2.1.

Microscopy and Image Processing

Figure 1 shows a schematic of the optical and electrical impedance measuring apparatus. A SR830 lock-in amplifier (Stanford Research Systems, Sunnyvale, California) circuit generated a current through the ITO electrode and measured the resulting electrode voltage. A computer equipped with a Personal Computer Memory Card International Association card and a LABVIEW-based data acquisition program controlled the amplifier and performed frequency scans. Cells were kept viable using an incubator (WeatherStation, Olympus, Center Valley, Pennsylvania) that kept the temperature $(37°\mathrm{C})$ , humidity, and $\mathrm{C}{\mathrm{O}}_2$ (5%) levels constant. These controlled conditions were crucial to keep cells alive during long time-lapse measurements. The imaging system consisted of a $20\times$ -long working distance objective lens with a numerical aperture (NA) equal to 0.4, an Olympus model IX-71 inverted microscope with a polarizer (IX-LWPO), a DIC prism (IX2-DIC20) in the long working distance DIC condenser (IX2-LWUCD, $\mathrm{NA}=0.55$ ), a transmitted Nomarski prism (U-DICTS), an analyzer (IX2-AN), a $100\text{-}W$ halogen lamp house (U-LH100-3, Olympus), and a Hamamatsu $14\text{-}\text{bit}$ electron multiplier (EM) cooled and intensified–charge-coupled device (CCD) digital camera. Additionally, a mechanical shutter was synchronized with the CCD camera to minimize any effects the halogen lamp may have had on cell growth.¹⁹ The total magnification used in the current imaging system was $32\times$ , resulting from a $20\times$ objective lens and a $1.6\times$ lens, which approximately fit the entire ITO working electrode. Differential interference contrast image and impedance measurements were taken simultaneously at $3\text{-}\min$ time intervals.

Fig. 1

Integrated optical and electrical impedance measurement system schematic. DICM images are acquired simultaneously with electrical impedance measurements. An environmental chamber maintains a humidified, $37°\mathrm{C}$ , and 5% $\mathrm{C}{\mathrm{O}}_2$ atmosphere for dynamic long time-lapse measurements. The measured impedance is a function of the current flow under the cells, between the cells, and the capacitively coupled current through the cell membranes.

Image processing was performed using a series of deconvolution and edge detection filters to identify the cell boundaries and to estimate the total cell-covered electrode area. Each image was deconvolved using a Lucy-Richardson deconvolution algorithm with a Gaussian point spread function, and the background was removed from the cell-covered areas using high and low threshold limits.²⁰ Binary images were obtained using Canny and Sobel filters. An overlay image was subsequently constructed to compare the digitally processed image with the original image. This permitted the cell-covered area to be quantified by determining the total number of pixels that had not been eliminated by the filters.

2.2.

Electrical Impedance Measurements

Figure 2 shows a schematic of the impedance measuring circuit. Electrical impedance measurements were performed using a SR830 lock-in amplifier. A $1\text{-}{\mathrm{V}}_{\mathrm{rms}}$ ac signal was generated and connected to an electrode array via a $1\text{-}\mathrm{M}\Omega$ resistor. Although this provided an approximately $1\text{-}\mu \mathrm{A}$ current source, corrections based on the circuit model were made to estimate the actual impedance from the electrode voltage measurement. The lock-in amplifier had an input impedance characterized by a parallel combination of a $10\text{-}\mathrm{M}\Omega$ resistor and a $25\text{-}\mathrm{pF}$ capacitor. The ac voltage source resistance was $50\Omega$ , and the capacitance of each coaxial lead, $C_{pv}$ and $C_{ps}$ , was approximately $86\mathrm{pF}$ . Fabrication of the ITO electrodes is explained in detail in Choi ¹⁸

Fig. 2

Impedance measuring circuit configuration schematic. A $1\text{-}\mathrm{V}$ ac signal is applied to an electrode via a $1\text{-}\mathrm{M}\Omega$ resistor, $R_{cc}$ . The source resistance, $R_s$ , is $50\Omega$ and the input impedance to the lock-in amplifier is equivalent to a parallel combination of a $10\text{-}\mathrm{M}\Omega$ resistor and $25\text{-}\mathrm{pF}$ capacitor. The reference cable coaxial leads and lock-in amplifier coaxial leads introduce into the circuit parallel capacitances $C_{ps}$ and $C_{pv}$ , respectively. The electrode impedance, $Z_c$ , is estimated by voltage measurements and a voltage-to-impedance conversion based on this circuit.

