TMPRSS2:ERG Gene Fusions in Prostate Cancer of West African Men and a Meta-Analysis of Racial Differences

Abstract

The prevalence of fusions of the transmembrane protease, serine 2, gene (TMPRSS2) with the erythroblast transformation-specific–related gene (ERG), or TMPRSS2:ERG, in prostate cancer varies by race. However, such somatic aberration and its association with prognostic factors have neither been studied in a West African population nor been systematically reviewed in the context of racial differences. We used immunohistochemistry to assess oncoprotein encoded by the ERG gene as the established surrogate of ERG fusion genes among 262 prostate cancer biopsies from the Ghana Prostate Study (2004–2006). Poisson regression with robust variance estimation provided prevalence ratios and 95% confidence intervals of ERG expression in relation to patient characteristics. We found that 47 of 262 (18%) prostate cancers were ERG-positive, and being negative for ERG staining was associated with higher Gleason score. We further conducted a systematic review and meta-analysis of TMPRSS2:ERG fusions in relation to race, Gleason score, and tumor stage, combining results from Ghana with 40 additional studies. Meta-analysis showed the prevalence of TMPRSS2:ERG fusions in prostate cancer to be highest in men of European descent (49%), followed by men of Asian (27%) and then African (25%) descent. The lower prevalence of TMPRSS2:ERG fusions in men of African descent implies that alternative genomic mechanisms might explain the disproportionately high prostate cancer burden in such populations.

Prostate cancer is a significant public health burden and a major cause of morbidity and mortality among men worldwide. Global geographic variation remains substantial for both prostate cancer incidence and mortality, with disproportionately higher rates among men of African descent (1, 2). Such racial and geographic differences can be attributed partly to variation in somatic mutations (3, 4), given that the prevalences of certain driver oncogenes (e.g., PTEN deletion) of prostate cancer vary by racial group and are associated with prostate cancer prognosis (5–7). Understanding population diversity in somatic mutations might promote the development of blood and urine diagnostic tests for prostate cancer as well as improve the predictive accuracy of prostate cancer progression.

In 2005, the fusion of the 5′-untranslated region of transmembrane protease, serine 2, gene (TMPRSS2), an androgen-responsive gene, to erythroblast transformation-specific (ETS)-related gene (ERG), an oncogenic transcription factor, was first identified as a the most common chromosomal aberration in prostate cancer (8), explaining the mechanism of the frequently overexpressed ERG proto-oncogene in prostate cancer (9). The rearrangement occurs either through chromosomal translocation or intergenic deletion (10, 11) leading to the overexpression of a chimeric mRNA that encodes ERG oncoprotein with N-terminal 32 amino acid deletion. More than 90% of prostate cancers that overexpress ERG harbor TMPRSS2:ERG fusions, rendering highly concordant results using detection methods of fluorescence in situ hybridization, polymerase chain reaction, and immunohistochemistry with great sensitivity and specificity (12–16).

There is a wide range (approximately 8%–83%) of reported prevalence of TMPRSS2:ERG fusions in prostate cancer. The variation may be partly due to: 1) different study cohorts, tissue acquisition, and zonal origins (17, 18); 2) intra- and intertumor heterogeneity (19, 20); 3) variation between laboratories and assays; and 4) racial diversity (21, 22). A previous meta-analysis summarized the prevalence of TMPRSS2:ERG fusions by geographic region and found the highest proportion in Europe (54%) followed by North America (48%) and then Asia (23%) (22). However, this study did not directly address racial differences and mixed hormone-naive tumors with those receiving neoadjuvant endocrine therapy, the latter of which may suppress formation and/or expression of TMPRSS2:ERG fusions (10, 23–25). Associations between TMPRSS2:ERG fusions and prostate cancer prognostic factors are complicated by varying prevalence of the gene rearrangements. The previous meta-analysis demonstrated that only tumor stage was positively associated with TMPRSS2:ERG fusions in radical prostatectomy specimens (22). However, whether race modifies these associations has not been systematically reviewed. In a small comparative study, investigators reported that European Americans were more likely to have ERG overexpression in high-grade (Gleason score >7/4 + 3) prostate cancers than African Americans (26).

