Quantitative real-time RT-PCR for detection of disseminated tumor cells in peripheral blood of patients with colorectal cancer using different mRNA markers



The detection of disseminated tumor cells in peripheral blood from colorectal cancer patients by RT-PCR could be an attractive method for selecting patients for adjuvant therapy. We here report on real-time RT-PCR assays (LightCycler) to quantitate potential mRNA markers. We investigated specimens from colon carcinoma and normal colon mucosa tissues, cell lines, blood samples from 129 patients with colorectal cancer (all stages) and 58 reference blood samples (healthy donors, persons suffering from inflammatory bowel or infectious diseases). The expression profile in tissues showed high values for CEA and CK20, whereas in cell lines ProtM was predominant. All markers were detected in reference and patient blood samples (ProtM, 22, 17%; CEA, 84, 86%; CK20, 85, 88%). After quantitative analysis, the definition of cutoff values for each marker and the combination of markers, 13% of patients were judged to have elevated marker concentrations in their blood, from which only 6 had values significantly differing from cutoff value. There were no differences between stages of disease. In the case of 19 patients, investigated prior to and 1 week after surgery, 2 samples revealed a significant postoperative increase in CEA or CK20 mRNA concentration. In spite of high expression levels in tissues and cell lines, we were not able to differentiate satisfyingly mRNA markers originating from tumor cells and those from illegitimate transcription in hematopoetic cells in blood. We conclude that either copy numbers of analyzed markers in circulating tumor cells are not sufficient for detection or, more probably, peripheral blood is not a suitable compartment for detection of tumor cells in colorectal cancer. © 2003 Wiley-Liss, Inc.

The molecular monitoring of circulating tumor cells by RT-PCR, routinely applied in patients with certain leukemias and lymphomas,1, 2, 3, 4 is still under debate for patients with solid malignancies. In colorectal cancer, where indication for adjuvant therapy without metastasis is yet performed by histologic investigation of lymph nodes,5 the immunocytochemical identification of epithelial cells in bone marrow6 encouraged the detection of epithelial mRNA markers in blood and bone marrow by RT-PCR.7 However, a series of subsequent investigations by a number of groups with conventional nested PCR led to conflicting reports on both the frequency of gene transcripts in blood of patients and the specificity of the method.8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20

The recent availability of real-time PCR equipment has obviously changed the situation.21, 22, 23 The quantification of low-level background transcription allows the definition of cutoff values for marker expression in blood and thus improves specificity.16, 22, 24, 25 Furthermore, the reliable quantification of housekeeping gene expression allows excellent quality control on a per-sample basis and relates absolute marker concentration to sample quality.

We now developed quantitative real-time RT-PCR assays to measure housekeeping gene expression and the expression of 3 colorectal cancer-associated mRNAs. Two of them, CEA and CK20, have been described and largely used before.9, 13, 14, 26, 27, 28, 29, 30, 31 The recently identified gene of protease M (Zyme, Neurosin) belongs to the serine protease gene cluster located on chromosome 19q13 and codes for a trypsin-like serine protease isolated from colon and ovarian cancer cell lines.32, 33, 34 It is expressed in a variety of tissues, including colorectal tumors, and due to its sequence homology with several kallikreins and prostate-specific antigen has been suggested to play a role in cancer diagnosis.34, 35 Here we report on the expression of the 3 mRNA markers (ProtM, CEA, CK20) in colon cancer tissues, cell lines and peripheral blood samples of 129 patients with colorectal cancer. Quantitative PCR is used for differentiation between marker background expression in blood and those probably arising from epithelial cells.


CEA, carcinoembryonic antigene; CK20, cytokeratin 20; GTC, guanidinium thiocyanate; MNC, mononuclear cell; NSM, number of standard molecules; PBGD, porphobilinogen deaminase; ProtM, protease M; qRT-PCR, quantitative reverse transcriptase-polymerase chain reaction; UICC, Union International Contre le Cancer.


