Genomic Characterization of Human and Rat Prolactinomas

Abstract

Although prolactinomas can be effectively treated with dopamine agonists, about 20% of patients develop dopamine resistance or tumor recurrence after surgery, indicating a need for better understanding of underlying disease mechanisms. Although estrogen-induced rat prolactinomas have been widely used to investigate the development of this tumor, the extent that the model recapitulates features of human prolactinomas is unclear. To prioritize candidate genes and gene sets regulating human and rat prolactinomas, microarray results derived from human prolactinomas and pituitaries of estrogen-treated ACI rats were integrated and analyzed. A total of 4545 differentially expressed pituitary genes were identified in estrogen-treated ACI rats [false discovery rate (FDR) < 0.01]. By comparing pituitary microarray results derived from estrogen-treated Brown Norway rats (a strain not sensitive to estrogen), 4073 genes were shown specific to estrogen-treated ACI rats. Human prolactinomas exhibited 1177 differentially expressed genes (FDR < 0.05). Combining microarray data derived from human prolactinoma and pituitaries of estrogen-treated ACI rat, 145 concordantly expressed genes, including E2F1, Myc, Igf1, and CEBPD, were identified. Gene set enrichment analysis revealed that 278 curated pathways and 59 gene sets of transcription factors were enriched (FDR < 25%) in estrogen-treated ACI rats, suggesting a critical role for Myc, E2F1, CEBPD, and Sp1 in this rat prolactinoma. Similarly increased Myc, E2F1, and Sp1 expression was validated using real-time PCR and Western blot in estrogen-treated Fischer rat pituitary glands. In summary, characterization of individual genes and gene sets in human and in estrogen-induced rat prolactinomas validates the model and provides insights into genomic changes associated with this commonly encountered pituitary tumor.

Prolactinoma is the most common adult pituitary tumor accounting for 60% of functional pituitary adenomas (1, 2). Prolactinoma causes hyperprolactinemia, resulting in impaired reproduction, decreased libido, amenorrhea, and galactorrhea. Adenoma growth leads to compressive mass effects resulting in headache, visual disturbances, cranial nerve palsies, and hypopituitarism. Prolactinomas express abundant levels of dopamine D2 receptor (D2R) and can be effectively treated with dopaminergic drugs, reducing both prolactin levels and tumor volume (1). However, about 20% of prolactinomas may exhibit either dopamine agonist resistance or high recurrence rates after surgery (3), indicating a need for better understanding the disease and developing improved therapies.

Evaluation of new drugs and understanding prolactinoma development have been assessed using high-mobility group protein A (HMGA)-1 and HMGA2 transgenic mice, D2R knockout mice and estrogen-treated rats (3–7). HMGA1 and HMGA2 transgenic mice develop pituitary tumors that secrete both GH and prolactin (PRL) at approximately 12–16 months of age (4, 5). D2R-deficient mice form pituitary lactotroph adenomas at approximately 17–20 months (6). In contrast to these transgenic models, some rat strains rapidly develop prolactinoma when exposed to estrogen (7). The Fischer 344 (F344)-inbred rat is the most sensitive strain to form prolactinomas after receiving estrogen (7). Continuous estrogen treatment of F344 rats induces rapid pituitary growth within a few days, and the pituitary enlarges up to 10-fold after 8–12 wk of treatment. PRL overproduction by these estrogen-induced pituitary tumors results in circulating hyperprolactinemia increasing up to approximately 220-fold (7).

Rapid prolactinoma development in rats makes them useful for drug evaluation and for studying tumorigenesis. Drug studies have included those evaluating cysteamine (8), SMS 201–995 (9), fumagillin and its analog TNP-470, terguride, flutamide, and tamoxifen (10), estrogen receptor (ER) antagonist ICI-182780 (11), thalidomide, octreotide (12), and most recently epidermal growth factor receptor (EGFR) tyrosine kinase inhibitors. Bromocriptine and cabergoline potently suppress prolactinoma growth, validating the usefulness of this model in evaluating drug efficacy (12, 13). Aberrant pathways discovered using this model include expression of estrogen-induced pituitary tumor transforming gene (PTTG1) and fibroblast growth factor as early events in prolactinoma pathogenesis (14), bone morphogenetic protein-4 (BMP4) promotion of PRL secretion and lactotroph proliferation (15), reduced retinaldehyde dehydrogenase 1 synthesis (16), and melatonin induction of apoptosis of prolactinoma cells (17).

Although estrogen-induced rat prolactinomas have been widely studied, the underlying pathogenesis of human prolactinoma formation remains elusive. Importantly, the fidelity by which rat models recapitulate human prolactinoma is not clear. Accordingly, we used global profiling with microarray technology to investigate these questions in rat and human prolactinomas and found that both species share expression of several genes. We also identified novel genes and gene sets, which may be important for prolactinoma development.

Materials and Methods

Microarray data

Experiments to yield rat pituitary microarray data were previously reported (7), and microarray data was downloaded from Gene Expression Omnibus (access no. GSE4028). Briefly, male ACI or Brown Norway (BN) rats were previously treated with diethylstilbestrol (DES; a synthetic nonsteroidal estrogen, 5 mg for 12 wk), and pituitaries were analyzed using Affymetrix Rat Genome 230 version 2.0 arrays (Affymetrix, Santa Clara, CA). For each strain, four pituitaries of the DES-treated and three pituitaries of the vehicle-treated were analyzed.

The microarray data for human prolactinomas were derived from the Oyesiku laboratory (18). Briefly, human prolactinomas (n = 4, three males and one female) were obtained during transsphenoidal surgery as part of an ongoing accession of human pituitary tumors (18). The study was approved by the Institutional Review Board of Emory University, and informed consent was obtained for all subjects. Tumors were microdissected and removed using the surgical microscope, rinsed in sterile saline, snap frozen in liquid nitrogen, and stored (−80 C) until analysis. Each tumor fragment was confirmed independently by a neuropathologist by histology and immunohistochemistry before molecular analysis. Three cadaveric pituitary glands (two males and one female) were obtained from the National Resource Center (National Disease Research Interchange, www.ndriresource.org) and confirmed to be normal by histology. Each human tissue sample was analyzed using Affymetrix Human Genome U95Av2 arrays and data uploaded to Gene Expression Omnibus (access no. GSE36314).

