Detection of Rare Disseminated Tumor Cells Identifies Head and Neck Cancer Patients at Risk of Treatment Failure

Despite improvements in surgery and reconstructive techniques, 20–30% of patients presenting with early tumors, and more than one-half of those with advanced head and neck cancers, fail to survive 5 years posttreatment. However, although the prognosis for these cases remains poor, there has been a change in the pattern of clinical disease such that, today, fewer cases succumb as a result of development of locoregional recurrence, but more develop distant metastases, which are an increasing cause of morbidity and mortality (reviewed by Partridge, see Ref. 1). These findings highlight the requirement to develop strategies to detect DTCs² to enable clinicians to identify cases most likely to benefit from early systemic treatment and to stem the rising tide of distant metastases.

Two approaches are commonly used to detect DTCs. One is based on immunocytochemical examination of cells sedimented onto microscope slides, the other uses RT-PCR for tumor-specific or epithelial-specific targets. The primary advantage of the ICC assay is the availability of stained cells for morphological examination, but this approach is very time-consuming, particularly when large numbers of cells are sedimented onto microscope slides and scored manually. Most protocols, incorporating ICC to screen samples of blood or BM aspirates obtained from patients with breast or colon adenocarcinoma for DTCs, examine 2 × 10⁶ mononuclear cells obtained after density gradient separation of leukocytes (2). However, when these gradients are used to analyze blood and BM obtained from patients with head and neck carcinoma, malignant cells sediment with both the mononuclear and granulocyte cells, reflecting the different densities of squamous tumor cells (3). In addition, because DTCs occur at only low frequency, a large number of leukocytes need to be screened to detect these rare events. Our previous report identified one or two tumor cells per 6 × 10⁷ leukocytes (3), and it is not feasible to score this number of leukocytes manually for every case without the addition of a tumor enrichment technique. Studies incorporating positive or negative IMS of BM demonstrate greater tumor cell recovery when the negative approach is applied to samples obtained from head and neck cancer patients (3).

Keratin polypeptides are commonly used as the target for detection of tumor cells by ICC. Expression of these antigens on head and neck squamous tumors is heterogeneous, such that a pan-cytokeratin antibody is commonly used to maximize the chances of detecting these targets (3). However, the use of these reagents may generate false-positive immunoreactive cells (3, 4). This is because rare hematopoietic cells illegitimately express cytokeratins, or because of cross-reactivity of the antibody resulting in a positive reaction in nonepithelial cells. This means that detection of a cell reactive with a pan-cytokeratin antibody may not be sufficient to conclude that this represents a disseminated epithelial tumor cell (3, 5, 6, 7). Thus, to ensure high specificity of ICC for tumor cell detection, these protocols must incorporate appropriate controls and morphological assessment of immunostained cells to demonstrate that the immunoreaction is tumor-specific. Despite incorporating these safeguards, on occasion it can be very difficult to distinguish an immunostained tumor cell from a cytokeratin-positive cell with the morphological characteristics of a hematopoietic cell (4, 8). On the basis of this knowledge, studies have attempted to define the morphological criteria for scoring immunostained cells as tumor more precisely (3, 8). Our investigation incorporating these modifications confirmed our earlier report of the finding of one or two tumor cells per 6 × 10⁷ leukocytes derived from preoperative BM aspirates (3) and revealed that approximately one-half of these cases develop distant metastases (9).

In view of the effort needed to assure the reliability of approaches based on ICC, studies have investigated the feasibility of using RT-PCR assays to detect DTCs by the finding of gene transcripts that are not normally present in the HCC. It is also claimed that RT-PCR may be more sensitive than ICC for tumor cell detection, because theoretically this approach can amplify a single molecule of target in a given sample. However, this extreme sensitivity also confers an inherent tendency to produce false-positives because of expression of epithelial cell transcripts by nonhematopoietic cells, detection of genomic sequences in contaminating DNA, and detection of transcripts of other genes. Whereas the latter problems can be controlled for by careful primer design and separating the components of the RT and PCR, ectopic expression of the target sequences remains a major problem. Thus, whereas illegitimate gene transcription (the finding of transcripts from any gene in any cell) is low and estimated at one molecule per 100–10,000 cells (10), this results in unwanted or false-positivity because of the high sensitivity of the PCR method.

