Molecular phenotyping of circulating tumour cells in patients with prostate cancer: prediction of distant metastases

Authors


Michael Giesing, Onkokonsult, Zum Herzfeld 5, D-49536 Lienen, Germany. e-mail: Giesing@onkokonsult.com

Abstract

What's known on the subject? and What does the study add?

The role of circulating cancer cells in metastogenesis is generally accepted. Two forms of these cells have been reported in a number of studies, cancer cell clusters (CCCs) and individual epithelial cancer cells. Clusters appear at higher frequencies in the blood. CCCs have been reported to be rich in vimentin and poor in E-cadherin expression The resulting epithelial to mesenchymal transition, a prerequisite for metastasis formation, occurs in CCCs. We have developed a new set of biomarkers, namely the antioxidant genes GPX1, SOD2 and TXNRD1, specific to cell trafficking in the blood.

Firstly, the study shows that diagnosis of distant metastases is feasible by applying molecular phenotyping with a five gene test that has 94% sensitivity and 81% accuracy. Again SOD2 and GPX1 showed the highest sensitivities. Secondly, the study shows the efficacy of palliative chemotherapy in clearing the blood of CCCs overexpressing diagnostic genes. Clinically the overall lifespan ranged from 5 to 99 months under taxotere. We aimed to investigate the molecular reasons and found that MDR1 overexpression worsened survival by 31%. To the best of our knowledge, this is the first study to show the clinical impact of drug targeting and the counter-effect of drug resistance in CCCs on overall survival. The findings may, therefore, add a novel tool for clinicians in tailoring therapies individually.

OBJECTIVES

  • • To find the molecular phenotype in circulating cancer cells from patients with prostate cancer (PCa) in order to predict distant metastases.
  • • To determine genes affecting the study endpoints of overall survival and time to progression.

PATIENTS AND METHODS

  • • Twenty-five urologists in several clinics participated in the study, with 51 patients with metastatic and 77 with non-metastatic PCa.
  • • Molecular analysis was carried out in two forms of circulating cancer cells, cancer cell clusters (CCCs) and individual epithelial cancer cells (CECs).
  • • Gene expression was studied using real-time reverse-transcriptase PCR.
  • • Cycle threshold values were normalized with glyceraldehyde 3-phosphate dehydrogenase in cancer cells and mononucleated cells, yielding comparative specific expression values from the relative quantification method with the help of the standard curve method for each patient and each gene locus.

RESULTS

  • • Preclinical validation was performed using aggregated and non-aggregated SW480 cells showing the independence of CCCs and CECs.
  • • Prediction of metastases was achieved with five genes showing the highest sensitivity, SOD2, GPX1, AR, cyclin B and bFGF.
  • • The following results were obtained: 94% sensitivity, 65% specificity, 76% positive predictive value and 89% negative predictive value. The prevalence was 63%. Test accuracy was 81% with an odds ratio of 32 (P < 0.001).
  • • Overall survival was worsened by preceding chemotherapies when leaving insufficient GPX1 clearance in blood.
  • • Drug resistance genes were found to worsen the endpoints, among them MDR1 (P = 0.003; hazard ratio: 1.31; 95% CI: 1.09–1.58).

CONCLUSIONS

  • • SOD2, GPX1 and AR represent a novel biomarker set for circulating cancer cells (clusters and scattered individual cells) in PCa.
  • • The clinical usefulness of these biomarkers ranges from the prediction of clinical tumours to disease prognostication, therapy monitoring and therapy outcome prediction (hormonal therapies, chemotherapies).
  • • The presence of CCCs and CECs after batch isolation allows the addition of genes for intensive studies, e.g. drug resistance.
Abbreviations
PCa

prostate cancer

CCC

cancer cell cluster

CEC

epithelial cancer cell

MNC

mononucleated cell

AR

androgen receptor

Ct

cycle threshold

GAPDH

glyceraldehyde-3-phosphate dehydrogenease

CSE

comparative specific gene expression

OS

overall survival

PPV

positive predictive value

NPV

negative predictive value

OR

odds ratio

TTP

time to progression

ROC

receiver-operating characteristic

HR

hazard ratio.

INTRODUCTION

Prostate tumours possess a special propensity to spread to bone. As a consequence, bone metastases are highly prevalent in patients with prostate cancer (PCa). Whereas bone metastases are infrequent in newly diagnosed PCa, very high levels of frequency (>65%) have been reported in hormone-refractory PCa accompanied by a high mortality rate [1,2]. The formation of bone metastases requires trafficking of malignant cells via the circulatory or lymphatic system in a multistep process [3]. Two types of such cells exist, namely small, scattered epithelial cells and large cells appearing mostly in clusters [4]. Circulating epithelial cancer cells (CECs) have been reported to be associated with bone metastases, although predictive data remain unknown [5]. The test used for CECs is of limited use owing to its low sensitivity (49%); therefore, we attempted to find a novel CEC marker with higher sensitivity, resulting in quantitative reverse-transcriptase (RT)-PCR of the androgen receptor (AR) [4]. In addition, we have recently isolated circulating cancer cell clusters (CCCs) and large tumour cells (≥20 µm) in an antigen-free manner from patients with PCa. These cells contain fewer epithelial markers than EPCAM-positive CECs. The malignant nature of CCCs has been demonstrated by DNA aberrations [4]. By contrast to CEC enumeration [6], quantification of the expression of the antioxidant genes GPX1, SOD2 and TXNRD1 in CCCs yielded high predictive values for the detection of primary PCa [4]. For the prediction of bone metastases in the present study, we studied the oncogenic phenotype of CCCs and CECs using qRT-PCR in 20 functional, not structural gene loci emerging from previous microarray experiments. For this purpose, the sensitivity for overexpression in candidate loci has been determined in CCCs and CECs obtained from metastasized patients with PCa. Loci exhibiting the highest sensitivity or with association to androgen deprivation failure were selected for predicting bone metastases. As a result a five-gene test, four in CCCs and one in CECs, served as the basic tool. The findings may have impact on therapy strategies in the future.

