Profile of chimeric RNAs and TMPRSS2-ERG e2e4 isoform in neuroendocrine prostate cancer

Specific gene fusions and their fusion products (chimeric RNA and protein) have served as ideal diagnostic markers and therapeutic targets for cancer. However, few systematic studies for chimeric RNAs have been conducted in neuroendocrine prostate cancer (NEPC). In this study, we explored the landscape of chimeric RNAs in different types of prostate cancer (PCa) cell lines and aimed to identify chimeric RNAs specifically expressed in NEPC. To do so, we employed the RNA-seq data of eight prostate related cell lines from Cancer Cell Line Encyclopedia (CCLE) for chimeric RNA identification. Multiple filtering criteria were used and the candidate chimeric RNAs were characterized at multiple levels and from various angles. We then performed experimental validation on all 80 candidates, and focused on the ones that are specific to NEPC. Lastly, we studied the clinical relevance and effect of one chimera in neuroendocrine process. Out of 80 candidates, 15 were confirmed to be expressed preferentially in NEPC lines. Among them, 13 of the 15 were found to be specifically expressed in NEPC, and four were further validated in another NEPC cell line. Importantly, in silico analysis showed that tumor malignancy may be correlated to the level of these chimeric RNAs. Clinically, the expression of TMPRSS2-ERG (e2e4) was elevated in tumor tissues and indicated poor clinical prognosis, whereas the parental wild type transcripts had no such association. Furthermore, compared to the most frequently detected TMPRSS2-ERG form (e1e4), e2e4 encodes 31 more amino acids and accelerated neuroendocrine process of prostate cancer. In summary, these findings painted the landscape of chimeric RNA in NEPC and supported the idea that some chimeric RNAs may represent additional biomarkers and/or treatment targets independent of parental gene transcripts.


Background
Neuroendocrine prostate cancer (NEPC) is a type of rare but lethal tumor in men, which shares similar histology with small cell lung cancer or other small cell carcinomas [1,2]. NEPC is reported to be derived from treatment resistance and positive for staining of Synaptophysin (SYP), Chromogranin A (CHGA) and Enolase 2 (ENO2) [3,4]. Moreover, its two characteristics: tendency for distant metastasis, and treatment resistance; give patients worst prognosis than other types of PCa [5,6]. Unfortunately, the existing platinum-based chemotherapy only has short-term effect on NEPC and the majority of patients will die within 1 year [7][8][9]. Although numerous studies have been conducted on NEPC, and several important genes (such as CXCR2, MUC1-C and LIN28B) have been reported to be involved in the process of NEPC [5,10,11], the role of chimeric RNAs has not yet been clarified.
Chimeric RNAs, as fusion transcripts composed of exons, or fragments of exons from different genes, have been validated as cancer diagnostic markers and therapeutic targets for many years [12]. For example, the first discovered gene fusion, BCR-ABL in chronic myeloid leukemia (AML) [13], the famous oncogenic fusion, TMPRSS2-ERG in prostate cancer [14], and EML4-ALK in lung adenocarcinomas [15], etc. Chimeric RNAs can be products due to chromosomal rearrangement as the well-known examples listed above, but they can also be generated due to intergenic splicing such as SLC45A3-ELK4 in prostate cancer [16], ASTN2-PAPPA antisense in esophageal cancer [17], and BCL2L2-PABPN1 in bladder cancer [18].
Chimeric RNAs can affect tumor progression through a variety of mechanisms, including acting as long noncoding RNAs, coding for fusion proteins, and misregulating parental gene expression [13]. They represent a new repertoire of the transcriptome by expanding the functional genome, and contribute to novel mechanisms of tumorigenesis. In this study, we deep-mined chimeric RNAs from Cancer Cell Line Encyclopedia (CCLE) prostate RNA-seq dataset and characterized the landscape of chimeric RNAs in different PCa types. We then identified and validated a number of chimeric RNAs in NEPC. In the end, we found four chimeric RNAs specifically expressed in NEPC cells NCI-H660 and LAS-CPC-01. Among them, TMPRSS2-ERG (e2e4) was expressed higher in tumors and its expression predicted poor prognosis in TCGA prostate cancer study, whereas its parental genes had no such association. Importantly, compared to the most frequently detected TMPRSS2-ERG form (e1e4), e2e4 encodes additional 31 amino acids and accelerated neuroendocrine process of prostate cancer. All above supported that chimeric RNAs represent a new source of potential biomarkers or therapy targets for NEPC.

