Introduction

Prostate cancer (PCa) is the most frequent tumor entity in men in developed countries [1]. From the perspective of the clinician, a differentiation between clinically relevant and insignificant tumors is the most urging questing in localized PCa. In metastatic and castration-resistant tumors, clinical focus moves on to the choice of the most suitable treatment approach in a growing armamentarium of therapeutic options. For both questions, molecular markers beyond prostate-specific antigen (PSA) serum level offer great opportunities.

Recent large studies have aimed to elucidate the genomic, transcriptomic and epigenetic landscape of PCa [2, 3]. Unlike other tumors, such as colorectal carcinoma or breast cancer, which typically present driver gene mutations in a broad number of cases, PCa is merely characterized by larger genomic alterations, being ETS-gene fusions (around 50%, with most of these cases having a TMPRSS2:ERG fusion), loss of PTEN (20–40%) and changes in the epigenetic profile [4, 5]. In addition to genetic markers, a large number of potential protein markers have also been described [6]. Gene expression analyses can be used to identify different tumor expression patterns and divide them into different molecular subtypes [3]. Such classifications can be decisive for risk stratification and therapy decision-making. For example, in urinary bladder carcinoma, tumors with a basal phenotype are more likely to respond to neo-adjuvant therapy, resulting in better clinical [7]. Such strategies are currently discussed in PCa for neo-adjuvant therapy and androgen deprivation [8, 9]. Furthermore, molecular profiling can result in the identification of novel biomarkers.

Of these, fatty acid-binding protein 5 (FABP5), which is typically associated with epithelial tissues, has been shown to be overexpressed in PCa compared to benign prostatic hyperplasia (BPH) using immunohistochemistry (IHC) [10,11,12]. In addition, protein-level overexpression of FABP5 in transurethral resection of the prostate (TUR-P) is associated with tumor stage and patient survival. Yet, the value of FABP5 mRNA gene expression has only been studied in a small patient cohort [13].

In the present study, we compared FABP5 mRNA expression determined by qRT-PCR analysis in tissue samples from patients who underwent radical prostatectomy or palliative TUR-P (pTUR-P) with benign controls. Furthermore, FABP5 expression was correlated with ERG expression. To validate our findings, we reanalyzed the mRNA expression datasets and correlated FABP5 expression with TCGA molecular subtypes and druggable downstream pathway components.

Materials and methods

Cohorts and patient samples

A cohort of 57 patients who underwent radical prostatectomy (mean age ± SD 62.7 ± 7.1 years) and 50 patients (mean age ± SD 75.9 ± 6.8 years) who underwent a pTUR-P in the Department of Urology of the Medical Faculty Mannheim of the University Heidelberg were analyzed (MA cohort). Histologically proven tumor-free prostate tissue specimen from 14 patients who underwent cystoprostatectomy or TUR-P served as controls (mean age ± SD 66.6 ± 11.9). In addition, a commercially available cDNA array (Origene, Rockville, MD, USA) consisting of 40 PCa (mean age ± SD 62.8 ± 8.2) and 8 benign control samples (mean age ± SD 64.0 ± 10.9) were used to determine the expression of FABP5 using qRT-PCR. Patient characteristics are shown in Supplementary Table 1. All experiments conducted in this retrospective analysis were in accordance with the institutional ethics review board (ethics approvals 2013-845R-MA, 2014-592 N-MA).

For further validations, three large microarray and sequencing datasets from cBioPortal (www.cbioportal.org) were reanalyzed (MSKCC, Cancer Cell 2010 n = 131 localized PCa, n = 19 metastatic; TCGA, Cell 2015 n = 333 and mCRPC, Nature Medicine 2016 n = 114) [2, 3, 14].

RNA extraction, cDNA synthesis and qRT-PCR from patient samples

Tumor-bearing or tumor-free formalin-fixed paraffin-embedded prostate tissue specimen of patients treated in our department were sectioned, stained with hematoxylin and eosin, and reviewed by board-certified uro-pathologists (CAW, MG). Areas with at least 70% of tumor or tumor-free areas from control patients were marked and macrodissected from subsequent unstained 10-μm cuts. RNA was extracted using the XTRAKT FFPE kit (Stratifyer, Cologne, Germany), as recommended by the manufacturer. Finally, the RNA was eluted in 100 μl of elution buffer. RNA samples were stored at − 80 °C.