MATERIALS AND METHODS

Preclinical validation has been achieved with two types of spiked disease cells: clusters and individual cells. Cancer cell fractions were isolated from 7.5 mL blood, and molecular analysis was carried out in two populations of circulating disease cells, namely EPCAM+ CECs and CCCs. Both classes of circulating tumour cells have been enriched from mononucleated cells (MNCs) as described recently [4,7,8]. As published by our laboratory in 2000 [8], spiking experiments in blood with aggregates or individual, dispersed EPCAM+ epithelial cells from SW480 cultures have shown that the two cell types to be isolated represent two different classes of circulating disease cells. Aggegates were harvested from specific culture conditions in silanized glass tubes. Each cluster harboured a mean of 7 cells, 30–70 µm in size. Individual scattered cells were collected after trypsinisation of cultures. They were 16.5 µm in diameter. The process used is shown in Fig. 1. The appearance or absence of SW480 cells Ki-ras in Codon 12 (GGT [Gly]→ GTT [Val]) was chosen to be detected by restriction-fragment-length-polymorphism as a measure for CCCs (Fig. 2). The methods were published by our laboratory in detail in 2000 [8]. Briefly, clusters were captured in the filtrate matrix whereas individual cells would pass through the matrix and appear in the filtrate; therefore, scattered non-aggregated SW 480 cells were applied to determine their presence in CCCs (Table 1). The appearance of wild-type Ki-ras served as a purity measure in CCCs (Fig. 2) and dispersed individual cells were also measured in the CCCs (Table 1).

Figure 1.

Steps used in CCC preparation (Ki-ras codon 12).

Figure 2.

Wild-type Ki-ras wt and CD 45 RNA in CCCs from spiking experiments with constructs of SW480 aggregates.

Table 1. Capture of Ki-ras mutation in CCCs from scattered SW480 cells
No. of SW480 cellsKi-ras mutation in CCCs
  1. n.d., not detectable.

1500n.d.
300n.d.
150n.d.
30n.d.
0n.d.

Twenty-five urologists in several institutions participated in the open investigation. Physicians were asked to fill out anamnesis forms. A total of 128 patients were enrolled for the study after giving written informed consent (Table 2). In all, 51 patients had distant relapse tumours, of whom 46 had bone metastases (M1b) and five had other distant organ tumours, comfirmed by imaging techniques. Patients at the metastatic stage were compared with patients without tumour relapse (n = 77). The risk profile of pathological variables was higher in the metastatic group. Cell fractions from patients with metastatic disease were molecularly phenotyped, with and without chemotherapy. Overall survival (OS) data were collected in a prospective fashion. Observation was continued after clinical metastasis diagnosis and chemotherapy for a median (range) time of 21.5 (2–44) months.

Table 2. Demographic data of patients included in the study
 Median (range)
Group A: no tumour relapse, n = 77
 Age, years64 (49–81)
 Follow-up, months31 (1–216)
 Gleason score7 (2–10)
 Tumour stagepT2 (1–4)
 Node stagepN0 (0–1)
Group B: distant relapse tumours, n = 51 (bone metastases n = 46; other distant organs n = 5)
 Age, years65 (45–79)
 Follow-up, months14 (1–44)
 Gleason score8 (6–10)
 Tumour stagepT3 (1–4)
 Node stagepN1 (0–3)

At each time point, circulating disease cells were collected. In brief, peripheral blood was collected in heparinized vacutainers (Becton Dickinson, Franklin Lakes, NJ, USA). MNCs were purified over a density gradient using Nycoprep 1.077 (Nycomed, Oslo, Norway). MNCs were washed twice in PBS (0.2% BSA; Life Technologies, Germany) and resuspended finally in 10 mL PBS. CCCs were enriched by size using a column containing a polyester mesh (width of 20 µm) [8]. Epithelial cells were enriched using anti EpCAM-conjugated immunomagnetic beads [4]. Enriched cells were lysed with Trizol and stored at −80 °C until molecular phenotyping was carried out. This technology removes genomic DNA, thus avoiding interference with the RNA test systems. Total RNA was prepared from the aqueous phase which is free from genomic DNA remaining in the organic phase. Reverse transcription was then carried out on the collected RNA, as previously described [4]. Molecular phenotyping was carried out using real-time PCR. The efficiency of cDNA synthesis was monitored using real-time quantitative RT-PCR for glyceraldehyde-3-phosphate dehydrogenease (GAPDH) using K562 as a reference cell line. For target genes, cell lines as indicated were applied to obtain a normalized cDNA standard curve [7]. Target gene primers and probes designed by using the Primer Express software 1.0 (Perkin-Elmer, Norwalk, CT, USA) were purchased from TIB Molbiol (Berlin Germany). Primer designs spanned two exons without intron sequences, thereby avoiding DNA contamination.

