Doublet discrimination in DNA cell-cycle analysis

Authors


  • This article is a US government work and, as such, is in the public domain in the United States of America.

Abstract

Differences in doublet analysis have the potential to alter DNA cell-cycle measurements. The techniques for doublet determination are often used interchangeably without regard for the complexity in cell shapes and sizes of biological specimens.

G0/1 doublets were identified and quantitated using fluorescence height versus area and fluorescence width versus area pulse measurements, by enumerating the proportion of G2 + M cells that lack cyclin B1 immunoreactivity, and modeled in the DNA histograms by software algorithms. These techniques were tested on propidium iodide-stained whole epithelial cells or nuclei from asynchronous cultures, or after exposure to chemotherapeutic agents that induced cell-cycle arrest and were extended to human breast tumor specimens having DNA diploid patterns.

G0/1 doublets were easily discernible from G2 + M singlets in cells or nuclei that are generally homogenous and spherical in shape. Doublet discrimination based on pulse processing or cyclin B1 measurements was nonconcordant in some nonspherical cell types and in cells following cell cycle arrest. Significant differences in G0/1 doublet estimates were observed in breast tumor specimens (n = 50), with estimates based on pulse width twice those of pulse height and nearly five times greater than computer estimates. Differences between techniques are attributed to difficulties in the separation of the boundaries between G0/1 doublets and G2 + M singlet populations in biologically heterogeneous specimens.

To improve reproducibility and enhance standardization among laboratories performing cell cycle analysis in experimental cell systems and in human breast tumors, doublet discrimination analysis should best be accomplished by computer modeling. Shape and size heterogeneity of tumor and arrested cells using pulse-processing can lead to errors and make interlaboratory comparison difficult. Cytometry (Comm. Clin. Cytometry) 46:296–306, 2001. Published 2001 Wiley-Liss, Inc.

Debris and aggregates can be prominent components of DNA histograms, affecting the accuracy and reproducibility of cell-cycle estimates (1–4). Debris originate from the damage and disintegration of cells following apoptosis or the fragmentation associated with the slicing of cells or nuclei during mechanical disaggregation. Clumping may be due to the incomplete disruption of tissues by mechanical or enzymatic means into single-cell or nuclear suspensions, by the use of alcohol-based fixatives that yield DNA histograms with low coefficients of variation of the G0/1 peak but induce clumping, or by centrifugation. In addition, clumping may be an inherent attribute of some cell types, e.g., keratinocytes.

Aggregates can be composed of large clusters of cells or nuclei or two or more G0/1 (2N) events adhered together (G0/1 doublets) that are indistinguishable from particles with 4N, 6N, or 8N DNA content. Large clumps can be removed by nylon mesh filtration (typically 35–53 μm), sheared apart by passage through small gauge needles, or identified on the basis of forward-angle light scattering (2). Strategies to separate overlapping G0/1 doublets from the G2 + M1 population have utilized the gating of correlated measurements of integral DNA fluorescence pulse (≈ area) with either peak pulse height (brightness) (5) or pulse duration (width) (6–9), gating on the G2 + M cells that lack cyclin B1 protein expression (10), and computer algorithms to model aggregate probability distributions in DNA histograms (4, 11–14).

G0/1 doublet discrimination based on signal processing (pulse, width, area) is dependent on the geometry of the illumination beam and particle size (review in Shapiro (15)). Under conditions where the cell size approximates laser beam width, G0/1 doublets are not in the beam at the same time, resulting in a broadened height signal (that may be bimodal, depending on nuclear size and beam width) comparable to the height pulse of a G0/1 singlet, but with the integrated fluorescence of a G2 + M event. The width signal increases with increasing particle diameter, since it takes longer for a larger particle to traverse the laser beam. Although a G2 + M cell has twice the volume of a G0/1 cell, diameter only increases by ≈26% (based on the formula relating the volume V of a sphere to its diameter D, i.e., V = (3.14/6)D3, since if VG2 = 2VG1, DG2 = (2)1/3 DG1 = 1.26DG1), by contrast, the combined diameter of a G0/1 doublet is at least twice that of a single G0/1 event, providing that it aligns in the direction of flow. In the case of uninucleate cells, the diameter of a G2 nucleus compared to a G0/1 nucleus correlates very well with that predicted by this theory (16).

