etry data for ploidy and cell cycle kinetics ... eral reviews, 43 with ploidy analysis, 14 with ploidy and cell kinetic analyses, and one with cell kinetic analysis only.
0 1988 Alan R. Liss, Inc.
Cytometry 9:494-498 (1988)
How Reproducible Are Flow Cytometry Data From Paraffin-Embedded Blocks?' T.E. Kute,' B. Gregory, J. Galleshaw, M. Hopkins, D. Buss, a n d D. Case Departments o f Medicine (T.E.K., B.G., J.G.) and Pathology (M.H., D.B.), Center for Prevention Research and Biometry (D.C.), Bowman Gray School of Medicine, Winston-Salem,North Carolina 27103 Received for publication February 17, 1988; accepted April 28,1988
The purpose of this technical report is to determine the reproducibility of flow cytometry data for ploidy and cell cycle kinetics using paraffin-embedded blocks of breast cancer tissue. One block from each of 39 tumors was studied in this report with each block having multiple sections analyzed independently. All of these sections gave ploidy analyses, while only 34 gave cell kinetic values. The standard deviation for the DNA index value in the multiple analysis study was less than 0.1 in all but three cases. The cell kinetic values gave larger variability, and the actual
A review was performed on 65 articles published between 1983 and 1986 that had cited Hedley's procedure (6) on the use of paraffin-embedded (PE) sections for DNA analysis. These reports cited at least 26 different tissues, with colon, breast, and lymphoma being the most frequently analyzed types. There were seven general reviews, 43 with ploidy analysis, 14 with ploidy and cell kinetic analyses, and one with cell kinetic analysis only. This would indicate that cell kinetics is used in only 26% (15/58)of these studies. Because of the interest in this procedure, the following study was performed in order to establish objective criteria to evaluate the flow cytometry data obtained from PE blocks. This report provides information obtained from a reproducibility study of paraffin-embedded breast cancer tissues but could be applied to other tissues.
values were dependent on the method of analysis. Comparison of the variability for each method of analysis could not predict which procedure was superior. These results would indicate that ploidy is a reproducible value, while cell kinetic parameters should only be used as an indicator of proliferative activity that has been normalized to the mean or median of a large set of observations processed and analyzed by the same procedure. Key terms: Breast cancer, cell kinetics, ploidy analysis, variability of analysis
propidium iodide (PI), and 37 pg/ml of RNAase (75k unitdmg). This solution was made fresh daily in order to prevent loss of RNAase activity. The stained nuclei were analyzed on a fluorescence-activated cell sorter (FACS) 440 flow cytometer with a n argon ion laser as the excitation source (i.e., 448 nmj, and the parameters for the emitted light were analyzed for forward and 90" light scatter, pulse width (to discriminate doublets), and red fluorescence (630 nMj of PI to determine DNA content per nucleus. In most cases 40,000 events were measured per analysis. The resulting data were acquired in list mode for discriminative analyses such as the use of various gating procedures to eliminate debris, doublets, and normal cells if a n aneuploid group was present. Ploidy status was determined by a DNA index, which was defined as the ratio of the peak channel of the tumor Gl,o peak to the peak channel of a diploid control PE tissue or the diploid cells in the tumor itself. A DNA index of 1.0 i-.1was defined as diploid. The actual data for each assay are given in Figure 1. Cell cycle kinetics were determined for each aneuploid population, and, if only diploid populations were obtained, the analysis was done on the diploid group. Several techniques were used
MATERIALS AND METHODS Thirty-nine PE tumor samples from primary biopsies were selected from a larger study of 650 node-positive breast cancer patients. Three to four 50-pm consecutive sections were sliced from each PE block, processed, stained, and analyzed independently by Hedley's procedure (6)except for the use of Fisher's Hemo-DE instead lSupported by grants from Gaston County Cancer Association and of xylene to dissolve the paraffin. Following hydration, Oncology Research Center at Bowman Gray School of Medicine. pepsination, and centrifugation, the single-cell pellet was 'Address reprint requests to Timothy E. Kute, Ph.D., Department of stained with a nuclear isolation media (10) containing Medicine. Bowman Gray School of Medicine. 300 S. Hawthorne Road. PBS with C a + + and Mg++, 0.6% NP-40, 50 pglml of Winston-Salem,NC27io3.
REPRODUCIBILITY IN PARAFFIN-EMBEDDEDBLOCKS
2.0
495
1
1.8'
$ 1.6TL ! *
-
2 1.4' n
1.2'
0
.
