Cell Manipulation Is A Fundamental Process Biology Essay

Published: Last Edited:

This essay has been submitted by a student. This is not an example of the work written by our professional essay writers.

Cell manipulation is a fundamental process in many biology and biotechnology areas1, such as cell-based screenings for basic science, surface immunophenotyping for diagnosis, studies of how cell morphology affects differentiation, and detecting pathogenic bacteria in food supplies.

Cell sorting techniques are used to separate cells according to their properties. There are many cell sorting techniques, including panning, fluorescence activated cell sorting (FACS), magnetic cell sorting (MACS) and recent developments based on dielectrophoresis (DEP) to move cells in microfluidic devices2,3. In this paper, we present an automated method to track and quantify the rotation of pigmented cells with diameter in the order of 10µm. Accurate analysis of cell rotation can be potentially used as a sorting criterion for cell identification and separation, based on our group's recent discovery of the self-induced rotational motion of pigmented biological cells in a dielectrophoretic force field4.

The generation of DEP force from a non-uniform electric field is a well-known phenomenon and is briefly described below. When a polarized object is exposed to a non-uniform electric field, a dipole moment is induced and the object will move towards the maxima or the minima of the electric field depending on its relative polarizability with respect to the medium5. According to the theory of electromagnetism, the DEP force acting on a spherical particle, such as cells suspended in a fluid is given by (1):


where R is the radius of the particle, Erms is the root mean square (rms) value of the electric field and K(ω) is the real part of the Clausius-Mosotti (CM) factor. The CM factor is related to the particle dielectric constant and indicates the relative magnitude and direction of the force experienced by the particle and is given by (2):


where εp* and εm* are the complex permittivities of the particle and the medium, respectively. ε* is defined as (3):


where ε and σ refers to the permittivity and the conductivity, respectively.

In practice, during biological experiments, cells may not be treated as isolated, single particles. Usually, there are large numbers of cells in a given sequence of microscopic video frames that captured and stored for cell-motion analyses. Any manual method used to ascertain the cell rotation speed for cell sorting would be slow, difficult, and inaccurate. Therefore, we have developed an automated procedure to analyze microscopic image sequences using algorithms to estimate cell rotation speed. We envision that this procedure will eventually enable the development of a fully automated system for identification, manipulation, and sorting of cells based on the self-induced rotation phenomenon of specific cells. The fabrication process for the microfluidic chips and the DEP experimental setup are detailed in an earlier paper4. This paper will present the DEP-based cell rotation experiments along with details of the rotation speed estimation algorithm with translation motion compensation.

Cell Rotation Experiments

Cell Behavior in a DEP Field

Cells, similar to other micro- and nano-scale entities such as carbon nanotubes, carbon nanoparticles, Au nanoparticles and microbeads, mainly experience two types of forces in an appropriately applied AC electric field: a positive-DEP force (p-DEP), i.e., an attractive force; and a negative-DEP force (n-DEP), i.e., a repelling force. Based on the distinctive electrical phenotypes of different cells, it is possible that by applying the same DEP field, one type of cell would experience an n-DEP, while the other type of cells experiences a p-DEP force. The factors that would affect the cell response in the DEP field include (1) the physical properties of the cells such as shape, size, and mass distribution of the cells, (2) the dielectric property of the cells, as well as (3) the dielectric property of the medium.

Fig. 1 shows the response of typical cells (B16 cells) under a DEP force in a microchannel. In Fig. 1 (a), the cells are experiencing a p-DEP force, and are attracted to the microelectrodes. Pearl chains are formed between the tips of electrodes. In Fig. 1 (b), the cells are experiencing an n-DEP force, and are repelled from the electrodes. The DEP parameters applied during the experiment were 16V voltage and 22MHz frequency. Under these conditions, the cells typically experienced a p-DEP force and are attracted to the electrodes. Then, when the voltage is fixed and the frequency is gradually decreased, the cells become disconnected with the Au electrodes.

DEP Parameters. Self-rotation is observed only when the cells are experiencing a p-DEP force, as applied between the Au microelectrodes. A voltage larger than 5V and a frequency higher than 10M provide a sufficient DEP force for cell rotation.

