Sensors and Actuators A: Physical: Sciencedirect
Sensors and Actuators A: Physical: Sciencedirect
Sensors and Actuators A: Physical: Sciencedirect
a r t i c l e i n f o a b s t r a c t
Article history: This paper illustrates the potential of microwave frequencies for biological analysis and cell discrim-
Received 30 September 2013 ination. Microwave electric fields have the capability to penetrate inside cells and interact with their
Received in revised form 25 February 2014 intracellular content. Hence, measuring cell electrical properties in the Gigahertz frequency spectrum
Accepted 10 March 2014
may allow distinguishing intracellular differences between cells without requiring any labeling or dena-
Available online 4 April 2014
turation. Here, we present how microwave resonators can be used as biosensors to measure individual
cell dielectric permittivity and obtain characteristic electromagnetic signatures as a function of the cell
Keywords:
type considered, in particular their cancer cell stage (i.e. aggressiveness level). Five cell lines, all derived
Colorectal cancer
Biosensors
from low to high grade human colorectal tumors, were cultured on-chip or simply deposited in water
Microwave resonators droplets on microwave biosensors. For this preliminary study, the sensors only operated in air for fre-
High frequency dielectric spectroscopy quencies ranging from 5 GHz to 14 GHz, allowing establishing a representative electromagnetic signature.
Cancer detection During our experiments, sensors were fully dried, and a fixation step allowed the reproducible character-
ization of cells outside their culture medium, in ambient air, while maintaining their intracellular content
unmodified. Results showed significant electromagnetic signature differences between cell lines, espe-
cially between cells with low and high aggressiveness levels. We also show that a correlation might be
envisioned between the measured electromagnetic signatures and the potential aggressiveness stage. At
this level, we stand at the proof-of-concept stage, and a clinical investigation will be required to establish
from several patients both the reliability and the reproducibility of the potentially existing relationship
between electromagnetic signatures identified in this study and actual cancer progression.
© 2014 Elsevier B.V. All rights reserved.
http://dx.doi.org/10.1016/j.sna.2014.03.022
0924-4247/© 2014 Elsevier B.V. All rights reserved.
406 L.Y. Zhang et al. / Sensors and Actuators A 216 (2014) 405–416
the early diagnosis of cancer is crucial for cancer prevention and content close to live state while measuring the cells outside of their
for significantly decreasing the associated mortality rates [10], par- aqueous and ionic medium.
ticularly for cancers that currently present a low recovery rate or In the following sections, we will discuss the biosensor opera-
affect a large number of people every year. In this context, this work tion principle, design and fabrication. Material and methods used
focuses on colorectal cancer, which is the third most common can- for this study will also be detailed. Finally the data collected though
cer worldwide and unfortunately causes an estimated one million experiments performed on the five selected colorectal cancer cell
deaths annually [11]. lines will be presented and discussed.
Conventionally, the cancer diagnosis is based on the staging
of tumors. The defined stages range from 0 to IV depending on
the properties of cells that make up the tumor. When the stage 2. Label-free biological cell sensing sing microwaves
increases, the malignant cells are more aggressive, have a propen-
sity to proliferate and can produce metastases in remote tissues. At 2.1. Microwave detection principle and sensor design
this stage, the efficiency of available treatments generally decreases
significantly and the chances of recovery are very low. The microwave spectrum has long been considered as a promis-
Up to now, biologists have no specific label available to accu- ing frequency domain for cellular analysis. Indeed, water-based
rately determine the tumor stage. The diagnosis is based on samples, such as biological cells, are known to exhibit strong dielec-
macroscopic and microscopic observations of tissues and/or cells tric property dispersion at such frequencies [17]. Hence, according
from tumor biopsies. Tumor staging systems classify the micro- to their tissue of origin and/or pathological state, it has already
scopic cell appearance abnormalities according to changes in been shown that cells seem to display significant differences in
morphology, increase in cell proliferation index or nucleus size. their responses to microwave stimuli [14,18]. In previous works
This method is efficient but presents some disadvantages: (i) it is [23–25], the authors have developed microwave biosensors able to
complex and invasive since a biopsy or a resection is required; (ii) achieve dielectric spectroscopy analysis at the cellular scale. Once
it is not an automated process and only a few sections (thin slices) carefully calibrated and properly used, such sensors allow measur-
of the suspicious tissue are observed which lacks sensitivity due to ing the average intracellular dielectric permittivity dispersion over
visual-only observation; (iii) finally, this is a relatively expensive frequency, and thus establishing associated electromagnetic (EM)
method which requires time and double reading of tissue slides by signatures for selected cells. Fully label-free, this method allows an
two pathologists in order to limit errors. accurate dielectric intracellular analysis performed at the unicellu-
That is why an alternative and automated process would be of lar scale.
great interest to help pathologists to complete their diagnosis. The To design these sensors, a meandered inductor and two inter-
discrimination between different stages of tumor would be based digitated comb (IDC) capacitors are set in parallel and implemented
for example on detection of very aggressive cells present inside the in a microwave coplanar waveguide as illustrated in Fig. 1 pho-
tumor. Such cells should present intracellular content differences tographs. From an electromagnetic point of view, such device
(i.e. change in ionic composition, water proportion, shape and size works as a stop-band LC resonator for which the transmitted signal
of nucleus, etc.) compared to normal or less active malignant cells. through the device is strongly attenuated when the sensor res-
Hence one can notice that a highly sensitive detection method is onates. Therefore, the LC resonator S21 parameter, which relies on
required to achieve such discrimination between cells. sensor frequency response, shows a minimum value at this specific
In this context, considering that the cell plasma membrane is resonance frequency. Indeed, S21 is the ratio between amounts of
transparent at microwave frequencies [12], it seems particularly RF power transmitted though the sensor and amounts of power
pertinent to use such electric fields, in order to probe the cell cyto- applied at the sensor input.