The width signal is the sum of the width of the laser beam and cell or nuclear diameter. When the vertical beam size is large as compared to particle width (used to improve measurement sensitivity by increasing the duration of particle illumination and collection), the ability to discriminate differences in pulse widths between cells or between G0/1 singlets and doublets may be marginal. While this problem can be solved by real-time subtraction of the beam width using high-precision biased amplifiers, this is rarely done with the needed degree of precision on most commercial instruments. Similarly with a large vertical beam, the resolving power of peak height versus area measurements is minimized, since the height signals are optically integrated, which decreases its difference from the pulse area signal that is electronically integrated.

G0/1 doublet discrimination using pulse height versus area and pulse width versus area measurements are sometimes interchangeably used among laboratories. In samples where there is substantial variability in cell/nuclear size or shape, doublet determination might be variable. This study critically evaluates G0/1 doublet discrimination signal processing techniques in homogenous and heterogeneous biological specimens, using commercial instruments that have standard wide-beam geometry. Results are compared to those obtained from the modeling of doublets, triplets, and quadruplets in software doublet correction modeling. Human breast tumor specimens, having diploid DNA profiles, were examined because of the strong association between increased proliferative activity (S-phase fraction) and DNA ploidy and poor clinical outcome (17–24).

MATERIALS AND METHODS

Sample Acquisition

Human breast carcinoma specimens were collected from samples submitted to surgical pathology in accordance with Institutional Review Board (IRB) regulations. Tissue samples were analyzed for DNA ploidy and the histograms classified as aneuploid when two separate G0/1 peaks were present (25). Tumor specimens with diploid DNA patterns were used in this study (n = 50). Histological examination of tissue sections verified the predominance of tumor cells (>85%) in these samples from patients who presented with primary untreated node-negative breast adenocarcinoma. All unique patient identifiers were removed prior to analysis, making additional clinical follow-up parameters, such as survival, unavailable for further study. Specimens were rapidly frozen at −70°C and stored at this temperature until analysis.

A549 and LoVo cells, derived from a human lung carcinoma and metastatic colorectal adenocarcinoma, respectively, were obtained from the American Type Culture Collection (Manassas, VA) and grown in Iscove's modified Eagle's medium supplemented with 10% fetal calf serum (FCS), 25 mM Hepes, 100 U/ml penicillin G, and 100 μg/ml streptomycin. Both cell lines have epithelial morphology. Human epidermal keratinocytes were purchased from Clonetics (Walkersville, MD) and grown in KGM-2 media. To ensure proper sampling of all cell cycle phases, media containing nonadherent floating cells were combined with the cells harvested by trypsinization (0.05% trypsin, 0.02% EDTA). For DNA cell-cycle analysis, 1 × 106 cells were centrifuged at 400g for 10 min, resuspended in either 0.5 ml FCS, and stored at −70°C, or 70% ethanol or acetone/methanol mixture, and stored at −20°C until DNA staining.

DNA Staining

Cells frozen in FCS and frozen solid tumor specimens were stained for DNA cell-cycle analysis, using the rapid nuclear isolation medium (NIM) procedure (3). Nuclear preparations yield higher quality DNA histograms in terms of the coefficient of variation (CV) of the G0/1 peak than those based on whole cells, and are the method of choice for DNA cell-cycle analysis of tumor specimens. The NIM technique was used due to its stability (2 days after staining versus 3 h using detergent-trypsin prepared nuclei technique (26)). Tumor samples were thawed in approximately 0.5 ml of cold NIM buffer and finely minced using a pair of iridectomy scissors. The nuclear suspension was filtered through a piece of 47-μm nylon mesh, and counted with a hemocytometer, and the concentration was adjusted to 2 × 106 nuclei/ml. For DNA staining, 0.5 ml of the nuclear suspension were mixed with an equal volume of NIM buffer containing 50 μg/ml propidium iodide PI (Calbiochem-Behring, LaJolla, CA) and 15 μl of freshly prepared RNAse (dissolved in water; 500 U/ml; Type IV, Sigma Chemical Co., St. Louis, MO). Stained samples were incubated at room temperature in the dark for a minimum of 1.5 h or then stored overnight at 4°C before analysis. FCS-frozen tissue culture cells were thawed at room temperature, after which NIM-PI buffer and RNAse were added as described above.

