Improved Flow Cytometric Determination of Proliferative
Activity (S-phase fraction) from Paraffin-Embedded Tissue
DONALD L. WEAVER, M.D., C. BRUCE BAGWELL, M.D., PH.D., SHELLY A. HITCHCOX, RT (CSLT),
SHERRY D. WHETSTONE, B.S., DAVID R. BAKER, B.S., DONALD J. HERBERT, HTL (ASCP),
AND MICHAEL A. JONES, M.D.
Recent studies suggest that proliferative activity (S-phase fraction |SPF|) may have greater prognostic significance than total
nuclear DNA content; however, relatively few studies have examined SPF from paraffin-embedded tissue because of significant
contamination of histograms with debris. In this study, cell cycle
analysis was performed on 124 matched tissue specimens. Fresh
tissue was divided into two equal portions; one portion was frozen,
whereas the other portion was processed and embedded in paraffin. S-phase could be determined for both frozen and paraffinembedded tissue in 81 cases. Correlation between SPF from frozen and paraffin-embedded tissue was demonstrated (r = 0.80)
when debris was subtracted from histograms with the use of two
new subtraction algorithms referred to as multicut and singlecut. Unlike other debris-subtraction algorithms, the quantity and
distribution of debris calculated by these algorithms are dependent on the magnitude and position of histogram peaks. A lesser
degree of correlation was demonstrated with the use of a standard
exponential debris subtraction algorithm (r = 0.67). Correlation
of SPF for aneuploid cases was greater when SPF was calculated
as a percentage of the aneuploid cell population rather than as
a percentage of the entire cell population. This was attributed
to the observation that the proportion of aneuploid cells from
paraffin-embedded tissue was less than that from frozen tissue.
The results of this study indicate that SPF can be calculated
from paraffin-embedded tissue with values comparable to those
obtained from frozen tissue. The ability to calculate SPF reliably
from paraffin-embedded tissue should allow additional evaluation
of this parameter as a prognostic indicator. (Key words: Flow
cytometry; DNA; Human neoplasms; Cell cycle analysis; Computer modeling) Am J Clin Pathol 1990;94:576-584
SEVERAL RECENT ARTICLES have reviewed the
prognostic significance of cytometrically determined DNA
content in a variety of neoplasms, 101418 and some studies
have found S-phase fraction (SPF) to be a more important
prognostic indicator than DNA content.5-6-819 The use of
paraffin-embedded tissue allows DNA analysis to be performed on cases in which fresh tissue cannot be spared
Received December 4, 1989; received revised manuscript and accepted
for publication March 12, 1990.
Supported in part by the Maine Medical Center Research Fund, the
Maine Medical Center Annual Fund, the Davenport Foundation of
Maine, and the Maine Cancer Research and Education Foundation.
Dr. Weaver is a Maine Medical Center Annual Fund Fellow in Cytometry.
Dr. Bagwell, who developed the software used for data analysis in this
study, works part time for Verity Software House, Topsham, Maine,
which markets that software.
Address reprint requests to Dr. Weaver: University of Vermont Department of Pathology, Burlington. VT 05405-0068.
Maine Cytometry Research Institute and Maine Medical
Center, Portland, Maine, and Department of Pathology,
The University of Vermont College of Medicine,
Burlington, Vermont
for flow cytometry (FCM) or for retrospective prognostic
studies. Despite the large number of studies using paraffinembedded tissue, only a small percentage have attempted
to examine SPF,10 presumably because of lack of confidence in S-phase estimates resulting from contamination
with debris in the S-phase region. In fresh tissue, SPFs are
comparable to thymidine labeling indices when debris is
subtracted from FCM histograms with the use of algorithms that model debris as an exponential.916 The validity
of calculating SPFs from paraffin-embedded tissue must
also be verified if clinical application is intended; however,
relatively few studies have compared SPFs from fresh and
paraffin-embedded tissue. 12131517 The current study is
unique in that it compares SPFs from frozen and paraffinembedded tissue derived from a single block of fresh tissue,
the purpose being to minimize the effect of intratumor
variation in both DNA index and SPF, which has been
observed by other investigators.113 Different combinations
of mathematical models for SPF and debris were used
and their effect on the correlation for paired samples was
examined.
