To assess the diversity of gene expression in the mouse prostate, we constructed and characterized cDNA libraries representing distinct stages of prostate development and maturation that began with the urogenital sinus (E16.5), and continued with prostates from two day old (neonatal), ten day old, 20 day old, 35 day old (puberty), three month old (normal adult), and 14 month old (aged adult) animals (Figure 1). A total of 62,046 ESTs were generated, of which 50,562 passed stringent quality assessments. Of these, 4,299 ESTs annotated to the mitochondrial genome. Assembly of the remaining prostate ESTs into distinct clusters or transcription units (TUs) was performed using the sequence assembly program Phrap 14 and each TU was assigned an annotation by BLAST comparisons against sequences in public databases. In accordance with definitions used by the FANTOM and RIKEN teams in assembling a mouse transcriptome using full-length cDNAs 15, we use the term TU to designate a cluster of transcripts or ESTs that contain a common core of sequence information transcribed from a segment of the genome. A TU may be defined by a single cDNA sequence or by multiple overlapping ESTs that share a common core sequence. Assembly and annotation of the 46,263 mouse prostate ESTs produced 15,009 unique TUs. Of these, 4,550 TUs are composed of more than one EST and 10,459 TUs are represented by a single EST. Of the 4,550 TUs that contain more than one EST, 3,567 are represented by sequences derived from more than one mouse prostate library. All sequences are archived for public accession in the mouse prostate expression database (mPEDB) 1617.

To identify genes potentially involved with prostate development and maturation, we determined temporal expression changes by calculating transcript abundance levels in cDNA libraries constructed from different stages of prostate development. Pairwise comparisons of the datasets from each timepoint were performed using Audic and Claverie's analysis method 18. At a significance level of p = 0.001, 285 genes were differentially expressed between one or more time points (see Additional data file 1, Table 1). Applying the Bonferroni correction for multiple comparisons identified a cohort of 69 genes with highly significant differential expression between one or more developmental stages. Hierarchical clustering of these genes demonstrated distinct temporal partitioning of gene expression with notable cohorts increasing over time and others decreasing over time. In Figure 2, two clusters of genes are highlighted which show progressive increases or decreases in expression over time from urogenital sinus to adulthood. Several of these genes encode proteins involved in specialized prostate secretory activity such as spermine binding protein (Sbp) 19 and serine protease inhibitor Kazal type 3 (Spink3) 20. Several are also known to be regulated by androgenic hormones, such as probasin (Pbsn), and increase in expression following the onset of puberty 21. Other TUs with temporal changes in expression have no currently identified functional roles in prostate development.

The rodent prostate gland is comprised of four distinct anatomical lobes: dorsal, lateral, ventral and anterior, each with unique ductal branching patterns and different responses to androgenic hormones (Figure 3) 2223. Studies of secretory products from rodent prostates have identified specific expression patterns that can be assigned to individual lobes 242526. To further explore the functional differences between mouse prostate lobes, we compared the profiles of genes expressed in each lobe and identified 34 genes that exhibited statistically significant differential expression. This group was reduced to 17 genes after correcting for multiple comparisons (Table 2). The expression patterns of these genes clearly support conclusions of prior anatomical and biochemical studies demonstrating functional heterogeneity between lobes of the gland. Genes highly expressed in the anterior prostate relative to the other lobes included onzin (alias placenta specific 8; Plac8), pancreatic family type A ribonuclease 1 (RNAse1), and experimental autoimmune prostatitis antigen (Eapa1; a putative prostate transglutaminase), as well as several uncharacterized transcripts. Spermine binding protein (Sbp), and serine protease Kazal type 3 (Spink3) are relatively highly expressed in the ventral lobe, while transcripts with increased expression in the dorsolateral prostate include receptor activity modifying protein 2 (Ramp2), a transcript with similarity to acrosomal vesicle protein-1, and several uncharacterized transcripts. We determined the lobe-specific expression patterns of several transcripts using quantitative PCR (QPCR), and found that the EST and QPCR methods produced qualitatively similar results (Figure 4).

