Detection of circulating tumor cells in blood of metastatic breast cancer patients using a combination of cytokeratin and EpCAM antibodies.

Abstract

Circulating tumor cells (CTCs) are detectable in peripheral blood of metastatic breast cancer patients (MBC). In this paper we evaluate a new CTC separation method based on a combination of anti-EpCAM- and anti-cytokeratin magnetic cell separation with the aim to improve CTC detection with low target antigen densities.
Blood samples of healthy donors spiked with breast cancer cell line HCC1937 were used to determine accuracy and precision of the method. 10 healthy subjects were examined to evaluate specificity. CTC counts in 59 patients with MBC were measured to evaluate the prognostic value on overall survival.
Regression analysis of numbers of recovered vs. spiked HCC1937 cells yielded a coefficient of determination of R(2) = 0.957. The average percentage of cell recovery was 84%. The average within-run coefficient of variation for spiking of 185, 85 and 30 cells was 14%. For spiking of 10 cells the within-run CV was 30%. No CTCs were detected in blood of 10 healthy subjects examined. A standard threshold of 5 CTC/7.5 ml blood as a cut-off point between risk groups led to a highly significant prognostic marker (p < 0.001). To assess the prognostic value of medium CTC levels we additionally considered a low (CTC-L: 0 CTC), a medium (CTC-M: 1-4 CTC) and a high risk group (CTC-H: ≥5 CTC). The effect of this CTC-LMH marker on overall survival was significant as well (p < 0.001). A log-ratio test performed to compare the model with 3 vs. the model with 2 risk groups rejected the model with 2 risk groups (p = 0.026). For CTC as a count variable, we propose an offset reciprocal transformation 1/(1 + x) for overall survival prediction (p < 0.001).
We show that our CTC detection method is feasible and leads to accurate and reliable results. Our data suggest that a refined differentiation between patients with different CTC levels is reasonable.

Full text

Page 1

TECHNICAL ADVANCEOpen Access
Detection of circulating tumor cells in blood of
metastatic breast cancer patients using a
combination of cytokeratin and EpCAM
antibodies
Ulrike Weissenstein1*†, Agnes Schumann2†, Marcus Reif2, Susanne Link3, Ulrike D Toffol-Schmidt1and
Peter Heusser4
Abstract
Background: Circulating tumor cells (CTCs) are detectable in peripheral blood of metastatic breast cancer patients
(MBC). In this paper we evaluate a new CTC separation method based on a combination of anti-EpCAM- and anti-
cytokeratin magnetic cell separation with the aim to improve CTC detection with low target antigen densities.
Methods: Blood samples of healthy donors spiked with breast cancer cell line HCC1937 were used to determine
accuracy and precision of the method. 10 healthy subjects were examined to evaluate specificity. CTC counts in 59
patients with MBC were measured to evaluate the prognostic value on overall survival.
Results: Regression analysis of numbers of recovered vs. spiked HCC1937 cells yielded a coefficient of
determination of R2=0.957. The average percentage of cell recovery was 84%. The average within-run coefficient of
variation for spiking of 185, 85 and 30 cells was 14%. For spiking of 10 cells the within-run CV was 30%. No CTCs
were detected in blood of 10 healthy subjects examined.
A standard threshold of 5 CTC/7.5 ml blood as a cut-off point between risk groups led to a highly significant
prognostic marker (p<0.001). To assess the prognostic value of medium CTC levels we additionally considered a
low (CTC-L: 0 CTC), a medium (CTC-M: 1–4 CTC) and a high risk group (CTC-H: ≥5 CTC). The effect of this CTC-LMH
marker on overall survival was significant as well (p<0.001). A log-ratio test performed to compare the model with
3 vs. the model with 2 risk groups rejected the model with 2 risk groups (p=0.026). For CTC as a count variable, we
propose an offset reciprocal transformation 1/(1+x) for overall survival prediction (p<0.001).
Conclusions: We show that our CTC detection method is feasible and leads to accurate and reliable results. Our
data suggest that a refined differentiation between patients with different CTC levels is reasonable.
Background
In recent years results about the clinical relevance of cir-
culating tumor cells (CTCs) in peripheral blood of
patients with metastatic breast cancer (MBC) and other
tumor types have accumulated [1-6]. Cristofanilli et al.
demonstrated in 177 MBC patients that the number of
CTCs before treatment is an independent predictor of
progression-free and overall survival [2,3]. Elevated CTC
levels during therapy further indicated subsequent rapid
disease progression and mortality for MBC patients [4].
The correlation between CTC count and prognosis has
been confirmed by several studies [5,6].
The main approaches to analyze CTCs derived from
blood are immunological and PCR-based molecular assays.
The frequency of tumor cells among normal blood cells is
assumed to range from 10−5to 10−8[7,8]. Because of this
rareness,CTCsneed to
usually achieved by immunomagnetic separation. As
standard markers for the immunocytochemical detection
of CTCs the epithelial cell adhesion molecule (EpCAM)
be enrichedwhichis
* Correspondence: u.weissenstein@vfk.ch
†Equal contributors
1Society for Cancer Research, Hiscia Institute, Arlesheim, Switzerland
Full list of author information is available at the end of the article
© 2012 Weissenstein et al.; licensee BioMed Central Ltd. This is an Open Access article distributed under the terms of the
Creative Commons Attribution License (http://creativecommons.org/licenses/by/2.0), which permits unrestricted use,
distribution, and reproduction in any medium, provided the original work is properly cited.
Weissenstein et al. BMC Cancer 2012, 12:206
http://www.biomedcentral.com/1471-2407/12/206

