Heterogeneous Expression of MAGE-A Genes in Occult Disseminated Tumor Cells: A Novel Multimarker Reverse Transcription-Polymerase Chain Reaction for Diagnosis of Micrometastatic Disease¹

The most common types of human cancer bear a considerable risk of systemic recurrence even when early diagnosed and despite curative resection of the primary tumor. Those patients with resected localized cancer who finally progress to lethal metastatic disease would be eligible for adjuvant therapy if reliable prognostic parameters were available to predict individual clinical outcomes. Therefore, systemically disseminated tumor cells have become the subject of intensive research as the presumed seminal precursors of later distant metastasis, which may persist in a state of dormancy for many years. Immunocytochemical techniques based on monoclonal antibodies against cytokeratins and other differentiation markers have been applied to identify rare, disseminated cells of epithelial tumors in bone marrow aspirates of carcinoma patients. The presence of cytokeratin-positive cells correlates with a significantly higher risk of future distant metastasis, as shown for patients with breast cancer (1) and other carcinomas (reviewed in Ref. 2). Nevertheless, the microscopic, preferably double-blinded examination of cytocentrifuged bone marrow cells is laborious and observer dependent, thus complicating routine use.

As an alternative, mRNA transcribed in rare, disseminated tumor cells from genes encoding differentiation markers or tumor-associated antigens could be detected in blood, bone marrow, or lymph nodes by sensitive RT-PCR⁶ assays in various types of cancer (3). However, low-level gene expression in nonmalignant cells appears to limit the specificity of most candidate PCR markers, with only a few exceptions, including PSA in prostate cancer (4). As to the sensitivity of detecting rare, disseminated tumor cells with molecular techniques, approaches that rely on a single marker are particularly susceptible to the expression loss or down-regulation frequently connected with tumor cell heterogeneity (5, 6). Therefore, several different multimarker assays have recently been developed for the sensitive detection of tumor cells in the peripheral blood (7, 8, 9) as well as in lymph nodes (10, 11) of patients with various malignancies.

In the study reported here, we made use of the exceptionally restricted expression of the MAGE-A family to develop a highly sensitive and tumor-specific multimarker nested RT-PCR comprising independent amplification of MAGE-1, -2, -3/6, -4, and -12, respectively. Expression of these MAGE genes has been found in many different types of tumors (12, 13), but not in normal adult tissues with the exception of testicular germ cells (14). In particular, frequent expression of MAGE genes had been reported in primary colorectal (15), lung (16), and breast carcinomas (17). The MAGE genes encode proteins of yet unknown physiological function that were originally described as tumor antigens recognized by specific T cells in malignant melanoma (18). Therefore, MAGE gene products have become promising target antigens for tumor vaccination (19, 20). Despite high homology among the MAGE genes, we could identify unique pairs of PCR primers that exclusively amplified individual members of the gene family with enhanced sensitivity. The ability of the MAGE assay to detect rare, disseminated tumor cells was analyzed with bone marrow aspirates from 99 patients with breast, lung, colorectal, and prostate cancer and compared with that of cytokeratin-based immunocytochemistry. In the particular case of prostate cancer, the bone marrow aspirates were also subjected to an established PSA RT-PCR. The data show frequent and heterogeneous MAGE expression considerably overlapping with the presence of rare, cytokeratin-positive tumor cells and concordant with PSA mRNA expression in prostate cancer. Moreover, MAGE-positive bone marrow samples from a small group of seven sarcoma patients demonstrate the relevance of our multimarker RT-PCR in nonepithelial tumors. In sharp contrast to the results obtained with specimens from cancer patients, we could not find any MAGE transcripts in bone marrow aspirates from 30 healthy donors. For a more in-depth characterization of the multimarker MAGE RT-PCR, we chose a cohort of 30 patients with clinically localized prostate cancer because of the opportunity to use PSA mRNA amplification as one of the few truly specific PCR standards (4) with established prognostic significance (21, 22). The analysis of double-sided bone marrow aspirates showed that the multimarker MAGE assay detects disseminated tumor cells in twice as many patients as the PSA RT-PCR or cytokeratin-based immunocytochemistry. Moreover, the expression of MAGE transcripts in bone marrow correlated significantly with established risk factors predicting metastatic relapse.