Geometry and dimensions of the micro-electrodes. Several patterns of electrodes have been used in attempting to induce rotation of cells, including four-probe orthogonal electrode pattern (as shown in Fig. 1), circular, rectangular, and interdigitated. Results showed that even with the same DEP parameters, these patterns do not deliver the same results, as no self-rotation of the pigmented cells is observed in all tested patterns except for the four-probe orthogonal pattern. In addition, we also tried the same four-probe orthogonal electrode pattern with electrode widths of 150µm, 40µm and 20µm. Theoretically, the shrinkage in electrode size should provide a larger DEP force, which could potentially accelerate the self-rotation process. However, the experimental results failed to follow this expectation, as the self-rotation of the pigment cells does not exhibit any dramatic improvement in terms of either rotation speed or the total number of cells that undergo self-rotation motion. On the contrary, a probe electrode with a 150µm width shows better results for the observation of this phenomenon. The probable reason is that when a sufficient DEP force is applied, the larger surface area of a larger electrode would attract a larger number of cells. This induces the possibility of self-rotation of more cells without compromising the rotation speed. In other words, since the diameter of pigment cells was ~10-15µm, therefore, for smaller electrodes, the electrode surface area may not have sufficient force field to induce cell rotation. Also, for 40µm and 20µm electrodes, we only observed alignment of cells between the electrodes, but rarely any rotation..

On the other hand, for non-pigmented cells, which are not as susceptible to the DEP force induced self-rotation, the smaller dimensions of the electrodes exhibit an improved response to the DEP force in terms of cell alignment and attraction. For instance, HaCaT cells (non-pigment cells) are not susceptible to 150µm microelectrodes during DEP manipulation, i.e., no apparent attraction to the electrodes even when a maximum DEP force is applied. However, a cell chain of HaCaT cells could be formed using 40µm microelectrodes. Therefore, for the cell rotation experiments reported in this paper, the dimension of the microelectrodes is 150µm in width, unless otherwise specified.

Medium in the microchannel. Both the viscosity and the dielectric property of the solution are critical for cell manipulation. Ideally, the permittivity of the solution should be as small as possible in order to create a distinct difference between the cell itself and the surrounding solution. Also, the viscosity of the solution should be carefully controlled as the cell movement and rotation could slow down or even completely stop if the solution is too viscous. The solution used for our experiments was 0.2 M sucrose, i.e., 6.84% sucrose solution. Based on prior literature, this solution's estimated conductivity is ~ 1mS/m, and the kinematic viscosity at 20 degree Celsius is ~1.2Ã-10-6 m2/s. (Guanglie: you need to add the reference here)

Pigmented Cells versus Non-pigmented Cells

From our experiments, pigmented cells are more susceptible to DEP force in terms of alignment and attraction to the electrodes, i.e., pigmented cells are able to form a pearl chain of cells between the microelectrodes faster than the non-pigmented cells. Also,as aforementioned, stable and repeatable self-rotation is only observed in pigmented cells. For the other three types of cells, including two types of skin cells (Keratinocyte and HaCaT) and one type of lung cells (A549), the self-rotation phenomenon only occurs occasionally. For instance, self-rotation in a small number (< 5µm) of A549 cells is observed. However, it only lasts for a few seconds and quickly stops after the cells reached a stable position. Also the rotation speed is much slower when compared to the pigmented cells. From our observations, the self-rotation of non-pigmented cells may occur but the duration is short, and the rotation is unstable, unrepeatable and highly susceptible to the flow rate compared to the rotation of pigmented cells.

Although we are still working on a solid theoretical explanation of this self-rotation phenomenon, we highly suspect that it is related to the semiconductive electrical property of the melanin located inside the pigmented cells. Typically, cells in the fluids are modeled as a single or multi-shell sphere with an evenly distributed dielectric property on the cell surface. It is possible that the existence of the melanin would break this "balance" and create an unevenly distributed model with the part containing melanin exhibiting a different electrical property than the rest of the cells. Therefore, this imbalance could lead to the self-rotation of the cells in an AC electrical field, as observed.