In this analysis of the Ghana Prostate Study, we assessed ERG expression—an established surrogate of ERG fusion genes—in relation to demographic and clinicopathological characteristics using 262 prostate cancer biopsies. We further conducted a systematic review and meta-analysis of the prevalence of TMPRSS2:ERG fusions and their associations with race, Gleason score, and tumor stage, with inclusion of the results from the Ghana Prostate Study.

METHODS

Ghana Prostate Study

Study and tissue background

Detailed descriptions of the Ghana Prostate Study have been reported elsewhere (27, 28). In brief, it is a case-control study that included incident prostate cancer cases from a population screening and recruitment of clinical cases from the greater Accra region of Ghana (which has a population of approximately 3 million). A probability sample of 1,037 men aged 50–74 years were enrolled from the community during 2004–2006 to screen for prostate cancer. They underwent an in-person interview at enrollment, had a digital rectal examination, and provided an overnight fasting blood sample for a prostate-specific antigen test. Men who had total prostate-specific antigen concentration of ≥2.5 ng/mL and/or a positive digital rectal exam subsequently underwent a sextant transrectal ultrasound-guided needle biopsy at the Korle Bu Teaching Hospital. A total of 73 patients with histologically confirmed prostate cancers were identified through screening. Meanwhile, a total of 741 clinical prostate cancer patients were newly diagnosed and recruited at the Korle Bu Teaching Hospital during 2003–2012; 676 (91%) of them underwent an in-person interview within a median of 140 days after prostate cancer diagnosis. The interview ascertained epidemiologic information, including demographics, benign urinary symptoms, medical histories, health-care utilization, and lifestyle factors. Height and weight were measured at the interview. Medical records were extracted to elicit clinicopathological information. Biopsy specimens were fixed in 10% buffered formalin and embedded in paraffin at the Pathology Department of the Korle Bu Teaching Hospital. Unstained slides were sent and stored at the National Cancer Institute biorepository. The Ghana Prostate Study was approved by institutional review boards of the University of Ghana and the National Cancer Institute.

Tissue processing and assay

We included 428 prostate cancer patients from the Ghana Prostate Study after exclusion of 386 patients who had no tissue sections available. The included patients were diagnosed during 2004–2012 and had formalin-fixed, paraffin-embedded biopsy sections available for analysis. We randomly sampled 1 unstained tissue section (5–6 μm) for each of the 428 eligible patients and used immunohistochemistry to detect ERG expression. Following deparaffinization, sections were dehydrated and blocked in 0.6% hydrogen peroxide in methanol for 20 minutes followed by microwaving in ethylenediaminetetraacetic acid buffer pH 9.0 for 25 minutes. Sections were then blocked in 1% horse serum for 40 minutes followed by incubation with the ERG-MAb mouse monoclonal antibody (9FY, available from Biocare Medical, Concord, California) at a dilution of 1:80 for 60 minutes at room temperature. Sections then were incubated with a biotinylated horse antimouse antibody at a dilution of 1:200 (Vector Laboratories, Burlingame, California) for 30 minutes followed by treatment with the ABC Kit (Vector Laboratories) for 30 minutes. The color was developed using the VIP Peroxidase Substrate Kit (Vector Laboratories) for 5 minutes followed by counterstaining with hematoxylin for 1 minute. ERG-positive nuclear immunostaining of normal endothelial cells was used as an internal control. Only nuclear immunostaining of prostate cancer cells was deemed ERG-positive. The presence of ERG expression and Gleason scores of the section were evaluated by a genitourinary pathologist (I.A.S.). A total of 262 tissue samples were evaluable for ERG expression, after excluding 166 due to insufficient target tumor tissue or tissue failure during the staining procedure.

