Cell analysis system based on compact disk technology

Abstract

Background

A cell analysis system was developed to enumerate and differentiate magnetically aligned cells selected from whole blood. The cellular information extracted is similar to the readout of musical information from a compact disk (CD). Here we describe the optical design and data processing of the system. The performance of the system is demonstrated using fluorescent-labeled cells and beads.

Materials and Methods

System performance was demonstrated with 6-μm polystyrene beads labeled with magnetic nanoparticles and allophycocyanin (APC) and immunomagnetically aligned leukocytes, fluorescently labeled with Oxazine750 and CD4-APC, CD8-Cy5.5, and CD14-APC/Cy7 in whole blood.

Results

The sensitivity of the system was demonstrated using APC-labeled beads. With this system, beads containing 333 APC molecules could easily be resolved from the background. This level of sensitivity was not achievable with a commercial flow cytometer. A maximum of 20,000 immunomagnetically labeled cells could be aligned and analyzed in between 0.6 m of Ni lines, distributed over a surface area of 18 mm² and extracted from a blood volume that depended on the height of the chamber. The utility of the system was demonstrated by performing a three-color CD4-CD8-CD14 assay.

Conclusions

We built a cell analysis system based on immunomagnetic cell selection and alignment and analysis of fluorescent signals employing CD-technology that is as good or better than current commercial analyzers. The cell analysis can be performed in whole blood or any other type of cell suspension without extensive sample preparation. Cytometry 47:173–182, 2002. © 2002 Wiley-Liss, Inc.

Recently we introduced CellTracks (1), a new cell analysis method based on immunomagnetic selection, extraction, and aligning of cells on a glass substrate, which forms the upper surface of a sample chamber. A complete description of the magnetic principles and considerations are given elsewhere in this issue (2). Once the cells are aligned, they are scanned with a 635-nm laser using compact disk (CD) technologies, and are counted and characterized by means of fluorescence.

During the last decade, alternative cell analysis systems have been developed that overcome some of the limitations of flow cytometry. One of these is the volumetric capillary cytometer (3), which was developed to obtain a direct measurement of the absolute count of cells in combination with minimal sample handling. This was realized using a glass capillary with a precisely known volume. In the cytodisk (4) and the laser scanning cytometer (5), cells are present on a solid base, which provides unique features for cell analysis (6, 7).

The CellTracks system combines the benefits of these alternative systems in a single system, which results in a simple and versatile cell analysis platform. The magnetic forces extract the cells from a precisely known volume to the upper glass surface of a sample chamber, and the number of cells present on this surface directly yields an absolute number of cell subsets. The system is capable of performing white blood cell analysis, including immunophenotyping, in the presence of all blood constituents (8). Furthermore, the magnetic force holds the cells in position, which creates the possibility of revisiting the cells for further analysis and studying changes as a function of time with or without changing the buffer in which they reside (9).

Here we give a description of the optical design and data analysis combined with measurements that demonstrate the performance of the system.

MATERIALS AND METHODS

Chamber and Magnet Design

The pole pieces of two magnets, made of neodymium iron boron with an internal magnetization of 13,700 Gauss (Crumax Magnetics, Inc., Elizabethtown, KY), are positioned at 3 mm from each other, and their faces make an angle of 70° with the z-axis. The chamber is placed 2 mm below the top of the magnet, and has a width of 3 mm and a height of 0.5 mm (Fig. 1). The chamber consists of a glass shaped bottom that fits between the pointed magnets. Two pieces of double-sided tape (Core Series 2-1300, 3M, St. Paul, MN), with a thickness of 0.5 mm, form the walls of the chamber.

Figure 1. Schematic illustration of magnet sample-chamber configuration. Inset: Enlarged view of cross section of chamber. Lines show paths along which cells move to the upper surface and the final alignment of cells between Ni lines.

The upper glass surface (7740 Pyrex®, Corning International, Germany) of the chamber contains ferro-magnetic lines of Ni, with a thickness of 200 nm, a line width of 22 μm, and a line spacing of 8μm, resulting in a line pitch of 30 μm. These Ni lines are patterned onto the Pyrex® wafer by positive-resist Olin 907/12. This is exposed to ultraviolet (UV) light using the appropriate mask and followed by dipping in chloro-benzene, rinsing with demineralized water, and development in OPD 4262. Next, 200 nm Ni are sputtered onto the surface for 35 min, using an argon flow. The residue is removed by acetone to obtain the Ni lines.