After receiving written consent, we obtained unilateral pelvic bone marrow aspirates from 33 lung cancer patients, 14 patients with colorectal cancer, and 7 sarcoma patients as well as bilateral pelvic bone marrow aspirates from 30 patients with clinically localized prostate cancer (Wiesbaden cohort). Formalin-fixed, paraffin-embedded needle biopsies of the primary tumors were available from 17 and peripheral blood from 12 prostate cancer patients of the Wiesbaden cohort. Peripheral blood was also obtained from 12 other patients with metastasized prostate cancer. Moreover, we had access to a tissue bank at the Section of Urological Surgery/Department of Surgery at the University of Chicago that contains frozen specimens of benign prostate hyperplasia and primary prostate cancer from patients who had undergone radical prostatectomy. From a previous study, cDNA was available from conventionally processed bone marrow aspirates from 22 breast cancer patients (4).

Bone marrow aspirates from 30 healthy allogeneic bone marrow donors and peripheral blood samples from 20 healthy volunteers served as negative controls.

The cell lines Mz2-Mel and LB23-SAR were cultured in HEPES-buffered DMEM supplemented with 10% FCS, 0.55 mm l-arginine, 0.24 mm l-asparagine, 1.5 mm l-glutamine, 1 mg/ml d-glucose, 100 IU/ml penicillin, 100 mg/ml streptomycin, 10 μg/ml transferrin, and 2.83 mg/ml insulin. The other cell lines were grown in RPMI 1640 supplemented with 10% FCS, 2 mm l-glutamine, 100 IU/ml penicillin, 100 mg/ml streptomycin, 10 μg/ml transferrin, and 2.83 mg/ml insulin. Cells were harvested from subconfluent cultures into PBS containing 0.05% trypsin-0.02% EDTA.

One ml of each native bone marrow aspirate or peripheral blood sample was immediately mixed with 5 ml of nucleic acid extraction buffer [4 m guanidine isothiocyanate, 0.5% sarcosyl (N-laurylsarcosine sodium salt), 25 mm sodium citrate (pH 7.0), 0.7% 2-mercaptoethanol] and stored at −20°C until needed. The RNA was isolated by ultracentrifugation in SW40 centrifuge tubes at 35,000 rpm for 18 h at 15°C, using a 5-ml cushion of CsTFA solution (Pharmacia, Freiburg, Germany) containing 0.1 m EDTA (pH 7.0). After centrifugation, the upper layer of nucleic acid extraction buffer/bone marrow or blood mixture and the lower layer of CsTFA solution were removed by aspiration and discarded. The RNA pellet was dissolved in 300 μl of diethyl pyrocarbonate-treated water and reextracted sequentially with 300 μl of a phenol-chloroform-isoamyl alcohol mixture (25:24:1, v/v/v) and 300 μl of chloroform. After precipitation with 300 μl of isopropanol, 40 μl of 3 m sodium acetate (pH 5.0), and 20 μg of glycogen at −20°C overnight, the RNA pellet was dissolved in 5 μl of diethyl pyrocarbonate-treated water. cDNA was generated with a first-strand cDNA synthesis kit (Life Technologies, Eggenstein, Germany) in a final volume of 10 μl containing 50 mm Tris (pH 8.3), 75 mm KCl, 3 mm MgCl₂, 10 mm DTT, 0.5 mm total deoxynucleotide triphosphate, 1.6 μg of random hexamer primers (Roche), and 100 units of Superscript II (Life Technologies).

Alternatively, instead of using ultracentrifugation with a CsTFA gradient, we prepared RNA according to the method of Chomczynski and Sacchi (23) from whole blood or bone marrow lysed directly in nucleic acid extraction buffer. RNA resulting from this alternative protocol was reverse-transcribed into cDNA by use of an equimolar mixture of the outer antisense primers of MAGE-1, -2, -3/6, -4, and -12 and the histone antisense primer (Table 1), rather than random hexanucleotides.

Standard RNA preparations from cell lines, fresh frozen tissues, and mononuclear cells of blood or bone marrow were obtained by Ficoll density centrifugation, according to the method of Chomczynski and Sacchi (23), followed by random-primed cDNA synthesis. Total RNA from formalin-fixed, paraffin-embedded needle biopsies of primary prostate carcinomas was isolated from 25-μm sections according to the method of Stanta et al. (24). The resulting RNA was reverse-transcribed into cDNA via random priming.

