Comparison of molecular and immunocytochemical methods for detection of disseminated tumor cells in bone marrow from early breast cancer patients

Despite a continuous effort to improve cancer diagnostics and treatment, breast cancer remains a leading cause of death among women worldwide. Current adjuvant treatment decisions are dependent on well-known prognostic factors including TNM-staging and histological grade, as well as the estrogen receptor (ER), progesterone receptor (PgR), human epidermal growth factor receptor 2 (HER2), and more recently Ki-67-status [1]. The search for better prognostic factors, as well as predictors of the effect of adjuvant treatment, has led to a thorough evaluation of disseminated tumor cells (DTCs) and their persistence in bone marrow (BM) [2–5]. Moreover, DTCs have been shown to provide independent prognostic information in breast cancer patients [2–5]. However, more research is needed before the implementation of BM status in routine clinical practice. The predictive value of BM status as a tool in making adjuvant treatment decisions has yet to be investigated in randomized phase III trials. Furthermore, the detection of tumor cells in the BM does not always lead to disease relapse. Many patients with positive DTC status do not relapse, and DTCs can be detected in patients with ductal carcinoma in situ [6]. The mechanisms behind tumor dormancy and the possibility of tumor cell re-awakening are poorly understood. Interestingly, increasing evidence has emerged in the last few years supporting that the addition of bisphosphonates in the adjuvant treatment both reduces the risk of persistent DTCs and improves survival [7–10]. This supports the biological relevance of DTCs and the importance of methods to accurately assess the DTC-status when selecting patients for adjuvant treatment.

However, different methods are used to assess DTCs in the BM, and there is a clear need for standardization. Due to the very low frequency of DTCs in the BM, different methods are used to enrich tumor cells in the BM samples before detection. The enrichment can be based on density gradient centrifugation, flow cytometry, immunomagnetic beads, and membrane filtration [11]. Protocols based on immunocytochemistry (ICC) and reverse transcription quantitative PCR (RT-qPCR) are the most commonly used methods for DTC detection. When ICC is used for DTC detection, the results will be affected by the choice of keratin antibodies, as discrepancies between different antibody mixtures have been reported [11–13]. Similarly, the choice of mRNA markers, as well as different assays and platforms, affect the performance of RT-qPCR based DTC detection [4, 14–19]. Thus, the comparison of studies based on different detection methods is challenging. Nevertheless, a few studies report the concordance between ICC-based and RT-qPCR-based DTC detection in breast cancer patients to be about 70-80% [20–22]; although, these numbers are primarily reflecting that the majority of patients have negative BM-status with both methodologies.

In the present study we compared a multimarker (MM) RT-qPCR assay, consisting of keratin 19 (KRT19), TWIST1, and mammaglobin A (hMAM), with ICC using the AE1-AE3 mAb for the detection of DTCs in 267 early breast cancer patients previously treated with adjuvant fluorouracil, cyclophosphamide, epirubicin (FEC) chemotherapy.

This study was undertaken to compare ICC with a MM RT-qPCR assay for the detection of DTCs in BM after adjuvant chemotherapy in early breast cancer patients. Our study revealed a markedly higher frequency of positive samples by both the MM RT-qPCR assay and the individual mRNA-assays compared with ICC. Multiple mRNA markers clearly contributed to a higher number of positive samples compared to only using single markers. The relatively high (61%) concordance between ICC and RT-qPCR is primarily because a large fraction of the samples were negative by both methods. Accordingly, the kappa observer agreement value was only 0.04, suggesting that the apparent concordance was primarily due to chance. In principle, the ICC assay should stain, among others, KRT19 positive cells. Thus, we expected better concordance between the ICC and the KRT19 mRNA results. However, only 4 out of 313 samples were positive by both methods and as many as 42 ICC-negative samples were positive for KRT19 mRNA. One possible explanation for this is that the KRT19 mRNA assay is more sensitive than the ICC assay. The 19 ICC-positive samples that were not detected by the KRT19 mRNA assay might be explained by detection of KRT19-negative DTCs that express other keratins detected by the ICC approach.

The low concentration of DTCs in BM samples may affect reproducibility in both detection methods. Many samples had levels near the detection limit for KRT19 mRNA; whereas, the ICC-assay was able to detect only a single cell in the majority of positive samples. It follows from the Poisson distribution of rare events that there is a roughly 35% risk that a second sample would be a false negative. Hence, the reproducibility of DTC detection might be enhanced by analyzing larger sample volumes. On the other extreme, it was recently shown that screening a very large volume of peripheral blood by leukapheresis revealed DTCs in 90% of non-metastatic breast cancer patients [32]. Such high numbers of DTCs does not correlate with the risk of relapse for this patient group, and thus implies a dramatic increase in detection of clinically irrelevant cells.

The hMAM mRNA assay was only positive in a very small number of samples (3/313); therefore, it might be of limited value in combination with KRT19 mRNA in the post-adjuvant treatment setting. Indeed, 2 of the 3 positive hMAM samples were also positive for KRT19 mRNA and by ICC, with convincing DTC-counts of 2 and 46 by ICC. The remaining hMAM positive sample was TWIST1-positive and KRT19- and ICC-negative. Thus, it seems that hMAM contributes to the identification of a very small subgroup of patients, possibly those with a very high risk, consistent with our previous report on hMAM [4, 18].

