Assessment of circulating tumor cells and serum markers for progression-free survival prediction in metastatic breast cancer: a prospective observational study

Introduction Circulating tumor cells (CTC) have been recently proposed as a new dynamic blood marker whose positivity at baseline is a prognostic factor and whose changes under treatment are correlated with progression-free survival (PFS) in metastatic breast cancer patients. However, serum marker levels are also used for the same purpose, and no clear comparison has been reported to date. Methods The IC 2006-04 enrolled prospectively 267 metastatic breast cancer patients treated by first line chemotherapy and confirmed that CTC levels are an independent prognostic factor for PFS and overall survival (OS). A secondary pre-planned endpoint was to compare prospectively the positivity rates and the value of CTC (CellSearch®), of serum tumor markers (carcinoembryonic antigen (CEA), cancer antigen 15.3 (CA 15-3), CYFRA 21-1), and of serum non-tumor markers (lactate deshydrogenase (LDH), alkaline phosphatase (ALP)) at baseline and under treatment for PFS prediction, independently from the other known prognostic factors, using univariate analyses and concordance indexes. Results A total of 90% of the patients had at least one elevated blood marker. Blood markers were correlated with poor performance status, high number of metastatic sites and with each other. In particular, CYFRA 21-1, a marker usually used in lung cancer, was elevated in 65% of patients. A total of 86% of patients had either CA 15-3 and/or CYFRA 21-1 elevated at baseline. Each serum marker was associated, when elevated at baseline, with a significantly shorter PFS. Serum marker changes during treatment, assessed either between baseline and week 3 or between baseline and weeks 6 to 9, were significantly associated with PFS, as reported for CTC. Concordance indexes comparison showed no clear superiority of any of the serum marker or CTC for PFS prediction. Conclusions For the purpose of PFS prediction by measuring blood marker changes during treatment, currently available blood-derived markers (CTC and serum markers) had globally similar performances. Besides CEA and CA 15-3, CYFRA 21-1 is commonly elevated in metastatic breast cancer and has a strong prognostic value.


Introduction
Several serum markers have been developed in different types of cancer as tools for non-invasive assessment of the tumor burden, mostly in metastatic patients. Quantitative variations of serum markers are, therefore, often used in several cancer types as noninvasive tools to assess treatment efficiency in metastatic patients. However, the use of serum tumor markers faces several issues and unanswered questions: their specificity and sensitivity are considered as low and no clear consensus exists on what threshold and/or variation should be considered clinically significant and which serum marker to follow.
In breast cancer, the commonly used serum markers are carcinoembryonic antigen (CEA), cancer antigen 27.29 (CA 27.29), and cancer antigen 15.3 (CA 15-3) [1]. CEA is a cell surface glycoprotein involved in cell adhesion, normally not present in the blood of healthy adults. CA 27.29, mostly used in North America, and CA 15-3, mostly used in Europe, correspond to two different epitopes of the same protein, MUC1, which is also a cell surface glycoprotein involved in cell adhesion. CYFRA 21-1, which is a commonly used serum marker in lung cancer [2], consists of cytokeratin-19 fragments which are specifically recognized by two monoclonal antibodies originally derived after injecting breast cancer cells (MCF7 cell line) into mice [3]. Small published reports, together with an empiric background at the Institut Curie, suggested that CYFRA 21-1 is commonly elevated and may be used for the management of metastatic breast cancers [4,5]. The 2007 American Society of Clinical Oncology update on tumor markers in breast cancer [6] reported that for monitoring patients with metastatic disease during therapy, CA 27.29 or CA 15-3 can be used in conjunction with other monitoring tools, such as tumor response radiological assessment.
In 2004, circulating tumor cells (CTC) detection in blood by the CellSearch ® system (Veridex, Raritan NJ, USA) was reported to be a prognostic marker in a study by Cristofanilli et al. [7]; moreover, changes in CTC count after one cycle of chemotherapy were associated with progression-free survival (PFS) [8]. On the basis of this first study, the FDA cleared the use of the Cell-Search ® system as a tool for monitoring chemotherapy in metastatic breast cancer patients, although no comparison with serum marker was initially reported. The prospective multicentric IC 2006-04 study was initiated in 2006 as a confirmatory study of the Cristofanilli's study, with the correlation of CTC changes and survival as the primary objective. We have recently reported this confirmatory objective of the study, which was clearly reached: CTC changes are a strong prognostic factor for both PFS and overall survival (OS) [9]. We report here, for the first time, the comparison of CTC with different serum tumor markers (CEA, CA 15-3, CYFRA 21-1) and non-tumor markers (lactate deshydrogenase LDH, alkaline phosphatase ALP), which was a prospectively planned secondary objective of the IC 2006-04 study.

