The impact of CAG repeats in exon 1 of the androgen receptor on disease progression after prostatectomy

Abstract

BACKGROUND

The authors examined the impact of the number of CAG repeats in exon 1 of the androgen receptor on disease progression among men with prostate carcinoma after prostatectomy. This polymorphism has been associated with alterations in activity of the androgen receptor in in vitro systems and with the risk of clinically diagnosed prostate carcinoma in some epidemiologic studies. An earlier series found that, among men at low risk of progressive disease, a small number of CAG repeats predicted a high risk of recurrence, and the impact of CAG repeats varied among men with different risks of progressive disease.

METHODS

The authors analyzed specimens from a large clinical series of fixed tissue specimens from men who underwent prostatectomy at a single institution, including 413 American white men (WM) and 298 African-American men (AAM), with 5–10 years of available clinical follow-up.

RESULTS

There was little association between the number of CAG repeats and extent of disease, Gleason score, and preoperative PSA level at diagnosis. Overall, patients who had > 18 CAG repeats had a greater risk of recurrence compared with patients who had ≤ 18 CAG repeats (hazard ratio [HR] = 1.52; P = 0.03). Excess risk was not found among men who were at low risk of recurrence (HR = 0.93; P = 0.96); however, among men who were at high risk of recurrence, the risk elevated for WM (HR = 1.75; P = 0.28), AAM (HR = 1.49; P = 0.06), and both races combined (HR = 1.53; P = 0.03).

CONCLUSIONS

Overall, men with prostate carcinoma who had > 18 CAG repeats had an estimated 52% increased risk of disease recurrence. The increased risk could be attributed to men who were at high risk of recurrence. Cancer 2005. © 2005 American Cancer Society.

Prostate carcinoma is an androgen-dependent disorder. Androgen function is mediated by the androgen receptor (AR), a ligand-dependent steroid hormone-transactivation factor located on the X chromosome. Males have one copy of the gene. The transcription-activation domain of the AR is in a highly polymorphic N-terminal region of exon 1.1 A CAG trinucleotide repeat in that region codes for glutamine.2 In vitro studies indicate that the length of the polyglutamine tract is related inversely to transcription activity,3, 4 suggesting that the potency of function of this receptor is related inversely to the length of this repeat.

The increased risk of prostate carcinoma within families suggests that genetic factors are responsible in part for the initiation and/or the progression of the disease. The number of CAG repeats ranges from 11 to 30 repeats in most individuals. African Americans tend to have fewer CAG repeats in the AR gene and higher incidence and mortality rates for prostate carcinoma.5 Several epidemiologic studies found that the number of CAG repeats in the AR was associated inversely with the risk of prostate carcinoma, especially the risk for disease characterized by high Gleason score and advanced stage.6–9 More recent studies were less consistent, and the topic has been the subject of several reviews.10–14 Giovannucci suggested the positive studies tended to include predominantly patients who were diagnosed without prostate-specific antigen (PSA) screening, whereas the studies that showed little effect largely included patients who were diagnosed by PSA screening.13

The epidemiologic studies cited above investigated associations between CAG repeats and the risk of newly diagnosed prostate carcinoma. They often included patients from multiple centers. Data on staging and grading were not uniform, and little follow-up data were available, limiting the ability of those studies to investigate the impact of CAG repeats on disease progression. Recently, a small series of studies focused on the impact of CAG repeats on disease progression and prognosis in patients with prostate carcinoma. Nam et al.15 studied men (presumably primarily white men [WM]) from Ontario, Canada, after radical prostatectomy. Those authors found an association with CAG repeats that differed according to their risk of recurrence, as estimated by traditional prognostic variables (disease stage, Gleason score, and preoperative PSA level). Among patients who were at low risk of recurrence, the presence of fewer CAG repeats increased the risk of recurrence. Among patients who were at a high risk of recurrence, the presence of greater numbers of CAG repeats increased the risk of recurrence. Only for low-risk patients was the association statistically significant (P = 0.003), but the authors reported a statistically significant interaction between risk group and the category of CAG repeats (P = 0.004).15 Two smaller studies of relatively small groups of patients found little impact of CAG repeats on disease-free survival: Edwards et al.16 studied 178 British WM, and Cude et al,17 studied 131 patients at the National Cancer Institute.