=
1.0-
0)
0.8 0.6
I
-.4
I
I
I
-.2
I
1
Mean
Mean f
1-
+.2
+.4
SD
FIG. 1. Variability of the mean DNA index values. Three to four the block mean and standard deviation values were calculated. Total consecutive samples from a PE block were analyzed individually by number of samples was 39. flow cytometry;the DNA index was determined for each sample, and
for cell cycle analysis. These included the curve-fitting routine available on our instrument, rectangular S fit analysis (l), and area fit analysis. The latter is a gating procedure of G1/O area and G2-M area based on the observable gaussian curves; %S activity is the proportion of cells not in the gaussian distributions. The mean, standard deviation (SD), and coefficient of variation (CV = [SD/mean] x 100) were calculated for each set of triplicate or quadruplicate sections. Differences in the means and variability of the various methods were assessed using paired analyses.
RESULTS Our first studies of flow cytometry used frozen or fresh tumor tissue for analysis (8).The processing of this tissue was usually mechanical disruption and/or enzymatic digestion of the nuclei prepared from these cells. These tumors gave DNA histograms that were easy to interpret, but obtaining long-term clinical data on these patients was not possible. Others have shown that cells obtained via the PE procedure give DNA histograms similar to those obtained from frozen or fresh tissue (6,9),In particular, our group observes a good correlation (i.e.?0.91) between the values for the DNA index in 13 tumors analyzed by both methods (Data not shown). Figure 1 illustrates the variability of the DNA index from paraffin-embedded tissues. Three to four slices from the PE blocks were analyzed independently, and the resulting means of the DNA index (k one standard deviation) are plotted in the figure. The standard deviation was less than 0.1 in all but 3 cases. In most cases, a diploid standard from samples of a normal node, liver, or spleen had a Gl,o peak that coincided with the peak
that was defined. as diploid cells in the breast tumor. These paraffin-embedded diploid standards were useful in interpreting DNA histograms that might otherwise be confusing. However, several tumors had single peaks that did not coincide with the defined normal diploid populations. These tumors were hard to interpret for ploidy status, but they are reproducible from slice to slice. This abnormality probably reflects a fixation artifact since this is never observed using fresh or frozen tissue. The percent aneuploid population in this group will vary depending on the interpretation of the results. If single Gllo modal peaks are defined as diploid, the percent aneuploid of this group is 49%. If a diploid tumor is defined with a DNA index of 1.0 f .l, the percent aneuploid increases to 64%. This definition of diploid was based on f2 times the pooled estimate for the standard deviation of the DNA index. Five of the 39 tumors could not be analyzed for cell cycle kinetics. Two had too few cells, and three had aneuploid peaks that overlapped the diploid population or had poor quality DNA histograms. In Figure 2, one observes the broad spectrum of values for percent S activity obtained by either the area fit procedure or the rectangular S phase procedure. In general, these two methods gave similar relative results for each patient's tumor, but the rectangular S fit analysis was usually larger because it included cells that were by definition in the Gllo or Gz/M Gaussian distribution areas. The mean percent was 5.8% for the S area fit procedure and 11.4%for the rectangular S fit procedure ( P < -001).The pooled estimate of variance was also larger for the rectangular fit compared to the area fit (5.62 vs. 1.35). However, variance increased with the mean. A regres-
KUTE ET AL.
496 40-
30'
v,
* C
,
W
2?! 2 0 C
0 W
.
I 10-
-
&
I .
1
f
Rectangular Fit
Area Fit Mean
* SD
FIG.2. Variability of %S values. Tissues were analyzed by flow cytometry and the %S activity was determined by area fit and rectangular S fit. The means and standard deviations for each sample are presented. Total number of samples was 34.