To receive a greater yield of target-specific transcripts and to reduce contamination with other amplified cDNA sequences, a multiplexed specific cDNA synthesis with equimolar pooling of transcript-specific reverse PCR primers (housekeeping gene Calm and target genes FABP5, ERG and AR, Supplementary Table 2) was used. As reverse transcriptase, Superscript III (Life technologies, Carlsbad, CA, USA) was used at 55 °C for 120 min, followed by an enzyme inactivation at 70 °C for 15 min. cDNA was immediately used for qRT-PCR or stored at − 20 °C. qRT-PCR analyses of the cDNA array were performed using the same primers. 40 cycles of amplification with 3 s of 95 °C and 30 s of 60 °C were conducted on a Step One Plus qRT-PCR cycler (Applied Biosystems, Waltham, MA, USA). mRNA expression in all samples was calculated using the 40−(ΔCt) method.

In silico validation and statistics

FABP5 expression was correlated with ERG-fusion status, TCGA molecular subtypes, known cancer driver mutations and the expression of druggable downstream pathway components. ANOVA, Mann–Whitney test, χ2 test and Spearman correlation were used as appropriate for statistical analysis in Prism 5 (GraphPad Software, La Jolla, CA, USA).

Results

FABP5 mRNA expression is elevated in patients with localized and advanced PCa

A significantly higher expression of FABP5 was observed in localized (n = 56, gene expression was not measurable in one patient, p = 0.010) and advanced (n = 50, p = 0.002) PCa tissue samples from the MA cohort compared to benign (n = 14, median expression with IQR: 39.46 (38.70–40.68) and 39.64 (38.55–40.31) vs. 38.68 (38.04–39.00); ANOVA p < 0.001, Fig. 1a). In these localized tumors, FABP5 expression was independent of Gleason score, T stage, the presence of lymph node metastases and PSA level (Supplementary Figure 1A–D). Furthermore, the FABP5 expression was analyzed in 40 patients of the Origene cohort with localized PCa by qRT-PCR in a cDNA array. FABP5 expression was detectable in 34 patients and showed an overexpression compared to benign controls (median expression with IQR: 38.40 (37.60–40.16) vs. 36.76 (36.06–37.63), p = 0.003, Fig. 1b), again independent of T stage and Gleason score. To validate the expression of FABP5 in a larger cohort, the mRNA expression microarray dataset by Taylor et al. (MSKCC) was analyzed in silico [2]. In this dataset, 131 patients with primary PCa showed an FABP5 overexpression compared to benign controls (dashed line) independent from T stage (Fig. 1c) and Gleason score (Fig. 1d) as well as the presence of lymph node metastases and serum PSA level (Supplementary Figure 1E–F). FABP5 expression was not significantly associated with clinical outcome in patients with localized PCa from the MA cohort (BCR-free log-rank p = 0.151, Supplementary Figure 1G) as well as in the MSKCC dataset (BCR-free log-rank p = 0.483; Supplementary Figure 1H).

Fig. 1
figure 1

FABP5 is overexpressed in patients with localized and advanced PCa (ANOVA p < 0.001) (a) and in tissue samples with localized PCa (n = 34) independent of tumor stage (p = 0.003) (b). The MSKCC cohort with 131 primary PCa was reanalyzed and showed an overexpression of FABP5 compared to benign controls, regardless of T stage (c) and Gleason score (d), (dashed line: baseline z-score = 0 as reference of average expression in benign controls). The Mann–Whitney test was used for statistical analysis (*p < 0.05; **p < 0.01)

Full size image

FABP5 is overexpressed in ERG fusion-negative tumors which carry FABP5 amplifications more frequently

