Persistence of Disseminated Tumor Cells in the Bone Marrow of Breast Cancer Patients Predicts Increased Risk for Relapse—A European Pooled Analysis

Disseminated tumor cells (DTC) can be detected in the bone marrow (BM) up to 30% to 40% of breast cancer patients (1–5). The strong independent prognostic significance of DTCs at the time of primary diagnosis has already been confirmed by a large pooled analysis including more than 4,700 breast cancer patients (1). On the basis of these results, it was hypothesized that DTCs reflect the presence of minimal residual disease (MRD) and may be the precursor of subsequent metastatic disease (6). The success of adjuvant therapy is based on the ability to eradicate MRD before it becomes clinically evident. Currently, no diagnostic tools are available to monitor treatment response after the completion of adjuvant treatment and identify patients with the need for secondary adjuvant therapy due to persistent tumor cell load. Reevaluation of BM status may be a promising procedure because the presence of DTC is a possible surrogate marker for persistent MRD. To date, only few small studies indicated that a positive BM status during follow-up may be associated with worse outcome (7, 8).

To elucidate the role of persistent DTCs in a larger cohort, clinical follow-up data of 676 patients from 3 academic breast cancer centers were pooled. The aim of the analysis was to assess the prevalence of tumor cells in BM of early breast cancer patients during clinical follow-up and evaluate the clinical significance of the BM status for the individual residual risk after primary treatment of breast cancer.

Three academic breast cancer units in Oslo (Norway), Munich (Germany), and Tuebingen (Germany), which had previously investigated DTCs during follow-up of breast cancer patients, contributed individual patient data for analysis. Earlier data of parts of the patient cohorts (Oslo and Munich) have been published with shorter follow-up (7, 8). For this analysis, the follow-up was limited to 10 years. The study was approved by the Internal Review Boards (Germany) and the Regional Ethic Committee (Norway).

Patients were eligible if they had completed surgery for invasive breast cancer (stages pT1–4, pN0–3, and M0) and agreed to undergo BM aspiration (BMA), after prior written informed consent, during clinical follow-up without evidence of relapse.

The tumor stage at primary diagnosis was classified according to the revised AJCC (American Joint Committee on Cancer) tumor node metastasis (TNM) classification (9), and histopathological grading of the primary tumors was performed according to Elston and Ellis (Oslo cases; ref. 10) or the Bloom–Richardson system (Munich and Tuebingen cases; ref. 11).

Primary surgery consisted of either breast conservation (55.6%) or mastectomy (44.4%) leading to R₀ resection in all cases. Axillary dissection was performed in all patients, except for 7 women having cN0, where the surgeon decided not to do the procedure (age/other conditions). Radiotherapy was done according to the respective national guidelines.

In Oslo, systemic treatment followed the Norwegian guidelines 1995 to 1998 and was given to pT2pN0G2–3 or pN+ patients (5). Chemotherapy [CMF (cyclophosphamide, methotrexate, and fluorouracil)] was administered if age was less than 55 years or if age 55 to 65 years and negative hormone receptor (HR) status. HR-positive patients received tamoxifen for 5 years. In the German centers (Munich and Tuebingen), systemic treatment followed the St. Gallen Treatment Recommendations 1998 and 2001 (12, 13).

We have previously published our semiquantitative assay for BM preparation (14–17). Tumor cell isolation and detection were done on the basis of consensus recommendations (18, 19). In summary, 5 to 20 mL of BM was aspirated and processed within 24 hours. Immunostaining of cytospins from the BM preparations using the pan-anticytokeratin mAbs (monoclonal antibodies) A45-B/B3 (Munich and Tuebingen) or AE1 and AE3 (Oslo) and controls has been described in detail elsewhere (8). The cytospins were screened for DTCs manually or by an automatic device (MDS 1, Applied Imaging) by skilled pathologists. The determination of the presence of DTC was based on consensus criteria (19, 20).

Following the same procedures, the specificity of the method has been determined. In Oslo, BM slides from 98 healthy donors were analyzed with both AE1/AE3 and control antibody. Four of 98 BMs had 1 or more positive cell detected, without similar cells in the negative control. In Tuebingen, among 100 patients without evidence of malignant disease, 1 patient was detected positive. In Munich, as previously published, BM aspirates from 191 patients with nonmalignant disease were also analyzed in a blinded fashion, before the final histopathological result was disclosed. In 2 patients in this group—1 with a chronic benign inflammation of the breast and the other with a benign cystadenoma of the ovary–specifically stained cytokeratin-positive cells were detected. Therefore, the mean overall false-positive rate in the 3 institutions was 1.8%.