Wubah et al. have reported the cloning and tissue distribution of a transcript termed ventral prostate predominant 1 (Vpp1) that was identified through mRNA differential display between different mouse prostate lobes 27. QPCR and virtual expression analysis confirmed that the expression of Vpp1 is primarily localized in the ventral prostate, with lower but detectable levels of expression in the dorsolateral and anterior prostate lobes (Figure 4).

Our EST results demonstrated that the probasin transcript was most highly expressed in the anterior prostate with lower - but still high - transcript abundance levels in the dorsolateral lobes, and minimal expression in the ventral lobe (Figure 4). QPCR quantitated the highest probasin expression in the dorsolateral lobes followed by the anterior prostate, with lowest levels in the ventral lobe. These results differ from studies of expression localization using the rat probasin promoter to drive transgene expression in mouse prostate epithelium where the highest levels were observed in the lateral and dorsal lobes, with lower expression in the ventral prostate and very low to absent expression in the anterior prostate 1128. Subsequent immunohistochemical and Western analyses with antibodies recognizing the mouse probasin protein demonstrated high levels of expression in the anterior as well as the dorsolateral mouse prostate lobes 29. These results suggest that the rat probasin promoter may confer slightly different cellular specificity compared with the native mouse promoter, or that transgenic constructs alter the normal regional distribution of probasin expression.

The comparative analysis of cDNA libraries representing the repertoire of genes expressed in the mouse prostate provided an opportunity to identify genes whose expression is enhanced or restricted to the mouse prostate relative to other normal tissues. The human prostate gland expresses several transcripts and corresponding proteins in a highly tissue-restricted manner including prostate specific antigen (PSA) 30, human glandular kallikrein 2 (hK2) 31, prostase/KLK4 32 and prostate specific membrane antigen (PSMA) 33. Studies of these proteins have provided insights into normal prostatic function, mechanisms of hormone-regulated gene expression and the development of prostate pathology. To identify transcripts preferentially expressed in the mouse prostate, we assigned each prostate TU to a UniGene cluster and then determined the tissue sources of all ESTs comprising the cluster. In total, 776 prostate TUs were mapped to UniGenes containing ESTs derived from three or fewer other tissues (see Additional data file 2). Only 28 of these TUs represent characterized genes (Table 1). Included among the 28 are NKX3.1 and probasin, both of which are known to have prostate-restricted patterns of expression 2934. Most of the prostate-enhanced TUs are represented by uncharacterized ESTs and full-length cDNAs sequenced by the RIKEN mouse transcript sequencing project 15. Eighty TUs contained ESTs from at least two different prostate libraries. Of these, eight represent spliced gene products based on sequence comparisons with the mouse draft genome sequence. An additional 20 TUs had interesting features, such as high sequence conservation with human sequences, and may represent orthologous genes.

One TU, originally designated TU23348 after the original library clone identifier, was found to be expressed in multiple mouse prostate cDNAs libraries, lack representation in the UniGene or dbEST sequence databases, and exhibit sequence similarities to genes in the β-defensin family. We reasoned that since TUs specifically originating from the mouse prostate were not represented in the public sequence databases, TU23348 could represent a β-defensin with expression restricted to the prostate. To evaluate this possibility, we measured the relative abundance of TU23348 transcripts in multiple mouse tissues by QPCR. These results confirmed high levels of TU23348 expression in the prostate, with very low levels in testis and skeletal muscle, and no detectable transcripts in a wide range of other tissues (Figure 5a). A multiple tissue mRNA blot confirmed the RT-PCR findings (Figure 5b), and northern analysis identified a transcript size of approximately 0.4 kb. Northern analysis and QPCR also indicated that TU23348 expression is further compartmentalized within the mouse prostate having highest expression levels in the dorsolateral and ventral lobes, with very low to absent expression in the anterior prostate (Figure 5c).