Our center has a large series of patients who underwent prostatectomy > 5 years ago for whom surgical blocks and follow-up data are available. Compared with previously reported follow-up studies in which most patients were WM, nearly 50% of the patients who undergo prostatectomy at our center are African-American men (AAM). We previously reported careful multivariable analyses of the impact of clinical variables, including age, PSA levels, Gleason scores, and disease stage on race-specific recurrence rates.18, 19 Those reports found that younger AAM who were diagnosed during the early 1990s had a poorer prognosis compared with WM. The clinical variables were significant risk factors for recurrent disease but did not account for the racial disparity in recurrence-free survival.

We were able to extract DNA from the stored surgical material, so that we did not have to wait for follow-up data to study the impact of polymorphisms on disease progression. The patients reported herein had clinically localized prostate carcinoma treated with radical prostatectomy and overlap patients who were included in previous survival analyses.18, 19 The diversity of patients, careful pathologic staging, long follow-up, and availability of specimens facilitated investigation of the factors responsible for the difference in prognosis among the races. In the current study, we examined whether the number of CAG repeats was associated with staging factors (Gleason score, invasion of seminal vesicles and lymph nodes, PSA levels at diagnosis) or with biochemical progression-free survival (PFS) after prostatectomy.

MATERIALS AND METHODS

Study Participants

Consecutive radical prostatectomy specimens of clinically localized prostate carcinoma from 1009 patients were examined by 2 experienced urologic pathologists at Wayne State University, Harper Hospital (Detroit, MI) from January, 1991, through June, 1996. The study was conducted according to procedures approved by the Wayne State University Institutional Review Board. Characteristics of patients in this series and the methods used were described in previous reports.18, 19 Patients were seen on a physician-referral and self-referral basis. We excluded 46 patients from other countries, 27 patients who underwent salvage prostatectomy, 6 patients for whom clinical data were not available, and 107 patients who did not have PSA levels that declined to < 0.4 ng/mL and/or who received neoadjuvant therapy. This left 823 patients who were eligible for the study. Race was established by self-assignment. Five Asian patients and one patient of unknown race were excluded from further analysis. Sufficient follow-up information was not available for 46 patients, tissue blocks were not available for DNA extraction for 31 patients, and genotyping was not successful for 33 patients. This resulted in 711 patients with both CAG analysis and follow-up data, including 413 WM and 298 AAM.

PSA levels were determined prior to digital rectal examination or (for < 1% of patients) > 10 days after tumor biopsy. Biochemical recurrence after prostatectomy was defined as a PSA level > 0.4 ng/mL that persisted for > 1 reading.

Surgical pathologic specimens were staged according to the 1992 TNM classification system. Pathologic examination was performed according to a carefully established protocol by two highly experienced pathologists. The margins and capsular surface of the prostatectomy specimens were marked thoroughly by India ink, and the gland was sectioned serially at 3–5-mm intervals. This produced consecutive cross-sectional slices from the apex to the base perpendicular to the posterior surface. The specimens were inspected closely and were palpated to identify tumor nodules macroscopically. Abnormalities were mapped on a preliminary diagram. Microscopic evaluation of the entire gland was used to create a final mapping of tumor distribution, the extent of tumor, and the status of the prostatic capsular and apical, anterior, posterior, lateral, and proximal margins. Histologic grading of the primary tumor was scored according to the Gleason grading system. Locally advanced/nonorgan-confined prostate carcinoma included specimens with capsular penetration and positive surgical margins with or without extracapsular extension and/or seminal vesicle invasion.

Data concerning these patients have been assembled in a computerized data base. Selected variables include race, age, clinical stage, preoperative PSA level, postoperative evaluation (including disease stage, lymph node status, Gleason score, capsular margin, or seminal vesicle involvement). Postoperative PSA levels were used to determine recurrence-free survival and PFS survival after prostatectomy.

DNA Isolation

DNA was isolated from fixed tissues by a modified procedure using QIAamp tissue kits (Qiagen, Inc., Valencia, CA). Paraffin slices were incubated overnight and were mixed continuously at 57 °C with proteinase K. This was followed by treatment with protease for 3 hours, binding the DNA to QIAamp silica membrane, washing, and elution into a polymerase chain reaction (PCR)-compatible buffer according to the manufacturer's recommendations.

