TruPloidİ INTRODUCTION PRIVATE Quality control and standardization are important to the clinical application of DNA cell- cycle analysis. One aspect of cell-cycle quality control that has recently gained attention is fluorescence drift. Drift, in this context, is defined as any non-random process which results in a change in the apparent fluorescence intensity of events (cells or nuclei) as a function of their relative acquisition time. These changes could arise from alterations in dye binding to DNA (either dissociation or accumulation), fluctuations in instrument alignment, fluidics, electronics or laser intensity. The potential problems caused by drift are exacerbated by the nature of the information extracted from DNA histograms. Drift during acquisition may result in high CVs, shoulders on peaks or distinct artificial peaks, depending on the duration and amplitude of the drift. Each of these features, when present on a cell-cycle histogram, can have biological or clinical significance. A current technique to identify fluorescence drift is through the collection of a Time vs Fluorescence dot-plot (1). Drift is detected by a visual examination of the plot. Although this is a useful technique it has a number of drawbacks: 1) The flow cytometer must have a TIME parameter, 2) dot-plot interpretation is subjective and therefore inconsistent, 3) data is non quantitative, 4) documentation of fluorescence stability is difficult. TruPloidİ is a computerized analysis method for detecting and quantifying fluorescence drift during post acquisition list-mode analysis. TruPloidİ, quantifies drift from list-mode files without the necessity of an acquired time parameter by taking advantage of the sequential organization of data inherent in Flow Cytometry Standard (FCS) list-mode files. The ultimate purpose of TruPloidİ is to identify DNA cell-cycle data with potential histogram artifacts caused by fluorescence drift. Once identified, the samples can be re- acquired and re-analyzed. Alternately, it is possible to time gate samples with TruPloidİ, removing cells acquired during transient drift periods. This is only advisable in those situations where the problem was of sufficiently short duration that the majority of the data acquisition was stable. Drift Analysis Theory: Clinical samples stained for DNA content may have greater than a ten-fold range in cell- to-cell fluorescence variation. Because of this heterogeneity, it is not possible to track drift by performing statistics on the fluorescence intensity of individual cells. However, TruPloidİ examines the individual cell fluorescence intensities in batches, calculating a fluorescence mean for each batch of cells. If the batches are sufficiently large, the batch fluorescence means can be compared to each other and to the total-mean (the mean fluorescence of all cells in the list-mode file). In these comparisons, the only assumptions necessary are that the cells are being sampled randomly and in a great enough number to assure representation by all possible values. Probability theory indicates that for any continuous distribution, a sample size of 200 has a 0.994 probability that 97.5% of the range of possible values will be represented in the 200 events (2). This probability does not rely on any assumed structure for the distribution. The "distribution" in our case is a DNA histogram. Therefore, calculating a mean fluorescence for successive batches of 200 cells in the list-mode file should give similar values if no systematic drift has occurred during sample acquisition. To detect the variety of potential drift situations, three separate statistical tests were used to detect fluorescence drift: 1) Slope Test: To measure long term, gradual changes in fluorescence sensitivity, a best-fit line through the batch-means is calculated by linear regression. The relative amount of change is reported as a percent change from the beginning of a sample acquisition to the end. To determine the accuracy of the slope measurement, a 99% confidence interval for the slope is calculated. This gives a 0.99 confidence that the interval contains the true slope. Any file with a slope yielding >5% change in fluorescence mean and a confidence interval that was not sufficiently wide to include a slope of 0% (flat) is flagged as drift failure by TruPloidİ. 2) Bulge Test: The slope of a best-fit line can be insensitive to transient drift occurring within a long sample acquisition. To detect these transient shifts, a test based on "runs" was used. A run of length s on one side of the median is a sequence of s values above or below the median preceded and followed by a value on the opposite side of the median (or the beginning or end of the sample). Under the null hypothesis, the probability of a run length of at least s can be calculated using the formula provided by Mosteller (3). Our null hypothesis is that the batch-means should be randomly distributed about the total-mean. In this work, the total-mean was substituted for median since the sample sizes are large enough that the mean and median should be similar. Our test is to reject the null hypothesis at a significance level a if the probability under the null of the longest run length is less than or equal to a. The Mosteller formula shows, for samples of size 70 to 110 (batch means), that rejecting the null hypothesis if a run of ten or more is present is consistent with choosing the a level closest to 0.05. Simply stated, TruPloidİ declares a bulge failure when ten or more sequential batch-means are on the same side of the total- mean. 3) 3SD Test: In a sequence of random measurements from a normal distribution, only 3 per 1000 should be greater than 3 standard deviation (SD) away from the mean (99.7% of events should be within 3SD of the mean). TruPloidİ calculates the fraction of batch- means which exceed the 3SD limit. If that fraction is greater than 0.3%, a Standard Deviation failure is generated. Similarly, in any sequence of <1000 measurements, no event should be further than 4SD from the mean. Therefore, if a single batch-mean lies further than 4SD from the total-mean, a standard deviation failure occurs. This test is designed to detect rapid transients which are too short to cause a bulge failure or a drift alert, but may be indicative of acquisition artifacts in the data. Other artifacts, besides instrument drift, may also be detected using these techniques. For example, unequal settling of cells in the sample tube during acquisition may change the sampling frequency of one population of cells in relation to another (i.e. DNA aneuploid to DNA diploid or G2 cells to G0/G1 cells). This would change the mean fluorescence and be detected as a slope failure. Although this is not strictly drift, because the fluorescence measurement of each cell remains accurate, this situation can alter cell- cycle analysis results and should be detected and avoided. REFERENCES 1. Muirhead, K: Quality Control In: Clinical Flow Cytometry : Principles and Applications, Bauer, KD, Duque, RE and Shankey, V (eds). Williams and Wilkins, Baltimore,1993,pp.177-199. 2. Montgomery DC. Introduction to statistical quality control. John Wiley & Sons, Inc., New York, 1985. 3. Mosteller F. Annals of mathematical statistics, Vol. XII:228-232, 1941. Also see Fluorescence Drift Detection as a Novel QC Procedure for DNA Cell-Cycle Analysis by Larry C. Seamer and K.K. Altobelli; CYTOMETRY(Communications in Clinical Cytometry) vol.22:60-64 (1995)