Page 2

and cytokeratins (CK) are used. Several CTC assays used
today are based on enrichment with anti-EpCAM anti-
bodies and subsequent detection with anti-cytokeratin, for
example the FDA approved CellSearchTMsystem [9].
EpCAM as well as cytokeratin expressing cells can be
found in peripheral blood of advanced cancer patients but
are rare in healthy donors [1,10]. EpCAM is overexpressed
100- to 1000-fold in primary and metastatic breast cancer
relative to normal breast cells [11]. Breast cancer cells of
all grades typically express the epithelial cytokeratins CK7,
CK8, CK18 and CK19 [12-14]. On the other hand expres-
sion of these antigens can vary widely in breast cancer cells
and there is growing concern about consequences of this
heterogeneity for CTC detection [15-18].
Deng et al. demonstrated the advantage of combining
anti-EpCAM and anti-cytokeratin antibodies for CTC
enrichment which compensates low or missing expres-
sion of either EpCAM or cytokeratins [15].
For the present study we modified a commercially
available tumor cell enrichment and detection assay to
combine anti-EpCAM and anti-cytokeratin for immuno-
magnetic CTC enrichment. Blood samples of healthy
donors spiked with breast cancer cell line HCC1937 were
used to determine accuracy, precision and specificity of
the method. CTC levels of 59 MBC patients were mea-
sured and the prognostic significance regarding overall
survival (OS) was examined. For survival analysis the
conventional threshold of 5 CTCs/7.5 ml was used. But,
only recently a discussion has started about the right
way to use CTC measurements for risk assessments
[19,20]. We particularly examine a further division of the
0–4 CTC group into a low risk group with 0 CTC and a
medium risk group with 1–4 CTC and provide a sensi-
tivity analysis with CTC as a count variable. The defin-
ition of a medium risk group, which was also used by
Botteri et al. [19], was motivated by the assumption that
already a single detected CTC reflects a higher probabil-
ity for forthcoming death.
All analyses were performed according to the RE-
MARK criteria [21,22].
Methods
Patients and blood collection
Our investigation was in compliance with the declaration
of Helsinki. The study was accepted by the Ethics Com-
mittee of Basel, Switzerland (EKBB).
Blood samples were drawn after gathering informed
consent from 10 healthy donors and from 59 MBC
patients in the Lukas Clinic, Arlesheim. Patients with
other tumor diagnoses before the breast cancer diagnosis
were not considered eligible to avoid ambiguities over
the onset of the disease. The blood samples were
obtained between 08/2007 and 08/2008. Last update of
survival information was made in April 2010. The
sample size was dependent on the number of MBC
blood samples send by doctors available within the time
frame.
Blood was collected into Cell Save Preservative Tubes
(Immunicon, Huntington Valley, PA, USA), maintained
at room temperature until processing within a maximum
of 96 hours.
Sample preparation and analysis protocol for circulating
tumor cell detection
CTC samples were prepared using the Carcinoma Cell
Enrichment in combination with the Carcinoma Cell De-
tection Kit supplemented
MicroBeads (Miltenyi Biotech GmbH, Bergisch-Glad-
bach, Germany).
The protocol was carried out according to manufac-
turer’s instructions. In brief, 7.5 ml anti-coagulated pe-
ripheral blood was centrifuged at 400xg for 35 minutes
without brake and afterwards leukocyte enriched inter-
phase (buffy coat) was carefully collected in a volume of
3.5 ml. Cells were permeabilized, fixed and incubated
with FcR blocking reagent for 30 minutes. After the si-
multaneous incubation with anti-cytokeratin (specific for
CK 7/8) - and anti-EpCAM-MicroBeads, anti-cytokera-
tin-alkaline phosphatase (specific for CK 7, 8, 18 and 19)
was added. For the detection and quantification of un-
specific binding due to Fc receptor binding or other pro-
tein-protein interactions mouse IgG1/IgG2a isotype
controls (Miltenyi Biotech GmbH) were used at identical
concentrations and staining conditions as the target pri-
mary antibodies.
Samples were applied to positive selection columns
and placed in the magnetic field of a QuadroMACSTM
separation unit (Miltenyi Biotech GmbH). After washing
with PBS, columns were detached from the cell separator
and targeted cells were eluted.
Eluted target cells were spinned on Silane-Prep Slides
(Sigma-Aldrich Logistik GmbH, Buchs, CH). Cell spots
were dried and incubated with the freshly prepared sub-
strate solution for the alkaline phosphatase color reac-
tion. After mounting and drying, slides were now ready
formicroscopicvisualization
microscope.
Criteria for classification of a cell as circulating tumor
cell were round or oval morphology, positive staining for
cytokeratins and negative corresponding isotype control.
CTC measurements of all patients were performed
blinded to the study endpoint overall survival.
withCD326(EpCAM)
usingabrightfield
Accuracy, precision and specificity of circulating tumor
cell detection
To estimate the accuracy and precision EpCAM+/CK+
breast cancer cell line HCC1937 (German Collection
of Microorganisms and Cell Cultures, Braunschweig,
Weissenstein et al. BMC Cancer 2012, 12:206
http://www.biomedcentral.com/1471-2407/12/206
Page 2 of 8