For cell analysis, the 40 central lines in the chamber are used. The overall length is the product of a single line length (15 mm) times the number of lines (40) scanned, giving a total line length of 0.6 m. The diameter of a leukocyte is on the order of 10 μm, resulting in a theoretical maximum of 60,000 cells positioned between the lines. The height times the scanned surface provides the scanned volume and directly yields the number of leukocytes/ml.

Sample Preparation

To measure system linearity, 500 μl of EDTA blood were incubated with 250 μl of CD45 immunomagnetic particles (diameter 175 nm, 100 μg/ml; Ferrofluids, Immunicon, Huntingdon Valley, PA), 75 μl of 0.01 M Oxazine750 (Exciton Corp., Dayton, OH) in PBS, and 3 μl of 0.006 mg/ml CD4 (Leu3a)-APC (Becton Dickinson Immunocytometry Systems (BDIS), San Jose, CA). After 15 min, the magnetically labeled cells were isolated using a magnetic separator (Immunicon), and the cells were resuspended into 500 μl of PBS. From this, a dilution range was made, with dilution factors ranging from 1–500.

For the three-color leukocyte immunoassay, 100 μl of EDTA blood, 50 μl CD45 ferrofluids, of 100 μg/ml (Immunicon) were added together. At the same time, 1 μl of 0.006 mg/ml CD4 (Leu3a)-APC (BDIS), 4 μl of 2 μg/ml CD8-Cy5.5, and 6 μl of 2 μg/ml CD14 (Leu-M3) APC/Cy7 (Caltag, Burlingame, CA) were added. The monoclonal antibody CD8 (Leu2a, BDIS) was conjugated to Cy5.5 according to the protocol provided by the manufacturer (Amersham, Arlington Heights, IL). After a 15-min incubation, 300 μl of PBS were added, and 50 μl of this blood mixture were injected into the chamber. The chamber was placed in the magnet, and after 2 min the feedback system was switched on and the measurement was started.

RESULTS

Magnetic Cell Selection and Alignment

Immunomagnetic particles with a diameter of 175 nm and labeled with an antibody (Ferrofluids) that binds specifically to the cell type of interest were added to blood. This cell suspension was injected into a chamber that was placed between two pointed magnets that were yoked together to form a single magnet (Fig. 1). A small section of the chamber with a random selection of gradient lines which showed the paths along which the magnetically labeled cells moved to the upper surface was plotted. A complete description of the magnetic principles is given elsewhere in this issue (2). The upper surface of the chamber contained ferro-magnetic lines of Ni, which were magnetized by the magnetic field of the external magnets. In the plane of the Ni lines, the magnetic field was parallel to this plane (x-direction), and the individual Ni lines were magnetized in the x-direction. At 50 μm from the top of the chamber, the cells started to feel the additional magnetic gradient produced by the induced magnetization of Ni lines. The net gradient forced the cells to move in between the Ni lines, where they stuck, to be analyzed optically. The average time for a cell to reach the upper surface and align was 85 s.

Optical Design

A schematic representation of the optical system is shown in Figure 2. A 635-nm laser diode contained beam-shaping optics emitting 8 mW with a 4-mm diameter Gaussian beam profile (ACM 08, Power Technology, Little Rock, AR), and was focused onto the sample by two identical cylindrical lenses, with a focal length of 40 mm (Spindler & Hoyer GMBH, Göttingen, Germany), a dichroic long-pass mirror (650drlp, Omega Optical, Inc., Brattleboro, VT), and a regular CD player objective. The objective (KSS210a, Sony, Japan; see insert, Fig. 2) had an NA of 0.45 and a diameter of 4 mm, and was optimized for 780 nm. The chamber and magnets are placed on a XY-stage (Fig. 2), and upon moving the stage in the y-direction, the cells passed the laser-focus one after the other. After scanning a line, the laser focus was shifted to the next line, and the stage was moved in the opposite y-direction. During scanning, a feedback system kept the laser focused on the aligned cells to maintain optimum illumination. The two identical cylindrical lenses, with a focal length of 40 mm, were placed at a distance to each other slightly larger than two times their focal length. A cylindrical lens with an effective focal length of 1 m was formed in this way, and made the beam converge in the z-direction while it remained parallel in the x-direction. Focusing this beam with the CD-player objective lens resulted in two focal planes, 15 μm apart from each other. For scanning, the plane where the parallel beam was in focus was used, as indicated by the position of Ni lines in Figure 3a. The purpose of creating this astigmatic system was twofold:

Figure 2. Layout of CellTracks system. Lower left: Photograph of CD-player objective.