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 Applied Biosystems, Foster City, CA, USA). Normalized reference expression values were calculated as cell equivalents related to target-gene-enriched reference cell lines and MNCs after cell counting in a Neubauer chamber. Details of cell line preparation and of the PCR reaction conditions have been given elsewhere [4,7]. Interpretation of the cycle threshold (Ct) values was achieved by the relative quantification method using the standard curve method [9]. This was a better method than the ΔΔ Ct calculation, taking into consideration that the exact doubling of cDNA or RNA with each PCR cycle is not always met in scientific studies, which possibly leads to erroneous results. In fact, doubling varies between 1.5 and 2.5, therefore, calibrator curves were established for each gene using appropriate cell cultures as indicated. Standard curves were run for each gene and each patient. Target and endogenous amplifications were carried out in separate tubes. The target values were normalized with the housekeeping gene GAPDH. GAPDH expression also served as a cell counter using K562 cells as a calibrator. Because cancer cell fraction obtained from blood can be efficiently separated from a patient's benign cells, i.e. MNCs or CD45+ cells, these were introduced as a second internal control. Thus, each biological effect could be monitored in tumour cells as well as in the patient's benign blood cells. GAPDH-normalized reference values (cell equivalents target/cell equivalents GAPDH) obtained are expressed as comparative specific gene expression (CSE), calculated from the expression values in enriched circulating prostate cells compared with patients' MNCs.

CEC Marker-RNA: The CSE values for the AR (Ref.Seq. NM 000044; Location: Xq12) were studied and calculated. Details of the PCR-based test design have been published by our laboratory [4]. The lower detection limit was 2–3 cells.

CCC Marker-RNAs: AOX testing served as the basic marker set for CCCs, as previously published [4]. Test designs and test systems for the marker AOX genes have been described in detail. The expression of SOD2 (NM 000636; 6q25), TXNRD1 (NM_003330; 12q23-q24.1) and GPX1 (NM 201397; 3p21.3) was given as CSE values using reference cell lines as indicated. CSE values of bFGF2 and cyclin B1 served as additional marker-RNAs.

bFGF2 (basic), (NM_002006.4) forward: 5′-GAAGCGGCTGTACTGCAAAA-3′; reverse: 5′-CCTTTGATAGACACAACTCCTCTCTC-3′; TaqMan probe: 5′- (TGAAGTTGTAGCTTGATGTGAGGGTCGC-3′. (ref. cell line K562) cyclin B1 (CCNB1) (NM_031966): forward primer 5′-TTG AGG AAG AGC AAG CAG TCA G-3′, reverse primer 5′-TGA CAC CAA CCA GCT GCA G-3′, TaqMan probe: 5′- GGC ACA CAA TTA TTC TGC ATG AAC CG-3′. (ref. cell line K562) CCC and CEC were controlled for contamination by blood cells using real-time RT-PCR of PTPRC (CD 45) gene expression (Y00062; 1q31-q32). The maximum threshold values have been given elsewhere [4].

Determination of the oncogenic phenotype was achieved in CCCs and CECs from the blood of patients with metastatic disease. Gene loci selected from preceding microarray experiments were reanalysed by real-time PCR. The technical principles – sense primer and probe – are listed in Table 3. Expression values in CCCs and CECs and sensitivity from patients with metastatic PCa are shown in Table 4.

