ERG rearrangement and protein expression in the progression to castration-resistant prostate cancer

Abstract

Background

Approximately half of the prostate carcinomas are characterized by a chromosomal rearrangement fusing the androgen-regulated gene TMPRSS2 to the oncogenic ETS transcription factor ERG. Aim of this study was to comprehensively analyze the role and impact of the ERG rearrangement and protein expression on the progression to castration-resistant (CR) disease.

Methods

We used a tissue microarray (TMA) constructed from 114 hormone naive (HN) and 117 CR PCs. We analyzed the ERG rearrangement status by fluorescence in situ hybridization and the expression profiles of ERG, androgen receptor (AR) and the proliferation marker Ki67 by immunohistochemistry.

Results

Nearly half of the PC tissue specimens (HN: 38%, CR: 46%) harbored a TMPRSS2-ERG gene fusion. HN PCs with positive translocation status showed increased tumor cell proliferation (P<0.05). As expected, TMPRSS2-ERG gene fusion was strongly associated with increased ERG protein expression in HN and CR PCs (both P<0.0001). Remarkably, the study revealed a subgroup (26%) of CR PCs with ERG rearrangement but without any detectable ERG protein expression. This subgroup showed significantly lower levels of AR protein expression and androgen-regulated serum PSA (both P< 0.05).

Conclusions

In this study, we identified a subgroup of ERG-rearranged CR PCs without detectable ERG protein expression. Our results suggest that this subgroup could represent CR PCs with a dispensed AR pathway. These tumors might represent a thus far unrecognized subset of patients with AR-independent CR PC who may not benefit from conventional therapy directed against the AR pathway.

Introduction

Prostate cancer (PC) is the most frequently diagnosed cancer among males in western countries and the second leading cause of cancer-related death.¹ Although the mortality of PC has decreased mainly due to earlier detection, this disease still accounts for 9% of the total cancer deaths. Most PCs are nowadays diagnosed at an early stage. They initially depend on androgens for their growth and are thus referred to as hormone naive (HN) PC. Based on this dependence, the standard treatment for patients harboring these tumors is androgen-deprivation therapy (ADT). Although this therapy is initially effective, most of the treated tumors recur after a few months or years as castration-resistant (CR) PC. Mechanisms responsible for this progression are not fully understood.

PC research was revolutionized by the discovery of the TMPRSS2-ERG gene fusion in 2005.² Later on, it was realized that this rearrangement was part of a whole family of gene fusions that connect the promoter region of androgen-regulated genes, most frequently the TMPRSS2 (transmembrane protease inhibitor 2) with transcription factors of the ETS (erythroblastosis virus E26 transforming sequence) family of transcription factors.^3,4 Of these fusions, the rearrangement involving the genes TMPRSS2 and ERG is by far the most common (> 90%) and is present in approximately 50% of prostate tumors.⁵ The two involved genes are <3 Mb apart on chromosome 21, and their fusion can occur through various rearrangements mechanisms, most frequently deletion of the intervening region on chromosome 21 (reviewed in Tomlins et al.⁶ and Perner et al.⁷). This rearrangement results in an androgen regulation of the ERG gene, leading to the overexpression of this gene in prostatic adenocarcinoma (reviewed in Sreenath et al.⁸). Despite the extensive studies about the role of the ERG rearrangement and expression, its clinical significance remains controversial.^9,10 Recently, Minner et al.¹¹ did not observe any prognostic impact in a larger cohort of radically operated PCs.

In CR PC, ERG rearrangement has been shown to prevail in 34– 45% of the tumors.^12,13 Very recently, we observed a higher frequency of ERG rearrangements (45%) in recurrent CR PC specimens and a lower frequency of 25% in metastatic CR PCs.¹³ In contrast to the rearrangement, which is present on a genomic level, ERG protein expression is more dynamic, as it depends on the presence and activation of the androgen receptor (AR). In the CR disease state, the tumor may adapt to very low levels of androgens. Thus, it is not evident if these levels are sufficient for the activation of ERG transcription. Data from these investigations have provided controversial results: whereas in some CR PC xenograft experiments ERG mRNA expression was not detectable,¹⁴ others have shown ERG protein expression in rearranged CR PC samples and xenografts.^15,16

In the present study, we used a tissue microarray (TMA) consisting of 231 locally advanced PCs that were collected either before (HN) or after recurrence to ADT (CR). We used this TMA to comprehensively interrogate and characterize the ERG protein expression and rearrangement comparing HN and CR PCs. We included standard markers into our analyses known to be relevant in PC, such as AR protein expression and Ki67 labeling. Here, we show that a considerable fraction of ERG-rearranged CR PCs loses ERG protein expression. We hypothesize that this might be due to a dispensed AR pathway.

