Background: Evidence of the chemopreventive effects of the dietary antioxidants α-tocopherol (vitamin E) and l-selenomethionine (selenium) comes from secondary analysis of two phase III clinical trials that found treatment with these antioxidants reduced the incidence of prostate cancer. To determine the effects of selenium and vitamin E in blood and prostate tissue, we undertook a preoperative feasibility study complementary to the currently ongoing Selenium and Vitamin E Cancer Prevention Trial.

Methods: Forty-eight patients with clinically localized prostate cancer enrolled on this 2 × 2 factorial design study were randomized to take selenium, vitamin E, both, or placebo for 3 to 6 weeks before prostatectomy. Sera were collected from patients before and after dietary supplementation. Thirty-nine patients were evaluable, and 29 age-matched disease-free men served as controls. Mass profiling of lipophilic serum proteins of lower molecular weight (2-13.5 kDa) was conducted, and mass spectra data were analyzed using custom-designed software.

Results: Weighted voting analyses showed a change in sera classification from cancerous to healthy for some patients with prostate cancer after dietary intervention. ANOVA analysis showed significantly different treatment effects on prediction strength changes among the four groups at a 95% confidence level. Eliminating an outlying value and performing post hoc analysis using Fisher's least significant difference method showed that effects in the group treated with the combination were significantly different from those of the other groups.

Conclusion: In sera from patients with prostate cancer, selenium and vitamin E combined induced statistically significant proteomic pattern changes associated with prostate cancer–free status.

No reliable and sensitive method currently exists to diagnose prostate cancer or to monitor therapeutic interventions to treat it. Despite the widespread clinical use of the serum prostate–specific antigen test, it is neither sufficiently sensitive nor specific for early detection of prostate cancer, and it is not predictive of biological behavior of tumors (1-3). Because of the inherent heterogeneity of prostate tumors, it is likely that a panel of proteins could serve better than the prostate-specific antigen as “signatures” for modulation of the molecular and cellular events occurring during carcinogenesis and therapeutic intervention. In fact, studies in several different tumor types, including prostate, brain, gastrointestinal, breast, ovarian, and hepatocellular cancers, have shown that serum protein patterns could be used to differentiate cancer from controls (4-9). Proteomics, because it considers all serologic proteins, could potentially identify multiple biomarkers instead of only one. Although none of these proteins alone merits total reliance, together their profiling pattern, assisted by statistical analyses, could yield a quick and reliable diagnosis of prostate cancer (4, 10, 11). Furthermore, this qualitative approach is able to reveal the molecular and cellular effects of therapeutic or dietary intervention in a period as short as several weeks.

Selenium and vitamin E are important intracellular antioxidants, but their function in preventing prostate cancer was initially discovered in studies intended to investigate their chemopreventive role in other cancers. Selenium's effect on prostate cancer, for example, was found during a study of its effects on skin cancer (12). In a randomized controlled study, 1,312 patients with skin cancer were given 200 μg of selenium orally (in the form of selenized yeast) or placebo daily. The study did not find that selenium treatment protected against the development of skin cancer, but a secondary analysis of the results revealed a 63% reduction of prostate cancer in patients treated with selenium. In an update from the Nutritional Prevention of Cancer trial, investigators reported finding a continued protective effect by supplemental selenium in data through the end of the blinded treatment in 1996. Selenium reduced the overall incidence of prostate cancer by 49%, albeit the effect was limited to patients with lower baseline prostate-specific antigen and plasma selenium concentrations (<123.2 ng/mL; ref. 13).

Likewise, findings of the chemopreventive effects of vitamin E (α-tocopherol) were also from secondary analysis of another study. The α-tocopherol, β-carotene study, which evaluated 29,133 male smokers between 50 and 69 years of age from southwestern Finland, was initially designed to find vitamin E's chemopreventive effect against lung cancer as measured in incidence. Results showed no reduction in lung cancer incidence after 5 to 8 years of dietary supplementation of α-tocopherol or β-carotene (14); however, a secondary study of the 6-year follow-up showed a significant reduction in prostate cancer incidence among participants treated with vitamin E. Inasmuch as these studies were initially designed for lung and skin cancer, findings related to prostate cancer must be treated with caution. However, a prospective randomized clinical trial with reduction in the incidence of prostate cancer as an a priori hypothesis is still needed to evaluate the role of vitamin E and selenium as chemopreventive agents for prostate cancer.