Materials and methods
This prospective study was approved by the national ethics board, identified as DGS 2006-A00523-48 (France) and NCT00898014 (USA) and was conducted in five different French comprehensive cancer centers.

Patients and treatment
The main eligibility requirements for this study included the patient's written informed consent, metastatic breast cancer, patients entering first-line chemotherapy (chosen by clinicians) with or without targeted therapy, life expectancy of at least three months, and measurable or evaluable disease. Clinical evaluations were conducted as usual, but were mandatory at inclusion (before cycle 1, that is, week 0) and at the first radiological evaluation before cycle 3 or 4 (C3/4, that is, weeks 6 to 9). Radiological evaluations (RECIST [10]) were mandatory at inclusion and before C3/4. The following blood tests were obtained at inclusion, before cycle 2 (C2), C3/4, and at progression: complete blood count, liver function (including LDH and ALP), serum calcium and serum markers: CEA, CA 15-3 and CYFRA 21-1. These blood tests were processed locally and disclosed to clinicians. CTC were analyzed at four different time-points in an experienced laboratory (Institut Curie, Paris): at inclusion (before starting treatment), before C2, before C3/4 and at disease progression (not shown). CTC counts were not disclosed to clinicians. Technical details of the CellSearch ® technique have been described elsewhere [11].

Objectives and statistics
The study was powered for its main objective that consisted of detecting among patients with ≥ 5 CTC/7.5 ml at baseline, a 35% difference in six-month PFS rates between patients with a CTC count < or ≥ 5 CTC/7.5 ml before C2. The study was not specially powered for its planned secondary objectives, including this comparison with serum tumor markers. Data are expressed as means or numbers (%). Categorical variables were compared by chi-square or Fisher's exact test, and continuous variables were compared by Student's t-test or Wilcoxon's rank-sum test. Kaplan-Meier progression free survival (PFS) curves were computed according to CTC counts (< or ≥ 5 CTC/7.5 ml) and other tumor markers (< or ≥ Upper Limit of Normal Value (ULNV)), and were compared using log-rank tests. Progression was defined as local or distant relapse or death by any cause. Concordance index was also computed for each marker as continuous variables, PFS as outcome. Univariate analysis was performed using hazard ratios and 95% CI for PFS for each clinical characteristic and for each tumor marker or CTCs separately. A concordance index for PFS based on a clinical model (a model including all the above clinical characteristics) was used to show by how much each tumor marker or CTCs improved the performance of the clinical model in a multivariate analysis. Significance was defined as P ≤ 0.05. Statistical analyses were performed using R2.12.1 software (Wien University, Wien, Austria) [12] Results Detection rates at baseline CTC and serum marker values at inclusion repartition in percentile, mean, median range are given in Table 1 and Figure 1. Values for serum markers are expressed in ULNV: upper limit of normal value. Table 2 shows elevated serum marker (> ULNV) and CTC (≥ 5 CTC/ml) incidence rates at baseline: CA 15-3 and CYFRA 21-1 were the two most commonly elevated serum markers. Serum markers and CTC were highly correlated to performance status, number of metastatic sites and to each other. Table 3 shows the percentage of patients who had at least one marker elevated at baseline according to different marker combinations. As expected, this percentage globally increases with the number of markers assessed. However, the combination of CA 15-3 and CYFRA 21-1 retrieved almost the same positivity rate (86%) than all four markers (90%).