Microarray data analysis

Microarray data were imported into Genespring GX11 (Agilent Technologies, Palo Alto, CA) according to the manual. All genes were normalized to their median, and data quality was assessed using a principle component analysis and sample clustering. Differentially expressed genes were identified by parametric testing (not assuming equal variance). Results were subjected to multiple testing correction using the Benjamini and Hochberg method. Genes with a false discovery rate (FDR) less than 0.01 were considered as differentially expressed. The gene symbols of differentially expressed genes were used to query the PubMed database together with key words such as prolactinoma, invasion, drug resistance, and recurrence using LocoySpider 2010 software (Locoy, Hefei, China) according to the manual. Microarray results were validated by real-time PCR.

Gene set enrichment analysis

Analyses were performed using gene set enrichment analysis (GSEA) desktop software from the Broad Institute (Massachusetts Institute of Technology, Cambridge, MA) (19). Briefly, the data matrix was exported from Genespring GX11 and further analyzed according to the manual (Agilent Technologies). Gene sets for curated pathways (c2.all.v2.5.symbols.gmt) and gene sets derived from motifs of transcription factors (c3.tft.v2.5.symbols.gmt) were used for analysis. Results were ranked according to the FDR. The FDR of 25% was used as a cutoff as suggested in the user guide of the GSEA software (Broad Institute).

Rat prolactinoma model

Animal protocols were approved by the Institutional Animal Care and Use Committee at Cedars-Sinai Medical Center (Los Angeles, CA) (20). After isoflurane inhalational anesthesia, 17β-estradiol-filled SILASTIC brand capsules (Dow Corning Corp., Midland, MI; medical grade tubing special; length 3 cm; outer diameter 0.125 in.; inner diameter 0.062 in.) were implanted sc into 4- to 5-wk ovariectomized female F344 rats (Harlan Sprague Dawley, Inc., Indianapolis, IN). Five hundred microliters of blood were collected by retroorbital bleeding every 2 wk for hormone assessment. Rats were euthanized 2 months after estrogen implantation and cardiac blood and pituitary glands collected and analyzed. Fragments of each pituitary were fixed in formalin, embedded in paraffin, and either preserved in RNA Later solution (Ambion, Grand Island, NY) or frozen in liquid nitrogen.

Western blot

Protein extracts were resolved by Nupage 4–12% Bis-Tris Gel (Invitrogen, Grand Island, NY), samples were electroblotted onto polyvinyl difluoride membrane (Invitrogen), and membranes blocked and incubated with primary antibody. Antibody targeting C-Myc (5605, 1:1000) was purchased from Cell Signaling (Danvers, MA). E2F1 (sc-193X, 1:500) and Sp1 (sc-59, 1:200) antibodies were from Santa Cruz Biotechnology (Santa Cruz, CA). Donkey antirabbit (1:2000) or antimouse (1:2000) (GE Healthcare, Waukesha, WI) antibodies were conjugated to horseradish peroxide to reveal immunocomplexes by enhanced chemiluminescence (Pierce, Rockford, IL). Detected bands were quantified using Image J version 1.43 (National Institutes of Health, Bethesda, MD) as instructed in the software manual.

Real-time quantitative PCR

Total RNA was isolated by an RNAeasy kit (QIAGEN, Hilden, Germany) according to the manual. Reverse transcription was carried out by a Superscript III first-strand cDNA synthesis kit (Invitrogen) and PCR amplifications with SYBR Green PCR master mix. Real-time quantitative PCR was performed according to the manufacturer's protocol (Bio-Rad Laboratories, Hercules, CA) in a Bio-Rad IQ5 multicolor real-time PCR detection system (Bio-Rad). A standard curve was used to quantify expression levels and the amount of each gene normalized to glyceraldehyde-3-phosphate dehydrogenase (GAPDH) mRNA. Primer sequences are available on request.

Results

Rat prolactinoma

Estrogen-treated ACI and BN rats exhibited distinct gene expression patterns as indicated by principle component analysis (Fig. 1A). A total of 4545 genes (FDR < 0.01) were differentially expressed in pituitaries of DES-treated ACI rats compared with vehicle treatment (Fig. 1B and Supplemental Fig. 1, published on The Endocrine Society's Journals Online web site at http://endo.endojournals.org). In contrast, only 629 pituitary genes (FDR < 0.01) were differentially expressed in DES-treated BN rats (Fig. 1B and Supplemental Fig. 2). A total of 472 genes were found to be differentially expressed in both DES-treated ACI and BN rats (Fig. 1B). Because DES-treated BN rats did not develop prolactinomas, these overlapping genes may not be critical for estrogen-induced prolactinoma development. Of the 4545 genes, 2711 were up-regulated and 1834 down-regulated in DES-treated ACI rats (Supplemental Fig. 1). When searching the literature, 58 of 4545 genes are known to be aberrantly expressed in prolactinomas (Table 1), including Pttg1, Gal, Egf, Egfr, Ccnb1/b2, Igfbp3, Nrg1, and Myc.

Fig. 1.

Microarray analysis of prolactinomas. A, Microarray data derived from pituitaries of either vehicle- or DES-treated ACI or BN rats were imported into Genespring GX11. Principle component analysis was performed according to the software manual. B, Differentially expressed genes were identified using Genespring GX11 and a Venn diagram generated to unmask overlapping genes. C, Genes related to invasion, recurrence, and drug resistance were identified and applied to the Venn diagram. D, A total of 38 genes was selected from 4545 differentially expressed genes in DES-treated ACI rats (empty bar). Expression of these genes was validated using real-time PCR (solid bar) in prolactinomas from estrogen-treated Fischer rats. The y-axis represents fold change in arithmetic scale.

Table 1.