Patient selection

All patients had given informed consent for the analysis. Clinical disease status was determined according to the Union International Contre le Cancer (UICC) guidelines. A hundred twenty-nine colorectal cancer patients entered the study and had complete tumor evaluation at the time of blood withdrawal and during follow-up. Ninety-two patients underwent surgery for their tumors and blood sampling was performed prior to surgery on the same day. From 19 operated patients, additional blood samples were taken 1 week after surgery. In 38 patients with advanced disease, samples were obtained at the occasion of outpatient clinic visits, regardless of current treatment. Follow-up samples were obtained in 10 of them on the occasion of outpatient clinic visits. Negative controls were blood samples of healthy volunteers (n = 45) who had no evidence of any clinically detectable disease at the time of blood withdrawal and of patients with inflammatory bowel or infectious diseases (n = 13).

Processing of blood samples

Samples of 10 ml blood were collected in EDTA-containing tubes. Sample processing was performed within 2 hr after blood withdrawal. Blood was transferred into a 50 ml Falcon tube and centrifuged at 1,600g for 10 min. Serum was removed and the cell pellet resuspended in 5 ml guanidinium thiocyanate (GTC) buffer. After mixing well, the GTC buffer/blood cell mixture was stored at −80°C.

Processing of tissue specimens, cell lines, blood-spiking samples

Tissue samples (15 patients with colorectal cancer pathologically diagnosed as adenocarcinomas and 4 normal mucosa specimens) were obtained by surgical resection, collected and dissected under stringent sterile conditions to prevent RNA contamination and immediately frozen in liquid nitrogen. All cell lines (COLO 205, LS-174-T, CX 2, CX 94, HCT 116, HT 29, CaCo2) were provided by the American Type Culture Collection (Rockville, MD) or the Cell Line Service (Heidelberg, Germany). Cells were grown in RPMI medium supplemented with 10% fetal bovine serum at 37°C in a 5% CO2 air environment. For blood-spiking experiments, COLO 205 cells were serially diluted in PBS and mixed with 10 ml blood obtained from healthy volunteers to give concentrations of tumor cells between 1 and 106 cells per 10 ml blood.

RNA extraction and reverse transcription

Total RNA of peripheral blood samples was isolated by acid guanidinium thiocyanat/phenol chloroform extraction36 using Phase Lock Gel Heavy tubes (Eppendorf, Hamburg, Germany) and further purified by the High Pure RNA Isolation Kit (Roche Diagnostics, Mannheim, Germany). Total RNA of cell lines and tissues was isolated by RNeasy Mini Kit including RNase-Free DNase Set (Qiagen, Hilden, Germany). RNA integrity was checked electrophoretically and quantified spectrophotometrically. For reverse transcription, 2 μg of RNA was diluted in 15 μl RNase-free water, incubated 5 min at 65°C and placed on ice. In 54 cases, RNA from a second blood sample was processed with higher RNA concentration in cDNA synthesis. A 7.5 μl mixture containing 2 μl oligo-p(dT)15 primer (0.8 μg/μl), 2 μl dNTP (5 mM), 0.5 μl RNase Inhibitor (40 U/μl), 1 μl Omniscript Reverse Transcriptase (4.5 U/μl), and 2 μl RT buffer (×10) was added to the diluted RNA. After incubation at 37°C for 1 hr, Omniscript Reverse Transcriptase was inactivated for 5 min at 95°C and cDNA was stored at −20°C. All RT reagents except oligo-p(dT)15 primers and RNase Inhibitor (Roche Diagnostics) were purchased from Qiagen.

Quantitative real-time PCR (qRT-PCR)