Ethanol or acetone-methanol fixed samples were centrifuged (300g, 10 min), washed in PBS (lacking Ca2+, Mg2+ salts, Gibco/BRL, Rockville, MD), and counted, and 1 × 106 cells were stained in 1 ml of PBS containing 50 μg/ml PI and 1 mg/ml RNAse for 1 h at room temperature.

For evaluation of cyclin B1 protein expression across the cell cycle and enumeration of G0/1 doublets, the cells were fixed for at least 24 h in a cold (−20°C) 1:1 acetone-methanol mixture (HPLC grade solvents), followed by washing in 70% (v/v) ethanol, and PBS, and permeabilized for 5 min on ice with 0.25% Triton X-100 in PBS containing 1% bovine serum albumin (BSA). Cells (1 × 106) were incubated with a fluorescein-isothiocyante (FITC)-conjugated antibody to cyclin B1 (Pharmingen, San Diego, CA) for 1 h at room temperature, then washed once with PBS containing 1% BSA and resuspended in PBS containing PI (10 μg/ml) and RNAse (1 mg/ml) for 30 min prior to analysis (27). FITC-conjugated mouse IgG (Pharmingen) was used as negative control for antibody staining. Comparable results could be obtained with acetone-methanol fixed cells stored at −20°C for up to 2 weeks or cells fixed in 70% ethanol and stored at −20°C. Attempts were made to perform cyclin B1 staining on isolated nuclei and solid tumor specimens previously frozen at −70°C. In the former case, cyclin B1 labeling was absent in NIM-prepared nuclei from A549 cells, and the discrimination between G1 doublets and G2 + M was minimal for A549 nuclei isolated using a sucrose-buffer-based technique (www.dcs.nci.nih.gov/branches/medicine/flowcore/nuclei.htm) and stained with cyclin B1 antibody. Cyclin B1 staining was present in nuclei prepared using the washless KI-67 technique (28) (data not shown); however, the number of G1 doublets was low (<0.5%) as measured by cyclin B1 staining, fluorescence peak/width signals, and microscopic evaluation, making statistical comparisons difficult. Thawed, previously frozen solid tumor specimens that were mechanically disaggregated and fixed in acetone/methanol showed no reactivity with the cyclin B1 antibody (data not shown).

Flow Cytometry

DNA content was measured using three different flow cytometers: FACScan™, FACScalibur, and FACSStar Plus™ (Becton-Dickinson, Mountain View, CA). FACScan™ and FACScalibur™ are the most commonly used nonsorting flow cytometers in the research and clinical environments. Both instruments utilize a quartz flow cell for sensing (430 × 180 μm, the only difference being a slight change in the internal geometry of the flow cell on the FACScalibur®). The FACSStar Plus™ was equipped with jet-in air sensing, using a 70-μm nozzle. Beam geometries for all three instruments are similar: 20 × 64 μm (FACScan™), 22 × 66 μm (FACScalibur™), and 20 × 60 μm (FACSStar Plus™). Sheath fluids were either isotonic saline (Fisher Diagnostics, Swedesboro, NJ) or Biosure sheath fluid (Riese Enterprises, Grass Valley, CA) on the FACSstar™, FACSflow on the FACScalibur™, or isotonic saline or Haema-Line 2 (Serono-Baker Instruments, Allentown, PA) on the FACScan™. Instruments were aligned with unstained and stained fluorescent polystyrene microspheres (CaliBRITE beads, Becton-Dickinson), or, in the case of FACSStar Plus™, 6.4-μm (6 peaks) Rainbow Calibration Particles (Spherotech, Inc., Libertyville, IL). Chicken erythrocyte nuclei (CENs) were used to check the linearity of the fluorescence pulse detectors. On FACScan™ and FACScalibur™, red fluorescence of PI-stained nuclei was collected through a 650-nm longpass filter into the fluorescence 3 (FL3) photomultiplier tube on linear scale at a flow rate of 12 μl/min (low), corresponding to 50–150 events/s. Some data were acquired on an older FACScan™ that lacked the ability to acquire width measurements in the FL3 channel. In this case, the high voltage and signal cables from FL2 and FL3 were exchanged prior to alignment and standardization with CENs. PI fluorescence was measured thru a 660 ± 20 nm bandpass filter on the FACSstar™ on linear scale at similar sample rates, and excited at 488 nm with 200 mW of laser power. For dual-parameter flow cytometric measurements of cyclin B1 expression and DNA content, green FITC fluorescence was measured with a 530 ± 30 nm bandpass filter (FL1) in addition to the FL3 signal.