For the first round of PCR, 10-μl reactions containing 1 μl of random-primed cDNA, 1 μl of 10× PCR buffer [100 mm Tris (pH 8.3), 500 mm KCl, 10 mm MgCl₂, 1 mm deoxynucleotide triphosphate], 0.4 μm each of the outer MAGE primers, 5 μg of BSA (Roche), and 0.6 units of Taq DNA Polymerase (Roche) were run according to the following cycle profile: denaturation at 94°C for 6 min, annealing at 60°C for 30 s, and extension at 72°C for 2 min for the first cycle; denaturation at 93°C for 40 s, annealing at 60°C for 30 s, and extension at 72°C for 20 s for 14 cycles; denaturation at 93°C for 40 s, annealing at 60°C for 30 s, and extension at 72°C for 30 s for 50 cycles; terminal extension at 72°C for 2 min. For the second round of PCR, 1 μl of the first reaction, rather than cDNA, was transferred into a new reaction mixture as described above except that the inner MAGE primers replaced the corresponding outer primers. Thirty more cycles were performed with an initial denaturation at 93°C for 40 s, annealing at 58°C for 30 s, and elongation at 72°C for 30 s, followed by a final extension at 72°C for 2 min. For the PSA RT-PCR, random-primed cDNA was used with primers specific for exons 2 and 3 common to the three PSA variants as described previously (4). The presence of cDNA was monitored by a control PCR amplifying p53. cDNA from specifically primed reverse transcriptions was also amplified by nested MAGE PCR with a total of 95 cycles, except that amplification of histone replaced p53 in the control PCR. Adequate precautions to prevent cross-contamination and negative control reactions that included mock preparations were performed routinely. PCR products were separated on a 1.8% agarose gel and stained with ethidium bromide. We confirmed negative as well as positive findings by repeating the assay with a second aliquot of each original total RNA sample. Reproducibility was almost 100% in negative cases and >90% in positive cases. The specificity of the amplification product obtained for each MAGE gene was multiply monitored by DNA sequencing.

Mononuclear cells were isolated by density gradient centrifugation with Ficoll-Hypaque (Pharmacia) at 400 × g for 30 min and were deposited onto glass slides by cytocentrifugation at 150 × g for 3 min. To detect tumor cells in bone marrow, we used the monoclonal antibody A45-B/B3 (IgG1; Micromet, Munich, Germany), which detects an epitope on a variety of cytokeratin components, including cytokeratin 8, 18, and 19 (25). Bound antibody was detected with the alkaline phosphatase-antialkaline phosphatase technique combined with new fuchsin (Sena, Heidelberg, Germany). For each bone marrow aspirate, two slides containing a total of 10⁶ mononuclear cells were examined.

Statistical analysis was performed with JMP (version 3.1.6; SAS Inc.). Cross-tabulation was analyzed with Fisher’s exact test (two-tailed P).

A highly sensitive assay for the detection of rare, disseminated tumor cells based on nested RT-PCR amplification of mRNA transcribed from six members of the MAGE-A gene family was established in two rounds of optimization. In the first round, five optimal double pairs of PCR primers, each specific for the transcripts of MAGE-1, -2, -3/6, -4, and -12, respectively, were experimentally identified from a total of almost 800 preselected primer combinations. To avoid cross-amplification among the different MAGE gene transcripts because of the high level of homology, preselection of PCR primers was limited to those pairs of oligonucleotides differing at least by the terminal nucleotide of one 3′ end from the cDNA sequence of each of the other MAGE genes. Because MAGE-3 and -6 exhibit 99% homology to each other, the corresponding primer pairs were designed to amplify the cDNA of both genes. Amplification of genomic sequences was prevented by use of (a) primers localized in different exons or (b) primers spanning different neighboring exons, thus restricting hybridization to cDNA only. Decreasing numbers of cells from the cell lines Mz2-Mel and LB23-SAR, known for their strong expression of MAGE-1, -2, -3/6, and -12 or MAGE-4, respectively, were mixed with a constant number of mononuclear cells from the peripheral blood of healthy donors (peripheral blood mononuclear cells), followed by standard preparation of total RNA and cDNA synthesis. In Table 1, those two pairs of (outer and inner) primers are listed for each MAGE gene that yielded specific amplification products with the lowest relative amounts of MAGE cDNA in a one-step PCR comprising 65 cycles. Control preparations of cDNA from unspiked peripheral blood mononuclear cells were consistently negative with the same PCR. In general, cDNA primed by random hexanucleotides gave higher amplification yields than did oligo(dT)-primed cDNA.