PCR conditions for the LightCycler (Roche Diagnostics) were summarized in Table I. Two μl of each cDNA was diluted to a volume of 20 μl PCR mix (LightCycler Faststart DNA Master Hybridization Probes, Roche Diagnostics) containing a final MgCl2 concentration as listed in Table I, 0.5 pmol of each primer, and 0.2 pmol of each probe. Primer sequences specific for CEA transcripts were originally described by Gerhard et al.26CK20-, ProtM- and PBGD-specific primers were carefully designed using the oligo 6.0 software. The PCR primers and probes have been positioned to span exon-intron boundaries, reducing the risk of detecting genomic DNA. Furthermore, the specificity of the product is ensured by hybridization probes selectively binding to marker-specific amplification products of cDNA, 2 of them included in each run with either a donor fluorophore at its 3′ end (Fluorescein) or an acceptor fluorophore (LC Red 640) at the 5′ end and phosphorylated at its 3′ end to prevent probe extension. Primers were purchased from Metabion (Martinsried, Germany) and probes from TIB Molbiol (Berlin, Germany) as well as Metabion. For amplification, an initial denaturation at 95°C for 10 min, followed by PBGD: 0 sec at 95°C, 12 sec at 65°C and 10 sec at 72°C; ProtM: 0 sec at 95°C, 12 sec at 67°C and 10 sec at 72°C; CEA: 10 sec at 95°C, 12 sec at 65°C and 6 sec at 72°C; CK20: 10 sec at 95°C, 10 sec at 65°C and 5 sec at 72°C, for 50 cycles with a final extension of 2 min at 72°C was used. The expected size of the PCR products was confirmed by agarose gel electrophoresis.

Table I. Primer and Hybridization Probe Sequences, Conditions and Annealing Temperature of Real-Time RT-PCR
Marker Primer/hybridization probe sequenceAmplicon (bp)Annealing temperature (°C)MgCl2 concentration (mM)

All samples were analyzed in duplicate. The average value of both duplicates was used as quantitative value. If only one of the duplicates gave a positive signal, the positive result was taken. In each individual case, we have checked that the positive value was near detection limit (i.e., nondetection was a sporadic event) and housekeeping gene value was in the normal range.

Plasmid controls, standard curve

PCR products generated from ProtM, CEA, CK20 and PBGD cDNAs were cloned into the vector pCR2.1-TOPO (Invitrogen, Karlsruhe, Germany). Recombinant vectors, linearized with EcoRV, were serially diluted in water containing 0.4 μg/μl polyadenylic acid (Pharmacia Biotech, Freiburg, Germany). A standard curve with 3 plasmid dilutions in duplicates of TOPO 2.1-ProtM, TOPO 2.1-CEA, TOPO 2.1-CK20, and TOPO 2.1-PBGD was included in each respective PCR run.


To reduce risk of contamination, thermocycling and post-PCR steps were performed in separate laboratories to that used for RNA extraction, cDNA synthesis and preparation of the PCR mixture. PCR mixtures were set up in a template tamer (Oncor Appligene, Heidelberg, Germany). All reagents for cDNA synthesis were prepared with RNase-free water. For all RT-PCR steps, negative controls were performed, including a reverse transcriptase negative sample control for every sample and a water control for every PCR run.

Data analysis

With the LightCycler software (version 3), crossing points (beginning of the PCR exponential phase) were assessed by the second derivative maximum algorithm and plotted against the concentrations of the standards. Sample concentration was calculated using the plasmid standard curve, resulting in marker concentrations expressed as copy number of corresponding standard molecules/μl (NSM/μl). The relative sample amount was expressed as ratio marker (ProtM, CEA, CK20 [NSM/μl])/(PBGD [NSM/μl]). The statistical analysis was performed using the Kolmogorov-Smirnov 2-sample test (Statistica software, release 4.1).


Quantitative range of RT-PCR assays

To establish quantitative range and principal detection limit, serial 10-fold dilutions of marker-specific recombinant TOPO 2.1 plasmids were assayed in duplicate. Real-time RT-PCRs of serially diluted plasmid standards provided quantitative data for all markers (Fig. 1). The detection of one copy was possible for all markers, whereas precise quantification would require at least 10 molecules per assay. The dynamic range of quantitation using plasmid standards was 8 orders of magnitude for all markers.

Figure 1.

Correlation between theoretical and measured standard concentrations for at least 8 plasmid dilutions of TOPO 2.1-ProtM, TOPO 2.1-CEA and TOPO 2.1-CK20 implemented in duplicate. For regression, all plasmid dilutions with more than 10 molecules per assay were used.

