Clinical utility of antioxidant gene expression levels in circulating cancer cell clusters for the detection of prostate cancer in patients with prostate-specific antigen levels of 4–10 ng/mL and disease prognostication after radical prostatectomy


Professor Dr Michael Giesing, Onkokonsult, Zum Herzfeld 5, D-49536 Lienen, Germany.



To test antioxidant genes (AOX) expression in circulating cancer cell clusters (CCC). A novel method using molecular, polymerase chain reaction (PCR)-based detection of CCC was applied for predicting prostate cancer and to assess the effect of radical prostatectomy (RP) on reducing CCC and for prognostication of relapse-free survival (RFS), as serum total prostate-specific antigen (tPSA) has limited specificity at 4–10 ng/mL.


In all, 240 patients were enrolled in the study, 129 for tumour diagnosis and 111 after RP for disease prognostication. Filtration assay in previously fractionated mononuclear cells (MNC) was used to enrich the CCC and large cells, which were retained in a mesh of 20 µm width. To establish the malignant nature of these cells they were analysed for genomic imbalances detected via PCR-assays of loss of heterozygosity in tumour suppressor loci and of DNA amplification in protooncogen loci. As a screening test in daily practice, real-time reverse transcription (RT)-PCR of AOX was introduced to overcome the laborious and expensive DNA tests. The AOX chosen were glutathione peroxidase (GPX1), Mn-dependent superoxide dismutase 2 (SOD2) and thioredoxine reductase (TXNRD1); selected from 67 marker candidate genes according to sensitivity and specificity data. AOX overexpression in CCC serves as a general marker for solid tumours needing, however, organ markers to relate to the organ of origin. Androgen receptor (AR), PSA and prostate-specific membrane antigen mRNAs served as organ markers for the prostate. Signals were detected in patients’ MNC and to a minor level in CCC, rendering to CCC a substantial loss in epithelial features equivalent to a lower grade of epithelial differentiation. Organ markers in the MNC fraction were positive in <85% of AOX testing.


The AOX test was tumour predicting (P < 0.001) with a sensitivity of 86%, specificity 82%, positive predictive value 69%, negative predictive value 92%, accuracy 83% and odds ratio (OR) of 28. SOD2 and TXNRD1 expression correlated to tumour size and Gleason score. Objective assessment for the evaluation of the molecular cell markers was achieved by receiver operating characteristic (ROC) curves. The areas under the ROC curve values of the AOXs were 0.7–0.9. RP was followed by a complete clearance of AOX-expressing cells. After RP, a subgroup of patients had residual CCC over-expressing only SOD2 and GPX1 indicating incomplete clearance by RP. Sustained overexpression of SOD2 and GPX1 accounted as risk factors for distant tumour recurrence (P = 0.003) mainly for bone metastases (97% M1b) as evaluated by Kaplan–Meier curves. In univariate analysis the tumour size had a limited effect on the probability of RFS (P = 0.05). In multivariate analysis tumour size, nodal status and Gleason score had no effect. This can partially be attributed to the higher risk level of pathological variables in the AOX over-expressing group but also to ineffective endocrine therapy resulting in marked overexpression of ARs and GPX1, the lead prognosticator gene. The AOX expression level allowed the identification of patients with high progression risk, who have more favourable pathological variables.


The AOX testing of CCC is a novel method with excellent prognostic and predictive properties enabling the monitoring of therapies, e.g. effects of RP and endocrine therapy. We speculate that the continuing elevated expression of AOX with special emphasis on GPX1 acts as survival and defence mechanism in CCC required in an atypical environment prone to escape from immune surveillance.


circulating cancer cell clusters and large cancer cells


antioxidant genes


glutathione peroxidase


Mn-dependent superoxide dismutase 2


thioredoxine reductase


total serum PSA level


(positive) (negative) predictive value


radical prostatectomy


mononuclear cells


androgen receptor


prostate-specific membrane antigen


cytokeratin 20


circulating epithelial cells


positron emission tomography


glyceraldehyde-3-phosphate dehydrogenease


comparative specific gene expression


receiver operating characteristic


reactive oxygen species.


