A novel approach to identify driver genes involved in androgen-independent prostate cancer

To identify candidate genes involved in the progression to AIPC we used a replication-incompetent LV, LV-SFFVEGFP (Figure 1A) that has a strong spleen focus-forming virus promoter known to dysregulate genes[20], and also includes a bacterial origin of replication and a kanamycin resistance gene to allow identification of integration sites by rescue of shuttle vector plasmids in E. coli. Our approach uses random shearing of genomic DNA to avoid restriction site bias[21] and does not require any PCR amplification so it also eliminates PCR-based bias. The LV expresses EGFP gene from the spleen focus-forming virus promoter, allowing for efficient tracking of transduced cells in vitro and in vivo (Figure 1B). The androgen-dependent human PCa cell line, LNCaP, was transduced in triplicate with LV-SFFVEGFP resulting in three independent cultures of LNCaP cells denoted shuttle vector-mutagenized (SVM) -A, -B, and -C. The transduction frequency was over 99% as assessed by EGFP expression.

The clonality of these cultures prior to selection for androgen independence was evaluated to ensure that cells used for the mutagenesis screen were polyclonal. We reasoned that a highly polyclonal pool of integrations, in essence a library, would improve the ability of our screen to identify AIPC progression genes. High-throughput sequencing of shuttle vector rescued plasmids was performed to identify provirus-chromosome junctions (Figure 1C,D). Sequences were aligned to the human genome to identify the provirus integration site in the human genome. A clonality calculation was performed to approximate the number of unique integrations present in each sample prior to selection for androgen-independence. By counting both the number of new integration sites in each survey, and accounting for the number of previously identified integration sites, the approximate number of unique integration sites in a population could be calculated. A sample calculation for SVM-A is shown in Additional file1: Figure S1. In this example, five surveys were performed and a clonality of 7.2 × 10³ unique integration sites was calculated. The more surveys that are performed, the larger N will become, improving the accuracy of the clonality estimate, so this is a minimum estimate. Clonality was similar for cultures SVM-B and SVM-C. Sequencing confirmed that there was no evidence of cross contamination between independent cultures, as no identical integration sites were identified between the three mutagenized cultures. These data showed that we had a highly polyclonal starting population prior to selection for androgen independence.

To model the clinical progression to advanced AIPC[22]in vitro we used a previously established method to select for cells that become androgen-independent in culture[11, 14, 15]. In this model androgen-dependent LNCaP cells are cultured in charcoal/dextran-treated fetal bovine serum (CT-FBS) which is essentially devoid of androgen. This selects for those cell clones that have a proliferative advantage in an androgen-deficient environment (Figure 1B). The human LNCaP cell line has several advantages for our screen including expression of androgen receptor, androgen-dependent growth, a demonstrated ability to develop androgen-independent growth[11, 14, 15], and the ability to form tumors in vivo[16, 23] to allow us to explore genes and gene pathways that mediate progression in vivo. Preliminary experiments showed that transfer of LNCaP cells into media supplemented with 10% CT-FBS led to a loss of cells which would have reduced the clonality of our LV-mutagenized library of insertion sites. We found that initial culture in 9.75% CT-FBS with 0.25% untreated FBS minimized cell loss. We thus cultured LV-mutagenized and control cells in media supplemented with 9.75% CT-FBS with 0.25% untreated FBS for approximately 140 days prior to moving cultures to media supplemented with 10% CT-FBS. At this time shuttle-vector mutagenized androgen-dependent LNCaP cells were maintained in an androgen-deficient environment with CT-FBS. After 211 days, the SVM-A culture showed an increase in growth rate compared to control cultures, and was deemed androgen-independent (Figure 1B). We hypothesized that in the SVM-A culture, cells with LV proviruses near genes that influenced progression to AIPC had a selective advantage and this led to androgen-independent growth in the SVM-A culture prior to control cultures. This is expected to lead to an over-representation of cells with proviral integrants near genes that influence AIPC. Thus, analysis of provirus integration sites in these androgen-independent cells should identify dysregulated genes near vector proviruses that may mediate progression to AIPC. Genomic DNA was isolated from LV-transduced androgen-independent cells to identify candidate AIPC progression genes by shuttle vector rescue (Figure 1C).