Materials and Methods
Specimen Acquisition
One hundred twenty-four tissue specimens were obtained from either the Maine Medical Center or the Medical Center Hospital of Vermont from October 1988
through April 1989. Samples were selected from routine
surgical specimens and were restricted to cases that were
received in a fresh, unfixed state and that were of sufficient
quantity that tissue in excess of that needed for diagnosis
was available. Approximately 1 cm 3 of tissue was minced
into 3-4-mm cubes and randomly divided into two equal
portions. One portion was fixed in 10% (volume/volume
[v/v]) neutral buffered formalin (NBF), processed, and
576
Vol. 94 • No. 5
S-PHASE FROM PARAFFIN-EMBEDDED TISSUE
embedded in paraffin in the usual manner (see below),
whereas the other portion was frozen at - 7 0 °C in a 5mL aliquot of buffered DMSO/sucrose.21 In cases in which
a delay occurred between tissue acquisition and mincing,
the specimen was held in RPMI 1640 (a cell culture media)
tissue culture medium (Sigma Chemicals, St. Louis, MO).
In no case was the delay longer than 30 minutes.
Tissue Processing
Maine Medical Center specimens (n = 103) were fixed
in 10% (v/v) NBF (one part Anatech CB® Formalin Concentrate #111, four parts distilled water) and processed
automatically (Innovated Medical Systems LX-120®)
according to the following programmed schedule: 10%
NBF X 2; 80% (v/v) ethanol; 95% (v/v) ethanol X 2; 100%
ethanol X 3; and xylene X 2—each for 60 minutes at 34
°C—then Paraplast X-TRA® (Curtin-Matheson, Wilmington, MA) X 3 for 45 minutes each at 58 °C. The last
two Paraplast steps were at 0.5 atmospheric pressure. Tissues were embedded in Paraplast X-TRA.
Medical Center Hospital of Vermont specimens (n
= 21) were fixed in 10% NBF (130 g sodium phosphate
dibasic, 80 g sodium phosphate monobasic, 18 L water,
2 L formaldehyde, pH adjusted to 7.1) and processed automatically (Ames/Miles Scientific VIP®) at 130 mmHg
vacuum according to the following programmed schedule:
alcoholic formalin (ten parts formaldehyde, nine parts
95% [v/v] ethanol) for 5 minutes; alcoholic formalin for
90 minutes; 70% (v/v) ethanol for 20 minutes; 95% (v/v)
ethanol for 20 minutes; 95% (v/v) ethanol for 45 minutes;
100% ethanol for 45 minutes X 3; toluene for 60 minutes;
toluene for 90 minutes—each at room temperature—then
Paraplast X-TRA for 30 minutes X 2, 60 minutes, then
30 minutes, each at 56 °C. Tissues were embedded in
Paraplast X-TRA.
Paraffin DNA Method
Paraffin-embedded tissue was deparaffinized and dissociated according to a modification of a method reported
by Hedley and associates.'' Three to four 50-/im sections
of the paraffin block were cut on a standard microtome
and deparaffinized with two 3-mL washes of Americlear®
(Stephens Scientific Division, Oak Ridge, NJ). Tissue sections were rehydrated by sequential 10-minute washes in
3 mL of 100%, 95%, 70%, and 50% (v/v) ethanol, then
washed in distilled water and held overnight in 3 mL of
distilled water at RT. The tissue was dissociated the following morning in 1 mL of a 1% (w/v) pepsin solution
(1,100 units/mg protein, Sigma) at 37 °C for 30 minutes,
vortexing every 5 minutes, then centrifuged at 1,000 X g
for 5 minutes, and the cell pellet resuspended in 1-3 mL
Earles Balanced Salt Solution (Sigma). The sample was
577
filtered through a 100-Mm nylon mesh (Small Parts, Inc.,
Miami, FL) and recentrifuged at 1,000 X g for 5 minutes.
