Prostate cancer is the most frequently diagnosed cancer as well as the second leading cause of cancer death among American men [1]. Androgen ablation therapy for patients with advanced prostate cancer inevitably fails as the disease progresses to an androgen-independent stage [2]. Few effective treatment options are available to these patients, and these increase survival by only a matter of months [3, 4]. We examined an in vivo human prostate cancer tumour model to identify the underlying molecular events involved in hormonal progression. The LNCaP hollow fibre model differs from xenograft models by growing the LNCaP human prostate cancer cell line within fibres that are implanted subcutaneously in host mice [5]. These fibres prevent host cells from infiltrating, and contaminating, the tumour cell population. Upon castration of the host the LNCaP cells progress to an androgen-independent stage as determined by a rising titre of serum prostate-specific antigen (PSA), mimicking this aspect of clinical disease [5].

Many genes important in the development and progression of cancer have been identified by first detecting their altered expression at different stages of the disease. It has thus become desirable to perform high-throughput gene expression analyses to quickly assay the expression status of large numbers of genes in a given model or treatment condition. A variety of techniques are available for monitoring gene expression profiles, with microarrays and Serial Analysis of Gene Expression (SAGE) being the most widely used. However, microarray experiments are only able to monitor the expression of genes for which prior knowledge of the transcript sequence is available, and they also lack the sensitivity to detect transcripts expressed at very low levels. The SAGE technique is capable of detecting novel transcripts [6–9], but SAGE is also not optimal for detecting low abundance transcripts. In contrast, suppression subtractive hybridization includes a normalization step that enriches for rare transcripts in a population of RNAs [10, 11]. Subtractive hybridization is also able to detect entirely novel transcripts for which no previous annotation exists [12, 13]. Thus, subtractive hybridization is a powerful tool to detect less abundant transcripts and the novel transcripts that tend to be expressed at low levels. In support of this concept, a significant proportion of the transcripts identified by subtractive hybridization were shown to be expressed at levels below the detection limit of Affymetrix GeneChip^® arrays [14]. Additionally, subtractive hybridization identified a number of novel transcripts which were not represented on these arrays [14].

Gene expression changes occurring with the hormonal progression of prostate cancer have been examined in various systems (see [15–17], for example). Our goal was to utilize the LNCaP hollow fibre model to identify genes that had not been previously associated with prostate cancer. The application of subtractive hybridization resulted in the identification of a number of novel expressed sequences in this model. These sequences exhibit low protein-coding potential and low conservation across species, but RT-PCR experiments confirmed their expression in samples of prostate cancer and in a variety of human tissues.

While the precise number of genes in the human genome remains unknown, it is clear that an even greater number of transcripts are produced by a myriad of alternative splicing events. The recognition of non-coding RNAs has also resulted in a greater focus on transcripts not containing open reading frames. Thus, even with the completed sequence of the genome further experiments are required to fully annotate the functional units transcribed from the genome in the different cell types during normal growth and development, as well as in diseased tissues. High-resolution tiling microarrays have been used to generate predictions of likely transcripts, many of which mapped to intergenic and intronic regions of the genome that were not previously annotated [22–24]. Similarly, approaches focused on sequencing full-length cDNAs continue to identify a large number of novel transcripts, many of which appear to be non-protein coding [25, 26]. Our results here provide further evidence that a significant number of genes and transcripts remain to be identified.

Studies now suggest that the majority of the human genome is transcribed, but the function(s) of most of these transcripts has not yet been demonstrated (reviewed in [27]). The lack of defined function for these transcripts has led some to propose that they arise indirectly through spurious transcription of the genome. However, new functions of non-coding transcripts continue to be identified, indicating that these "spurious" transcripts likely have functions that have yet to be characterized. For example, steroid receptor RNA activator (SRA) was demonstrated to act as an RNA transcript to regulate the transcriptional activity of steroid nuclear receptors, including the androgen receptor [28]. Another non-coding transcript, expressed at low levels in various tissues, was recently demonstrated to regulate the nuclear trafficking of nuclear factor of T cells (NFAT), and has been renamed non-coding RNA repressor of NFAT (NRON) [29]. Such studies confirm that low-abundance, non-coding, RNA transcripts perform diverse functions and regulate multiple biological processes.

Subtractive hybridization has been used to characterize gene expression changes associated with prostate cancer [12, 15, 30–33], and some of these studies have also characterized novel genes that were identified from their subtracted libraries [12, 13, 30]. However, these reports examined only a few novel genes that were differentially expressed in their experimental systems, and few studies have used subtractive hybridization to examine changes with hormonal progression in an in vivo model [15]. Our study is unique in that we sequenced all the clones arising from the subtractive hybridization experiments performed with in vivo samples from the LNCaP hollow fibre model. This approach identified a large number of transcripts that had not previously been detected in prostate cancer cells and may be of prognostic or therapeutic value.