Figure 3. a: Position of Ni lines with respect to two focal planes that are spaced 15 μm from each other. Scanning takes place in the plane in which the parallel beam is focused. b: Schematic representation of change in intensity on split detector when the objective moves with respect to Ni lines in the x-direction while the z-position is fixed. c: Change in intensity on split detector when the objective moves in the z-direction with respect to Ni lines while the x-position is fixed.

• 1
  To maintain optimum illumination during scanning, a tracking signal is needed to provide the position of the sample with respect to the objective in the x-direction, and a focus signal providing the position of the objective in the z-direction is needed. To obtain this signal, the long axis of the elliptical laser focus was chosen to be larger than the spacing of the Ni lines. Light falling on the Ni lines, which act like mirrors, was reflected and captured by the objective. It passed the λ/4 plate for the second time and was separated from the incoming laser light by a polarizing beam-splitter cube (PBS) and imaged with a spherical lens (f = 100 mm) onto a split detector (SPOT 2DMI, UDT Sensors, Inc., Hawthorne, CA) consisting of two separate detectors A and B. The difference between signal A and B provided the tracking signal (A − B), and the sum of signal A and B provided the focus signal (A + B).
• 2
  To avoid the possibility that more than one cell was present in the laser focus, the short axis of the laser focus had to be smaller than or equal to the diameter of a cell. Because of astigmatism, the laser focus had an elliptical shape of 14 × 5 μm. Typically, white blood cells have a diameter ranging from 5–20 μm, and only one cell can be present in the focus of the laser.

The fluorescence light emitted by the cells was captured by the CD-objective, passed the dichroic mirror, and was focused by an achromatic lens with a focal length f = 150 mm (Spindler & Hoyer GMBH) onto a 300-μm pinhole. After passing the pinhole, a combination of dichroic mirrors and band-pass filters (Omega Optical, Inc.) separated the emission spectra of the individual dyes. A 690drlp dichroic long-pass mirror in combination with a 670df20 differential band-pass filter (central wavelength, 670 nm; band-pass, 20 nm) separated the APC emission spectrum from the Cy5.5 and APC/Cy7 spectra. The emission spectrum of Cy5.5 was separated from the APC/Cy7 spectrum using a 730drlp dichroic long-pass mirror in combination with a 705df40 differential band-pass filter. Red-sensitive Hamamatsu R636-10 photomultipliers (PMTs; Hamamatsu City, Japan) were used as fluorescence detectors, and transimpedance type preamplifiers converted the PMT signal currents to voltage pulses.

The focal plane where the parallel beam was in focus was used for scanning aligned cells. Therefore, the fluorescent light beam exiting the objective was parallel. The feedback loop in combination with the objective and achromatic lens focused the captured fluorescent light at the same spot and in the same focal plane as where the pinhole was positioned. This was independent of the x- and z-position of the sample when the system was in feedback mode.

A forward light scatter detector (not shown) was placed below the sample and collected the forward-scattered light at 20–30°. Forward-scattered light can only be detected for clear samples. For measurements on whole blood, the forward-scattered light was obstructed by the erythrocytes present in the chamber; as a consequence, no coherent signal was detected.

The depth of focus was dependent on the Ni line spacing, since the lines formed an aperture on the sample. An experimentally determined depth of focus of 25 μm was obtained for a 8 μm line spacing: the smaller the line spacing, the smaller the depth of focus becomes (2).