TWIST1 was shown to add prognostic information to a DTC MM panel described by Tjensvoll et al. [4]. Interestingly, we noted that a substantially higher number of patients had elevated TWIST1 mRNA levels in our present study [4]. The clinical follow-up will ultimately help determine the relevance of this discrepancy. As TWIST1 is a proposed epithelial-mesenchymal-transition (EMT)-marker [33], the higher number of positive samples might indicate that a substantial portion of patients have DTCs not expressing keratins [34]. The number of TWIST1 positive samples, however, exceeds the anticipated number of clinical relapses. Thus, our assay might be too sensitive, or the cut-off level needs refinement to reveal only clinically relevant information. ROC analysis in relation to clinical outcome data, when available, may reveal an optimal cut-off value. However, this will require confirmation in a validation cohort.

The high number of DTC positive samples by the RT-qPCR approach is in part explained by the high number of TWIST1 positive samples. In later years, there has been much focus on mesenchymal markers to detect cells that have undergone EMT as part of the metastatic process. This is thought to be a reversible process in which the cancer cells gain mesenchymal properties to be able to infiltrate different tissues and give rise to micro- and ultimately macro metastases [34]. Yu et al. showed that in circulating tumor cells a shift towards higher expression of EMT-markers is associated with tumor progression [35]. Thus, we might speculate whether cells transiently expressing mesenchymal genes, like TWIST1, comprise the subgroup of DTCs with stem-cell properties and, therefore, the proportion of DTCs that harbor metastasis-generating abilities [36]. The loss of epithelial characteristics may imply that these cells are difficult to detect by most commonly used ICC DTC assays.

The discrepancy between the RT-qPCR based and the ICC-based DTC detection is not surprising based on previous studies. Becker et al. found agreement between ICC (with the A45-B/B3 mAb) and KRT19 mRNA detection in 73% of the 385 cases, in line with our KRT19 qPCR results (81% agreement). Although, the results are biased since the majority of patients were negative by both methods. In fact, a kappa value of 0.39 can be computed based on their reported data, confirming this suspicion to some extent. Moreover, they demonstrated a 35% positive rate for both ICC and KRT19 mRNA and 49% of the patients were positive by at least one of the methods. The time of BM-collection might be an important difference between their study and the present one. We collected BM after adjuvant chemotherapy; whereas, Becker et al. collected the majority of samples prior to surgery and only a few (n = 63) after surgery and chemotherapy. This may have contributed to the much lower number of positive samples based on both ICC and KRT19 mRNA in our study. Others have reported concordance in the same range as in our study. Benoy et al. reported concordant results in 75% of the samples; whereas, Slade et al. found agreement between the methods in 71% of the samples [21, 22]. Molloy et al. compared a MM RT-qPCR assay with ICC in a large population of 733 patients and found both to be significantly predictive of poorer outcome. However, the RT-qPCR assay was applied to blood samples (circulating tumor cells) and the ICC to BM samples (DTCs). Thus, a direct comparison with the current study is difficult because the samples were collected from different body regions in addition to being analyzed by two different methods [37].

A general issue regarding mRNA-based DTC detection is the background level of epithelial transcripts in white blood cells. However, comparison with blood samples from a normal control cohort may compensate for this issue, allowing threshold values for pathological marker levels in blood to be established. The latter strategy was utilized in the current study to minimize the number of false positives due to such background expression in leukocytes.

Despite clear evidence that DTCs in BM in early breast cancer patients predict a poor outcome, a better understanding is needed for these analyses to be implemented in the routine clinical management of patients. Braun et al. found, in their large pooled analysis of 4703 patients, a significant prognostic value of BM DTC-status in all patients including the lymph-node negative subgroup [3]. On the other hand, several smaller studies, e.g. by Langer et al., did not find any significant DTC-specific difference in overall and breast cancer specific survival in 411 clinically lymph node negative patients, a result that may be caused by the low number of patients in the study [3, 38]. However, this result emphasizes that the prognostic value of BM DTC-status might be strongest in already defined high-risk patients. In the current analysis, only intermediate/high-risk patients, i.e. patients with higher-risk node negative or node positive disease, were included. Thus, this may be a group where the BM-DTC status may add clinically relevant prognostic information.

Few studies have investigated the impact of BM-DTCs after initial therapy. In a pooled analysis, Janni et al. demonstrated that DTCs can be detected several years after diagnosis [39]. The persistence of BM DTCs after neoadjuvant treatment was also associated with worse prognosis in a recent study [40], and our previous results showed reduced survival for patients with persistent DTCs, assessed by our mRNA MM-assay (hMAM, KRT19, and TWIST1), after surgery [41]. The majority of studies so far have used ICC for the detection of DTCs in the BM; although, some studies also used RT-qPCR. The DTC-detection by ICC is largely based on pan-cytokeratin antibodies, but different antibody combinations have also been used with varying results. Effenberger and colleagues found that ICC detection and prognostic relevance were different for the two most commonly used pan-cytokeratin antibody combinations (A45-B/B3 (A45) and AE1-AE3 (AE)) [12]. The AE mAb was more prognostic for the lymph node positive patients. Accordingly, the AE mAbs were utilized in the current SATT-trial, consistent with the inclusion of a higher risk population.

In conclusion, this study is to our knowledge the largest comparison between ICC- and RT-qPCR-based DTC detection methods in BM samples collected after adjuvant chemotherapy in a defined high-risk early breast cancer population. We detected more positive BM samples with RT-qPCR assays, based on KRT19, hMAM, and TWIST1 mRNAs, than with ICC. The clinical implications of these findings, however, await future clinical follow-up. Due to a potential shift in DTC phenotype, we included the mesenchymal TWIST1 mRNA marker in an attempt to detect the subpopulation of DTCs lacking epithelial characteristics. Hopefully, this might help identify additional patients with clinically relevant DTCs. The current findings support that the different means of detection could be complementary and that both RT-qPCR and ICC should be further studied as methods for DTC detection in early breast cancer patients.