plasm and estimate its dielectric characteristics. It is well known The main reason for using resonant microwave sensors for this
that changes in a cell, for example from a healthy to a pathological study is their ability to achieve an accurate microwave dielectric
state, result in variations of its electrical properties. As an exam- spectroscopy analysis even if a very low biological samples num-
ple, the electrical conductivity and permittivity of normal tissues ber are involved. Indeed as illustrated in Fig. 2 example where only
are lower than those of cancerous tissues [13,14]. Indeed con- three cells load the sensor, such design allows benefiting from high
ductivity and permittivity of biological tissues depend on several detection sensitivity insofar as any dielectric perturbation on the
factors: concentration of relevant ions (potassium, sodium, chlo- comb capacitors results in a sensor resonance change that can be
ride, magnesium, etc.) [15], proteins and other macromolecules, easily observed and measured: the resonance frequency is shifted
water content, nucleus/cytoplasm ratio [16], etc. As a consequence, to a lower frequency and a slight microwave signal attenuation
tumor stage differences in individual malignant cells may also most change occurs at the resonance. Hence, to assess the own cell intra-
likely result in electrical property differences. cellular medium permittivity with a relatively good accuracy at
Biological investigations at the tissue scale with microwaves this scale, we monitor sensor resonance frequency changes, once
have been done for several decades [17,18]. Recently some pub- cells are deposited or cultured on sensing IDC capacitances. In fact,
lished works have demonstrated the possibility to detect and when sensor capacitors are loaded with few dielectric particles, the
analyze dielectric properties with RF/microwave-based biosensors resulting resonant frequency shift depends on three main param-
at the cell scale [19–21]. Especially, microwave sensors working eters: the number of particles involved, their individual size and
with a resonant principle have been tested exhibiting a superior associated volume, and the average relative dielectric permittivity
sensitivity allowing the single cell detection and analysis [22]. value of their content considered at the sensor resonance frequency
The experiments presented in this paper focus on the microwave [23]. Now, taking into account these three parameters, the reso-
analysis of five colorectal tumor cell lines derived from cancerous nant biosensors can be calibrated to become sensitive to the nature
tissues at different stages. For this study, a cell electromagnetic sig- and pathological state of analyzed cells. Using an appropriated
nature has been either established or initiated for these five cell modeling, the dielectric properties of individual cells can be then
populations, using microwave sensors following similar topologies estimated at the resonance frequency of each sensor used. With
and design principles than those presented in [23–25]. In this frame, such approach, an EM signature over frequency can be established
these cell lines were characterized in ambient air with several reso- for investigated cells. As it is shown in this paper, this signature
nant biosensors. An appropriate cell fixation procedure was applied seems to be representative of the original type and pathological
prior to microwave measurements to maintain their intracellular state of cells.
L.Y. Zhang et al. / Sensors and Actuators A 216 (2014) 405–416 407
Fig. 1. Main characteristics of few passive resonant sensors used in this study, including for each of them: a microscope view illustrating geometry design changes between
sensors, the unloaded LC resonator frequency response (before cell deposition), and the associated equivalent electrical model with computed circuit element values.
Fig. 2. Resonance frequency shift measured on S1 type passive sensor once loaded with 3 DLD-1 cells; also including a bottom microscope (upper panel) and a scanning
electronic microscope (SEM) (lower panel) views of the interdigitated capacitor onto which cells have been deposited, fixed and dried.
To collect the experimental data presented in this work, two different passive sensor designs of which four are shown in Fig. 1.
biosensor design families were employed. The first one is based This sensor type, allows characterizing the intracellular cell dielec-
on a passive LC resonator design that presents a fixed reso- tric permittivity only at a single frequency. As discuss previously,
nance frequency. For this study, we have worked with around ten this frequency actually corresponds to the sensor resonance. To
408 L.Y. Zhang et al. / Sensors and Actuators A 216 (2014) 405–416
Table 1 500 MHz continuous frequency tuning. In fact for present sensors,
Measured resonance frequency vs varactor bias voltage for active sensors.
this tuning range is intrinsically limited by the moderate capac-
Bias (V) itance tuning ratio (∼2) achievable with MA46H146 varactors.
However, other alternative tuning components can be considered
0 0.5 1 1.5 2 2.5 3 4 5
for this application. As an example, it would be interesting to use
Sensor SA
digitally tunable capacitors with fine tuning step and high Q perfor-
f0 (MHz) 4954 5107 5212 5284 5344 5398 5437 5506 5557
Sensor SB mance in order to enlarge the analysis sensor bandwidth. Such high
f0 (MHz) 5668 5782 5872 5941 5995 6043 6088 6121 6151 performance capacitor banks are indeed now available on CMOS
Silicon-On-Sapphire [26] or RF-MEMS technologies [27], and their
integration on sensors would result in a compromise between tun-
obtain wideband EM signatures on a cell line, several cells of this able component size and connectivity constraints, and the sensor
line have to be analyzed with different sensors working at differ- frequency tuning, sensitivity and detection capabilities.
ent frequencies. In this case, we used sensors presenting different Another solution is to combine several sensors with different
meander inductor geometries (i.e. different inductor length) that inductor geometries. Indeed, with a proper design, the covered
allow us spreading their resonance frequency over the 5–14 GHz analysis spectrum can be easily enlarged using sensors presenting
band for the present study. adjacent frequency bandwidths as in the example of the SA and
Actually, in order to limit the number of sensors required to SB sensors shown in Fig. 3. Hence, the number of sensors required
establish a cell line EM signature, frequency tunable resonators to establish a representative and wide band EM signature can be
were developed. This second sensor generation integrates an active strongly reduced.