A full-length TU23348 transcript was obtained through assemblies of mouse prostate cDNAs and by 5' and 3' rapid amplification of cDNA ends (RACE) reactions using mouse prostate cDNA as a template. The coding sequence comprises 219 nucleotides encoding a predicted polypeptide of 73 amino acids (Figure 6a). The nucleotide sequence comprises a 72 nt 5' UTR and 111 nt 3' UTR with putative polyadenylation sequences located 99, 72 and 51 nt upstream of the polyA tail. Protein sequence alignments with members of the mouse and rat β-defensin family demonstrate a high level of sequence similarity including conservation of the majority of the cysteine and positively-charged residues that are predicted to modulate antimicrobial activity (Figure 6) 3536. Alignments of the TU23348 transcript with the draft mouse genome sequence indicates that TU23348 is located on chromosome 8 and is comprised of two exons; the first encoding 19 amino acids and the second 54 amino acids. The exons are separated by a 7.3 kb intron (Figure 6). The gene is located between two other members of the β-defensin family, β-defensin 1 and 2. We named this gene Pbd1 for prostate β-defensin 1. Subsequently, in accordance with the Human Genome Organization (HUGO) nomenclature committee, this gene carries the official designation of Defβ37 for defensin beta 37 as it is the 37th β-defensin reported. The sequence is represented in GenBank under accession number AY387658.

Male C57Bl6J mice (Jackson Labs, Maine, USA) of defined ages were sacrificed in accordance with institutional protocols. Whole urogenital sinus (UGS) or individual prostate lobes were dissected, pooled and snap frozen in liquid nitrogen. Tissues were homogenized using a Polytron PT-MR-2100 rotor stator homogenizer (Kinematica, Switzerland) and total RNA was purified using Qiagen RNeasy purification system (Qiagen, Valencia, CA, USA). For libraries made from whole prostate, equal amounts of RNA from each lobe were pooled prior to cDNA synthesis. Approximately 1 μg of total RNA from each tissue source was used in first strand cDNA synthesis reactions using the SMART cDNA library construction kit in accordance with the manufacturer's protocol (Clonetech, CA, USA). Second strand cDNA was synthesized and amplified through 18-21 cycles of the PCR, digested with SfiI and directionally ligated into the lambdaTriplEx2 phagemid. Phagemids were converted to pTriplEx2 plasmids in the BM25.8 Escherichia coli strain. A total of 14 cDNA libraries were constructed. From each library, between 3,000 and 15,000 bacterial clones were picked into 384-well plates using a Q-pix robot (Genetix, OR, USA) and grown overnight at 37°C in LB media supplemented with 8% glycerol. We used 384-pin-replicators (Genetix) to transfer directly to 10 μl PCR reactions containing Triplex2 forward PCR primer (5' CTCGGGAAGCGCGCCATTGTGTTGGT 3') and Triplex2 reverse PCR primer (5' ATACGACTCACTATAGGGCGAATTGGCC 3'). PCR reactions were incubated with 1.5 units of exonuclease I (Epicentre) and 0.5 units of shrimp alkaline phosphatase (USB) for 15 minutes at 37°C to remove primers and nucleotide triphosphates. Pin-replicators were used to transfer ~0.2 μl of purified PCR product to 5 μl fluorescence-based sequencing reactions containing 0.6 μM Triplex2 5'Seq primer (5' CTCGGGAAGCGCGCCATTGTGTTGGT 3'), 0.5 μL Big Dye Terminator (ABI), 4 mM Tris pH 9.0 and 1 mM EDTA. One urogenital sinus library (UGS02) was derived from approximately 500 Balb/c strain fetal mice and was synthesized using both oligo-dT and random hexamer priming followed by size-selection for cDNAs greater than 500 basepairs. These cDNAs were ligated to BstXI and EcoRI adaptors, cloned into pcDNAII plasmids (Invitrogen, CA, USA) and sequenced as described above but with the following primers: VN26 PCR (5' TTTCCCAGTCACGACGTTGTA 3'), VN27 PCR (5' GTGAGCGGATAACAATTTCAC 3'), M13R 5'Seq (5' GGAAACAGCTATGACCATG 3'). Detailed library descriptions are available at the mouse prostate expression database website 1617.