Expression of ProtM, CEA and CK20 in colon carcinoma cell lines and tissues

In all cell lines, CEA-, CK20- and ProtM-specific mRNA was detected (Fig. 3). The mRNA expression levels varied in colon carcinoma cell lines over 5 (CEA) and 3 (ProtM, CK20) logs each, whereas the median expression of ProtM was highest with a marker/PBGD ratio of 2.6 and 2 logs higher than that of CEA (0.055) and CK20 (0.018).

Figure 3.

Level of ProtM, CEA and CK20 mRNA by real-time RT-PCR of colon carcinoma cell lines (n = 7). Ratio marker/PBGD: the relative sample amount was expressed as ratio marker (ProtM, CEA, CK20 [NSM/μl])/(PBGD [NSM/μl]). The sample concentration was calculated using the plasmid standard curve.

One colon carcinoma tissue specimen in our series was completely negative for ProtM and showed a significantly lower expression for the other 2 markers (Fig. 2). For colon carcinoma tissues, the median expression of CEA was highest with a marker/PBGD ratio of 330, followed by CK20 (73), and 3 logs lower by ProtM (0.12). As to CEA and CK20, PCR results of the 4 normal mucosa tissue samples fell in line with that of the colon carcinoma tissue samples, whereas in 2 of 4 normal colon specimens ProtM was not detectable and the mean expression level in positive samples was about 2 logs lower than that of carcinoma samples.

Figure 2.

Level of ProtM, CEA and CK20 mRNA by real-time RT-PCR of colon carcinoma (n = 15) and normal colon mucosa tissues (n = 4) snap-frozen after surgical resection. Ratio marker/PBGD: the relative sample amount was expressed as ratio marker (ProtM, CEA, CK20 [NSM/μl])/(PBGD [NSM/μl]). The sample concentration was calculated using the plasmid standard curve.

Detection of markers in blood of healthy persons

In 22.5%, 83.7% and 84.6% of healthy volunteers, ProtM, CEA and CK20 mRNA expression was detectable with a level of up to 8.5 × 10−5, 81 × 10−5, and 59 × 10−5 ratio to housekeeping gene, respectively. Low-level expression of all markers in the peripheral blood of healthy volunteers required the introduction of cutoff marker/PBGD ratios. We considered 2 cutoff strategies.

In strategy 1, we assumed the data to be samples from a log-normal distribution. The fitting results for CK20 and CEA are shown in Figures 4 and 5. As cutoff marker/PBGD ratios, we have taken the 99 percentile, corresponding to a cutoff ratio CEA/PBGD of 93 × 10−5 and CK20/PBGD of 95 × 10−5. For protease M, the number of positive blood samples was too small to be a reasonable basis for a statistical distribution. The cutoff ProtM/PBGD ratio was therefore related to maximum value in healthy control blood. Positive findings for ProtM were only arbitrarily judged as positive if they exceeded the healthy volunteer background by a factor of 2, resulting in a cutoff ratio ProtM/PBGD of 17 × 10−5. In strategy 2, marker/PBGD ratios in patient's blood were judged as positive when exceeding the maximum value of healthy volunteer background.

Figure 4.

Cumulative frequency for CEA in blood of healthy persons (diamonds). The solid line represents the fit of data to a log-normal distribution. The relative sample amount was expressed as ratio (CEA [NSM/μl])/(PBGD [NSM/μl]). The sample concentration was calculated using the plasmid standard curve.

Figure 5.

Cumulative frequency for CK20 in blood of healthy persons (diamonds). The solid line represents the fit of data to a log-normal distribution. The relative sample amount was expressed as ratio (CK20 [NSM/μl])/(PBGD [NSM/μl]). The sample concentration was calculated using the plasmid standard curve.