The voltage was sampled at a rate of 32 samples per second using a filter time constant of $32\mathrm{ms}$ and a $12\text{-}\mathrm{dB}$ /decade roll-off every $3\min$ . The averages and standard deviations of the 32 sampled voltages were calculated for each set of measurements. Frequency sweeps using 15 logarithmically spaced frequency samples between $10\mathrm{Hz}$ and $100\mathrm{kHz}$ were obtained for the naked electrode and the cell-inoculated electrode during cellular attachment.

2.3.

Cell Culture

PPAECs were isolated from fresh porcine arteries obtained from a local abattoir. The endothelial cells were removed from the intimal artery walls by carefully scraping them off with a scalpel. The cell culture media consisted of M199 (GibcoBRL, Baltimore, Maryland) and 10% fetal bovine serum (Hyclone, Logan, Utah) supplemented with basal medium eagle (BME) vitamins (Sigma-Aldrich, St. Louis, Missouri), L-glutamine (GibcoBRL), penicillin and streptomycin (GibcoBRL), and BME amino acids (Sigma). The culture was maintained until the cells reached confluence, and then they were passaged. Passaging generally occurred once a week, and passages 4 through 8 were used for testing in this study. Trypsin–ethylenediamine tetraacetic acid (1X, GibcoBRL) was used to detach cells for passaging and electrode inoculation. Cells were cultivated in an incubator that kept the temperature at $37°\mathrm{C}$ and the $\mathrm{C}{\mathrm{O}}_2$ level at 5%. Cells suspended in M199 were inoculated directly onto a sterilized $\mathrm{ITO}\text{-}{\mathrm{Si}}_3{\mathrm{N}}_4$ microelectrode that was not previously coated with any adhesion molecules such as fibronectin. High and low cellular inoculation densities were chosen to produce either a subconfluent pattern or a confluent pattern following the initial attachment phase.

3.1.

Electrical Scan Summary

Figure 3 shows the time-dependent changes in normalized resistance and reactance at 1.0, 1.77, 3.16, and $5.62\mathrm{kHz}$ during the attachment of PPAECs to a $250\text{-}\mu \mathrm{m}$ diameter ITO electrode. The zero time point corresponds to the cell inoculation time. The normalized resistance, $(R_c-R_n)∕R_n$ , and the normalized reactance, $(X_c-X_n)∕X_n$ , were calculated from the cell covered and naked electrode resistances, $R_c$ and $R_n$ , and reactances, $X_c$ and $X_n$ , respectively. An initial peak in the normalized resistance and reactance occurred approximately $4\mathrm{h}$ after cell inoculation on the ITO electrodes. Following the initial peak, the normalized resistance and reactance stabilized and slowly decreased over several hours. The normalized resistive component of the naked electrode was consistent with small amounts of drift. In addition, changes in both the resistive and reactive components were observed during the first several scans until the environmental chamber parameters, such as the $\mathrm{C}{\mathrm{O}}_2$ level, pH, and temperature stabilized.

Fig. 3

Normalized resistances (solid lines) and normalized reactances (dashed-dotted lines) as a function of time. The terms $R$ and $X$ represent the resistance and reactance, respectively, and the subscripts $c$ and $n$ indicate cell-covered and naked scans, respectively. Among the 15 frequency scans acquired, 4 representative frequencies are selected to illustrate the cellular attachment of PPAECs to a $250\text{-}\mu \mathrm{m}$ ITO electrode. For the sake of clarity, every 20th data point is marked with symbols: ◻ for $1.0\mathrm{kHz}$ , ▵ for $1.77\mathrm{kHz}$ , ◇ for $3.16\mathrm{kHz}$ , and ○ for $5.62\mathrm{kHz}$ .

Figure 4 summarizes the time-dependent evolution of the normalized resistance and reactance as a function of frequency. The normalized resistance peaked around $1\mathrm{kHz}$ , and the normalized reactance appeared to increase with increasing frequency following an initial attachment phase. The resistance and reactance changes were a complex function of the endothelial cell attachment, spreading, and micromotion. Although the impedance measurements were a sensitive function of the cellular attachment parameters, these measurements alone could not sufficiently evaluate the dynamic cellular morphology changes coupled to the cell-cell and cell-substrate adhesion, and cell-covered area on the working electrode.