Table 3. Molecular test systems
Gene lociAccession NoSense-primerProbeReference cell line
ARNC_0000235′- CCATCTTGTCGTCTTCGGAAA-3′5′- ATGACTCTGGGAGCCCGGAAGCT GAA-3′MCF7
bcl-2NM_0006335′- GAGCTCTTCAGGGACGGG-3′5′- ACTCAGTCATCCACAGGGCGATGTTGT-3′ES-2
bFGF (FGF2, FGFB),NM_0020065′-GAAGCGGCTGTACTGCAAAA -3′5′- TGAAGTTGTAGCTTGATGTGAGGGTCGC- 3′K562
COX2 (COII, MTCO2)NC_0129205′- TGGAACATGGAATTACCCAGTT-3′5′- CACCAGCAACCCTGCCAGCA-3′ES-2
Cyclin B1 (CCNB1,CCNB)NM_0319665′-TTg Agg AAg AgC AAg CAg TCA g-3′5′- GGC ACA CAA TTA TTC TGC ATG AAC CG-3K562
DCK (MGC117410)NM_0007885′- CCC GCA TCA AGA AAA TCT CC -3′5′- AAG GGA ACA TCG CTG CAG GGA AGT CT-3′Colo320
DPD DYPD, DHP)NM_0001105′- GAGCTCGCCGAGTGTTCATC-3′5′- TGTCCCTGAGGAGATGGAGCTTGCTAA-3′ES-2
ER (ESR, ESR1)NM_0010402755′- GCCCACAATACAGCTTCGAGT-3′5′- AGTAGTTGTGCTGCCCTTCCATTGCC-3′ES-2
erb B2 (TKR1, Her2)NM_0010058625′- TGAGACTGATGGCTACGTTGC -3′5′- AGCCCCCAGCCTGAATATGTGAACC -3′EFM 192
ERCC1 (UV20, COFS4)NM_0019835′- CCAGCAAGGAAGAAATTTGTGAT-3′5′- AATAAGGGCTTGGCCACTCCAGGAGGGA-3′EFM192
GCS (GCLC, GCL)NC_0014985′- TGG GAT TTG GAA TGG GCA -3′5′- TTG CTG TCT CCA GGT GAC ATT CCA AGC T-3′H82
GPX1NM_0005815′- CTCGGCTTCCCGTGCAA -3′5′- AGTTTGGGCATCAGGAGAACGCCAAGAA -3′BT474
HSP 27 (MKBP, HSPB2)NM_0015405′- GCGTGTCCCTGGATGTCAAC -3′5′- TCGTGCTTGCCGGTGATCTCCACC -3′K562
MDR1 (ABCB1, CLCS)NM_0009275′- CAGCGCTTCGCTCTCTTTG -3′5′- AAGGAGCGCGAGGTCGGGATGGAT -3′MES
MGMTNM_0024125′-CTGATGCAGTGCACAGCCTG- 3′5′-TGTCTGGTGAACGACTCTTGCTGGAAAAC- 3′MCF7
RAF-1 (CRAF, RAF1)NM_0028805′- CACCAGTATCTGGGACCCA -3′5′- AAAAACAAAATTAGGCCTCGTGGACA -3′MCF7
SOD2 (MNSOD)NM_0006365′- GTCACCGAGGAGAAGTACCAGG -3′5′- CGTTGGCCAAGGGAGATGTTACAGCCC -3′EFM192
Topo IIA (TOP2, TP2A)NM_0010675′- GGTGTGGTTGGGAGAGACAA-3′5′- AGCTTCTCATAAGCAGATCATGGAAAATGC-3′MES
TP (TYMP, ECGF)NC_0000225′- GTCAGCCTGGTCCTCGCA-3′5′- ATGTGGCTGCAAGGTGCCAATGATCA-3′ES-2
TSNM_0010715′- GTCTGGGTTCTCGCTGAAGCT -3′5′- TCACGGGCCTGAAGCCAGGTGACTTTA -3′ZL 697
TXNRD1 (TR, TXR1)NM_0033305′- GGAGGGCAGACTTCAAAAGCTAC -3′5′- TTGGGCTGCCTCCTTAGCAGCTGCCA -3′MES
UP (UPASE, UPP)NC_0000075′- CCACTAGCAGACACAATTTCCCA-3′5′- TTGGAGATGTGAAGTTTGTGTGTGTTGGT-3′ES-2
VEGF (VEGFA, VPF)NC_0000065′- CCACTAGCAGACACAATTTCCCA-3′5′- TTGGAGATGTGAAGTTTGTGTGTGTTGGT-3′ES-2
Table 4. Expression values of selected gene loci in CCCs and CECs from patients with metastatic PCa
Gene lociMean CSE value−95% CI, Min.*+95% CI, Max.*Sensitivty, %
erb B2 (TKR1, Her2)14.72.832.18
bcl-21.951.312.5810
GCS (GCLC, GCL)2.671.35.617
RAF-1 (CRAF, RAF1)1.501.31.725
ERCC1 (UV20, COFS4)1.601.41.825
MDR1 (ABCB1, CLCS)8.530.6317.7030
Topo IIA (TOP2, TP2A)3.350.576.1232
DCK (MGC117410)3.21.45.040
ER (ESR, ESR1)3.561.465.6546
TS4.801.38.348
TXNRD1 (TR, TXR1)2.921.853.9858
COX2 (COII, MTCO2)9.354.414.362
VEGF (VEGFA, VPF)4.361.66.266
HSP 27 (MKBP, HSPB2)3.63.44.166
bFGF (FGF2, FGFB),37.631.8473.4269
MGMT2.480.634.3271
Cyclin B1 (CCNB1,CCNB)6.852.211.572
DPD DYPD, DHP)3.760.0037.5372
UP (UPASE, UPP)2.901.624.1772
TP (TYMP, ECGF)3.002.003.9972
SOD2 (MNSOD)3.782.555.0079
AR29.204.0119.691
GPX16.815.488.1397

STATISTICAL ANALYSIS

All data were entered into Microsoft Office Excel spreadsheets. Statistical analyses were performed using Statistica 9 software (StatSoft, Germany). Molecular tests were compared between patient groups using the t-test for independent variables. Sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), accuracy, odds ratio (OR) and prevalence were calculated using the 2 × 2 contingency table method applying the chi-squared test for statistical evaluation.