sion model of %S standard deviation on %S mean was CONCLUSIONS fitted, and a signed rank test on the residuals was used to compare variability between methods adjusting for In this technical report we examined the variability of mean differences. Once adjustment is made for the mean, ploidy and cell cycle kinetic data from PE tissues. It is the difference in variability between the two methods is clear from these data and others that DNA values are not statistically significant ( P = .72). highly reproducible between consecutive PE sections Percent S values were also obtained by a curve-fitting from an individual block (Fig. 1).A definition of diploid analysis procedure. The fitting program (6) gave a large as a tumor with a DNA index of 1 0.1 is conservative. variability, even in cases where the raw histogram data It is quite possible that combining multiple sections for were superimposable (data not shown). Therefore, we a single analysis would provide less variability in the feel that these computer-fitted procedures do not provide final results. These experiments have not been done, information that is superior to the above procedures. It and the results might vary depending on the nature of is quite possible that other programs might do a better the tumor in the block. Use of diploid cells in the tumor job. tissue and standardized diploid controls (i.e., paraffinThe percent proliferating cells (i.e., S + G2-M) is fre- embedded diploid tissue) do aid in the more exact charquently used to define cell cycle kinetics. This value is acterization of ploidy. As demonstrated here (Fig. 1) and easily obtained by subtracting % Gl/o from 100%. In from a large group of 220 tumors obtained from the Figure 3, one observes the distribution of percent Gl/o node-positive breast cancer tissues, there are some DNA activity for each of the 34 patients determined by two histograms that are not directly interpretable. One such procedures. The area fit gave a mean of 85.8% with a histogram is that which has a unimodal Gl/o peak that range of 62.8-98.7%while the curve fitted program gave does not coincide with the external diploid standard a mean of 74.3%with a range of 54.8-93.5% (P < .001). cells. We found 13%of this type in a study of 220 tumors. The pooled estimate of variance was higher for the curve It is our interpretation that these are diploid tumors fit compared to the area fit (34.0 vs. 7.0). For Gl/o, the that may have been fixed differently from the control variance increased as the mean decreased. As above, a tissue. This is quite possible since these PE blocks were signed rank test on the residuals obtained from the obtained from surrounding local hospitals. Preliminary linear model of percent Gl/o standard deviation on per- data have indicated that the relative fluorescence of the cent mean was used to compare variability adjust- stained nuclei (peak channel #) from the same tissue is ing for the mean. After adjustment, the difference in higher in glutaraldehyde-fixed tissue compared to forvariability between the two methods was of borderline malin-fixed tissue. At this time, we have no explanation for this artifact. A second histogram type that is difficult significance ( P = .033).
REPRODUCIBILITY IN PARAFFIN-EMBEDDEDBLOCKS
497
100’
90 -
F
80-
0 Y
K
w 70-
2 e a
c 600
al 5040 -
30 -20
-10
Mean
+10
-20
+20
-10
Mean
410
+20
Program Fit
Area Fit
Mean
* SD
FIG.3. Variability of %GI values. Tissues were analyzed by flow cytometry, and the %GI activity was determined by area fit and hy a computer program fit procedure (6).The means and standard deviations for each sample are presented. Total number of samples was 34.
to evaluate is one comprised of a population of tetraploid cells with low proliferative activity. The third hardest case includes those tumors that have a low percent of cancer cells extractable by the pepsin treatment. Our previous results from frozen tissue indicated that aneuploidy was related to percent cancer cells in the tumor (8). When the tissue sections were analyzed histologically and by flow cytometry, only 43% of the tumors were aneuploid when they contained fewer than 10% cancer cells. This percentage increased to 69% as the percent of cancer cells increased to > 50% (8). Seventy-two blocks from a large study were analyzed for flow cytometry and tumor content. We found that ten of 12 (83%)tumors containing 20% or fewer cancer cells were diploid, whereas only 28 of 60 (47%) tumors with >20% cancer cells were classified diploid. The overall mean and range of percent tumor were 36% and 0-85% for the diploid group versus 41% and 5-75% for the aneuploid group. There was one tumor of the group that did not have any cancer cells and was diploid with a n 7% S phase activity. This work indicates heterogeneity in solid tumors and demonstrates the need for pathological examination in order to avoid potential sampling artifacts. The range for %S or %G~,,Jis great for tumors obtained from different patients, but within individual patient samples, one is able to observe a good reproducible value. This value is dependent on the procedure of analysis. At the present time, little, if any, information indicates that one method is better than any other. Our conclusion
from this study is that curve-fitting programs present in the package of some flow cytometers tend to overestimate %S.It is, therefore, critical to use cell cycle values as the signal or descriptor of tumor kinetics and not as a n absolute number of cellular activity. Using the estimated variability, one can define a confidence interval for a n individual %S value, One can calculate a n approximate 95% confidence interval as the mean & 2 standard deviations. This has been done by Dressler et al., and their results demonstrate a range of 53.0% from the mean value (3). Our results indicate that it would be better to define the confidence interval (0based on the CV since the SD varies with the mean. This would be as follows: 95%CI = X -t 2 (X) CV, where CV is the estimate of the within-patient CV based on a set of replicated data, and X is the observed value. Our estimate of the CV for the rectangular S fit procedure is 0.21. Therefore, a n approximate 95% confidence limits for a n S value of 5% is 5 & 2.1% (i.e., 2*5*.21). The upper and lower interval would then be 2.9-7.1%. Our estimates of the CV for %S via area fit and program fit were .20 and .31 respectively. The estimates of the CV for proliferative index (i.e., total-Gl,o) for area fit and program fit were .19 and -23, respectively. With these numbers one can get a better appreciation for the confidence of these cell kinetic values. The potential of measuring dynamic cell cycle kinetics using bromodeoxyuridine (BrdUrd) uptake might overshadow these studies (7) because the interpretation of the data is more direct. However, there are only a few
498
KUTE ET AL.