Patients were followed at the hospitals' outpatient departments or by family physicians/private gynecologists, at 3 to 12 months interval, and included clinical examination (each visit), mammography (yearly), and (if present) symptoms-driven examinations. Information on disease recurrence was obtained from the patient records. Deaths (including cause) were verified with the regional Cancer Registries (Germany) or the National Mortality Registry (Norway).

The association of DTCs in BM with patient characteristics was tested by the chi-squared test. For survival analysis, breast-cancer–related death, death due to any cause, distant metastasis, and any disease recurrence were separately investigated. If no endpoint was reached, data were censored at last follow-up.

For patients surviving more than 10 years, the follow-up data were censored at 120 months after diagnosis (35 patients, none with events after 120 months). Survival time was measured from the time of surgery to the time of death or first evidence of recurrence. As BM aspiration was an eligibility criterion for this study, left truncation was used to correct for the fact that patients could have died before having had a chance to determine their DTC status. Thus, in the statistical analysis, patients came under observation with regard to the endpoint of interest starting only from the time of BM aspiration.

Meta-analysis techniques were used to compute a summary estimate of the HR and 95% CI for recurrence or death with DTCs as the sole variable on the basis of the effect estimates of each study calculated from the individual patient data. The Q test was done to assess heterogeneity between studies (21).

For univariate significance of DTCs, Kaplan–Meier curves were plotted (22) and the log-rank statistic calculated. Incidence rates and mortality were calculated as the number of disease recurrences or deaths per 1,000 person years. Mortality ratios, incidence–rate ratios, and 95% CI were estimated. Univariate results for all covariates are given in the Supplementary Table.

Cox proportional hazards regression was used to evaluate the simultaneous effect of factors potentially influencing survival (23). The categorical variables were tested for trend across strata. If separate categories did not improve the fit of the model, a linear trend was preferred. A test for interaction between pairs of variables in the final models was done, and the effect of each variable assessed with the Wald test and described by the HR with a 95% CI. All estimates were stratified according to study center, and all reported P values are 2-sided.

The initial model included age at diagnosis, menopausal status, grade, histological tumor type, HR-, pT- and pN-status, and whether a patient had received adjuvant therapy (endocrine and/or cytotoxic). Subjects with any missing values were excluded from modeling. The final model was developed by dropping each variable, in turn, and conducting a likelihood ratio test to compare the full and the nested models. A significance level of 0.10 was used as cutoff to exclude a variable from the model. Finally, the variable of DTCs was added to the model to test the resultant model against that without the variable.

As we observed that curves on the Kaplan–Meier graphs dispersed during the first year of follow-up and then showed less divergence, the assumption of proportional hazards was not met over the entire follow-up period. We therefore opted for a piecewise Cox model with a cutoff point set at 5 years. For both the first and second intervals, separate Cox models were fit. The proportional hazards assumption was formally tested for each interval and separate regression estimates are given (24).

Individual patient data from a total of 676 histologically confirmed invasive breast cancer patients from 3 centers were included in this study (Table 1). The median age was 56 years. BMA was done at a median time of 37 months after primary diagnosis. Overall, 105 patients (15.5%) had DTCs during follow-up. There was no association of persistent DTCs with clinicopathological characteristics (Table 2). The detection rate of DTC decreased to some extent over time. While DTCs were found in 21% of patients who underwent BMA in the first year of follow-up, DTCs were detected in only 6% of patients who had BMA done after the 4th year of follow-up (trend not significant; Table 3).

The meta-analytic HR was 4.87 (CI: 2.04–11.61) for overall survival (OS) and 3.10 (CI: 1.70–5.65) for disease-free survival (DFS) during the first 5 years of follow-up. For these endpoints, the individual HRs calculated for each single study ranged from 4.33 to 5.33 and from 1.58 to 4.30, respectively. The 95% CIs were significant for all but the smallest study (Tuebingen) and showed considerable overlap, indicating a similar effect of DTCs in all studies. The Q test for statistical heterogeneity showed no significant interstudy variation among the estimated HRs (P = 0.976 for OS and P = 0.510 for DFS).

Median follow-up time was 89 months from diagnosis. Seventy-one patients (10.5%) relapsed: 54 with distant metastasis (76.1%) and 17 with locoregional relapse (23.9%). DTCs were detectable in 19 patients (35.2%) with distant metastasis and in 1 patient (5.9%) with locoregional relapse.