Receiver-operating characteristic (ROC) analysis and areas under the curve were used as objective measures to evaluate the lead biomarkers. Estimation of independent variables influencing the binary variable metastasis was achieved using the logistic regression Quasi-Newton calculation. Residual analysis was performed to create predicted and residual values for all observations and for calculation of the OR. The percentage of correct classification of cases was calculated for metastases and non-metastases. Spearman's correlation by rank was calculated for different pair variables as indicated. Thus the correlation of predictive molecular tests and pathological variables with each other and with metastases could be elucidated. For analysis of OS and of time to progression (TTP), Kaplan–Meier curves were used. Patients without relapse tumours were censored. Differences between groups were calculated using the log-rank test. Genes affecting survival were investigated using the Cox proportional hazard method.

RESULTS

Preclinical validation was published by our group in 2000 in detail [8]. We used the blood of healthy volunteers for spiking with SW480 cells harbouring a Ki-ras mutation in codon 12. As shown in Fig. 1, the cells were applied either in aggregate or in non-aggregate form; the latter serving as a model for CECs. Then CCCs were isolated by filtration of the MNC fraction and the integration or the absence of the mutation was measured in cells retained on the filter matrix. Wild-type Ki-ras were continuously removed from CCCs with an increasing number of spiked aggregates (Fig. 2). The best results were obtained when the cells retained on the filter were collected by a reflux design with PBS. Also contamination of CCC with CD45+-expressing cells was the lowest in the reflux system. When blood was spiked with SW480 without aggregates, no mutated cells could be observed in the CCC (Table 1). The data confirm that CCCs can be differentiated from CECs.

Demographic data for the 128 patients enrolled in the study are listed in Table 2. Patients with metastatic PCa had a higher Gleason score, tumour size and pelvic lymph node status (Group B). ‘Risk patients’ were also observed in the non-metastatic group, albeit with a lower incidence (Group A). Of the 51 patients with metastatic PCa in group B, 18 were chemotherapy-naïve at the time of molecular analysis and 41 patients had received adjuvant endocrine therapy. Of the patients in the non-metastatic group A, 26 were pretreated with endocrine drug regimen, 20 were not pretreated, and no information was available in 31 patients. The mean serum PSA concentration in the metastatic group was 31.0 ng/mL (± 95% CI 15.7–46.4) and in the non-metastatic group 0.6 ng/mL (± 95% CI 0.1; 36.7). The best choice for oncogenic molecular markers was achieved in isolates of CCC and CEC obtained from patients with metastatic PCa through quantitative RNA expression measurement. From preceding microarray studies, 23 functional gene loci were selected for obtaining CSE values and sensitivity data in patients with metastatic PCa. Technical details, including design details, are shown in Table 3. For each gene, a set of preclinical measurements was taken (details not listed) using the relative quantification method. Expression values and sensitivity are listed in Table 4. We did not investigate in greater depth structural gene loci, because the moiety of cytokeratin values did not exceed a sensitivity of 30–40% (unpublished results), which was regarded as unsatisfactory. The results from functional gene loci led to the choice of SOD2 and GPX1 as lead biomarkers in CCCs (sensitivity 79 and 97%, respectively) and AR in CEC (sensitivity 91%). Comparing the values with MNCs, a CSE threshold of 1.2 was used, taking into consideration a se of 10–15%.

Determination of inter- and intra-assay precision was smaller than the biological variation of cell numbers in CCCs and CECs respectively. The range of cells (GAPDH assay) varies from a few hundred to a few tens of thousands, exceeding by far results from spiking experiments. Introducing a marker under these circumstances requires a great stability of values in a stratified patient group. As one example, we chose GPX1 expression in CCCs from patients with metastatic disease without chemotherapy (Fig. 3). Three groups with different cell numbers were chosen, varying by a factor of >100. In contrast to cell numbers in CCCs the expression of GPX1 (CSE) was fairly stable. No significant difference was found. The inter-assay variability is presented as ± 95% CI. In addition, GPX1 expression was studied comparing cell numbers among patients with non-metastatic PCa (Fig. 4). Cell numbers in CCCs did not differ between stable and progressive disease, but GPX1 expression differed significantly when applying various statistical procedures. This finding prompted us to establish a set of metastases-predicting genes from the list shown in Table 4.

Figure 3.

Cell number and GPX1 expression in CCCs from patients with metastatic PCa before chemotherapy.

Figure 4.

Validation of cell count and GPX1 expression in CCCs from patients with PCA. Met, metastasis.