In 2001, a phase III chemoprevention trial in prostate cancer—the Selenium and Vitamin E Cancer Prevention Trial (SELECT)—was initiated. The primary end point of SELECT is the detection of a 25% reduction in the incidence of prostate cancer for selenium or vitamin E alone with an additional 25% decrease for the combination. This multicenter, randomized, placebo-controlled, double-blind, phase III trial of supplemental selenium and vitamin E is expected to enroll >32,000 men over 7 to 15 years. Creating a complementary trial employing the identical formulations of selenium and vitamin E, we undertook a preprostatectomy placebo–controlled feasibility study to determine the biologic effects of these agents in blood and prostate tissue. Using serum samples from 39 patients evaluable for blood biomarker correlative study and 29 healthy controls, we examined whether differential serum proteomic patterns could be induced by short-term antioxidant supplementation.

Patients

Forty-eight presurgical patients with clinically organ-confined prostate cancer at The University of Texas M.D. Anderson Cancer Center in Houston were enrolled in a randomized, double-blind trial of vitamin E and selenium between February 2001 and April 2002. Twenty-nine age-matched men, who consented to participate at the University of Massachusetts, served as prostate cancer–free controls, providing serum samples but receiving no treatment. The protocol and informed consent forms were approved by the respective universities' Institutional Review Boards, and an informed consent form was signed by all patients and controls before participation.

Eligibility criteria required patients to have clinical stage T1c/T2 disease, a prostate-specific antigen level <10 ng/mL within 3 months of study registration, a Gleason score ≤7, and a prostatectomy scheduled 3 to 6 weeks after the start of treatment. Patients taking >50 μg of selenium and/or ≥300 IU of vitamin E over 3 consecutive days within 1 month of registration were excluded. Once enrolled, patients received 400 IU of vitamin E (all rac α-tocopheryl acetate), 200 μg of selenium (l-selenomethionine), a combination of vitamin E and selenium, or placebo daily for 3 to 6 weeks. The formulations of selenium and vitamin E were the same as those in the ongoing SELECT. Each patient also received a multivitamin and 250 mg of vitamin C daily. Each multivitamin contained high-selenium yeast (30 μg) and d-α-tocopherol (30 IU), contributing to incremental increases in supplemental selenium and vitamin E of 15% and 7.5%, respectively. Serum samples were collected from the patients before and after treatment.

Proteomics Analysis

The venous blood samples were collected in serum separator tubes from patients participating in the study at M.D. Anderson before and after treatment. The samples were initially allowed to clot at room temperature and then were centrifuged to separate the cellular component from the supernatant within 2 hours of collection. Aliquots of supernatant were promptly frozen and stored at −80°C until analyzed. The samples were archived for 11 to 26 months. The samples from the 29 prostate cancer–free controls from the University of Massachusetts were processed following similar standard procedures. These control samples were archived at −80°C for 1 to 12 months until analyzed. All samples for the training set were analyzed at the same time, and none of the samples were discarded on the basis of turbidity or hemolysis. All samples used in the experiment were thawed only once.

Lipophilic proteins/peptides were extracted from each serum sample with UltraMicroSpin C18 columns (The Nest Group, Southboro, MA) following a previously published protocol (4). In brief, sera were diluted by mixing 4 μL crude serum of each sample with 200 μL solution containing 5% acetonitrile (MeCN) and 1% trifluoroacetic acid, which was in turn passed through the C18 column and washed thrice with 50 μL of the 5% MeCN/1% trifluoroacetic acid solution. The mixture was finally eluted with 30 μL of 75% MeCN/1% trifluoroacetic acid solution. Each washing and elution cycle was done by centrifugation at 2,500 × g. Five microliters of the final eluted solution were mixed with an equal volume of 50% saturated matrix, consisting of sinapic acid in 50:50 5% MeCN/1% trifluoroacetic acid (4).

All data were obtained using the Ciphergen Protein Biosystem II with delayed extraction in conjunction with Ciphergen 2.1 software. The dried droplet method was used to crystallize 1 μL of the matrix/sample mixture on a blank sample probe (4). An automatic protocol was designed to accomplish the data collection.