Clinical samples
Fresh tumor and adjacent normal tissues of 32 PCa patients from Sun Yat-sen Memorial Hospital were obtained to explore the full length of natural existence of TMPRSS2-ERG (e2e4). All samples were immediately snap-frozen in liquid nitrogen and stored at −80 °C until required. The use of tissues and clinical information in this study was approved by the Sun Yat-sen University's Committees for Ethical Review of Research Involving Human Subjects (approval no. SYSEC-KY-KS-2020-201). All patients submitted their written informed consents.

RNA extraction, qRT-PCR and touch-down PCR
Total RNA from cells was extracted using TRIzol reagent (Invitrogen, United States) as previously described [20]. The complementary DNA was synthesized with random hexamer primer using Verso cDNA Synthesis Kit (Thermo Fisher Scientific, United States). Quantitative real-time PCR (qRT-PCR) was carried out on ABI StepOne Plus real time PCR system (Applied Biosystems, United States) using SYBR mix kit (Thermo Fisher Scientific, United States) as previously described [21]. Primers for the 80 NEPC related chimeric RNAs were listed in Additional file 8: Table S1. Touch-down PCR (TD-PCR) was carried out using Platinum Taq High Fidelity Kit Invitrogen, United States. Primers for TMPRSS2-ERG (e1e4) and TMPRSS2-ERG (e2e4) were listed in Additional file 12: Table S5.

Agarose electrophoresis and sanger sequencing
2% Agarose Gel was made for separating 100-300 bp DNA products, and 1% Agarose Gel for longer DNA products. In detail, mix agarose (Thermo Fisher Scientific, United States) powder with 1 × TAE (Trisbase, Acetate and EDTA solution) in a microwavable flask. Microwave for 1-3 min to completely dissolve agarose followed by adding ethidium bromide (EtBr). Pour the agarose into a gel tray with the well comb in place and wait for 20-30 min until the gel completely solidified. Carefully load PCR products into the wells of the gel and run the gel at 120 V for 30 min. Axygen ® AxyPrep DNA Gel Extraction Kit (Thermo Fisher Scientific, United States) was used for gel extraction and DNA purification and followed by Sanger sequencing at Genewiz.

Protein isolation and western blotting
Protein isolation and western blotting were performed as described previously [6]. Primary antibodies: Flag Cell proliferation assay, cytotoxicity assay, colony formation assay, and migration assay For cell proliferation assay, cells (3,000 for LNCaP and 2,000 for C4-2 cells per well) were seeded in 96-well plates and cultured for 5 days. We detected the absorbance of each well at 450 nm every day using CCK8.
For colony formation assay, 20,000 C4-2 cells were seeded in six-well plates and cultured in incubator for 7 days to form macroscopic clones. After staining with 0.1% crystal violet, we compared the difference among different groups.
For cytotoxicity assay, the CCK8 assay (K1018, APExBIO, China) was used to test the viability of LNCaP and C4-2 cells treated with Enzalutamide (S1250, Selleck, China) or Docetaxel (S1148, Selleck, China). In brief, cells (4,000 for LNCaP and 3,000 for C4-2) were seeded in 96-well plates with different concentrations of Enzalutamide or Docetaxel and cultured for 96 h. Then, we calculated the IC50 according to the absorbance at 450 nm.
The 24-well Transwell chamber (8 mM, 353,097; Corning, United States) was used for the migration assay. In brief, 100,000 LNCaP cells (80,000 for C4-2) in 200 mL of 1% FBS medium were seeded in the top insert chamber, and 600 mL of medium containing 10% FBS was added into the lower chamber. The top chamber was fixed with 4% paraformaldehyde and stained with 0.2% crystal violet after 48 h incubation (12 h for C4-2). The migrated cells on the lower membrane surface of the top chamber were detected under a microscope (Nikon, Tokyo, Japan).

Statistical analyses
Quantitative results in this study were assessed by Student's t test (GraphPad, La Jolla, CA, USA). Spearman tests were used to analyze the correlation of chimeric RNAs with other genes. The Kaplan-Meier method was used to describe recurrence-free survival in those patients from TCGA and P < 0.05 was considered statistically significant after Log-rank test.