Genotyping of AR CAG Microsatellite

The number of repeats in the CAG microsatellite of exon 1 of the AR gene was determined using nested PCR analysis. Briefly, genomic DNA was amplified using PCR primers 5′-GCC TGT TGA ACT CTT CTG AGC A-3′ (forward) and 5′-TGC TGA AGG AGT TGC ATG GT-3′ (reverse) followed by a heminested PCR with 6FAM-GAT TCA GCC AAG CTC AAG GAT-3′ and the outer reverse primer using AmpliTaq Gold. The amplified product was electrophoresed in polyacrylamide (5.0% Long Ranger; FMC BioProducts, Rockland, ME) using an ABI 377 (Genomics Core Facility, Karmanos Cancer Institute, Wayne State University) and ROX-labeled internal size standards. The sizes of the PCR products were determined using GeneScan software (PE Applied Biosystems, Foster City, CA). Replicate amplifications of the same DNA sample yielded identical results. Data were obtained from 97% of the tissue blocks studied. Using this system allows size discrimination to a single base pair. Electrophoresis controls were established for the length polymorphisms of the CAG repeats in the AR by sequencing PCR products of differing lengths to confirm accurate sizing. They were then used for quality control in subsequent electrophoresis runs. Sixty-six of the samples were sequenced for confirmation and quality control.

Statistical Methods

To investigate associations between the number of CAG repeats and clinical variables, the mean number of repeats was calculated for strata defined by age (younger, ≤ 65 years; older, > 65 years), preoperative serum PSA level (low, 0–10 ng/mL; high, 10–20 ng/mL; very high, > 20 ng/mL), pathologic Gleason score (low, < 7; moderate, 7; high, > 7), tumor pathology (organ-confined disease, positive margins, extraprostatic extension, positive seminal vesicles, and positive lymph nodes), disease stage (organ-confined vs. locally advanced disease, with locally advanced disease including all men with positive margins, extraprostatic extension, positive seminal vesicles, or positive lymph nodes) and were compared in nonparametric tests. For comparisons with the data from Nam et al.,15 patients were divided into risk groups in which low risk included patients with organ-confined disease, Gleason scores < 7, and preoperative PSA levels ≤ 10 ng/mL; and high risk included patients with locally advanced disease, Gleason scores ≥ 7, and preoperative PSA levels > 10 ng/mL. Kendall τ B correlation coefficients were used to assess associations between CAG repeats and clinical variables (Table 1). Both median and mean values are provided in Tables 1–3. Mean values and standard deviations (SDs) were included to present the small differences among groups, whereas nonparametric procedures were employed for statistical testing, because staging information was ordinal, several other variables were not distributed normally, and it avoided heavily weighting extreme observations. A comparison of the number of CAG repeats among patients who developed recurrent disease with the number of CAG repeats among patients who did not develop recurrent disease was conducted using Wilcoxon rank-sum tests for strata defined by race, extent of disease (Table 2), and risk group (Table 3).

Table 1. Association Between CAG Repeats, Patient Age, Disease Stage, Gleason Score, and Preoperative Prostate-Specific Antigen Level by Race

Variable	No. of CAG repeats
	White men						African-American men
	No.	Median	Mean ± SD	Min	Max	P value	No.	Median	Mean ± SD	Min	Max	P value
SD: standard deviation; Min: minimum; Max: maximum; ext: extension; inv: invasion; PSA: prostate-specific antigen. a Wilcoxon rank-sum tests comparing white men with African-American men. b Race-specific Kendall τ B correlation for CAG and clinical variable. c Race-specific Kruskal–Wallis test for CAG classified by stage.
All patients	413	22	22.0 ± 3.0	8	33		298	20	19.8 ± 3.7	7	32	< 0.001a
Age
≤ 65 yrs	262	21	21.9 ± 3.1	8	33		159	19	19.7 ± 3.8	7	32
> 65 yrs	151	22	22.2 ± 2.6	17	32	0.19b	139	20	19.9 ± 3.6	10	30	0.86b
Stage:
Organ confined	206	22	22.0 ± 2.9	8	32		112	20	19.9 (3.6)	11	28
Positive margins	72	21	21.9 ± 3.3	15	31		78	19	19.4 (4.2)	7	31
Extracapsular ext.	68	21	22.2 ± 2.8	17	32		47	20	20.3 (3.7)	13	32
Seminal vesicle inv.	37	22	22.1 ± 2.2	19	27		44	19	19.4 (3.3)	10	27
Positive lymph node	30	21	22.3 ± 3.8	14	33	0.89c	17	20	20.5 (3.0)	15	26	0.61c
Gleason score
< 7.0	153	22	22.0 ± 3.2	8	31		98	19	19.8 ± 3.9	7	31
7.0	207	21	21.8 ± 2.9	14	33		152	19	19.8 ± 3.7	7	32
> 7.0	51	23	22.8 ± 2.7	18	29	0.72b	48	20	19.8 ± 3.1	13	27	0.76b
Preoperative PSA level
≤ 10 ng/mL	279	21	21.8 ± 3.0	8	32		172	20	19.8 ± 3.8	7	32
10–20 ng/mL	93	22	22.6 ± 3.1	17	33		59	20	19.7 ± 3.6	10	28
> 20 ng/mL	41	22	22.1 ± 2.6	17	27	0.18b	67	20	19.8 ± 3.5	10	27	0.70b