Single assessment and association with PFS
When elevated, all markers tested at baseline had a negative prognostic impact in univariate analysis, as shown in Figure 2. The concordance index (and 95% confidence interval), which quantifies the quality of ranking and is a common performance measure for assessing learned models in survival analysis, was calculated for each marker, with PFS as the outcome. Concordance indexes with PFS were, at baseline, C-index ALP = 0.56 (0.50 to 0.61) (n = 238), C-index CA15-3 = 0.56 (0.51 to 0.61) (n = 244), C-index CEA = 0.58 (0.52 to 0.64) (n = 210), C-index CTC = 0.61 (0.56 to 0.66) (n = 257), C-index CYFRA21-1 = 0.67 (0.62 to 0.72) (n = 190) and C-index LDH = 0.68 (0.63 to 0.73) (n = 218). These results showed that the markers' performances were globally close to each other, with ALP having the worst result and CYFRA 21-1 and LDH the best. Similar results were obtained when markers were assessed during treatment, before cycle 2 (three to four weeks later, Figure 3) and before cycles 3 to 4 (six to nine weeks later, not shown).
In a univariate analysis for PFS, performance status, number of metastatic sites and triple-negative status were statistically significant for PFS (Table 4). At multivariate analysis, only performance status and triple negative tumor status were independent clinical prognostic factors and were included in a clinical model. CTC, LDH, CYFRA21.1, CEA and CA15.3 were statistically significant for PFS and not ALP. When the concordance index was calculated for each tumor marker or CTCs added to the clinical model, all of them, except ALP, slightly improved the performance of the clinical model. The C-index of the clinical model (0.697) increased from 0.71 to 0.723, except with ALP.

Changes under treatment and association with PFS
By combining marker status (elevated or not) before cycle 1 and before cycle 2, four subgroups could be separated for each marker. Figure 4 shows that these early changes were highly correlated with PFS in univariate analysis for all the markers; globally, patients with non-elevated markers at both time points had longer PFS; patients with elevated markers at both time points had generally the worst PFS. However, even in these groups of worst prognosis, median PFS was longer than six months for every marker tested, suggesting that none of these marker's early changes can predict accurately very short PFS. Concordance indexes with PFS were C-index CTC = 0.57 (0.51 to 0.63) (n = 201), Cindex ALP = 0.50 (0.44 to 0.56) (n = 195), C-index CEA = 0.45 (0.38 to 0.51) (n = 157), C-index CA15-3 = 0.50 (0.44 to 0.56) (n = 193), C-index CYFRA21-1 = 0.57 (0.50 to 0.65) (n = 142) and C-index LDH = 0.63 (0.55 to 0.70) (n = 136). Again, these results showed that marker performances were extremely close to each other.
As a spurious early rise of serum markers may occur during the first weeks of a new chemotherapy and may lower the above serum markers' concordance indexes, we also studied the changes that occur before cycle 1 and before cycles 3 to 4; Figure 5 shows that these changes were also significant for all the markers. We calculated the concordance indexes by combining marker status before cycle 1 and before cycle 3/4 (weeks 6