In the course of this study, it was noted that chick red blood cells, frozen in FCS at −70°C and stained using the NIM procedure could be used in place of the detergent-treated, alcohol-fixed CEN supplied by Becton-Dickinson. Peripheral blood lymphocytes were stained in a similar manner as the tumor samples, and cell lines were used as diploid DNA standards. The coefficient of variation (CV) of the G0/1 peak was calculated by the full-width method and averaged <1.5% for diploid human lymphocytes.

Histogram Analysis

List mode files, consisting of forward and side scatter, integrated (area), peak (height) and pulse (width) red fluorescence, were analyzed using CellQuest™ (Becton-Dickinson). Raw data were minimally gated by excluding only channels 1–5 and 1020–1024, respectively, during sample acquisition. Cell-cycle analysis of the DNA histograms of integrated red fluorescence was performed with Modfit (Verity Software, Inc., Topsham, ME), using the F_DIP_T1 model (fresh, diploid, aggregrates) with the S-phase fraction modeled as multiple broadened trapezoids, and Multicycle (Phoneix Flow Systems, San Diego, CA), using a zero-order polynomial to model the S-phase fraction. Both software packages include algorithms for estimating multicut debris, and in the case of doublets, they model doublets, triplets, and quadruplets as probabilistic events (29). Debris subtraction was performed in all samples prior to cell-cycle analysis. List mode data were displayed as two-parameter correlated histograms of peak width, peak height, and peak area for reprocessing, using different gating strategies.

Statistical comparison of data was performed using the Statgraphics statistical software package (STSC, Rockville, MD). Variance between doublet estimates and cell-cycle fractions was tested using the chi-square test at the 95% confidence limit. Data are reported as the mean ± standard deviation.

RESULTS

Doublet Discrimination in Biologically Homogenous Samples

Identification of G0/1 doublets in asynchronously growing A549 cells cultured in vitro (Fig. 1) or human peripheral blood lymphocytes (data not shown) stained for cell-cycle analysis was straightforward, using fluorescence pulse processing signals. In dot plots of height versus area or width versus area, doublets were clearly discriminated from other stained cells (Fig. 1A). Independent of the peak-processing methods, doublets can be detected using cyclin B1 staining; this technique is based on the rationale that cyclin B1 protein expression is restricted to the G2 + M phases of the cell cycle, hence, G2 + M cells lacking cyclin B1 immunofluorescence are G0/1 doublets (Fig. 1B). Doublets identified by cyclin B1 staining fell within the “doublet” regions of height versus area or width versus area dot plots (Fig. 1B, insert). Overall, all three strategies yielded comparable results (Fig. 1C), irrespective of the type of flow-chamber sensing (jet-in-air or closed quartz).

Figure 1.

G0/1 doublet discrimination in A549 cells, using fluorescence pulse processing and cyclin B1 protein expression. A: Pulse height versus area (left) and pulse width versus area (right) dot plots. G0/1 doublets are depicted by rectangular region. B: Cyclin B1 immunofluorescence versus DNA content (area). Immunofluorescence of cells stained with FITC-conjugated mouse IgG1 is shown by the trapezoidal area and reflects the staining of cells lacking cyclin B1 protein expression. G0/1 doublets are identified in the ellipsoidal region. Insets: Positions of cyclin B1-identified G0/1 doublets in either the height versus area or width versus area dot plots. C: Comparison of percentage of G0/1 doublets identified by three techniques using different flow cytometers. N = 15 for each determination. Dot plots represent 25,000 cells. No differences are observed (P > 0.8).

Doublet Analysis in Biologically Complex Specimens

In some tumor cell lines, primary cultures, and cells exposed to cell-cycle inhibitors, pulse processing and cyclin B1 immunofluorescence G0/1 doublet detection techniques were not mutually inclusive. G0/1 doublets could be clearly identified in PI-stained unperturbed LoVo cells, using either height versus area or width versus area measurements (6.5 ± 1.8% height versus area, as compared to 6.9 ± 1.3% width versus area; n = 5, P > 0.8) (Fig. 2A). However, cyclin B1 protein expression was diminished in these cells, resulting in an overestimation of number of G0/1 doublets by this technique (24.2 ± 2.8% versus an average of 6.2 ± 1.6% for the height versus area or width versus area measurements combined; n = 7, P < 0.05). Reanalysis of the position of doublets identified by cyclin B1 staining in width versus area dot plots showed that both contained G2 + M and G0/1 populations (Fig. 2A, inserts).

