• imaging virtual cytometry;
  • digital slide;
  • scanning fluorescence;
  • slide-based cytometry;
  • laser scanning cytometry;
  • flow cytometry


  1. Top of page
  2. Abstract


Flow cytometry (FCM) and laser scanning cytometry (LSC) are the routine techniques for fluorescent cell analysis. Recently, we developed a scanning fluorescent microscopy (SFM) technique. This study compares SFM to LSC (two slide-based cytometry, SBC, techniques) and FCM, in experimental and clinical settings.


For the relative cell-frequency determinations, HT29 colorectal cancer cells and Ficoll separated blood mononuclear cells (FSBMCs) were serially diluted (from 1:1 to 1:1,000) and measured by each of the three techniques. For the absolute cell number determinations (only for SBC) FSBMCs were smeared on slides, then HT29 cells were placed on the slide with a micromanipulator (5–50 cells). Tumor cells circulating in the peripheral blood were isolated by magnetic separation from clinical blood samples of colorectal cancer patients. All samples were double-stained by CD45 ECD and CAM5.2 FITC antibodies. For slides, TOTO-3 and Hoechst 33258 DNA dyes were applied as nuclear counter staining.


In the relative cell frequency determinations, the correlations between the calculated value and measured values by SFM, LSC, and FCM were r2 = 0.79, 0.62, and 0.84, respectively (for all P < 0.01). In the absolute cell frequency determinations, SFM and LSC correlated to a high degree (r2 = 0.97; P < 0.01).


SFM proved to be a reliable alternative method, providing results comparable to LSC and FCM. SBC proved to be more suitable for rare-cell detection than FCM. SFM with digital slides may prove an acceptable adaptation of conventional fluorescent microscopes in order to perform rare-cell detection. © 2004 Wiley-Liss, Inc.

In the last decades, there has been a growing demand for multiparametric quantitative characterization of clinical and biological samples (1). Fluorescent assays have the highest versatility for these purposes. There are many fluorescent dyes with different excitation and emission characteristics and comparable biological specificity. These facts make it possible to create multiple staining and use different filter combinations to analyze cells. The early interest of cytometrists concentrated on cell types with high frequency (e.g., hematological samples, cell cultures, mechanically dissociated solid tissue cells). Recently, enhanced focus has been set on the rare-cells of the human organism (e.g., physiologic and pathologic stem cells, fetal cells, circulating tumor cells) (1–3).

Flow cytometry (FCM) and laser scanning cytometry (LSC) are used in practice for routine fluorescent-labeled cell analysis. FCM has proved to be a reliable, high-throughput technology (4), with the disadvantage that high cell numbers are needed, and that it is incapable of providing high-resolution morphological analysis. In the case of circulating tumor cell detection, unsatisfactory sensitivity was found (5). Slide-based cytometry (SBC) may be an alternative tool. LSC, which is applicable to fluorescent-labeled specimens, cells and tissue immobilized on slides, is also a state-of-the-art alternative for fluorescent cell analysis (6–8). It can be used for relatively small specimens, and its application for rare cell detection has also been proven (9, 10). A disadvantage is the relatively high investment and working costs, manual sample handling, and moderate measuring speed.

Fluorescent microscopy is mostly used for qualitative analysis or for documentation (“observe and judge”) (11). Automated single-parameter scanning was first introduced for histochemistry studies (12, 13). In the early 1990s, as imaging technology increased, higher resolution became available. This was paralleled by developments in image storage and digitization technology, and multiparameter techniques (14, 15). In the mid- and late 1990s, dedicated programs were developed for automated FISH analysis and rare-cell detection (16–18).

In addition to qualitative analysis by fluorescent microscopy, digitizing the fluorescence signals, and creating multichannel digital slides, quantitative analysis can be done.