Table 2. Association between CAG Repeats and Recurrence Status By Disease Status (Organ-confined vs. locally advanced disease) and Race

Recurrent disease	Organ-confined disease						Locally advanced disease
	No.	Median	Mean ± SD	Min	Max	P valuea	No.	Median	Mean ± SD	Min	Max	P valuea
SD: standard deviation; Min: minimum; Max: maximum. a Wilcoxon two-sample test.
White men
Yes	11	22	22.6 ± 2.9	19	28		76	22	22.4 ± 3.3	14	33
No	195	22	21.9 ± 3.0	8	32	0.55	131	21	21.9 ± 2.8	17	32	0.19
African-American men
Yes	8	20	20.4 ± 3.6	16	25		93	20	19.9 ± 3.6	7	28
No	104	20	19.9 ± 3.6	11	28	0.84	93	19	19.5 ± 4.0	7	32	0.24

Table 3. Association between CAG Repeats and Recurrence Status by Risk Group (High-risk vs. low-risk diseasea) and Race

Recurrent disease	Low-risk disease						High-risk disease
	No.	Median	Mean ± SD	Min	Max	P valueb	No.	Median	Mean ± SD	Min	Max	P valueb
SD: standard deviation; Min: minimum; Max: maximum; NA: not applicable. a As defined by Nam et al, 2000.15 b Wilcoxon two-sample test.
White man
Yes	0	No events					87	22	22.4 ± 3.3	14	33
No	94	22	22.2 ± 3.3	8	30	NA	232	21	21.8 ± 2.7	14	32	0.09
African-American men
Yes	2	20	20.0 ± 2.8	18	22		99	20	20.0 ± 3.6	7	28
No	49	19	19.4 ± 3.5	11	28	0.75	148	19	19.8 ± 3.9	7	32	0.52

The race-specific distributional characteristics of CAG repeats were examined graphically (Fig. 1). When analyses required stratification of CAG results, results were grouped by ≤ 18 repeats, 19–22 repeats, and ≥ 23 repeats, to provide roughly equally sized patient groups, or into 2 groups, a group with ≤ 18 repeats and a group with > 18 repeats, to allow for comparisons with the previous study by Nam et al.15 Nonparametric Kaplan–Meier survival function estimates for PFS distributions after radical retropubic prostatectomy were obtained for patients overall and were stratified by category for several key factors. PFS was defined as the time from surgery until the first disease recurrence. For patients without disease recurrence, PFS was right-censored at the date of the last PSA follow-up. Then, log-rank tests were employed to test the homogeneity of survival distributions between the categories of the factor. The Kaplan–Meier plots and log-rank tests were used to assess the homogeneity of survival distributions across strata defined by individual variables and high risk versus low risk. These analyses were conducted both for all men and stratifying men by race.

Figure 1. Frequency distribution of the number of CAG repeats in exon 1 of the androgen receptor by ethnicity for African American men (black bars; n = 298) and white men (white bars; n = 413).