The detection of prostate cancer is generally achieved by DRE, TRUS and serum total PSA level (tPSA) measurements. The methods differ in sensitivity, specificity and predictive values [1]. The detection rate of prostate cancer increases with serum tPSA rendering the 4–10 ng/mL level as a diagnostic grey zone associated with a positive predictive value (PPV) of 20–30%[1]. This group of men accounts for 82% of men undergoing radical prostatectomy (RP). A threshold value of tPSA of 4 ng/mL is most commonly used for prostate biopsy. Consequently, ≈70% of biopsies remain without tumour diagnosis at tPSA levels of 4–10 ng/mL. Half of the tumours may be missed [2]. The ability to detect cancer is determined by sensitivity, which is relatively high for tPSA (60–80%). However, the specificity is poor (25%) leading to false-positive findings and unnecessary biopsies in ≈75% of the cases. Improvement of PSA specificity by measuring PSA density, velocity or various molecular forms has been a matter of ongoing research for many years as reviewed by Stenman et al.[3]. In daily routine practice a considerable portion of men with low tPSA levels are hesitant to undergo biopsy, asking for further evidence of disease [4,5]. Despite professional counselling a high proportion of suspect individuals in those studies rejected biopsy. A major argument was being afraid that manipulation of the gland might lead to the iatrogenic intravasation of prostate cells. Schamhart et al.[6] have reviewed the shedding of epithelial prostate cells through various intervention strategies as studied by reverse transcription (RT)-PCR of tPSA and prostate-specific membrane antigen (PSMA). However, the specificity of such events and the consequences for survival remain to be determined. Moreover, the proof of the malignant nature of epithelial prostate cells as shown by DNA aberrations has not yet been shown in detail to a satisfactory extent [7]. In addition, scattered epithelial cells are not the only form of cells appearing in the blood stream. Another type of circulating cell has been observed in the form of cell clusters [8–10]. The malignant nature of these cells has not yet been determined. The objective of the present study was to investigate the clinical utility of circulating cancer cell clusters (CCC) for cancer diagnosis, the efficacy of RP and the prognostic role of minimal residual disease after RP. In the context of cancer diagnosis and prognosis, we generated data according to the requests of scientific societies and regulatory authorities concerning molecular cancer markers [11]. For the first time CCC were enriched by size as previously described for breast malignancies [12,13]. The molecular marker for CCC was based on the real-time RT-PCR of normalized mRNA of the antioxidant genes (AOX) glutathione peroxidase (GPX1), Mn-dependent superoxide dismutase 2 (SOD2) and thioredoxine reductase (TXNRD1), calculated from CCC and patients’ mononuclear cells (MNC). AOX were chosen from comparative studies of medium density microarrays and real-time RT-PCR performed at the Institute for Molecular Nanotechnology (unpublished results). The correlation of results between the two platforms was poor, e.g. Spearman’s correlation by rank for SOD2 expression between microarray and Taqman was P = 0.98. In daily routine practice real-time RT-PCR was the method of choice. Overexpression of the three chosen AOX in CCC had the highest sensitivity and specificity of the 67 candidate genes (unpublished data). There were similar findings for extravasating large cells (≥20 µm in diameter) and CCC from other tumour entities in our laboratory suggesting the general applicability of the AOX-test system (patent file No. WO2004/019037A2). To confine the AOX results to the prostate gland, we studied the appearance of epithelial and organ markers in CCC. We used RT-PCR of cytokeratin 20 (CK20), tPSA, prostate-specific membrane antigen (PSMA) and androgen receptor (AR). To date only circulating epithelial cells (CEC) have been fractionated by anti EpCAM-conjugated immunomagnetic beads to study the cellular distribution of molecular signatures [14].

We describe for the first time the role of CCC marker genes in improving prostate cancer diagnosis in patients with tPSA levels of 4–10 ng/mL, in showing the efficacy of RP by reducing minimal residual CCC and in identifying patients remaining at risk for disease progression. We also classify the cancer nature of CCC.


In all, 240 patients were enrolled in the study after giving informed consent. Demographic data are listed in Table 1. The patients were recruited by >30 urologists in their offices in cooperation with clinical urology departments and by clinicians, respectively. Patients were divided into two main cohorts. The first cohort consisted of 129 patients with tPSA levels of 4–10 ng/mL before diagnosis enrolled for molecular testing of CCC. This group represents 82% of patients undergoing RP in the clinic participating in the present study. Although a biopsy before molecular testing was not an exclusion criterion, 31 patients had previously had a negative biopsy. After molecular testing an additional seven patients were excluded from further evaluation because they were reluctant to undergo clinical diagnosis. After molecular testing clinical diagnosis was made by biopsy and in a subgroup of 12 patients in one centre by 11C-choline positron emission tomography (PET) followed by RP, if both tests, imaging and AOX, were highly positive. Magnetic resonance thermal imaging was used in one patient.

Table 1.  Demographic data
VariableMedian (range)P*
  • *

    t-tests for independent variables comparing groups D and C.

  • †The cohort after RP was subdivided into patients by the load of remaining CCC: group C had complete clearance of CCC while group D had incomplete clearance characterized by highly significant enhanced AOX expression. n.s., not statistically significant.

Before tumour diagnosis (n = 129)
Age, years64 (41–82) 
Time to diagnosis, months 6 (1–28) 
Gleason score 7 (6–9) 
Tumour size, pT 2 (1–4) 
Nodal status, pN 0 (0) 
Group A, tumour (n = 42)
Group B, no tumour (n = 87)
After RP (n = 111)
Group C, low progression risk (n = 77)
 Age, years65 (49–89) 
 RFS, months31 (1–216) 
 Gleason score 6 (2–10) 
 Tumour size, pT 2 (1–4) 
 Nodal status, pN 0 (0–1) 
 Surgical margins, R1 6 
 tPSA, ng/mL 0.6 (0.1–36.7) 
 Clinical diagnosis of previous local relapse, n 0/77 
Group D, high risk for progression (n = 34)
 Age, years67 (57–80)n.s.
 Time to relapse, months36 (0–146)n.s.
 Gleason score 8 (4–10)0.04
 Tumour size, pT 3 (1–4)n.s.
 Nodal status, pN 1 (0–2)<0.001
 Surgical margins, R1 4 
 tPSA, ng/mL 5.7 (2.1–38)0.4
 Clinical diagnosis of previous local relapse, n 11/28<0.001

The second cohort consisted of 111 patients after RP. Six additional patients were excluded from evaluation because of missing clinical data. In all, 77 patients remained with no tumour recurrence within the follow-up period. In all, 34 patients relapsed, 14 (41%) with local relapse, nine (26%) with nonregional lymph nodes (M1a), 33 (97%) with bone metastases (M1b), and four (12%) with metastases in other sites (M1c).

Peripheral blood was collected in heparinized Vacutainer systems (Becton Dickinson) and processed within 24 h.

To enrich circulating prostate cells, MNC were purified over a density gradient using Nycoprep 1.077 (Nycomed, Norway). MNC were washed twice in PBS (0.2% BSA; Life Technologies, Germany) and re-suspended finally in 12.2 mL PBS. MNC served for the enrichment of two circulating prostate cell fractions. MNC were rinsed 10 times using 5 mL PBS each. The final volume was 12.2 mL.