Blood-spiking experiments

As an example of how the detection of tumor cells is influenced by target mRNA amount in these cells, we spiked different amounts of COLO 205 cells into blood samples from healthy volunteers (Fig. 6). Using ProtM as marker, the differentiation of tumor cells from background is possible for concentrations higher than 36 cells in 10 ml blood (corresponding to about 1 cell in 2 × 106 mononuclear cells, or MNCs). As to CEA, only more than 2,300 tumor cells in 10 ml blood can be differentiated from background, corresponding to about 1 cell in 4 × 104 MNCs, while in the case of CK20, it would be necessary to have 105 tumor cells in 10 ml blood (results not shown). The difference in detection sensitivity for ProtM and CEA is about 70, resulting from a different expression of the corresponding marker in COLO 205 cells (ProtM:CEA:CK20 = 1,000:66:3.8, indicated by the shift of the regression line to the right) and different background levels (cutoff marker/PBGD ratios) in blood. This result is cell line-specific and should be similar for CX2 but completely different for the other cell lines (Fig. 3).

Figure 6.

Detection of COLO 205 cells in blood using ProtM and CEA mRNA. Different amounts of COLO 205 cells were spiked in blood of healthy persons. For regression (solid lines), only values above cutoff (broken lines) were used. The relative sample amount was expressed as ratio (ProtM, CEA [NSM/μl])/(PBGD [NSM/μl]). The sample concentration was calculated using the plasmid standard curve.

Expression of ProtM, CEA and CK20 mRNA transcripts in blood samples of patients with colorectal cancer

ProtM transcripts were detected in 17%, CEA in 86% and CK20 in 88.4% of patient blood samples (Figs. 7–9). Detection frequencies are thus in line with those found for healthy volunteers. With cutoff strategy 1 (strategy 2 in parentheses), in 1 (2), 15 (18) and 5 (20) out of 237 blood samples from 129 patients, ProtM, CEA and CK20 mRNA expression was over background. Only one blood sample (rectum carcinoma UICC IV) was positive for more than one marker (CEA, CK20). For one patient (colon carcinoma UICC IV), one blood sample was positive for CEA and a second one for CK20. In 17% (CEA), 4% (CK20) and 37% (ProtM) of samples, one of the duplicates gave no signal. Marker/PBGD ratios for these cases are all far below the cutoff marker/PBGD ratio even under the assumption of a hypothetical failure of 10% (relative standard deviation) due to our data management. Strong increased expression levels (more than 5-fold above background level) were found in 5 samples for CEA and in the 1 positive for ProtM. When all blood samples positive for at least one marker are considered, 17 out of 129 patients (13.2%) showed increased marker expression levels (Table II). By cutoff strategy 2, this proportion would be 24.8%, also including 1 of the inflammatory disease samples and 3 of the adenomas. In both cases, a correlation of marker detection with stage of disease, either as to percentage of positive blood samples or to marker expression levels, is lacking.

Figure 7.

Detection of ProtM in blood of healthy persons (H), of patients with inflammatory bowel or infectious diseases (I), and patients with colorectal cancer in dependence on UICC stage (circles, preoperative blood samples; diamonds, postoperative blood samples; cross, blood samples of patients under chemotherapy or vaccination). The relative sample amount was expressed as ratio (ProtM [NSM/μl])/(PBGD [NSM/μl]). The sample concentration was calculated using the plasmid standard curve.

Figure 8.

Detection of CEA in blood of healthy persons (H), of patients with inflammatory bowel or infectious diseases (I), and patients with colorectal cancer in dependence on UICC stage (circles, preoperative blood samples; diamonds, postoperative blood samples; cross, blood samples of patients under chemotherapy or vaccination). The relative sample amount was expressed as ratio (CEA [NSM/μl])/(PBGD [NSM/μl]). The sample concentration was calculated using the plasmid standard curve.

Figure 9.

Detection of CK20 in blood of healthy persons (H), of patients with inflammatory bowel or infectious diseases (I), and patients with colorectal cancer in dependence on UICC stage (circles, preoperative blood samples; diamonds, postoperative blood samples; cross, blood samples of patients under chemotherapy or vaccination). The relative sample amount was expressed as ratio (CK20 [NSM/μl])/(PBGD [NSM/μl]). The sample concentration was calculated using the plasmid standard curve.