The discovery pipeline of chimeric RNAs in PCa
To discover chimeric RNAs in PCa, we downloaded raw RNA-seq data from CCLE, which contained eight prostate related cell lines. A total of 4,232 unique chimeric RNAs were predicted after Ericscript analysis. Based on the junction sites of two parental genes, we categorized these chimeric RNAs into four types. E/E: both junction sites fall onto the end/begin of known exon boundaries. M/M: both junction sites are in the middle of exons. E/M or M/E: one junction site is located at the end/begin of exon boundaries and the other in the middle of exon. We first filtered out the M/M chimeric RNAs because their lower validation rate [22]. We then removed the chimeric RNAs we previously identified in normal tissues and cells from Genotye-Tissue Expression (GTEx) study [19]. Because HPrEC LH was established from primary prostate epithelial cells, we further removed 127 chimeric RNAs which was found in this non-cancer line. After additional confirmation using UCSC genome browser, we were left with 457 chimeric RNAs, predicted specifically in PCa cell lines (Fig. 1A).
Androgen deprivation therapy (ADT) was the firstline treatment for PCa, and most of PCa patients benefited from this therapy [23]. However, it is inevitable that PCa will progress to CRPC within 2-3 years because of androgen resistance [24]. What's worse is that more than 25% of CRPC patients will evolve into a more aggressive NEPC after they become resistant to new therapeutics, such as Abiraterone or Enzalutamid [25]. Therefore, from HSPC to CRPC to NEPC is essentially an 'evolutionary' process started when patients receive clinical treatment. We hypothesized that chimeric RNAs may also play important roles in this process.
Among these PCa cell lines, three of them are androgen dependent cell lines (hormone sensitive prostate cancer, HSPC), three are considered androgen independent cell lines (castration resistance prostate cancer, CRPC), and NCI-H660 is a NEPC cell line. The numbers of chimeric RNAs found in each cell line and each type of cell line are also shown in Fig. 1A. Circos plots were used to depict the non-M/M chimeric RNAs in all PCa cell lines (Fig. 1B).

The landscape of chimeric RNAs in PCa
We subsequently analyzed the landscape of three groups of chimeric RNAs (HSPC, CRPC, and NEPC) from three angles. First, we categorized chimeric RNAs into three types, based on the chromosomal locations of parental genes. Read-Through: two parental genes are neighboring Fig. 1 The discovery pipeline and landscape of chimeric RNAs in prostate cancer. A The pipeline for discovering prostate cancer chimeric RNAs. The CCLE prostate related cell sequencing data were used for analysis. After filtering out of "M/M" fusions, GTEx fusions and PrEC LH fusions, 864 chimeric RNAs remain. 457 prostate cancer biased chimeric RNAs were identified after UCSC confirmation. B Circos plot depicting all identified chimeric RNAs in each prostate cell line. C Distributions of chimeric RNAs from HSPC, CRPC and NEPC. Chimeric RNAs were categorized based on their fusion type, junction position, and fusion protein coding potential. D Venn diagram shows the overlapping and specific chimeric RNAs among HSPC, CRPC, and NEPC. E-G Gene ontology analyses of parental genes involved in chimeric RNAs specific for HSPC, CRPC, and NEPC genes transcribing the same strand. Intra-chromosomal: two parental gens are non-neighboring and/or opposite strand genes on the same chromosome. Inter-chromosomal: two parental genes are located on different chromosomes. Similar to our previous study [18], interchromosomal was the most frequent, and read-through was the least frequent categories. In HSPC, the percentage of read-through was 6.9%, with 31.8% for intra-chromosomal, and 61.3% for inter-chromosomal. In CRPC, the percentage of read-through was 10.7%, with 33.7% of for intra-chromosomal, and 55.6% for inter-chromosomal. In NEPC, the percentage of read-through was 7.5%, with 41.9% for intra-chromosomal and 51.6% for inter-chromosomal (Fig. 1C).
Secondly, as described earlier, we divided the chimeric RNAs into E/E, M/M, E/M and M/E according to the junction location. In HSPC, the percentage of E/E was 35.5%, with 16.1% for M/E, and 48.4% for E/M. In CRPC, the percentage of E/E was 34.9%, with 9.5% for M/E, and 55.6% for E/M. In NEPC, the percentage of E/E was 39.8%, with 5.4% for M/E, and 54.8% for E/M. Of note, E/M was the most frequent in these three different types of cell lines, hinting that the 5ʹ junction sites tend to be more faithful of using canonical splicing donor sites (Fig. 1C).
Lastly, the chimeras were categorized into three categories according to the reading frame as described in our previous study [26]. In frame: the known reading frame of the 3′ gene was the same as the 5′ gene. Out of frame: the known reading frame of the 3′ gene was different from the 5′ gene. NA: both parental genes were lncRNA and junction sequences of the chimeras fall into untranslated regions (no predicted effect on the reading frame of parental genes). In HSPC, the percentage of in frame was 12.4%, with 10.1% for out of frame, and 77.4% for NA. In CRPC, the percentage of in frame was 8.3%, with 8.3% for out of frame, and 83.4% for NA. In NEPC, the percentage of in frame was 15.1%, with 10.7% for out of frame, and 74.2% for NA. Of note, NA was the most frequency in all three categories, suggesting that chimeric RNAs are more frequently affecting parental gene expressions or work as lncRNAs (Fig. 1C).