Differentially expressed pituitary genes in DES-treated ACI rats that are also aberrantly expressed in human prolactinomas

Gene symbol	Gene title	ACI
		Fold	P value
Gal	Galanin prepropeptide	214.0	1.16E-04
Vip	Vasoactive intestinal peptide	28.7	2.36E-03
Cfp	Complement factor properdin	12.8	6.13E-03
Lpl	Lipoprotein lipase	12.8	1.63E-03
Nupr1	Nuclear protein 1	10.7	2.70E-03
Nov	Nephroblastoma overexpressed gene	8.0	1.17E-03
Pgr	Progesterone receptor	7.9	2.41E-04
Sstr3	Somatostatin receptor 3	7.5	1.97E-04
Mki67	Antigen identified by monoclonal antibody Ki-67	7.0	2.25E-03
Cdc2	Cell division cycle 2, G1 to S and G2 to M	6.5	2.52E-03
Tnf	TNF (TNF superfamily, member 2)	6.4	4.74E-03
Bag1	BCL2-associated athanogene	5.8	1.74E-04
Lgals3	Lectin, galactoside-binding, soluble, 3	5.4	3.73E-04
Myc	Myelocytomatosis oncogene	4.9	1.82E-03
Pttg1	Pituitary tumor-transforming 1	4.8	2.31E-03
Ntrk2	Neurotrophic tyrosine kinase, receptor, type 2	4.7	1.66E-04
Mpg	N-methylpurine-DNA glycosylase	4.3	3.31E-04
Egfr	Epidermal growth factor receptor	3.7	8.57E-03
Egf	Epidermal growth factor	3.4	4.20E-04
Ccnb1	Cyclin B1	3.3	6.32E-04
Igfbp3	IGF binding protein 3	3.2	5.05E-03
Sla	Src-like adaptor	3.0	1.96E-04
Ccnd3	Cyclin D3	2.9	2.58E-03
Igf1	IGF-I	2.9	7.64E-03
Cope	Coatomer protein complex, subunit-ϵ	2.7	3.09E-03
Aga	Aspartylglucosaminidase	2.5	9.19E-04
Tgfb3	TGFβ3	2.4	7.01E-04
Bid	BH3 interacting domain death agonist	2.3	8.34E-04
Cs	Citrate synthase	2.2	4.66E-03
Pts	6-Pyruvoyl-tetrahydropterin synthase	2.1	9.92E-03
E2f1	E2F transcription factor 1	2.0	8.57E-03
Guk1	Guanylate kinase 1	1.9	1.97E-03
Poll	Polymerase (DNA directed), λ	1.9	3.63E-03
Grb2	Growth factor receptor bound protein 2	1.6	2.97E-03
Chgb	Chromogranin B	1.5	8.98E-03
Prkar1a	Protein kinase, cAMP dependent regulatory, type I, α	1.5	1.94E-03
Preb	Prolactin regulatory element binding	1.4	8.11E-03
Cdkn1b	Cyclin-dependent kinase inhibitor 1B	1.4	1.18E-03
D2R	Dopamine D2 receptor	1.4	4.53E-03
Ncl	Nucleolin	1.4	6.79E-03
Gpi	Glucose phosphate isomerase	1.4	6.39E-03
Ctnna1	Catenin (cadherin associated protein), α1	−1.5	3.54E-03
Pou1f1	POU class 1 homeobox 1	−1.5	4.80E-03
Slc20a1	Solute carrier family 20 (phosphate transporter), member 1	−1.7	1.40E-03
Pitx1	Paired-like homeodomain 1	−1.8	9.03E-03
Dst	Dystonin	−1.9	3.56E-03
Jup	Junction plakoglobin	−2.1	6.92E-03
Cp	Ceruloplasmin	−2.3	4.02E-03
Ghr	GH receptor	−2.7	3.19E-03
Gstm1	Glutathione S-transferase μ 1	−2.7	5.06E-03
Chga	Chromogranin A	−2.9	1.36E-03
Sms	Spermine synthase	−3.1	1.94E-03
Prlr	Prolactin receptor	−3.7	8.18E-03
Nrg1	Neuregulin 1	−5.0	7.01E-04
Rab3b	RAB3B, member RAS oncogene family	−18.2	6.35E-04
Ret	Ret protooncogene	−21.3	3.75E-03
Spp1	Secreted phosphoprotein 1	−44.5	1.56E-03

Analysis of human prolactinomas identified 1177 differentially expressed genes (FDR < 0.01) (Fig. 1B and Supplemental Fig. 3), of which 32 were previously reported in prolactinomas. When combining the results derived from pituitaries of DES-treated ACI rats and human prolactinomas, 268 genes were identified (Fig. 1B), whereas 44 genes were identified when combining results derived from pituitaries of DES-treated BN rats and human prolactinomas (Fig. 1B), suggesting that pituitary glands derived from DES-treated ACI rats are more similar to human prolactinomas. Of the 268 genes identified, 235 are well annotated. A total of 145 of 235 genes were concordantly expressed between pituitaries of DES-treated ACI rats and human prolactinomas, and the remaining genes were inversely expressed. Concordantly expressed genes included Myc, Ccng1, Igf1, Faim3, Atf6, E2f1, Akt2, ATF1, Ccne1, Bcr, Hes1, Mdm1, Sox2, Cdh1 and Cdh2, Igfbp5, Gpc3, Dlk1, Cebpd, and Nnat. A representative list is presented in Table 2 (see Supplemental Fig. 4 for a full list). The inversely expressed genes included Gal, Igfbp3, Pik3ca, Gpc4, stat2, Nrg2, and Acin1. Other genes that are likely important for prolactinoma development but not differentially expressed in the ACI rat pituitary include Hmga1 and Hmga2.

Table 2.