We identified 25 completely unannotated clones, and an additional 57 clones matching previously sequenced ESTs that were otherwise unannotated. The isolation of these ESTs specifically from prostate cancer cells may prove informative at a later date. Furthermore, we considered any overlap of an EST or annotated gene with one of our clones to signify that that clone derived from a previously identified transcript. However, several of the clones matching ESTs and annotated transcripts displayed only partial overlap with the known sequence, suggesting that the subtracted clone may still represent an unidentified splice variant of the known transcript. Most of these novel transcript fragments exhibited low sequence conservation amongst vertebrate species, though this may indicate that these transcripts are human- or primate-specific. Prostate cancer is only known to spontaneously occur in humans, rats, and some species of dogs even though the prostate organ is present in all mammals. Furthermore, the human prostate exhibits numerous morphological differences. Thus, it is likely that many transcripts required for prostate development and prostate cancer progression would not be extensively conserved throughout evolution. In support of this, the KLK3 gene, encoding prostate-specific antigen (PSA), is only present in primate genomes [34]; PSA is a well known clinical marker used to monitor prostate cancer progression and response to therapy. This demonstrates that evolutionary conservation alone is not predictive of potential clinical utility.

The functional relevance of these novel transcripts in the hormonal progression of prostate cancer remains to be elucidated. In our experiments, 4 of the 6 clones tested were expressed only in the normal prostate and testes; this limited tissue expression profile suggests these novel transcripts may function specifically in these organs. This also suggests the expression of these transcripts is tightly regulated, as would be expected for a functional transcript. Furthermore, a related publication from our group identified a novel variant of TMEFF2 which encodes a secreted form of the protein [35]. This alternate form of the protein was identified after a novel clone from our subtractive hybridization library (clone 2A06) was mapped to the fourth intron of the TMEFF2 gene. The expression of TMEFF2 has been shown to increase with progression to androgen independence [15, 36], consistent with our subtractive hybridization experiments. TMEFF2 is also currently being investigated as a target for antibody-based therapy in the treatment of prostate cancer [37, 38], confirming that our approach identified novel transcripts which may be of interest in the study of prostate cancer. Changes in expression of some of these transcripts may also be valuable as a marker for disease progression.

To characterize the function(s) of these novel transcripts it will first be necessary to identify the full-length cDNA sequence. Multiple techniques are available to recover full-length cDNA molecules starting from only cDNA fragments. For example, RACE is widely used to obtain full-length cDNA sequences [39]. RACE was successful in isolating further 3' sequence information for 3 of the 4 clones for which it was attempted. As in the case of clone 1E05, it is possible that all three of these clones actually derive from large 3' untranslated regions of protein coding transcripts. The large size and low expression levels of these transcripts increase the difficulty of identifying their full 5' sequence. Alternatively, we may have already identified the majority of the sequence in these novel, non-coding transcripts. Another technique to identify full-length sequence for these novel transcripts would be to screen existing cDNA libraries [6, 13]. However, given the relatively low expression of these transcripts this approach would likely require extensive screening of such libraries. Two recent studies have used Northern blot analysis to detect expression of similar low-abundance, non-coding, novel transcripts [22, 26]. These groups found that only 20–30% of the novel transcripts were detectable by Northern blot analysis, even when using large amounts of poly-A+ RNA, indicating that the remaining transcripts fell below the detection limit of this technique.

Margulies et al. [40] recently described a highly parallel sequencing by synthesis approach that demonstrated an increased throughput for sequencing of genomic DNA. Our group has combined sequencing by synthesis and random shotgun analysis to generate ESTs and characterize the transcriptome of LNCaP cells grown in tissue culture [41]. This study isolated approximately 180,000 ESTs, and of these, 1,900 (1.0%) mapped to the human genome in regions not previously annotated in the Ensembl database. One of these ESTs mapped directly to clone 2A03 described here, while 12 more mapped within 1.5 kb of clones 1cD03 and 2dB01. However, the remaining 22 novel clones identified here were still not detected by this alternative approach.

SAGE has also been used to identify potentially novel transcripts [6–9]. Unfortunately, while SAGE provides sufficient sequence information to accurately map the tags to the human genome, there is often little other information available to aid in the design of experiments to derive more sequence data from these novel transcripts. In contrast, subtractive hybridization provides longer sequence fragments, but it is not possible to determine the orientation of these fragments. Our data also suggest that subtractive hybridization was able to detect transcripts that had not previously been found using SAGE or high throughput sequencing by synthesis. It is possible that subtractive hybridization may be more sensitive to detecting transcripts expressed at low levels. Alternatively, subtractive hybridization may isolate those transcripts that can not be efficiently detected by SAGE, for instance transcripts lacking an NlaIII restriction site. Common results from multiple techniques gives greater confidence in the identification of novel transcripts and underlines the value of using complementary techniques to achieve a more thorough analysis of the human transcriptome.

Our subtractive hybridization experiments have identified novel transcripts that are specifically expressed in the prostate and/or the urogenital tissues. It may be of clinical value to further develop these novel transcripts as prognostic or therapeutic markers for prostate cancer and hormonal progression. Additionally, characterizing such novel transcripts and transcript variants may aid in identifying and understanding the processes important in the development of androgen independent disease.