In a first attempt to examine the relevance of a multimarker MAGE RT-PCR in different types of cancer, we amplified MAGE-1, -2, -3/6, -4, and -12 from 28 cell lines of epithelial, mesenchymal, hematopoietic, and neuroectodermal origin by running 35 PCR cycles with the respective outer primers. As shown in Table 2, highly heterogeneous expression patterns were found in different cell lines. However, 27 of 28 cell lines exhibited expression of at least one MAGE gene corresponding to an overall positivity rate of 96%. Only one cell line (A-498) derived from a renal cell carcinoma was completely negative. As shown in Fig. 1, for each of the selected MAGE genes, the best double pair of outer and inner PCR primers was combined in a nested PCR comprising a total of 95 cycles. To establish the feasibility of detecting rare, disseminated tumor cells through the sensitive analysis of MAGE gene expression, we applied the five different nested MAGE PCRs to cDNA from bone marrow aspirates from 22 breast cancer patients, which had been shown in a previous study to contain cytokeratin-positive epithelial tumor cells as detected by immunocytochemistry (4). The cDNA had been synthesized by reverse transcription of total RNA from mononuclear bone marrow cells obtained by centrifugation on a Ficoll density gradient. Three of 22 probes were double positive for MAGE-1 and -2, one probe was double positive for MAGE-3/6 and -4, and four cDNA preparations showed PCR signals for MAGE-12 only. Fourteen of 22 probes exhibited no MAGE signals as did control cDNA from bone marrow mononuclear cells of tumor-free patients with bone fractures. Thus, as a matter of principle, detection of rare, disseminated tumor cells proved to be feasible with the nested multimarker MAGE PCR. However, the method’s recovery of signals from cytokeratin-positive tumor cells was still limited to 8 of 22 bone marrow aspirates typed positive by immunocytochemistry, suggesting that further optimization of parameters may substantially influence the performance of the assay.

Accordingly, the second round of optimization was focused on preventing the loss of rare, disseminated tumor cells and RNA degradation during sample transport and preparation. Usually RNA from blood or bone marrow is prepared from the fraction of mononuclear cells obtained by centrifugation on a Ficoll density gradient. Thus, the time from blood or bone marrow aspiration to full protection of the RNA in a highly denaturing extraction buffer may range from 30 min, when the samples can be processed immediately, to several hours or even an entire day, when transport of the samples is required. We therefore established protocols by which high-quality RNA can be prepared from whole blood or bone marrow lysed immediately after aspiration and prior to transport in a denaturing extraction buffer, providing full RNA protection. As a further advantage of these protocols, cell separation or elimination steps, which may cause the loss of rare, disseminated tumor cells prior to RNA preparation, are no longer required. These hypotheses could be experimentally confirmed by mixing decreasing numbers of Mz2-Mel or LB23-SAR cells into 1 ml of peripheral blood from a healthy donor, followed by (a) density centrifugation on Ficoll-Hypaque, (b) ammonium chloride-mediated lysis of erythrocytes by Coulter Q-Prep, or (c) direct lysis of the unmodified sample in denaturing extraction buffer prior to preparation of RNA, respectively. The subsequent analysis with the nested multimarker MAGE RT-PCR revealed a sensitivity of 1–10 tumor cells detectable in 1 ml of blood when the RNA was prepared after direct lysis of the whole blood sample in denaturing extraction buffer (Fig. 2). In contrast, 500-1000 tumor cells in 1 ml of blood were required to obtain MAGE amplification with the same nested RT-PCR when density centrifugation or erythrocyte lysis preceded the preparation of RNA. Peripheral blood from 20 healthy donors prepared through direct lysis in denaturing extraction buffer remained completely negative in the nested multimarker MAGE RT-PCR, despite the enhanced overall sensitivity of the assay. Unless stated otherwise, this optimized protocol was also used for the investigations described below.

At present, cytokeratin-based immunocytochemistry of cytocentrifuged mononuclear cells from bone marrow is used as standard method for the diagnosis of systemic tumor cell spread in patients with epithelial tumors. To compare the nested multimarker MAGE RT-PCR with this standard assay, bone marrow aspirates from 47 lung and colorectal cancer patients staged free of distant metastasis (M₀) were examined in parallel with both tests. As an essential prerequisite, however, even minimal expression of MAGE genes in normal bone marrow had to be excluded. We therefore analyzed samples from 30 healthy individuals who donated bone marrow for allogeneic transplantation. Remarkably, despite the high level of proliferation found in bone marrow, none of the 30 probes revealed any expression of MAGE-1, -2, -3/6, -4, or -12, as tested with the highly sensitive nested RT-PCR. In contrast, unilateral bone marrow aspirates from 33 patients with lung cancer showed heterogeneous expression of at least one MAGE gene in 11 cases (33%), i.e., in 6 of 20 squamous cell carcinomas, 3 of 10 adenocarcinomas, 1 of 2 large cell carcinomas, and in the only tested case of small cell lung cancer (Table 3 and Fig. 3). Cytokeratin-positive epithelial cells were found in 8 of 33 lung cancer patients (24%). The analysis of unilateral bone marrow aspirates from 14 patients with colorectal cancer revealed heterogeneous MAGE expression in 9 (64%) and cytokeratin-positive epithelial cells in 4 cases (29%). Regarding colon cancer (n = 10), cytokeratin-positive epithelial cells were found in three patients, of whom two also expressed MAGE-1 or MAGE-1 and -4, respectively. Three additional colon cancer patients, who were negative in cytokeratin-based immunocytochemistry, exhibited expression of MAGE-1 only. The four remaining patients with colon cancer were negative with both tests. In patients with rectal cancer (n = 4), the only one with cytokeratin-positive epithelial cells was single-positive for MAGE-1. Of the remaining three rectal cancer patients, two showed expression of MAGE-1 and -3/6 and one expressed MAGE-3/6 only. The presence of cytokeratin-positive cells considerably overlaps with MAGE expression in both lung and colorectal cancer. However, the nested multimarker MAGE RT-PCR appears to detect rare, disseminated tumor cells more frequently than does cytokeratin-based immunocytochemistry. To investigate the relevance of the nested multimarker MAGE RT-PCR in nonepithelial tumors, we analyzed unilateral bone marrow aspirates from a small group of patients with different sarcomas. MAGE expression could be detected in four of seven cases, indicating the presence of rare, disseminated tumor cells undetectable by cytokeratin-based immunocytochemistry because of their mesenchymal differentiation.