The variables for the protocol were set as follows: the mass window was set to acquire data from 0 to 20 kDa and was optimized between 3 and 10 kDa, yielding an extraction time delay of 600 ns. Each mass spectrum is the result of averaging 200 optimal shots. The intensity of the UV nitrogen laser (337 nm) was set at 240 V, the filter average was set to 0.2 times the expected peak width, and the baseline smoothing was set to 25 points before baseline correction. On the day of the experiment, external calibration was carried out with the following protein standards: [M + 1]+ and [M + 2]+ peaks from each standard-bovine insulin (5,733.6), bovine ubiquitin (8,564.8), and bovine cytochrome c (12,230.9). An internal calibration was completed with a combination of external protein standards and the extracted serum proteins in order to mark proteins found in each patient sample that encompassed the range of molecular weights of interest. In this way, each spectrum was internally calibrated with endogenous proteins present in each sample.

Data Processing and Analysis

Using Peakfit version 4.0 to perform the processing, we narrowed the mass spectra frame to capture data between 2 and 13.5 kDa, eliminating data that, from our experience, do not play a significant role in the prediction strength calculation. We then smoothed the spectra display data with a Gaussian convolute algorithm, adjusted background using a second derivative zero technique, normalized the spectra to 1.0, and zeroed the negative data. The processed mass spectra data were then exported to a text file.

The resulting text file was in turn imported into the data-processing application software MOCS,8

8

Modulized Classification System, Sun and Ho, unpublished.

which had been custom-designed and implemented for the weighted voting method. MOCS did all the necessary subsequent steps in the analysis automatically with the input of several variables.

First, MOCS lined up the data points from all smoothed and normalized spectra against a single mass-to-charge (M/Z) value array, which was calculated to the mass accuracy of the internally calibrated spectra of 150 ppm. The aligned data were further standardized to set all intensity values between 0 and 1. The standardization was done using the equation

\[\mathit{S}_{\mathrm{standardized}}\ =\ (\mathit{S_{x}}\ {-}\ \mathit{S}_{\mathrm{min}})/(\mathit{S}_{\mathrm{max}}\ {-}\ \mathit{S}_{\mathrm{min}}),\]

where Sx is the data point that needs to be standardized, Smin is the lowest intensity of all spectra for the M/Z value, and Smax is the highest intensity of all spectra for this M/Z value. By using this standardization strategy (4), we gave all intensities equal significance.

The standardized spectra data contained the intensity of all M/Z values from 2 to 13.5 kDa. We optimized the training set data and chose the top 1,000 M/Z values that had the most influence on the clustering. To determine the top 1,000 M/Z values, we calculated the weighted value for each M/Z. The training set consisted of two groups: pretreatment samples from 39 men who had prostate cancer and samples from the 29 controls who did not have prostate cancer. Those M/Z values with larger distances between the means of intensity of the two groups and smaller SDs within each group were granted values of higher weight. The equation used to calculate the weighted value, WM/Z, was

\[\mathit{W_{M/Z}}\ =\ (\mathit{M}_{\mathrm{h}}\ {-}\ \mathit{M}_{\mathrm{c}})/(\mathit{d}_{\mathrm{h}}\ +\ \mathit{d}_{\mathrm{c}}),\]

where Mh is the mean of the healthy group, Mc is the mean of the group with prostate cancer, dh is the SD of the healthy group, and dc is the SD of the group with prostate cancer (15). This way, the M/Zs with the most significant values differentiating serum of those with prostate cancer from serum of controls have the highest weights.

The 1,000 M/Zs with the highest weighted values were selected and used for comparison against the posttreatment samples in the weighted voting analysis. For each of the posttreatment samples, the training set voted on each of the 1,000 top-weighted M/Z values using the equation,

\[\mathit{V_{M/Z}}\ =\ \mathit{W_{M/Z}}\{[(\mathit{S}_{\mathrm{standardized}}\ {-}\ \mathit{M}_{\mathrm{all}})/\mathit{d}_{\mathrm{all}}]\ {-}\ \mathit{B_{M/Z}}\},\]

where WM/Z is the weighted value for M/Z, Sstandardized is the standardized intensity value of the posttreatment sample at the M/Z value, Mall is the mean of all training set values at this M/Z, and BM/Z is the statistically derived border of this M/Z (15), defined as

\[\mathit{B_{M/Z}}\ +\ (\mathit{M}_{\mathrm{h}}\ +\ \mathit{M}_{\mathrm{c}})/2,\]