In addition to the CCC and CEC marker genes already known from their sensitivity, we selected two additional biomarkers for further characterization of CCC: cyclin B (sensitivity 72%; (mean CSE value 6.85; ± 95% CI 2.2; 11.5) and bFGF (sensitivity 69%; mean CSE value 37.6; ± 95% CI 1.84; 73.42). Thus, we were ready to apply the five-gene test in groups A and B for clinical validation. SOD2 and GPX1 were significantly overexpressed in CCCs from the blood of patients with metastatic PCa (Fig. 5), whereas TXNRD1 did not differ. The two AOX genes showed an excellent performance at outlined in ROC curves (Fig. 6). Cyclin B and bFGF expression in CCCs differed significantly from patients with non-metastatic PCa, albeit with a large range of CSE values (Fig. 7). AR expression in CECs was higher in patients with disease relapse than in group A (P = 0.04). Logistic regression analysis was used to identify the moiety of predictors of bone metastases (Table 5). Among the earlier therapies that patients had undergone, surgery and endocrine regimen were predictors of metastases (P < 0.05). Among the pathological variables, Gleason score and node stage were highly significant predictors. The best molecular predictors were GPX1, SOD2 and AR and a combination thereof. The diversity of clinical predictors made it unlikely that all of them act through the same biomechanism, e.g. molecular markers. This aspect was studied by calculating Spearman correlations by rank (Table 6). We could not find any correlation of predictive pathological variables (node stage and Gleason score) with GPX1, the leading molecular predictor of metastases. If, however, the antioxidant genes and AR are overexpressed to the level shown in Figs 5 and 7, the clinical search for metastases may be indicated. Taking together the molecular findings, a five-gene test appeared useful for generating a complete set of predictive data stemming from the 2 × 2 contingency table method (Table 7).

Figure 5.

AOX expression in CCCs isolated from patients with PCa with bone metastases before chemotherapy compared with that in patients with non-metastatic PCa.

Figure 6.

SOD2 and GPX1 expression at the stage of distant tumour recurrence shown as ROC curves.

Figure 7.

Expression of AR, cyclin B and bFGF in circulating tumour cells (n = 16 for each group). AR-CSE expression was studied using CEC, cyclin B and bFGF. CSE values were obtained from CCCs.

Table 5. Logistic regression analysis
Independent variableDependent variable: metastasisCorrect classification of cases, %
OverallMetastasisNo metastasisOR P
Therapies     
 Radiationredundant independent variable0.8
 Endocrine therapy66.318.896.15.70.02
 Radical prostatectomy72.721.294.84.90.01
Pathological variables     
 Tumour sizeredundant independent variable0.1
 Node stage79.253.888.69.00.001
 Gleason73.29.096.72.90.03
Molecular biology     
 GPX180.045.594.815.2<0.005
 SOD271.812.197.45.20.02
 AR83.366.793.816.00.009
Best combination     
 GPX1-SOD2-AR 100100 0.003
Table 6. Spearman's correlation by rank
Paired variablesSpearman's correlationt(N-2) P
Metastasis and GPX10.4081304.60275<0.001
Metastasis and node stage0.5060083.84696<0.001
Metastasis and Gleason score0.4444722.935430.006
GPX1 and node stage0.2123321.424850.161
GPX1 and Gleason score−0.151919−0.909320.369
Table 7. Performance of gene expression in CCCs and CECs (AR) for predicting distant metastases (CSE threshold values: SOD2 ≥2.4; GPX1 ≥3.4; AR ≥1.1; cyclin B ≥ 2.2; bFGF ≥ 3.8)
GeneSensitivity, %Specificity, %PPV, %NPV, %Prevalence, %Accuracy, %OR P
GPX17165577740674.3<0.001
SOD25170536840632.50.01
AR907569923581280.001
Cyclin B868686865086350.007
bFGF679189715278200.004
At least 2/5 genes94657689638132<0.001

Threshold values for each gene were introduced as indicated. The results of predictive molecular variables for each of the five genes are shown in Table 7. If two out of five genes exceeded the threshold values the best predictive values resulted: 94% for sensitivity, 65% for specificity, 76% for PPV and 89% for NPV. The prevalence was 63%. Test accuracy was 81% with an OR of 32 (P < 0.001). The data indicate that metastases formation is a rather complex phenomenon involving the protection of CCCs against reactive oxygen species, escape from immune surveillance in the bloodstream through SOD2 and GPX1 [9], drug resistance, cell cycle and angiogenesis [4,9,10–18]. In addition, the phenotype seems to indicate hormone refractory cancer.

Prediction of metastases raises further questions and answers regarding the molecular signature of palliative chemotherapy and OS prognostication. To obtain reliable data, gene expression at the metastatic level was compared between different patients before and after chemotherapy. Results are shown in Fig. 8. Cell marker genes SOD2 and GPX1 for CCC and AR for CEC were partially reduced after chemotherapy to the level of those of patients with non-metastastic PCa. This finding may mimic the complete success of the therapy regimen; however, the OS range is wide and dependent on the drug applied (Table 8). The idea emerging was that the action of drugs may be counteracted by the appearance of overexpressed resistance factors. Even for a single drug, the lifespan varies considerably, e.g. for taxotere it is between 5 and 99 months. Hence, we became interested in genes affecting the length of OS. At first, drug resistance factors were thought to play a key role. In the present study, we report the effects of MDR1 expression on OS. MDR1 overexpression was greatly enhanced in the course of chemotherapy (Fig. 8) in contrast to the cell marker genes. As a consequence, continued overexpression of GPX1 or MDR1 after chemotherapy worsens OS in a significant fashion (Fig. 9) if followed from the diagnosis of metastases to the end. As shown in Table 9, 62.9% of patients are still alive after 840 days when GPX1 is enhanced, whereas only 23.1% survive MDR1 overexpression. Putting together the data in the form of COX regression analysis, OS is worsened by 31% through MDR1 overexpression (hazard ratio [HR]: 1.31; 95% CI: 1.09–1.58; P = 0.003).