Representative pituitary genes concordantly expressed in DES-induced ACI rat and human prolactinomas

Gene symbol	Gene title	ACI rat		Human
		Fold	P value	Fold	P value
Cgref1	Cell growth regulator with EF hand domain 1	211.3	1.9E-05	1.3	3.2E-02
Plaur	Plasminogen activator, urokinase receptor	49.0	1.2E-04	1.3	8.4E-03
Ptprc	Protein tyrosine phosphatase, receptor type, C	13.3	8.3E-03	1.2	1.4E-02
Spc25	SPC25, NDC80 kinetochore complex component, homolog (Saccharomyces cerevisiae)	5.6	1.1E-03	1.1	6.1E-03
Myc	Myelocytomatosis oncogene	4.9	1.8E-03	1.4	1.8E-02
Ccng1	Cyclin G1	4.3	3.0E-04	1.4	2.5E-02
Scamp5	Secretory carrier membrane protein 5	3.5	7.9E-03	1.4	3.5E-02
Ptprn	Protein tyrosine phosphatase, receptor type, N	3.2	4.4E-04	1.8	1.4E-03
Faim3	Fas apoptotic inhibitory molecule 3	3.0	3.9E-03	1.3	8.1E-03
Igf1	IGF-I	2.9	7.6E-03	1.2	1.2E-02
Ern1	Endoplasmic reticulum to nucleus signaling 1	2.8	1.8E-04	1.1	2.2E-02
Eif4ebp1	Eukaryotic translation initiation factor 4E binding protein 1	2.5	1.8E-03	1.2	3.7E-02
Spink5	Serine peptidase inhibitor, Kazal type 5	2.4	6.0E-03	1.1	1.7E-02
Atf6	Activating transcription factor 6	2.3	2.3E-03	1.2	3.3E-02
Atp2a3	ATPase, Ca2+ transporting, ubiquitous	2.2	9.9E-03	1.1	2.2E-02
Polr2e	Polymerase (RNA) II (DNA directed) polypeptide E, 25 kDa	2.1	3.0E-04	1.3	3.5E-02
Rala	V-ral simian leukemia viral oncogene homolog A (ras related)	2.1	4.2E-04	1.4	8.5E-03
E2f1	E2F transcription factor 1	2.0	8.6E-03	1.2	1.3E-02
Akt2	V-akt murine thymoma viral oncogene homolog 2	2.0	7.9E-03	1.2	3.4E-02
Atf1	Activating transcription factor 1	2.0	4.8E-03	1.3	5.3E-03
Mcm6	Minichromosome maintenance complex component 6	1.9	1.9E-03	1.4	4.9E-02
Npr2	Natriuretic peptide receptor B/guanylate cyclase B (atrionatriuretic peptide receptor B)	1.8	4.3E-03	1.2	2.3E-02
Dnajb5	DnaJ (Hsp40) homolog, subfamily B, member 5	1.7	6.0E-03	1.8	1.3E-02
Fntb	Farnesyltransferase, CAAX box, β	1.7	3.3E-03	1.2	3.8E-02
Bcr	Breakpoint cluster region	−1.4	5.0E-03	−1.3	2.4E-02
Add1	Dducing 1 (α)	−1.6	1.8E-03	−1.1	2.4E-03
Hes1	Hairy and enhancer of split 1 (Drosophila)	−1.8	3.7E-03	−1.6	4.3E-03
Ddit4	DNA damage-inducible transcript 4	−1.9	3.0E-03	−3.4	6.6E-03
Map3k4	MAPK kinase kinase 4	−2.0	6.4E-03	−1.4	9.6E-03
Plch2	Phospholipase C, η2	−2.0	8.4E-03	−1.5	4.1E-03
Hspa2	Heat shock protein 2	−2.0	5.8E-03	−2.8	4.4E-06
Mdm1	Mdm1 nuclear protein homolog (mouse)	−2.2	3.2E-03	−1.8	1.1E-02
Sox2	SRY (sex determining region Y)-box 2	−2.3	9.0E-03	−1.3	1.1E-02
Ghrhr	GHRH receptor	−2.3	7.0E-03	−1.8	2.0E-02
Prkci	Protein kinase C, iota	−2.4	6.7E-03	−1.7	4.8E-02
Cdh2	Cadherin 2	−2.5	3.2E-03	−1.9	7.1E-03
Rorb	RAR-related orphan receptor B	−3.0	8.5E-03	−1.2	1.1E-02
Igfbp5	IGF binding protein 5	−3.3	2.6E-03	−10.8	1.4E-03
Gpc3	Glypican 3	−3.8	2.6E-03	−12.0	3.0E-05
Lpin1	Lipin 1	−4.3	1.7E-04	−1.9	2.0E-02
Dlk1	δ-Like 1 homolog (Drosophila)	−4.4	1.3E-03	−18.9	1.1E-04
Cdh1	Cadherin 1	−4.7	5.5E-03	−2.6	7.3E-04
Dhcr24	24-Dehydrocholesterol reductase	−5.1	5.8E-04	−1.3	1.3E-02
Wfdc2	WAP four-disulfide core domain 2	−5.3	5.1E-04	−2.4	1.6E-06
Itpr1	Inositol 1,4,5-triphosphate receptor, type 1	−6.4	2.6E-04	−2.2	3.6E-03
Stmn2	Stathmin-like 2	−6.6	1.7E-04	−2.2	4.4E-06
Fam107a	Family with sequence similarity 107, member A	−8.2	1.4E-03	−1.7	2.4E-03
Tmem30b	Transmembrane protein 30B	−8.5	1.9E-03	−1.7	5.7E-03
Cebpd	CCAAT/enhancer binding protein (C/EBP), δ	−9.8	1.4E-03	−4.3	4.9E-05
Nnat	Neuronatin	−12.8	6.8E-04	−9.9	2.8E-04
Efemp1	EGF-containing fibulin-like extracellular matrix protein 1	−52.5	8.9E-03	−1.4	1.3E-02

Genes related to prolactinoma invasion, drug resistance, and recurrence

The major challenges of prolactinoma treatment include drug resistance and rapid recurrence after surgery. We searched the literature to determine candidate genes that may relate to tumor invasiveness, drug resistance, and rapid recurrence. Among 4545 genes, 787 appear related to invasiveness (Fig. 1C), 959 to drug resistance (Fig. 1C), and 627 to tumor recurrence. The presence of tumor stem cells may contribute to drug resistance and tumor recurrence, and of 4545 genes, 1225 were known stem cell-associated markers. A total of 376 genes were identified from combining genes related to drug resistance and recurrence, of which 360 were also present in the list of 1225 stem-cell related genes (Fig. 1C). Notable gene expressions associated with stem cell regulation and tumor invasion included Notch2, Cd44, Cd14, Cd55, Pten, Tgfb3, Mdm2, Bcr, Abr, stat1, sox2, and Sp1 (Table 3 and Supplemental Fig. 5).