element allowing adjustment of the LC resonator’s capacitor value,
making the sensor resonance frequency tunable (Fig. 3). Hence, 2.2. From measured sensor resonance frequency shift to cell
the sensor working frequency can be continuously adjusted on a electromagnetic signature plot
predefined band. It means that, for each frequency of this band,
it becomes possible to assess the expected cell dielectric proper- In this section, the methodology used to establish the cell EM sig-
ties measured this time starting from the same biological samples. nature is detailed and discussed. Actually, to obtain a representative
In the present case, the active component is a surface-mounted signature, we have seen that several sensors are needed in order to
semiconductor diode varactor (MA46H146 from MA-COM). As ensure cell dielectric property measurements on a sufficiently wide
illustrated in Fig. 3, this tunable capacitor is set in parallel with frequency band. Hence during this measurement campaign, each
the LC resonator. of these sensors is loaded with cells derived from the same line. At
The capacitance value can be changed by applying an appropri- the end, the dielectric permittivity of the intracellular content of
ated DC voltage to the diode through an integrated biasing network selected cell is computed using to the resonance frequency change
(biasing resistor and DC block capacitance). More details on the sen- detected and recorded on each of these different sensors.
sor design and operation can be found in [24]. As detailed in Table 1, During microwave measurements, there is a quasi-uniform
the frequency tunable sensors used for this study exhibit at least electromagnetic field between the used sensing electrodes. Indeed,
Fig. 3. Microscope view of the three tunable resonance frequency sensors used in this work with their tunable unloaded LC resonator frequency responses (before cell
deposition) and their associated equivalent electrical models with computed circuit element values.
L.Y. Zhang et al. / Sensors and Actuators A 216 (2014) 405–416 409
Table 2
Main characteristics of investigated cell lines.
Cell morphology
as it will be detailed later on in this paper, this is made possible Insofar as they come from the same origin and during their culture
thanks to the polymer coating which covers the sensor surface. As they were not subject to any specific treatment, they should keep
one can see in Fig. 5, in this polymer a window is specifically opened the same pathological state. We also consider an identical intracel-
above each sensing capacitor to form a small micro-chamber. Actu- lular volume for each cell coming from the same line. This volume
ally, regarding the design of this chamber, all parts of IDC electrodes is computed from size measurements done on a representative
which present a non-uniform electromagnetic field distribution cell population (20,000 up to 50,000 cells) using a Coulter Counter
are hidden under the polymer. As a result, whatever the cell loca- apparatus. One should notice that there is inevitably a size distri-
tion between and along the comb capacitance fingers, its effect bution on such living cell population; actually standard deviation
and influence on sensor will be absolutely the same. Moreover, generally ranges between 1 and 2 m on cell diameter measure-
one can observe in scanning electronic microscope image shown ments. Nevertheless, as we can notice in data reported in Table 2,
in Fig. 2 that once cells have adhered on the substrate right at the difference between the mean cell diameter and its median value
the micro-chamber bottom; they generally entirely fill the air gap is very small (<2.5%) for the investigated cell lines. Its means, that
width between IDC electrodes. This is mainly due to the fact that the cell size dispersion actually follows very narrow Gaussian dis-
the sensor drying favors a cell drop down and flattening in the tribution and a large cell number may have very close intracellular
bottom of capacitor interdigitated gaps. Hence, once cells are load- volume. Indeed, immortalized cell lines are well known to show
ing a sensor, the applied microwave electromagnetic field should homogeneous morphological characteristics unlike cells coming
interact with the whole cell cytoplasm, including proteins, nucleic from primary cultures. In the end, it seems that the induced error
acids, electrolytes and water content. To fulfill this objective, the on the cell dielectric permittivity value might remain acceptable if
current sensor configuration demands to work outside of an aque- we consider the same intracellular volume for one cell to another
ous environment. Therefore, we chose to perform measurements as long as they come from the same cell line.
on fixed cells, which are more likely to handle such environ- Now, to establish representative EM signatures of selected cell
mental constraints without alteration. For our experiments, we lines, we focused on computations of the real part of the cell relative
selected a formaldehyde fixation protocol that is considered as a permittivity, which gives relevant information on the intracellular
gold standard by biologists to study cell morphology, organelles, content specificities as shown in the next section. The imaginary
proteins (antigens) and nucleic acids, even if not the best suited part of the cell permittivity, which relies on the intracellular dielec-
for some particular applications [29]. Aldehydes are commonly tric loss and ionic conductivity, would be also very interesting to
employed in biology to prepare tissues and cells for manipulation know. However, our sensor detection capabilities do not currently
outside of a survival medium. They act by crosslinking proteins allow us to measure its effect on the sensor frequency response
and thus “freezing” a tissue or cell in its native state. Here, we are with a satisfactory and reproducible accuracy.
working in a single-cell rather than a whole tissue context. How- In our previous works, 3D fullwave EM finite element model-
ever, contrary to whole tissues, the effects of fixation procedures ing of both unloaded and cell loaded sensor were used to extract
at a single-cell level, including formaldehyde, have been relatively the real part of the cell relative permittivity from measurements
understudied. The protocol that we have chosen ensures that water [22,23]. Such EM computation allowed accurately fitting the sen-
content does not leak out of the cell through an altered membrane, sor measured frequency response with the cost of a quite long
at least for a timeframe that we will discuss later. simulation time. However, for this study, an analytical model was
Finally, the induced perturbation occurring on the sensor res- developed to compute the real part of the intracellular permit-
onance is the result several effects: the amount of water present tivity in real-time during experiments. In this model, the cell is
in cell cytoplasm, its ionic and protein concentrations, the size and assimilated to a uniform dielectric block located between two inter-
own electrical properties of the nucleus and other organelles, etc. digitated capacitor fingers. As illustrated in Fig. 4, to an electrical
The experimental data we can collect with our biosensors do not point of view, this dielectric block acts in fact as an additional capac-
allow us discriminating one of these influences from the others. itor CCell that loads the IDC sensing capacitor and increases its initial
That is why we will consider in following calculus only average value. And, if several cells are involved, their capacitive influence
dielectric properties for the whole cell. Indeed, we model cells as will be cumulative. As a consequence, the LC resonator sees its reso-
homogeneous dielectric particles, defined with their own dielec- nance shifting from frequencies f0 to f1 ; both given by the following
tric permittivity; this dielectric permittivity is supposed to follow equations:
a dispersive frequency dependence as it is commonly observed in
any biological samples holding water [17,30]. Moreover, this model 1
f0 = (1)
does not consider the thin cell membrane, since it can be seen as 2 LCU
transparent in the frequency domain of interest [12].