The cell button was resuspended in 1 mL propidium iodide (PI) staining solution and processed as described below under "Staining Methods."
Frozen DNA Method
Samples were stored at —70 °C and thawed just before
processing by swirling in a 37 °C water bath until ice
crystals just disappeared. The tissue was removed from
the freezing solution (250 mmol/L sucrose, 40 mmol/L
trisodium citrate, 5% [v/v] dimethylsulfoxide [DMSO])
and placed in a Petri dish with 5 mL supplemented RPMI
(RPMI 1640 with 2 mmol/L L-glutamine, 5.96 g/L
HEPES buffer, and 50 mg/mL gentamicin sulfate). Each
sample was then teased into smaller pieces with the use
of toothed forceps, poured through a 425-/*m stainless
steel sieve (American Scientific Products, McGaw Park,
IL), and remaining tissue gently forced through the sieve
with a 10-mL syringe plunger to obtain fine aggregates
and single cells. Additional supplemented RPMI was used
to rinse the sieve and bring the total cell suspension volume to 14 mL. Cells were centrifuged at 400 X g for 5
minutes in a 15-mL conical polystyrene tube. The resulting pellet was washed with 14 mL supplemented
RPMI, centrifuged at 400 X g for 5 minutes, and resuspended in 1 mL of phosphate-buffered saline with 1 %
(v/v) fetal calf serum (PBS/FCS). The cell concentration
was determined with a hemocytometer and the final concentration adjusted to 1.0 X 107 cells per milliliter with
PBS/FCS. A 200-/uL aliquot of each specimen was added
to a polystyrene tube containing 1.0 mL of PI staining
solution and processed as described below.
Staining Methods
Cell suspensions were stained with PI (Sigma) according
to the method described by Bauer.4 Cell suspensions from
frozen tissue or cell pellets from paraffin-embedded tissue
were added to 1.0 mL of low salt stain (3.0 g polyethylene
glycol [PEG 8000®, Fisher Scientific, Pittsburgh, PA], 5
mL of 1 mg/mL PI, 5 mL of 3,600 units/mL RNAse A
in PBS, 1 mL of 10% [v/v] Triton X-100/PBS, 89 mL of
4 mmol/L sodium citrate buffer) and gently vortexed.
Samples were incubated for 20 minutes at 37 °C, then
1.0 mL of high salt stain (3.0 g PEG, 5 mL of 1 mg/mL
PI, 1 mL of 10% [v/v] Triton X-100®/PBS, 94 mL 400
mmol/L sodium chloride) was added to each tube and
gently vortexed. Each specimen was stored at 4 °C overnight, filtered through 60-nm nylon mesh (Small Parts)
into a new polystyrene tube, then analyzed by FCM for
DNA content. Staining solutions were prepared in advance and stored at —70 °C in 3-mL aliquots until just
before use.
578
WEAVER
Flow Cytometry
All samples were run on an EPICS-C® flow cytometer
(Coulter Corporation, Hialeah, FL) at 400 mW of laser
power with a 2-W argon ion laser. A 550-nm dichroic
short pass mirror and 570-nm long pass mirror were used
in front of the red photomultiplier tube (PMT). The signal
from the red PMT was split and fed into two amplifiers.
One fluorescence amplifier was set at a gain of 5, and the
PMT high voltage was adjusted to bring the diploid peak
to approximately channel 80 in a 256-channel histogram.
The other fluorescence amplifier was set to a gain of 2 to
yield a "down-scale" version of the DNA histogram, which
was used for visualizing triplets and tetraploid tumors.
The following parameters were collected in list mode
on all samples: log forward angle light scatter, log side
scatter, red fluorescence, and down-scale red fluorescence.
The 256-channel red fluorescence histograms were selfgated above channel 10. All samples were briefly vortexed
just before running, and data were acquired at a flow rate
of 100 events per second. A maximum of 20,000 events
was acquired for each sample.