Investigation of the Self-Rotation Speed of Pigmented Cells

As is reported in the previous section, pigmented cells exhibit the cell rotation phenomenon under a sufficient p-DEP force. Fig. 2 shows a sequence of photos of a single Melan-A cell self-rotating for one revolution, which takes 320ms. Furthermore, we found that the rotation speed of the cells could be changed by adjusting the DEP parameters, i.e., different applied frequencies and voltages result in variations in the rotation speed. Video records of cell rotation are examined frame by frame, and the rotation speeds under different conditions are calculated. The resulting changes in self-rotation speed under different frequencies while the voltage is kept constant at 16V, as well as the rotation speed of the same cell under different voltages while the frequency is kept constant at 22MHz are shown in Fig. 3 (A) and (B), respectively. The curves in the figure prove that a higher frequency or a higher voltage leads to a high rotation speed of up to 150rpm.

In addition, we also apply different waveforms (i.e., sine wave, square wave and triangular wave) of the AC electrical field to the microfluidic chip and analyze the self-rotation speed of the pigmented cells. The results show that cells rotated much faster under a square waveform (~240 rpm) compared to a sine waveform (~130 rpm) or a triangle waveform (~110 rpm). This phenomenon could be explained by the fact that a square wave is composed of an infinite number of harmonics, i.e., a fast Fourier transformation (FFT) of a square wave can be expressed as an infinite series of sinusoidal waves. Comparing a sine wave and a square wave at the same frequency, the latter waveform contains higher frequency components than the designated frequency of the wave6. The quantified rotation speed is summarized in Fig. 4. This plot also shows that this variation in rotation speed variation is both controllable and repeatable. Also,from our observation, the rotation direction of the cells was random, i.e., we have no control of the clockwise or counter-clockwise of the cell rotation on the surface plane (x-y plane).

Automatic Cell Sorting Algorithm

In order to automatically estimate the cell self-rotation speed for rapid cell sorting, we have developed an algorithm based on the analysis of image sequences. The acquired image sequences are captured by a charge coupled device (CCD)-based microscope system. The cell rotation speed estimation method is divided into three stages, as described below. First, image pre-processing, including noise reduction and contrast adjustment, enhances the performance of the analysis. The noise filtering improves image quality and a contrast adjustment is performed with histogram equalization enhancing the contrast between the cells and the background. A cell rotation estimation algorithm is used to analyze two basic motion patterns, namely, cell self-rotation about its yaw axes, and the other is rotation with translational motion along the chip surface. In order to estimate the cell rotation speed about the yaw axis accurately, translation compensation is adopted in the algorithms. After this compensation, the cells' rotation cycles can be counted through a pixel patch correlation calculation. Summarizing, the cell rotation speed estimation method is split into three stages: image pre-processing, translational motion tracking and a pixel patch correlation-based calculation that yields the cell self-rotation speed. Fig. 5 shows a process flow diagram of the aforementioned algorithm.

Image pre-processing

The cell rotation image sequences are captured by a CCD microscope system. There are several sources of noise in microscope imaging: photon noise, thermal noise, readout noise and quantization noise7. Image pre-processing methods with noise reduction and contrast enhancement are the basic methods to improve the performance of later image sequence analysis. Noise reduction improves the fidelity of the original acquired image. After preliminary noise reduction, we then apply contrast enhancement to the image sequences. Image contrast enhancement is applied to images to improve the visibility of melanin in the cells.

Noise Reduction. Noise reduction processing is a fundamental operation in biomedical image processing applications. Any subsequent operation will benefit from noise reduction processing. The noise level can affect the performance of the image analysis algorithm, especially in separating objects from the background. In a microscope video sequence of a cell, light objects are on a dark background. We could, for example, look for the lowest intensity value in the object and separate the object from the background by searching for all pixels that have an intensity value higher than this threshold value. However, noise has changed the intensities in the background and in the object, so that there are some very high intensities within the background and some very low intensities in the object. Image pre-processing can reduce this noise distribution effect.