Recently, we reported on a digital slide and virtual microscopy technique, the scanning fluorescent microscope (SFM) (19). A digital slide is composed of mosaicked digital images of all fields of view on a manually-selected part of a slide; these images are made and matched by the computer to create a continuous image. Based on this digital slide, virtual microscopy software simulates the basic functions of the microscope (magnification and stage movement). This technique was modified in order to perform SFM and to support cytometric evaluation of fluorescent calibration beads (20). SFM technology proved to be linear and stoichiometric on fluorescent calibration beads using Hg arc lamp excitation inhomogeneity-compensated fluorescent images, and has shown its feasibility for cytometric measurement. The advantage of the SFM method over the previously reported fluorescent microscopy scanning techniques, is the application of multichannel digital slides (images of the same field of view are made in the different parts of the light spectrum representing different fluorochromes, parameters, or channels). Off-line standard cytometry techniques (e.g., gating, scatter plots, frequency histograms, cell galleries) are possible, even via Internet protocols.

The aims of the study were to use SFM as an SBC instrument and to compare the analytic accuracy of SFM with FCM and LSC on high and low cell concentrations (between 1:1 and 1:107 cell frequency) in artificial and clinical specimens.


  1. Top of page
  2. Abstract

High Concentration Samples for Relative Cell Frequency Determination

Peripheral blood mononuclear cells (PBMCs) were isolated from EDTA anticoagulated blood (from young, cancer-free patients after informed consent and approval of the Ethical Committee of the Semmelweis University Budapest, Hungary) with a standard density gradient centrifugation (Histopack 1.077; Sigma-Aldrich, St. Louis, MO). The isolated PBMCs were washed three times in phosphate buffered saline (PBS; pH 7.4). HT29 human colorectal cancer cells cultured in RPMI 1640 (Sigma-Aldrich) supplemented with 10% fetal calf serum (Sigma-Aldrich) were detached by 0.25% trypsin, 0.02% EDTA solution (T4049; Sigma-Aldrich) and washed three times in PBS.

HT29 and PBMC cells were counted manually with a hemacytometer and mixed at different ratios (HT29 to PBMC ratios: 1:1, 1:2, 1:4, 1:8, 1:20, 1:50, 1:100, 1:500, and 1:1,000) in three replicates. 100,000–200,000 cells in 50–100 μl were used per mixture. A total of 5 μl anti-human CD45 ECD (pan-leukocyte marker, PE-Texas Red tandem dye; Beckman Coulter–Immunotech, Krefeld, Germany) and 5 μl anti-cytokeratin antibody (CAM 5.2-FITC, BD-Biosciences, San Jose, CA) antibodies were added and incubated 30 min in the dark at room temperature. After 2× washing in PBS by centrifugation at 1,500 rpm (5 min, Janetzki T32), the cells were fixed for 20 min in 0.5% phosphate buffered paraformaldehyde. Cells were pelleted, and the supernatant was discarded. Then the cells were resuspended in the remaining 100 μl solution. The DNA-specific fluorescent dyes TOTO-3 (Molecular Probes, Eugene, OR) in 5 μmol final concentration and Hoechst 33258 (Molecular Probes) in 100 nmol final concentration in PBS were added and cells were stained for 20 min, in the dark at room temperature. Cells were then washed in PBS, and the supernatant was discarded. Then samples were divided into two parts for FCM and SBC measurements, respectively.

From 10 μl of the resuspended pellet, smears were placed onto a spot with a diameter of 5–6 mm in the middle of conventional glass slide. Smears were covered by the Pro-long antifading agent (Molecular Probes). It was used as recommended by the manufacturer on dry smears to stabilize the fluorescent signal for a long time and permit multiple scanning. The same slides were used for SFM and LSC measurements.

Low Concentration Samples for Absolute Cell Frequency Determinations (SBC Measurements) (Table 1)

Table 1. Refinding of Tumor Cells in the Low Concentration Samples I*
No. of HT29 cells placed on the slideNo. of HT29 cells determined by SFM mean ± SD, (No. of slides)No. of HT29 cells determined by LSC mean ± SD, (No. of slides)
  • *

    CAM 5.2-FITC labeled HT29 cells were placed by micromanipulator on the slide smears contained Ficoll separated, CD45-PE-TexasRed labeled, peripheral blood mononuclear cells.