The characteristics of parental genes forming chimeric RNAs
We subsequently merged the chimeric RNAs in the three different stages of PCa progression and found 202 specific chimeric RNAs in HSPC, 155 in CRPC and 80 in NEPC (Fig. 1D).
Gene Ontology term analyses were performed for parental genes of these chimeric RNAs. For HSPC, the parental genes were mostly enrichment in cell component and metabolic related process (Fig. 1E), presumably because the distinct metabolic aberrations in prostate adenocarcinoma driven by the androgen receptor (AR) [27]. This is also consistent with reports showing glycolysis and lipid metabolism being the possible reasons for tumorigenesis [28,29]. Similarity, the parental genes from CRPC related chimeras were also mostly enrichment in metabolic related process (Fig. 1F). For NEPC, the parental genes were mostly enriched in GO terms such as neuron projection morphogenesis, synaptic vesicle transport, consistent with neuroendocrine related processes (Fig. 1G).

Validation of NEPC biased chimeric RNAs
We decided to focus on NEPC, because of its most malignant nature, with no effective treatment method. We chose all the 80 chimeric RNAs that are unique for NEPC for validation. Primers annealing to parental genes and flanking the fusion junction site were designed (Additional file 8: Table S1). We mixed cDNAs from NCI-H660 and LASCPC-01 for qRT-PCR. 22 out of 80 chimeric RNAs were amplified with bright single bands. These PCR products were extracted and submitted for Sanger sequencing ( Fig. 2A). Finally, 15 chimeric RNAs were confirmed (Fig. 2B, and Additional file 1: Figure S1).
Further in silico AGREP analysis was performed to validate if these 15 chimeric RNAs were also detected in other independent studies. Unsurprisingly, 14 out of 15 chimeric RNAs were detected in a separate NCI-H660 RNA-seq dataset, and 10 chimeric RNAs were found in another NEPC cell line MSKCC-EF1 (Table 1). In addition, in a dataset (GSE118206), 14 out of 15 chimeric RNAs were found in small cell prostate cancer, while few chimeric RNAs were found in transformed prostate basal epithelial (three chimeric RNAs) or prostate adenocarcinoma (two chimeric RNAs) [30] (Table1). Of note, six chimeric RNAs were found in bone metastases clinical samples (GSE31528), suggesting that the detection of these chimeric RNAs may have some diagnostic/prognostic value. We further calculated NE activities in seven of the eight clinical samples (one sample failed to download) from GSE31528 by using the following formula: Read counts (CHGA) × Read counts (NSE) × Read counts (SYP). Similarity, the chimeric RNA scores of these seven clinical samples were calculated by multiplying the read counts of these 15 chimeric RNAs (read counts = 0 was defined as 1 to avoid the result is 0). We defined chimeric RNA score ≥ 1000 as the high score group, in which four samples are included, and < 1000 as the low score group which contains the rest three samples. Consistent with our prediction, the high chimeric RNA score group has higher NE activities than the low score group (Additional file 2: Figure S2).
When we performed additional qRT-PCR of the 13 specific chimera on separate NCI-H660 and LAS-CPC-01 samples, four chimeric RNAs, TMPRSS2-ERG (e2e4), EEF2-SLC25A42, SNX13-ATP2C1 and FXYD2-DSCAML1 were detected in both NEPC lines (Fig. 3C). It should be mentioned that a lower band of ANAPC13-VIT could also detected in LASCPC-01, but further Sanger sequencing validated that it was a non-specific amplification. The TMPRSS2-ERG fusion involves the joining of the exon 2 (end) of TMPRSS2 and the exon 4 (start) of ERG. The parental genes of TMPRSS2-ERG were both on the negative strand of chromosome 21, and the fusion has been reported to be the product of intra-chromosomal deletion or insertional gene rearrangement [31][32][33]. The EEF2-SLC25A42 fusion involves the joining of the exon 11 (end) of EEF2 and the exon 7 (start) of SLC25A42. The parental genes of EEF2-SLC25A42 were both on the chromosome 19 but on different strands, so this chimeric RNA is a potential product of chromosomal rearrangement or trans-splicing [34]. The SNX13-ATP2C1 fusion involves the joining of the exon 14 (end) of SNX13 and the exon 23 (middle) of ATP2C1. The parental genes of SNX13-ATP2C1 are from different chromosomes, another potential product of chromosomal rearrangement or trans-splicing. The FXYD2-DSCAML1 fusion involves the joining of the exon 5 (end) of FXYD2 and the exon 2 (start) of DSCAML1. Its parental genes are adjacent to each other on the same chromosome region, so this fusion is likely to be a product of interstitial deletion or cis-splicing between adjacent genes (cis-SAGe) [22]. The detail information of these four candidate chimeric RNAs was presented in Fig. 3D and Additional file 9: Table S2.