In vitro models lack the ability to identify genes involved in processes required only in vivo such as vascularization. We thus also performed our LV shuttle vector screen in vivo using a LNCaP xenograft model[16, 23]. NOD.Cg-PrkdcscidIl2rgtmlWjl/SzJ (NSG) mice are severely immunocompromised, and allow for efficient engraftment of human cells[24]. Androgen-dependent control LNCaPs and mutagenized SVM-A cells prior to selection in vitro were injected into male NSG mice (Figure 2A). Tumors developed from the injection of both SVM-A and control LNCaP cells. Similar to androgen-dependent tumors in PCa patients, it was expected that tumor volumes would regress following androgen deprivation therapy. In patients, this is done by either a surgical or chemical castration[22]. In the in vivo model, this environment was created by surgical removal of the testes which are the primary source of androgens. It was expected that tumor size would decrease immediately following castration and similar to the in vitro model, the androgen-deficient environment would select for androgen-independent cells modeling what occurs in PCa patients[11, 12]. LNCaP cells were injected in male NSG mice and formed in 6 of 7 injected mice (Figure 2B). Tumors did regress following castration and following the castration, tumor growth resumed. Tumors were allowed to grow until they reached volumes larger than the tumor size prior to castration, at which point tumors were harvested and genomic DNA was obtained for shuttle vector rescue analysis.

Rescue of genomic DNA from the androgen-independent culture SVM-A in vitro samples post-selection recovered a total of 21 unique sites (Additional file2: Table S1). From two in vivo tumors analyzed post-castration we identified a total of 54 insertion sites, 27 from one tumor and 28 from the second with one site found in both tumors (Additional file2: Table S1). All sites were single insertion sites. Custom PERL computer programs were used to analyze whether the provirus integrated within a RefSeq gene and to also provide the distance from the integration site to the nearest 3 Refseq gene transcription start sites (TSS). Only genes that had a TSS within 100 kb of the vector provirus were considered.

We hypothesized that our approach should enrich for vector integration sites near regions known to be involved in PCa progression. The 75 integration sites (Additional file2: Table S1) were mapped relative to previously described PCa susceptibility loci (Additional file3: Table S2) and compared to 412 control integration sites obtained from transduced LNCaP cells prior to selection for androgen-independence, and also to a random in silico generated data set of 10,000 sites (Figure 3). The percent of vector integrations within PCa susceptibility loci for LNCaP cells cultured in androgen-deficient medium was significantly higher than from the LNCaP cells prior to culture in androgen-deficient culture medium (p < .001) and also significantly higher than from the random sites (p < .001). This demonstrates a significant enrichment of vector proviruses near loci previously associated with PCa when cells are grown under androgen-deficient conditions.