Table 3.

Representative pituitary genes involved in stem cell regulations, regulation of invasion, tumor recurrence, and drug resistance

Gene symbol	Gene title	Fold	P value
Gale	UDP-galactose-4-epimerase	12.3	1.1E-04
G6pd	Glucose-6-phosphate dehydrogenase	2.9	1.6E-04
Efs	Embryonal Fyn-associated substrate	−4.0	2.3E-04
Ccr5	Chemokine (C-C motif) receptor 5	22.7	4.1E-04
Dcn	Decorin	−20.3	4.6E-04
Cd68	Cd68 molecule	5.1	5.8E-04
Il4ra	IL-4 receptor, α	28.6	6.8E-04
Cd14	CD14 molecule	4.3	8.2E-04
Cd44	Cd44 molecule	2.9	8.5E-04
Cad	Carbamoyl-phosphate synthetase 2, aspartate transcarbamylase, and dihydroorotase	3.0	1.0E-03
Abr	Active BCR-related gene	−2.1	1.1E-03
Cd55	Cd55 molecule	18.5	1.3E-03
Crop	Cisplatin resistance-associated overexpressed protein	−2.0	1.3E-03
Dlk1	δ-Like 1 homolog (Drosophila)	−4.4	1.3E-03
Spn	Sialophorin	−2.6	1.4E-03
Mdm2	Mdm2 p53 binding protein homolog (mouse)	1.5	1.8E-03
Notch2	Notch homolog 2 (Drosophila)	−4.9	1.9E-03
Numb	Numb homolog (Drosophila)	−1.6	2.1E-03
Cd9	CD9 molecule	−2.0	2.2E-03
Lox	Lysyl oxidase	6.1	2.3E-03
Evi1	Ecotropic viral integration site 1	−3.0	2.3E-03
Ace	Angiotensin I converting enzyme (peptidyl-dipeptidase A) 1	−5.4	2.5E-03
App	Amyloid β (A4) precursor protein	2.1	2.6E-03
Anxa5	Annexin A5	1.6	2.6E-03
Tnfrsf11a	TNF receptor superfamily, member 11a	10.4	2.7E-03
Ptpn11	Protein tyrosine phosphatase, nonreceptor type 11	−1.7	2.8E-03
Ky	Kyphoscoliosis peptidase	−4.2	3.2E-03
Hmox1	Heme oxygenase (decycling) 1	9.5	3.6E-03
Hes1	Hairy and enhancer of split 1 (Drosophila)	−1.8	3.7E-03
Dcx	Doublecortin	−4.3	3.9E-03
Cdk4	Cyclin-dependent kinase 4	1.6	4.1E-03
Ptpn6	Protein tyrosine phosphatase, nonreceptor type 6	1.4	4.1E-03
Sp1	Sp1 transcription factor	2.1	4.5E-03
Rars	Arginyl-tRNA synthetase	1.7	4.7E-03
Itgb2	Integrin-β2	1.9	4.8E-03
Bcr	Breakpoint cluster region	−1.4	5.0E-03
Dhfr	Dihydrofolate reductase	1.8	6.0E-03
Lef1	Lymphoid enhancer binding factor 1	−1.9	6.6E-03
Ckap4	Cytoskeleton-associated protein 4	3.9	6.9E-03
Stat1	Signal transducer and activator of transcription 1	−1.7	7.5E-03
Ptprc	Protein tyrosine phosphatase, receptor type, C	13.3	8.3E-03
Pten	Phosphatase and tensin homolog	2.3	8.8E-03
Sox2	SRY (sex determining region Y)-box 2	−2.3	9.0E-03
Bdnf	Brain derived neurotrophic factor	11.4	9.6E-03
Rad51	RAD51 homolog (RecA homolog, E. coli) (S. cerevisiae)	2.1	9.6E-03
Wars	Tryptophanyl-tRNA synthetase	1.7	9.6E-03
Cdc42	Cell division cycle 42 (GTP binding protein)	1.2	9.6E-03
Ril	Reversion induced LIM gene	−2.2	9.7E-03
Rac1	Ras-related C3 botulinum toxin substrate 1	2.0	9.8E-03

To validate the microarray results, the expression of 38 genes was measured by real-time PCR. As depicted in Fig. 1D, this technique validated microarray expression data for 36 of 38 investigated genes observed in rat prolactinomas, suggesting approximately 95% validation using real-time PCR. Due to low availability of human prolactinoma tissues, the microarray results were not validated with quantitative RT-PCR. To avoid overinterpretation of the human tissue results, we performed a conservative data analysis by analyzing genes with a fold change larger or smaller than 2. This analysis indicated 24 genes (Igsf1, Gja1, Ddit4, Blcap, Cd9, Hspa2, Mdk, Dsp, Igfbp5, Aldh2, Cnn3, Gpc3, Tshb, Dlk1, Cdh1, Wfdc2, Itpr1, Stmn2, Ppl, Tbx19, Cebpd, Nnat, Fshb, and Lhb), which exhibit convergent expression between human prolactinomas and rat models.