Cox proportional hazard regression models were used to determine of the impact of CAG repeats on disease-free survival.20 Preliminary models determined the effects of clinical variables, including race, age, disease stage, Gleason score, and preoperative PSA level, either grouped as presented in Table 1 or stratified by disease stage or risk as described above and in Tables 4 and 5. Subsequent models controlled for these factors. Because CAG repeats differed by race (Table 1), models were stratified by race. The proportional hazards assumption of the Cox model was tested by studying interactions of each factor with PFS.21 Effect modification of any genotype/PFS association was ruled out by examining all first-order interaction effects among pairs of the predictors. Goodness of fit of the Cox models was evaluated using plots of deviance residuals.21

Table 4. Disease Progression Hazard Ratios for a One-Category Increase in CAG Repeats (≤ 18, 19–22, and > 22): Analysis for All Men and for 8 Subsets of Men

Patient group	HR	95%CI	P value
HR: hazard ratio; 95%CI: 95% confidence interval. a Analysis stratified by race, controlling for age, preoperative prostate-specific antigen (PSA) level, Gleason score, and stage. b Controlling for age, preoperative PSA level, Gleason score, and stage. c Analysis stratified by race, controlling for age, preoperative PSA level, and Gleason score. d Controlling for age, preoperative PSA level, and Gleason score.
All mena	1.11	0.90–1.38	0.32
White menb	0.95	0.66–1.36	0.78
African-American menb	1.21	0.93–1.57	0.15
Organ-confined diseasec	1.01	0.52–1.96	0.97
Locally advanced diseasec	1.14	0.91–1.42	0.27
White men
Organ-confined diseased	1.17	0.39–3.48	0.78
Locally advanced diseased	0.92	0.63–1.35	0.67
African-American men
Organ-confined diseased	0.76	0.32–1.83	0.54
Locally advanced diseased	1.25	0.95–1.64	0.11

Table 5. Disease Progression Hazard Ratios for Men with > 18 CAG Repeats and Men with ≤ 18 Repeats. Analysis of All Men and for 8 Subsets of Men

Patient group	HR	95%CI	P value
HR: hazard ratio; 95%CI: 95% confidence interval. a Analysis stratified by race, controlling for age and risk. b Controlling for age and risk. c Analysis stratified by race, controlling for age. d Controlling for age.
All mena	1.52	1.03–2.23	0.03
White menb	1.75	0.64–4.80	0.28
African-American menb	1.48	0.98–2.25	0.07
Low-risk diseasec	0.93	0.06–14.91	0.96
High-risk diseasec	1.53	1.04–2.26	0.03
White men
Low-risk diseased	No events
High-risk diseased	1.75	0.64–4.80	0.28
African-American men
Low-risk diseased	0.93	0.06–14.91	0.96
High-risk diseased	1.49	0.98–2.28	0.06

RESULTS

Figure 1 presents the distribution of CAG repeats by race among 711 men with prostate carcinoma. AAM had fewer CAG repeats than WM, with a distribution skewed and shifted to the smaller CAG values. Most men with < 18 CAG repeats were found among AAM, whereas high values (> 27 repeats) tended to be distributed equally between the races.

CAG repeats for AAM, on average, were 2.2 fewer than the number in WM (Table 1). This difference was highly statistically significant (Wilcoxon rank-sum test; P < 0.001). Considering all men, the numbers of CAG repeats did not vary significantly between those who developed recurrent disease versus those who remained free of disease: The 188 men who developed recurrent disease averaged 21.1 (SD, ± 3.6) CAG repeats, whereas the 523 men who did not develop recurrent disease averaged 21.1 ± 3.4 repeats (P = 0.93; Wilcoxon two-sided test). Results for each race separately revealed that the 87 WM who developed recurrent disease averaged 22.4 ± 3.3 repeats, whereas the 326 WM who did not develop recurrent disease averaged 21.9 ± 2.9 (P = 0.19). The 101 AAM who developed recurrent disease averaged 20.0 ± 3.5 repeats, and the 197 AAM who did not develop recurrent disease averaged 19.7 ± 3.8 repeats (P = 0.34). CAG repeats also did not vary significantly according to disease stage, Gleason score, or preoperative PSA values for either race (Table 1). Likewise, the number of CAG repeats did not differ among men with disease in different stages (organ-confined disease vs. locally advanced disease) or risk groups (low risk vs. high risk, as defined above [see Materials and Methods] based on the classification scheme used in an earlier report by Nam et al.15; data not shown).

