Quantitative Reverse Transcription-Polymerase Chain Reaction Detection of Cytokeratin 20 in Noncolorectal Lymph Nodes¹

Despite the histopathological diagnosis of tumor-free margins, many cancer patients suffer a recurrence of their disease (1). One belief is that these recurrences may arise from residual disseminated tumor cells (also termed “micrometastasis” or “minimal residual disease”) that at present are undetectable by conventional methods (2). The detection of these cells by molecular techniques could impact greatly on future patient management and therapy; this option has driven a growing field of research in recent years (3). The majority of studies rely on the principle of detecting a target of restricted expression in an extraneous environment. In many cases, this has taken the form of detecting tissue- or tumor-specific molecular markers in compartments of potential dissemination, such as blood, lymph nodes, and bone marrow.

CK20³ is a member of the intermediate filament protein family involved in cell structure and differentiation (4, 5). Early immunological and Northern blot studies found that CK20 expression was restricted primarily to gastrointestinal tissue, transitional cell carcinoma, and Merkel cells (4, 5, 6). Other adenocarcinomas, such as those from breast, endometrium, and lung, as well as squamous cell carcinomas were found not to express CK20. This restricted expression profile has formed the basis of many studies examining blood, lymph nodes, bone marrow, and urine for CK20 as a surrogate marker for disseminated tumor cells.

In recent years, many studies have taken advantage of the increased sensitivity of PCR assays to detect CK20 at levels previously undetectable by conventional molecular methods. However, these studies have produced many controversial findings. Whereas some studies have found associations between CK20 and stage (7, 8, 9), grade (10), and recurrence (8), others have not (11, 12). Critically, although some studies have reported specific CK20 detection (7, 13, 14, 15, 16, 17, 18, 19, 20), others have detected CK20 in blood, bone marrow, and urine from healthy donors or patients with nonmalignant disease (11, 12, 20, 21, 22, 23). A further confounding factor has been the recent detection of CK20 in samples from breast (13, 20, 24, 25), thyroid (19), endometrial (26, 27), lung (14), and pancreas (8, 14, 20) carcinoma and oral squamous cell carcinoma (16), although the reasons for this unexpected CK20 detection have not been examined. Together, these discrepancies have made it difficult to interpret the relevance of detecting circulating epithelial cells by CK20 RT-PCR.

One source of this uncertainty is that few studies have addressed the expression profile of CK20 mRNA detectable by PCR. A major reason for this has been the lack of standardized, quantitative PCR assays available for large-scale analysis. In recent years, the development of real-time, on-line fluorescence-monitoring PCR technology has promised to bring reliable and accurate PCR quantification (28). This technology monitors the entire PCR reaction by fluorescence detection, thereby allowing the beginning of the exponential phase of amplification (so-called “crossing points” or “threshold cycle”) to be measured. This reaction point is considered the most reliable point of the PCR reaction related to sample concentration (29). However, although the number of studies using this approach is increasing, few have examined the factors affecting the interpretation of results obtained by real-time PCR, and suitable approaches for quantification are yet to be established.

To clarify the expression profile and specificity of CK20 RT-PCR detection, we sought to compare CK20 levels in a collection of lymph nodes and tumors from various cancer types by quantitative PCR. To ensure correct interpretation, we developed a standardizable, quantitative real-time PCR assay and a quantification concept for high reproducibility and investigated factors influencing results obtained from these assays.

Ten colorectal and 11 breast tumors as well as 98 lymph nodes from various cancer types were collected from subjects undergoing surgery for their disease at Klinikum Kassel during 1999. The types and numbers of lymph nodes examined and their metastatic status are displayed in Table 1. Lymph nodes were assessed for metastases by routine histology after conventional H&E staining. For RNA extraction, 30–240 mg of tissue was added to 1 ml of TriPure Reagent (Roche Diagnostics, Mannheim, Germany) and homogenized with an Ultra-Turrax T25 homogenizer (IKA-Labortechnik, Staufen, Germany). RNA was extracted from the homogenate according to the manufacturer’s protocol and measured spectrophotometrically at 260 nm.