Materials and Methods

TMA and patients

The use of clinical specimens for the construction of the castration resistance TMA (crTMA) was approved by the ethical committee of the University and the University Hospital of Basel, Basel, Switzerland. The crTMA was manufactured as previously described.¹⁷ Briefly, tissue cylinders with a diameter of 0.6 mm were punched from the ‘donor’ tissue blocks containing the TURP specimens using a home-made, semi-automatic robotic precision instrument. Three cores from each specimen were arrayed. The composition of the crTMA has been previously described and is summarized in Supplementary Table S3.¹⁸ Briefly, it is composed of 697 spots from 231 TURPs from a total of 202 patients treated with advanced, locally obstructive PC. In addition, it contains 12 specimens from BPH. Castration resistance was defined as locally obstructive recurrence and/or PSA-recurrence during ADT.

Immunohistochemistry (IHC) and fluorescence in situ hybridization (FISH)

IHC was performed according to the standard indirect immuno-peroxidase procedures. The primary antibody was omitted for negative controls. All slides were read manually by an experienced pathologist (LB). Data from AR and Ki67 were available from a previous study on the same TMA block.¹⁸ Briefly, the antibodies M3562 and M7240 (both DAKO, Carpinteria, CA, USA) were used for AR and Ki67 staining, respectively. The anti-ERG mouse monoclonal antibody 9FY was from Biocare Medical (Concord, CA, USA).¹⁹ FISH analysis for detection of ERG rearrangement was performed as previously reported.¹³ Images were obtained by usage of the AXIO Imager.A1 microscope equipped with an AxioCam and the AxioVision 4.6 software (all from Zeiss, Jena, Germany).

Cutoffs, data analysis and statistics

For protein expression analysis of AR, Ki67 and ERG, the percentage of positive tumor cells was noted by an experienced pathologist (LB) and used as score.¹⁸ For dichotomous stratification of ERG, samples with any specific positivity were considered as ERG positive (that is, cutoff >0) and were considered negative in reference to endothelial ERG-positive staining.^19,20 Cutoffs for definition of low or high for Ki67 labeling index were used as previously described.¹⁸ For correlation studies between different markers, every evaluable spot was considered for the analysis, that is, the analyses were performed on a ‘spot-by-spot’ basis. All other analyses (that is, descriptive tables, association with clinical data, such as treatment status, cM, cT and survival data) were performed on a ‘one-value-per-biopsy’ basis, thereby considering only one value per biopsy/specimen. If more than one spot/value per biopsy/specimen was evaluable, the spot with the maximal score was included in the analysis.

Statistical analysis was performed with the R Framework Version 3.0.121 including the ‘coin’ package.²² Differences between two groups were analyzed with the Wilcoxon's rank-sum test; differences between more than two groups were analyzed using the Kruskal–Wallis rank-sum test for metric variables, for example, expression score. χ² and Fisher's exact test were used to analyze contingency tables. Survival curves were plotted by usage of the Kaplan–Meier method, and differences were assessed using the log-rank test. P-values < 0.05 were considered as statistically significant.

Results

ERG expression and TMPRSS2-ERG rearrangement in HN and CR PC and association with clinicopathological features

To interrogate ERG protein expression and rearrangement by IHC and FISH, respectively, in the context of progression to castration resistance, we used the recently described crTMA that was constructed for this purpose.¹⁸ In addition, we included IHC data for AR and Ki67 expression from a previous study.¹⁸