We used data from previously published microarray analysis of patient tumors to identify the genes in our dataset that were most likely to contribute to AIPC based on dysregulation in patient tumors. This approach has the advantage that multiple data sources may improve the power of the screen, and may improve the ability to identify genes that are clinically relevant due to their dysregulation in patient tissues. We used Oncomine™ (Compendia Bioscience, Ann Arbor, MI)[25] to compare microarray data from 16 different studies[26‐39] that evaluated gene expression in normal prostate tissue vs. PCa tissue (prostate carcinoma or prostate adenocarcinoma) to evaluate the candidate progression genes identified near vector proviruses listed in Additional file2: Table S1. We reasoned that candidate driver genes identified in our screen were likely to be over or under-expressed in PCa tissue and that meta-analysis of previously published microarray studies would be a powerful way to screen our candidates for driver genes likely to be potential biomarkers or therapeutic targets. To evaluate over or under-expression we used both the rank and a multiple comparisons-corrected p-value as calculated by Oncomine™. Candidate genes with p-values < .005 and ranks less than 1000 in either the over- or under-expression multiple comparisons were considered as potential driver genes (Table 1). Promising candidates identified from the androgen-independent in vitro culture were ATPAF1 (ATP synthase mitochondrial F1 complex assembly factor 1) and TRPM4 (transient receptor potential cation channel, subfamily M, member 4). Promising candidates identified from the androgen-independent in vivo tumor were GCOM1 (GRINL1A complex locus 1), MEX3D (mex-3 RNA binding family member D), and PTRF (polymerase I and transcript release factor). Three of these genes, ATPAF1, MEX3D and TRPM4, are over-expressed in PCa tissue, while two are under-expressed, GCOM1 and PTRF (Figure 4 and Additional file4: Figure S2). The locations of the five vector integrants were mapped relative to genomic loci using the University of California, Santa Cruz (UCSC) Genome Browser (Additional file5: Figure S3 and Additional file6: Figure S4).

To validate the effects of ATPAF1 and PTRF on androgen-independence we used LV-mediated RNAi. For ATPAF1 and PTRF a set of commercially available pGIPZ LV vectors targeting different sites on each gene were first screened to find an efficient shRNA target (Additional file7: Figure S5). The pGIPZ vector with the most effective shRNA was then used to transduce LNCaP cells. Cells were then cultured with puromycin to eliminate untransduced cells. Knockdown was confirmed by RT-PCR (Additional file7: Figure S5C) and then puromycin-resistant cells were cultured under androgen-deficient conditions to select for androgen-independence. PTRF is under-expressed in PCa tissue, so PTRF knockdown was expected to lead to more rapid proliferation under androgen-deficient conditions. Conversely, ATPAF1 is overexpressed in PCa tissue, so knockdown of this gene should impair proliferation under androgen-deficient conditions. After 31 days of culture in androgen-deficient medium, as expected knockdown of PTRF resulted in higher cell numbers than cells transduced with the control vector, however the increase did not reach statistical significance (p = 0.25) (Figure 5). Knockdown of ATPAF1 resulted in significantly fewer cells in androgen-deficient medium than the control by day 31 of culture (p < 0.05).

The androgen-dependent human prostate carcinoma cell line LNCaP-FGC (ATCC CRL-1740) was cultured in RPMI-1640 supplemented with 10% FBS (Atlanta Biologicals, Lawrenceville, GA) at 37°C in 5% CO₂. The LV shuttle vector, LV-SFFVEGFP, has self-inactivating long terminal repeats (LTRs), an internal spleen focus-forming virus promoter driving EGFP expression, and R6Kγ origin of replication and a neomycin phosphotransferase gene. Vesicular stomatitis virus glycoprotein pseudotyped vector stocks were made by PEI-mediated transfection of HEK-293T cells as previously described[64]. Functional titers were determined by transduction of HT-1080 fibrosarcoma cells. Cells were cultured for 14 days post vector exposure prior to use in experiments.

Genomic DNA was isolated using the Puregene Cell & Tissue kit (Qiagen Inc., Valencia, CA) and was sheared using a Hydroshear (DigiLab Inc., Marlborough, MA). The ends of the DNA were repaired using the Terminator End Repair Kit (Lucigen Corporation, Middleton, WI). Sheared fragments were ligated using T4 DNA Ligase (New England Biolabs, Inc., Ipswich, MA) and transformed by electroporation. Shuttle vector plasmids from kanamycin-resistant colonies were sequenced using primers specific to the LV LTR. A total of 288 colonies were sequenced for the in vitro cultures and 96 colonies were sequenced for each tumor. The junction between the integrated provirus and the chromosome was identified and integration sites in the human genome (hg19) were determined using PERL bioinformatics programs[21] that invoke a standalone BLAT program[65]. To be considered as an integration site the alignment score had to have a canonical LTR-chromosome junction and meet additional strict criteria as previously described[21]. A control dataset of 10,000 random sites was generated as previously described[65]. The Schnabel method of multiple-census mark-recapture was used to approximate the number of unique integration sites in the entire population prior to selection for androgen independence[66, 67].