Figure 8.

Expression of key genes during disease progression. Met, metastasis.

Table 8. Chemotherapy and OS of the patients included in the study. 15 patients have died: nine under Taxotere, 2 nothing, 5 others
 Median (range) OS, months
Chemotherapy 
 Taxotere33 (5–99)
 Others4.5 (2–7)
 None24 (21–28)
Figure 9.

Overall survival in disseminating CCCs with overexpressed GPX1 and MDR1: comparison of lethal vs alive in patients with metastastic PCa after chemotherapy.

Table 9. Cumulative OS in patients with metastatic PCa
Days after diagnosisOS rate, %
MDR1GPX1
9100100
84023.162.9
16710.000037.8
25020.000027.0
33330.00009.0
41650.00000

A second endpoint in such studies is TTP. In the present study, TTP could only be calculated retrospectively. All patients with progressive disease had previously undergone various chemotherapy regimens before we analysed their blood. As a consequence, GPX1 expression failed to explain TTP (Fig. 10). The spectrum of different chemotherapies prompted us to look into various drug resistance candidates in the form of a Cox regression calculation (Table 10). Some drug resistance indicators seem to be persistently overexpressed, e.g. the effector gene for topoisomerase II blocking agents such as doxorubicin. After therapy, Topoisomerase remained overexpressed, as did MDR1, suggesting drug resistance. The role of overexpressed bFGF and MGMT as drug resistance factors requires further elucidation. The measurement of drug effector genes and drug resistance genes seems to be a method of describing drug effects and predicting drug outcomes.

Figure 10.

Time to progression and GPX1 overexpression.

Table 10. Cox regression for overexpression of drug targets and drug resistance candidates for TTP
Gene P HRCI −95%CI +95%
Topo II0.061.470.972.21
bFGF0.411.180.791.78
MDR10.131.130.961.31
MGMT0.271.200.722.0
 Multivariate analysis
Topo II0.151.410.872.2
bFGF0.61.120.71.8
MDR10.171.230.971.34
MGMT0.561.210.632.3

DISCUSSION

We hypothesize that molecular phenotyping of circulating tumour cells, namely CCCs and CECs can predict distant metastases. This strengthens the need to identify the independence of each of the two populations of circulating disease cells. We achieved this goal by spiking the blood of healthy volunteers with SW480 cells carrying a mutation in Ki-ras codon 12. Two cell types were used. Cell aggregates formed clusters composed of a mean of 7 cells, 30–70 µm in size. Individual scattered cells were collected after trypsinisation of cultures. They were 16.5 µm in diameter. In essence, clusters were completely captured on the mesh (20 µm) whereas scattered cells passed through the filter. Scattered cells, therefore, do not appear on the mesh. These appear in the filtrate awaiting further isolation techniques. There is no crossover of the two cell populations, although they stem from the same cell line. In vivo clusters show a gene expression pattern that is different from scattered cells [4]. Scattered cells of epithelial origin (CECs) harbour more epithelial messages than CCCs. CCCs have more mesenchymal messages, e.g. vimentin. In addition, circulating malignant cells in patients with sarcoma could be captured in CCCs (data not shown).

Gene expression was studied in cancer cells and in benign cells, i.e. MNCs or CD45+ expressing cells using quantitative RT-PCR. Interpretation of ct values was achieved by the relative quantification method. For each analysis and each gene a standard curve was run. Thus, we could avoid the mistakes associated with the ΔΔ ct method emerging from doubling values which are not precisely 2. Normalization with the housekeeping gene GAPDH has an additional advantage, namely total cell counting. Expression values were calculated in cancer cell fractions and in MNCs. The ratio is called CSE and allows the identification of cancer cell-specific traits absent in MNCs. In fact CSE values give much better results than does total cell counting only (Fig. 4). CSE values are considerably stable although the cell number in cancer cell fractions varies by three orders of magnitude (Fig. 3).

Distant metastases formation in PCa is believed to be related to circulating tumour cells [3]. Proving the hypothesis depends largely on the sensitivity of biomarkers. In the present study, we tested 23 candidate functional gene targets selected from microarray experiments. Improved data could be obtained by real-time PCR. The test designs are shown in Table 3. Expression level and sensitivity were studied by qRT-PCR in CCCs and CECs in advanced PCa, thus determining the oncogenic phenotype (Table 4). The highest sensitivity for overexpressed biomarker candidates was found for GPX1 (97%), SOD2 (79%) and AR (91%). This is in accordance with a previous study demonstrating the capability of three AOX genes (GPX1, SOD2, TXNRD1) in CCCs to predict primary PCa in patients with a serum PSA level 4–10 ng/mL [4]. Applying the analytical system for the prediction of metastases, however, resulted in changes of the pattern of predictive genes comparing patients without metastases and bone metastases. Firstly, TXNRD1 turned out to have lost the predictive capability since expression level remained unchanged (Fig. 3). Secondly specificity, PPV and test accuracy for SOD2 and GPX1 were considerably lower in metastases prediction than for primary PCa [4]. This unexpected phenomenon prompted us to consider other gene loci that may be involved in metastasis prediction. It seemed reasonable that these genes may reflect endocrine therapy which is known to fail in many metastatic cases [19,20]. We selected three gene loci from our expression/sensitivity gene pattern (Table 4). The mean CSE value for AR was 29.2 the sensitivity was 91%, cyclin B (6,85; 72%) and bFGF (37,63; 69%).