Table II. Percentage of Blood Samples Positive for at Least One Marker in Patients with Colorectal Cancer and in Reference Persons in Dependence on Cut-Off Strategy
 All patientsColon-CaRectum-CaReference
All0–IIIIIIVAllI-IIIIIIVHealthInflammatory bowel or infectious diseases
Number of patients12969301524601918234713
% positive by cutoff I13.215.913.320.016.710.010.511.
% positive by cutoff II24.826.123.326.729.223.315.822.

In UICC stage IV, 36 out of 47 patients had chemotherapy or vaccination and 11 underwent primary surgery of their tumors. The percentage of patients with positively judged blood samples in both groups by cutoff strategy 1 was 11.1% (chemotherapy) vs. 9.1% (surgery, preoperative blood). From 10 patients under chemotherapy or vaccination, blood samples were obtained at more than one time point with at least 1-week difference. For 8 of them, we obtained the same results for all time points (7 negative, 1 positive). One patient was negative for 2 investigations in distance of 1 week and positive after 3 months. A second was negative at beginning, positive after 3 and 4 months, negative after 5 months and positive again after 6 months.

For 19 patients, we compared expression levels of ProtM, CEA and CK20 in blood samples taken pre- and postoperatively. All preoperative samples are negative for any of the markers on applying cutoff strategy 1. One week after surgery, the blood samples of one patient with colon carcinoma UICC III became positive for CEA or CK20 and of a second patient with colon carcinoma UICC IV positive for CEA only.

For 70 patients, we collected 2 blood samples at one time point. In 59 cases, both samples were negative and in 1 case positive. For 10 patients, one sample was positive and a second one negative. Considering all patients with double blood sampling, 8.6% of blood samples have been judged as positive (UICC 0 to III, 10%; UICC IV, 5.7%), whereas for patients with one collected blood sample, 9 out of 97 (9.3%) blood samples (UICC 0 to III, 7.1%; UICC IV, 13.6%) were positive. When related to number of patients with positive blood samples, 18% of those with 2 samples and 10% with 1 sample were positive (18.6% vs. 7.9% in UICC 0 to III; 17.6% vs. 12.9% in UICC IV), suggesting that multiple sampling might be advantageous.

PBGD housekeeping gene transcript levels

For estimation of sample quality, we considered PBGD concentration per μg implemented RNA in cDNA synthesis. The median PBGD content per μg RNA in the 273 blood samples described in detail above was 18,578 NSM, with limited variation, namely, a 10% percentile of 9,034 and a 90% percentile of 41,311. The median in reference blood samples was slightly lower (15,293; p = n.s.).

To analyze the potential influence of PBGD concentration (sample quality) on the detection of ProtM, CEA or CK20, we compared the distribution of total PBGD values of the 20 PCR-positive samples (cutoff strategy 1) with that of all samples (Fig. 10). PBGD values of PCR-positive samples are evenly distributed over the range of PBGD values below the 75% percentile (80,226) with a median value in positive samples of 29,556 (near the 25% percentile; 31,102). There were 5 positive samples below the 10% percentile of PBGD values (< 21,867), 10 positive samples between the 10% percentile and the median (50,949), 5 between the median and the 75% percentile (< 80,226).

Figure 10.

Housekeeping gene concentration (PBGD) in reference (healthy persons, patients with inflammatory bowel or infectious diseases) and patient (colorectal cancer) blood, and in those blood samples judged as positive for at least one of the mRNA markers by RT-PCR. The sample concentration was calculated using the plasmid standard curve.


The goals of RT-PCR assays for detecting circulating tumor cells include the definition of patient groups with increased risk for development of hematogenous tumor spread, monitoring of circulating tumor burden and of antigen expression in circulating tumor cells. The scope of this study is an evaluation of this method using quantitative RT-PCR as to its applicability for blood samples in colorectal carcinoma.