Beside clinical and histomorphological classifications, PCa can be subdivided into molecular subtypes. According to the TCGA classification system [3], three of these subtypes are characterized by ETS-gene fusions, namely ERG-, ETV1 and ETV4-fusions. ERG fusions, being the largest of these groups, typically go along with an ERG overexpression. Thus, the median of the ERG expression was determined by qRT-PCR and patients with localized and advanced PCa from the MA cohort (n = 106 with detectable expression) were grouped into ERG low and ERG high depending on their ERG expression. PCa with low ERG expression showed a trend towards a higher expression of FABP5 compared to PCa with high ERG expression (median expression with IQR: 39.60 (38.78–40.74) vs. 39.41 (38.39–39.88), p = 0.063, Fig. 2a). In a subgroup analysis, no difference in the FABP5 expression was seen among localized tumors (median expression with IQR: 39.41 (38.46–40.85) vs. 39.46 (38.74–40.56), Supplementary Figure 2A), but among patients who underwent pTUR-P, a significantly higher FABP5 expression was observed in PCa with low ERG expression (median expression with IQR: 40.25 (38.92–40.76) vs. 39.41 (38.33–39.83), p = 0.013, Supplementary Figure 2B). Across all tumor samples, expression of FABP5 and ERG showed no correlation (Rho = − 0.139, Mann–Whitney test p = 0.159, Supplementary Figure 2C). To further investigate these results, the MSKCC cohort was reanalyzed for FABP5 expression in ERG fusion-positive and -negative tumors. In primary PCa (n = 131), FABP5 showed a significant overexpression in ERG fusion-negative tumors compared to ERG fusion-positive tumors (median expression z-score with IQR: 3.25 (0.77–6.66) vs. − 0.05 (− 1.12–0.93), Mann–Whitney test p < 0.001, Fig. 2b). In a small group of metastatic tumors from the same cohort (n = 19), FABP5 gene expression was slightly higher in ERG fusion-negative tumors as well, but without reaching significance (median expression z-score with IQR: 4.97 (3.00–6.74) vs. 1.67 (− 1.07–5.30), p = 0.149). In primary tumors, Spearman correlation revealed a strong negative correlation between FABP5 and ERG expression (Rho = − 0.5898, p < 0.001). To validate this observation, two additional datasets from cBioportal were reanalyzed. In the TGCA dataset, which encompassed 333 cases with primary PCa, FABP5 was significantly overexpressed in ERG fusion-negative tumors (median expression z-score with IQR: 0.06 (− 0.30–0.82) vs. − 0.39 (− 0.43–0.32), p < 0.001, Fig. 2c). Figure 2d describes the FABP5 expression pattern dependent on the ERG-fusion status in a cohort of mCRPC patients. The expression of FABP5 was low in most of the tumors, yet an elevated FABP5 gene expression (z-score > 2) was seen in 9 ERG fusion-negative tumors and not in ERG fusion-positive tumors (median expression z-score ± IQR: − 0.4 ± 0.33 vs. − 0.37 ± 2.39, p = 0.736).

Fig. 2
figure 2

In patients with localized and advanced PCa, FABP5 showed a trend towards a higher expression in ERG low PCa compared to ERG high PCa (p = 0.063) (a). Significant overexpression of FABP5 was observed in ERG fusion-negative primary tumors compared to ERG fusion-positive tumors in the MSKCC cohort (p < 0.001) (b). Reanalysis of 333 cases with localized PCa from the TCGA dataset showed differential expression of FABP5 depending on ERG-fusion status (p < 0.001) (c). FABP5 expression was elevated in some ERG fusion-negative tumor samples in mCRPC (d). Furthermore, ERG fusion-negative tumors from the TCGA dataset had a higher frequency of FABP5 amplifications (Chi2p = 0.002) (e). The higher FABP5 expression is associated with the occurrence of gene amplifications (p < 0.001) (f). The Mann–Whitney test was used for statistical analysis (***p < 0.001)

Full size image

Next, all cancer datasets (n = 227), listed in cBioPortal were screened for genetic and genomic alterations of FABP5. Among these, amplifications were far most frequent. Amplification rates of more than 10% of altered cases were almost exclusively observed in PCa and breast cancer datasets. Only datasets of the ASC project (14.29%), malignant peripheral nerve sheath tumors (13.33%) and liver cancer (10.63) showed a similar amplification rate (Supplementary Figure 3A). Mutations and deletions were rarely observed. In PCa datasets the FABP5 amplification frequency varied between 1.52% and 40.35%. The highest frequencies were observed in highly advanced CRPC (Supplementary Figure 3B). In the TCGA dataset, ERG fusion-negative tumors had a higher frequency of FABP5 gene amplifications (χ2p = 0.002, Fig. 2e). More than one-third (37.02%) of these ERG fusion-negative tumors were affected. In contrast, only 17.77% of the ERG fusion-positive tumors carried gene amplifications. In addition, compared to diploid tumors, the FABP5 expression was significantly higher when a FABP5 amplification was present (median expression z-score with IQR: amplification 0.02 (− 0.33–0.75) vs. diploid − 0.35 (− 0.41–0.11), p < 0.001, Fig. 2f).