Page 3

Germany) was spiked into the blood of healthy donors in
duplicates or triplicates with approximately 10, 30, 85
and 185 cells per 7.5 ml blood. Before spiking, the actual
cell number of HCC1937 cell line was determined using
BD TruCount tubes (BD Biosciences, San Jose, CA,
USA), containing a known number of fluorescent beads
and by running of samples on a flow cytometer (FACS
Calibur, BD Biosciences, San Jose, CA). All 28 tubes were
processed within 4 days after blood collection according
to the sample preparation and analysis protocol for cir-
culating tumor cell detection.
To investigate the specificity of the CTC detection
method, 7.5 ml peripheral blood samples of 10 healthy
volunteers (6 female and 4 male) were collected and ana-
lyzed according to the sample preparation and analysis
protocol.
Statistical analysis of CTC as a prognostic marker
Overall survival probability estimates for risk groups
were visualized by Kaplan-Meier plots and compared by
a standard log-rank test or a log-rank test for trend
where appropriate [23]. Baseline blood collection was
taken as the point of origin from which on survival was
estimated. Cox proportional hazards model [23] was
used to calculate p-values, hazard ratios (HR) and to
compare models by a likelihood-ratio test. Median scores
were used for a trend test in the Cox model. As a sensi-
tivity analysis CTC were also included as a count variable
in the Cox model using the supremum test for functional
form [24] to choose an appropriate transformation. The
model assumptions were checked by the supremum test
for proportional hazards [24] and by a calculation of cu-
mulative incidence rates [25] considering loss of patients
due to unknown reason as a competing risk to survival.
Fisher’s exact test was applied to check whether there
were differences in early drop out reasons between risk
groups. Demographic data, therapy information and
other prognostic markers were traced retrospectively
from patient's medical record and compared among risk
groups by Fisher’s exact test and Cochrane-Armitage
trend test. An assessment of the univariate prognostic
value of each of the baseline parameters is performed
and used as a basis for a multivariate Cox proportional
hazards model. The final multivariate model was chosen
by automatic variable selection. All calculations in this
section were done with SASW9.2 for Windows. Statis-
tical two-sided P-values <0.05 were considered signifi-
cant. The statistical analysis of the recovery experiments
was performed using MedCalc for Windows, version
9.3.8.0 (MedCalc Software, Mariakerke, Belgium).
Results
Accuracy, precision and specificity
The results of the recovery tests performed by spiking
varying numbers of HCC1937 breast cancer cells into
blood samples of healthy donors are summarized in
Table 1. The relationship of the number of recovered vs.
the number of spiked tumor cells was linear and regres-
sion analysis yielded a coefficient of determination of
R2=0.957 (P<0.001). The
HCC1937 cell recovery was 84% (95% CI=78–90), the
within-run coefficient of variation for the recovery rate
lay between 12% and 30%. None of the 7.5 ml peripheral
blood samples of the 10 healthy subjects analysed was
found to have CTCs.
averagepercentageof
Prognostic value of CTCs detected in peripheral blood
samples of MBC patients
CTC levels of 59 MBC patients were measured to assess
their prognostic value for overall survival prediction. 20
patients had no CTCs, 15 patients had 1-4 CTCs and 24
had ≥5 CTCs/7.5 ml blood. During the follow-up period
26 of the 59 patients died. Patients were followed as long
as they were accessible to the clinician (17 patients) or
until the end of the overall study (14 patients); 2 patients
were lost because a study clinician left the clinic. The
median follow-up time for the patients still alive at the
end of the study was 85 weeks (range 7-134 weeks).
Following the suggestion in [2] using the Veridex Cell
Search system, we could confirm that differentiating be-
tween patients with ≥5 CTC vs. <5 CTC led to a very
strong risk marker for overall survival (p=0.00006 log-
Table 1 Accuracy and Precision of the CTC detection method
Expected
CTC
count
NN Observed CTC count % Recovery
donorssamples MeanSD95% CIMean SD 95% CI% CV
102682.46–10802455–10530
303825322–27821074–9012
8538761465–88901676–10418
1852615118131–1718210 71–9212
TOTAL528 8415 78–9018
Observed vs. expected numbers of CTCs and recovery measured in blood samples of 5 healthy donors spiked with defined numbers of HCC1937 tumor cells.
SD=standard deviation, CI=confidence interval of the mean, CV=coefficient of variation (SD/mean).
Weissenstein et al. BMC Cancer 2012, 12:206
http://www.biomedcentral.com/1471-2407/12/206
Page 3 of 8