Gene set enrichment analysis

Cellular processes are affected by sets of genes acting in concert (19). Using 3272 gene sets derived from curated pathways (pathway gene sets), 669 were found up-regulated in DES-treated ACI rats, of which 278 gene sets had a FDR less than 25% and 19 had a FDR less than 10% (Table 4 and Supplemental Fig. 6A). A total of 526 gene sets were up-regulated in vehicle-treated ACI rats, with 211 gene sets having a FDR less than 25% and 67 having a FDR less than 10% (Supplemental Fig. 6B). In the BN rat pituitary gland, 622 gene sets were up-regulated after vehicle treatment with 47 gene sets at FDR less than 25%, and two at FDR less than 10% (Table 4 and Supplemental Fig. 6C). A total of 573 gene sets were up-regulated in DES-treated rats, with 15 gene sets at FDR less than 25% but none at FDR of 10% (Supplemental Fig. 6D). In human prolactinomas 615 gene sets were up-regulated but none at FDR less than 25% (Table 4 and Supplemental Fig. 6E). A total of 753 gene sets were up-regulated in the normal rat pituitary, with 102 sets significantly enriched at FDR less than 25% with five at FDR less than10% (Table 4 and Supplemental Fig. 6F).

Table 4.

Enriched pathway gene sets for each group

Source	Type	Pathway gene sets			Transcription factor gene sets
		Total enriched	FDR <25%	FDR <10%	Total enriched	FDR <25%	FDR <10%
ACI rat	DES	669	278	19	186	59	7
	Vehicle	526	211	67	404	260	111
BN rat	DES	622	47	2	310	57	0
	Vehicle	573	15	0	280	41	0
Human	Prolactinoma	615	0	0	279	0	0
	Normal pituitary	753	102	5	309	14	1

Of the top 20 pathway gene sets enriched in DES-treated ACI rats, 19 were also found to be enriched in pituitary gland of DES-treated BN rats (Supplemental Fig. 6G), suggesting that the two samples are similar. In contrast, only nine were also enriched in human prolactinomas. None of the 19 gene sets enriched in DES-treated BN rats passed the FDR 25% cutoff, suggesting the degree of alteration of these gene sets may determine development of rat prolactinoma. Altered gene sets were grossly categorized into four categories: 1) cell death/transformation/invasion (VERNELL_pRB_CLSTR1, BCL2_FAMILY_AND_REG_NETWORK, OLDAGE_DN, CMV_24HRS_UP, CROONQUIST_RAS_STROMA_UP, and EMT_UP); 2) cell differentiation (HSC_INTERMEDIATEPROGENITORS_FETAL, LIAN_MYELOID_DIFF_RECEP-TORS, HSC_INTERMEDIATE-PROGENITORS_SHARED, and HSC_INTERMEDIATE-PROGENITORS_ADULT); 3) epigenetic disruption (HDACI_COLON_BUT30MIN_DN, TSA_PANC50_UP, TSA_CD4_UP, and HDACI_COLON_SUL16HRS_DN); and 4) metabolism (NADLER_OBESITY_HYPERGLYCEMIA and HSA01030_GLYCAN_STRUCTURES_BIOSYNTHESIS_1). Enrichment of the VERNELL_pRB_CLSTR1 gene set was of special interest because Rb is known to play an important role in pituitary tumorigenesis (21). A representative figure for the VERNELL_pRB_CLSTR1 is shown in Fig. 2A.

Fig. 2.

Microarray data for either rat or human prolactinomas were exported from Genespring GX11 and imported to GSEA software according to the manual. GSEA was performed using pathway gene sets. A, VERNELL_pRB_CLSTR1 gene set enrichment. B, Western blot for c-Myc, E2F1, and Sp1 proteins in pituitary glands from vehicle- or estrogen-treated Fischer rats.

Because Rb regulates E2F activity, enrichment of the Rb pathway gene set is also supported by observed enrichment of the E2F1 gene set (REN_E2F1_TARGETS, rank 56; Supplemental Fig. 6A). Enrichment of pituitary gene sets was attributed to altered expression of corresponding genes. For example, enrichment of the epidermal growth factor (EGF) gene sets (EGFPATHWAY, rank 33; CARDIACEGFPATHWAY, rank 51; and EGF_HDMEC_UP, rank 264) supports the overexpression of EGF (3.4-fold, P = 4.20E-04) and EGFR (3.7-fold, P = 8.57E-03). Enrichment of the WNT gene set (LIN_WNT_UP, rank 106; Supplemental Fig. 6A) supports the up-regulation of WNT4 (13.3-fold, P = 6.07E-4). Enrichment of Myc gene sets (SCHUMACHER_MYC_UP, rank 61; ZELLER_MYC_UP, rank 119; MYC_TARGETS, rank 141; LEE_MYC_TGFA_UP, rank 227; MYC_ONCOGENIC_SIGNATURE, rank 255; YU_CMYC_UP, rank 260; and FERNANDEZ_MYC_TARGETS, rank 262; Supplemental Fig. 6A) supports the up-regulation of Myc (4.9-fold, P = 1.8E-3) in prolactinomas. Changes in the ATF/CREB pathway were supported by enrichment of the CREB gene set (CREBPATHWAY, rank 117; Supplemental Fig. 6A). Down-regulation of CEBPD (9.8-fold, P = 1.4E-3, Fig. 2B) was reflected in enrichment of the CEBP gene set (HALMOS_CEBP_DN, rank 123; Supplemental Fig. 6A). Up-regulation of TNF (6.4 fold, P = 4.7E-3) was supported by enrichment of the gene set (SANA_TNFA_ENDOTHELIAL_UP, rank 208; NEMETH_TNF_UP, rank 236). Moreover, enrichment of the stem cell gene sets (STEMCELL_COMMON_DN, rank 118; and STEMCELL_COMMON_UP, rank 154; Supplemental Fig. 6A), suggesting involvement of tumor stem cell in rat prolactinoma development. Enrichment of the EMT gene sets (EMT_UP, rank 20; and JECHLINGER_EMT_UP, rank 98; Supplemental Fig. 6A) support involvement of invasion pathways.