Figure 2.

G0/1 doublet discrimination in complex samples, using (left to right) cyclin B1 immunofluorescence, pulse height versus area, and pulse width versus area. Insets: Positions of G0/1 doublets in the two other corresponding dot plots after gating of the primary (larger) dot plot (e.g., width versus area and cyclin B1 versus DNA content, gated on height versus area G0/1 doublets). Shaded areas in height versus area and width versus area dot plots depict positions of singlet events (G0/1, S, and G2 + M). Immunofluorescence of cells stained with FITC-conjugated mouse IgG1 is shown in trapezoidal areas, and reflects staining of cells lacking cyclin B1 protein expression. G0/1 doublets are identified in the ellipsoidal region in the cyclin B1 or height versus area dot plots, or in the rectangular region in width versus area dot plots. Dot plots represent 25,000 cells and are representative of at least three experiments. A: LoVo cells. B: Human epidermal keratinocytes. When gated on the basis of height versus area, G0/1 doublets contained both cyclin B1-positive (arrow) and -negative populations. C: A549 exposed to SKF-96365, an M-phase blocker. D: Comparison of percentage of G0/1 doublets in SKF-96365-treated cells by instrument. N = 9 separate determinations. P < 0.05, comparing G0/1 doublets estimated on the basis of cyclin B1 staining and pulse-processing techniques. E: Comparison of percentage of G0/1 doublets in taxol-treated A549 cells, using FACScan and FACScalibur. N = 5 separate determinations.

Conversely, while width versus area and cyclin B1 doublet detection methods yielded nearly equivalent numbers of G0/1 doublets in samples of PI-stained primary human keratinocytes (Fig. 2B; 4.5 ± 0.8% width versus area, as compared to 4.9 ± 0.3% cyclin B1; n = 5, P > 0.8), or Hs578T cells (a human breast cancer cell line that exhibits a high degree of chromosomal instability; data not shown), doublet discrimination using peak versus area detection was poor (23.4 ± 4.4%.). Not surprisingly, putative height versus area-identified G0/1 doublets contained mixtures of both G0/1 doublets and G2 + M cells in the width versus area or cyclin B1 versus DNA content dot plots (Fig. 2B, inserts).

The correlation between the doublet detection techniques was poorest in cells exposed to agents that induce cell cycle arrest and may alter cyclin B1 protein expression. Compared to untreated cells (above), in both the height versus area or width versus area dot plots, G0/1 doublets and G2 + M cells populations overlapped after treatment with SKF-96365, an agent that induces M-phase arrest (27) (Fig. 2C; similar patterns were observed in colcemid-, vinblastine-, and nocodazole-treated cells, data not shown), and showed large variation with the number of G0/1 doublets estimated from cyclin B1 protein expression (Fig. 2D). Notably, this uncertainty in positioning of the G0/1 doublet gate in height versus area or width versus area dot plots underestimated the number of doublets (average ≈1%, as compared to 4% by microscopic examination). In A549 cells, exposure to 10 nM taxol induced apoptosis (30); again, the number of G0/1 doublets estimated from cyclin B1 protein expression varied (Fig. 2E).

Doublet Analysis in Biologically Heterogeneous Specimens- Human Breast Tumor Specimens