Page 4

rank test). Median overall survival of the ≥5 CTC group
was 60 weeks compared to >100 weeks in the <5 CTC
group; quartile survival times were 22 vs. 85 weeks, re-
spectively. The hazard ratio HR of ≥5 CTC vs. <5 CTC
estimated by Cox model was 4.79 (p=0.0002, 95%
CI=[2.09,11.0]).
To further investigate the relationship between CTC
number and prognosis we defined 3 risk groups, a low
(CTC-L: 0 CTC), a medium (CTC-M: 1-4 CTC) and a
high risk group (CTC-H: ≥5 CTC). The 3-valued marker
CTC-LMH proved to be significant as well (p=0.00002
log-rank test for trend, p=0.0001 trend test with Cox
model). The differentiation between the low and the
medium risk as well as between the medium and the
high risk groups were both significant in the Cox model
(CTC-M vs. CTC-L: HR=4.94, 95% CI=[1.02,23.8],
p=0.046; CTC-H vs. CTC-M: HR=2.55, 95% CI=
[1.04,6.3], p=0.042). A log-ratio test performed to com-
pare the model with 3 vs. the model with 2 risk groups
rejected the model with 2 risk groups (p=0.026). The
median survival was shortest in the high risk group
(60 weeks), followed by the medium (>78 weeks) and
low risk group (>100 weeks). Corresponding quartile
survival times were 22, 32 and 100 weeks (Table 2).
Kaplan-Meier estimates (KM) of survival probabilities
for each risk group are shown in Figure 1. There were
no major differences in early drop out reasons between
risk groups (p=0.63) and the cumulative incidence
curves were similar to 1-KM estimates.
As sensitivity analysis we included CTC as a count
variable into the Cox model without dichotomization.
Due to the positively skewed CTC distribution, the sig-
nificance of CTC as a survival predictor is lost when
CTC counts are used without transformation (p=0.056).
Considering the supremum test for functional form, we
propose an ‘offset reciprocal’ transformation y=1/(1+x)
(p=0.0002, HR 0.063 for a change of 1 on transformed
scale). On our data, the log(x+1)-transformation used
by Botteri et al. [19] did not pass the supremum test for
functional form. Using the ‘offset reciprocal model’ for
successive deterioration from 0 to 5 CTC, we get hazard
ratios of 3.98 (1 vs. 0 CTC), 1.58 (2 vs. 1 CTC), 1.26 (3
vs. 2 CTC), 1.15 (4 vs. 3 CTC) and 1.10 (5 vs. 4 CTC).
This sequence rapidly converges to 1 indicating that the
definition of risk groups is more effective in the low level
range.
From a practical point of view, it is helpful to define
risk groups and describe them by mean characteristics
and respective course of disease. We compared them in
regard to demographic data, treatment and several prog-