Statistical method

We compared the distribution of demographic and clinicopathological characteristics by ERG positivity using χ² tests, and we estimated prevalence ratios and 95% confidence intervals of ERG expression in relation to clinicopathological characteristics using robust Poisson regression (29, 30) with adjustment for potential confounding variables, such as age at diagnosis and calendar year of diagnosis/tissue preservation. We also assessed whether lifestyle factors (i.e., body mass index (calculated as weight (kg)/height (m)²) (31–34), history of diabetes (35), smoking status (36), and alcohol consumption (36), which have been associated with circulating sex hormone concentrations) modified associations between ERG expression and clinicopathological characteristics. To assess potential case selection bias, we performed a sensitivity analysis in which we compared patient characteristics according to tissue section availability and immunohistochemistry quality-control status. Statistical analyses were performed using SAS, version 9.3 (SAS Institute, Inc., Cary, North Carolina). Two-sided P values of <0.05 were considered statistically significant.

Systematic review

To compare the prevalence of TMPRSS2:ERG fusions and its associations with Gleason score and tumor stage across racial groups, a systematic review was conducted in 4 major scientific literature databases (PubMed (National Center for Biotechnology Information, US National Library of Medicine, Bethesda, Maryland), EMBASE (Elsevier BV, Amsterdam, the Netherlands), Scopus (Elsevier BV), and Web of Science (Thomson Reuters, New York, New York)) through August 2015 by a trained clinical informationist (J.W.). Search terms incorporated a wide array of variables related to TMPRSS2:ERG or TMPRSS2:ETS and prostate cancer (search strategies are available from the authors on request). Two reviewers (C.K.Z. and M.B.C.) independently assessed titles, abstracts, and keywords and then selected potentially relevant studies from retrieved references, followed by full-text reviews. Bibliographies of retrieved full-text articles were also reviewed to identify references that might have been missed or absent from the databases used. To be included in this meta-analysis, a published study met the following criteria: 1) based on human prostate cancer tissue; 2) had ≥50 prostate cancer cases for at least 1 racial group; 3) reported no endocrine or radiation therapy prior to tissue acquisition; 4) studied population- or hospital-based, nontrial patients; and 5) included self-reported race or took place in historically racially homogeneous regions. When multiple reports were published with substantial geotemporal overlap in the base population, we included the most recent or best-powered publication. Two reviewers (C.K.Z. and S.B.C.) independently extracted data from selected articles according to a standard form created a priori for this study.

Meta-analysis

To pool proportions of TMPRSS2:ERG fusions, we used Freeman-Tukey arcsine transformation in a random-effects model with 95% confidence intervals estimated by score test (37). Prevalence was summarized by ancestral group, geographic region, tissue acquisition, midyear of diagnosis, and detection method. We also summarized study-specific risk estimates of TMPRSS2:ERG fusions in relation to Gleason score and tumor stage using random-effects models. We then performed metaregression by separately including study-level covariates, such as ancestral group, geographic region, tissue acquisition, midyear of diagnosis, and detection method. Heterogeneity was assessed with Cochran’s Q test and quantified using I² and its 95% uncertainty interval (UI) (38). Publication bias was visually inspected in funnel plots and quantified using Begg’s rank correlation test and Egger’s linear regression test. Meta-analysis was performed using Stata, version 14 (StataCorp LP, College Station, Texas).

RESULTS

Table 1 shows patient characteristics according to ERG staining status in the Ghana Prostate Study. We found that 47 of the 262 (18%) prostate cancers were ERG-positive. ERG-negative tissues were more likely to be diagnosed/preserved recently (P = 0.04) and to have a higher Gleason score (P = 0.03), in particular, a higher primary Gleason pattern (P = 0.03).

Table 1.