Feedback System

If the sample moved with respect to the CD-objective in the x-direction, the ratio of light intensities measured on the detectors A and B of the split detector changed. This is schematically presented in Figure 3b. By subtracting the measured intensities of A and B, the tracking signal was obtained. The point where this signal crossed zero is the point where the focus of the objective was perfectly aligned in the center of the spacing between two Ni lines on the aligned cells. Figure 3c shows a schematic representation of the intensity profile, observed when the sample moved in the z-direction with respect to the objective. Because of astigmatism, the shape of the laser focus changed along the z-axis (Fig. 3a), causing a change in the amount of light reflected by the Ni lines, which was measured by A and B. Adding the measured intensities of A and B gave the focus signal. Out of focus, the sum of intensities was zero and increased when approaching focus. Close to focus it decreased again, since the intensity focused on the Ni lines compared to the intensity focused in between the lines decreased, and less light was reflected. The reflected intensity was at its minimum when the long axis of the focal spot was aligned with the direction of the lines. Moving further through focus, the size of the focus increased again and light fell on the Ni lines, resulting in an increase in signal, which approached zero when out of focus. The measured signal displayed in Figure 3c was inverted, and an offset was added to make the signal cross zero, needed for feedback. The positive slope was used for feedback, and at the point where the signal crossed zero, sample and laser focus were in the same focal plane.

The CD objective could be moved in the x- and z-direction by applying currents to coils present in the objective assembly (10). Figure 4 shows the focus and tracking feedback loops, for the x- and z-direction to keep the objective in the desired position with respect to the wires. The error signal, E₁, was the difference between the signal at the set point and the tracking signal A − B. E₂ was the error signal for the focus feedback, and was the difference between the set point and the focus signal A + B. The error signals were converted to currents by regulator electronics, and in the case of the focus feedback loop, were directly applied to the coils that moved the CD-objective in the z-direction, keeping the error signal minimal. The output of the tracking feedback loop was split into a high- and a low-frequency (<1 Hz) part. The high-frequency part was directly applied to coils that were responsible for the x-movement of the objective, keeping the laser focused on the aligned cells. The low frequency part was applied to a DC motor that drove the x-direction of the XY-stage that moved sample and magnets (Fig. 4). If the objective moved too far from its zero position to stay on the aligned cells, in for example the positive x-direction, the motor pushed the stage with the sample in the negative x-direction. In this way, the objective stayed aligned with the incoming laser light. The maximum error in the alignment of the objective with the incoming laser light was 40 μm, and was negligible with respect to the incoming laser-beam diameter of 4 mm.

Figure 4. Feedback loops for focusing and tracking.

After scanning one line of cells, the laser focus was shifted to the next line by moving the objective in the x-direction. To get to the next line, the feedback for both focusing and tracking should be switched off. This was achieved by switching the feedback loops to ground at the positions indicated in Figure 4, and could be done because the error signals E₁ and E₂ in feedback were zero. At the same time, a pulse was applied to the coils that moved the objective in the x-direction to the next line. The next line was reached in 40 ms, the feedback loop was closed again, and the stage was moved in the opposite y-direction. The analog-out connectors of the ADC board provided the pulses needed for shifting the laser to the next line.

The measured error signals are shown in Figure 5a,c. The focus error signal was measured while the tracking feedback loop was active, and x-position was fixed. The tracking signal was measured while the focus feedback was active, and the z-position was fixed. In practice, both feedback loops are active at the same time. The slope of the tracking signal was determined using the Ni line periodicity and was 0.73 μm/V. To bring the objective into the feedback working range, it was connected to a spindle with a micron scale (not shown). This scale was used to determine the focus signal as a function of the z-position. A slope of 6.13 μm/V was obtained.

Figure 5. a: Tracking signal and focus feedback loop are active while stage is moved in x-direction. b: Frequency spectrum of tracking error signal E₁. Scanning speed in y-direction was 10 mm/s. c: Focusing signal and tracking feedback loop are active while objective is moved in z-direction. d: Frequency spectrum of focusing error signal E₂. Scanning speed in y-direction was 10 mm/s.;6>

The performance of the feedback loops was demonstrated by measuring the frequency spectra of the error signals while both loops were active and the stage was moving at a speed of 10 mm/s in the y-direction. The spectra of error signals for the tracking and focusing feedback loops are depicted in Figure 5b,d. The positional error was determined by integrating the spectrum over the bandwidth of the feedback loop, which was approximately 2 kHz. For the tracking feedback loop, the positional error was 250 nm_rms, and for the focusing feedback the positional error was equal to 120 nm_rms. These positional errors were both negligible to the width of laser focus, width of Ni lines, and depth of focus.