CEC were enriched from 5 mL filtrate using Ber-EP4 monoclonal antibody (mAb)-conjugated Dyna-Beads (Dynal, Norway) according to the manufacturer’s instructions

Circulating large cancer cells (≥20 µm in diameter) and CCC derived from 5 mL of the final MNC suspension were enriched by size using a column containing a polyester mesh (width 20 µm in diameter) as previously described [12]. Cells retained on the mesh were lysed with Trizol (Life Technologies, Germany) and processed to RNA analysis.

Genomic imbalances in CCC were analysed from Trizol extracts. DNA amplification of the loci erb B2 and c-myc was analysed as previously described by our laboratory using Trizol extracts [12]. Briefly, both targets were co-amplified with β-globin in circulating prostate cells using CD45-expressing cells as controls. Sense primers were labelled with fluorogenic molecules. The PCR products were analysed by capillary gel electrophoresis (Genescan genetic analyser 310, Applied Biosystems). The peak area threshold was ≥2.0 calculated from target:β-globin in prostate cells. For studies of allelic loss, the threshold for the peak integral value was ≥2000 and the peak noise/ratio was ≥50. Loss of heterozygosity was studied in the loci APC, RB, p53 and DCC as described elsewhere [12,15,16]. The primer sequences for loss of heterozygosity detection in the loci D7 S522 (GeneBank Accession Z17100), D7 S523 (Z17102) and D11 S1344 (Z24193) were obtained from the genome database. PCR and gel electrophoresis were performed as described by our laboratory [12].

For gene expression analysis, total RNA was prepared and processed to reverse transcription (RT) as previously described [17]. The efficiency of cDNA synthesis was monitored by real time quantitative RT-PCR for glyceraldehyde-3-phosphate dehydrogenease (GAPDH) using reference cell lines as indicated below to obtain a normalized cDNA standard curve. Then GAPDH expression values in cell fractions enriched from the blood of patients served as molecular cell counter. PCR amplification was performed in the presence of target-specific, double-fluorescence-labelled probes and quantified using the ABI PRISM 7700 Sequence Detection System (PE Biosystems, USA). Normalized reference expression values were calculated as cell equivalents related to target gene enriched reference cell lines and MNC after cell counting in a Neubauer chamber. Details of cell line preparation and of the PCR reaction conditions have been given elsewhere [12]. GAPDH-normalized reference values (Target/GAPDH) obtained are expressed as comparative specific-gene expression (CSE) calculated from the expression values of circulating prostate cells compared with patients’ MNC. CSE represents the unit of gene expression level of tumour cells retained in the filter mesh in comparison with patient’s MNC. CSE is a measure of prediction as well as of prognostication for CCC. Both features can be related to changes in cancer cell number, phenotype or both. Target gene primers and probe designed by using the Primer Express software 1.0 (Perkin-Elmer, USA) were purchased from TIBMOLBIOL, Germany. Primer designs are overspanning two exons without intron sequences thereby avoiding DNA contamination.

CCC marker RNAs: SOD2 (NM 000636; 6q25) sense: 5′-GTCACCGAGGAGAAGTACCAGG-3′ antisense: 5′-GGGCTGAGGTTTGTCCAGAA-3′ probe: 5′- CGTTGGCCAAGGGAGATGTTACA GCCC- 3′; ref. cell line: EFM192); TXNRD1 (NM_003330; 12q23-q24.1) sense: 5′-GGAG GGCAGACTTCAAAAGCTAC-3′antisense: 5′-ACAAAGTCCAGGACCATCACCT-3′ probe: 5′-TTGGGCTGCCTCCTTAGCAGCTGCCA-3′; ref. cell line: (MES), GPX1 (NM 201397; 3p21.3) sense: 5′-CTCGGCTTCCCGTGCAA-3′ antisense: 5′-TGAAGTTGGGCTCGAACCC-3′ probe: 5′-AGTTTGGGCATCAGGAGAACGCCAAGAA-3′; ref. cell line: EFM192).

The AOX-test (SOD2, TXNRD1, GPX1) in CCC was selected out of 67 candidate genes using microarray technology (Genestick®). Real-time RT-PCR was chosen because the data did not correspond with Genestick®-microarry data (overlapping <30%) thus generating the necessary sensitivity and specificity (data not shown).

Organ marker RNAs: AR (Ref.Seq. NM 000044; Location: Xq12) sense: 5′-CCATCTTGTCGTCTTC GGAAA-3′, antisense: 5′-CTGGGTTGTCTCCTC AGTGGG-3′, probe: 5′-ATGACTCTGGGAGCC CGGAAGCTGAA-3′; ref. cell line: MCF7); kallikrein 3, alias PSA (NM 145864; 19q13.41) sense: 5′-TGCGGCGGTGTTCTG-3′, antisense: 5′-ATGAAACAGGCTGTGCCGAC-3′, probe: 5′-AGCAAGATCACGCTTTTGTTCCTGATGCAGT-3′; ref. cell line: LNCaP-FGC); folate hydrolase, alias PSMA (NM 004476; 11p11.2) sense: 5′-TGTTCATCCAATTGGATACTA TGATGC-3′, antisense: 5′-TTGAGACTT CCTCTCTCCAGCTGC-3′, probe: 5′-AGAAGCTCCTAG AAAAAATGGGT GGCTCAGCA-3′; ref. Plasmid-DNA); keratin 20, alias CK20 (NM 019010; 17q21.2) sense: 5′-AGTGGTACGAAACCAACGCC-3′, antisense: 5′-GAGATCAGCTTCCACTGTTAGACG-3′, probe: 5′-AGCTGCGAAGTCAGATTAAGGATGCTCA-3′; ref. cell line: NCI H508.

Organ-confined target genes were analysed in CCC, Epcam-positive cell isolates (CEC) and MNC as indicated. In the latter case, MNC of apparently healthy male volunteers served as controls and calibrator. Target gene expression was considered positive if the cycle threshold value was at least one log lower compared with controls. AR expression in CEC was quantified (CSE) as indicated.