where Mh is the mean of all samples in the healthy group and Mc is the mean of all samples in the group with prostate cancer. In this study, a positive vote indicated that at this M/Z value, the training set classified the sample into the group associated with being disease-free, whereas a negative vote classified the sample into the group associated with having prostate cancer. The resulting vote (VM/Z) for each of the 1,000 M/Z values was used to calculate the sum of all positive votes (Vp) and the sum of all negative votes (Vn), respectively. A larger absolute value of Vp classified the sample into the group associated with being disease-free, and a larger absolute value of Vn classified the sample into the group associated with having prostate cancer. To describe the chance that the classification was accurate, we used a variable called prediction strength as a quantitative measurement (4, 15). The prediction strength is calculated with the equation

\[\mathrm{PS}\ =\ (|\mathit{V}_{\mathrm{p}}|\ {-}\ |\mathit{V}_{\mathrm{n}}|)/(|\mathit{V}_{\mathrm{p}}|\ +\ |\mathit{V}_{\mathrm{n}}|).\]

The prediction strength values range from −1 to 1. A positive prediction strength value is associated with being disease-free, and a negative value is associated with having prostate cancer. The closer the value is to 0 (the border of the two groups), the less chance that the classification is accurate; the closer it is to either −1 or 1, the more likely that classification is accurate. An arbitrary cutoff can be set to exclude certain samples. Setting such a cutoff increases the reliability of the prediction at the cost of lowered usability, because with this method, more samples fall into the category of nonclassifiable. In our study, no arbitrary cutoff value was applied; thus, any sample with a positive prediction strength value was classifiable as disease-free, and any sample with a negative prediction strength value was classifiable as prostate cancer.

Prediction strength values that resulted from weighted voting were calculated for the patients before and after the dietary supplementation treatment. The changes in the prediction strength values after the dietary supplementation were calculated for each sample, and the data set was analyzed using one-way ANOVA and Fisher's least significant difference method. Compared with other samples in the placebo group, sample 9 had an unusually high prediction strength value change toward the positive direction. It contributed 49.5% of the total statistical significance in the ANOVA analysis and lay outside the 95% confidence interval of the samples in the placebo group. In line with common practice, this outlier was excluded from the final analysis of the data set.

Prediction strength values determined after weighted voting analyses showed that before dietary supplementation, most serum samples from the patients with prostate cancer had negative prediction strength values, values that are associated with prostate cancer (Table 1; Fig. 1). However, after dietary supplementation, many of the serum sample values obtained from patients given the combination were positive prediction strength values, which are commonly associated with the absence of disease (Fig. 2). Generally speaking, the closer the prediction strength value to 1 or −1, the more likely that the classification is correct.

ANOVA analysis on the entire data set (Table 2), as well as on the set without sample 9 (Table 3), both showed significant statistical differences in prediction strength value changes among the four groups at the 0.05 level. Analysis on the data without sample 9 yielded an F ratio of 3.75 (Table 3). The probability of the result, assuming the null hypothesis (that all diet supplement treatment had the same effects on prediction strength change) is 0.020, which suggests that the effect on each arm was significantly different from the others at the 0.05 level. In the post hoc analysis of data without sample 9, the results at the 0.05 significance level using Fisher's least significant difference test suggest that the different treatment methods had significantly different effects: significant differences were found between the group receiving the combination and all other three groups (Table 4). There were no significant differences in effect among the groups which received selenium treatment, vitamin E treatment, or the placebo group (Table 4).

Furthermore, ANOVA evaluation indicated no significant differences among the groups on the basis of age (F ratio = 0.638), duration of treatment (F ratio = 1.210), pretreatment prostate-specific antigen level (F ratio = 0.630), relation to biopsy Gleason score (F ratio = 1.085), relation to prostatectomy Gleason score (F ratio = 0.144), or pretreatment performance status (F ratio = 0.256).

Because evidence of the chemopreventive properties of selenium and vitamin E comes from secondary analysis of two phase III clinical trials, researchers conducted SELECT, a multicenter, randomized placebo-controlled, double-blind phase III trial with reduction of the incidence of prostate cancer as an end point. In order to use proteomics to characterize the biological effects of these agents in blood, we used serum samples of patients treated with the selenium and vitamin E formulations identical to those in SELECT for 3 to 6 weeks preprostatectomy and evaluated serum samples from 29 disease-free age-matched men as controls. Because this was a study in preprostatectomy patients with prostate cancer in whom the benefits of short-term intervention of antioxidants was unknown, and because it was anticipated that recruitment of patients to a placebo-controlled study would be difficult, we decided to provide a multivitamin and extra vitamin C daily. Each tablet contained high-selenium yeast (30 μg) and d-α-tocopherol (30 IU), contributing to incremental increases in supplemental selenium and vitamin E of 15% and 7.5%, respectively. Thirty-nine of 48 patients were evaluable. Weighted voting analysis of serologic protein mass profiling data indicated a change in classification mostly in patients who had received the selenium and vitamin E combination. These patients' values changed from levels associated with disease to those associated with its absence. In post hoc analysis, the shift in serum classification, protein profiling as quantified by changes in prediction strength value, indicated that effects in the groups treated with the selenium and vitamin E combination were significantly different from those of the groups treated with selenium, vitamin E, or placebo.