Data Acquisition and Analysis

After preamplification, the PMT signals were filtered and if necessary amplified with a linear amplifier. For a speed of 10 mm/s in the y-direction, a pulse width on the order of 1 ms was obtained, and a first-order low-pass filter, with bandwidth DC-2 kHz, was used for filtering. A 12-bit analog to digital converter (ADC; PCI-MIO-16E-4, National Instruments Austin, TX), which corresponded to 4,096 binary levels or 3.6 decades, digitized the signals. The sampling frequency was determined by the scanning speed in the y-direction and was set to 2,000 points/mm, which corresponded to 20 kHz for a speed of 10 mm/s. With this sampling frequency, approximately 30 sampling points per event were obtained, as shown in Figure 6a.

Figure 6. a: Sampled PMT signal obtained in scanning a CD4+ lymphocyte, selected and aligned using CD45 ferrofluids and fluorescent labeled with CD4-APC. b: Block diagram of data acquisition and signal processing.

A dedicated program, written in the graphical programming environment of Labview (National Instruments, Austin, TX), controlled the ADC board and communicated over the serial port with the driver for the stepper motor that moved the stage in the y-direction. The digitized signals were stored in computer memory, and peak detection was performed on the channel that was set as trigger channel. Peak-detection was performed using the standard Labview peakdetection routine (Labview version 5.1, peakdetection.vi). This routine fit a quadratic polynomial through a number of data points (= points used in quadratic fit, in Fig. 6b) if the signal was above trigger level, and determined its peak value and peak position. The number of data points used in the fit can be set in the program; the best fit was obtained if the number of data points was equal to half the number data points present in the total peak. The peak positions found in the trigger channel were used to determine the peak values in the nontrigger channels. For this, the peak detection routine was applied to data subsets of the nontrigger channels that contained a number of points equal to two times the number of points used in the quadratic fit for the trigger channel. The center of these subsets was the determined peak position of the trigger channel. This is schematically presented in Figure 6b. If no peak value was found in these subsets, the signal value at peak position of the trigger channel was taken as peak value.

After peak detection, the pulse width was determined using the determined peak values and their positions. The routine started at the peak position and took subsequent data points until the value was smaller or equal to the half-maximum value. This was done for both sides of the peak. The data-points belonging to these half maximum values, which can be coverted to microns since sample frequency and scan speed are known, provided the pulse width (Fig. 6a). For a scan speed of 10 mm/s and a sampling frequency of 2,000 pnts/mm, the distance between each sampled data point was 0.5 μm. Compensation for the spectral overlap was performed in software on the linear amplified stored signals.

Fluorescence Sensitivity of the System

The fluorescence sensitivity and linearity of the system was tested using 6 μm polystyrene beads labeled with magnetic particles and containing five different amounts of the fluorochrome APC (Immunicon). The MESF (molecules of equivalent soluble fluorochrome) values per particle were: blank, 333, 925, 5,370, and 39,444. Ni lines with a spacing of 6 μm were used, and the forward light scatter signal was set as trigger signal. The obtained histogram is presented in Figure 7a, which shows that all five populations could be resolved. For whole-blood measurements, no forward light scatter signal could be obtained, and one of the fluorescence channels was used as trigger channel. In Figure 7b, the histogram for the same APC beads using the APC fluorescence channel as trigger is shown. In this case it was not possible to discriminate between a true blank event and no event at all. This explains the very large peak for the blank. When the same set of beads was analyzed on a FACS Calibur (BDIS) flow cytometer equipped with a 635-nm laser diode, and a red-sensitive PMT with a 670-nm longpass filter, only the two brightest populations of the same beads could be resolved (Fig. 7c).

Figure 7. Histograms of APC-labeled beads. MESF values per particle were blank, 333, 925, 5,370, and 39,444. a: Histogram obtained with CellTracks system. Forward scatter used for triggering. Obtained CVs: M1 = 17%, M2 = 20%. b: Histogram obtained with CellTracks system. APC fluorescence channel used for triggering. c: Histogram obtained with FACScalibur flow cytometer. Forward scatter used for triggering. Obtained CVs: M1 = 18%, M2 = 24%. d: Fluorescence intensity of APC-labeled beads plotted against MESF value.