CCC were controlled for contamination by blood cells using real-time RT-PCR of protein tyrosine phosphatase receptor (CD45) gene expression (Y00062; 1q31-q32): sense: 5′-TGGAAGTGCTGCAATGTGTC-3′, antisense: 5′-CAAATGGTAACGTTCATGGG-3′, probe: 5′-ACTAAAAGTGCTCCTCCAAGCCAGGT-3′ (Ref. MNC of healthy volunteers). Patient’s matched CD45-positive lymphocytes were enriched from 0.2 mL of the MNC fraction using anti-CD45 mAb-coated magnetic beads (Dynal). The contamination allowed was 0.1% of total CD45 gene expression of matched lymphocytes for the CCC-fraction and 1.0% for CEC.

The t-test for independent variables was used to compare molecular tests between patient groups. Sensitivity, specificity, PPV, and negative predictive value (NPV), accuracy, odds ratios (ORs) and prevalence were calculated using the 2 × 2 contingency table method applying the chi-square test for statistical evaluation.

Receiver operating characteristic (ROC) analysis and areas under the curve were used as objective measures to evaluate the molecular cell markers. The Hanley–McNeil test was used for calculations of standard errors. The ROC calculation software was additionally programmed by StatSoft, Germany.

For examination of Gleason score, tumour size, nodal status and molecular cell markers before RP, Spearman’s correlation coefficient by rank was calculated.

For survival analysis after RP of the relapse-free survival (RFS) Kaplan–Meier curves were used. Patients with no relapse tumours were censored. Differences between groups were calculated using the log-rank test. Univariate and multivariate analysis for comparing patients’ tumour size, nodal status, Gleason score with progression of the disease was done using the Cox proportional hazards model. P values were derived from Wald’s chi-square test with P < 0.05 considered to indicate statistical significance. Molecular analyses for prospective long-term observations were partially performed from archived CCC, CEC and MNC [12]. The collection of further prospective data was achieved by questionnaires forwarded to urologists and patient interviews by telephone. A considerable portion of patients visited the laboratory for interviews.


Molecular AOX analysis in CCC was performed in 240 men at various clinical stages. Two cohorts were investigated (Table 1). The first cohort was analysed before tumour diagnosis and then subdivided into group A (tumour clinically confirmed) and group B (no tumour clinically diagnosed). The second cohort after RP was subdivided into patients depending on the load of remaining CCC. Group C had complete clearance of CCC while group D was characterized by incomplete clearance. Pathological variables indicated that group D represented patients with a high risk of progression. Gleason score as well as the lymph node status were significantly higher. Four patients in group C were found to be lymph node positive (pN1), six in group D were pN1, two pN2, and four in group D were pM1b or pM1c upon entering the study. Two of these patients underwent RP plus orchidectomy and two had no surgery. Local gland resection was incomplete (R1) in six patients in group C (pT3 and pT4), and in four in Group D. At ‘risk’ patients, i.e. those who were N+, R+, T3/4, Gleason >7 underwent adjuvant treatment (endocrine therapy and/or radiation) shortly after RP according to the recommended guidelines. Therefore, these patients were not available to assess without adjuvant treatment. The tPSA was elevated in both groups, but higher in group D than in group C, albeit not reaching statistical significance.

The molecular data obtained from these patients in AOX expression levels in CCC are summarized in Fig. 1. The molecular test system allowing the substratification of patients was designed to show and quantify the presence of CCC based according to the expression level of AOX (GPX1, SOD2, TXNRD1). The test system was chosen after microarray testing of 67 candidate RNAs in parallel to quantitative real-time RT-PCR. From the results obtained, real-time RT-PCR was chosen as the analytical method of choice in a daily routine setting.

Figure 1.

AOX expression in CCC at the tumour diagnostic level (A and B) and after RP (C and D). Relative specific expression (CSE of target gene/GAPDH in CCC/patient MNC) is given. Group A was diagnosed with prostate cancer, group B is no cancer in patients with tPSA of 4–10 ng/mL. After RP 33/34 patients (97%) in group D developed bone metastases (M1b) and had incomplete clearance of CCC by RP leading, therefore to a ‘high risk’ status prone to progressive disease unless additional therapies reverse the overexpression of CCC genes in group D to the level of Group C.

Figure 1 shows the CSE values (the ratio of target/GAPDH in CCC vs MNC) of three AOX before clinical diagnosis. Patients with tumours (group A) had higher values of each gene compared with patients with no tumours (group B; P < 0.001).

Prostate cancer was diagnosed in 42/129 patients (32.5%). The mean period from molecular analysis to clinical diagnosis and RP, respectively, was 7.2 months (95% CI 5.4–9.0). Different clinical diagnostic procedures were used in the present study. Biopsies were taken in 30/42 patients, 11C-choline PET was used in 11/42 and magnetic resonance thermal imaging was used in one patient. The mean Gleason score was 7 (95% CI 6–9). The distribution of the tumour size was T2 (83%), T3 (12%) and T4 (5%).

Predictive data of CCC genes were calculated from groups A and B for tumour prevalence, sensitivity, specificity, PPV, NPV, OR and test accuracy (Table 2). CCC genes are highly significant predictors (chi-square test). The best tumour diagnostic values, based on test NPV and test accuracy, were obtained with all three genes in combination. The ranking among individual AOX was SOD2 = GPX1 > TXNRD1. The data for AOX are by far superior to tPSA values that in the present study were not specific as a predictor.