The HT29 colon carcinoma cell line (American Type Culture Collection, Manassas, VA) was grown and maintained in DMEM with 0.9% glucose, and the HL60 promyelocytic leukemia cell line (American Type Culture Collection) was grown and maintained in RPMI 1640. Both cell lines were supplemented with 10% FCS and 1% (v/v) 2 mm glutamine and incubated at 37°C in 5% carbon dioxide in a humidified chamber. Five HT29 cell suspensions (100 μl) containing 10⁴, 10³, 10², 10, and 0 cells, respectively, in 1× PBS were each mixed with 100 μl of 10⁶ HL60 cells. The entire 200 μl was loaded onto columns from the HighPure RNA Isolation Kit, and RNA was extracted according to the manufacturer’s instructions. All reagents were from Roche Diagnostics (Mannheim, Germany) unless specified.

cDNA was produced in a 20-μl reaction from 500 ng of tissue RNA or 10 μl of a 100-μl cell line RNA eluate. The other components were 20 units of AMV reverse transcriptase, 1× AMV reaction buffer, 10 mm deoxynucleotide triphosphates, and 3.2 μg of random hexamers. The reactions were incubated at 25°C for 10 min, 42°C for 1 h, and 94°C for 5 min. All reagents were from Roche Diagnostics.

Each reaction contained 2 μl of cDNA, 1× Detection Mix, and 1× DNA Master Hybridization Probes Mix (Roche Diagnostics) in a 20-μl volume. For the amplification of CK20 and the housekeeping gene, porphorynbilinogen deaminase (PBGD), separate detection mixtures were made. Each 10× Detection Mix included 5 μm each primer (forward and reverse), 2 μm each hybridization probe (fluorescein and LC-Red640), and 40 mm MgCl₂. The forward and reverse PCR primers for CK20 were placed in exons 1 and 2, respectively, amplifying a 125-bp amplicon, and those for PBGD were in exon 1 and 3, respectively, giving a 151-bp amplicon. One of the two hybridization probes for CK20 spanned the exon 1/2 boundary, whereas one for PBGD spanned the exon 2/3 boundary to ensure detection of only mRNA sequences. We cannot reveal the primer and hybridization probe sequences because of patent restrictions; however, all of the components used in this assay can be readily found in a commercially available kit (Roche Diagnostics).

In the LightCycler instrument, each reaction capillary underwent a 1-min incubation at 94°C before 50 cycles of 94°C for 0 s, 60°C for 10 s, and 72°C for 5 s. The PCR run was concluded with a 40°C incubation for 30 s. Fluorescence was monitored at the conclusion of each 60°C incubation. The PCR reactions for CK20 and PBGD were conducted in separate capillaries. The fluorescence detected in channel F2/F1 was analyzed by the LightCycler Analysis Software at the end of the run. The crossing points (beginning of the PCR exponential phase) for each reaction were determined by the Second Derivative Maximum algorithm and arithmetic baseline adjustment. For size estimation, amplified product was examined on 2% agarose gels with 1 mg/ml ethidium bromide (Roth, Karlsruhe, Germany) with a 100-bp DNA molecular weight marker XIV ladder (Roche Diagnostics).

Four methods for quantification were evaluated. The first two methods used a standard curve. Partial DNA fragments of CK20 and PBGD were amplified and cloned into separate pSPTBM21 cloning vectors (Roche Diagnostics). After linearization, the copy numbers were estimated by spectrophotometry with the assumption that 1 mol represents 6 × 10²³ molecules. Dilution series containing 10², 10³, 10⁴, and 10⁵ copies were used as PCR templates to construct a standard curve based on the relationship between the crossing point and the logarithm of copy number. Separate standard curves for CK20 and PBGD were included in each PCR run. In all cases, the correlation coefficient for the curves was −1. From this linear relationship, the copy numbers were calculated from the standard curve and crossing points of each sample. The first quantification method calculated the copy numbers of CK20 per ng of RNA. The second method calculated a CK20 relative ratio, namely the ratio of CK20 to PBGD copies in a sample relative to CK20 to PBGD copies in a calibrator sample. The calibrator was 10 ng/μl RNA from the HT29 colorectal carcinoma cell line and was included with each RT-PCR run. The final result was multiplied by 1,000,000, setting the calibrator in each run to this value. Therefore, a sample with a ratio of 100,000 had 10 times less CK20 than the calibrator.

For the third and fourth methods, the calculation of the CK20 relative ratio did not require a standard curve, but instead was a function of the PCR reaction efficiency and the crossing points of CK20 and PBGD of the unknown and calibrator samples. For the third method, the CK20 relative ratio was calculated assuming a reaction efficiency of 2, which corresponded to a perfect doubling of PCR product with each cycle according to the following equation:

where CP is the crossing point; U is the unknown sample, and C is the calibrator.