Fig. 4

Representative surface plots of normalized resistance (a) and reactance (b) attach scans of PPEACs on a $250\text{-}\mu \mathrm{m}$ diameter ITO electrode as a function of frequency and time following a confluent inoculation density. The surfaces consist of approximately 300 frequency spectrum scans that were sampled at $3\text{-}\mathrm{m}$ time intervals. Plots (c) and (d) show normalized resistance and reactance, respectively, for PPEACs’s attachment at $t=9$ , 60, 120, 180, 270, 600, and $900\min$ . (c) The normalized resistance as a function of frequency shows a monotonic increase with time up to approximately $600\min$ at which point saturation occurs. (d) The corresponding normalized reactance also shows an increase with time during the cellular attachment phase.

3.2.

Image Segmentation and Cell-Covered Area Estimation

Figure 5 shows a time series of endothelial cell DICM images of a $250\text{-}\mu \mathrm{m}$ diameter ITO electrode following cellular inoculation densities that ultimately produced a subconfluent and confluent endothelial cell pattern. The first row shows the time series leading to a subconfluent pattern, and the second row shows the series leading to a confluent pattern. During the endothelial cell attachment and spreading process, the cells start with a globular morphology and then gradually flatten. For the case that produced a subconfluent pattern, approximately 25 cells became attached to the working electrode. Six minutes after inoculating the cells onto an ITO electrode, two cells began to spread, but the rest retained their globular morphology for approximately $30\text{to}33\min$ . In the case that produced a confluent pattern, approximately $6\min$ after inoculation, a few cells began spreading, but most cells remained globular for approximately $30\text{to}33\min$ . After this time, the cell-covered area remained almost constant. However, the cells became more tightly packed and much flatter over time. There were approximately 55 and 65 cells on the working electrode at $t=450\min$ and $t=900\min$ , respectively. For inoculations that produced subconfluent and confluent endothelial cell patterns, the time required for the initial cell spreading stayed relatively constant. Another comparison made for these two cases was the settling time, defined as the initial time when the number of cells remained constant for a period of four consecutive images. The settling time for the case that produced a confluent pattern was approximately $15\text{to}18\min$ , while the settling time for the case that produced a subconfluent pattern was approximately $21\text{to}27\min$ .

Fig. 5

DICM images of PPAECs on a $250\text{-}\mu \mathrm{m}$ diameter $\mathrm{ITO}\text{-}{\mathrm{Si}}_3{\mathrm{N}}_4$ electrode. The top row shows a set of images for a subconfluent inoculation, and the bottom row shows a set of images for a confluent inoculation. The magnification is $32\times$ providing a $250\text{-}\times \text{-}250\text{-}\mu {\mathrm{m}}^2$ field of view. The initial inoculation density was carefully chosen to produce either a subconfluent or confluent endothelial cell attachment pattern.

Figure 6 summarizes the image processing results used to evaluate the cell-covered electrode area in this study. A deconvolved image (a) was first obtained using a Gaussian point spread function. The following image (b) removed a portion of the background from the cell-covered areas using high and low threshold limits and a complete binary image (c) was separately obtained using a Canny or Sobel filter. The local gradients were compared to high and low threshold values, either provided by the user or internally calculated, to roughly detect the cell boundary. A pixel-by-pixel comparison of images (b) and (c) was then made to more accurately define the cell membrane boundaries. Generally, a portion of each cell was removed using the threshold filter because some of these pixels had similar intensities compared to those of the surrounding medium. This is a general characteristic of DICM. The intensities in a cell varied from darker to brighter or brighter to darker along the shear axis. The diagonal filter compensated for this eliminated area and specified single cells. Another filter, called a stitch filter, was employed to completely fill holes in cells to produce image (f). Finally, a removal filter deleted the defects that appeared as cells but were possibly small air bubbles or optical artifacts. The final overlay image (h) was constructed from images (a) and (g) to compare the digitally processed image with the original image. The last image (h) shows the area covered by cells using the image processing algorithm and quantifies any eliminated or overestimated areas produced by the algorithm.

Fig. 6

Image processing flow chart and digitally processed images following each filter step. To automate the cell-covered area estimation, a sequence of image processing steps is carried out. (a) Deconvolved image with examined area outside electrode set to 0. (b) Image after applying threshold filtering with high and low threshold limits. (c) Binary image created using a Canny filter. (d) Pixel comparison step of images (b) and (c). (e) Image after applying diagonal filter along the shear axis. (f) Image after the filling filter. (g) Image after removal filter. (h) Overlaying image with (a) and (g) for the comparison of actual and calculated areas.