Control and LV-SFFVEGFP transduced LNCaP cells were maintained in RPMI-1640 supplemented with 9.75% charcoal/dextran-treated FBS (CT-FBS) and 0.25% FBS (Atlanta Biologicals, Lawrenceville, GA), before being moved to and maintained in RPMI-1640 supplemented with 10% CT-FBS. Cells were counted using a Cellometer Auto T4 (Nexcelom Bioscience Inc., Lawrence, MA) and replated approximately every 2-3 weeks.

All protocols involving the use of animals were approved by the Washington State University Institutional Animal Care and Use Committee and institutional guidelines for the humane use of animals in research were followed. Male 4-8 week old NSG mice were obtained from The Jackson Laboratory (Bar Harbor, Maine). LV-mutagenized or control LNCaP cells were inoculated via subcutaneous injection. Between 1-5×10⁶ cells were suspended in 100 μL of RPMI-1640 plus 5% FBS plus 100 μL Matrigel (BD Biosciences, Bedford, MA) and injected via a 25 gauge needle into the right flank. Tumors were visually monitored or measured once to twice weekly and their volumes were calculated using the formula L×W×H×0.5236[17].

pGIPZ lentiviral shRNA vector sets of 3-6 shRNA vectors were obtained from Thermo Fisher Scientific (Waltham, MA) for the selected target genes. Gene knockdown was evaluated using the psiCHECK™-2 system (Promega Corporation, Madison, WI) using a synthesized sequence (GenScript USA, Inc., Piscataway, NJ) containing target sites for each target gene shRNA cloned into the psiCHECK™-2 vector. psiCHECK™-2 vectors containing these target fragments were co-transfected with pGIPZ shRNA vector sets for each specific target gene. The efficiency of the knockdown for each shRNA for each gene was determined with the Dual-Luciferase® Reporter Assay System (Promega Corporation, Madison, WI). The shRNA which demonstrated the most efficient knockdown of luciferase activity for each target gene was selected. Vesicular stomatitis virus glycoprotein pseudotyped vector stocks were made by PEI-mediated transfection of HEK-293T cells as described above. LNCaP cells for in vitro validation were transduced at a multiplicity of infection of 5, selected in 2.5 μg/mL puromycin, and expanded. RT-PCR was used to quantitate the expression levels of these genes in the pGIPZ transduced LNCaP cells. Total RNA from LNCaP cells transduced with a LV carrying a shRNA sequence against the candidate genes or vector with no shRNA (EV) was isolated using TRIzol reagent (Invitrogen, Carlsbad, CA). cDNA was synthesized from the total RNA using Transcription First Strand Synthesis Kit (Roche Diagnostics, Indianapolis, IN). cDNA was used as a template to amplify gene specific products using primers ATPAF1 (forward: 5′- GCCAACCAAGTTCAGCTCTT-3′ and reverse 5′-GGTTCTGGGCACATTTCAGT-3′), PTRF (forward: 5′-GAAGAGCTGATCAAGTCGGACC-3′ and reverse: 5′-GCTTCACTTCATCCTGGTAGATCA-3′) and Glyceraldehyde-3-phosphate dehydrogenase (GAPDH; forward: 5′-ATCCCATCACCATCTTCCAG-3′ and reverse: 5′- CCATCACGCCACAGTTTCC-3′). Transduced puromycin-resistant LNCaP cells were cultured in androgen-deficient medium and cell number determined as described above.

To assess the enrichment of proviruses in PCa susceptibility loci after selection for androgen-independence, a χ² test of the frequency of integration sites was used. To assess the effects of shRNA knockdown of PTRF and ATPAF1 on androgen-independence a Student’s t-test was used.