For ERG expression analysis, 78 (68%) and 88 (77%) out of 114 HN and 117 CR TURPs, respectively, were evaluable (Figure 1). Of note, only cases with unequivocal nuclear staining for ERG in endothelial cells were considered as evaluable. ERG FISH analysis was successful in 94 (83%) and 94 (81%) of the 114 and 117 HN and CR PCs, respectively. ERG protein positivity, as well as the presence of ERG rearrangement, showed similar distributions between HN and CR PC (Table 1a). We found ERG protein positivity in 47% (37/78) and 40% (35/88) of the HN and CR PC samples. Similarly, 38% (36/94) and 47% (44/94) of the same samples showed ERG rearrangement. High-grade prostatic intraepithelial neoplasias were not present in this TMA and thus not analyzed in this study. We did not observe ERG positivity in the 10 evaluable BPH samples present on this TMA. In addition, the crTMA comprises a unique set of 36 matched PC samples from the same patients before (HN) and after hormonal ablation therapy (CR). The analysis of this subset revealed a change of ERG status in individual patients to be rare (1/21 and 2/30 for IHC and FISH, respectively; Supplementary Table S1).

Figure 1.

Representative images of ERG-stained prostate samples from the castration resistance tissue microarray (crTMA). Endothelial cells (black arrows) were used as positive control for the ERG staining.

Table 1. Overview of the ERG status on the castration resistance tissue microarray (crTMA).

(a)
					NS
	BPH		All PC		HN		CR
	n	%	n	%	n	%	n	%
FISH
Not rearranged	10	100	108	57	58	62	50	53
Rearranged	0	0	80	43	36	38	44	47
Immunohistochemistry
ERG negative	10	100	94	57	41	53	53	60
ERG positive	0	0	72	43	37	47	35	40
(b)
Gleason Pattern	NS				*P-value< 0.05
	FISH				Immunohistochemistry
	Not rearranged		Rearranged		ERG negative		ERG positive
	n	%	n	%	n	%	n	%
HN
3	7	58	5	42	1	14	6	86
4	45	58	33	42	27	44	34	56
5	61	70	26	30	44	67	22	33
CR
3	0	—	0	—	0	—	0	—
4	21	54	18	46	17	49	18	51
5	80	58	58	42	90	70	39	30

Abbreviations: CR, castration resistant; FISH, fluorescence in situ hybridization; HN, hormone naive; NS, not significant; PC, prostate cancer.

(a) ERG status was significantly different between BPH and all PCs, but not between HN and CR. Fisher's exact tests were used for comparisons.

(b) HN and CR prostate cancer samples without ERG protein expression are characterized by higher Gleason pattern. This association was not true for ERG rearrangement. The χ² test was used for comparison between the groups: Not rearranged vs rearranged and ERG + vs ERG — in HN samples. Fisher's exact test was used for CR samples.

We next investigated a potential association between ERG status and clinicopathological features, such as cM and cT stages, and Gleason pattern. ERG status was not differentially distributed across different cM and cT stages (data not shown). Interestingly, only ERG protein expression but not ERG rearrangement revealed a significant decrease of positivity toward higher Gleason pattern. This was true in HN (P=0.004) as well as in CR PCs (P = 0.019) (Table 1b). As PCs of higher Gleason pattern are characterized by higher tumor cell proliferation, we investigated a potential correlation between ERG status and Ki67 labeling index. We did not observe a correlation between ERG protein expression and increased tumor cell proliferation. This was also true for ERG rearrangement. However, stratification into HN and CR revealed that the proliferation index in ERG-rearranged HN was significantly higher than in those HN where ERG was not rearranged (55% vs 38%, P< 0.05, Supplementary Table S2).

No significant association of ERG status with overall survival of HN or CR PC patients

We analyzed the potential impact of ERG protein expression and rearrangement on overall survival. In both cohorts, HN as well as CR, neither ERG staining nor ERG rearrangement were related to patient prognosis in Kaplan–Meier survival analysis (Supplementary Figure S1).

Decreasing correlation of TMPRSS2-ERG translocation with protein expression of ERG in CR PC

It is well established that ERG protein expression is dependent on the presence of an ERG rearrangement in prostatic adenocarcinoma. Here we investigated the power of this correlation in the cohort of the crTMA, which is composed of highly advanced PCs before (HN) and after ADT (CR). As expected, a high correlation between ERG rearrangement and ERG protein expression was observed (P<0.0001). This was also true if PC samples were stratified according to their hormonal treatment status HN and CR (P<0.0001, Table 2). Intriguingly, whereas in HN PCs, the number of FISH-IHC discordant results were minimal (7% FISH positive, but ERG negative and 9% FISH negative, but ERG positive), in CR PCs, 26% (13 spots) of the ERG-rearranged samples did not show detectable ERG protein expression (Table 2). This surprisingly large group of ERG-rearrangement positive, but ERG-protein-negative PC samples in the CR, but not in the HN group, can hardly be explained by a technical phenomenon. These findings rather suggest that losing the high concordance between ERG FISH and ERG IHC toward more advanced PC samples may be due to the existence of a specific subset of CR PC patients whose tumors have lost the ability of expressing the ERG protein despite the presence of an ERG rearrangement. Of note, these 13 spots were from 11 different TURPs from 10 distinct patients.