Figure 7 shows four digitally processed representative overlay images of PPAECs and the normalized cell-covered area as a function of time. The image processing software identified single cells and their boundaries until they reached confluence. Overlay images were used to visually check cell-occupied areas overestimated or discarded from digital image processing. The graph validates the results obtained by digital image processing by comparing them with those acquired by manual area estimation. The average difference between digital image processing and manual area estimation for each experiment was observed to be less than 5%.

Fig. 7

Overlay images comparing the cell-covered areas estimated using the image processing algorithm and manual cell-covered area estimates. The overlay images are used to visually check the areas that are discarded or overestimated after image processing. The graph validates the cell-covered area estimates obtained from digital image processing by comparing those obtained by manually tracing the cell boundaries. The maximum difference between the two methods is approximately 8% for this sample where cells reached confluence at approximately $100\min$ . The term $A_c$ denotes the cell-covered area and $A_e$ denotes $0.043\text{-}{\mathrm{mm}}^2$ electrode area.

3.3.

Combined Electro-Optic Analysis

Figure 8 compares the time-dependent changes in the normalized resistance and normalized cell-covered area for a subconfluent and confluent cell inoculation. In both cases, the normalized resistance did not completely reflect the amount of the electrode covered with cells. Changes in cell-matrix and cell-cell attachment, which altered the resistance, complicated the correlation between cell-covered area and resistance. In both cases, however, their time-dependent patterns are similar.

Fig. 8

Normalized resistance (dashed lines) and normalized cell-covered area (solid lines) vs time for the (a) subconfluent and (b) confluent cases shown in Fig. 5. The normalized resistance and the normalized cell-covered area have similar time-dependent patterns. Both electrical impedance and DICM image data were simultaneously acquired at $3\text{-}\min$ time intervals.

Figure 9 illustrates the changes in normalized resistance with normalized cell-covered area for both the subconfluent and confluent cases. Three distinct intervals are considered and will be referred to as the settling zone, the linearly correlated zone, and the saturated zone. For the settling zone, the normalized cell-covered area first increases abruptly due to the increasing number of cells. A total of 25 cells in the subconfluent case and 65 cells in the confluent case do not appreciably change the normalized resistance. In the linearly correlated zone, the normalized cell-covered area proportionally increases with the normalized resistance showing both subconfluent and confluent cases have an analogous starting time, which is around $t_1$ . Also, a linear relationship between the normalized cell-covered area and normalized resistance is shown. For the saturated zone, in the subconfluent case, the normalized cell-covered area slightly decreases while the normalized resistance remains relatively constant as the cell-substrate adhesion is much tighter and flatter. In the confluent case, however, the normalized cell-covered area remains constant while the normalized resistance continuously increases as a possible result of increasing cell-substrate and cell-cell adhesion. The cells look much flatter and more tightly packed in these cases. Changes in normalized resistance and reactance are a complicated function of the normalized cell-covered area, cell-cell adhesion, and cell-substrate adhesion. All of these factors can be considered both optically and electrically with a transparent conductive ITO electrode. Increasing cell-cell and cell-matrix adhesion states, however, are consistent with the more tightly packed and flatter cell morphologies. This confirms cellular attachment with ITO silicon nitride bioelectrodes as well as demonstrates that the ITO electrode is a good tool for both electrical measurement and optical visualization.

Fig. 9

Normalized resistance vs normalized cell-covered area for the (a) subconfluent and (b) confluent cases shown in Fig. 5. The error bars shown for selected data points indicate the difference between manual area estimates and those calculated using the digital image processing algorithm. Estimated errors for the normalized resistance are smaller than the symbol size. $t_1$ and $t_2$ represent the times necessary for most cells to spread and the time to reach approximately 5% of their maximum electrode covered area, respectively. For the confluent case, the maximum covered area is approximately 99% of the total electrode surface area, and for the subconfluent case, the maximum covered area is approximately 70% of the total electrode area.

Each subconfluent and confluent cellular inoculation needs to be considered on a case-by-case basis. For the confluent case, which lasted up to $19\mathrm{h}$ , the cell-covered electrode morphology is relatively constant after reaching confluence. However, the population of cells was observed to fluctuate. In another confluent case considered in this study, after $13\mathrm{h}$ some of the cells appeared to migrate from the working electrode surface to the insulating layer area thus leaving bare areas. In all the subconfluent cases, the cells migrate more freely than the confluent case, and as a result, there are greater fluctuations in the total cell-covered area.