For the fourth method, because PCR product in some reactions does not precisely double with each cycle, adjusted CK20 and PBGD efficiencies for the calculation of the relative ratio were used. The efficiency was calculated as 10^−1/slope, where the slope of the line was determined from the relationship between the PCR crossing point and the logarithm of concentration. The adjusted CK20 and PBGD efficiencies were the mean efficiencies determined from seven individual tissue samples diluted over four 10-fold different concentrations. The mean efficiencies were 2.115 (± 0.029) for CK20 and 2.101 (± 0.141) for PBGD and were used for the calculation of the CK20 relative ratio of the tissue samples. With these efficiencies, the relative ratio was calculated according to the following equation:

This fourth method was selected for the quantification of tissue samples.

To evaluate the reproducibility of the quantities derived from each quantification method, the coefficient of variation, which calculates the percentage of the SD of a data series to its mean, was determined. The association between CK20 detection in noncolorectal lymph nodes and epithelial invasion determined histologically was examined using the χ² test. Differences between the mean CK20 quantities in lymph nodes from subjects with colorectal and noncolorectal cancers were tested for by a two-tailed t test. All calculations were performed using Excel 97 (Microsoft, Seattle, WA).

A protocol for the quantification of CK20 using the real-time PCR LightCycler instrument was developed. For the analysis of each sample, a single aliquot of RNA underwent a two-step RT-PCR procedure with random hexamers used for RNA priming. Aliquots from the same cDNA sample were subjected to real-time PCR for CK20 and PBGD in separate capillaries, and their crossing points were measured by the LightCycler software and used for quantification. Accompanying the samples in each RT-PCR run was a calibrator sample comprising RNA from HT29 cells, whose CK20 and PBGD crossing points were also measured for quantification.

All RT-PCR reaction reagents and concentrations used in this protocol were selected after rigorous optimization experiments. Where possible, reagents were combined in mixed solutions to minimize the number of manipulations. The LightCycler hybridization probes format was selected to provide an additional level of specificity and to allow fluorescent monitoring of the PCR kinetic reaction (29). Primers and hybridization probes for CK20 and PBGD were designed to amplify and detect short amplicons spanning exon-intron boundaries, thereby allowing reduction of cycling times and preventing the detection of genomic DNA. As a reference gene, after testing a variety of housekeeping genes (results not shown) we selected PBGD because of its relatively low expression and lack of pseudogenes. With the rapid cycling capabilities of the LightCycler instrument (30), the total run time for the reverse transcription (45 min) and PCR (30 min) steps was 75 min.

In plasmid dilution experiments, concentrations down to 1 copy of CK20 or PBGD was detectable by this assay (Fig. 1). CK20 could be detected in HT29 cells to a dilution of 10 cells mixed with 10⁶ HL60 cells, whereas no signal was detected from 10⁶ HL60 cells alone (Fig. 1). Agarose gel electrophoresis, sequencing, and LightCycler melting curve analysis of the amplified fragments showed the correct size (Fig. 1), sequence, and melting pattern, respectively (results not shown).

We assessed the reproducibility of the four methods for quantifying CK20 (see “Materials and Methods”), using a single tissue sample in a total of 27 replicates for each method, involving 3 serial dilutions, 3 separate reverse transcriptions, and 9 individual PCR assays (Table 2). The coefficient of variation from the first method, which determined absolute CK20 copy number per ng of RNA via a plasmid-derived standard curve, was 43%. For the second method, which also used a standard curve but calculated a relative CK20 ratio, the coefficient of variation was 32%. The coefficient of variation for the slope of the standard curve from the nine PCR runs was 2.3% for CK20 and 2.5% for PBGD the Y-intercept was 0.9% for CK20 and 1.3% for PBGD. For the third method, which used an algorithm for calculation of the CK20 relative ratio and assumed reaction efficiencies of 2, the coefficient of variation was 12%, whereas for the fourth method, which used adjusted individual efficiencies for the CK20 and PBGD reactions, it was 13%.

Of the tumors analyzed, 10 of 10 (100%) colorectal and 8 of 11 (73%) breast carcinomas were CK20 positive. The frequency of CK20 detection in the different types of lymph node samples is summarized in Table 1. All 9 (100%) colorectal lymph nodes examined were CK20 positive, whereas 41 of 89 (46%) noncolorectal lymph nodes had detectable CK20 mRNA. There was a strong correlation (P < 0.0001) between CK20 detection and lymph node epithelial invasion in the noncolorectal lymph nodes.