We also assume that the large majority of cells, coming from the with CU = C0 + C0 or CU = C0 + CVar considering respectively passive
same line, possess identical dielectric permittivity characteristic. or active sensors where C0 is initial capacitance of IDC capacitor
410 L.Y. Zhang et al. / Sensors and Actuators A 216 (2014) 405–416
Fig. 4. Modeling of cell capacitive influence on sensor IDC capacitor; used sensor comb fingers are 4–5 m thick and spaced 13 m (=WIDC ) from each other.
before cell loading (Fig. 1) and Cvar is the varactor capacitance value where ε0 is permittivity of free space.
(Fig. 3). Finally, combining (3) and (4), the real part of the relative per-
1 mittivity of cell can be computed using the following equation:
f1 = (2)
2 LCL WIDC 2 (f0 − f1 )(f0 + f1 )
εr = CU · (5)
Cell
f12 · ε0 · VCell · NCell
with CL = CU + Ccell where CU and CL respectively represent the
overall sensor capacitor without and with cells loaded on and L Regarding this formula, one can notice that the accurate mea-
models the resonator inductive behavior. surement of sensor resonant frequencies f0 and f1 is an important
Hence, knowing the unloaded and loaded sensor resonant fre- issue. In order to limit resulting errors on the cell permittivity
quencies (f0 and f1 ) and the unloaded initial sensor capacitor CU , computation, a polynomial curves fitting technique is applied on
the resulting cell capacitor CCell can be computed using Eq. (3) that sensor microwave measurement data. Combined with the 1 MHz
results from combining (1) and (2). frequency resolution used during RF measurement, this method
CU (f0 − f1 )(f0 + f1 )
allows us determining the sensor resonance frequency with a
CCell = Ncell · CCell = 2
(3) satisfactory accuracy (i.e. error less than 5 MHz). During sensor
f1 measurements, the exact number of cells loaded on sensing capaci-
where NCell represents the number of cells actually loaded on the tors was systematically controlled under microscope. As previously
sensor. discussed, their exact location inside the gap is not an important
In practice, the CU capacitance value is precisely extracted issue and does not change the sensor resonance shift. Consequently,
thanks to dedicated test structures based on LC resonator design the main source of uncertainty comes from the accuracy of the
for which the meander inductor was removed. intracellular volume estimation as it has been previously discussed.
Now, assuming that the cell volume VCell remains constant dur-
ing measurement and assuming that all cells entirely fill the air gap 3. Materials and methods
width (WIDC ) between two IDC fingers, the real part of the relative
permittivity is then given by: 3.1. Cells preparation
CCell · WIDC 2 Experiments presented in this paper were performed with five
εr Cell = (4)
ε0 · VCell · NCell distinct cell lines (WiDr, SW480, SW620, DLD-1 and Colo 205)
L.Y. Zhang et al. / Sensors and Actuators A 216 (2014) 405–416 411
Table 3 -12
Computed relative permittivity of SW620 cell line from collected experimental data
obtained with on chip culture and droplet deposit techniques.
S21 (dB)
Loaded sensor (5 SW620 cells)
94 MHz
Droplet deposit technique Unloaded sensor
f1 (MHz) 6022 7228 7700 7741 9300 -18 0.39 dB Loaded sensor (5 WiDr cells)
εr 54 ± 5 40 ± 4 43 ± 4 39 ± 4 29 ± 3
Unloaded sensor
0.32 dB
One should notice that the experiments achieved on the SW620 64 MHz
cell line and summarized in Fig. 7 were led using the two deposition -21
8 9 10 11 12
techniques: on-chip cell culture and floating cell droplet deposi-
Frequency (GHz)
tion. We can see in Table 3 that whatever the technique, collected
experimental data both were in good agreement: the extracted rel- Fig. 8. Measured resonance frequency shifts at 22 ◦ C on SW620 and Colo 205 cells
ative permittivity was in the same range for the two approaches. were loaded onto the sensors.
These results gave us confidence on the insignificant influence
of the deposition method on EM signatures extracted from tumor (grade II). For these characterizations, only passive sensors
measurements. have been used, with the on-chip cell growth technique. Fig. 8
focuses on two measurements done using two sensors with compa-
4. Results and discussion rable unloaded resonance frequencies, and each sensor loaded with
the same number of cells. Indeed, the first one was loaded with five
A tumor stage is defined with respect to anomalies at the tissue WiDr cells, whereas the second one was loaded with five SW620
scale and invasion of suspect cells inside healthy tissue. Among the cells. As detailed in Table 2, these two cell lines present similar
large number of malignant cells involved in tumor tissues, some intracellular volumes and measurement conditions were the same
of them show stronger aggressiveness than others. They may have for both sensor S-parameter measurements (i.e. room tempera-
the ability to proliferate very quickly, regenerate a whole tumor, ture, VNA calibration, and equipment). Consequently, considering
and even produce metastases in remote tissues. The aggressive- Eq. (5), the observed frequency shift difference between these two
ness of cells making up a tumor may be correlated to the tumor sensors mainly relies on intrinsic dielectric properties differences
stage. Hence, high-grade cancer generally involves a large num- between the two cell lines.
ber of highly aggressive cells [31]. The main objective of this work As for SW620 line, the estimated dielectric permittivity for WiDr
was to determine if electromagnetic signatures of malignant cells cell followed a dispersive behavior as frequency increases (Fig. 9).
derived from low to high grade tumors could present some notice- As discussed in the previous section, this phenomenon stems from
able differences and if these signatures could be correlated to a cell intracellular content dielectric properties, largely dominated by
aggressiveness level. For demonstration purpose, five cancer cell
lines (WiDr, SW480, SW620, DLD-1 and Colo 205) derived from
human colorectal cancer at different stages have been selected.