The ideal target for RT-PCR would be expressed at high levels in all tumor cells and not be present in the HCC. To date, only the p53 gene has been found to be commonly mutated in many head and neck tumors, but methods to screen for mutated p53 DNA at the level of only a few tumor cells in blood or BM are not currently available (11). As a consequence, epithelial-specific targets have been used for tumor cell detection. The E48 (Ly-6D) antigen has also been proposed as a target for the detection of DTCs. This is a surface protein anchored to the cell membrane by a glycosylphosphatidyl-inositol anchor involved in signal transduction and cell-to-cell adhesion of normal and malignant squamous epithelia. When applying the E48 RT-PCR assay, Brakenhoff et al. (12) detected one tumor cell in 2 × 10⁷ leukocytes, and false-negative results were minimized by incorporating an internal standard and testing each cDNA four times. However, false-negatives may still occur because of sampling errors, the presence of inhibitors of the PCR process in tissues, and, particularly, the down-regulation of the target gene on single disseminated cells. More than 90% of head and neck tumors express the E48 antigen and approximately two-thirds of these cases show expression of this target on more than 50% of the malignant cells (13).

Some studies have suggested that a more reliable assay may be developed by testing postoperative rather than preoperative samples. The basis of this hypothesis is that tumor cells may simply overflow into the circulation when the tumor burden is high preoperatively, but any that remain after surgery are recognized and destroyed by the immune system, such that those persisting after 3 months probably represent a true minimal residual disease situation (14).

Despite problems inherent in the development of approaches based on both ICC and RT-PCR, there is evidence that detection of DTCs produces data that are clinically relevant, identifying patients who are developing distant metastases and having reduced survival (for examples, see Refs. 9 and 15). However, at present there is a lack of consensus as to which is the most appropriate method to detect DTCs for patients with head and neck cancer. The present study was designed to address this point by analyzing samples of blood and BM taken preoperatively, intraoperatively, and 3 months postsurgery using approaches incorporating ICC and RT-PCR, and following the cases to establish the contribution of these remaining malignant cells to distant relapse.

UM-SCC-22A (16) were maintained in DMEM with 10% fetal bovine serum. Monolayers were washed in EDTA, incubated with 0.25% trypsin or with EDTA, and single-cell suspensions were prepared by mechanical disruption.

BM (up to 20 ml) was obtained from 40 patients with SCC who consented to undergo aspiration biopsy at the time of surgery. Four cases had recurrent tumors and 36 were undergoing treatment for a primary malignancy (Table 1). The aspirate was taken from the iliac crest, once the patient was anesthetized but before placement of the tracheostomy tube. Contamination of the sample by skin cells was avoided by making a skin incision and separating the skin edges before inserting the trochar. The first 5 ml of sample, collected with 5000 units heparin/5 ml of BM, was discarded, and the needle was repositioned three to four times while taking the sample.

Venous blood was taken from a central line, situated as close to the right atrium as feasible, and was collected with heparin. One sample was collected before surgery and a second intraoperative sample was collected after the resection was complete. Radiotherapy was given according to standard protocols, and a sample of PVB was collected 3–4 months postoperatively, once radiotherapy was complete. The BM aspiration was repeated at this time under local anesthetic. Ethical Committee approval for this project was granted at all collaborating centers.

PVB and BM were also collected from 34 cases previously treated for hematological malignancy, myelodysplastic syndrome, who were undergoing BM biopsy for other diagnostic purposes, and from 16 healthy controls. All of the samples were collected in a single tube and were divided into two aliquots. One aliquot of the samples were processed for RT-PCR and another for ICC as described previously (3).