Using the results of Figure 7a, the mean fluorescence intensity of each of the five populations was plotted versus the MESF values in Figure 7d. A line was fitted through the four MESF highest values and was extrapolated to the mean fluorescence value of the blank particle distribution, resulting in a value of 65 MESF for the blank particles.

System Linearity

For cells with a diameter of 10 μm, the theoretical maximum number of cells present on 600 mm of Ni line was 60,000. The experimental maximum was determined by measuring different concentrations of leukocytes labeled with CD4-labeled APC and Oxazine 750 (8, 11). The sample was prepared as described in Materials and Methods. For every dilution factor, 40 lines of 15 mm were measured at a speed of 10 mm/s. Figure 8 shows the total number of leukocytes measured against the dilution factor. The system remained linear up to 20,000 leukocytes, which corresponded to an average of 1 leukocyte on every 30 μm of line or to a cell density in the chamber of 2,200 cells/μl for a chamber with a height of 0.5 mm. For higher cell densities, the cells landed on top of each other, leading to coincidences, and the system deviated from linear. For a total number of 40,000 cells, the error in the absolute count was 15%.

Figure 8. Number of measured leukocytes, aligned between Ni lines, against dilution factor.

CD4-CD8-CD14 Immunoassay

Whole blood was incubated with CD45-labeled immunomagnetic particles, CD4-APC, CD8-Cy5.5, and CD14-APC/Cy7. Forty lines of 15 mm each were scanned at a speed of 10 mm/sec. The fluorescence signals from the PMTs were filtered, using a first-order 2-kHz low-pass filter and sampled with a frequency of 2,000 points/mm, which corresponded to 20 kHz. The APC channel was set as the trigger channel, and peak detection and pulse width measurement routines were applied. The scatter plots, obtained after compensating for the spectral overlap, are presented in Figure 9. In Figure 9a, CD4-APC is plotted versus CD8-Cy5.5, and three populations are clearly visible: CD4+ monocytes, CD4++ lymphocytes, and CD8+ lymphocytes. In Figure 9b, the CD4-APC versus CD14-APC/Cy7 staining is shown, and the CD14-APC/Cy7 and CD4+ monocytes form an obvious cluster that can be distinguished from the CD4+ and CD4++ lymphocytes (12–14). In Figure 9c, CD8-Cy5.5 versus CD14-APC/Cy7 staining is shown, confirming that no CD8 was present on the CD14+ monocytes (15, 16). The pulse width versus peak intensity of the CD4-APC channel is shown in Figure 9d. The two clusters represent the monocytes and CD4+ lymphocytes. The average pulse width at FWHM of the monocytes was 19.5 (CV = 10%) sampling points, and for the lymphocytes it was 17 (CV = 8%), which corresponded respectively to 10.0 and 8.5 μm. These obtained pulse widths did not directly correspond to the cell diameter but were the result of a convolution of laser focus with cell diameter, where the fluorochromes were expressed at the cell membrane.

Figure 9. Scatterplots of leukocytes captured and aligned in whole blood by CD45-labeled immunomagnetic particles. CD4-APC, CD8-Cy5.5, and CD14-APC/Cy7 were used as fluorescent labels. After compensation for spectral overlap, the presented scatterplots were obtained. a: Scatterplot of CD8-Cy5.5 versus CD4-APC. b: Scatterplot of CD14-APC/Cy7 versus CD4-APC. c: Scatterplot of CD14-APC/Cy7 versus CD8-Cy5.5. d: Scatterplot of pulse width in sampling points against peak height of APC channel. Sampling frequency was 2,000 points/mm; 40 points correspond to 20 μm.

DISCUSSION

We have built an instrument where CD-technology in combination with immunomagnetic selection and aligning are combined, resulting in a new cell analysis system, CellTracks. High fluorescence sensitivity, minimal sample handling, a direct measure of the absolute count of cells, and the possibility of revisiting a specific event are features which make this instrument a very versatile system which is as good as the current line of cell analyzers.

The utility of the system was demonstrated by performing a CD4-CD8-CD14 white blood cell analysis in whole blood, using the dyes APC, Cy5.5, and the tandem dye APC/Cy7. Excellent separation of the different populations was obtained. Pulse-width measurements gave a FWHM of 10 μm for the monocytes and 8.5 for the lymphocyte population.