Each cDNA was sequenced once to produce an EST averaging 500 bp in length. Each EST was screened for sequence comprising vectors, E. coli, repetitive elements and low complexity regions using the RepeatMasker algorithm. Sequence quality was determined using Phred basecalling software 51. Following quality assessment, ESTs comprised of >100 unmasked nucleotides with 80/100 bases demonstrating a quality score exceeding 20 were included in the analysis. Each EST was compared against the mouse reference sequence database using the Basic Local Alignment Search Tool (BLAST) 52 and assigned a corresponding gene annotation if the BLAST score exceeded 1,100. A BLAST score of 1,100 represents a minimum match of 555 nucleotides with 100% identity, or a longer sequence match with relatively few mismatches or gaps in the alignment; a stringency suitable for matching ESTs against curated gene sequences. The remaining ESTs not matching a reference sequence with high stringency were assembled into gene clusters using the Phrap assembly software with minimum match length parameter set to 60 and minimum score parameter set to 90 53. Consensus sequences were determined and compared with sequences in the UniGene, GenBank, and dbEST databases 54 using BLAST. ESTs that received a BLAST score of 200 or better were assigned a database annotation. The remaining ESTs were classified as unannotated. The number of ESTs from each library corresponding to the same annotation were enumerated and used for subsequent comparative determinations of gene expression measurements.

ESTs and EST assemblies not receiving annotations in UniGene, GenBank or dbEST were further evaluated for characteristics supporting their classification as genes as opposed to artifacts such as genomic DNA. Sequences represented by only one EST were not further evaluated. ESTs represented in ≥ 2 different libraries were compared against the mouse genome assembly (February 2003 freeze) using the BLAT algorithm to assess for splicing and gene predictions 5556. These sequences were further evaluated for protein coding regions by conceptual translations of six possible reading frames using a custom perl script. Sequences with greater than 200 amino acids of translated sequence in any reading frame were analyzed on the BLOCKS server 57, run with default settings, to identify possible motifs. Conceptual translations were also checked for protein similarity using the PSI- and PHI-BLAST algorithms 58.

To identify genes with restricted or enhanced expression in the prostate relative to other mouse tissues, we searched the UniGene database with each prostate TU and selected those UniGene clusters containing sequences derived from two or fewer non-prostate tissues. For the purposes of this screen the terms 'urinary bladder', 'vesicular gland', 'Wolffian duct includes surrounding region', 'testicles', 'Wolffian duct and mullerian duct' were grouped into a 'genitourinary' category. In addition, 'tumor biopsy sample', 'tumor gross tissue', 'pooled lung tumors', 'tumor metastatic to mammary', 'infiltrating ductal carcinoma', 'spontaneous tumor metastatic to mammary, and 'stem cell origin' were grouped under 'tumor'.

To identify genes differentially expressed in prostate development or between mouse prostate lobes, we utilized the program Identification of Differentially Expressed Genes 6 test statistics (IDEG6) 5960. IDEG6 performs a normalization calculation to adjust for libraries with different numbers of ESTs. Six different algorithms are applied to the datasets to determine those TUs with statistical differences in abundance levels between two or more libraries. Romualdi et al. identified the Audic-Claverie approach 18 as the best statistical method for identifying differentially expressed genes in pairwise comparisons 60, and we report the results of this analysis using a significance level of p = 0.001 with the Bonferroni correction to adjust for multiple comparisons. The DLP01, CG02, VP01 and VP02 libraries were evaluated in a pairwise fashion for the lobular comparison. Libraries included in the developmental pairwise comparison were UGS01, UGS02, NEONATAL, DAY10, DAY20, DAY35, a pooled library for MONTH3 (represented by combined TUs from the lobular comparison) and MONTH14. Genes and ESTs exhibiting differential expression between distinct developmental time points were grouped by hierarchical clustering using the Cluster software 61. Clusters were visualized using Treeview 61.

Unannotated, putatively novel sequences determined to be differentially expressed between prostate lobes were further evaluated using PSI-BLAST (Position-Specific Iterated BLAST) and PHI-BLAST (Pattern-Hit Initiated BLAST) 58. Nucleotide sequences were translated using the ExPASy Translate program 62 to identify open reading frames. ESTs were also compared against the assembled mouse genome (February 2003 freeze) through the University of Santa Cruz's BLAST-Like Alignment Tool (BLAT) 56 to identify spliced sequences and predicted genes.