The poor separation between G0/1 doublets and G2 + M cells in the height versus area and width versus area dot plots from cells treated with cell cycle inhibitors might have been due to heterogeneity in nuclear size and/or shape. To test this hypothesis, PI-stained nuclei were prepared from normal human breast tissue and tumors, and the percentage of G0/1 doublets was evaluated using the pulse height and width techniques. Breast tumors are heterogeneous (31) and represent a human cancer where DNA cell-cycle and ploidy measurements have clinical relevance (reviewed in Wenger and Clark (24)). As shown in Figure 3, G0/1 doublets present in nuclear preparations of normal breast tissue are clearly separated from the G2 + M fraction in either the height versus area and width versus area dotplots. By contrast, the width signals of G0/1 doublets in tumor nuclei samples are broad, making the clear identification of the boundary between the G0/1 doublets and G2 + M population difficult. To exclude the possibility that these observations were due to the use of nuclei for cell-cycle analysis, height and width pulse-processing doublet-detection techniques were tested on PI-stained nuclei from A549 cells prepared using the NIM method. No differences were observed in the number of G0/1 doublets estimated from height versus area and width versus area dot plots analyzed either on the FACScalibur™ (4.7 ± 2.2% height versus area; 5.1 ± 2.2% width versus area; n = 20, P > 0.8) or FACScan™ (3.5 ± 2.3% height versus area; 3.3 ± 2.1% width versus area; n = 20, P > 0.8), nor do the results significantly differ from those obtained with whole cells (Fig. 1C; P > 0.5).

Figure 3.

G0/1 doublet discrimination in human breast normal and tumor tissue using fluorescence pulse processing. Left: Height versus area and width versus area dot plots from normal tissue. In both dot plots, G2 + M singlets and G0/1 doublets are clearly separated (arrow in width versus area). Dot plots represent 25,000 cells and are representative of at least three experiments. Right: Tumor sample with near-diploid aneuploid stemline. Doublets appear better separated from singlet events in the height versus area dot plot than in width versus area (large arrow). This sample also illustrates that both the diploid and aneuploid G0/1 (grey arrow) and G2 + M singlet width signals are a broad continuum.

Microscopically, nuclei prepared from breast tumor tissue using the NIM technique appear as a continuum of various shapes, ranging from the spherical to the markedly elongated. Consequently, this might account for the indistinct separation of G0/1 doublets from the G2 + M fraction by pulse-width signals. To further explore this possibility, the percentage of G0/1 doublets in 50 breast tumor specimens having diploid DNA histogram patterns was evaluated, using the pulse-processing techniques, and compared to the results from computer modeling of doublets. Figure 4 illustrates the height versus area and width versus gating strategies used for doublet analysis in a representative tumor specimen. In height versus area dot plots (shaded area, Fig. 4A), region R1 contained DNA singlet events (G0/1, S, G2 + M; upper red fluorescence area histogram), while region R2 contained G0/1 doublets (lower red fluorescence area histogram). Analysis of the position of G0/1, S, G2 + M singlets and G0/1 doublets in the width versus area dot plot, based on the R1 and R2 height versus area gates (upper right dot plot, Fig. 4A), indicated that R1 contained only G0/1, S, and G2 + M singlets. Importantly, R2 contained a mixture of both G2 + M singlets and G0/1 doublets (lower right dot plot, Fig. 4A). By contrast, the width versus area dot plot of the same sample (shaded area, Fig. 4B), was divided into three regions: R1 contained DNA singlet events (G0/1, S, and G2 + M; upper red fluorescence area histogram); R2 contained putative G0/1 doublets (middle red fluorescence area histogram); and R3 was composed of a mixture of singlet, doublets, and aggregates (lower red fluorescence area histogram). In the width versus area dot plots, no discrete boundary was present clearly separating G0/1 doublets from G2 + M nuclei. Corresponding analysis of the position of G0/1, S, and G2 + M singlets and G0/1 doublets in the height versus area dot plots, based on the R1, R2, and R3 width versus area gates (right dot plots, Fig. 4B) indicated that R1 contained both G0/1, S, and G2 + M singlets and G0/1 doublets (upper right dot plot, Fig. 4B), R2 both G0/1 doublets and G2 + M singlets (middle right dot plot, Fig. 4B), and R3 G0/1, S, and G2 + M singlets, and G0/1 and G2 + M doublets (lower right dot plot, Fig. 4B).

Figure 4.

Comparison of fluorescence pulse height versus area and width versus area doublet-discrimination techniques in DNA diploid human breast tumor specimens. A: Height versus area. B: Width versus area. The primary dot plot is shown in the shaded area and depicts gating regions. Dot plots at right show the corresponding gated dot plot (i.e., width versus area for gating on height versus area regions). The integrated red fluorescence (area) histograms from the gating regions are shown below each shaded dot plot. The primary dot plot represents data from 25,000 nuclei. B: Region R3 corresponds to right side of dot plot and includes R2. The rationale for positioning R2 and R3 with respect to singlet events was based on the uppermost width signal of G0/1 singlets and was approximately channel 560. Data are representative of results from 50 tumor samples.