Computer Modeling
All histograms were analyzed with the use of Verity:
ModFit® (Verity Software House, Inc., Topsham, ME),
a cell cycle analysis program. The results were entered
into a database and analyzed with the use of Excel® (Microsoft Corporation, Redmond, WA). S-phase was modeled as a single broadened rectangle2 and expressed as a
percentage of its respective GOGl, S, and G2M fitted
model components. The standard deviation of the G 2 M
Gaussian peak was always assumed to be twice the standard deviation of the respective GOG 1 Gaussian peak. If
there was not a distinct G 2 M peak, its position was assumed to be twice the GOGl peak position. No attempt
was made to model the diploid S-phase for DNA aneuploid histograms and no attempt was made to model diploid G 2 M for tetraploid tumors. Debris from the paraffinembedded and frozen samples was modeled with the use
of "single-cut" and "multicut" algorithms,3 respectively
(see below and "Discussion"), as well as an exponential
algorithm.9 Values for all other model components were
allowed to be determined by the Marquardt nonlinear
least-squares procedure.7
Debris Subtraction Algorithms
The single-cut and multicut debris subtraction
algorithms3 assume that debris is generated by cut nuclei
and the quantity and distribution of debris are dependent
on the magnitude and position of histogram peaks. Nuclei
are stabilized in paraffin and some of the nuclei are cut a
single time by the microtome blade. The probability dis-
ET AL.
A.J.C.P. • November 1990
tribution of the two resulting nuclear fragments can be
expressed in terms of channel numbers:
PS(J, x) =
4
,
=
7rjVl-((2x/j)-l)2
where
Ps(j> x) = the probability of a cut nuclear fragment from
channel j falling into channel x.
The single-cut debris model component is expressed as
follows:
n
Debris s (x)sk
2
j(,/3)Y0bs(j,Ps(J, x)
j = x+l
where,
Debriss (x) =
k =
n=
j, x =
Yobs(j) =
P s (j, x) =
number of debris particles in channel x
amplitude constant
number of channels
channel numbers
observed events in channel j
single cut nucleus probability function
A proportion of fresh and frozen nuclei are shattered by
sample preparation techniques that are identical to randomly cutting a nucleus into multiple pieces. The probability distribution of the multiple nuclear fragments can
also be expressed in terms of channel numbers:
P m (j, x) = ke~kx
where,
P m (j, x) = the probability of a nuclear fragment from
channel j falling into channel x
k = exponential rate constant
The multiple-cut debris model component is expressed
as follows:
n
Debris m (x) ^ ae~ kx
£
Y obs(j)
j = x+l
where,
DebriSm (x) =
a =
k=
n=
j, x =
Y0bs(j) =
number of debris particles in channel x
amplitude constant
exponential rate constant
number of channels
channel numbers
observed events in channel j
Exclusion Criteria
Cases were excluded from the study for the following
reasons: (1) total analyzed events in the histogram less
than 2,900, (2) GOGl peak coefficient of variation (CV)
greater than 10.5%, (3) ploidy mismatch (e.g., aneuploid
Vol. 94 • No. 5
S-PHASE FROM PARAFFIN-EMBEDDED TISSUE
tumor in frozen tissue with unresolvable near-diploid or
"shouldered" peak in paraffin-embedded tissue), (4) multiploid tumor, (5) nuclear aggregates in S-phase, (6) greater
than 30% debris in frozen histogram, (7) aneuploid population in paraffin histogram less than 15%.
