance of innate lymphoid cells (32). Intriguingly, Ryr1 encoding an intracellular Ca²⁺ channel protein important in maintaining cellular Ca²⁺ homeostasis plays a pivotal role in determining cell survival-versus-death fate (33) and is variably expressed in prostate cancer cell lines (34). Because these afore-discussed hub genes were concordantly regulated by both AGN and equimolar D/DA, their associated actions on cancer energy metabolism, androgen-dependent gene expression, cellular structure, cell proliferation, migration, and invasion (Table 2) should be attributable to D/DA actions, but not the other AGN components.

In addition, we bring attention to the gene expression changes in AGN DLP, but not in D/DA DLP (Table 2). These “AGN-specific” hub molecules were likely mediated by the non-D/DA AGN components and could contribute additional benefits over D/DA at controlling precancerous epithelial lesion cell composition and fate and TME. Ntrk2 gene belongs to the RAF family and is highly expressed in human metastatic prostate cancer tissues. NTRK2 and another RAF family MERTK were considered novel therapeutic targets for prostate cancer bone and visceral metastasis (35). For the 5 upregulated hub genes only in AGN DLP (Table 2), Acta1 and Ckm together with Ryr1 were related to muscle feature. Acta1 was identified to be downregulated in malignant prostate cancer cells (36). The nuclear membrane protein product of Lmna was increased in advanced stage prostate cancer compared with benign controls (37). In the TRAMP and other genetically modified models of prostate carcinogenesis, deficits of muscle expression signatures have been observed in the prostate lesions (38, 39). Defects in Fbxo32 and Lmna were also related to some types of muscular dystrophy. Known to be involved in ubiquitination, epigenetically silencing of Fbxo32 was observed in multiple cancers and its restoration could inhibit tumor proliferation both in vitro and in vivo (40). Although Stat3 was activated in a variety of cancers, including TRAMP neuroendocrine carcinoma (7), recent studies reported a tumor suppressor role of STAT3 in human prostate cancer (predominantly adenocarcinoma), glioblastoma and intestinal tumor. Indeed, low expression of STAT3 was linked to high risk of cancer recurrence and poor outcome in 204 prostate cancer patient samples (41). The TNFα-related death molecule Traf3 was lower in multiple cancers than their respective originating tissues. Traf3 was involved in T-dependent immune response (42) and restraining normal and malignant B cells through suppressing Pim2/c-Myc oncogenes (43). Relevant to modulating PIN-like lesion function, six hub genes in TRAMP DLP, that is, Bach1, Flii, Gja1, Gsk3b, Sin3a, Tuba1a might interact with AR signaling. Flii and Sin3a were repressors of AR transcription as discussed above while cytoplasmic abundance of Gja1 was associated with androgen depletion (44).

Furthermore, it is of mechanistic importance to speculate on Sin3a as the only qRT-PCR verified hub gene in D/DA-TRAMP DLP (Table 2). Formation of a repressor complex among Sin3a, Ebp1, and HDAC2 was shown to suppress E2F1-regulated and AR-regulated genes such as KLK3 (PSA) and inhibit the proliferation of prostate cancer cells (45). Because D/DA was provided to match equimolar to those present in the AGN dosing, the failure to observe increased Sin3a in the AGN DLP might suggest a potential “antagonism” by the other non-D/DA components against the particular D/DA action on this hub gene.

The following discussion will address hub genes affected by AGN in the neuroendocrine carcinoma (Table 3), all except one through downregulation. Best known as an apoptosis executor caspase, Casp7 has also a nonapoptotic function for the regulation of cell cycle at mitosis and silencing of Casp7 could prevent cell proliferation through mitotic arrest (46). Paradoxically, elevated CDKN1A (P21Cip1) expression was associated with increased recurrence in many prostate cancer cases and poor patient survival (47). Consistent with a cancer cell survival role of P21Cip1, neuroendocrine carcinoma genesis in TRAMP model was significantly reduced in Cdkn1a-deleted cross-bred mice (48). Col1a1 encodes the components of type I collagen and its increased level could enhance the invasion and metastasis of prostate cancer (49). eIF3a has been recognized for oncogenic activities in several organ sites and as a potential cancer drug target (50). Grb10 was upregulated in epinephrine-induced neuroendocrine carcinoma, and in prostate cancer specimens; and acted as a major downstream tumor-promoting effector of PI3K because knockdown of Grb10 in a murine model could delay the onset of acute leukemia possibly by interacting with tyrosine kinases such as Bcr-Abl (51). Hmga1encodes a nonhistone chromosomal protein and promotes androgen-independent prostate cancer growth and invasion by responding to inflammatory signals (52, 53). Irf8 is an inflammatory mediator in macrophage function and other myeloid cell functions to contribute to the well-known immune suppressive TME of neuroendocrine carcinoma. The histone acetyl transferase Kat5 (Tip60) promotes growth, migration, and invasion of prostate cancer cells (37, 54).