Table 2.  Performance of tumour predictive AOX gene expression in CCC (CCC gene expression, threshold values: −95% CI)
GeneSensitivity, %Specificity, %PPV, %NPV, %Prevalence, %Accuracy, %ORP
TXNRD1487851763370 3.30.003

The relative effectiveness of the predictive CCC genes has been depicted in the form of ROC curves (Fig. 2). Areas under the ROC curves were ranked GPX1 > SOD2 > TXNRD1 ranging between 0.7 and 0.9. The expression of TXNRD1 correlated with tumour size and that of SOD2 with Gleason score (Table 3). There were no correlations for GPX1.

Figure 2.

ROC curves for the test performance of AOX gene expression in predicting primary prostate cancer.

Table 3. 
Spearman’s correlation by rank
Paired variablesSpearman’s RP
Tumour size (pT) + SOD2−0.120.64
Tumour size (pT) + TXNRD10.590.02
Tumour size (pT) + GPX1−0.060.83
Gleason score + SOD20.820.003
Gleason score + TXNRD10.330.36
Gleason + GPX10.250.49

The effect of RP on reducing of CCC was studied by molecular analysis of AOX RNAs in the patient cohort (n = 111) after RP (Fig. 1C,D). This analysis was carried out on at a mean of 16 months (95% CI 8.8–23.7) after RP. This period of tumour dormancy was chosen, to avoid unstable results after RP [6], as well as disequilibrium of CCC in bone marrow and blood, the first compartment probably nourishing blood with life. Consequently, additional data on adjuvant therapies could be collected within these 16 months. We collected data on AOX expression levels and divided them in to two groups depending on CCC clearance after RP. The first group C (n = 77) had a complete clearance of CCC AOX (Fig. 1C; P = 0.02 for TXNRD1 and P < 0.001 for GPX1 and SOD2 compared with values before RP) reaching the CSE level of patients with no tumours (group B vs C, not statistically significant). Only TXNRD1 remained slightly elevated. In this group, 21/77 patients (T1–3; N0, Gleason 4–10) had no adjuvant therapy suggesting a long-term effect of RP alone.

However, in 34 patients (31%, group D), the loss of CCC after RP was incomplete (group D vs C). Thus, group D had only a partial clearance of AOX-expressing CCC. The load of SOD2- and GPX1-expressing cells in group D was significantly higher than in group C. TXNRD1 expression was the same as in group C. The difference between group C and D after RP appeared to suggest varying risks for disease progression, group C representing a low progression risk and D representing a high risk. Hence, both groups were monitored for distant relapse recurrence.

The prognostic clinical impact of groups C and D after RP for distant metastases is shown as Kaplan–Meier diagrams using the − 95% CI CSE limits in group D as threshold values: 5.5 for GPX1; 2.2 for SOD2; 1.3 for TXNRD1 (Fig. 3). CCC genes (SOD2 and GPX1; Fig. 3A) were highly significant prognosticators for RFS when calculated as a whole (P = 0.003, log-rank test). GPX1 was the leading prognosticator (P < 0.001, Fig. 3B)). TXNRD1 (P = 0.67, Fig. 3C) was not a prognosticator or predictor gene for distant metastases. The tumour size has a limited effect on the probability of RFS in univariate analysis. In multivariate analysis tumour size, nodal status, and Gleason score had no effect. The difference in tPSA values in groups C and D also did not correlate with AOX expression, e.g. for GPX1 according to Spearman’s correlation by rank the P value in group C was 0.9 and in D 0.7. By contrast to tPSA the difference between both groups for GPX1 expression was highly significant (P = 0.004). The mean CSE values for group C were 2.9 (95% CI 1.5–4.3) and for D 8.9 (95% CI 3.5–14.1). More importantly, we approached the question of causes underlying the differences in AOX expression between the low-progression-risk group C and group D harbouring the risk for progression.

Figure 3.

Probability for distant tumour recurrence in AOX-expressing CCC. The −95% CI values of group D (Fig. 1) were chosen as threshold values.

Figures 1 and 3 are summary figures including adjuvant therapies according to the recommended guidelines for patients with pathological risk factors (N+, T3/4, Gleason >7 and R1) in both the low-progression and high-progression-risk groups C and D plus lower risk patients with no adjuvant therapies. Adjuvant therapy was initiated shortly after RP in patients with pathological risk, i.e. N+, R+, T3/4 and Gleason >7 in groups C and D, i.e. before the molecular analysis was conducted. Postponing adjuvant therapy in high-risk patients until the designated time of the molecular analysis would be considered unethical. Instead, we compared patients with lower risk pathological variables that had no adjuvant therapy in groups C and D and patients with higher pathological risk factors with adjuvant therapy also in these groups. In fact, we found that pathological factors were less important than endocrine adjuvant therapies, which play a decisive role in stimulating tumour progression. The results for further insight into risk determination by substratification of AOX expression after tumour size, nodal status, Gleason score, and margin status are set out in Fig. 4, although the case numbers become correspondingly small. Adjuvant endocrine therapy was used in 26/77 patients in group C and 26/34 patients in group D allowing the study of AR and GPX1 expression in more patients. Only seven patients in D did not receive endocrine therapy or radiation. In group C 21 patients had no preceding adjuvant therapy, neither radiation nor endocrine drug regimen. No records on therapy could be obtained in 30 patients in group C, in group D only one patient who underwent endocrine therapy and four who had radiation therapy were lacking relevant information. Chemotherapy was not used in the groups C and D.

Figure 4.