Chimeric RNA TMPRSS2-ERG (e2e4) is associated with worse outcome in PCa
To investigate the role of these four chimeric RNAs in the progress of PCa, further AGREP analysis was performed to quantify their expression by searching for the junction sequence of these chimeras in TCGA PCa and normal RNA-seq data. Only TMPRSS2-ERG had reasonable read counts (Additional file 10: Table S3), therefore we focused on it for further analysis.
The intra-chromosomal translocation of TMPRSS2-ERG was the most prevalent fusion occurring in about 50% of PCa cases [31]. At least 17 different variants of TMPRSS2-ERG were reported in previous studies [35]. We conducted further EricScript analysis on TCGA PCa RNA-seq data and found seven different variants of the fusion (Additional file 11: Table S4). Among these isoforms, the most frequently detected is the form joining exon 1 of TMPRSS2 to exon 4 of ERG (e1e4), consistent with other studies [36]. Although many studies have been reported regarding the e1e4 form in PCa progress [37][38][39], there is very little known about the e2e4 form.
We first compared the expression difference between tumor and matched normal tissues in TCGA, and found that the e2e4 fusion was expressed significantly higher in tumor samples than in the paired normals (Fig. 4A). In addition, we performed AGREP analyses on parental TMPRSS2 gene expression using the junction sequence of its exon 2 and 3, and parental ERG gene expression using the junction sequence of its exon 3 and 4. We detected a higher level of TMPRSS2 in the tumors (Fig. 4B), presumably because the activated AR in prostate cancer directly bind to its promoter region [40]. Unsurprisingly, no statistically significant difference of parental ERG was found between tumor and normal tissues (Fig. 4C). Importantly, neither parental gene expression had a statistically significant correlation with TMPRSS2-ERG (e2e4) (Fig. 4D and E), suggesting that it is regulated differently from its parental gene expression.
Furthermore, we divided the clinical cases into two groups according to the fusion read counts and found that higher TMPRSS2-ERG (e2e4) fusion RNA expression predicted a worse outcome (Fig. 4F). Differently, even though TMPRSS2-ERG (e1e4) was also overexpressed in tumor samples, it has no statistically significant effect on the survival of PCa patients (Additional file 3: Figure  S3), that is consistent to other studies reported [41,42]. e1e4 is the downstream of AR, considering that nearly all the PCa patients will be subject to ADT, thus it is predicted that e1e4 has no significant effect on the survival of PCa patients. Additionally, RFS-free survival analysis showed that both parental TMPRSS2 and ERG had no effect on the prognosis of prostate cancer, suggesting that TMPRSS2-ERG (e2e4) may be an independent prognostic factor for PCa ( Fig. 4G and H).  To validate that the fusion indeed contains the first two exons of TMPRSS2 and the last nine exons of ERG, we designed different primers located on different exons of ERG (Fig. 5A and Additional file 12: Table S5). Due to different isoforms of ERG, we designed three reverse primers for parental ERG (Additional file 4: Figure S4A). Touch-down PCR and Sanger sequence were performed on NCI-H660 and PCa mix samples ( Fig. 5B and C), and the sequence from exon 4 to exon 12 of parental ERG was validated (Additional file 4: Figure S4B, C). The same product was also found in another small lung cancer cell line NCI-H526 (Additional file 4: Figure S4D). However, no product was amplified when we use R12-3 as the reverse primer, suggesting that ERG part matches the ENST00000398919.6 transcript. In addition, we failed to detect the same signal using adjacent normal mix samples (Fig. 5D).