Genomic and expression changes of AR have been reported to be related to hormone refractory tumours [10,11]. The aberration is sustained in bone metastasis [12]. Overexpression of AR may sensitize cells for low levels of androgens [13]. The AR has been reported to regulate cellular proliferation by control of CDK and cyclins at the transcriptional level and by post-translational modifications that influence cell cycle protein activity, including overexpression of cyclin B in recurrent tumours [14]. A further facet of cyclin B overexpression has been reported, namely, to sensitize malignant prostate cells to apoptosis induced by chemotherapy [15]. BFGF has been reported to have several functions in advanced PCa. A variety of angiogenic factors including bFGF are overexpressed while cells convert from the premalignant status to malignancy [16]. The net result of increased bFGF signalling includes enhanced proliferation, resistance to cell death, increased motility and invasiveness, increased angiogenesis, enhanced metastasis, resistance to chemotherapy and radiation and androgen independence, all of which can enhance tumour progression and clinical aggressiveness [17]. The role of bFGF in chemoresistence has been also detected in other solid tumours [18]. Overexpression of all three genes in circulating target cells obtained from patients with metastatic PCa reached significance, suggesting that cells have acquired endocrine therapy resistance (Fig. 7). AR was found to be overexpressed in CEC, bFGF and cyclin B in CCC obtained from metastatic patients.

Introducing CSE threshold values of these five different RNA served to distinguish group A (non-metastatic) and B (metastatic) patients. Each individual gene was already a significant predictor of distant metastasis (Table 7). When a minimum of 2/5 genes exceeding threshold values were considered, improved prediction values were reached as compared with AOX genes alone resulting in highly significant P values. The data indicate that metastases formation is a rather complex phenomenon as compared with primary PCa [4], which involves various mechanisms, e.g. drug resistance, cell replication and angiogenesis. We believe that the present study may also have some impact on therapy regimen, e.g. inhibition of bFGF or blockade of the AR or downstream effectors. Further interventional studies are, however, required to investigate the therapeutic effects in CCCs and CECs.

We investigated potential metastases predictors including therapies and pathological variables (Table 5). Radical prostatectomy and endocrine therapy were found to be predictors of metastases. Also node stage and Gleason score are significant predictors of bone metastases, as are SOD2, GPX1 and AR overexpression. Pairing GPX1 with pathological variables using Spearman's rank, however, showed no correlation (Table 6) suggesting that AOX overexpression can predict bone metastases irrespective of pathological variables and therapies. A previous study from our laboratory has already shown that AOX genes prognosticate recurrence-free survival independently of pathology variables [4]. High-risk pathological variables may lead to metastases in cases of GPX1 overexpression or may not in cases of low GPX1 expression. In fact, GPX1 overexpression worsened OS significantly (Fig. 9) when monitored in a prospective fashion. The expression of GPX1, SOD2, AR and MDR1 is governed by chemotherapy. Genes predicting metastases showed a marked reduction in CSE values (Fig. 8). This may indicate partial clearance from these cells. By contrast, MDR1 expression was greatly enhanced, suggesting the development of drug resistance under chemotherapy beyond primary resistance. As a consequence MDR1 overexpression worsens OS by more than 30% (Fig. 9; Table 9). We calculated a HR of 1.31 (95% CI: 1.09–1.58).

We also studied patients who have had tumour progression before gene expression analysis.

Some of them had been pretreated with topoismerase II blockers, but also with other drugs. Four overxpressed genes after therapy were studied. Overexpressed topoisomerase II in CCCs after therapy reduced TTP by 47% (Table 10), suggesting either drug resistance (MDR1) or underdosage. The other drug resistance candidates MGMT and bFGF showed no effect.

The data show that the cell and the analytical system can be used as a platform for the predicition of clinical tumours. Studying a drug-sensitive cell biomarker and at the same time drug targeting and drug resistance would be a novel approach that makes use of circulating cancer cells.

In conclusion, the present study provides a novel insight into metastases formation and drug treatment beyond the serum PSA level alone. AOX genes reflect the advanced tumour type in as much as only SOD2 and GPX1 are predictrive, but not TXNRD1 as observed for the predicition of primary PCa. In addition the molecular phenotype (AR, cyclin B, bFGF) in CCCs and CECs from hormone refractory PCa has been described for the first time. The molecular signature of hormone therapy in the adjuvant situation as well as chemotherapy regimen at the metastatic stage can be applied in routine practice and in research. We believe that this subject warrants further clinical studies.

ACKNOWLEDGEMENTS

This study was supported by NRW, Germany, grant No. PTJ – 9804v02.

CONFLICT OF INTEREST

Michael Giesing is inventor and owner of several patents underlying this work; in Europe these are EP 1100 873, EP 1 532 444. The patents have been granted in the USA and in Canady in part. Bernd Suchy is co-inventor of EP 1 532 444. The patents stem from the grant NRW, Germany, grant No. PTJ – 9804v02.

Ancillary