56.5 ± 2.1 (4)5.2 ± 2.6 (4)
109.1 ± 1.5 (8)14.5 ± 4.4 (8)
2021.6 ± 3.7 (3)21.3 ± 11.9 (3)
5048 (1)28 (1)
Slides prepared by micromanipulation of tumor cells from tissue culture.

Histopack 1.077 separated mononuclear cells were CD45 ECD-labeled with the above protocol, smeared in the middle of the slide and dried. From a CAM5.2-labeled HT29 cell suspension, cells were sucked up and dropped on the slide among leukocytes by a micromanipulator (Carl Zeiss, Gottingen, Germany) under fluorescent microscopy control. The number of HT29 cells placed on the slide was visually counted (5–50 cells were placed on a slide). The same slides were used for SFM and LSC measurements.

Slides prepared from the blood of tumor bearing patients.

In another experiment, peripheral blood of colorectal cancer patients with different Dukes stages (B:2, C:3, D:5) were evaluated by the standard protocol (Miltenyi Biotech, Bergisch Gladbach, Germany). Shortly, mononuclear white blood cells and circulating tumor cells were isolated from 20 ml EDTA anticoagulated peripheral blood by a standard density gradient centrifugation (Histopack 1.077; Sigma-Aldrich). Cells were washed three times by PBS and labeled with the anti-human epithelial antigen (HEA-125) magnetic microbead conjugated antibody (Miltenyi Biotech, Bergisch Gladbach, Germany) with FCR-blocking reagents (Miltenyi Biotech). Enrichment of HEA-125–expressing tumor cells was achieved using magnetic cell separation columns (Miltenyi Biotech). After isolation in the magnetic field (MiniMacs, Miltenyi Biotech) the enriched cell fractions were labeled according to the above immunocytochemical labeling protocol for CD45 ECD, CAM 5.2 FITC, and nuclear DNA. After washing, cells were cytocentrifuged (STATSPIN E802/22 centrifuge, STATSPIN, Norwood, MA, at 700 rpm, 10 min) on the slide. After air-drying, cells were covered with Pro-long medium. The identical slides were used for SFM and LSC measurements.

Flow Cytometry (FCM)

The leukocyte and the HT29 cell suspensions were filtered through 50-μm nylon mesh (Partec GmbH, Münster, Germany). The analysis was performed on a FACScan flow cytometer (BDIS), equipped with a Macintosh Quadra 650 computer (Apple Computer Inc., Cupertino, CA) and CellQuest software (BD-Biosciences), using 488-nm argon-ion laser excitation at 15 mW. With FSC triggering, the fluorescence signals were detected at 530 ± 15-nm bandpass (BP) filter (FITC) and 650-nm longpass (LP) filter (ECD). A total of 10,000–15,000 cells were measured per sample.

Laser Scanning Cytometry

LSC (Compucyte Inc., Cambridge, MA) was equipped with a 20× UPLANFL (Universal plan semiapochromatic) NA 0.5 objective (Olympus). Fluorochromes were excited by a 488-nm argon-ion (5 mW) and a 633-nm HeNe (5 mW) laser, and fluorescence signals detected at 530/30 nm BP (FITC), 625/28 nm BP (ECD), and 670/20 nm BP (TOTO-3) filters. The TOTO-3 signal was used for triggering.

Scanning Fluorescent Microscopy

The Hoechst 33258 fluorescence signal (with the strongest fluorescent light) was used for focusing. After charge-coupled device (CCD) camera dark current and Hg arc lamp excitation light inhomogeneity calibration, each of the microscopic fields of view was digitally recorded in different fluorescent channels. The digital slide was evaluated using virtual microscopy and standard cytometry techniques (Fig. 1). The technique is detailed elsewhere (20).