Comparison of GPX1 expression (CSE), the leading prognosticator gene (Fig. 3B), with pathological variables (N, T, Gleason score, R1) with and with no adjuvant endocrine therapy in patients with high progression risk (group D) and low progression risk (group C) from Fig. 3. A, median value and range of T, N and Gleason score in both groups with no adjuvant endocrine therapy. GPX1 expression (CSE) in the high-progression-risk group D is significantly elevated whereas the pathological variables do not differ significantly. B, corresponding values for GPX1 expression in the low-progression-risk (C) and high-progression-risk (D) patients corresponding to risk pathology variables: N+ T3/4, Gleason score >7 and R1. All patients had had previous adjuvant endocrine therapy according to the recommended guidelines. There was high GPX1 expression regardless of pathological variables in group D patients. The results prompted us to study AR expression in patients in groups C and D. C, CSE values of AR and GPX1 in both risk groups after adjuvant endocrine therapy. Patients with high progression risk (Group D) had a highly significant increase in AR and GPX1 expression.

Figure 4A shows the data for GPX1 expression, the lead gene for disease prognostication, in patients with no adjuvant endocrine therapy. Tumour size (range T1–3) and Gleason score (range 4–10) was matched for Group C (low progression risk) and D (high progression risk). The range for nodal status was 0 in C and 0–1 in group D. GPX1 expression remained significantly higher in the high-progression-risk group D.

Figure 4B shows correspondingly the GPX1 expression values after adjuvant endocrine therapy in the low-progression-risk group C vs high-progression-risk group D. The patients groups were matched for high-risk pathological variables N+ (N1 in C; N1–2 in D), T3/4, R1 and Gleason score >7. GPX1 expression remained higher in the high-progression-risk group D as compared with group, C reaching a highly significant difference except for N+.

Figure 4C shows the effect of adjuvant endocrine therapy in groups D and C. The high-progression-risk group D had selective overexpression of AR and GPX1, indicating drug resistance. In contrast, in group C the expression of both genes was normal, suggesting the absence of therapy resistance. The differences for AR and GPX1 between both groups differed significantly. These findings may suggest that positive lymph nodes (N+) may have some influence on GPX1 overexpression in group D with no adjuvant therapy. The other pathological markers appeared to have no influence (Fig. 4A). When a hormone-ablation regimen is used in patients with risk pathological markers, failures in the form of AR plus GPX1 overexpression appear to be prerequisites for tumour progression.

Patients with low GPX1 expression do not necessarily need adjuvant therapy, whereas those with high expression need adjuvant treatment, thus measurement of GPX1 expression should identify patients requiring therapy that would not be evident from tPSA levels alone. The type of investigation presented here cannot be regarded as an epidemiological study, as the authors actively collected patients attributable to inclusion in groups C and D. The emerging results show that tumour progression can occur in any patient, i.e. with a fraction of high risk as well of low risk pathological variables. The technology used in the present study helps to identify patients with high AOX expression, which allows for additional adjuvant treatment in the state of tumour dormancy for preventing tumour recurrence. Adjuvant radiation had no effect on SOD2 and GPX1 expression in either group suggesting trafficking of CCC in the blood stream independent from the radiated tissues.

In view of the diagnostic and prognostic capabilities of the AOX testing in CCC two further questions were addressed: the malignant nature of CCC and the relation with the tissue of origin. The first question of the malignant nature of these cells was addressed by identifying genomic imbalances in a selection of protooncogene and tumour suppressor loci (Table 4). The appearance of genomic imbalances proves the cancerous nature of the cells exhibiting a tendency of genomic imbalances accumulation in metastatic disease. The increase was mainly from a loss of heterozyosity in the loci APC and D7 S523, albeit not significantly. Secondly, AOX testing in CCC vs MNC was designed as a general marker applicable for solid tumours giving therefore no information about the location of the malignant tissue. Thus, the cellular distribution of tissue-specific and epithelial-marker genes was investigated in CCC and MNC. The distribution of AR, PSA, PSMA, CK20, and CD45 mRNAs in CCC and MNC is given in Table 5. None of the markers accumulated in CCC as compared with MNC. CCC had only minute amounts of CD45-expressing blood cells contaminating the cancer cell fraction. For relating the molecular findings in CCC to the prostate in the clinical setting, we therefore used the appearance of prostate markers in MNC. In the present study at least one organ-typical positive signal in MNC was found in 82% of the AOX tests: AR 63%, PSA 7%, PSMA 51%. After RP a positive signal from organ markers was missed in 9.5% in Group C (Fig. 1) and 12% in Group D patients. Thus, these organ markers are helpful to confine the AOX data to the tumour tissue predicted or prognosticated.

Table 4. 
Distribution of genomic imbalances from nine different loci in CCC
Genomic imbalancesNo relapse
(n = 39), %
(n = 40), %
≥2 8200.1
Table 5.  Cellular distribution of epithelial gene expression
Cell fractionGene expression, 95% CI (% of patients)
CCC (n = 20)MNC (n = 40)
  • *

    P < 0.05 vs MNC.

AR   2–32* 60–85
PSA   0–15  3–21
PSMA   4–46* 39–67
CK20   0–15* 30–48
CD 45 (% of MNC)0.05–0.14100


PSA is the landmark protein for the diagnosis, staging and follow-up of prostate cancer. However, the indiscriminate use of PSA testing in elderly men, followed by biopsies, may result in substantial over diagnosis and over treatment [3]. Lowering the threshold values for biopsy, currently 4 ng/mL in most recommendations, will increase the medium time from a tPSA level rise to death from prostate cancer. Also, the cause of death may be erroneously ascribed to prostate cancer. In the context of the ongoing discussions about the adequate use of PSA testing the conceptual idea underlying this work was to look at the disease from a systemic and cellular point of view, i.e. to establish companion diagnostics on the basis of molecular analysis of CCC before and after RP. As for other solid tumours, the migration of cells away from the primary tumour and entry into the systemic vasculature or lymphatics has been also hypothesized for prostate cancer [18]. Hence, studying CCC could be a valuable tool to identify patients with tumour at tPSA levels of 4–10 ng/mL when the marker exerts high sensitivity, low specificity and to stratify patients at risk for disease progression after RP for AOX expression, considering pathological risk factors, serum tPSA and adjuvant endocrine therapy.