Fig. 7 illustrates the results obtained from experiments led on
SW620 line (derived from grade III tumor). In this case, data was
recorded during four measurement campaigns following strictly
identical cell preparation conditions. Data was obtained using fif-
teen passive sensors (data labeled with squares) and four tunable
frequency sensors (data labeled with circles, triangles or dia-
monds). To make reading easier, error bars were not plotted for
these last ones; nevertheless, they were within the same range
as passive sensors. More details related to active sensors results
are reported in Table 4 regarding the two sensors previously
introduced in Fig. 3 and Table 1. In the end, data extracted using pas-
sive or tunable frequency sensors were in good agreement, while
tunable frequency sensors allowed collecting a much larger amount
of data from a single experiment.
Similar measurement campaigns have been conducted on a Fig. 9. Real part of relative permittivity extracted from measurements for WiDr and
WiDr cell line derived from a lower aggressiveness stage colon SW620 cells.
Table 4
Real part of SW620 cell dielectric permittivity (εr ), computed from data collected using two tunable frequency sensors.
Bias (V)
Sensor SA
f1 (MHz) 4924 5074 5179 5248 5308 5362 5401 5470 5521
εr 62 ± 7 62 ± 7 59 ± 6 61 ± 6 59 ± 6 58 ± 6 56 ± 6 54 ± 5 53 ± 5
Sensor SB
f1 (MHz) 5647 5761 5851 5920 5974 6022 6064 6130
εr 57 ± 6 54 ± 5 51 ± 5 50 ± 5 55 ± 6 54 ± 6 53 ± 6 52 ± 5
414 L.Y. Zhang et al. / Sensors and Actuators A 216 (2014) 405–416
Table 5
Used Cole–Cole equation parameters to model measured cell permittivity dispersion
as function as frequency.
εr∞ 3 3 3 3 3
ε 87 87 97 97 97
34 ps 30 ps 27 ps 23 ps 17 ps
˛ 0.005 0.005 0.005 0.005 0.05
established for SW480 cells also clearly differs from SW620. One
can notice that the related dielectric permittivity of SW480 cells is
much lower than SW620, as in the WiDr case. This allows us assum-
ing that the tumor grade of the tissue of origin may influence the
dielectric permittivity detected with our sensors.
This hypothesis was further supported by results obtained on
two other cell lines derived from higher grade tumors: DLD-1
(grade III) and Colo 205 (grade IV). Indeed, as presented in Fig. 10a,
computed permittivity frequency dispersion trends clearly show
differences between the DLD-1 and Colo 205 lines; this is also
true for SW480, WiDr and SW620. These additional signatures
confirmed that malignant cells derived from high tumor grade
exhibited the highest dielectric permittivity values, while cells
derived from low grade models displayed much lower permittiv-
ity values. Table 5 summarizes the modeling parameters used to
fit experimental data with Eq. (6) derived from a Cole–Cole model.
These curves are plotted in Fig. 10a for each investigated cell line.
Form this data, it is difficult to draw a general conclusion. Still, one
can notice a trend regarding the time constant parameter used to
model the frequency dependence of the different lines investigated.
Indeed, in the 5–14 GHz band, the frequency dispersion of dielec-
tric properties (i.e. the variation of measured permittivity values vs
frequency) appears to be lower for cells derived from higher grade
tissue than for cells from lower grade tissue.
Fig. 10. (a) Real part of relative permittivity extracted from measurements and
However, the differences in EM signature observed between
Cole–Cole modeling for WiDr, DLD-1, SW620 and Colo 205 cells. (b) Representa-
tion of the possible correlation occurring between malignant cell EM signatures and cells derived from tissue staged at the same grade are currently
their potential aggressiveness degree. not fully explained. It may rely on intracellular features proper
to each considered biological model. Especially, water content
variation related to the overall cell volume or even changes in
the amount of water in the cell cytoplasm. An interesting learn- organelles/nucleus (size, number, physical properties, etc.) may
ing comes from a failed characterization campaign achieved on explain the differences in dielectric characteristics we observed
WiDr cells, which is also illustrated in Fig. 9. During this campaign, for the investigated cell lines. Additional experiments involving
some problems occurred during the cell fixation step; cells were nucleus vs cytoplasm volume ratio measurements or careful anal-
not properly fixed, and they ended up strongly degraded once the ysis of cytoplasmic ions, macromolecules and proteins might help
sensors had dried. Indeed, for the different sensors used, we never gaining more insight into our results.
succeeded in observing stable resonance frequency. Instead, sen- Nevertheless, we believe that the preliminary results presented
sors showed unstable frequency shift that rapidly changed until in this paper are very promising; as they tend to show that the
almost disappearing as if intracellular content evaporated during intracellular EM signature may provide information about the
the few seconds after complete sensor drying. Sensor measure- aggressiveness potential of cells known to be malignant. Indeed, as
ments have been done once S-parameters were stabilized. The illustrated in Fig. 10b, it may be possible to define some dielectric
extracted resulting dielectric permittivity was very low (i.e. less permittivity ranges that may be associated a cell aggressiveness
than 10) and did not show any frequency dispersion. These results degree. In fact, such aggressiveness degree may be correlated to
suggest that measured permittivity differences between cell lines a more or less high probability that tumor holds some cells with
are due to noticeable changes within the intracellular medium. highly invasive capabilities. It is clear that larger investigations in
Regarding the error margins that were considered, uncertain- the frame of a clinical study supported with statistical analysis will
ties on the cell volume and positioning in the sensing IDC slots be required to properly define these ranges and evaluate the rele-
as well as on accurate resonance frequency shift measurements vance of this idea. However, if these results are confirmed, it may
were taken into account. Fig. 9 shows that the computed WiDr be possible to use the analysis of cell dielectric property as a novel
line relative dielectric permittivity was always less than SW620 prognosis indicator for the course of the disease that might be com-
between 5 and 14 GHz. Consequently, as we initially assumed, it plementary to conventional anatomo-pathologic tumor diagnosis.