In lymph nodes with detectable CK20 mRNA, the mean relative CK20 quantity was significantly higher in colorectal than in noncolorectal specimens (P < 0.05; Fig. 2,A). Comparison of the mean CK20 relative ratio between tumors and lymph nodes showed similar relative levels according to tissue type (Fig. 2 B). Mean CK20 levels in tumor and lymph nodes were both relatively high in colorectal and both low in breast carcinoma samples.

A major obstacle to accurately evaluating the clinical relevance of detecting circulating epithelial cells by CK20 RT-PCR has been the underlying uncertainty over its expression profile and specificity. We sought to clarify this uncertainty by measuring CK20 levels in a collection of lymph nodes and tumors from subjects with a range of cancer types by quantitative PCR.

One factor contributing to the uncertainty about CK20 has been the lack of standardizable, quantitative assays amenable to large-scale analysis for its assessment. The majority of previous assays have often involved nested conventional PCR followed by detection through visualization on agarose gels (7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27), requiring long analysis times and providing only a limited qualitative analysis. Two competitive PCR assays for CK20 quantification have been described (31, 32); however, the many manipulations required for these assays make them difficult to apply to a large-scale analysis. A third assay using real-time PCR technology has also been reported; however, it detected CK20 in 100% of blood samples from healthy donors (21), creating uncertainty over its specificity.

An essential part of this study was the development of a convenient and reliable protocol for CK20 quantification using real-time PCR technology. Numerous design elements have been selected and optimized to provide controls for influencing factors and to facilitate standardizable and large-scale analysis. Rapid analysis times, including 30 min for the PCR, were made possible through the use of the LightCycler instrument, which has rapid cycling thermodynamics (30), and the selection of short amplicons. Convenience has been enhanced by combining reagents and applying a nonnested PCR procedure. Along with the LightCycler closed capillary system, reducing the number of manipulations also minimizes the risk of cross-contamination. Selection of a two-step RT-PCR procedure enables conservation of samples with multiple PCR analyses possible from a single RNA aliquot. Random hexamer priming for the reverse transcription step has been selected to provide flexibility to sample sources.

In addition to the reduction of cross-contamination, other design elements help to ensure specificity. Selection of the LightCycler hybridization probes format means that amplification and detection of CK20 and PBGD occurs only after the annealing of four sequence-specific oligonucleotides. These PCR primers and hybridization probes have been positioned to span exon-intron boundaries, reducing the risk of detecting genomic DNA. Agarose gel electrophoresis, sequencing, and LightCycler melting curve analysis of the amplified fragments consistently showed the correct size (Fig. 1), sequence, and melting pattern (results not shown) of the amplicons. The sensitivity of this assay is comparable to those reported previously (15, 21, 31), allowing the detection of 1–10 plasmid copies of CK20 and PBGD and of 10 plasmid copies of HT29 colorectal cells in a background of 10⁶ HL60 promyeloleukemic cells (Fig. 1).

As with assay design, the method of translating crossing points into a quantitative result can have a significant impact on data interpretation. After careful consideration of numerous methods, we selected a method that calculates a relative ratio of CK20 levels with an adjusted efficiency. The final result obtained by this method is a ratio of CK20 to PBGD in a sample relative to the ratio of CK20 to PBGD in a calibrator. The calculation of PBGD levels adjusts for RNA loading and integrity within the PCR run. This makes it a more relevant normalizer than spectrophotometric assessment and obviates the requirement to estimate sample concentrations prior to PCR analysis. The analysis of calibrator RNA in each run provides adjustment for inter-PCR variations and also sets a standard reference point. Additionally, efficiency values adjusted for CK20 and PBGD were used for the calculation of the relative ratio to account for differences in efficiency between the two reactions. Together these methods provide a quantification amount that allows for direct comparison of CK20 levels while adjusting for factors that may influence quantity determination.