thumbnail image

Figure 1. Image data analysis steps by the SFM program. A: A virtual microscopy evaluation field in the multichannel digital slide made by the SFM showing a clinical sample for rare cell detection. Of the six available channels, three are used. The first is applied for the Hoechst 33258 channel, the second for ECD, and the third for FITC. Magnification is 100×. A single CAM 5.2 positive cell is shown in the middle (arrow), it is surrounded by CD45 positive cells. B: Cell recognition by SFM, found objects are indicated by a circle. C: Quantitative data of cells generated by image analysis of the multichannel digital slide could be analyzed by SFM as standard cytometric results. In the scatterplot, FITC-positive rare tumor cells are gated. The x-axis shows the ECD channel intensity, the y-axis is the FITC channel intensity value. Six cells are located inside the gate. (Arrow indicates a point on the dot plot which represents a cell that was relocated on A.) D: Gallery of the six gated tumor cells with electric pseudomagnification. The Hoechst 33258–stained nuclei and the CAM5.2-FITC–labeled surface are clearly visible. The highlighted cell (arrow) in the cell gallery window is shown by the SFM in A.

Download figure to PowerPoint

The digital slide images were acquired with an Axioplan 2 MOT (motorized) microscope (Carl Zeiss Vision GmbH, Hallbergsmoos, Germany), equipped with a 100-W mercury lamp. A 12-bit Axiocam CCD camera (Carl Zeiss Vision GmbH, Hallbergsmoos, Germany), motorized object desk, and filter changer were controlled by SFM.

An entire smear area, or at least 1,000 cells, was scanned automatically at 20× magnification. Scanning sequence was Hoechst 33258, ECD, and FITC. Fluorescent light was detected for Hoechst 33258 at filter set No. 2 (all filters from Carl Zeiss) using an excitation filter (G 365), a beam splitter (FT 395) and an emission filter (LP 420 nm); for ECD the filter set No. 20 was used (excitation filter BP 546/12 nm, beam splitter FT 560 nm, emission filter BP 575–640 nm); and for FITC the filter set No. 10 (Excitation filter BP 450–490 nm, beam splitter FT 510 nm, emission filter BP 515–565 nm).

The digitization times for the single fields of view were different in the three channels, and depended on the strength of the fluorescent signals (100 ms for Hoechst 33258, 1,000 ms for ECD, and 400 ms for FITC). The spatial resolution of the system was 1 μm2/pixel. Local area network remote accessibility of the digital slides for SFM analysis was supported by the transmission control protocol/Internet protocol (TCP/IP).

Visual Fluorescent Microscopy Analysis of Rare Cells

The manual screening and evaluation of low concentration specimens was performed by two independent observers using an Axioplan 2 Imaging microscope equipped with triple path and single path filters for FITC, ECD and Hoechst 33258 staining and an Axiocam camera. The images of the single cells were recorded with the positions using the Axiovision software (V.3.0, Carl Zeiss Vision GmbH).

Statistical Analysis

The statistical analysis (linear regression, determination of correlation coefficients) was done using Sigmaplot from Statistical Program for Social Sciences (SPSS) program (SPSS V.8.0, SPSS Inc., Chicago, IL).


  1. Top of page
  2. Abstract

In our study, for the comparison, we wanted to use the same sample for all three techniques. Therefore, we had to use two different DNA dyes and ECD labeling (with broad excitation [460–600 nm] and emission [555–680 nm] spectra) to overcome the problem of the different excitation and emission spectra of the detection modalities of each instrument (SFM, LSC, and FCM). The FITC- and ECD-labeling was detectable by all three instruments (Fig. 2). For nuclear counterstaining, TOTO-3 and Hoechst 33258 were detectable by the LSC (CV range 12–14%) and the SFM (CV range 6–8%), respectively. The relatively high CV value in TOTO-3 staining did not impede finding cells by the LSC. In our experiments, we did not use permeabilization and ribonuclease (RNase) treatment, so DNA staining could be heterogeneous and TOTO-3 could bind to RNA also (21), giving broad variance in TOTO-3 fluorescence.

thumbnail image

Figure 2. Differentiation of peripheral blood cells and HT29 tumor cells by the three modalities. Scatterplots show clearly distinguishable peripheral blood cells (R2) and the HT29 cell population (R3), by SFM (A), LSC (B), and FCM (C). x-axis, CAM5.2 FITC labeling; y-axis, CD45 ECD labeling. With SFM and LSC, a relatively high aspecific red fluorescence of the CAM5.2 FITC–labeled tumor cells was detected. Therefore, the HT29 cell population shows unexpectedly high ECD values. This fluorescence was compensated for in the LSC and FCM measurements.