We also analyzed the data using 615 gene sets derived from those with a transcription factor (TF) motif (TF gene sets). One hundred eighty-six gene sets were up-regulated in the DES-treated ACI rats, with 59 gene sets at FDR less than 25% and seven at FDR less than 10% (Table 4 and Supplemental Fig. 7A). Four hundred four gene sets were up-regulated in the vehicle-treated ACI rats with 260 at FDR less than 25% and 111 at FDR less than 10% (Table 4 and Supplemental Fig. 7B). Analyzing the BN rats indicated that 310 gene sets were up-regulated in vehicle-treated rats with 57 gene sets at FDR less than 25% (Table 4 and Supplemental Fig. 7C). Two hundred eighty gene sets were up-regulated in the DES-treated rat with 41 gene sets at FDR less than 25% (Table 4 and Supplemental Fig. 7D). None of the gene sets was enriched in the vehicle- or DES-treated BN rats at FDR less than 10% (Table 4). In the human data set, 279 gene sets were up-regulated in prolactinomas but none at FDR less than 25% (Table 4 and Supplemental Fig. 7E). Three hundred nine gene sets were up-regulated in the normal rat pituitary, with 14 gene sets at FDR less than 25% and one gene set at FDR less than 10% (Table 4 and Supplemental Fig. 7F).

Enrichment of the TF gene sets suggested that activities of the corresponding transcription factor may be altered. The top 20 enriched gene sets (Supplemental Fig. 7G) correspond to 13 known transcription factors. Three of 13 (ETS1, ATF1, and CREB) known to regulate prolactin expression (20, 22, 23) suggest that multiple pathways are involved in regulating prolactin expression in estrogen-induced rat prolactinoma. Analysis of transcription factor gene sets were supported by gene expression microarray results, which indicated that gene expression levels of eight of 13 transcription factors (c-Myc, Sp1, ATF1, AP4s1, CREB3, USF1, ETS1, and CEBPD) were significantly altered (Supplemental Fig. 7H). Altered expression of the eight transcription factors were further validated by real-time PCR (Supplemental Fig. 7H). Our results indicated an increased E2F1 mRNA and enrichment of the E2F1 gene set, and expression of Sp1, c-Myc, and E2F1 was validated at the protein level by Western blotting (Fig. 2B). These results were also supported by previous analysis of pathway gene sets. For example, enrichment of the Myc TF gene set (V$MYC_Q2) was supported by several gene sets for the Myc pathways (rank 61, FDR = 0.14; rank 119, FDR = 0.17; rank 141, FDR = 0.18; Supplemental Fig. 7A). The E2F TF gene set (V$E2F_Q6_01) was ranked 56 in the enriched TF gene sets (FDR = 0.23) and is supported by pathway gene sets VERNELL_pRB_CLSTR1 (rank 3) and REN_E2F1_TARGETS (rank 56; Supplemental Fig. 7A). The enrichment of the CEBPD (V$CEBPDELTA_Q6) TF gene set was supported by the pathway gene set (HALMOS_CEBP_DN, rank 123; Supplemental Fig. 7A). Enrichment of the ATF1 (V$ATF_01) and CREB (V$CREB_01) TF gene sets were supported by the CREB pathway gene set (CREBPATHWAY, rank 117; Supplemental Fig. 7A).

Discussion

This study integrally analyzed microarray results derived from rat and human prolactinomas at the single gene as well as gene set levels and provides insights in the genomic profile of rat and human prolactinomas.

When using single gene-based analysis, results obtained from pituitaries of estrogen-treated rats exhibited commonly regulated genes also observed in human prolactinomas, suggesting that the microarray captured some common prolactinoma features, which may play roles in regulating prolactinomas. Microarray data derived from DES-treated ACI rats confirmed that several genes such as Pttg1, Egf, Egfr, E2f1, Myc, Ccnb1, and Ccnb2 are aberrantly expressed in prolactinoma. PTTG1 overexpression is found in most pituitary adenomas and particularly in invasive hormone-secreting tumors (24). EGF increases prolactinoma cell proliferation and prolactin expression (25). Inhibition of EGFR or its family members using small molecular inhibitors blocks cell proliferation as well as prolactin expression. Earlier studies found that Ccnb2 was overexpressed in pituitary tumors (26, 27), and this gene was induced by HMGA1 and HMGA2 (27). The E2F1 pathway is frequently enhanced in pituitary tumors (28) and Rb knockout induces these tumors in transgenic mice (28). Up-regulation of these genes was further supported by subsequent GSEA analysis, suggesting that these genes affect tumor growth in a gene set-specific manner.

By searching the literature, we also identified 376 genes that might play important roles in regulating drug resistance and prolactinoma recurrence. Most genes (360 of 376) also regulate stem cells, suggesting that prolactinoma tumor stem cells may play a role in prolactinoma treatment resistance. These stem cell related genes include Notch2, Cd44, Cd14, Pten, Tgfb3, Mdm2, and Abr. Involvement of stem cells in prolactinoma development was also supported by the pathway gene sets analysis, which exhibited enrichment of both stem cell gene sets and cell differentiation gene sets.

HMGA1 and HMGA2 regulate pituitary tumor development (4, 5, 27). HMGA1 and HMGA2 transgenic mice develop PRL- and GH-secreting pituitary tumors associated with a high expression of Ccnb2 (4, 5, 27). We detected up-regulation of Ccnb1 and Ccnb2 but no change in HMGA1 and HMGA2, suggesting that other pathways may be involved in regulating Ccnb1/2. It is also interesting that some genes are discordantly expressed in rat and human prolactinomas. For example, galanin (Gal) knockout mice exhibit reduced prolactin, and transgenic mice overexpressing Gal develop pituitary adenomas and increased secretion of PRL and GH (29). Gal was up-regulated 214-fold in DES-treated ACI rats but was down-regulated 55-fold in human prolactinomas. Similarly, expression of Igfbp3, Pik3ca, Gpc4, stat2, Nrg2, and Acin1 was discordant in rat and human prolactinomas. These results might reflect either differences in tumor types or aberrant species-specific mechanisms involved in prolactinoma development.