seems that some EM signature differences exist between these cell Our current circuit designs can be greatly improved for prac-
lines. Therefore, a relationship between intracellular content and tical clinical use. We also believe that microfluidic technologies
tumor aggressiveness stage might be established. This hypothesis can be helpful for single cell analysis in a flow condition. To our
appears to be confirmed by data collected on the SW480 line that point of view, it should be a very good approach for characteriz-
originates from the same patient as SW620 cells, but from a lower ing a large number of cells required to be representative of the cell
grade tissue (grade II). As reported in Fig. 10a, the EM signature population heterogeneity that makes up a tumor. Challenges are
L.Y. Zhang et al. / Sensors and Actuators A 216 (2014) 405–416 415
that sensors have to ensure both accurate flowing cell detection [17] S. Gabriel, R.W. Lau, C. Gabriel, The dielectric properties of biological tissues. II.
Measurements in the frequency range 10 Hz to 20 GHz, Phys. Med. Biol. 41 (11)
in dispersive liquid media and analysis of their own intracellular
(1996) 2271–2293.
dielectric properties. [18] A.P. O’Rourke, et al., Dielectric properties of human normal, malignant and cir-
rhotic liver tissue: in vivo and ex vivo measurements from 0.5 to 20 GHz using
a precision open-ended coaxial probe, Phys. Med. Biol. 52 (2007) 4707–4719.
5. Conclusion
[19] A. Ferrier, et al., A microwave interferometric system for simultaneous actua-
tion and detection of single biological cells, Lab Chip 9 (23) (2009) 3406–3412.
This paper demonstrates the capability of microwave resonant [20] Y. Yang, et al., Distinguishing the viability of a single yeast cell with an ultra-
sensitive radio frequency sensor, Lab Chip 10 (5) (2010) 553–555.
sensors to distinguish some differences in colorectal tumor cells
[21] T. Chen, et al., Microwave biosensor dedicated to the dielectric spectroscopy
from cell scale permittivity measurements. More specifically, the of a single alive biological cell in its culture medium, in: IEEE IMS Conference,
presented work is a proof-of-concept study based on experiments 2013.
achieved on immortalized cancerous cell lines. It demonstrates [22] C. Dalmay, et al., Ultra sensitive biosensor based on impedance spectroscopy
at microwave frequencies for single cell analysis, Sens. Actuators A: Phys. 162
the potential of intracellular dielectric permittivity analysis at (2) (2010) 189–197.
microwave frequencies to evaluate the degree of aggressiveness [23] C. Dalmay, et al., Microwave sensors for stem cell identification and discrim-
on malignant cells. The results presented here show a potential ination, in: IEEE MTT-S International Microwave Symposium Digest, 2010,
pp. 620–623.
relationship between the tumor grade and the resulting EM signa- [24] A. Landoulsi, et al., Tunable frequency resonant biosensors dedicated to
ture of these cells. We are currently working on extending these dielectric permittivity analysis of biological cell cytoplasm, in: IEEE MTT-S
preliminary results with investigations on primary culture cells or International Microwave Symposium Digest, 2013.
[25] L. Zhang, et al., Label-free colorectal cancer cell line bio-sensing using RF res-
cells coming from dissociated fresh cancer patient tissues in order onator, in: Transducers-Eurosensors Conference, 2013.
to prepare future clinical trials. [26] R. Whatley, et al., CMOS based Tunable Matching Networks for cellular handset
The high detection sensitivity achieved on such microwave res- applications, in: IEEE MTT-S International Microwave Symposium Digest, 2011,
1, 4.
onant biosensors allows working at the single cell level. As a result,
[27] S.P. Natarajan, et al., CMOS integrated digital RF MEMS capacitors, in: IEEE 11th
such label-free biosensing technology is really promising to target Topical Meeting on Silicon Monolithic Integrated Circuits in RF Systems (SiRF),
uncommon or rare cells and to provide an early prognosis with the 2011, pp. 173–176.
[28] L. Paavilainen, et al., The impact of tissue fixatives on morphology and antibody-
objective to assess and prevent tumor spreading and metastasis
based protein profiling in tissues and cells, J. Histochem. Cytochem. 58 (2010)
development. In this view, such biosensors could reveal a new and 237–246.
powerful tool to complete conventional medical analysis process, [29] J.N. Russell, J.E. Clements, L. Gama, Quantitation of gene expression in
particularly for early detection of cancer. formaldehyde-fixed and fluorescence-activated sorted cells, PLOS ONE 8 (9)
(2013) e73849.
[30] R. Buchner, et al., The dielectric relaxation of water between 0 ◦ C and 35 ◦ C,
Acknowledgments Chem. Phys. Lett. 306 (1999) 57–63.
[31] D.L. Trainer, et al., Biological characterization and oncogene expression in
human colorectal carcinoma cell lines, Int. J. Cancer 41 (February (2)) (1988)
This work was supported by the French Limousin Regional 287–296.
Council (Région Limousin) the European Commission with FEDER [32] P. Noguchi, et al., Characterization of WiDr: a human colon carcinoma cell line,
(Fonds Européen de Développement Régional). The authors are In Vitro 15 (June) (1979) 401–408.
[33] T.U. Semple, et al., Tumor and lymphoid cell lines from a patient with carcinoma
grateful to D. Passerieux for his help and expertise during RF mea- of the colon for a cytotoxicity model, Cancer Res. 38 (May) (1978) 1345–1355.
surements and F. Sidor and C. Guines for their assistance during [34] S.J. Vermeulen, et al., Did the four human cancer cell lines DLD-1, HCT-15,
biosensor fabrication. HCT-8, and HRT-18 originate from one and the same patient? Cancer Genet.