DFS and distant DFS (DDFS) were significantly shorter in patients with DTCs compared with patients with no DTC (log-rank test: P = 0.002 and P < 0.001, respectively; Fig. 1). The piecewise model revealed that the difference was significant only during the first 5-year interval of follow-up. There was no survival disadvantage for patients with DTCs during the follow-up period from 6 to 10 years.

In the multivariable model, DTC remained an independent indicator of poor prognosis for both endpoints during the first 5-year interval of follow-up (Table 4).

Overall, 47 patients (7.0%) died during follow-up. In 30 women (63.8%), death was related to breast cancer. Of these, 12 patients (40.0%) had DTCs in BM. Both OS and cancer-specific survival (CCS) were significantly shorter in patients with DTCs compared with patients with no DTC (log-rank test: P < 0.001). In the piecewise model, the survival difference was significant only during the first 5-year interval of follow-up (Fig. 1). In the multivariable model, DTC remained an independent indicator of poor prognosis for both survival endpoints during the first 5 years of follow-up (Table 4).

To specifically investigate the prognostic ability of DTCs after adjuvant therapy, patients who had received any kind of adjuvant therapy (endocrine and/or cytotoxic therapy) were selected for separate analysis. In this patient group, DTC was a predictor of poor outcome for all 4 endpoints (P values of log-rank test for OS, CSS, and DDFS: <0.001, DFS: 0.002; Fig. 1). In univariate models, the prognostic difference was significant for all endpoints during the first 5 years of follow-up but did not reach statistical significance after 5 years. In multivariable models, DTC remained an independent indicator of poor prognosis for all survival and DFS endpoints during the first 5-year interval of follow-up (Table 5).

In patients who did not receive adjuvant therapy, outcome was poorer in the group of patients with DTC (incidence rate ratio ranging from 1.60 for DFS to 3.53 for CSS), but the difference was not statistically significant (data not shown). Note that only few events were observed in this subgroup.

Information on vessel invasion was available of 574 patients. Although significant in univariate analysis (P = 0.018 for overall and P = 0.004 for DFS), vessel invasion was not significant in multivariable analysis, neither in the whole patient sample nor in the subgroup of patients with adjuvant therapy (data not shown).

This pooled analysis showed that DTC detected in BM of breast cancer patients during relapse-free follow-up is an independent prognostic factor for adverse patient outcome. The negative prognostic effect was seen in each of the 3 contributing studies alone. There was no statistical heterogeneity between studies. The use of pooled individual patient data, which is acknowledged as a reliable mode to carry out meta-analysis of survival data, allowed to both standardize inclusion criteria and investigate the effect of changing treatments over time on patient outcome. The prognostic impact of persistent DTC was valid for the first 5-year interval of follow-up after primary breast cancer diagnosis. This effect was also observed in patients who underwent BMA at the time of diagnosis (1). These findings, and particularly the fact that the poor prognosis was confirmed in our subgroup of patients with persistent DTC after adjuvant therapy, indicate that DTC is a marker of disease recurrence and persistent DTC at follow-up may serve as a surrogate marker for treatment failure in the adjuvant setting.

Prevalence of DTC during follow-up was 15.5%, which is overall lower than those 30% reported across studies investigating the DTC prevalence at primary diagnosis (1). Apart from disease stage–related causes, this might be due to a selection bias inherent in this particular study, although patients were required to be free of relapse for enrollment in the individual trials. One might assume that many patients with DTCs at time of primary diagnosis, and thus at high risk of recurrence, may have relapsed before they could be included into this study. This might also explain the lack of association between DTC detected during follow-up and clinical variables, such as advanced tumor stage and lymph node involvement. The assumption of early drop out of high risk patients is further supported by the fact that the prevalence of DTC to some extent decreased with increasing time intervals between primary breast cancer diagnosis and BMA during follow-up. On the other hand, from a clinical point of view, it is of value to observe the significant impact of DTC on patient outcome during the first year of follow-up.

Although DTC status at diagnosis is a strong prognostic marker, a substantial number of DTC-positive patient never recur (1). Performing BMA during follow-up could improve the prognostic accuracy, as the fate of patients with DTC presence at diagnosis would be expected to be affected by the DTC status at follow-up. In this study, the BM status of the patients at time of primary diagnosis is not available in the entire patient group for comparison. However, in a previous analysis of the Oslo cohort, patients with the presence of DTC at both diagnosis and follow-up had a very poor prognosis compared with those that turned DTC negative at follow-up (8).