By our quantitative RT-PCR assays, quantification is guaranteed over a wide concentration range down to 10 standard molecules within the reaction tube and detection up to one molecule. This high sensitivity makes it possible to abandon the concept of nested PCR often used for conventional PCR, which is time-consuming and can be coupled with a loss of specificity due to the danger of contamination.24, 25, 37 Using LightCycler, single-round PCR is as sensitive as nested PCR (results not shown), which was even described for conventional methods under optimized PCR conditions.38

A further advantage of quantitative PCR is the possibility to take into account variations in RNA and/or cDNA quality by quantifying housekeeping genes and subsequent normalization of marker concentration to that of the housekeeping gene. For our study, we choose PBGD as a low-abundant, usually unregulated constitutively expressed housekeeping gene with no known pseudogenes. The homogeneity of housekeeping gene expression of our samples is remarkably uniform, with the 10% and 90% percentiles only a factor of 4 apart. In preliminary experiments, we analyzed a second housekeeping gene (glucose-6 phosphate dehydrogenase). However, the range of data values comprised 4 orders of magnitude and the 10% and 90% percentiles were a factor of 12 apart. We therefore renounced to include marker/G6PD ratios in order to prevent false positives due to low G6PD values. Considering the detection of CEA, CK20 or ProtM transcripts in blood samples, we found only 9 samples to be completely negative for all of the markers and their PBGD concentrations were equally distributed over the whole range, implying that false negative samples due to low cDNA quality can be neglected in our study.

The next issue concerns the expression level of markers in tissues and cell lines. In spite of the high expression of CEA and CK20 in tissues and of ProtM in cell lines, the variation in expression levels was between 2 and 5 logs. This variation can obviously impact on diagnostic sensitivity in peripheral blood. By our cell-spiking experiment, we have shown that the PCR sensitivity is basically determined by the expression level of the marker gene and the corresponding background level in blood. Detection of less than 100 cells in 10 ml blood is only possible when the copy number in a single cell is high enough to ensure at least one copy/cell in NSM after performance. Considering further that the expression in circulating tumor cells might vary from tissue-resident cells,39 a multimarker assay could be advantageous.14, 15, 16, 39, 40 Tumor cell heterogeneity for the known markers rules out any quantitative correlation between marker concentration found in blood and tumor cell load.

PCR sensitivity is markedly limited by marker background level expression in blood, interpreted as illegitimate transcription in white blood cells, which in some cases is influenced by cytokines.20 We found for all markers background expression in blood from healthy persons. The expression of CEA and CK20 mRNA was higher in frequency and level (one log). In the literature, there are controversial reports concerning background expression of CEA and CK20, ranging from 0%12 to 100%.41 Beneath different PCR sensitivities, RNA handling (use of whole blood, isolation of total white blood cell fraction by erythrocyte lysis, or isolation of mononuclear cell fraction by density gradient centrifugation) is imputed to influence background expression.42 Due to inconsistent results concerning CK20 expression in mononuclear cells and granulocytes,41, 42, 43 the method of choice is still under debate. Since cell separation or red blood cell lysis prior to RNA isolation could lead to a loss of tumor cells,44, 45 we preferred the RNA extraction from whole blood. The newly developed enrichment of tumor cells by immunomagnetic bead selection could be advantageous in lowering background levels.10, 46, 47, 48

Background level expression requires cutoff strategies realized in conventional PCR by reduction of cycle number and/or probe volume.46 Since the cutoff in quantitative PCR is related to marker/housekeeping ratio, both cutoff strategies might give different results, especially for samples with low cDNA quality. The usual cutoff strategy is orientated at maximum background level.22, 49 For CEA and CK20, we improved quantification by fitting the data to an empirical distribution.