The time between histologic diagnosis of prostate cancer and definitive therapy (prostatectomy) provides an opportunity to test the biological effects of chemopreventive agents with minimal side effects. The phase II preprostatectomy cancer prevention trials employing short interventions with chemopreventive agents have been widely adopted, and in this setting, proteomics provides an effective method to discover biomarkers easily measurable in such biofluids as serum. In this study, enrolling patients slated for prostatectomy, rather than disease-free men as in the SELECT, was determined best because it offered full study of the biological effects of the study agents on the prostatectomy specimen.

Our data analysis is innovative in that, by using serologic protein patterns to evaluate the patients' status, potentially many biomarkers were used and had effects on the final results. Unlike other serologic protein mass profiling studies clustering methods, which are usually aimed at differentiating only two states (with cancer and without) as diagnosis results, the use of weighted voting analysis and prediction strength values provides a way to quantitatively describe the pattern changes among multiple groups and, in turn, facilitates other statistical analysis on the results, such as ANOVA and post hoc analysis.

The lengthy and unpredictable natural history of prostate cancer makes implementing a primary chemoprevention trial logistically difficult. Prostate cancer tumor progression from microscopic to metastatic disease is poorly understood. Oxidative stress–induced genomic and mitochondrial DNA mutations have been implicated as tumor promotion factors in recent population studies (16, 17). Offering a viable option in this setting is primary prevention, especially given that prostate cancer has as long latency period (18). Although undertaking a chemoprevention trial to evaluate a drug's efficacy with cancer incidence reduction as the primary end point in prostate cancer would be both expensive and time-consuming, it is logical to identify cellular and molecular intermediate biomarkers in a particular mechanistic pathway in prostate tissue or blood, which is more easily accessible. Such intermediate biomarkers could serve as surrogate end points for a reduction in the cancer incidence in large randomized trials.

The post hoc analysis using protected Fisher's least significant difference method, which is based on the previous rejection of the null hypothesis with F test, showed that different dietary supplementation treatments had significantly different effects on the groups. One sample (an extreme outlier) was omitted under the assumption that results within each group are normally distributed. The clinical data of this patient was checked, and no significant difference was found between his data and data of other patients in the group. There exists the possibility that the patient may have taken alternative treatment, which could have affected his prediction strength values and skewed the range upward. Such an extreme outlying value as his will not only affect the validity of the test results, but also, by increasing the variability of each group, significantly reduce test sensitivity.

The statistical analysis of serologic protein patterns produced by the mass spectrometry method yielded a quick and direct view of the effects of dietary supplement treatment without performing actual biopsy, a procedure with findings that cannot be matched in quality by other available methods. This approach, however, did not reveal details of how the combination of selenium and vitamin E affected the mechanisms of cancer progression. Furthermore, it did not correlate changes in protein patterns to tumor grade, pretreatment prostate-specific antigen, or treatment duration. The small sample size in each category would prevent meaningful analysis.

Despite efforts to ensure that the venous blood samples in both the Massachusetts and Texas studies were collected, processed, and stored similarly, differences in the duration of storage occurred. Samples from patients with prostate cancer were archived 11 to 26 months; those from cancer-free controls were archived 1 to 12 months. Although we are unsure what, if any, bias this difference may have introduced, it should be acknowledged that differences in storage duration occurred and that they may have affected the results of the study. Having the controls at another site, although not ideal, was not expected to affect results.

Studies using protein purification in serum and tissue microarrays to examine selected proteins that are most affected by the treatment are currently under way.

Grant support: Research grants CA88761 and CA88084 (to J. Kim) from the National Cancer Institute, NIH, Department of Human Services.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.

Note: J. Kim and P. Sun contributed equally to this work.