The results from these analyses are depicted in Figure 5 and show significant differences between the pulse-processing and computer-modeling techniques. On average, the number of G0/1 doublets estimated from the width versus area dot plots was nearly three times those from height versus area measurements. While no differences were observed using different modeling software programs (Modfit and Multicycle), these values were generally one-half and one-sixth those of the height versus area and width versus area measurements, respectively. For width versus area dot plots, decreasing the size of R2 (from ≈channel 560 to ≈channel 700) decreased the percentage of G0/1 doublets (to 7.3 ± 6.0% from 11.6 ± 9.3%; Fig. 5), nearly equivalent the doublet estimates (R2) based on the height versus area dot plot.

Figure 5.

Comparison of fluorescence pulse height versus area and width versus area doublet discrimination techniques and computer modeling in DNA diploid human breast tumor specimens. Data obtained from 50 tumors. *Observations that are statistically significant (P < 0.001). Regions in Figure 4 were used for doublet estimates, based on pulse-processing gates (R2 for height and width doublets). Narrow and wide designations refer to changing R2 in the width dot plot from channels 560–575 to channel 700, respectively.

DISCUSSION

In cell preparations stained with DNA-specific fluorochromes, G0/1 doublets can be identified and removed prior to cell-cycle analysis using fluorescence pulse height or width signals, or by the absence of cyclin B1 immunoreactivity, or modeled in the DNA histogram by computer algorithms. These techniques yield comparable results when the cells are spherical in shape, e.g., peripheral blood lymphocytes or trypsinized A549 cells, and express normal cyclin B1 protein levels. By contrast, the methods are not mutually inclusive and produce variable G0/1 doublet estimates in growth-arrested cells or in human breast tumor tissue, presumably due to heterogeneity in cell/nuclear shape and size, or when cyclin B1 protein expression levels are abnormal, as in perturbed cells. From the data presented here, G0/1 doublet discrimination in biologically complex breast tumor specimens, using width versus area pulse analysis, is confounded by uncertainty in the placement of the doublet gates due to the lack of a discrete boundary between G2 + M singlets and G0/1 doublets. The net result is overestimation of the number of G0/1 doublets and underestimation of the G2 + M fraction. G0/1 doublet estimates, using height versus area dot plot analysis, were nearly twice those obtained from computer modeling of DNA histograms. Rabinovitch (11) showed good correlation between computer modeling of G0/1 doublets and visual observation. For the breast tumor specimens studied here, the percentage of doublets varied from 1–3%, well within the range estimated from computer modeling. These results further support the objectiveness and reliability of computer models for G0/1 doublet detection.

Likewise, instrumental problems can affect G0/1 doublet estimates. Doublets may be correctly identified only if they are aligned sequentially and perpendicular to the laser beam. Consequently, errors in doublet detection may result when G0/1 doublets are oriented in parallel (side-by-side) or off axis to the excitation beam, as occurs in momentary hydrodynamic instability caused by clogs. However, under normal conditions of stable laminar flow, this does not happen, and doublets align well in the direction of flow. Likewise, the results here were obtained on instruments having a fixed, standard wide-beam geometry (FACScan, FACSCalibur, and FACStar). Instruments having non-comparable beam and cell sizes might yield dissimilar results.