Association of FABP5 overexpression with molecular subtypes, genetic alterations and downstream pathway components

Next, FABP5 expression was correlated with the PCa molecular subtypes proposed by the TCGA cohort. Overexpression of FABP5 was mainly found in tumors with SPOP and FOXA1 mutations and not in ETS-fusion subtypes (ERG, ETV1, ETV4) (Fig. 3a). In addition, tumors with a FABP5 expression z-score of > + 1 showed a predominance of SPOP and FOXA1 mutations. Other cancer driver mutations occurred only rarely (Fig. 3b).

Fig. 3
figure 3

Among the TCGA molecular subtypes, FABP5 overexpression was mainly found in the SPOP, FOXA1 and the subtype not associated with typical genetic alterations (a). Vice versa, tumors with a FABP5 expression z-score of > + 1 showed a predominance of SPOP and FOXA1 mutations. Other cancer driver mutations rarely occurred in these tumors (b). In the TCGA cohort, FABP5 is negatively correlated with ERG gene expression. No correlation with AR was observed, but a positive correlation with androgen signaling responsive genes KLK2 and KLK3 and the ERG-fusion partner TMPRSS2. Furthermore, PPARA, PPARD and VEGFC were positively and VEGFA negatively correlated with FABP5 expression (c). Spearman correlation showed that AR is significantly correlated with ERG (d) and FABP5 (e) in the MA cohort

Full size image

Furthermore, FABP5 was correlated with genes involved in the AR pathway and angiogenesis, which are crucial for progression and could serve as potential targets for therapeutic interventions. As in the MSKCC cohort, also in the TCGA cohort a strong negative correlation of FABP5 and ERG expression was observed (rho = − 0.566, p < 0.001). Interestingly, there was no correlation with AR, but a positive correlation with the androgen signaling responsive genes KLK2 and KLK3 and the ERG-fusion partner TMPRSS2 (Fig. 3c). As previously reported, AR and ERG showed a positive correlation. This was also the case in patients from the MA cohort (Rho = 0.219, p = 0.025, Fig. 3d) and the tumors of these patients showed also a positive correlation between FABP5 and AR (rho = 0.347, p < 0.001, Fig. 3e).

In the TCGA dataset, the mRNA of two of the three PPAR nuclear receptor subtypes (PPARA and PPARD), whose protein products are the main ligands of FABP5, showed an inverse correlation with FABP5 gene expression. Furthermore, FABP5 is negatively correlated with VEGFA, but positively correlated with VEGFC (Fig. 3c).

Discussion

Recent studies have shown that FABP5 is overexpressed in PCa on the protein level [10, 11, 15]. In the present study, FABP5 gene expression was analyzed by qRT-PCR, which is a possible alternative to IHC as it is sensitive, objective and not affected by inter-observer variability [16, 17]. Overexpression of FABP5 was observed both in prostatectomy and pTUR-P samples from our institution (MA cohort), which confirms previous studies in PCa [13].