Tables 2 and 3 present the average CAG repeats among men who did and did not develop recurrent disease stratified by race, disease stage, and risk group. The average number of CAG repeats differed by less than one repeat, and there were no significant differences in the number of repeats for any stratum.

A survival analysis with Cox proportional hazards regression models demonstrated significant associations between several of the categorized clinical variables and PFS. A significantly increased risk was found for preoperative disease stage (hazard ratio [HR] = 4.48; 95% confidence interval [95%CI], 2.72–7.37; P < 0.0001), Gleason score (HR = 2.40; 95%CI, 1.89–3.04; P < 0.0001), and preoperative PSA level (HR = 1.52; 95%CI, 1.27–1.82; P < 0.0001) but not for patient age (HR = 0.87; 95%CI, 0.65–1.16; P = 0.35). These findings are similar to findings from earlier analyses of patients at our institution that included many of the same study patients.19

The impact of CAG repeats was investigated using two approaches to grouping the CAG repeats and subsetting for the severity of disease. Figure 2 and Table 4 present results grouping the CAG repeats into three relatively equal groups and subsetting patients by race and extent of disease (organ-confined and locally advanced disease). The race/extent of disease-specific analysis presented in Figure 2 suggested a more favorable prognosis for AAM with locally advanced disease who had ≤ 18 repeats (Fig. 2D). The 3 survival curves shown in Figure 2D differ significantly (log-rank test; P = 0.04). No significant impact of CAG repeats was observed for other race/extent of disease subgroups (Fig. 2A–C). Proportional hazards modeling controlling for multiple clinical factors also suggested that, among AAM with locally advanced disease, the risk of recurrence increased 1.25-fold with successively greater CAG repeat numbers, but the association with the numbers of CAG repeats was not statistically significant (Table 4) (P = 0.11).

Figure 2. Kaplan–Meier survival curves showing the association of progression-free survival with CAG repeats in exon 1 of the androgen receptor for patients by race and disease stage (organ-confined or locally advanced disease, with locally advanced disease including all men with positive margins, extraprostatic extensions, positive seminal vesicles, and positive lymph nodes). (A) Organ-confined prostate carcinoma, Caucasian men (CM). (B) Locally advanced prostate carcinoma, CM. (C) Organ-confined prostate carcinoma, African-American men (AAM). (D) Locally advanced prostate carcinoma, AAM.

Figure 3 and Table 5 use grouping of CAG repeats (≤ 18 and ≥ 18) and risk stratification based on the approach used by Nam et al.15 For the risk stratification, the factors used were disease stage, Gleason score, and preoperative PSA level, as described above (see Materials and Methods). Among the patients in our series who were at low risk according to this classification, no WM and only 2 AAM (1 AAM with < 18 CAG repeats and 1 AAM with > 18 CAG repeats) experienced a recurrence. This precluded further analysis of men at low risk. An analysis of high-risk men, however, revealed that men of both races with ≤ 18 repeats had a more favorable prognosis (Fig. 3A,B). Using the log-rank test, this effect was not statistically significant for WM (P = 0. 26), was of borderline significance among AAM (P = 0.06), and was statistically significant (P = 0.03) when both races were combined (data not shown). Using Cox proportional hazard analysis and controlling for risk, as defined by Nam et al., men with > 18 repeats had an estimated risk of recurrence that was 52% greater among all men, 75% greater among WM, and 48% greater among AAM (Table 5). Among men with > 18 CAG repeats, both WM and AAM at high risk had elevated risks of recurrence, whereas AAM at low risk had decreased risks of recurrence (Table 5). Further statistical analysis investigating the possibility of an interaction effect of CAG repeat number category with risk group revealed that the interaction was not statistically significant in a model that included all men or in a model stratified on race (P > 0.35 for the interaction effect in each model).

Figure 3. Kaplan–Meier survival curves showing the association of progression-free survival with CAG repeats in exon 1 of the androgen receptor for high-risk Caucasian men (CM) and high-risk African-American men (AAM) with prostate carcinoma. High-risk included patients with locally advanced disease, Gleason scores ≥ 7, or preoperative prostate-specific antigen levels > 10 ng/mL. (A) High-risk CM. (B) High-risk AAM.