Aliquots of samples of blood and BM for E48 RT-PCR were processed as described by Brakenhoff et al. (12) and four 5-μl aliquots of each 20-μl RT reaction used for PCR in a final volume of 50 μl. Ten μl of each of the four PCR products were run on a 2% gel in 1× Tris-borate-EDTA and were denatured, and the PCR products were blotted onto nitrocellulose (Amersham, United Kingdom) by capillary transfer for 16 h. After blotting, the filters were neutralized by three rinses in 2× SSC for 5 min and were baked at 80°C for 2 h. The filters were prehybridized for 2 h at 65°C in buffer containing 0.5 m phosphate buffer (pH 7.2), 7% SDS, and 1 mm EDTA, and were hybridized overnight after addition of the probe. The cDNA probe consisted of nucleotides 29–772 of the E48 cDNA labeled with [γ-³²P]dCTP (Chalfont St. Giles, Amersham) by random primer elongation. The filters were washed twice with 2× SSC-0.1% SDS and twice with 0.2× SSC-0.1% SDS at 65°C. The bands were visualized by autoradiography with Kodak XAR-5 film (Kodak, Rochester, New York) using intensifying screens. The negative controls for the RT were reactions without reverse transcriptase and, for the RT-PCR, samples without RNA. The Abelson primer set (sense 5′-GTGATTATAGCCTAAGACCCGGAG-3′, antisense 5′-TTCAGCGGCCAGAGCATCTGACTT-3′) was used to check the quality of each RNA analyzed. Positive controls corresponding to 1 and 10 tumor cells in 7 ml of blood (5 and 50 pg of UM-SCC-22A RNA, respectively, in 5 μg of WBC RNA) were run in parallel, to show that the required level of sensitivity of one tumor cell in 7 ml of blood was reached in every experiment scored (Fig. 1 A). A test was scored as positive if one of the four PCRs was positive.

Five × 10⁷ leukocytes were negatively selected for tumor using 200 μl of M280 Dynabeads coated with anti-CD45, and nonrosetted cells sedimented onto microscope slides and stained with AE1/AE3 diluted 1:50 (Dako), or the isotype control, using an alkaline phosphatase anti-alkaline phosphatase (APAAP) technique as described previously (3). In addition, 4 × 10⁶ cells were stained with AE1/AE3 for each case without IMS. Positive controls were SCC-25 cells spiked into leukocytes obtained from healthy volunteers.

Immunoreactive cells were scored as tumor if there were no positive cells in the isotype-matched IgG1 control, if they lacked recognizable hematopoietic characteristics, and if they had obvious tumor morphology. Only cells with a clearly visible nucleus and an intact cell membrane were evaluated. The size of tumor cells varied from similar to more than three times the size of the surrounding hematopoietic cells, often with an irregular shape and a high nuclear:cytoplasmic ratio when compared with the hematopoietic cells. However, some tumor cell nuclei were the same size as adjacent hematopoietic cells. The presence of two to three immunoreactive cells meeting these criteria in a clinical sample was considered to be helpful when scoring these preparations as tumor-positive. Cytoplasmic staining for keratins varied from strong to weak but was typically evenly distributed in tumor cells, although heterogeneous patterns were seen in some weak- to moderately stained cells. Immunoreactive cells that could not be categorically identified as malignant were not scored as tumor.

After application of the criteria described, patient samples were scored as tumor-positive or -negative. BM was tested pre- and postoperatively with the E48 RT-PCR and ICC. CVB was tested preoperatively by E48 RT-PCR and ICC and, after the resection was complete, by E48 RT-PCR. Samples of BM and PVB were tested by E48 RT-PCR and ICC postoperatively. McNemar’s test was used to compare rates of a tumor-positive test result for pairs of assays. Cohen’s κ was used to summarize the concordance between the ICC and E48 RT-PCR assays carried out on the same sample. However, although the two tests were performed on the “same” sample of blood or BM aspirate, different aliquots of that sample were used, adding an additional source of variation. Log-rank tests and Cox regression were used to investigate the usefulness of the assays to predict the development of distant metastasis, DFS, and DMFS (in the latter, survivors might include cases with local or neck recurrence).