Table 2. Correlation of ERG rearrangement and protein expression.

	***P-value< 0.0001
	FISH
	Not rearranged		Rearranged
	n	%	n	%
Immunohistochemistry
HN
ERG negative	59	91	3	7
ERG positive	6	9	42	93
CR
ERG negative	63	93	13	26
ERG positive	5	7	37	74

Abbreviations: CR, castration resistant; FISH, fluorescence in situ hybridization; HN, hormone naive.

A highly significant correlation was found between ERG rearrangement and ERG protein expression in each of the subgroups. ERG FISH-positive CR prostate cancers showed by far the highest discordant rate (26%). Analyses were performed on a spot level by usage of the Fisher's exact test.

TMPRSS2-ERG-positive CR PCs without detectable ERG protein expression

We next interrogated the association between the AR protein expression and the ERG status. As previously described,¹⁸ AR protein expression was present in almost all analyzed PC samples and maximal (score = 100) in >90% of the specimens. Overall, we were not able to see a significant association between ERG rearrangement or positivity and AR expression (P> 0.05 both, data not shown). To analyze whether the ERG FISH vs IHC discrepancy in CR PC with ERG rearrangement but absent ERG protein is due to a loss of AR, we stratified the PCs into the different ERG subgroups according to the two treatment status. Although AR expression was present at high levels (score 90–100) in almost all PC samples, independent of the ERG status, only ERG-rearrangement-positive CR PCs with absent ERG protein were characterized by lower levels of AR protein (P = 0.002, Figure 2a). Further, we interrogated a correlation of ERG protein expression with serum protein levels of the AR target gene PSA in the subgroup of ERG-rearranged CR PCs. As expected, the group of ERG-rearrangement positive and ERG-protein-negative CR PC samples had lower PSA levels than ERG-rearranged- and ERG-protein-positive samples (P< 0.05, Figure 2b). However, it must be considered that PSA serum information was only available for four ERG-rearrangement-positive but ERG-protein-negative CR PC samples.

Figure 2.

Differential androgen receptor (AR) and PSA protein expression in ERG-rearranged castration-resistant (CR) prostate cancer (PC). (a) AR protein expression in hormone naive (HN) and CR PC. Only ERG-rearranged CR PCs without ERG expression (discordant samples) showed significantly lower levels of AR protein expression. (b) Serum PSA levels in ERG-rearranged CR PC. The discordant samples (see Figure 2a) showed reduced levels of serum PSA. Statistical test used: Wilcoxon rank sum. n.s., not significant.

Discussion

The rearrangement of the ERG gene² and its associated expression in PC23 has been the subject of numerous studies. Depending on the cohort used, the prevalence of the rearrangement and protein expression varies extensively (15–80%).¹³ Most of the studies have focused on the analysis of material from surgically resected prostates. In this study, we interrogated the ERG status on gene and protein levels in TURP specimens originating from HN and CR prostate tumors. For this purpose, we used a TMA specifically constructed for the analysis of disease progression.¹⁸