Results
Four cases were excluded before FCM analysis; paraffin
blocks could not be located for two specimens, and the
tissue had not been minced in two cases. Of the 120 cases
analyzed by FCM, 39 were excluded for reasons outlined
above and in Table 1. Cell-cycle analysis and S-phase
comparison between frozen and paraffin-embedded tissue
was possible in the remaining 81 cases. The cases represented a heterogeneous collection of normal tissues, benign neoplasms, and malignant neoplasms (Table 2). The
mean, range, and standard deviation of various parameters
are presented in Table 3. Routine hematoxylin and eosin
sections were cut from all paraffin blocks adjacent to the
section used for FCM and the slides reviewed by one of
us (D.L.W.) for confirmation of the tissue diagnosis and
to assure, in cases with a neoplastic diagnosis, that tumor
cells were well represented. Fifty-nine specimens were
diploid and 22 specimens were aneuploid. DNA index
showed excellent correlation between paired frozen and
paraffin-embedded tissue specimens (r = 0.96). All aneuploid cases had a malignant histologic diagnosis.
Typical histograms for frozen and paraffin-embedded
tissue are presented in Figure 1. Paraffin-embedded tissue
typically contained more debris (Fig. 1D) than frozen tissue (Fig. \A). In most cases the debris from paraffinembedded tissue formed a plateau that merged with the
G0G1 peak (Fig. ID). Debris in channels immediately
below GOG 1 was greater than the sum of debris and SPF
immediately above GOG 1, creating a "step" as the GOG 1
peak was traversed. Very low fluorescent debris occasionally approximated an exponential function (data not
shown); however, even in these cases the plateau and step
prevailed nearer the G0G1 peak. When the SPFs from
frozen and paraffin-embedded tissue were compared
Table 1. Cases Excluded from Study
Exclusion Criteria
Number of Cases
Less than 2,900 events
Ploidy mismatch
Technical (no staining)
G0G1 C V > 10.5%
Multiploid
Debris > 30% in frozen
Aneuploid < 15% in paraffin
Aggregates in S-phase
9
7
6
5
4
3
3
2
Total
39
579
Table 2. Cases Included in Study
Tissue
Total No.
Malignant*
Breast
Colon
Kidney
Lung
Lymph nodes and
metastases
Ovary
Placenta
Uterus
Othersf
13
12
4
9
7
11
4
9
—
—
—
—
—
11
9
5
10
8
9
6
2
3
—
—
—
—
2
3
5
1
1
Total
81
57
14
10
7
4
Benign*
3
1
Normal*
3
* Classification determined by histologic review.
t Pancreas, parotid, small bowel, testis, thyroid, ureter.
without subtraction of the superimposed debris (Fig. 2),
only moderate correlation was demonstrated (r = 0.67).
The SPF correlation did not improve (r = 0.67) when
debris was subtracted with the use of an exponential algorithm (Fig. 3). Over subtraction of debris in the S-phase
region produced a large number (n = 25) of cases in which
the calculated SPF from paraffin-embedded tissue was
zero. Significant improvement in SPF correlation (r
= 0.80) was achieved when debris was subtracted with the
use of two new debris subtraction algorithms, referred to
as single-cut and multicut 3 (Fig. 4).
Multiple combinations of different debris and S-phase
models were used to analyze the data in an attempt to
discover a combination with improved correlation (Table
4). The multicut, single-cut, trapezoid S-phase combination gave the best correlation coefficient (r = 0.81);
however, the trapezoid S-phase model caused unpredictable results in two cases that when corrected produced a
correlation coefficient identical to the rectangle model (r
= 0.80). Quite surprising was the polynomial S-phase
model in which calculated SPFs were frequently negative
and the overall correlation was unsatisfactory (r = -0.33).
Calculated SPFs were uniformly higher from paraffinembedded tissue as evidenced by the regression curve
slopes that were significantly less than 1.0 in all the combinations tested (Table 4). The slopes varied somewhat
from tissue to tissue in the current study; however, the
slope for colonic tissue was substantially lower (y = 0.38x
+ 0.9; r = 0.77; n = 12) than that of all other tissues
combined (y = 0.73x - 0.9; r = 0.80; n = 69).
Several important points were noted in the aneuploid
cases. SPFs in aneuploid cases were calculated as a percentage of the aneuploid population. The correlation was
poorer when SPF was calculated as a percentage of the
entire population (diploid and aneuploid) (Table 5). Paraffin-embedded tissue preparations produced smaller
percent aneuploid populations than the corresponding
frozen tissue sample (Fig. 5). The greater the difference
580
A.J.C.P. • November 1990
WEAVER ET AL.