Discussion
The need for novel independent prognostic factors in metastatic breast cancer patients is much lower than the need for dynamic blood markers, which can indicate the treatment efficiency in a reliable and early fashion. Serum tumor markers are an easy, quick, cheap, but rather imprecise and sometimes misleading tool, to monitor the treatment efficacy. However, they are  particularly valuable for treatment monitoring in patients that have disease that cannot be evaluated by radiology. CTC count by the CellSearch ® system is a validated prognostic factor at baseline, but is also used for treatment monitoring [13][14][15]. This test is, however, more expensive than serum tumor markers and has never been directly compared with them. The IC 2006-04 is the first large prospective study in which a comparison between CTC and five different serum markers was pre-planned. At first, our study tried to answer a recurrent issue for medical oncologists, that is "which serum marker should  I ask for?" To be considered as informative and usable for further determination under therapy, a serum marker should be elevated at baseline, before the start of therapy. Historically, comparison between serum markers mostly used positivity rates (> ULNV) at baseline in metastatic patients [16]. For serum markers and CTC, incidence rates were in line with those previously reported at first metastatic relapse [17,18], together with the low incidence of CEA in triple-negative cancers [19].
Our study shows that repeated CTC counts are Single marker positivity rates are given for each of the markers, Table 2.  informative mainly for patients who have an elevated level at baseline (≥ 5CTC/7.5 ml), but not for patients initially CTC negative (< 5CTC/7.5 ml), as only two of these patients will present an increase (≥ 5CTC/7.5 ml) during treatment (Figure 3). Due to the very small number of these patients, no conclusion can be drawn on the prognostics of this subgroup. On the basis of baseline positivity rate comparison, assessing only CEA and CA 15-3, the two most commonly used breast cancer serum markers retrieve around 75% of patients with at least one marker elevated at baseline. Our study has also investigated CYFRA 21-1 as a breast cancer serum marker, which turned out to be the most commonly elevated serum tumor marker (65% of patients), as already suggested in smaller studies [20,21]. Adding either CYFRA 21-1 or CTC to this CEA and CA 15-3 combination further increases this percentage to around 90% of patients, but assessing the full four-marker panel (CEA, CA 15-3, CYFRA 21-1 and CTC) did not further improve the overall positivity rate. Among the different markers tested, a drop in positivity rates between baseline and cycle 3/4 was mostly observed for CTC and CYFRA 21-1: 44 to 13% and 65 to 27% respectively ( Table 2).
We showed that serum markers and CTC positivity were highly correlated with other known clinical prognostic factors, such as poor performance status or the high number of metastatic sites. Interestingly, neither serum markers nor CTC detection were correlated with immunohistological subtypes (hormone receptors-positive, HER2-positive or triple-negative breast cancers). Unsurprisingly, each of the markers tested had a significant impact on PFS in univariate analysis. Multivariate analyses, including serum markers, CTC and known clinical prognostic factors, have been reported previously: CTC and CEA were the two blood markers independently associated with PFS, whereas CTC count was the only blood marker independently associated with OS [9].
Here, by comparing the early and late changes of five blood markers together with CTC changes for PFS prediction, we showed no clear superiority of CTC over the other serum markers. This result was, however, not the primary endpoint of our study, and the statistical power of these analyses may still be discussed, although performed in more than 200 patients. For this analysis, we used the "prognosis-optimized" threshold of ≥ 5 CTC/7.5 ml, which was initially defined as the best dichotomizing threshold for PFS and OS prediction by CTC at baseline and under treatment [7]. Contrarily, for serum tumor markers, the upper limits of normal values (ULNV) were historically defined by healthy-donor analyses, and no cut-off has been optimized previously for survival prediction. In this study, it was pre-analytically planned to use ULNV as the positivity threshold at both baseline and before cycle 2 and, as for CTC counts, to divide the population into four subgroups according to the marker at baseline and before cycle 2. Other models may have been used for serum markers, in order to take into account the relative changes of elevated marker (for example, 50% decrease between baseline at cycle 2...), but testing all the possibilities for serum marker analyses would have led to multiple testing biases favoring serum markers over CTC.

Conclusion
Finally, the IC 2006-04 study suggests that CTC count should not be considered alone, but should be put into the current clinical context of metastatic breast cancer. We previously demonstrated that CTC, but not serum markers, has an independent prognostic value in multivariate analysis. This prognostic value did not interact with any tested subgroup. However, when focusing on the different blood markers assessable for PFS prediction before and during treatment, we report here that there is no clear superiority of CTC count over serum markers. Markers predicting or monitoring treatment efficiency are much more needed for patients with metastatic breast cancer than prognostic serum markers. A further prospective comparative study would be required to obtain a definite conclusion with a predefined statistical power. As, to our knowledge, no such study is ongoing, we believe that the IC 2006-04 will remain the largest prospective study with a CTC/serum marker comparison. The interest of implementing CTC count into the routine management of metastatic breast cancer chemotherapy will be finally assessed in the two ongoing randomized trials SWOG500 (USA) and CirCe01 (France).