One type of noise arises from the camera sensors. In the camera incoming photons are transformed into an electrical charge by a charge coupled device or CCD. However some electrons are created within the CCD randomly. This randomly distributed noise is added to the signal. Since the background noise should be homogeneous with a random distribution of intensities, a Gaussian low-pass filter can be used to remove readout noise. The Gaussian filter kernel function is described as follows,


The Gaussian filter can be applied using standard convolution methods with a suitable kernel function. The convolution method is performed to reduce the filtering computation time, since the 2D isotropic Gaussian equation can be separated into two orthogonal components. Therefore the 2D convolution can be performed by two convoluting operations in two orthogonal directions.

Contrast Enhancement. The noise-reduced grayscale image in Fig. 6(A) lacks details since the range of luminance is limited to a narrow band of gray levels, as shown by the image's histogram in Fig. 6(C). The image histogram simply plots the frequency at which each gray level occurs from black to white. It shows that the majority of the gray levels in the image are grouped between about 90 and 240. Image contrast enhancement is required to produce a clear image through a redistribution of these brightness intensity values.

In order to enhance contrast between the cells and the background, a histogram equalization is applied to the image sequences which enhances the contrast of the grayscale image8. Histogram equalization is one the most well-known methods for contrast enhancement. This approach is generally useful for images with a poor intensity distribution. Histogram equalization expands the luminance within the image to fill the entire gray-scale spectrum. To do this, the cumulative frequencies are calculated within the image. The cumulative frequency ofa grey level is defined as the sum of the histogram data. So the equalized histogram keeps the profile of the original histogram, although it is now extended to the entire spectrum. The cumulative frequency graph makes the grey level frequencies distribute evenly within the image.

Histogram equalization is applied in the following manner. A given grayscale image {x} is composed of L discrete gray levels, denoted as {Xi}, the probability density function is defined as


for i = 0 , 1 , . . . , L-1, where ni is the number of times that the level Xi appears in the image {x} and n is the total number of samples in the image. Based on the probability density function, a cumulative density function cdf(x) is defined as


where cdf(XL-1) = 1 by definition. Thus, the corrected pixel value of the grayscale image transform function is based on the cdf(x) and is expressed as


An image that is intensity corrected using a homogeneous histogram equalization has enhanced contrast between the cells and the background. The corrected image is much clearer and details within the cells are much sharper. The equalized image with the histogram and cumulative frequency graphs is shown in Fig. 6(D). After noise reduction and contrast enhancement, image qualities are greatly improved and ready for subsequent algorithms, as seen in Fig.6 (B).

Rotation Estimation Algorithms

There are two basic observed motion patterns for a cell, one is self-rotation about its yaw axes and the other is a translational motion along the chip surface. In order to estimate cell rotation, translation compensation needs be adopted in the algorithms. After this compensation, the number of cell rotation cycles is counted through a pixel-patch correlation calculation.

Translational Motion Tracking. The most popular translational motion compensation method is the block-matching algorithm (BMA), which is a block-based motion estimation method. The best matching block is found for a reference block within a search area, and then a motion vector is calculated as the displacement of the best matching block to the position of the current macro-block. This translational motion vector describes the location of the matching block from the previous frame with reference to the position of the target block in the current frame.

Although there is no zooming motion in the sequence of microscope images of the cells in our DEP experiment, and the brightness and contrast of the sequences are nearly stable, self-rotation of cells could affect the matching accuracy due to non-overlapping blocks needed for the general block-matching algorithm.

To improve accuracy, we propose using a rotating-circle matching template to replace the non-overlapping matching block template. Fig. 7 depicts the rotating-circle matching template generation process. The process involves three stages: (1) converting the original image to a grayscale; (2) producing a binary image using an adaptive thresholding method9; (3) estimating the center and radius of the fitted-circle mask. For a rotation over an angle θ, a point X(x, y) in the original image is mapped onto the point X'(x', y') in the resultant image. The relation between the points is:


where Tθ(∙) is a rotation operator and .