The analysis speed is 1,000 times slower compared to flow cytometers, and permits one to digitize and store the complete PMT waveforms in memory or file. This enables the use of software routines instead of analog hardware to perform peak detection and pulse-width analysis. The software routines used in this study were simple and only permitted the selection of one trigger channel. At present, the software provides the selection of multiple channels, and logical gating strategies greatly reduced false negatives in the detection process. The PMT waveforms of events of interest clearly contain relevant information (9), and we are exploring more complex signal processing, e.g., correlation, Fourier transforms, and pulse skewness to improve performance (17).

At the current level of development, the sample throughput is too low for large-scale sample analysis. However, the system can readily be engineered to process multiple samples simultaneously, enabling high-throughput, multiparameter cell analysis.

The minimum sample handling, in combination with the possibility of revisiting a specific event, will make the system a powerful tool for the detection of rare events. The lower detection limit is determined by the sampling volume of 9 μl. Increasing the height of the chamber and increasing the Ni length will increase the sampling volume and will lower the lower limit. For a chamber with a height of 5 mm and a line length of 60 mm, the scanned volume is increased to 360 μl. This will result in a theoretical lower detection limit of 1 cell/360 μl. In the design of experiments for CellTracks, it is important to know what cell population one is interested in, as this has a direct impact on the number of cells subjected to further analysis. For example, if one is to study cells expressing CD4 (T-cell subset), CD19 (B-lymphocytes), or CD34 (progenitor cells), these are the monoclonal antibodies (mabs) used for immunomagnetic selection, and in contrast with flow-cytometric analysis, only cells that express these antigens are analyzed.

The maximum cell density where the system still remains in the linear regime was determined to be 1 leukocyte on every 30 μm, resulting in a total number of 20,000 leukocytes that can be analyzed in a single run using the current sample-chamber dimensions. This corresponds to an upper volume density of 2,222 leukocytes/μl. For cells with higher densities, the height of the chamber can be lowered to decrease the number of cells per unit line length, which enhances the maximum cell density that can be measured.

The fluorescence sensitivity of our system compared to the flow cytometer was determined using magnetically labeled APC beads. The CV values for the two brightest populations showed no difference between the flow cytometer and our system. This indicates that the magnetic aligning is equal to the hydrodynamic aligning used in the flow cytometers. An excellent separation of the three dimmest APC beads containing 333, 925, and 5,370 MESF was obtained with CellTracks, whereas the flow cytometer could not resolve these three populations. This difference in fluorescence sensitivity could be due to the difference between magnetic and hydrodynamic alignment. In our system, only the magnetically labeled (selected) beads are analyzed, whereas in the flow cytometer the whole sample is measured, and this might contain nonmagnetic but fluorescent fractions that give a scatter signal which interferes with the dimmest APC beads. For blank particles, an extrapolated MESF value of 65 was determined. This was determined for beads in PBS where no free dye was present in the sample. But even in the presence of free unbound dye, as was the case in CD4-CD8-CD14 immunofluorescence analysis of white blood cells, the fluorescence sensitivity of the system was sufficient to separate the dim CD4-monocytes from the background (Fig. 9). Replacing the CD-objective by an objective with a higher numerical aperture will result in a smaller depth of focus and lower fluorescence background, but will finally also result in a loss of fluorescent signal if the depth of focus becomes too small. The depth of focus using the CD-objective was experimentally determined to be 25 μm, for a line spacing of 8 μm.

Using CD technology, we were able to design a feedback system that focused the laser onto the lines of cells with an accuracy of 250 nm_rms in the x- and 120 nm_rms in the z-direction. The feedback signal for focusing was obtained by adding the measured intensities measured on the split detector surfaces A and B. To use this signal for focusing feedback, an offset was subtracted. This offset was dependent on laser intensity and Ni line spacing. Another detector or a change in the optical scheme might result in a signal that is independent of these parameters.

Three-color immunofluorescence using APC, Cy5.5, and the tandem dye APC/Cy7 was demonstrated with a 635-nm laser diode as a light source. Additional light sources can be placed on the system to broaden the selection of dyes. The only limitation is the lens in the present CD-objective that cuts off below 380 nm.