The clinical utility of using epithelial prostate-specific marker RNAs, PSA and PSMA, is unclear as shown by others applying a similar clinical validation method [6,19]. Quantitative RT-PCR of PSA in MNC transformed into LNCaP cell equivalents as reported by others [20] may be more useful although predictive data are still missing. LNCaP equivalent is an artificial dimension for CCC because neither the number of CCC nor the normalized expression level can be exactly given. Also, pretest fractionation of CEC from blood specimens has been reported to be inconclusive for disease prognostication [21]. This is in contrast to the present findings of AOX overexpression in CCC. Pretest fractionation of CCC and comparison with MNC is a prerequisite for tumour prediction and disease prognostication. As discussed below the number of cancer cells is not decisive in contrast to the pheno- and genotype. In view of the vagaries associated with prostate epithelial-markers studies, we speculated that other forms of CCC might be more conclusive. We present for the first time in the present investigation a novel form of CCC enriched from patients with prostate cancer by a filtration assay (large cells of a diameter ≥20 µm are retained in the filter mesh whereas small CEC pass through). Large cells appear preferentially in the form of cell clusters as shown by our laboratory to date only for patients with breast cancer [12]. These cells harbour, similar to prostate (Table 5), much less organ-related and epithelial markers than detected in the MNC fraction suggesting a lower degree of epithelial differentiation. The malignant nature of these cells in patients with prostate cancer is shown by the presence of genomic aberrations (Table 4). We present the clinical utility of a novel screening test system for CCC based on real-time RT-PCR of the AOX genes SOD2, GPX1, TXNRD1. The AOX were chosen after comparing >60 candidate genes between a microarray platform and the TaqMan technology. The quantitative AOX data are expressed as CSE calculated from the transformation of TaqMan cycle threshold values into cell equivalents of the corresponding reference cell line as indicated vs GAPDH as the house-keeping and cell-counting gene in filtration enriched tumour cells and in patients’ MNC. The resulting ratio is termed CSE. CSE is different from cell equivalents as suggested by Straub et al.[20] for PSA-RNA. Target gene expression/GAPDH in the cancer cell fraction gives the exact expression value per cancer cell. CSE gives the expression values per cancer cell in comparison with benign MNC allowing the cellular localization of disease-related molecular events. GAPDH was used as the molecular cell counter in both cell fractions. The dynamics of CSE allow the prediction of tumours and the prognosis of the disease course. The prediction of clinical tumours as expressed by CSE values was not based on cancer cell numbers but a function of a substantial increase in AOX genes expression per cancer cell. For example, there was a 37-fold overexpression of GPX1 in enriched cancer cells in disease that did not progress and a 101-fold overexpression in patients with bone metastases (P = 0.008). The normalized expression of GPX1/GAPDH in MNC remained nearly constant with 14-times and 15-times overexpression, respectively.

The advantage emerging from this analytical system is the ability to study any DNA and RNA in cancer cells in a personalized manner or in clinically defined cohorts (Table 1). At the same time, systemic effects can be studied in patient’s MNC. Thus, genotype and phenotype of circulating tumour cells rather than cell numbers becomes the basis for prediction and prognostication. This appears to be in general accordance with the molecular concept of tumorigenesis. In more detail, we studied the clinical utility of the test system for predicting the primary prostate tumour, the effect of RP and prognostication after RP of RFS.

Each of the three genes was identified as a highly significant predictor of prostate cancer (Table 2). The best results were obtained for all three genes (sensitivity 86%, specificity 82%, PPV 69%, NPV 92%, accuracy 83%, OR 28). Before RP dissemination of CCC was significantly correlated with the pathological tumour status, namely SOD2 expression with nodal status and TXNRD1 expression with Gleason score (Table 3). Only GPX1 expression was not correlated. Most of the tumours detected were organ-confined (T2 83%, T3 and T4 17%; N0; M0). The functional analysis of the test systems as ROC curves resulted in areas under the ROC curves values of 0.7–0.9 (Fig. 2).

RP is accepted as the decisive clinical measure to prevent disease progression. Therefore, the beneficial effect must be expected to clear CCC from the blood. However, PSA-mRNA and PSMA-mRNA, has been shown in many studies to increase after RP [19] thus failing to attribute any clinical utility [20]. AOX testing of CCC would become another useless marker if RP had no effect. This is not the case. RP leads to a complete clearance of CCC to the level of patients with no tumour (Fig. 1, group C). Only TXNRD1 expression is slightly higher than in patients with no tumour. A few patients have incomplete clearance. We estimate that the group of incomplete clearance of CCC through RP is ≈15% of patients, which we do not think reflects a failure of the operative techniques. This group (group D) showed a partial clearance only for GPX1- and SOD2-expressing cells, i.e. CSE values were highly significantly elevated in comparison with the complete clearance in group C, suggesting differences in the disease course after RP. TXNRD1 expression was similar in groups C and D. The pathological variables in group D indicated a higher risk profile with respect to the nodal status and Gleason score, whereas the tPSA did not differ significantly from group C.

Both groups were studied for the likelihood for distant metastases. Using the − 95% CI CSE level of group D as the threshold values, high expression values of SOD2 and GPX1 increase the risk for distant metastases. The corresponding Kaplan–Meier diagram is shown in Fig. 3. The leading prognosticator was GPX1. TXNRD1 was not a prognosticator gene. Univariate analysis using the Cox proportional hazards model showed that the tumour size could be ascribed to RFS (P = 0.05), in multivariate analysis neither nodal status (pN), tumour size (pT) nor Gleason score remained as significant variables for survival.