Demographic and Clinicopathological Characteristics According to ERG Oncoprotein Expression Status Among 262 Men Diagnosed With Prostate Cancer, Ghana Prostate Study, 2004–2012

Characteristic	ERG Expression Status				P Valueb
	Negative (n = 215)		Positive (n = 47)
	No. of Patients	%a	No. of Patients	%a
Age at diagnosis, yearsc	70 (8.0)		69 (8.1)		0.537
45–64	48	22.3	12	25.5
65–74	99	46.0	24	51.1
≥75	68	31.6	11	23.4
Year of diagnosis					0.044
2004–2007	48	22.3	5	10.6
2008–2009	53	24.7	19	40.4
2010	51	23.7	14	29.8
2011–2012	63	29.3	9	19.1
Gleason score					0.033
<7	32	14.9	9	19.1
3 + 4	41	19.1	17	36.2
4 + 3	58	27.0	9	19.1
>7	81	37.7	11	23.4
Unknown	3	1.4	1	2.1
Primary Gleason pattern					0.030
3	82	38.1	27	57.4
4	114	53.0	15	31.9
5	16	7.4	4	8.5
Unknown	3	1.4	1	2.1
Secondary Gleason pattern					0.386
3	91	42.3	18	38.3
4	85	39.5	23	48.9
5	36	16.7	5	10.6
Unknown	3	1.4	1	2.1
PSA level at diagnosis, ng/mL					0.283
≤10	19	8.8	2	4.3
11–20	22	10.2	8	17.0
>20	166	77.2	36	76.6
Unknown	8	3.7	1	2.1
Clinical tumor stage					0.602
cT1	22	10.2	7	14.9
cT2	110	51.2	23	48.9
cT3	40	18.6	7	14.9
cT4	25	11.6	8	17.0
Unknown	18	8.4	2	4.3
D’Amico risk classificationd					0.736
Low to intermediate	24	11.2	6	12.8
High	184	85.6	39	83.0
Unknown	7	3.3	2	4.3
Ethnic group					0.918
Ga/Adangbe	56	26.0	13	27.7
Ewe	42	19.5	8	17.0
Other	117	54.4	26	55.3
Smoking status					0.518
Never	143	66.5	33	70.2
Ever	71	33.0	13	27.7
Unknown	1	0.5	1	2.1
Regular alcohol intakee					0.872
Never	120	55.8	26	55.3
Ever	92	42.8	21	44.7
Unknown	3	1.4	0	0.0
History of diabetes					0.838
No	181	84.2	39	83.0
Yes	34	15.8	8	17.0
Measured BMIc,f	25 (4.2)		27 (14.5)		0.413
<22	60	27.9	8	17.0
22–24	54	25.1	15	31.9
25–26	40	18.6	11	23.4
≥27	58	27.0	13	27.7
Unknown	3	1.4	0	0.0

Abbreviations: BMI, body mass index; ERG, oncoprotein encoded by erythroblast transformation-specific–related gene; PSA, prostate-specific antigen.

^a Column percentages may not add up to 100% because of rounding.

^bP values were calculated by χ² tests using nonmissing values.

^c Value is presented as mean (standard deviation).

^d Modified D’Amico risk classification: for low risk, PSA ≤10 ng/mL, Gleason score <7, and cT1–2; for medium risk, PSA greater than 10 and ≤20 ng/mL, Gleason score = 7, or cT2; for high risk, PSA >20 ng/mL, Gleason score >7, or cT3–4.

^e Regular alcohol intake was defined as having alcohol once per week in the past 6 months.

^f BMI was calculated as weight (kg)/height (m)² from measurements. BMI categories were quartiles of the continuous variable among the total eligible cases.

Table 2 shows associations between ERG staining status and tumor prognostic factors in the Ghana Prostate Study. After adjusting for age at diagnosis and calendar year of diagnosis, prostate cancer tissues with Gleason score >7/4 + 3 were less likely (prevalence ratio (PR) = 0.66; 95% confidence interval (CI): 0.47, 0.94) to be ERG-positive. ERG expression was not statistically significantly associated with clinical tumor (T) stage or modified D’Amico risk classification. Multiplicative interactions between ERG expression and each lifestyle factor were not statistically significant. Subgroup analyses within strata of each lifestyle factor did not materially change the estimates observed. The sensitivity analysis revealed a greater degree of missing data for clinicopathological characteristics among patients without tissue sections available for study (compared with those who had tissue sections). Among those with tissue sections available, patients whose tissue section failed immunohistochemistry quality control (excluded), versus those that passed immunohistochemistry quality control (included), were more likely to be diagnosed/preserved in earlier years (P = 0.03) and to have lower Gleason scores (P < 0.01) and earlier clinical tumor stage (P = 0.03) (Web Table 1, available at https://academic.oup.com/aje).