Finally, we considered whether the differences in HRs observed by our two modeling approaches resulted primarily from the different groupings of CAG repeats or from the different covariates used in the models. It appeared that the grouping of CAG repeats was most critical: HRs obtained with no covariates in the model that used 3 groups of CAG repeats (≤ 18 repeats, 19–22 repeats, and > 22 repeats) yielded an HR = 1.16 per incremental group (P = 0.41), whereas the model without covariates for 2 groups of CAG repeats (≤ 18 repeats vs. > 18 repeats) yielded an HR = 1.53 (P = 0.03). These result were similar to the adjusted HRs presented in Tables 4 and 5. To visualize further the impact of various groupings of CAG repeats, Figure 4 presents the distribution of CAG repeats among men at high risk of recurrence who did and did not develop recurrent disease during follow-up for this study. The distribution among men who remained disease free was symmetric, whereas the distribution among men who developed recurrent disease was skewed to the right with some suggestion of a bimodal distribution.

Figure 4. Frequency distribution of numbers of CAG repeats in exon 1 of the androgen receptor among men at high risk of developing recurrent prostate carcinoma who developed recurrent disease or remained disease free during follow-up.

DISCUSSION

A previous study of the impact of variability in CAG repeats on the survival of men with prostate carcinoma by Nam et al. had follow-up data for 253 men.15 Those authors found that the impact of CAG repeats varied according to the patient's risk. Among their patients who were at low risk for recurrence (Gleason score < 6, T2 tumor, and PSA level < 10 ng/mL), a CAG allele length of ≤ 18 repeats was associated with a relative risk of disease recurrence of 8.07 compared with patients who had > 18 repeats (P = 0.003). In contrast, for patients who were at high risk for disease recurrence (Gleason score ≥ 7, pT3/pT4 tumor, and PSA level > 10 ng/mL), Nam et al. reported a protective effect associated with fewer CAG repeats, in which the relative risk associated with ≤ 18 repeats was 0.72 (compared with > 18 repeats). This effect did not reach statistical significance (P = 0.41), but the results from tests for an interaction between the genotype and risk group by Cox proportional hazard models were highly significant (P = 0.004).

Two previous studies that included survival data had relatively few patients and did not stratify their analyses by risk. Cude et al. had follow-up data for only 91 patients.17 They reported no evidence of an association between CAG repeats and disease progression. Edwards et al. had data for 162 British WM.16 Those authors found that shorter alleles were associated with a mild increase in risk for recurrence, but the difference was not statistically significant.

Using the approach taken by Nam et al. to define high risk and low risk and a CAG cut-off level of 18 repeats, in our series, only 2 men at low risk developed recurrent disease, precluding further analysis. Among men who were at high risk, those with ≤ 18 CAG repeats tended to develop recurrent disease less often than men who had > 18 CAG repeats. This trend was observed for both WM and AAM. The results of log-rank tests for the effects of CAG repeats were not significant for WM (Fig. 3A) (P = 0.26); however, they approached statistical significance among AAM (Fig. 3B, P = 0.06) and were significant when men of both races were combined (P = 0.03). Proportional hazards models controlling for risk confirmed these observations (Table 5). (Note that, compared with Nam et al., in our models, we use the men with short CAG repeats as the reference group.)

To complement the current analysis, based on the risk groups and risk stratification approach used by Nam et al., we stratified patients according to whether they had organ-confined or locally advanced disease (Fig. 2) and grouped the CAG data into 3 categories with approximately equal numbers of patients (≤ 18 repeats, 19–22 repeats, and > 22 repeats) (Fig. 2, Table 4). Among AAM with locally advanced disease, recurrence risks differed by the numbers of CAG repeats (log-rank P = 0.04) (Fig. 2D), and men with higher numbers of repeats had an increased risk of recurrence, but the results were not statistically significant (HR = 1.25; P = 0.11) (Table 4).