Because amplification of PSA mRNA offered the opportunity to compare the multimarker MAGE RT-PCR with one of the few established RT-PCR assays that do not lose specificity with increasing sensitivity for detecting rare, disseminated tumor cells in blood or bone marrow, prostate cancer was chosen for a more in-depth characterization of the new method.

Bilateral bone marrow aspirates from 30 patients with clinically localized prostate cancer staged free of manifest distant metastasis (designated the Wiesbaden cohort) were analyzed in parallel with the multimarker MAGE RT-PCR, the established PSA RT-PCR, and cytokeratin-based immunocytochemistry. Combining the data obtained by RT-PCR from both iliac crest aspirates, MAGE-1 mRNA was detected in 12 patients (40%), MAGE-2 mRNA in 3 patients (10%), MAGE-3/6 mRNA in 5 patients (17%), and MAGE-4 and -12 mRNA in 6 patients each (20%), overlapping to various degrees (Table 4 and Fig. 4). At least one of the six analyzed MAGE mRNA species was found in 18 of 30 patients (60%). In contrast, at least one of two bone marrow aspirates showed detectable expression of PSA mRNA in eight patients (27%) and cytokeratin-positive tumor cells in nine patients (30%). As shown in Fig. 5, the results of all three independent assays overlapped considerably. In particular, it is noteworthy that all eight PSA-positive patients were included in the MAGE-positive group.

To correlate the bone marrow findings with MAGE expression in primary prostate cancer, we analyzed 17 available needle biopsies, all histologically confirmed to contain carcinoma tissue, from prostate cancer patients of the Wiesbaden cohort with the same nested multimarker MAGE RT-PCR. Overall, 13 of 17 biopsies were positive for at least one MAGE gene (77%). Table 5 A depicts the expression of the six tested MAGE genes for each patient. MAGE gene expression in bone marrow and corresponding primary tumors showed considerable concordance. Particularly, no MAGE expression was detected in the bone marrow of those patients with a MAGE-negative primary tumor. Moreover, with only a few exceptions, those MAGE transcripts detected in bone marrow were most often found to also be expressed in the respective primary tumor. Because the positivity rate of 77% appeared rather high compared with previously published data on the expression of MAGE-1 and -3 in primary prostate cancer (13), specimens from primary tumors of a second cohort of prostate cancer patients were tested in another laboratory at the Department of Urology/University of Chicago (Chicago cohort).

Random tissue samples from primary prostate carcinomas indeed showed rather low MAGE positivity rates (17% for at least one MAGE gene), similar to the rates found by van den Eynde and van der Bruggen (13). Microscopic examination of those random samples, however, showed that only a small percentage actually contained carcinoma tissue. Therefore, we analyzed samples from 21 prostate carcinomas that definitely contained cancerous tissue as confirmed by microscopic examination, revealing a PCR positivity rate of 76% for at least one expressed MAGE gene in the Chicago cohort (Table 5 B). Thus, MAGE expression analysis of histologically confirmed samples from primary prostate carcinomas led to almost identical results in both cohorts. As negative controls, tissue samples from 20 cases with benign prostate hyperplasia were investigated and found to be free of any MAGE transcript, as tested with our sensitive multimarker RT-PCR assay.