One prominent feature was that a third of the neuroendocrine carcinoma hub genes were involved in neurogenesis or signaling (Foxo1, Ldb1, Lmna, Ndn, Snca, Socs3). LDB1 promotes growth and metastasis of human head and neck cancer and maintains the self-renewal of invasive breast carcinoma cells (55). Lmna/c overexpression or knockdown impacted epithelial-to-mesenchymal transition (EMT) biomarkers in a cell model by direct regulation of β-catenin (56). Ndn is a downstream target gene of Stat3 and TP53. Ndn was upregulated in metastatic neuroendocrine cells from the prostates of CR2-Tag mice and promoted tumor growth possibly through delaying P53 responses (57, 58). Ppp1ca is a recognized potential oncogene for prostate cancer (59, 60). Genomic amplification of the PPP1CA gene was highly enriched in metastatic human prostate cancer and enhanced S6K/PP1α/B-Raf–MAPK signaling (61). Snca is primarily located in presynaptic terminals of brain neurons and its abnormal oligomerization into insoluble fibrils is thought to be a hallmark of Parkinson disease. Its expression in the TRAMP neuroendocrine carcinoma is likely a reflection of the neuroendocrine origin of the cancer. Socs3 was highly expressed in castration resistant prostate cancer specimens and its downregulation led to apoptosis of prostate cancer cells (62). Socs3 plays a major role in inflammation and infection signaling, likely through JAK kinase and STAT3 signaling (63). The overexpression of Vim is positively related to the motility, invasion, and metastasis of various tumors including androgen-dependent and -independent prostate cancer, breast cancer, malignant melanoma, and is a driver for EMT (64).

For the lone upregulated hub gene Foxo1 in neuroendocrine carcinoma by AGN, loss of this transcriptional factor gene cooperates with the prostate specific fusion gene TMPRSS2-ERG to drive prostate carcinogenesis and invasion (65). Foxo1 suppressed the invasion and migration of prostate cancer cells by inhibiting Runx2 expression (66) possibly through Cdkn1a/P21Cip1 expression (67). Taken together, AGN might exert it inhibitory effect against neuroendocrine carcinoma through a combination of anti-proliferation, anti-inflammatory, and antimetastatic activities as well as induction of tumor suppressor genes, reflected from the above discussed hub genes and the pathway enrichment analysis (Fig. 2C). Such a conclusion is consistent with our previous reported omics profiling of AGN-affected neuroendocrine carcinoma (6) and targeted survey detection of signature genes with qRT-PCR (7).

Our transcriptomic study has some caveats and limitations. First, due to the dearth of tissue, pooled DLP tissue samples were used and therefore no individual variability data for statistical comparison. Second, the DLP tissues and neuroendocrine carcinoma specimens were analyzed without further microdissection, therefore, they contained many cell/tissue types that made up the whole lesion or tumor. The transcriptomic data should be appreciated and interpreted with no knowledge of the specific cell types of origins. Furthermore, there was the inherent pitfall of the cDNA microarray for underestimating extent of transcript abundance. Compared with qRT-PCR, some 20% to 30% of signals estimated from microarray would be severely underestimated, as shown here in Tables 2 and 3 as well as in our previous publications (6, 39).

In conclusion, profiling DEGs and hub genes in TRAMP DLP lesions and neuroendocrine carcinoma in response to interception by AGN or equimolar D/DA informed potential molecular targets in malignancy lineage-specific manners. The broader scope of hub gene expression regulations by AGN than D/DA in the TRAMP DLP reinforced a similar conclusion that we had obtained in the neuroendocrine carcinoma through focused qRT-PCR detection of signature genes (7). These findings implicate the importance and contributions of other non-D/DA components toward efficacy and survival benefit and merit further mechanistic investigations using appropriate animal models for better translatability to human patients with cancer and high-risk populations.

No disclosures were reported.

S.-N. Tang: Conceptualization, data curation, validation, investigation, methodology, writing–original draft. P. Jiang: Conceptualization, data curation, software, validation, investigation, visualization, methodology, writing–original draft. S. Kim: Data curation, software, visualization, methodology, writing–original draft. J. Zhang: Conceptualization, resources, software, investigation, methodology. C. Jiang: Resources, supervision, project administration, writing–review and editing. J. Lü: Conceptualization, resources, supervision, funding acquisition, project administration, writing–review and editing.

This work was supported by R01 AT007395 grant from National Center for Complementary and Integrative Health (NCCIH; to J. Lu) and R01 CA172169 grant from US NCI (to J. Lu), and Penn State College of Medicine Start-up fund (to J. Lu).

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.