Eighteen of 36 evaluable cases had evidence of DTCs in the HCC preoperatively after testing BM aspirates and CVB with the E48 RT-PCR assay and ICC. The results were analyzed to establish whether any clinical or pathological features of the cases selected were associated with the finding of DTCs, but none were found. Five of 14 T₁ and T₂ tumors and 10 of 17 T₃ and T₄ cases were found to have DTC in the HCC (χ⁵⁰ = 1.64; df = 1; P = 0.200). Fourteen of 26 node-positive cases showed evidence of disseminated disease compared with 1 of 9 node-negative cases (Fisher’s exact test, P = 0.083), and no relationship was found between a tumor-positive test and the type of tumor differentiation, the site of the primary lesion, or the presence of extracapsular spread. Details of the cases examined and the results obtained are shown in Table 1.

Two cases found to have immunoreactive cells in the isotype controls were excluded from the analysis. Examination of 4 × 10⁶ cells without IMS did not reveal the presence of DTCs in any sample of blood or BM tested. However, using our protocol with anti-CD45 Dynabeads examining 5 × 10⁷ leukocytes, we found that 36 paired samples of blood and BM were evaluable preoperatively and that the number of tumor cells detected ranged from one to five per 5 × 10⁷ leukocytes examined. Significantly more tumor-positive cases were identified after the analysis of BM aspirates than after the testing of CVB [13 tumor-positive BM aspirates and 3 tumor-positive CVB samples were identified preoperatively (Table 1), with two cases scored as tumor-positive when both samples were tested; P = 0.022]. Thirty BM aspirates were evaluated pre- and postoperatively. Fewer tumor-positive samples were found postoperatively [12 tumor-positive cases preoperatively and 2 positive cases postsurgery (Table 1), with one sample positive on both occasions; P = 0.006].

E48 transcripts were not detected in CVB obtained from 16 healthy controls. However, they were detected in 8 of 34 BM aspirates from patients undergoing investigations for management of myeloma, myelodysplastic syndrome, anemia, neutropenia, viral infection, or frank leukemia. This information confirms that this assay provides a specific target for molecular detection of DTCs but also reveals that using samples from patients with hematological dyscrasias does not provide appropriate controls when developing sensitive and specific assays for tumor cell detection.

Thirty-eight paired samples of CVB and BM were evaluable preoperatively and 14 tumor-positive cases were identified [9 tumor-positive samples of BM and 11 of CVB with 6 samples scored as tumor-positive after testing of both samples (Table 1); P = 0.727]. Gel lanes representing duplicates of one of the four PCR reaction products are shown in Fig. 1,B. The number of cases found to have DTCs in CVB in the intraoperative sample was similar to the number identified preoperatively [11 preoperatively and 14 intraoperatively of 38 evaluable samples, with 7 positive on both occasions (Table 1); P = 0.549].

Thirty BM aspirates were evaluable pre- and postoperatively. The number of tumor-positive cases did not change significantly when the paired samples were compared [9 tumor-positive preoperatively and 10 tumor-positive postsurgery, with 6 samples positive at both time-points (Table 1); P = 1.0]. Thirty-six blood samples were evaluable at both time points, 10 tumor-positive cases were identified by testing the preoperative sample, and none postsurgery (Table 1).

Preoperatively, there was generally good agreement between the results obtained with the ICC and those obtained with the E48 RT-PCR applied to preoperative BM samples (κ = 0.62), but not between the results with the preoperative ICC and E48 RT-PCR analysis of CVB (κ = 0.12). However, postsurgery, the agreement between the two tests applied to BM aspirates was poor (κ = 0.25), because ICC identified significantly fewer tumor-positive cases than did the E48 RT-PCR (P = 0.008). None of the postoperative PVB samples was scored as tumor-positive with either test.

The patients were followed for a minimum of 24 months (range, 24–59 months; median, 36 months). During this period, four recurrences occurred at the operative site, two cases developed as locoregional recurrence, and seven neck masses developed (in one untreated and six treated necks; Table 1). Distant metastases were identified after chest radiography for six patients. Five cases with nodal recurrence had two or more lung nodules, and the diagnosis of SCC was confirmed by bronchoscopy for the case without locoregional recurrence. No metastases were identified by elevation of alkaline phosphatase or liver ultrasound.