In a first attempt to assess the prognostic significance of the detected MAGE transcripts, we tried to correlate the bone marrow findings in prostate cancer patients with established risk factors of developing systemic disease. Fuks et al. (26) statistically demonstrated that local failure of successfully treated primary prostate cancer strongly predicts later distant metastasis because it appears to build a highly efficient secondary source of systemic tumor cell spread. Regarding the primary tumor, the following independent risk factors of distant metastasis have been identified: A tumor size ≥T₃, an initial serum PSA >20 ng/ml, and poor differentiation, i.e., G₃ grading corresponding to a Gleason score of 8–10 (27). Accordingly, the combination of these three parameters was regarded as the highest risk of systemic tumor cell dissemination that could be defined in patients newly diagnosed with clinically localized prostate cancer (28).

Among the 30 patients, 8 had a high risk of later distant recurrence because of a local relapse at the time of bone marrow aspiration, whereas 3 patients presented with primary tumors combining T₃ tumor size and poor differentiation (Gleason score of 8–10) with an initial serum PSA >20 ng/ml. Thus, 11 of the 30 patients could be identified as having an exceptional risk of systemic tumor cell dissemination. Ten of these 11 high-risk patients (91%) exhibited MAGE transcripts. In contrast, of the remaining 19 patients lacking these clinical prognostic indicators of future distant metastasis, 8 (42%) had a positive MAGE PCR (P = 0.02, univariate analysis by Fisher’s exact test; Table 6).

To obtain further evidence for the prognostic significance of the multimarker MAGE RT-PCR, we compared the frequency of MAGE expression in peripheral blood of patients from the Wiesbaden cohort with clinically localized prostate cancer with that found in another group of patients with metastatic disease: only 3 of 12 prostate cancer patients staged free of distant metastasis were found to be MAGE positive (25%), whereas MAGE transcripts were detectable in 10 of 12 patients with metastatic prostate cancer (83%; Fig. 6), thus indicating an increase in the MAGE positivity rate with the stage of disease (P = 0.0058, Fisher’s exact test; Table 7).

The present work describes the design of a novel multimarker nested RT-PCR capable of detecting the expression of individual members of the MAGE-A gene family, MAGE-1, -2, -3/6, -4, and -12, by rare, disseminated tumor cells in blood and bone marrow. The main goals of this study were (a) to develop a sensitive assay capable of detecting rare, disseminated tumor cells in patients with different types of cancer; (b) to avoid the usual loss of specificity seen with the overwhelming majority of current PCR markers when detection levels reach the ultrasensitive range; and (c) to obtain a refined expression profile of rare, disseminated tumor cells in individual patients, comprising multiple markers that may serve as target antigens for adjuvant immunotherapy.

Selected members of the MAGE-A gene family appear to be ideal candidates for meeting these requirements because they are frequently expressed in many tumors almost irrespective of the histological origin but are completely silent in normal adult tissue with the single exception of testicular germ cells (14). Several MAGE gene products have been identified as promising target antigens for tumor immunotherapy and have already been used in cancer vaccination trials (19, 20). The search for PCR primers specific for individual MAGE genes was difficult because of the high homology among the different MAGE genes. Others have tried to solve this problem by use of consensus oligonucleotides that coamplify cDNA of several different MAGE genes (29). This approach, however, was not feasible in our case because it involves a dramatic loss of information on the MAGE expression profile of rare, disseminated tumor cells in individual patients, which may be essential for optimal vaccine design. Moreover, the prognostic impact of single members of the MAGE gene family may vary with different types of malignancies.

By systematic selection for high sensitivity of those specific primer pairs unable to amplify genomic DNA and mismatching, at least by the terminal nucleotide of one 3′ end, with the homologous sequence sections of the other MAGE genes, we could identify double pairs of oligonucleotides optimally suited for nested cDNA amplification of MAGE-1, -2, -3/6, -4, and -12, respectively. Although we could exclude cross-amplification of other MAGE genes by sequence analysis, the selected PCR primers certainly tend to cross-hybridize with other members of the MAGE family, thus reducing the amplification efficiency per PCR cycle. This may explain the high number of PCR cycles required for the detection of rare MAGE transcripts. However, increased numbers of PCR cycles, although not uncommon with sensitive RT-PCRs as reported by others (30), were accompanied by nonspecific amplification products when cDNA synthesis was random primed (Figs. 3 and 5). Nevertheless, the MAGE-specific bands could be clearly distinguished from nonspecific PCR signals, which almost disappeared with specific priming of cDNA synthesis (Fig. 6).