reports of measuring “in vivo” cell kinetics in human solid tumors. These experiments have their drawbacks, such as the need for fresh tissue, the availability of BrdUrd incorporpation into solid tumors, and the need for dual fluorescent parameter staining techniques. Finally, one is not able to perform retrospective studies with this particular procedure. Schutte et al. have suggested that cellular kinetics via PE sections are not reproducible (9). Their study did not compare the reproducibility of consecutive sections from PE blocks but based the variability of cell kinetics on a comparison of results from a single analysis of fresh versus PE blocks. Their results are not in disagreement with the observation of this study. Other studies comparing the validity of cell cycle kinetics indicate that %S phase activity is more in line with histologic examination (i.e., Feulgen staining) when the tumor is tetraploid (4). Finally Bin0 et al. (2) incidated that 3H-thymidine uptake is the better procedure, but the %S activity via flow cytometry analysis can also be valid if the right computer program is used. There are many other papers that have used flow cytometry for ploidy and cell kinetic analysis, but no other reports have assessed variability in consecutive sections of PE blocks. Dressler et al. (3) have demonstrated a technique of using a coarse powder of breast tissue as the source for DNA analysis. This eliminates any microheterogeneity of the sample. Our study indicates that one can still get a relatively reproducible result without having to make a frozen powder out of the entire tissue. In addition, a powder-standardized tumor is not available if PE sections are to be used in the study. This report demonstrates the variability of ploidy and cell kinetics obtained on three to four consecutive sections of PE blocks. Since some of this information is being used in clinical decisions for cancer treatment, it is important to know the limitations of the data. This report does not address the issue of variability between different PE blocks from the same patient. It also does not discuss the heterogeneity of flow cytometry data
between primary and metastatic disease. These two issues need to be addressed in order to define further how flow cytometry information can be used with more confidence.
ACKNOWLEDGMENTS We would like t o thank Anne Wallen for her helpful suggestions and Nan Hartman for typing this manuscript. This work was done on the Oncology Research Center’s flow cytometry shared facility.
LITERATURE CITED 1. Baisch H, Beck HP, Christensen IJ, Hartmann NR, Fried J, Dean PN, Gray JW, Jett JH,Johnston DA, White RA, Nicolini C, Zeitz S, Watson Jv: A comparison of mathematical methods for the analysis of DNA histograms obtained by flow cytometry. Cell Tissue Kinet 15235-249, 1982. 2. Bin0 DG, Bruni C, Koch G , Mazzini G, Costa A, Silvestrini R Validation of a mathematical procedure for computed analysis of flow cytometric DNA data in human tumors. Cytometry 6:31-36, 1987. 3. Dressler LG, Seamer L, Owens M, Clark GM, McGuire WL: Evaluation of a modeling system for S-phase estimation in breast cancer by flow cytometry. Cancer Res 47:5294-5302,1987. 4. Fallenius AG, Askensten UG, Skoog LK, Auer GU: The reliability of microspectrophotometric and flow cytometric nuclear DNA measurements in adenocarcinomas of the breast. Cytometry 8260266,1987. 5. Fox M H A model for the computer analysis of synchronous DNA distributions obtained by flow cytometry. Cytometry 1:71-77, 1980. 6. Hedley DW,Friedlander ML, Taylor JW, Rugg CA, Musgrove E: Method for analysis of cellular DNA content of paraffin-embedded pathological material using flow cytometry. J Histochem Cytochem 31:1333-1335,1983. 7. Kute TE, Muss H: Flow cytometry in solid-tumor prognosis. Lab Management 2521-32,1987. 8. Kute TE, Muss HB, Hopkins M, Marshall R, Case D, Kammire L: Relationship of flow cytometry results to clinical and steroid receptor status in human breast cancer. Breast Cancer Res Treat 6:113121, 1985. 9. Schutte B, Reynders MMJ, Bosman FT,Bligham GH: Flow Cytometric determination of DNA ploidy level in nuclei isolated from parafin embedded tissue. Cytometry 6:26-30, 1985. 10. Thornthwaite Sr,Sugarbaker H V , Temple W J Preparation of tisue for DNA flow cytometric analysis. Cytometry 1229-237, 1980.