nostic markers and evaluated the prognostic value of
each particular parameter for overall survival prediction
(Table 3). Among the predominantly female breast can-
cer patients there was one male patient who belonged to
the high risk group. Two patients in the low risk group
suffered from a secondary tumor (Ovarian/Uterus) diag-
nosed after the breast cancer diagnosis. The mean age at
baseline was 57.1±10 years (CTC-L: 57.6±9, CTC-M:
60.2±10, CTC-H:54.8±11). The fraction of patients with
positive lymph nodes at time of BC diagnosis increased
from the low to the high risk group (p=0.03). There was
a significant difference in the fraction of patients with
bone metastases which was highest in the high risk
group (p=0.02). The high risk group had significantly
more patients with an elevated level for the tumor mar-
ker Carbohydrate Antigen 15-3 (CA 15-3) (p=0.02). The
inflammatory marker C-reactive protein (CRP) displayed
a significant positive trend (p=0.03).
The significance of CTC-LMH is also supported by
multivariate models (Model 1: CTC-LMH (p<0.001),
ER/PR at least one positive (p=0.04), age at BC diagno-
sis >50 years (p=0.03) and T3/T4 at BC diagnosis
(p=0.002); Model 2 additionally stratified by N0 at BC
diagnosisa: CTC-LMH (p=0.001), ER/PR at least one
positive (p=0.005), age at BC diagnosis >50 (p=0.03)
and T3/T4 at BC diagnosis (p=0.08).
Discussion
Most immunocytochemical CTC detection technologies
are based on a separation of CTCs from normal blood
cells with EpCAM antibodies. The exact biological func-
tion of EpCAM is not fully understood and remains con-
troversial. In some publications EpCAM is argued to act
as an intercellular adhesion molecule, and loss of
EpCAM expression therefore reduces cell-cell adhesion,
thereby promoting dissemination of tumor cells [26]. In
contrast, Osta et al. [11] report that silencing EpCAM
gene expression in vitro decreases the proliferation,
Table 2 Survival data of 59 patients stratified into 3 risk groups according to baseline CTC counts
Risk Group Death N
(%)
Censored N
(%)
Quartile Survival
[weeks]
Median Survival
[weeks]
CTC-L (0 CTC) N=20 2 (10.0) 18 (90.0)100 95%CI=[85,+oo]
>100 95%CI=[85,+oo]
CTC-M (1-4 CTC) N=157 (46.7)8 (53.3) 32 95%CI=[12,78]
>78 95%CI=[20,+oo]
CTC-H (≥5 CTC) N=24 17 (70.8)7 (29.2)22 95%CI=[2,42] 60 95%CI=[27,73]
TOTAL N=59 26 (44.1)33 (55.9)46 95%CI=[20,70]100 95%CI=[70,+oo]
Quartile and median survival times are given together with confidence intervals (CI). N is the number of patients in the respective group.
Weissenstein et al. BMC Cancer 2012, 12:206
http://www.biomedcentral.com/1471-2407/12/206
Page 4 of 8