Overall sensitivity was clearly affected by sample handling; immediate fixation of blood or bone marrow samples as well as strict adherence to the established protocol to prevent loss of rare tumor cells was of critical importance. With all these precautions, we could detect 1–10 tumor cells/ml of blood for each of the five different nested MAGE RT-PCRs. Even at this high sensitivity, blood and bone marrow samples from healthy donors remained completely negative, confirming the exceptionally tight control of MAGE gene expression in normal tissue. Accordingly, the increased sensitivity of the MAGE assay was not accompanied by a concomitant loss in specificity, which frequently limits the utility of PCR assays amplifying differentiation markers [e.g., CK19 (31) and PSM (4, 32)] or other tumor-associated genes [e.g., CEA (33)]. In this context, it is noteworthy that the activation of MAGE expression is not controlled by appropriate transcription factors, but may be primarily attributable to the demethylation responsible for promoter activation of MAGE genes (34). As De Smet et al. (35) have reported, Ets binding sites in the MAGE-1 gene become accessible after demethylation.

In sharp contrast to samples from healthy subjects, MAGE was expressed frequently in bone marrow aspirates from 99 patients with breast, lung, colorectal, and prostate cancer as well as in bone marrow from 7 sarcoma patients. These findings confirmed the broad relevance of the multimarker MAGE RT-PCR in detecting rare, disseminated tumor cells of many different types of cancer. The expression patterns of the different MAGE genes were highly heterogeneous in both cell lines and bone marrow aspirates. Heterogeneous MAGE gene expression may reflect the heterogeneity of disseminated tumor cells in bone marrow, which would not be detected by a single-marker assay. Differences in the expression patterns found in aspirates from the right and left iliac crest (Table 4) support this possibility. On the other hand, single disseminated tumor cells may also express multiple MAGE genes, as seen with many tumor cell lines. Accordingly, identical expression patterns in both bone marrow aspirates, as found, e.g., in prostate cancer patient 3 (Table 4), may best be explained by the dissemination of closely related tumor cells expressing the same MAGE genes.

Because of its composition of pure hematopoietic and mesenchymal cells, bone marrow provides the perfect background for staining rare, disseminated cells from epithelial tumors with antibodies against epithelial differentiation antigens such as cytokeratins. Comparison of the multimarker MAGE RT-PCR with the present standard method of detecting rare, disseminated carcinoma cells in bone marrow by cytokeratin-based immunocytochemistry revealed a considerable overlap of positive results. When analyzing the data in lung cancer patients, it is noteworthy that cytokeratin-based immunocytochemistry detected rare tumor cells only in patients with squamous cell carcinoma, whereas the MAGE assay also yielded positive results in patients with adenocarcinoma, large cell carcinoma, and small cell lung cancer. In patients with localized colorectal cancer, the multimarker MAGE RT-PCR also was superior. Interestingly, MAGE was expressed in the bone marrow of all four patients with rectum carcinoma, whereas the MAGE positivity rate in colon cancer was 50%. This may reflect the higher risk of distant metastasis in rectum cancer because of its additional hematogenous route of tumor cell spread, circumventing the liver. Overall, the high representation of MAGE-1 was conspicuous in colorectal cancer and may be characteristic for this particular type of tumor. The broad applicability of the multimarker MAGE RT-PCR in many different types of cancer could be demonstrated most impressively with MAGE-positive bone marrow samples from a small group of sarcoma patients. Thus, disseminated cells from nonepithelial tumors, which are definitely not accessible to cytokeratin-based immunocytochemistry, were detectable by the MAGE assay.

A more in-depth characterization of the multimarker MAGE RT-PCR was carried out with 30 prostate cancer patients with clinically localized disease (Wiesbaden cohort) because of the opportunity of using PSA mRNA amplification as one of the few truly specific PCR standards (4) with established prognostic significance (21, 22). The results of simultaneous analysis of double-sided bone marrow aspirates by the multimarker MAGE RT-PCR, the established PSA RT-PCR, and cytokeratin-based immunocytochemistry overlapped considerably. The MAGE assay appears to detect disseminated tumor cells in twice as many patients as the two other tests. Importantly, all PSA-positive patients were included in the MAGE-positive group. Although this adds further evidence that the MAGE transcripts are indeed derived from systemically spread prostate cancer cells, it raises a question as to only 8 of 18 MAGE-positive cases were also positive in the PSA RT-PCR. Although we reproducibly ruled out illegitimate transcription of MAGE genes in control bone marrow aspirates, there is evidence that PSA as a differentiation antigen may be lost in more undifferentiated tumor cells (4). And indeed, PSA-negative epithelial tumor cells could be found by immunocytochemistry in the bone marrow of prostate cancer patients by a double-staining approach (36). On the other hand, the multimarker MAGE RT-PCR and the PSA RT-PCR were negative with both bone marrow samples of prostate cancer patients 13 and 14, although the immunocytochemical test revealed the presence of cytokeratin-positive cells. Because of the double-sided presence of cytokeratin-positive cells in bone marrow of prostate cancer patient 13 and the absence of MAGE expression in the corresponding primary tumor (Table 5A), the MAGE test may have yielded a false-negative result in this particular case, as presumably did the PSA RT-PCR. However, discrepancies between PCR and immunocytochemistry may be also explained by the fact that different bone marrow aliquots are analyzed in these two tests, leaving the possibility of a single rare tumor cell present in one aliquot but absent from the other. Alternatively, cytokeratin-based immunocytochemistry can yield false-positive results, e.g., because of cytokeratin expression in hematopoietic cells, which can rarely become detectable (2).