With respect to DNA cell-cycle analysis, does the inclusion or removal of G0/1 doublets have any real effect on cell cycle estimates? Without G0/1 doublet subtraction, the G2 + M fraction of the DNA diploid breast tumor specimens studied here (n = 50) averaged 12.1 ± 1.7%. Height versus area or width versus area doublet subtraction decreased the percentage of G2 + M cells to 5.6 ± 2.3% and 3.8 ± 2.6%, respectively (P < 0.05, comparing with and without subtraction), which was similar to the values from computer modeling (3.6 ± 2.4%, Multicycle; 4.6 ± 2.2%, Modfit). The lower number of G2 + M cells in width versus area-corrected DNA histograms, compared to height versus area doublet subtraction, is due to the inclusion of G2 + M singlets in the “doublet” region. Because cell-cycle percentages are usually calculated relative to each other (i.e., the sum of the G0/1, S, and G2 + M phases equals 100%), decreasing the number of G2 + M events (by overestimating G0/1 doublets) will affect the estimates of other cell-cycle compartments. The sum of the S + G2 + M phases is frequently used as a measure of the proliferative/growth fraction (30), and here ranged from 14.5 ± 4.0% (no doublet discrimination) to 8.8 ± 4.0% (height versus area) and 6.9 ± 3.9% (width versus area; P < 0.05 compared to height versus area). S-phase estimates after height versus area or width versus area doublet subtraction were similar (4.9 ± 3.5% and 5.6 ± 3.6%, respectively) and comparable to nonsubtracted values (4.9 ± 2.9%). Interestingly, S-phase estimates were also comparable, even though different models were used (2.8 ± 1.6%, Multicycle; 3.7 ± 0.8%, Modfit). This result may be dependent on the low CVs and percentage of S-phase cells in these specimens, and could be a source of variation in the cell-cycle analysis of tumor samples that do not share these characteristics. S-phase estimates based on computer modeling using Multicycle were significantly lower (P <0.05) than those obtained from pulse-processing techniques. As similarly noted by Bergers et al. (32), the decrease in S-phase estimates following computerized doublet correction has the clear potential to introduce interlaboratory discrepancies in the placement of clinical specimens into prognostic subgroupings. Alternatively, variations in doublet estimates could alter the ploidy classification of diploid/tetraploid DNA histograms (33) or affect the prognostic value of the percentage of G2 + M cells in prostate cancer (34). Because of the subjectivity involved in positioning the boundary between G2 + M events and G0/1 doublets in complex heterogeneous samples, particularly with fluorescence pulse width versus area measurements in fixed-beam commercial instruments, we suggest that G0/1 doublets be estimated by computer modeling. This appears to be the most conservative approach for improving reproducibility and enhancing standardization of doublet discrimination analysis. Perhaps electronic doublet-correction techniques, except for the simplest cases, should be reserved for situations such as cell sorting requiring real-time, rather than off-line, corrections for doublets. More complex types of cells, such as epithelial cells, make electronic doublet-correction difficult to reliably reproduce between laboratories that have different electronics of differing sensitivity and precision of calibration.

Human breast tumors are composed of mixtures of epithelial tumor cells, stromal tissue, and to a lesser degree, mononuclear cells. The use of cytokeratin antibodies to identify and remove diploid nontumor tissue or infiltrating lymphocytes from DNA histograms has been advocated as a method to improve histopathologic correlates in clinical DNA cell-cycle analysis (35–37). To examine whether this technique might improve the correlation between pulse-height and width-doublet discrimination, a limited number (n = 5) of mechanically disaggregated, ethanol-fixed intact (37) fresh breast tumor specimens were studied. Even after removal of the cytokeratin-negative elements from the height versus area or width versus area dotplots, G0/1 doublets still did not constitute a discrete population clearly separated from G2 + M cells. (data not shown). Again, software modeling appears to be the most objective rationale for doublet discrimination.

Infrequently, light scatter has reportedly been used to detect G0/1 doublets (38, 39). Backgating of pulse-identified or cyclin B1-identified G0/1 doublets indicated that neither forward nor 90° light scatter combined with a fluorescence pulse (area/height/width) measurement could be used to identify or exclude G0/1 doublets from DNA histogram analysis (data not shown). Consequently, this process most likely excludes nondoublet cells from DNA histograms, and may alter cell-cycle analysis.

Guidelines for DNA cell-cycle analysis of clinical samples (25) suggest that specimens be excluded from analysis when the combined levels of sample debris and aggregates exceed 20% of the total number of histogram events. Surprisingly, doublet discrimination is often absent or used sporadically in multicenter studies comparing interlaboratory variations in DNA measurements and S-phase fraction analysis (40–42). The discrepancies between doublet-discrimination techniques presented here and the differences in SPF estimates obtained by different models and debris corrections (3, 4, 32, 40, 43) add yet another level of complexity to the transposition of prognostic variables obtained from DNA cell-cycle analysis of breast tumors between laboratories (24).

Acknowledgements

The authors thank Leopold G. Koss for his critical evaluation of the manuscript and his role in stimulating numerous thought-provoking issues related to improving the potential of DNA flow cytometry as a reproducible, clinically relevant prognostic tool.

  • 1

    Single–parameter DNA histograms cannot resolve G2 from M phase cells, and in this report are considered to be a single entity.

Ancillary