To assess the suitability of this calculation method, we measured its reproducibility on 27 replicates of a sample and compared it with three alternative calculation methods (Table 2). The 27 determinations involved analysis over three different concentrations, three reverse transcription steps, and nine separate PCR reactions. The coefficients of variation for the current method using adjusted efficiencies and one using an assumed perfect efficiency of 2 were 13% and 12%, respectively, demonstrating the high reproducibility and the robustness of these methods to RNA loading. In this study, we favored the adjusted efficiency method because it would be more effective in the event of inefficient PCR reactions. The similarity of CK20 and PBGD reaction efficiencies makes the effect of this adjustment difficult to demonstrate; nevertheless, this result suggests that either method could be used with similar reproducibility for this CK20 assay. The variation for both approaches was lower than the coefficient of variation of 32% for the CK20 relative ratio obtained by a standard curve method. A major contributing factor to this lower reproducibility was the error introduced by the minor variations of the slope and Y-intercept of the standard curve between PCR runs. Furthermore, the coefficient of variation was 43% for a method that determines absolute copy numbers per ng of RNA via a standard curve, without factoring PBGD quantities and a calibrator ratio, which demonstrates the potential pitfalls of spectrophotometric RNA measurement and lack of calibration between RT-PCR runs.

Recent studies have found CK20 expression in samples from breast (13, 20, 24, 25), thyroid (19), endometrial (26, 27), pancreas (8, 14, 20), and oral squamous cell carcinomas (16), although these cancer types are not expected to express CK20. To explore this discrepancy, we used our quantitative PCR assay to compare CK20 expression levels in lymph nodes from subjects with colorectal cancer with those from subjects with a collection of noncolorectal cancers, namely breast, head and neck, and vulva squamous cell carcinoma and melanoma. We were surprised to find CK20 expression in lymph nodes from all of the noncancer types examined (Table 1), which suggests that CK20 expression may not be as restricted as previously thought. Four studies, including ours, have detected CK20 in samples from breast carcinoma (13, 20, 24, 25); however, this is the first report of CK20 in samples from head and neck and vulva squamous cell carcinoma and melanoma.

We examined CK20 expression levels in lymph nodes from subjects with noncolorectal cancer and observed extremely low levels relative to the expression in lymph nodes from subjects with colorectal cancer. The difference in CK20 levels between the highest quantity in a colorectal cancer lymph node (32,997,451) and the lowest quantity in a breast cancer lymph node (45) was ∼700,000-fold (Fig. 2,A). When we compared levels between lymph nodes and tumors from subjects with colorectal and breast cancer, we found similar quantitative differences (Fig. 2 B), which suggests that this result was not an artifact of lymph node analysis. CK20 levels were high in both lymph nodes and tumors from subjects with colorectal cancer and low in both lymph nodes and tumors from subjects with breast cancer. This result provides proof of large quantitative differences in CK20 expression detectable by quantitative PCR and may help to provide a possible explanation for discrepant observations of the CK20 expression profile. With these quantitative differences, less sensitive assays would be able to detect expression only in samples from colorectal cancers, whereas those more sensitive would detect CK20 in a larger range. To our knowledge, the expression profile of CK20 has been defined primarily by Northern blot and immunohistochemical analyses (5, 6). This result demonstrates that assumptions about expression profiles obtained by alternative methods should be interpreted with caution.

One possible explanation for CK20 detection in these cancer types is nonspecificity; however, we believe that this is highly unlikely. Along with assay design elements for specificity, a strong association was observed between CK20 detection and lymph node metastases demonstrated by histology (Table 1). Of the lymph nodes from noncolorectal cancer types, CK20 was detected in 87% (39 of 45) of cancer metastases and only 5% (2 of 44) of lymph nodes without histological evidence of metastatic infiltration.

The strong association between CK20 detection and metastatic invasion also suggests that detection of CK20 in lymph nodes may provide clinically relevant information in these tumor types. However, the small numbers of samples for each cancer type in this series prevents analysis of their association with other clinical parameters. Other studies in noninfiltrated lymph nodes from breast (13) and colorectal (18, 33, 34) cancer suggest that CK20 detection may help to identify cancer cell dissemination at a more sensitive level than conventional means. Rosenberg et al. (34) recently observed a nonsignificant trend between CK20 detection in lymph nodes from subjects with colorectal cancer and survival in a small sample series with a short follow-up time. Thus, further studies are clearly warranted to investigate the potential role of CK20 detection in lymph nodes.

In summary, we have described the development of a standardizable, quantitative CK20 PCR assay and have used this assay to characterize epithelial-specific, low-level CK20 expression in cancer types not expected to express CK20. This result helps to clarify uncertainty over the expression profile of CK20 and demonstrates the potential new information that quantitative PCR analysis may provide. Similar uncertainty surrounds the specificity of CK20 in blood, bone marrow, and urine from healthy donors. The potential benefit of a quantitative PCR analysis of these sample types awaits further analysis.