To find the motion vector for a specific cell in the current frame, a best matching image patch is searched within a predefined search window in the previous, reconstructed, reference frame as shown in Fig. 8.

There are many types of matching criteria for the block matching algorithms, such as the Sum of the Absolute Difference (SAD), the Mean Squared Difference (MSD)10, the Mean Absolute Difference (MAD)11, and so on. In this paper, we estimate the motion vector based on the SAD with consideration for optimizing computational efficiency. The motion vector (MVx, MVy) is defined in equations (10) and (11). The best matched patch in the search window of the current frame is found from the minimum value of the SAD with the rotating-circle matching patch in the reference frame and the target patch in the current frame.

The matching criterion is expressed as follows,


where (x, y)∊S and (i, j) ∊M, S is the search window and M is the defined image mask. The corresponding motion vector for the target window with the minimum SAD is then determined by,


In a pseudo-code format, the SAD calculation is performed based on the rotating-circle matching template as:

1. for (x,y) in search window S

2. { for( θ = 0 to 2π )

3. { calculate SAD(x, y, θ)

4. increment θ by a step Δ

5. }

6. increment x,y by 1

7. }

Image Patch Cross Correlation. After compensation for the translational motion in the images based on the motion vector, we then calculate the image patch cross correlation to estimate the cell rotation speed. The correlation coefficient shows the similarity between the selected template patch from the subsequent image patches at the same location. For a grayscale image patch, the correlation coefficient γ is defined as:


Where t(∙) is the selected template patch and f(∙) is the subsequent image patch; and are the mean of f(∙) and t(∙) respectively.

The correlation coefficient reflects the degree of linearity between two data sets. A larger value indicates a perfect positive linear relation between the two data sets, while a smaller value indicates a perfect negative linear relation. Therefore, we detect the local maxima values of the correlation coefficients in order to track the peak points and the rotation cycle can be estimated through the index of the peak points.

Algorithm Performance

For verification of the proposed rotation estimation algorithm, image sequences at different cell rotation speeds are recorded. The cell self-rotation speed range is varied from 50 rpm to 250 rpm by adjusting the DEP parameters, i.e. frequency, voltage and waveforms.

For quantitative verification of the performance of the translational motion compensation algorithm, the cell compensation estimation results are compared with the estimation without compensation, as shown in Fig. 9. We selected an image sequence over 1.12 seconds (28 frames at 25fps) for the test. The standard deviation of the motion compensation results is much less than the estimate without the motion compensation. That means that the proposed algorithm has performance stability in estimating the rotation rate of the cell.

The position change for a Melan-A cell for ~90 seconds is recorded and is plotted in Fig. 10, during which the voltage is first decreased from 16V to 1.5V gradually and then is increased back to 16V. Fig. 10 shows a comparison between using this algorithm and a manual method (i.e., by visual identification using human eyes) to analyze the same video frame-by-frame, which demonstrates that the different methods provide matching results. Also, the algorithm is capable of tracking changes in the rotation speed for a certain period, which is almost impossible to estimate manually.


Cell manipulation using a DEP technique is conducted in a transparent microfluidic chip with embedded Au microelectrodes fabricated using MEMS technology. Self-rotation of pigment cells is observed when a specific electrical potential is applied between the microelectrodes to generate a DEP force field. We have developed a novel computer vision algorithm to estimate the cell rotation speed automatically. This algorithm analyzes each frame of a video sequence taken from a CCD-based microscope using a rotating-circle template with a block matching method and a pixel patch correlation. Compared to the manual estimation process, the algorithm can more accurately calculate the DEP-induced rotation rate of the cells at various applied voltages, frequencies, and waveforms, and also reduce data processing time by at least a 100 times. Most importantly, the algorithm is accurate even when the cell has a simultaneous translational motion across the video image sequence. Also, the algorithm is capable of tracking changes in rotation speed over a long period of time by stably analyzing a massive data set of video image frames. Therefore, we envision that our automatic cell rotation analysis method can be used with DEP technology as an efficient procedure for automated cell sorting in the future.