In our sample series, we judged only 13% of patients with colorectal disease (17 out of 129; ProtM, 1; CEA, 11; CK20, 5) positive for tumor cells in peripheral blood. The higher number of positive patients by a more moderate cutoff strategy was coupled with a marked loss of specificity so that we preferred the stronger criteria of strategy 1. Remarkable was the low frequency in the group of patients with metastases, which was independent of therapy. The question is now to decide whether it is probable that marker values being greater than cutoff value are really reflecting tumor cells in blood and not false positives. First, the percentage of detection of markers in blood of healthy people and patients is similar. Second, the marker values of positively judged blood samples are in 72% near the corresponding cutoff point (CK20, some CEA values), except for 5 CEA and the one ProtM sample where the marker concentration exceeds the cutoff value more than 4-fold. Third, the percentage of positive patients in different disease stages was not significantly different. These facts indicate that marker values in patient's blood could arise from background expression as well as from tumor cells. Classification would only be possible by additional methods, e.g., flow cytometry.50 Similar uncertainties were observed after cytologic investigation of CEA-positive mesenterical blood probes,51 morphologic and immunofluorescent investigations of CK20-positive cells in bone marrow52 and CK20 detection in peripheral blood by qRT-PCR.16 As to detection frequencies, the results from other groups are very heterogeneous independently of the marker used (Table III), whereby some found similar low detection rates.47, 49, 53, 54 Moreover, high values as obtained by Castells et al.29 are paralleled by 55% positive blood samples in patients with benign inflammatory diseases. One frequently used explanation of detection failure is that circulating cells are not homogeneously distributed and noncontinuously shed into circulation.17, 29 Considering the results with our large patient cohort as well as those of other groups showing that even by multiple sampling (at one and different time points) PCR detection rates are not satisfying, the tumor cell shed should be a relatively rare event. Thus, the question arises as to whether peripheral blood is a suitable compartment. Other compartments such as bone marrow or mesenterical blood are known to provide higher detection rates, probably due to a larger number of tumor cells present.6, 51, 53 Accessibility, however, is more difficult and the problem of illegitimate transcription remains.22

Table III. Data from Literature for Detection of Tumor Cells in Peripheral Blood by Amplification of mRNA Markers
StudiesPatients with positive blood samples (%)Tumor stageMarker
Jonas et al.1784IVCEA
Mori et al.2835 (50)I–IV (IV)CEA
Guadagni et al.4069 (76)I–IV (IV)CEA
Castells et al.2941 (60)I–IV (IV)CEA
Ito et al.494 preoperation, 26 postoperation0–IIICEA
Patel et al.5569I–IVCEA, CK20
Weitz et al.1241I–IIICK20
Wharton et al.1474I–IVCEA, CK20
Wyld et al.1148IVCK20
Soeth et al.917I–IVCK20
Koch et al.5311I–IVCK20
Chausovsky et al.3063IVCK20
Weitz et al.5412.5I–IICK20
Hardingham et al.4720I–IIICK20, CK19, MUCI,II

It is often discussed whether tumor cells are shed into circulation during surgery of the primary tumor.12, 49 In our small group of 19 patients, 2 patients perhaps had tumor cell shedding due to surgery. On the other hand, the increase in detectable marker transcripts could also be a reflection of noncontinous tumor cell shed independent of the tumor resection.

Another issue concerns the suitability of our markers for detection of tumor cells in blood. The well-known markers CEA and CK20 are probably expressed in circulating tumor cells but devaluated by the relatively high background level expression. The newly introduced marker ProtM shows an encouraging high expression in all cell lines as well as a lower background expression in blood and a low expression in normal mucosa. In spite of these promising facts, the expression in circulating tumor cells seems to be insufficient, which underlines that colon cell lines are not representative with regard to the expression profile. Since the identification of an appropriate target gene is an important parameter, it remains to be seen whether other markers, such as genes of the MAGE family, would improve the situation.15 Preliminary results from our group, however, show that at least MAGE 3 and 6 are also expressed in blood and bone marrow from healthy people at considerable levels.

In summary, our results clearly question the suitability of the compartment peripheral blood for the detection of circulating tumor cells in colorectal cancer patients. We cannot give a final answer to the question as to whether nondetectability is a problem of marker or of the system itself, i.e., tumor cell shed into circulation, since the ideal marker (no illegitimate expression in blood, high expression in tumor cells) is not yet found. Nevertheless, by our experiments, convincing arguments are yielded underlining the latter hypothesis. We cannot, however, exclude the possibility that in the future, by using other markers, the situation would change. Thus, the search for an appropriate target gene remains an important concern.