Download figure to PowerPoint

High Concentration Samples for Relative Cell Frequency Determination by FCM and SBC Measurements

In the measurements we made, the most suitable fluorescence detection of cell surface markers was shown by the FCM, followed by SFM and LSC. In the FCM measurements on the dot plots, there are clearly distinguishable cell populations (Fig. 2C), which are less distinct with SFM and LSC (Fig. 2A and B). If the gating is not correct (for FCM), due to the lack of visual reinspection, several events can be classified as false-positives. However, the correlation between the measured and expected cell frequency (Fig. 3) was the highest (r2 = 0.84; P < 0.01) with FCM, followed by SFM (r2 = 0.79; P < 0.01) and LSC (r2 = 0.62; P < 0.01). In the FCM measurements, one of the three replicates series showed outlier results. If we exclude these data from the evaluation, even higher reproducibility can be achieved (Fig. 3C). LSC showed systematically higher relative tumor cell frequencies as the calculated value. SFM showed the highest variation in the measured cell frequencies. The measurement time for the evaluation of the sample on the flow cytometer took 3–5 min, on the LSC 10–15 min, and on the SFM 40–50 min.

thumbnail image

Figure 3. Correlation of determined and expected cell frequencies. Regression lines showing correlation between the expected (diluted, y-axis) and determined (x-axis) cell frequencies of HT29 cells to peripheral blood cells in dilution series (1:1 to 1:1,000), by SFM (A), LSC (B) and FCM (C), respectively. At each concentration, at least three samples were prepared and measured. For details see text.

Download figure to PowerPoint

Low Concentration Samples for Absolute Cell Frequency Determinations (SBC Measurements)

If we placed 5, 10, 20, or even 50 fluorescent-labeled cells into the measuring tube by micromanipulator (5-ml tube, 100-μl cell solution), we could not detect any FITC-labeled cells by FCM. In the case of the circulating tumor cells, we could not compare the same preparation because of lost cells after FCM. Therefore, in the low concentration specimen, we could compare only SFM and LSC with visual reevaluation of the same slides.

In the micromanipulation slides, the two systems showed very similar results (Table 1). The correlations between the number of placed tumor cells and that determined by SFM and LSC were very high (r2 = 0.96, P < 0.001; r2 = 0.95, P < 0.01 respectively).

In the circulating tumor cell samples of seven patients, the two systems showed similar results (Table 2). In the other three cases, LSC yielded a higher number of cells than SFM (749 versus 166, 355 versus 43, and 61 versus 5). In these studies, LSC overestimated the number of tumor cells. The number of the CAM 5.2 FITC-positive cells on the slide by SFM and LSC showed significant correlation to the visual counting (r2 = 0.99 and 0.97, respectively, all P < 0.01) and to each other (r2 = 0.97, P < 0.01).

Table 2. Finding of Tumor Cells in the Low Concentration Samples II*
Patient IDNo. of CAM5.2+ cells SFMNo. of CAM5.2+ cells LSCNo. of CAM5.2+ cells, visual observation
  • *

    Absolute numbers of magnetic isolated circulating tumor cells as detected by SFM, LSC, and microscopy in colorectal cancer patients.



  1. Top of page
  2. Abstract

Detection of circulating tumor cells is of substantial clinical importance and the prognostic value of these results has been demonstrated (22). Our results show that SFM is a powerful method for finding the fluorescent-labeled tumor cells contributing to the analysis and solution of this clinical problem, and that SFM is equivalent to LSC and FCM. Alternative, highly sensitive assays like RT-PCR were found to show false-positive results in tumor-free patients (23). According to our findings (24) cytokeratin, as a marker of the epithelial cancer cells, is also expressed in gastrointestinal inflammations in lymphocytes. Therefore RT-PCR or immunohistochemistry alone is not sufficient for reliable detection of micrometastases. Today U.S. Food and Drug Administration (FDA)-approved automated microscopy systems exist only for enzyme-based micrometastases detection (e.g., Chromavision Inc., San Juan, Capistrano, CA; Applied Imaging Inc., San Jose, CA).