Table 2.

Adjusted^a Prevalence Ratios of ERG Staining in Relation to Tumor Prognostic Factors Overall and Stratified by Lifestyle Factors Among 262 Men Diagnosed With Prostate Cancer, Ghana Prostate Study, 2004–2012

Factor	Gleason Scoreb >7 or 4 + 3 (n = 159)			Clinical Tumor Stagec = cT3–4 (n = 80)			D’Amico Risk Classificationd = High (n = 223)
	PR	95% CI	P for Interaction	PR	95% CI	P for Interaction	PR	95% CI	P for Interaction
Overall (either ERG status)	0.66	0.47, 0.94		0.97	0.61, 1.53		0.99	0.93, 1.05
BMIe			0.890			0.744			0.403
<25	0.65	0.42, 1.02		0.90	0.49, 1.65		0.97	0.89, 1.05
≥25	0.69	0.40, 1.17		1.05	0.52, 2.15		1.02	0.94, 1.11
History of diabetes			0.265			0.639			0.040
No	0.60	0.40, 0.91		1.01	0.62, 1.65		0.97	0.91, 1.04
Yes	0.95	0.53, 1.70		0.73	0.19, 2.86		1.09	1.02, 1.16
Smoking status			0.600			0.066			0.730
Never	0.59	0.29, 1.19		0.41	0.11, 1.62		1.02	0.94, 1.11
Ever	0.72	0.49, 1.07		1.27	0.78, 2.07		1.00	0.93, 1.07
Regular alcohol intakef			0.700			0.686			0.432
Never	0.62	0.37, 1.03		1.11	0.56, 2.21		0.97	0.88, 1.06
Ever	0.71	0.44, 1.13		0.91	0.49, 1.71		1.01	0.94, 1.10

Abbreviations: BMI, body mass index; CI, confidence interval; ERG, oncoprotein encoded by erythroblast transformation-specific–related gene; PR, prevalence ratio.

^a Adjusted for continuous age at diagnosis of prostate cancer and calendar year of prostate cancer diagnosis.

^b The reference category is Gleason score <7 or 3 + 4 (n = 99).

^c The reference category is cT1–2 (n = 162).

^d The reference category is D’Amico risk classification of low to intermediate (n = 30). Modified D’Amico risk classification was defined as: for low risk, PSA ≤10 ng/mL, Gleason score <7, and tumor stage cT1–2; for medium risk, PSA greater than 10 and ≤20 ng/mL, Gleason score = 7, or stage cT2; for high risk, PSA >20 ng/mL, Gleason score >7, or stage cT3–4.

^e BMI was calculated as weight (kg)/height (m)² from measurements.

^f Regular alcohol intake was defined as having alcohol once per week in the previous 6 months.

A flow chart for the systematic review is shown in Web Figure 1. In total, we included 41 studies (5, 11, 16, 17, 21, 24, 25, 39–71), including the Ghana results presented herein for qualitative and quantitative synthesis. Of the 41 studies, 1 study acquired prostate cancer tissues from 3 different sources (radical prostatectomy, cystoprostatectomy, and transurethral resection of the prostate) in a German population (17), and 2 studies assessed 2 ancestral groups (5, 63). Characteristics of included studies on TMPRSS2:ERG fusions are shown in Web Table 2. Four studies included men of African descent, including the Ghana Prostate Study (5, 21, 63), 12 included men of Asian descent (44, 46, 47, 49–52, 58, 60–62, 70), and 27 included men of European descent (5, 11, 16, 17, 24, 25, 39–43, 45, 48, 53–57, 59, 63–69, 71).