When using gene set-based analysis, we identified several gene sets significantly enriched (FDR < 10%) in DES-treated ACI rats but not enriched (FDR < 10%) in human prolactinomas. No overlap of gene sets between prolactinomas derived from ACI rats and human prolactinomas suggested that underlying pathways for prolactinoma development may differ between the two species. However, this could also reflect differences in experimental systems. Rat prolactinomas develop in homogenously inbred rat strains, allowing rigorous results with high statistical power using only a few samples. In contrast, human prolactinomas are highly heterogeneous, leading to low statistical power and subsequently constraining identification of pathways that might be important for disease pathogenesis. Therefore, analyses of human samples require greater sample size to gain genomic insights into the disease.

Despite the lack of gene set overlap between rat and human prolactinomas, gene sets enriched in estrogen-treated rats confirmed known pathways in pituitary tumor development. For example, the Rb/E2F pathway is known to play an important role in pituitary tumorigenesis (30). Rb heterozygous mice develop spontaneous pituitary tumors (21, 30). We identified Rb (VERNELL_pRB_CLSTR1) and E2F (REN_E2F1_TARGETS) pathway gene sets. Moreover, involvement of the Rb/E2F pathway was further supported by enrichment of the TF gene set derived from the E2F binding motif (V$E2F_Q6_01). Finally, increased E2F1 expression was validated by Western blot. Similarly, several Myc pathway gene sets were enriched in DES-treated ACI rats. Activation of Myc was also supported by the GSEA analysis using TF gene sets. Increased Myc expression was validated by real-time PCR and Western blot. The role of the EGF pathways was similarly supported by pathway gene set analysis. Altered in ATF1, CREB3, Sp1, and Ets1 were supported by pathway and TF gene set analysis and validated by real-time PCR. The CEBPD pathway involvement was supported by TF gene set analysis, and CEBPD down-regulation was validated by real-time PCR.

This study focused on gene sets enriched in prolactinomas. Because the pituitary gland comprises more than six cell types, gene sets enriched in control groups could be due to artificial effects introduced by unrelated, nontarget cells (31). Care should be taken to interpret gene set results in the normal pituitary, especially because microarrays may not be sufficiently sensitive and poor microarray probes may introduce errors. For example, in all microarray studies for prolactinoma, prolactin gene expression levels were not significantly changed, which was not consistent with real-time PCR results. Using next-generation sequencing may help address this challenge. Estrogen-induced rat prolactinomas may exhibit limited invasiveness and are responsive to treatment with dopamine agonists (12, 13), indicating that these identified changes in gene expression are not wholly sufficient to cause prolactinoma invasion or drug resistance. This analysis is based on previous knowledge of these genes, and genes with little previous study could therefore be missed. Importantly, because human pituitary microarray data were not validated with quantitative RT-PCR, all fold change of less than 2 should be considered with caution. When discarding genes with fold change less than 2, only 24 exhibited convergent expression between human prolactinomas and rat models. We validated the microarray data derived from DES-treated ACI rats with estrogen-induced prolactinomas in F344 rats, indicating that these results can be generalized to other rat strains. Indeed, rat strains exhibit different sensitivities to estrogen, and Quantitative trait locus (QTL) has been used to map loci responsible for observed differential estrogen sensitivity in ACI and BN strains (32). By using pituitary mass as a quantitative trait, two loci (Ept 5 and 7) on rat chromosome 4 were found to exert significant effects on 17β-estradiol (E2)-induced pituitary growth. D4Rat103, Ghrhr, Tgfa, Trh, Ghr1, Ret, Ccnd2, Cdkn1b, and D4Rat204 reside in the Ept5 locus and D7Rat44, Cdk4, Trhr, Myc, and D7Rat15 in the Ept 7 locus. Among these genes, five (Ghrhr, Ret, Cdkn1b, Cdk4, and Myc) exhibited altered gene expression in pituitaries of ACI rat models and two (Ghrhr and Myc) exhibited convergent expression between human prolactinomas and rat models (Table 2).

This study generated thousands of differentially expressed genes associated with prolactinoma development. Here we prioritized these candidate genes for further investigation. At the single-gene level, we identified 149 genes concordantly expressed in both rat and human prolactinomas. GSEA identified hundreds of enriched gene sets and dozens were directly supported by gene expression changes at the single gene level. These candidate genes or gene sets may be worthy of further investigation.

Several identified genes and gene sets have been shown to regulate pituitary tumorigenesis and/or prolactin expression. These include EGF/EGFR cross talk with the E2/ER pathway to up-regulate prolactin expression and lactotroph cell proliferation (33, 34), Rb/E2F1 regulation of pituitary tumorigenesis (35), CEBPD coordination of prolactin expression and lactotroph proliferation (20), Pit1 and Ets1 regulation of prolactin transcription (22), dopamine and D2R suppression of prolactin (36), and CREB induction of prolactin (37). We further searched the literature for evidence of interaction between identified genes and gene sets. Indeed, these genes and gene sets appear to regulate each other and form a complex signaling network (Fig. 3). E2/ER activates Rb/E2F1 (38), which may subsequently activate Myc (39). Myc also suppresses CEBPD expression (40), which may subsequently up-regulate prolactin and lactotroph proliferation (20). CEBPD in turn suppresses E2/ER and Rb/E2F1 activity by direct interaction (41). Interestingly, D2R expression is up-regulated in estrogen-induced rat prolactinomas, which may in turn constrain lactotroph tumorigenesis. How dopamine/D2R interacts with E2/ER-Rb/E2F1-Myc-CEBPD is unclear. CREB may activate CEBPB (42), which may subsequently activate CEBPD.

Fig. 3.

Proposed scheme for interactions between identified genes and gene sets in estrogen-induced rat prolactinomas.

In summary, we analyzed rat and human prolactinomas at the levels of single genes and gene sets. We confirmed known genes and pathways in the pituitaries of DES-treated ACI rats and also unraveled new candidate genes and gene sets that may play important roles in regulating prolactinomas, forming a complicated signaling network. This knowledge may result in enhanced use of rat models to investigate genomic profiles of human prolactinomas.