Cytogenet. 107 (1998) 76–79.
[35] K. Grenier, et al., Recent advances in microwave-based dielectric spectroscopy
References at the cellular level for cancer investigations, IEEE Trans. Microw. Theory Tech.
61 (May (5)) (2013) 2023–2030.
[1] C.J. Choi, et al., Comparison of label-free biosensing in microplate, microfluidic, [36] D. Berube, et al., A comparative study of four open-ended coaxial probe mod-
and spot-based affinity capture assays, Anal. Biochem. 405 (1) (2010) 1–10. els for permittivity measurements of lossy dielectric/biological materials at
[2] J.S. Daniels, N. Pourmand, Label-free impedance biosensor: opportunities and microwave frequencies, IEEE Trans. Microw. Theory Tech. 44 (October (10))
challenges, Electroanalysis 19 (12) (2007) 1239–1257. (1996) 1928–1934.
[3] A. Treizebre, et al., Terahertz spiral planar Goubau line rejector for biological [37] M.A. Stuchly, et al., Measurement of radio frequency permittivity of biological
characterization, Prog. Electromagnet. Res. M 14 (2010) 163–176. tissues with an open-ended coaxial line: Part II: Experimental results, Trans.
[4] H.J. Mulhall, et al., Cancer, pre-cancer and normal oral cells distinguished by Microw. Theory Tech. MTT-30 (1) (1982) 87–91.
dielectrophoresis, Anal. Bioanal. Chem. 401 (8) (2011) 2455–2463. [38] K. Nörtemann, et al., Dielectric properties of aqueous NaCl solutions at
[5] Y.I. Kim, et al., Biosensors for label free detection based on RF and MEMS tech- microwave frequencies, J. Phys. Chem. A 101 (37) (1997) 6864–6869.
nology, Sens. Actuators B: Chem. 119 (2) (2006) 592–599. [39] A. Vander Vorst, et al., RF/Microwave Interaction with Biological Tissues, IEEE
[6] L.A. Flanagan, et al., Unique dielectric properties distinguish stem cells and their Press/Wiley-Intersciences, 2006.
differentiated progeny, Stem Cells 26 (3) (2008) 656–665. [40] G.H. Markx, et al., The dielectric properties of biological cells at radiofrequen-
[7] C.E. Sims, N.L. Allbritton, Analysis of single mammalian cells on-chip, Lab Chip cies: applications in biotechnology, Enzyme Microb. Technol. 25 (August (3–5))
7 (4) (2007) 423–440. (1999) 161–171.
[8] S. Nagrath, et al., Isolation of rare circulating tumor cells in cancer patients by
microchip technology, Nature 450 (2007) 1235–1239. Biographies
[9] A. Rasooly, J. Jacobson, Development of biosensors for cancer clinical testing,
Biosens. Bioelectron. 21 (10) (2006) 1851–1858.
[10] M. Ponz de Leon, et al., Clinical and pathologic prognostic indicators in colorec- L.Y. Zhang (M’76–SM’81–F’87) was born on April 23, 1980, in Benxi, China. He
tal cancer, Cancer 69 (3) (2006) 626–635. received the M.S. degree in electronic science from the University of the Littoral
[11] A. Jemal, et al., Cancer statistics, CA: Cancer J. Clin. 59 (4) (2009) 225–249. Opal Coast, Calais, France, and the Ph.D. degree in electronics from the Univer-
[12] K. Asami, T. Yonezawa, Dielectric behavior of wild-type yeast and vacuole- sity of Western Brittany, Brest, France, in 2009. From 2010 to 2013, he has been
deficient mutant over a frequency range of 10 kHz to 10 GHz, Biophys. J. 71 a postdoctoral associate at the University of Limoges, Limoges, France. His research
(1996) 2192–2200. interests are focused on agility in microwave using ferroelectric thin films and on
[13] K.R. Foster, J.L. Schepps, Dielectric properties of tumor and normal tissues at characterization of microwave properties of dielectric materials.
radio through microwave frequencies, J. Microw. Power 16 (June (2)) (1981)
107–119. C. Bounaix Morand du Puch was born in 1981. He received his BSc in molecular
[14] D.-S. Yoo, The dielectric properties of cancerous tissues in a nude mouse biology from the College of Charleston (Charleston, SC, USA) and both his MSc in
xenograft model, Bioelectromagnetics 25 (October (7)) (2004) 492–497. biotechnology and his Ph.D. (2010) in biochemistry from the University of Greno-
[15] H. Scharfetter, Structural Modelling for Impedance Based Noninvasive Diagnos- ble (Grenoble, France). He has worked in academic institutions and independent
tic Methods, Thesis for the habilitation at the Faculty of Electrical Engineering biotechnology laboratories in USA, Switzerland and France, and currently holds
Technical University Graz, 1999, November. a project manager position at Oncomedics (Limoges, France), developing cancer
[16] M.J. Schroeder, et al., An analysis on the role of water content and state on primary models and techniques for ex vivo studies (chemotherapy sensitivity and
effective permittivity using mixing formulas, J. Biomech. Biomed. Biophys. Eng. resistance assays, drug candidate efficacy). His past and current research interests
2 (1) (2008) 1–11. include venomics as well as damaged-DNA interactomes and repair mechanisms,
416 L.Y. Zhang et al. / Sensors and Actuators A 216 (2014) 405–416
involving methods of cell and molecular biology, analytical chemistry (mass spec- After holding a teaching and research assistant position, he joined the Limousin
trometry) and biophysics (SPR). entrepreneur ship incubator to develop the Oncogramme, an innovative technology
dedicated to individualized cancer treatment profiling. In parallel, he also followed
C. Dalmay received the Ph.D. degree in electrical engineering from the University of the HEC Paris start-up institute program, a HEC School of management dedicated
Limoges, Limoges, France in 2009. She has hold a post doctoral position with the Cen- to the creation and development of innovative business. Dr. Lautrette has authored
tre National de la Recherche Scientifique (CNRS), SATIE–BIOMIS, IFR d’Alembert, ENS several research articles and is the main inventor of the Oncogramme patent. Dr.