Table 3 shows the prevalence of TMPRSS2:ERG fusions in prostate cancer using random-effects models. Overall, we found 41% (95% CI: 36, 46) of prostate cancer patients harbored the gene fusion, but there was substantial heterogeneity (I² = 96%; 95% uncertainty interval (UI): 95, 96). This proportion did not differ significantly by tissue acquisition (for heterogeneity between subgroups, P = 0.37). When we stratified the analysis according to ancestral group, the prevalence was highest in men of European descent (49%; 95% CI: 45, 52), followed by men of Asian (27%; 95% CI: 16, 41) and then African descent (25%; 95% CI: 18, 32). These racial differences were statistically significant (for heterogeneity between subgroups, P < 0.001), with moderate to high between-study heterogeneity within subgroups. The order of the prevalence held when further stratified by tissue acquisition (data not tabulated). Post hoc omission of the potentially influential study by Gao et al. (46) in the Asian subgroup resulted in a prevalence of 23% (95% CI: 19, 27), with slightly attenuated between-study heterogeneity (I² = 77%; 95% UI: 59, 87) in this subgroup. Geographical region–specific estimates were consistent with what we observed in ancestral groups. Pooled estimates did not vary significantly within strata of midyear of diagnosis (for heterogeneity between subgroups, P = 0.11) or detection method (for heterogeneity between subgroups, P = 0.20).

Table 3.

Prevalence of TMPRSS2:ERG Fusion or ERG Expression Using Random-Effects Models According to a Meta-Analysis of 41 Studies^a

	No. of Studies	No. of Cases	Prevalence	95% CI	I2, %	95% UI	P Valueb
Overall	41	16,259	0.41	0.36, 0.46	96	95, 96
Tissue acquisition							0.37
Biopsy	13	1,769	0.44	0.31, 0.58	97	96, 98
TURP	5	908	0.34	0.26, 0.43	85	67, 93
Radical prostatectomy	24	13,582	0.41	0.35, 0.46	95	94, 96
Ancestry							<0.001
European	27	13,281	0.49	0.45, 0.52	88	84, 91
Asian	12	2,457	0.27	0.16, 0.41	98	97, 98
African	4	522	0.25	0.18, 0.32	63	0, 88
Region							<0.001
North America	6	710	0.42	0.33, 0.52	85	73, 92
Europe	22	12,831	0.48	0.44, 0.52	89	85, 92
Asia	12	2,457	0.27	0.16, 0.41	98	97, 98
Africa	1	262	0.18	0.14, 0.23	–c	–c
Midyear of diagnosis							0.11
Before 2000	8	1,354	0.50	0.39, 0.60	93	88, 96
2000 onward	23	13,596	0.37	0.30, 0.44	97	97, 98
Detection method							0.20
FISH	19	3,069	0.38	0.32, 0.43	89	85, 92
IHC	16	12,423	0.42	0.33, 0.51	98	97, 98
PCR	6	767	0.52	0.37, 0.66	94	89, 97

Abbreviations: CI, confidence interval; FISH, fluorescence in situ hybridization; IHC, immunohistochemistry; PCR, polymerase chain reaction; TURP, transurethral resection of the prostate; UI, uncertainty interval.

^a The total does not always add up 41 due to some studies having assessed more than 1 ancestral group or source of tissue acquisition (studies include the present analysis as well as referenced studies (5, 11, 16, 17, 21, 24, 25, 39–71)).

^bP values were estimated for heterogeneity between subgroups.

^c The I² and 95% UI were not estimated because only 1 study was available.