Page 5

Figure 1 Kaplan-Meier plot estimating overall survival for 3 risk groups (0 CTC, 1–4 CTC, ≥5 CTC).
Table 3 Comparison of risk groups on demographic or treatment characteristics and other prognostic markers
Parameter Low Risk
(CTC-L)
CTC=0
(CTC-M) CTC1-4
Medium
Risk
High Risk
(CTC-H)
CTC>=5
TOTALAssociation with
LMH p-Value
Association
with OS
p-value (HR)
N/NG
16/20
% N/NG
13/15
% N/NG
14/24
% N/NG
43/59
%
Patients with age at TSE >5080 8758 73 n.s.n.s.
Patients with age at BC diagnosis >50 9/20 45 10/15 6711/24 4630/59 51n.s. n.s.
T3/T4 at BC diagnosis2/1612.5 4/1040 5/2025 11/4624n.s.
<0.01 (HR=5.1)
N0 at BC diagnosis (NX=missing value)7/16 44 2/1020 2/19 10.511/45 24 0.03 (CAX)0.001
ER/PR at least one positive15/1979 8/1361.518/228241/5476n.s.0.02 (HR=0.37)
HER2 overexpressed4/14 292/11 184/21 1910/46 22n.s.n.s.
CA 15–3 above normal value of 31.3
(≤ 30 days before TSE)
9/13 696/96718/18 10033/40 850.02 (FSHX) 0.03 (CAX)n.s.
CRP above normal value of 5
(≤ 30 days before TSE)
3/12255/862.513/19 6821/39540.03 (CAX) n.s.
Visceral metastases8/20 407/1450 10/2343.525/5744 n.s.n.s.
Nonvisceral metastases20/2010012/1486 22/23 96 54/5795n.s.n.s
Bone metastases14/20 706/14 4320/23 8740/57 70 0.02 (FSHX)n.s.
No. of metastatic sites>=2 9/20458/145711/234828/5749 n.s.0.02 (HR=2.8)
Surgery of primary tumor 17/208513/158719/2479 49/5983n.s.n.s.
Radiation therapy14/19749/13 6916/22 7339/54 72n.s.n.s.
Chemotherapy14/187810/147117/218141/5377n.s.n.s.
Mistletoe therapy 20/2010015/15 10024/2410059/59 100--
Anti-hormone therapy16/19 848/1361.519/23 8343/5578n.s. 0.01 (HR=0.34)
Bisphosponate therapy 11/17 65 6/144317/1989.534/50680.03 (FSHX)n.s.
Patients with ≥2 known therapy lines15/20756/154014/245835/5959n.s.n.s.
N is the number of patients, NGdenotes the number of patients with non-missing values in the respective group. P-values are given if significant (p<0.05), n.s.
denotes “not significant”. LMH denotes the prognostic marker which differentiates between low (L), medium (M) and high (H) CTC levels. Risk groups are
compared by the exact Cochrane Armitage trend test (CAX) or Fishers Exact Test (FSHX). Time of study entry (TSE) is defined as time of baseline blood draw. For
each parameter in the table the univariate prognostic value for overall survival (OS) prediction is given in the last column. P-Values and Hazard Rates (HR) in this
column are computed by Cox Proportional Hazard Regression; except for “N0 at BC diagnosis” the logrank test was used because of no events (deaths) in the
reference group.
Weissenstein et al. BMC Cancer 2012, 12:206
http://www.biomedcentral.com/1471-2407/12/206
Page 5 of 8