The patterns of MAGE gene expression in the available needle biopsies of primary tumors from the Wiesbaden cohort of prostate cancer patients were considerably concordant with those found in the respective bone marrow aspirates. The expression of at least one MAGE gene was observed in 77% of those primary prostate carcinomas. Although RNA preparations from formalin-fixed tissues usually suffer from rapid RNA degradation, we could nevertheless isolate RNA of considerable integrity, as demonstrated by the successful amplification of p53. This may be explained by the small size of the needle biopsies, which allows faster RNA protection during formalin fixation because of the short diffusion distances.

Because others had found an expression frequency of 15% for MAGE-1 and -3 in primary prostate cancer with other primers and different PCR conditions (13), primary tumors collected in the United States from another cohort of prostate cancer patients (Chicago cohort) were independently examined in another laboratory with our PCR assay. In this patient group, although certainly of different genetic background (37), we found an overall frequency of 76% MAGE-positive primary tumors (Table 5B). The only significant difference we detected between the Wiesbaden and the Chicago cohorts was the increased expression frequency of MAGE-1 in the former group. Because 70% of the MAGE-1-positive primary tumors in this cohort were confirmed by expression of at least one additional MAGE gene, the high positivity rate of MAGE-1, although varying to some extent, most likely represents a general feature of prostate cancer. For the study of MAGE expression in biopsies, it turned out to be critical that the presence of tumor cells in the individual test sample was histologically confirmed to reduce the frequency of false-negative results. We therefore assume that the overall positivity rate of MAGE gene expression in primary prostate cancer would be even higher than that obtained here with single biopsies if multiple biopsies of each primary tumor were tested. Importantly, 20 benign prostate hyperplasia samples were found to be MAGE negative, again underlining the cancer-restricted expression of the MAGE genes.

In a first attempt to assess the clinical relevance of the multimarker MAGE RT-PCR, we investigated whether established prognostic indicators of distant tumor recurrence in patients with localized prostate cancer were correlated with MAGE expression in bone marrow. Patients carrying an exceptionally high risk of distant metastasis at the time of marrow aspiration, because of either a local relapse (26, 38) or a primary tumor with multiple risk factors (28, 39), were more often MAGE positive than the remaining patients (P = 0.02, univariate analysis by Fisher’s exact test). Although the MAGE positivity rate of 42% in the latter group appeared to be rather high, it was not as surprising because we included several patients with single risk factors, such as T₃ tumor size. Further evidence for the prognostic impact of the multimarker MAGE RT-PCR came from the analysis of peripheral blood, which showed that MAGE expression was distinctly more frequent in 12 patients with metastatic prostate cancer (83%) than in 12 prostate cancer patients of the Wiesbaden cohort (25%) who were free of distant metastasis (P = 0.0058, Fisher’s exact test). We are presently initiating prospective clinical studies addressing the prognostic relevance of the MAGE marker in different types of cancer.

Because MAGE antigens can induce autologous cytolytic T-lymphocytes in vivo (40, 41, 42), they may function as rejection antigens when used as vaccines in active immunotherapies (19, 20). Thus, the multimarker RT-PCR to detect MAGE expression in patients with minimal residual cancer may serve a dual purpose in the future: to diagnose the presence of systemic tumor cell spread and at the same time to identify candidate vaccine targets for adjuvant immunotherapy that may cure those fully immunocompetent patients from their residual disease.

We gratefully acknowledge the technical support of Simone Baier, Tanja Siart, and Karin Öttrich. We would also like to thank Dr. Joel Shilyanski (Department of Surgery, University of Chicago, Chicago, IL) for his help and Drs. Walter M. Stadler and Marina Chekmareva (Department of Urology, University of Chicago) for providing prostate tumor specimens. We thank Dr. Francis Brasseur (Ludwig Institute for Cancer Research, Brussels, Belgium) for kindly providing the LB23-SAR and Mz2-Mel cell lines.