In 19 of the studies, investigators reported TMPRSS2:ERG fusions in relation to Gleason score (from biopsy or radical prostatectomy). Overall, the fusion was not statistically significantly associated with Gleason score (>7/4 + 3 versus <7/3 + 4; PR = 0.83; 95% CI: 0.67, 1.02) with moderate to high between-study heterogeneity (I² = 67%; 95% UI: 49, 79), as shown in Web Figure 2. There was mild evidence of publication bias, given Begg’s (P = 0.21) and Egger’s (P = 0.05) tests in addition to an asymmetric funnel plot (plot not shown). Ancestry did not seem to modify the association of TMPRSS2:ERG fusions with Gleason score, although the summary associations for Asian and African subgroups were statistically significant (Web Figure 2). Studies that used more recently diagnosed/preserved prostate cancer tissues or a detection method other than fluorescence in situ hybridization were more likely to report a negative association of the fusion with Gleason score (Web Table 3). In 18 studies, investigators reported TMPRSS2:ERG fusions in relation to tumor stage (clinical or pathological). Overall, the fusion was positively associated with distant/regional tumor stage (PR = 1.13; 95% CI: 1.04, 1.23), with low to moderate between-study heterogeneity (I² = 0%; 95% UI: 0, 44) (Web Figure 3). Publication bias was unlikely given the nonsignificant Begg’s (P = 0.62) and Egger’s (P = 0.99) tests and a symmetric funnel plot (plot not shown). Ancestry did not seem to modify associations between TMPRSS2:ERG fusions and tumor stage, given similar summary estimates within strata of ancestral groups (Web Table 3).

DISCUSSION

In this study of West African men diagnosed with prostate cancer, we have reported that 18% of the 262 prostate cancer patients expressed ERG using biopsy tissue sections, and biopsies being negative for ERG staining was associated with higher Gleason score. Lifestyle factors that have been associated with circulating sex steroid hormones did not modify associations between ERG expression and tumor prognostic factors in this population. Results from the meta-analysis provide compelling evidence of racial differences in the prevalence of TMPRSS2:ERG fusions in prostate cancer, with the highest being in men of European descent, followed by men of Asian and then African descent. Furthermore, the meta-analysis suggested that TMPRSS2:ERG fusions were not statistically significantly associated with Gleason score but were positively associated with tumor stage. There was limited evidence that race modified these associations.

Racial differences in androgenic activity have long been observed, with enhanced androgenic activity reported in men of African descent. Compared with their European-American counterparts, healthy adult African-American men tend to have higher free circulating testosterone concentrations (72), African-American prostate cancer patients tend to have shorter CAG repeats—the length of which is inversely correlated with the transcriptional activity of the androgen receptor yielding higher androgenic action (73)—and both benign and malignant prostate tissues in African-American men tend to express higher levels of androgen receptor (74). Such racial differences in androgenic activity may be expected to affect the likelihood of TMPRSS2:ERG fusions, as androgenic action plays a pivotal role in this chromosomal aberration. High testosterone concentrations induce spatial proximity of the TMPRSS2 and ERG genomic loci, facilitating the formation of gene fusions in vitro (75, 76), and higher concentrations of free circulating testosterone before prostate cancer diagnosis are associated with a greater risk of ERG overexpression (77). Prostate cancer with shorter CAG repeats are more likely to harbor the gene fusion (76), and high ERG expression coincides with androgen receptor overexpression (78). In sum, evidence suggests that prostate cancer among men of African descent may be more likely to harbor TMPRSS2:ERG fusions. However, our study showed that men of European descent had a higher prevalence of this rearrangement than men of African descent. Our results are supported by previous comparative studies that assessed prostate cancer tissues across ancestral groups (4, 5, 21, 63, 79, 80). Also, the overly simplistic assumption of a positive linear relationship between androgenic activity and the likelihood of TMPRSS2:ERG fusions has been challenged by a previous functional study, which reported that androgen receptor and ERG function in a negative autoregulatory loop, tightly controlling levels of both transcription factors (81, 82).