We thank Beth W. Allen for excellent editorial assistance.

1
Grossfeld GD, Carroll PR. Prostate cancer early detection: a clinical perspective.
Epidemiol Rev
2001
;
23
:
173
–80.
2
Pannek J, Partin AW. The role of PSA and percent free PSA for staging and prognosis prediction in clinically localized prostate cancer.
Semin Urol Oncol
1998
;
16
:
100
–5.
3
Stamey TA, Johnstone IM, McNeal JE, Lu AY, Yemoto CM. Preoperative serum prostate specific antigen levels between 2 and 22 ng/ml correlate poorly with postradical prostatectomy cancer morphology: prostate specific antigen cure rates appear constant between 2 and 9 ng/ml.
J Urol
2002
;
167
:
103
–11.
4
Mobley JA, Lam YW, Lau KM, et al. Monitoring the serologic proteome: the latest modality in prostate cancer detection.
J Urol
2004
;
172
:
331
–7.
5
Villanueva J, Philip J, Entenberg D, et al. Serum peptide profiling by magnetic particle-assisted, automated sample processing and MALDI-TOF mass spectrometry.
Anal Chem
2004
;
76
:
1560
–70.
6
Feldman AL, Espina V, Petricoin EF, Liotta LA, Rosenblatt KP. Use of proteomic patterns to screen for gastrointestinal malignancies.
Surgery
2004
;
135
:
243
–7.
7
Poon TC, Yip TT, Chan AT, et al. Comprehensive proteomic profiling identifies serum proteomic signatures for detection of hepatocellular carcinoma and its subtypes.
Clin Chem
2003
;
49
:
752
–60.
8
Li J, Zhang Z, Rosenzweig J, Wang YY, Chan DW. Proteomics and bioinformatics approaches for identification of serum biomarkers to detect breast cancer.
Clin Chem
2002
;
48
:
1296
–304.
9
Petricoin EF, Ardekani AMA, Hitt BA, et al. Use of proteomic patterns in serum to identify ovarian cancer.
Lancet
2002
;
359
:
572
–7.
10
Qu Y, Adam BL, Yasui Y, et al. Boosted decision tree analysis of surface-enhanced laser desorption/ionization mass spectral serum profiles discriminates prostate cancer from noncancer patients.
Clin Chem
2002
;
48
:
1835
–43.
11
Adam BL, Qu Y, Davis JW, et al. Serum protein fingerprinting coupled with a pattern-matching algorithm distinguishes prostate cancer from benign prostate hyperplasia and healthy men.
Cancer Res
2002
;
62
:
3609
–14.
12
Clark LC, Combs GF Jr, Turnbull BW, et al. Effects of selenium supplementation for cancer prevention in patients with carcinoma of the skin. A randomized controlled trial. Nutritional Prevention of Cancer Study Group.
JAMA
1996
;
276
:
1957
–63.
13
The α-Tocopherol, β Carotene Cancer Prevention Study Group. The effect of vitamin E and β carotene on the incidence of lung cancer and other cancers in male smokers. The α-Tocopherol, β Carotene Cancer Prevention Study Group.
N Engl J Med
1994
;
330
:
1029
–35.
14
Duffield-Lillico AJ, Dalkin BL, Reid ME, et al. Selenium supplementation, baseline plasma selenium status and incidence of prostate cancer: an analysis of the complete treatment period of the Nutritional Prevention of Cancer Trial.
BJU Int
2003
;
91
:
608
–12.
15
Golub TR, Slonim DK, Tamayo P, et al. Molecular classification of cancer: class discovery and class prediction by gene expression monitoring.
Science
1999
;
286
:
531
–7.
16
Karunasinghe N, Ryan J, Tuckey J, et al. DNA stability and serum selenium levels in a high-risk group for prostate cancer.
Cancer Epidemiol Biomarkers Prev
2004
;
13
:
391
–7.
17
Chen JZ, Gokden N, Greene GF, Green B, Kadlubar FF. Simultaneous generation of multiple mitochondrial DNA mutations in human prostate tumors suggest mitochondrial hyper-mutagenesis.
Carcinogenesis
2003
;
24
:
1481
–7.
18
O'Shaughnessy JA, Kelloff GJ, Gordon GB, et al. Treatment and prevention of intraepithelial neoplasia: an important target for accelerated new agent development.
Clin Cancer Res
2002
;
8
:
314
–46.