All of the six cases developing a lung tumor were found to have a tumor-positive BM or CVB with the E48 RT-PCR assay preoperatively and five of the six cases developing a lung tumor had a tumor-positive preoperative BM or CVB detected by ICC. Four of these cases also had evidence of malignant cells in the intraoperative CVB. Each of the four assays (ICC of BM and CVB and RT-PCR of BM and CVB) performed preoperatively was found to be related to DMFS using the log-rank test (see Table 2 and Fig. 2), and applying the E48 RT-PCR assay to the intraoperative CVB sample was also found to be significantly related to DMFS.

The results applying the E48 RT-PCR and ICC tests to preoperative CVB and BM were considered in combination with the use of Cox regression. No case that was scored as tumor-negative when applying the preoperative E48 assay to CVB developed distant metastases; therefore, the effect of this factor could not be estimated by the regression models. When the results obtained after testing of preoperative BM aspirates and CVB with ICC were considered together, each test was independently significant (tumor-positive BM: χ² = 6.74; df = 1; P = 0.009; tumor-positive CVB: χ² = 9.05; df = 1; P = 0.003), revealing that testing both samples contributes useful information when determining the risk of developing distant metastases. However, when the results after testing of preoperative BM aspirates with both the ICC and E48 RT-PCR assays were considered, neither test was significant (ICC: χ² = 2.53; df = 1; P = 0.11; E48 RT-PCR: χ² = 0.95; df = 1; P = 0.33), which suggested that either assay was a reasonable proxy for the other in predicting DMFS.

No sample of PVB was scored as tumor-positive postoperatively with either test; therefore, these results were not useful in terms of predicting an outcome for this patient cohort. The results after application of these tests to postoperative BM were not significant (Table 2). In addition, when these data were analyzed with tumor stage as an additional covariate, the results were significant, at least at the 10% level (data not shown).

Each of the four assays performed preoperatively was found to be related to DFS using the log-rank test (Table 2). Applying the E48 RT-PCR assay to the intraoperative CVB sample was also found to be significantly related to DFS.

When the results obtained after the testing of preoperative BM aspirates and CVB with ICC were considered in combination by using Cox regression, the BM assay was significant (χ² = 5.33; df = 1; P = 0.021) but the CVB assay was not (χ² = 2.41; df = 1; P = 0.12). Similarly, when the results obtained after the testing of preoperative BM aspirates and CVB with E48 RT-PCR were considered in combination, the BM assay was marginally significant (χ² = 3.60; df = 1; P = 0.058), but the CVB assay was not (χ² = 1.83; df = 1; P = 0.18). However, when the results of applying ICC and E48 RT-PCR to preoperative BM were considered together, neither test was significant, suggesting that either assay was a reasonable proxy for the other in predicting DFS (ICC: χ² = 1.50; df = 1; P = 0.22; E48 RT-PCR: χ² = 1.57l; df = 1; P = 0.21). Neither of the results of ICC or E48 RT-PCR applied to postoperative BM was significantly related to DFS (Table 2).

It is well recognized that DTCs are not always detected in the HCC even when advanced metastatic disease is present (17, 18, 19). These false-negative results are presumed to be the result of sampling errors, intermittent tumor-cell shedding, or effective immune surveillance or to reflect stochastic effects at the lower limit of sensitivity of the assay. Our earlier reports revealed that DTCs are found in the CVB and BM of head and neck cancer patients at a much lower frequency (one to five cells per 6 × 10⁷ leukocytes) than reported for many other tumor sites (2). Thus, detection of these cells is such a rare event that there is always an appreciable chance of not finding any of them when a single test is used. On the basis of this knowledge, a decision was made to incorporate the testing of BM and blood with ICC and with a molecular approach to establish whether or not testing multiple samples and/or applying different tests would increase the predictive value of screening tools to detect micrometastases.