PCa is associated with diverse genomic alterations such as fusion genes, gene amplifications and mutations. In 2005, Tomlins et al. described the fusion gene TMPRSS2:ERG as a predominant genomic alteration in PCa (around 50%) [4]. In contrast to relevant fusion genes in other malignant diseases such as the Philadelphia chromosome BCR:ABL in chronic myeloid leukemia, TMPRSS2:ERG does not lead to a novel fusion protein in PCa. The fusion with TMPRSS2 results in overexpression of ETS transcription factors such as ERG and ETV1, which influence the regulation of cellular processes such as cell proliferation, differentiation or apoptosis and are associated with an unfavorable outcome. To date, multiple studies have reported a correlation between ERG overexpression and an unfavorable PCa outcome [18, 19]. The reanalysis of the 333 cases with localized PCa from the TCGA dataset showed a negative correlation of FABP5 with ERG, but a positive correlation with the fusion partner TMPRSS2. In addition, ERG showed a positive correlation with the AR gene in the TCGA dataset and the MA cohort. The fusion gene TMPRSS2:ERG is mainly regulated by AR [4]. Interestingly, FABP5 did not show a correlation with AR in the TCGA dataset, but was positively correlated with the androgen signaling responsive genes KLK2 and KLK3. In contrast, FABP5 is positively correlated with AR in the MA cohort. These differences may occur due to the different cohorts (localized vs. localized/TUR-P) and technical differences of gene expression analysis (RNAseq, mRNA-microarray, qRT-PCR). Besides AR, other hormones play also a role in PCa progression. The estrogen receptor α triggers the tumor-promoting function of the TMPRSS2:ERG fusion. In another study, Senga et al. could show that FABP5 interacts with the estrogen-related receptor α (ERRα) [20]. This could indicate that PCa can bypass AR to promote its growth, using estrogens. In the MSKCC and the TCGA cohort, FABP5 and ERG correlate negatively. The overexpression of FABP5 in ERG fusion-negative tumors is associated with a higher copy number variation of FABP5 and with SPOP and FOXA1 mutations. Blattner et al. also described an inverse association of SPOP mutations and ERG rearrangement [21]. In 2012, another study postulated the FABP5 gene itself to be potentially involved in fusion genes in PCa, with KLK3, which is coding for PSA, being a potential fusion partner [22]. Since this study used only a bioinformatics approach, there is no experimental evidence to support this hypothesis, yet.

FABP5 is correlated with two of the PPAR receptors PPARA and PPARD as well as VEGFA and VEGFC, which are involved in the angiogenesis and are essential for tumor growth and progression [12, 23]. Pan et al. were able to show that FABP5 is higher expressed in hepatocellular carcinoma and that mRNA expression is positively correlated with VEGFA. Downregulation of FABP5 inhibits the IL6/STAT3/VEGFA pathway and angiogenesis [24]. Another study from Al-Jameel et al. showed that the chemical inhibitor SBFI26 of FABP5 suppresses proliferation, migration and invasiveness in vitro by affecting the signal axis of FABP5-PPARγ-VEGF [25]. FAPB5 appears to indicate increased activity in PPAR and VEGF signaling, which may act as a potential drug target in PCa. In preclinical studies, knockdown of the coding gene FABP5 resulted in a reduced growth of prostate cancer cells and xenografts [10, 23, 26]. This could be confirmed in stable knockout cell lines [11]. A suppression of FABP5 gene expression, along with suppression of cell growth and invasion, was also one of the main effects induced by several procyanidins, members of the tannin family, used as anti-neoplastic drugs in in vitro prostate cancer models [27]. Furthermore, FABP5 was shown to be a regulator of lipid composition and metabolism in highly aggressive prostate and breast cancer [28]. Both tumor entities are highly depending on stimulation with steroid hormones. Fitting to this FABP5 was shown to be a direct interaction partner of ERRα. This interaction leads to an increased expression of ERRα target genes with impact on the cellular energy metabolism [20]. Unfortunately, nothing is known yet about the expression of FABP5 and subsequent signaling cascades in dependency of antihormonal therapy.

A proteomic study identified FABP5 as a differentially expressed marker in lymph node-positive PCa tumor samples, confirming its potential relevance as a predictive marker [13]. A study by Fujita et al. identified FABP5 as a potential extracellular vesicle-based protein marker for the detection of high-risk PCa in urine samples [29]. In our own analyses, we could identify FABP5 to be present on extracellular vesicles of PCa cell lines [30].

In summary, the analysis of different cohorts with localized PCa revealed that FABP5 gene expression is not associated with clinical outcome, and therefore does not seem to be suitable as a single marker for risk stratification or outcome prediction. However, on the protein level data from the literature point to a potential role of FABP5 as a marker for PCa diagnosis and prediction of high risk localized PCa and the presence of lymph node metastases. Further studies, especially liquid biopsy based, are warranted to prospectively validate these findings.

Besides this, the characterization of new therapeutic targets, especially for the treatment of advanced PCa after failure of first- and second-generation antihormonal therapy is of clinical relevance. Due to its role in tumor cell metabolism and as it is part of the FABP5-PPAR-VEGF signaling axis, relevant for angiogenesis and tumor progression, FABP5 might serve as a novel therapeutic target, especially in ETS fusion-negative tumors, in which its coding gene FABP5 was shown to be frequently overexpressed.