TMPRSS2-ERG (e2e4) promotes docetaxel resistance and accelerates neuroendocrine process of prostate cancer
Some studies demonstrated that TMPRSS2-ERG could promote prostate cancer metastases and therapy resistance [39,43,44]. However, most of them focused on the e1e4 form. Different from e1e4, e2e4 encodes 31 more amino acids because translation starts in the TMPRSS2 exon 2 [45]. The first five amino acids are thus derived from the ORF of TMPRSS2 (exon2), and the followed 26 from the ORF of ERG (exon4) (Fig. 5E). We hypnotized that the differences of these amino acids may cause different functions. We thus constructed the overexpression plasmids, and verified their protein coding potential by Western blot (Fig. 5F). Consistently, we detected two bands with ERG antibody in two NEPC lines NCI-H660 and LASCPC-01. One is at the size of wild type ERG (e1e4 form), and the other has higher molecular weight which is consistent with the size of the e2e4 form (Additional file 5: Figure S5).
To investigate the functional difference between e1e4 and e2e4, further experiments were performed. Both forms promoted the migration of LNCaP and C4-2 cells (Fig. 6A), consistent with other reports [38,46]. However, neither of them influenced the proliferation or enzalutamide response in prostate cancer cells (Fig. 6B-D). This is consistent with the fact that 5' regulatory elements of TMPRSS2 enables the fusion to respond to AR pathway inhibition [47].
Elevated neuroendocrine markers and docetaxel resistance are the two notable features of NEPC [5,48]. We then tested whether e2e4 affects these features. In LNCaP derived CRPC C4-2 cells, expression of e2e4 promoted docetaxel resistance while e1e4 did not ( Fig. 6E and F). Moreover, e2e4 was turned on while e1e4 was not significantly changed in VCaP cells after docetaxel treatment ( Fig. 6G and Additional file 6: Figure S6). Importantly, neuroendocrine markers such as CHGA, NSE and SYP were significantly upregulated when e2e4 was overexpressed in LNCaP and C4-2 (Fig. 6H). Although e1e4 also caused the increase of some markers to certain extents, it was far less dramatic than e2e4.
We further evaluated the frequency and relative amount of the two forms of TMPRSS2-ERG in our clinical samples. We performed qPCR followed by agarose electrophoresis on 32 clinical samples and found that 23 samples expressed e2e4 specifically and six samples express both e1e4 and e2e4, while only three samples express neither form (example in Additional file 7: Figure  S7A). The detection rate of e1e4 is similar to the findings of Kong et. Al, who reported around 20% positive rate in Asia populations [49], which is much lower than that in the United States (around 50%) [50]. It is thus interesting that e2e4 could be detected in most of Asia population samples. We further used the absolute standard dilution method to determine the relative numbers of e1e4 and e2e4 in our clinical samples and NCI-H660 cell line as our previous study did [26,51]. The results showed that e1e4 and e2e4 shares a similar copy number in sample NO.1, 5 and 6. e1e4 has a higher copy number in sample NO.2, 3 and 4 than e2e4 (Additional file 7: Figure S7B), suggesting that the amount of e1e4 and e2e4 varies individually. In NCI-H660, the copy number of e2e4 is lower than that of e1e4, which is consistent with the finding above at the protein level (Additional file 4: Figure S4 and Additional file 7: Figure S7B).