In the present study, 10% of CVB samples and 20% of BM aspirates were scored as tumor-positive with the preoperative E48 RT-PCR, and 20% of CVB samples and 32% of BM aspirates were tumor-positive with the ICC test, in general agreement with our earlier reports (1, 3, 9). We did not find any evidence of DTCs in the postoperative PVB, likely reflecting the fact that a sample of CVB is taken from a large vein draining the tumor area, whereas the peripheral sample has traversed an enormous vascular bed, such that the chances of finding rare DTCs is remote. Although we cannot exclude the possibility that CVB samples might also have been tumor-negative at this time. The failure to find DTCs in the peripheral circulation supports the notion that malignant cells that are detected in BM have a distinct phenotype that causes them to arrest at that site, and that these micrometastases do not, or only rarely, circulate.

Preoperatively, there was good agreement between the number of tumor-positive cases identified after the application of both tests to BM. However, the concordance between the results when applying these tests to preoperative CVB was less good, a finding that may reflect the fact that different aliquots of the same sample were tested by RT-PCR and ICC or because of sampling errors.

Postoperatively, the agreement between the results obtained with the E48 RT-PCR and ICC was poor. A high number of cases were scored as having a tumor-positive BM with the molecular approach, but this result was not confirmed by a morphological assessment of immunostained cells. The lack of confirmation may be attributable to up-regulation the E48 gene in BM, perhaps as a response to blood loss during the surgical procedure, as an indirect radiotherapy reaction, or because of the presence of dying tumor cells in the HCC postradiotherapy. This highlights the benefit of using a second test that incorporates morphological assessment of immunostained cells when interpreting data from RT-PCR studies.

The finding of fewer tumor-positive cases postoperatively suggests that malignant cells overflow into the circulation when the tumor burden is high. Presumably most DTCs are trapped in the lung bed or are destroyed by the immune system, because they were only rarely detected in the BM 3 months postoperatively.

In the present study, the detection of DTCs, from examining either preoperative CVB or preoperative BM, predicted the risk of development of distant metastasis, strongly suggesting that a proportion of these residual tumor cells were true micrometastases. There was also evidence that the prediction could be improved by applying a single test to both BM and CVB or, possibly, by testing two different samples of BM or CVB, when making critical decisions about the need for additional treatment.

The intraoperative E48 RT-PCR assay was also significantly related to the development of distant metastases. Currently, more resections for head and neck cancer patients incorporate a selective rather than a radical neck dissection. At this stage, we can only speculate as to whether the failure to clamp the internal jugular vein at an early stage during this procedure contributes to intraoperative tumor dissemination. In the present study, four of the six cases developing lung tumors had evidence of DTCs in the CVB at operation, and lung metastases developed only in the right lung. None of these cases developed metastases in any other organ during the period of study. This supports the notion that these distant metastases may have arisen as a result of the shedding of malignant cells from the primary tumor, which were then were mopped up in the lung bed. Until recently it was generally believed that the barrier to survival of metastatic cells was their ability to escape from the circulation into a new organ after capillary trapping. However, recent studies reveal that not all DTCs need to extravasate from the HCC to grow, as Al-Mehdi et al. (20) showed that lung metastases may be initiated by attachment of tumor cells to the vascular endothelium and subsequent intravascular proliferation. This suggests that a combination of vascular trapping, and the shared phenotypic characteristics of the epithelial lining of the upper and lower airways, creates a microenvironment that is permissive for the survival and eventual outgrowth of DTCs, and it explains the preponderance of lung metastases in these cases.

An unexpected result in the present study was the finding that the presence of DTCs in preoperative BM and CVB and in intraoperative CVB correlates with DFS, as well as with DMFS, suggesting that detection of these micrometastases predict for the development of locoregional recurrence as well as distant metastasis. In addition, our findings suggest that DTCs predicts DFS independent of tumor stage, although the number of cases analyzed is relatively small, and a substantive conclusion awaits the analysis of more cases.

Future endeavors should encompass determining the genotypic and phenotypic characteristics of DTCs that are true micrometastases and are able to grow again at a new site. It is anticipated that such studies will reduce the number of false-positive tests so that information about the presence or absence of DTCs can be used to identify the subgroup of patients with the highest risk of developing metastases so that they can receive early adjuvant systemic treatment.

UM-SCC-22A was a kind gift from Dr. T. E. Carey (University of Michigan, Ann Arbor, MI).