Thus, the results from both the current series and the series reported by Nam et al. suggest that the number of CAG repeats has an impact on the recurrence of prostate carcinoma that is observed when patients are stratified according to their risk of developing recurrent disease. In the series by Nam et al.,15 in which data were available for patients both at low risk and at high risk of recurrence, the risk was in the opposite direction for low-risk men (in whom shorter CAG repeats have an adverse impact) and high-risk men (in whom longer CAG repeats have an adverse impact). In our series, only the men at high risk had a sufficient number of recurrences for meaningful study. Among those high-risk men, longer CAG repeats had an adverse impact. These strata-specific trends were statistically significant only for low-risk men in the series reported by Nam et al.15 (P = 0.003) and for high-risk men regardless of race in our current series (P = 0.03). Nam et al.15 also reported a highly significant genotype-risk group interaction effect (P = 0.004), whereby low numbers of CAG repeats had an adverse impact on men at low risk but a favorable impact for men at high risk of progressive disease. Using the same modeling approach with Cox proportional hazard models, our data did not show a statistically significant interaction; although, again, the current analysis was limited by the small numbers of low-risk patients who developed recurrent disease.

The results of the current study and the findings from Nam et al. are not consistent with a simplistic model of the role of this polymorphism in prostate carcinoma. A differential effect among patients at high and low risk suggests that the impact is not directly linked in all patients to mechanisms described in earlier biologic studies (in which smaller numbers of repeats were associated with increased signal transduction) or in the results from early case–control studies (in which the risk of diagnosis was related inversely to smaller numbers of CAG repeats). Although these inconsistencies raise questions about the biologic plausibility of the current findings, the high statistical significance of this interaction (P = 0.004) in the report from Nam et al. supports the existence of differential effects for men at different levels of risk, and the consistency of results among high-risk men from both races strengthens the reliability of our results. It is possible that the impact of the AR differs at different steps in the malignant process, perhaps because of changes in coregulators, transcription factors, the specificity of ligand requirements for receptor activation, interactions with other genes either linked or independent of this polymorphism, or other factors. Comparisons of the distribution of CAG repeats among men who developed recurrent disease versus those who did not suggest an especially high risk of recurrence for men with ≥ 25 CAG repeats and the possibility of different influences governing recurrence among subsets of patients (Fig. 4). Nonetheless, especially because our results stem from a reliance on analyses of subgroups in a setting of modest statistical power, they should be considered preliminary and will require confirmation by further research, preferably research supported by improved, clinically relevant biologic models, such as models developed from the results of expression or methylation profiling.

Almost 50% of the patients in the current series were AAM, compared with previous reports on the impact of CAG repeats on recurrence that included few AAM. AAM in our series had fewer CAG repeats (mean ± standard error of the mean, 19.8 ± 0.21 repeats) compared with WM (22.0 ± 0.15 repeats). Although their report did not include recurrence, Bennett et al. reported the numbers of CAG repeats by Gleason score, PSA level, and disease stage at diagnosis. Those authors found that the mean number of AR gene CAG repeats for white veterans was 21.9, compared with 19.8 repeats for African-American veterans (P = 0.001). Men with fewer CAG repeats were somewhat more likely to have Stage D prostate carcinoma (P = 0.09) but were not more likely to have a higher PSA concentration or Gleason score. Thus, Bennett et al. concluded that short CAG repeats on the AR gene were associated with African-American race and possibly with higher stage, but not with other clinical or pathologic findings.22 Kittles et al.23 reported fewer CAG repeats among Nigerians (mean, 16.7 repeats) compared with African Americans (mean, 17.8 repeats), Euro-Americans (mean, 19.7 repeats), and Asians (mean, 20.1 repeats). Those authors noted that, in Nigeria, prostate carcinoma is the most commonly diagnosed malignancy among men, with an incidence similar to that reported for African Americans. In addition, Kittles et al. reported greater allelic and haplotype diversity among populations of West-African descent compared with non-African populations.23

We have reported elsewhere that race/ethnicity was an independent predictor of disease progression in our total cohort of men who underwent radical prostatectomy.18, 19 Using specimens from several large, ongoing studies in metropolitan Detroit, we are studying associations between polymorphisms and the occurrence of high-grade prostatic intraepithelial neoplasia and latent malignancies in specimens obtained at autopsy; a population-based, case–control study of incident cases; and comparisons between normal tissue, primary tumors, and metastatic disease. These studies will allow us to examine the impact of polymorphisms at each stage of development of the disease from early hyperplastic lesions to advanced disease, with parallel studies of WM and AAM. Perhaps the major impact of fewer CAG repeats occurs at an earlier or later stage in prostate carcinoma progression than the stage addressed in the current study.

Acknowledgements

The authors thank Nimesh Patel for his tireless computer programming.