We observed an overall ERG positivity rate of 43% of both ERG rearrangement and IHC positivity across all PC samples. This is similar to recent reports that found ERG protein positivity in 47% and 52% of the PC samples^11,24,25 and ERG rearrangement in 47– 55%.^11,26,27 Stratification into HN and CR PC revealed a broader range (38–47%), but no significant differential positivity between these two groups could be detected. Concordantly, in the matched patient cohort, virtually all of the patients retained their ERG status after recurring under ADT. Although earlier reports that had focused on ERG RNA expression analysis or were based on tissues from xenografts had reported controversial prevalence rates in CR PC,²⁸ our findings are in line with a very recent study by Teng et al.²⁹ in which the authors observed the ERG expression in 37% of human CR PCs. These data strongly suggest that even lower levels of circulating androgens, as is the case under ADT therapy in patients with CR disease, are sufficient to sustain ERG expression in ERG-rearranged PC. Although no correlation of ERG status with clinico-pathological features, such as cM or cT stage, was found, we observed that at least for the protein expression, positive ERG status was associated with lower Gleason pattern (Table 1b). Of note, this TMA was not tailored for the analysis of the Gleason pattern, as most (97%) of the arrayed PCs show a high Gleason pattern (4 or 5) (Supplementary Table S3). In previous studies, TMPRSS2-ERG-negative PCs have already been associated to the highest Gleason category studied.^11,30 Similarly, we observed that a high fraction of tumors with Gleason Pattern 5 is ERG negative: 67% and 70% for HN and CR PCs by IHC, respectively, as well as 70% for HN PC by FISH (Table 1b). Interestingly, the lower number of ERG FISH-negative CR PCs with Gleason Pattern 5 (58%) might be explained by the higher number of ERG-discrepant CR PC samples in this study (see below).

As expected, we found a strong correlation between genomic rearrangement and protein expression in HN as well as in CR PCs, confirming that ERG expression depends on the presence of the ERG rearrangement, also in more advanced CR PCs. Stratification into the four different FISH (negative/positive) and disease (HN/CR) subgroups revealed that in the subgroup of ERG-rearranged CR PC the rate of discordant samples was surprisingly high (26%), suggesting that every fourth ERG-rearranged CR PC will no longer express the ERG protein. As the discordance rates in the other three groups were between 7% and 9%, the high discordance rate of 26% might be attributed to the defects of androgen signaling. Very recently, Teng et al.²⁹ had also reported a decrease in the consistency rate in the group of CR PC. However, their detected decrease was mainly due to ERG rearrangement negativity that needs to be further explored. Here, our findings suggest that up to 26% of the ERG-rearranged CR PC have lost their ability to express the ERG protein. These findings are consistent with a defective AR pathway.31,³² Indeed, only the discrepant samples of this subgroup (CR PC, ERG FISH positive) had significant lower levels of AR protein expression. Concordantly, this minor group of samples also had lower serum PSA levels. One could hypothesize that such patients with PCs who do not express androgenresponsive genes any longer might not be good candidates for a continuing ADT. However, it must be considered that serum PSA level information was only available for four patients of the subgroup of ERG-rearranged but ERG-protein-negative CR PCs. In a recently published study, we reported a subgroup of advanced CR PC patients whose tumors were characterized by the lack of AR expression and had a worst overall survival.18 Half of those tumors were classified as neuroendocrine prostate tumors, suggesting that they had circumvented AR dependency possibly by neuroendocrine-responsive trans-differentiation mechanisms.33 In contrast, in the subgroup of ERG-discrepant samples (CR PC, ERG FISH positive but IHC negative), only four out of the 13 stained positive for neuroendocrine markers, thus suggesting that neuroendocrine trans-differentiation alone cannot explain the characteristics of this subgroup. The four poorly differentiated neuroendocrine CR PCs included two small cell prostate carcinomas and two large cell neuroendocrine carcinomas. Further studies are needed to investigate the specific characteristics of this ERG FISH-positive but ERH IHC-negative subset of PCs on a molecular level and to define the role of ERG rearrangement and expression.

A limitation of our study is that our cohort comprises locally advanced and obstructive tumors from palliative TURPs. Materials from TURPs for TMA construction must be rigorously examined before construction to exclude areas with technical artifacts originating from the resection procedure (for example, heat/mechanical damage). However, PC specimens from these TURPs represent very valuable tissue samples, especially in the context of hormonal ablation. In this study, the stratification into different disease states (HN/CR) and FISH positivity groups limited the sample number in the different subgroups. Thus, studies with larger cohorts of HN and CR PC samples from TURPs are needed to further assess these findings and to evaluate the AR-downstream signaling pathways in the distinct HN/CR ERG subgroups. Despite these limitations, in this study we were able to show the prevalence of ERG positivity in HN and CR PC and that this positivity is not differentially distributed between these two disease groups. Importantly, we provide evidence for the existence of an ERG-rearranged PC subset of cases that has apparently lost the ability to express androgen-regulated genes.