Importantly, the information obtained from molecular analyses of the primary tumor has improved the prognostication and tailoring of adjuvant treatment in early breast cancer (25–27), but no routine method for modification of the treatment exists during follow-up. Our results indicate that detection of persistent DTC can identify patients who are at risk of relapse because of inadequate initial treatment. As repetitive BMAs were not carried out in this study, individual changes in DTC over time and the optimal time point of BMA cannot be estimated. Repeated BMAs over time could also provide insight into the fate of DTC in patients not suffering from relapse. We are also aware of the possibility of a false negative result when performing only 1 BMA. Multiple testing or more sensitive techniques combined with the analysis of DTC markers for malignancy might be able to improve the prognostic accuracy in the future (6).

Even though positive BM status was an indicator of early disease recurrence, far from all patients with DTC-experienced relapse. The presence of dormant persistent tumor cells has been reported as long as 15 years after primary diagnosis in otherwise tumor-free patients. However, the significance and characteristics of these cells remain unclear (28–30). In our study, outcome of patients with persistent DTC and more than 5 years of recurrence-free follow-up was not significantly worsened. Possibly, still larger studies and more events at later follow-up are needed for a better estimate of 10-year survival rates. It may also be of a great interest to extend the research beyond the presence and characteristics of DTC, and consider the stromal factors and their influence on DTC survival and their metastatic outgrowth.

In principle, patients with persistent DTCs during or after adjuvant therapy would be interesting candidates for clinical trials investigating the impact of treatment changes. Studies to properly address the switches in therapeutic regimens as a function of persistent DTC have been initiated or are under way. In the Norwegian SATT study (NBCG9), about 1,100 breast cancer patients receiving a taxane-free, antracycline-containing adjuvant chemotherapy regimen have been analyzed for DTC in BM at 1 year after surgery. Patients with detectable DTCs received taxane-containing chemotherapy as secondary adjuvant treatment and further monitoring of DTC and clinical outcome. Follow-up in this study is still ongoing. Bisphosphonates or endocrine therapy represent interesting candidates for cell-cycle–independent intervention, given the dormant state of persistent DTC in a substantial number of patients (31, 32). In a small pilot study reported by Rack and colleagues, 31 patients with persistent DTCs were treated with zoledronic acid. All patients but 4 were free of DTC 6 months after the end of zoledronate therapy. The reduction in cell numbers between first and second aspiration was statistically significant (P < 0.0001; ref. 33). In addition, new biological treatments and newer agents for antiresorptive bone treatment are candidate DTC intervention strategies to be explored in next-generation randomized clinical trials.

Because of the greater feasibility of peripheral blood sampling as compared with BM, many research groups are currently assessing circulating tumor cells (CTC) in clinical studies and have established CTC screening as a monitoring tool in the metastatic setting (34, 35). In contrast to patients with metastatic disease, much less information about the prognostic relevance of CTC in patients with early-stage disease is available (36). Studies using the CellSearch System or reverse transcriptase-PCR–based techniques have reported CTC persistence as an indicator of unfavorable outcome after completion of chemotherapy (37–40). However, available studies comparing CTC and DTC analysis so far show only partially overlapping results, and the prognostic information from DTC seems to be higher than that from CTC (41–43). The difference between CTC and DTC results might be explained by methodological and/or sensitivity-level issues. It is also likely that CTC and DTC provide complementary information. Parallel analysis of CTC and DTC by using the same methodology should be encouraged in future studies.

In conclusion, the results of this study further strengthen the prognostic significance and the clinical impact of DTC in BM of breast cancer patients. Our data imply that there is a clinical potential for monitoring treatment efficacy, which should be further explored in well-designed randomized clinical trials. Among the options for both trial and research hypotheses are the following: (i) utilization of persistent DTCs for adapting adjuvant treatment to modulated individual residual risk; (ii) phenotyping of DTCs for addressing the differential impact of tumor cell biology; and (iii) profiling DTCs for targeted therapy development. In this respect, we see a clinical potential of MRD indicators that by far outperforms that of established parameters currently used in breast cancer diagnosis and treatment.

No potential conflicts of interest were disclosed.

We are sincerely grateful to D. Gray and S. Gray for their editorial support.

Supported by a grant from the Friedrich-Baur Stiftung, Muenchen, Germany, the European Commission (DISMAL-project, contract LSHC-CT-2005-018911), and Norwegian Cancer Society.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.