TMPRSS2:ERG fusions generally occur early in prostate carcinogenesis with approximately 10%–20% of high-grade prostatic intraepithelial neoplasia (a histologic precursor to invasive prostate cancer) harboring this fusion (83–86). However, the role of TMPRSS2:ERG fusions in the progression of prostate cancer remains inconclusive, partly due to patient selection biases. The Ghana analysis found an inverse association between ERG expression in prostate cancer biopsies and Gleason score, consistent with data using radical prostatectomy specimens from African Americans (5, 63). Meta-analysis also showed an inverse association of TMPRSS2:ERG fusions with Gleason score in men of African descent—when these 3 studies were pooled—and in men of Asian descent but not in men of European descent, although the subgroup differences were not statistically significant. It remains unclear whether the lower prevalence of TMPRSS2:ERG fusions leads to more aggressive prostate cancer in African and Asian men or is a result of selecting cases from a higher proportion of symptomatic high-grade prostate cancers in populations where screening has historically been much less common. Meanwhile, although we found a null association between ERG expression and clinical tumor stage in Ghanaian patients, TMPRSS2:ERG fusions were overall positively associated with tumor stage in the meta-analysis, driven primarily by studies using radical prostatectomy tissues among men of European descent. The interpretation of this association must to be cautious given that late-stage prostate cancers may have been undersampled, because such patients are usually not referred for radical prostatectomy. This positive association between TMPRSS2:ERG fusions and tumor stage may not hold when extrapolated to metastatic foci, given the reported lower prevalence of TMPRSS2:ERG fusions in lymph-node metastases, implying unfavorable selection of fusion-positive tumor cells in the metastatic process (87).

Our study has several limitations. First, we used only 1 diagnostic biopsy slide per patient in the Ghana Prostate Study and were thus unable to assess tumor heterogeneity and multifocality. This is also a limitation of the majority of previous studies. The moderate to high between-study heterogeneity in our meta-analysis for prevalence may be explained partly by sampling variations due to tumor heterogeneity and differences in laboratory processing. Previous studies of the whole-mount prostatectomy specimens have shown that over 40% of prostate cancers with multiple, spatially separate tumor foci harbor discordant fusion status (19, 20). Incomplete sampling of tumor foci may underestimate the presence of TMPRSS2:ERG fusions. Second, although there were differential distributions of year of diagnosis, Gleason score, and clinical tumor stage in the sensitivity analysis by study inclusion/exclusion status, tissue availability and tissue quality are unlikely to be associated with ERG status, and thus we would not expect associations reported herein from the Ghana Prostate Study to be biased. Differences in year of diagnosis reflect the possibility of tissue degradation, particularly given the lack of a temperature- and humidity-controlled environment for tissue storage in Ghana. Third, the Ghana Prostate Study was not designed to follow prostate cancer patients prospectively to assess progression of disease due to the lack of an infrastructure to comprehensively capture it. Thus, we were unable to evaluate whether ERG expression was associated with fatal prostate cancer, although biopsy Gleason score and clinical tumor stage strongly predict prostate cancer survival in such an unscreened population. Last, only a small number of studies among African and Asian ancestral groups were available for inclusion in the meta-analysis, and the summary estimates are potentially subject to publication bias. Meanwhile, the moderate to high heterogeneity for the association between TMPRSS2:ERG fusions and Gleason score within the European ancestral group may be explained partly by different case sampling.

In conclusion, this is, to our knowledge, the first study of ERG expression among West African prostate cancer patients, and we found that the prevalence of ERG expression was comparable to that among African Americans and lower than that of men of European descent in a meta-analysis. Although urine TMRPSS2:ERG transcripts plus PCA3 score has been proposed as a diagnostic test for prostate cancer (88, 89), the clinical utility of urine TMRPSS2:ERG fusions may be lower in men of African descent given its lower prevalence. However, the validity and utility of these urine-based biomarkers in men of African ancestry await further studies. The relatively low level of ERG expression and its inverse association with higher Gleason score imply that there may be alternative mechanisms separate from, or in conjunction with, TMPRSS2:ERG fusions to explain the disproportionately high burden of prostate cancer in men of African ancestry.

Abbreviations

CI

confidence interval

ERG

erythroblast transformation-specific–related gene

ETS

erythroblast transformation-specific

PR

prevalence ratio

TMPRSS2

transmembrane protease, serine 2, gene

UI

uncertainty interval