Cachan working on the conception of microfluidic biochips for cell electroporation Lautrette is co-founder of Oncomedics.
and cell handling. She has currently an assistant professor position at the University
of Limoges, XLIM laboratory. Her main area of activity involves the development of
S. Giraud received her Ph.D. in biology and health from the University of Limoges
biodevices for cell analysis.
(Limoges, France) in 2005. After holding a teaching and research assistant position,
A. Lacroix was born on November 7, 1979. She received her BSc in biochemistry from she worked for 6 months as a clinical research associate, and in parallel she also
the University of Limoges (Limoges, France) and her MSc in infectiology from the helped developing the Oncogramme, an innovative technology dedicated to indi-
University of Tours (Tours, France). She obtained her Ph.D. in Virology and Cancerol- vidualized cancer treatment profiling. She is co-founder of Oncomedics and has
ogy (2008) from the University of Limoges. After holding a teaching and research been CSO of the company since its creation in 2006.
(notably in glycoproteomics) assistant position (2008–2010), her current research
since 2010 as a postdoctoral associate concerns isolation and characterization of A. Bessaudou received the Habilitation degree from the University of Limoges, Limo-
cancer stem cells on glioblastoma model. ges, France in 2001. She is currently a Professor at the University of Limoges, and
is leading research into XLIM research institute. Her current research activity is
A. Landoulsi was born in Tunis, Tunisia, in 1987. He received the Licence degree focused on innovating materials elaboration and more especially metal-insulator
from the University of Limoges, France, in 2010, and the Master degree in Sci- transition oxides.
ences and Technologies of information and communication, specialty electronics,
optics, telecommunications from the Limoges University, France, in 2012. In October P. Blondy (M’00) received the Ph.D. and Habilitation degrees from the University of
2012, he joined the XLIM-UMR 7252 laboratory, France, where started his Ph.D. in Limoges, Limoges, France, in 1998 and 2003 respectively. From 1998 to 2006, he was
the research and development of RF resonant sensors for the characterization of with the Centre National de la Recherche Scientifique (CNRS), as a Research Engineer
materials at the micro and nano scale. with XLIM Laboratory, where he began research on RF-MEMS technology and its
applications to microwave circuits. He is currently a Professor at the University of
J. Leroy was born on January 1, 1987. He received the Ph.D. degree in electrical engi-
Limoges, where he holds an Institut Universitaire de France research chair. He was
neering from the University of Limoges, France in 2013. Since November 2013, he
a visiting researcher at the University of Michigan, Ann Arbor, USA in 1997, and at
is a postdoctoral associate at the XLIM Institute, Limoges, France. His main area of
the University of California at San Diego, La Jolla, USA in 2006 and 2008. He was an
activity is focused on the characterization of cancer cells using microwave biosen-
Associate Editor for the IEEE Microwave and Wireless Components Letters in 2006.
sors.
He is a member of the IEEE International Microwave Conference Technical Program
C. Mélin was born on October 8, 1985. She received the Ph.D. degree in Biology Committee since 2003, and chair of the Technical Committee 21 on RF-MEMS. He
specialized in cancer stem cells from the University of Limoges, France, in 2012. received the “Outstanding Young Engineer Award” from the IEEE-MTT society in
After holding a teaching and research assistant position (2012–2013), her current 2010. He has authored or co-authored more than 150 papers in refereed journals
research since November 2013 as a postdoctoral associate concerns characterization and conferences, holds 4 patents.
of colorectal cells using microwave biosensors.
M.O. Jauberteau is Medical Doctor (University of Limoges) and Ph.D. (University
F. Lalloué received the Ph.D. degree in Biology and health Sciences from the Uni- of Paris VI) and now Medical professor of Immunology (since 2000 at Uni-
versity of Limoges, Limoges, France in 2001. He is currently a full-time assistant versity of Limoges, Faculty of Medicine). She is the director of the Research
professor at the University of Limoges in the department of Physiology and is lead- Team EA 3842, Homeostasis cellular and Pathologies, (HCP) created in 2004. The
ing research into the homeostasis cellular and pathologies team. His current research research is focused on Oncology, especially mechanisms of cell survival and cancer
activity is focused on Neurobiology and oncology such as normal neural stem cells stem cells.
and tumor stem cells in glioblastoma.
S. Battu was born in 1965. He was pharmacist and received the Ph.D. and Habili- A. Pothier received the Ph.D. degree in electrical engineering from the University
tation degrees from the University of Limoges, Limoges, France, in 1997 and 2003 of Limoges, Limoges, France in 2003. He is currently a full-time researcher with the
respectively. He is currently a full-time Assistant-professor at the University of Limo- Centre National de la Recherche Scientifique (CNRS), XLIM, University of Limoges.
ges, Faculty of Pharmacy, in the department of Analytical Chemistry and is leading His current research activity is focused on both RF-MEMS components application
research into the EA 3842 “Homeostasis cellular and pathologies” team. His cur- for analog communication modules and RF biosensors developments for cell anal-
rent research activity is focused on cancer cells and cancer stem cells sorting by ysis. He is a member of the IEEE International Microwave Conference Technical
Sedimentation Field Flow Fractionation. Program Committee since 2011 and co-chair the IEEE MTT-S Technical Committee
on Biological Effect and Medical Applications of RF and Microwave since 2013. He has
C. Lautrette is CEO of Oncomedics since its creation in 2006. He received his Ph.D. authored or co-authored more than 90 papers in refereed journals and conferences,
in biology and health from the University of Limoges (Limoges, France)in 2003. holds 2 patents.