The Kaplan–Meier diagram for AOX expression level was hampered by neglecting pathological risk factors, i.e. N+, R+, Gleason score >7, R1, with progression or not with and without adjuvant therapy. In addition, a period of 16 months (95% C.I. 8.8–23.7) was allowed before the molecular analysis after RP. On one hand, AOX values were stable then, allowing evaluation of RP alone, but on the other hand, adjuvant therapy might have been used. Analysis of GPX1, the lead prognosticator gene, was studied in patients with and with no adjuvant endocrine therapy in both the after RP groups, namely in group C as low GPX1 expressors and group D as high expressors prone to disease progression. As shown in Fig. 4A the absence of adjuvant therapy in patients with N = 0–1; T = 1–3 and Gleason score 4–10 did not affect the expression of the prognosticator gene GPX1 and, consequently, the assignment to no progression risk group C and to progression risk group D, respectively. Adjuvant endocrine therapy according to the recommended guidelines was used in most of the patients with risk factors, i.e. N = 1–2, T = 3–4, Gleason score >7 and R1 (Fig. 4B). Even then, GPX1 expression allocated to high-progression-risk group D was highly significantly elevated compared with group C. Only for N = 1–2 the P value of 0.5 did not reach statistical significance. In view of the adjuvant endocrine therapy used for patients with pathological risk factors, monitoring drug efficacy became obvious. Drug resistance is a major issue in hormone-ablation regimens. A couple of molecular mechanisms have been described that are associated with reduced efficacy of hormone-ablation strategies and disease progression, among them altered expression of the AR [22,23]. As shown in Fig. 4C, both patient groups with pathological risk factors had had previous endocrine therapy, but only in the GPX1-overexpressing high-progression-risk group D was the expression of the AR also overexpressed (P = 0004; median 9.05; range 0.17–55) as compared with group C (median 0.65; range 0–2.7). The data seemingly indicate that drug failure may be involved in ascribing adjuvant endocrine therapy to progression risk albeit not indicated by tPSA (Table 1).

Finally, AOX mRNA testing is a novel screening test for CCC. AOX overexpression in CCC can be described as a survival and defence mechanism required in an atypical environment. It is generally accepted that tumour cells produce increased reactive oxygen species (ROS) including superoxide, hydroxyl radical and H2O2. Comprehensive reviews have been published [24,25]. Most malignant cells are highly glycolytic resulting in high levels of ROS. Among the intracellular sources of ROS in cancer cells NADPH oxidase, glycerophosphate dehydrogenase, thymidine phosphorylase, DNA mutations in specific loci have been described [26]. Among blood cells, neutrophils and macrophages may contribute to ROS production in blood. Severe oxidative stress leads to apoptosis. Persistent oxidative stress at sublethal levels may also cause adaptive responses through up-regulating anti-ROS defence strategies. In prostate cancer, the AOX described in the present study have been reported as scavengers of ROS. The thioredoxin system composed of thioredoxin and TXNRD1 seems to act as a major scavenger candidate [27] associated with cell proliferation in the state of androgen independency [28]. SOD2 may have the opposite effect with possible tumour suppressor activity [29] expressed rather in benign epithelium than in prostatic intraepithelial neoplasia and cancer [30], whereas other authors describe the AOX prone to therapy resistance [31]. It should be noted that many studies have been done in prostate cancer cell lines. The clinical impact remains to be elucidated. There has been comparatively little attention to the expression of GPX1 in prostate cancer. With respect to the present findings, it is interesting to note that the AOX has been reported to protect from CD95-induced apoptosis suggesting that GPX1 expressing CCC may escape from immune surveillance [32]. This may explain why GPX1 becomes the leading marker gene for disease progression when surgical removal of the prostate ceases refilling the blood-borne cancer cell pool. GPX1 overexpression seems to be required for survival of CCC in blood. Ranking the predictive potency of AOX during the disease course, SOD2 and TXNRD1 are better predictors in before RP [30], whereas GPX1 and to a lesser extent SOD2 remain for disease prognostication and prediction of bone metastases after RP. Prediction of prostate cancer has been reported by various test systems in urine, e.g. prostate cancer gene 3 [33,34], annexin A3 [35], sarcosine [36] and hypermethylation of prostate-derived gene loci [37]. Most of these systems require preceding massage of the gland. Exfoliation of cancer cells into blood circulation has not been studied [6]. Also, no data on the prediction of local recurrence are available. Enumeration of epithelial cells in blood has been reported as inappropriate to predict prostate cancer [38]. However, this system is more appropriate for disease prognostication [39,40]. The obvious advantage of the AOX test is based on the ‘on-line’, noninvasive observation of CCC along the entire disease course, i.e. tumour prediction, disease prognostication and therapy monitoring including surgery and systemic drug regimens. AOX overexpression in CCC may indicate novel targets for cancer prevention therapies.

In conclusion, molecular staging in prostate cancer patients before and after clinical diagnosis is reported for a novel class of CCC enriched by filtration. Quantitative real-time RT-PCR of three AOX predicted prostate cancer in patients with serum tPSA levels of 4–10 ng/mL. If complete clearance of CCC occurred after RP then the prognosis was favourable. However, if clearance of CCC was incomplete the risk for bone metastases was significantly increased. The test system could be used as an additional tool for therapy monitoring, drug-resistance detection (here adjuvant endocrine therapy), the appearance of metastogenic factors and early detection of relapse tumours. The data suggest the potential usefulness for targeted prevention strategies.


This study was supported by North Rhine Westphalia, Germany, grant No. TJ – 9804v02.


Michael Giesing is inventor and owner of three patents underlying this work, US7,056,660, US7,232,653, EP03792393.5. Gerhard Driesal is co-inventor in US7,056,660. Bernhard Suchy is co-inventor in US7,056,660 and EP03792393.5. The patents stem from the grant of NRW, Germany, No. PTJ-9804102.