However, with more and more information, it appears that fluorescent cell detection with multiple labeling would have greater specificity and sensitivity than classical immunocytochemistry (25).

As fluorescent microscopy is becoming more and more widespread and is used in clinical and basic research, automation is of increasing necessity. Until now the automation was done using the traditional epifluorescent (25) or inverted microscopy stages (1, 26), which were designed for manual use, not for high-throughput use.

One of the most problematic issues in automation is autofocusing. Image analysis based autofocusing is time consuming. This is in agreement with our own experience. Automated slide loading also has not been solved in these applications. As user requirements become obvious for automation, and as algorithms are made available for automated scanning and evaluations, results with new automated and fast scanning hardware systems are expected (27).

In our study, we performed stoichiometric quantification of the fluorescence, which was based on the calibration procedure described in our previous work (20), using digital slide, compensation images, and standardization beads. Further results in this field have also been published by other groups (28). Currently, the acquisition speed for SFM is slower than that of LSC, while FCM is the fastest technology. Dedicated fluorescence microscopy hardware developments are expected in the near future from the groups of Bravo-Zanoguera et al. (27) and Tibbe et al. (29). Improvements are necessary from the manufacturers of the commercial all-around fluorescent microscopes, as well.

Comparative studies with FCM and LSC have shown that data obtained by both technologies are consistent with each other for the study of apoptosis or necrosis (30). LSC is a reliable tool for the intracellular staining of p53 protein (31) and DNA ploidy analysis (32). Moreover, SBC (LSC) provides a number of benefits that may make it more suitable for clinical laboratories than FCM (6). Even serial immunostaining could be performed for an increased number of fluorescence parameters (33).

In a comparison between LSC, FCM, and image cytometry, linear regression analysis of DNA ploidy data and SPF values of tumors also showed statistically significant agreement (34).

Our work is a comparative study using three methods. It showed that the fluorescent measurements on cell suspension or smears in dilution series yield similar results for the tumor cell/lymphocyte ratio, both in the high and low concentration ranges. Similar studies were performed with similar results using LSC and FCM (31), and image analysis and FCM (35).

In the very low frequency range of tumor cells, FCM is at somewhat of a disadvantage finding the one to two cells per sample. Similarly, the determination of the tumor cell concentration in the peripheral blood by FCM is unsatisfactory if the sample contains only one or two tumor cells.

FCM is a high-throughput technology, because data from many thousands of cells are rapidly acquired. It is also a high content (Cytomics) technology as many different parameters per cell are simultaneously acquired. On a dot-plot display events with similar signals form clusters that belong together. However, outliers from these clusters, such as rare cells, may not form clusters and are difficult to be taken into account. Cell aggregates could give signals out of scale and would therefore be lost from analysis. The measured objects would be unavailable for visual morphological inspection and would be lost after the acquisition. Since the cell would not be able to be relocated, we would have to believe the data that has been stored. Confirmation of the morphology could be done by sophisticated cell sorting or new in-flow imaging technology only (36).

After isolation of the cells from the blood, the whole sample volume should be measured. A problem arises when measuring the last drops of the cell suspension (looking for all tumor cells from the patient sample), because air bubbles in the flow channel may cause strong light scattering and many false signals are acquired, resulting in an analysis that is at least partially based on this artifact. Time gating of the data until the first air bubble could help the analysis.

SFM software using a commercial microscope and camera provides similar-quality results when finding fluorescence-labeled rare-cells as does the LSC. SBC proved to be more suitable for rare-cell detection than FCM. Due to interactive threshold determination, standard cytometry techniques, and human reevaluation of the digital slides, SFM may prove an acceptable adaptation of conventional fluorescent microscopy to perform rare-cell detection.


  1. Top of page
  2. Abstract