Discussion
In this study, we provided the landscape view of chimeric RNAs in PCa cells, focusing on NEPC. We uncovered 13 chimeric RNAs specific to NCI-H660, and four of them in both NCI-H660 and LASCPC-01. We then investigated the clinical implication of a less studied isoform of  TMPRSS2-ERG (e2e4) promotes docetaxel resistance and accelerates neuroendocrine process of prostate cancer. A Representative images and histogram of migration assays using LNCaP and C4-2 cells overexpressing TMPRSS2-ERG e1e4 or e2e4. B-C The CCK8 assay was used to measure cell viability in LNCaP and C4-2 cells when e1e4 or e2e4 was overexpressed. D The CCK8 assay was used to determine IC50 of enzalutamide in LNCaP cells with e1e4 or e2e4 overexpression. E Colony formation assay tested cell viability in C4-2 cells after docetaxel treatment when e1e4 and e2e4 were overexpressed. F The CCK8 assay was used to determine IC50 of docetaxel in C4-2 cells with e1e4 or e2e4 overexpression. G Gel images of RT-PCR product of e1e4 or e2e4 after using various concentrations of docetaxel treatment. H Representative image of the Western blotting analysis of CHGA, NSE and SYP levels after e1e4 or e2e4 overexpression in LNCaP and C4-2 cells. ****p < 0.0001, ***p < 0.001.**p < 0.01 TMPRSS2-ERG (e2e4 form). For chimeric RNA identification, RNA-Seq needs to have certain read length and reach sufficient read depth. Unfortunately, most RNA-Seq on NEPC clinical samples do not meet these requirements. This is the reason we started from CCLE dataset, followed by AGREP and RT-PCR to validate in silico and experimentally. This in a way limited the discovery for novel chimeric RNAs. In the future, when qualified clinical sequencing data become available, we envision more chimeric RNAs can be discovered.
It was reported that elevated neuroendocrine markers, treatment resistance and enhanced invasion capabilities were the three most important features of NEPC [5,48]. We here also found that TMPRSS2-ERG (e2e4) plays an important role on docetaxel resistance, migration ability and the alteration of neuroendocrine markers. Even though some published studies have validated that overexpressed TMPRSS2-ERG or ERG could promote cancer invasion ability and/or drug resistance [39,43,44], most reports on TMPRSS2-ERG are on the e1e4 form. However, since e1e4 form does not involve coding region of TMPRSS2, forming the fusion has been considered a mechanism to solely drive the overexpression of ERG [35]. In the situation of e2e4 form, protein coding sequence of TMPRSS2 and additional sequence of ERG are included. The detection of the additional band in Western blot supports the protein isoform. However, to prove the band is indeed the correct isoform, more experiments such as siRNA silencing specifically for the e2e4 form is needed. How exactly the new amino acids works in e2e4 form is one topic deserves further investigation. Considering the high detection rate of e2e4 form in our patient population and our finding above, we believe that e2e4, rather than e1e4, is a more important cause of prostate cancer progression at least in Asian populations. That is also a potential reason for the higher MR/IR (mortality-to-incidence rate ratio) in Asia population (40%) than that in Europe (18%), Northern America (10%) and worldwide (25%) [52].
We believe that the progression of prostate cancer to NEPC is a gradual process. Interestingly, we observed that with the increase of tumor malignancy, the number of related chimeric RNA expression also increases. For example, chimera SNX13-ATP2C1 could be detected in both small prostate cancer samples and bone metastasis prostate cancer samples (Table 1). SNX13-ATP2C1 is a novel chimeric RNA, first discovered in this study. SNX13 encodes Galpha(s)-specific guanosine triphosphatase-activating proteins which has been linked to heterotrimeric G protein signaling and vesicular trafficking [53]. ATP2C1 encodes a type of P-type cation transport ATPases catalyzing the hydrolysis of ATP coupled with the transport of calcium ions. It has been reported to be related to Hailey-Hailey disease, but not tumorigenesis [54]. The fusion is predicted to contain the first 14 exons (totally 26) of SNX13 and the last five (totally 27) exons of ATP2C1, encoding also an in-frame fusion protein. It is not yet known whether it plays a role in promote tumor progression.
We acknowledge that more in vitro studies and in vivo animal models investigating the implications of TMPRSS2-ERG e2e4 form on neuroendocrine prostate cancer conversion and drug resistance are needed. This is one of the areas of our future directions. Also in this study, we did not pursue the rest chimeric RNAs, which are also all novel, because of their relatively low read counts in publicly available NEPC dataset and scarce of NEPC clinical samples. It is thus worthwhile for future research to examine their expression and functions in large scale of NEPC datasets or clinical samples.