Table 3. Statistics for Selected Parameters
Paraffin-Embedded Tissue
Frozen Tissue
Diploid cases
(n = 59)
Events
Debris (%)
CV diploid (%)
S-phase (%)
Aneuploid cases
(n = 22)
Events
Debris (%)
CV diploid (%)
CV aneuploid (%)
S-phase (%)
Mean ± SD
(Range)
Mean ± SD
(Range)
14,141 ±4,298
4.6 ± 4.2
4.1 ± 1.4
5.9 ± 3.4
(2,922-19,214)
(0.3-25.4)
(2.1-7.3)
(1.0-18.0)
14,651 ±4,433
26.5 ± 12.0
5.7 ± 1.7
10.2 ±5.1
(3,440-19,521)
(10.3-53.4)
(3.1-9.9)
(2.5-26.9)
15,470 ±3,683
7.8 ± 4.2
4.2 ± 0.8
4.8 ± 1.3
16.9 ±7.4
(6,464-18,909)
(2.2-21.2)
(3.2-6.5)
(2.9-9.1)
(3.4-30.0)
16,620 ±3,171
32.8 ± 11.6
5.1 ±0.9
6.7 ±2.1
23.4 ± 9.4
(9,034-19,349)
(17.4-54.8)
(3.7-6.8)
(4.5-10.6)
(6.8-39.6)
Debris was modeled using the multicut algorithm for frozen tissue and the single-cut algorithm for paraffin-embedded tissue; S-phase was modeled as a single rectangle.
in percentage aneuploid cells between the paired samples,
the more likely their SPFs showed poor correlation (data
not shown).
Discussion
S-phase fraction has been found to be an important
prognostic indicator; however, SPF may be difficult to
calculate from FCM histograms because of debris superimposed on the S-phase region. It has been shown that
SPFs from fresh tissue correlate better with thymidine
labeling indices when debris is subtracted with the use of
an exponential algorithm.16 By inference, exponential debris subtraction has been applied to paraffin-embedded
tissue; however, in the current study the exponential debris
FIG. 1. Representative paired DNA histograms for frozen and paraffin-embedded tissue. A, B, and C. Frozen tissue histograms. A. Frozen tissue
histogram with low background debrisfluorescence.B. Histogram with superimposed DNA model that subtracts debris using an exponential algorithm.
Debris gradually decreases from low channels to high channels as an exponential function with the rate of decrease determined by the shape of the
curve in channels below GOGl. C. Histogram with debris modeled and subtracted using a multicut algorithm. Calculated debris is dependent upon
peak position and magnitude, which results in less debris subtracted from the S-phase fraction. D, E, and F. Paraffin-embedded tissue histograms.
D. Paraffin-embedded tissue histogram with typical lowfluorescencedebris forming a plateau below GOGl. E. Histogram with debris modeled and
subtracted using an exponential algorithm. Note that debris is under subtracted below GOGl and over subtracted in the S-phase region above GOGl,
resulting in cases with erroneous calculated S-phase fractions of zero. F. Histogram modeled and subtracted using a single-cut algorithm. Debris is
assumed to be the result of nuclei that have been randomly cut a single time; the debris curve is dependent on peak position and magnitude. Debris
decreases sharply as the GOG 1 peak is traversed, eliminating the over subtraction in the S-phase region seen with the exponential algorithm. (Cross
hatching = debris; vertical lines = S-phase; diagonal lines = GOG 1 and G2M; y-axis on frozen tissue histograms is magnified 1.4 times to demonstrate
debris.)
S-PHASE FROM PARAFFIN-EMBEDDED TISSUE
Vol. 94 • No. 5
581
40 -
r=0.80
35 -
n=81
%S-phase
US-phase
(Irozen)
(frozen)
30 •
•
25 •
-
20 •
•
•
•
D
&a
D
D
sf
15 •
D
a
10 -
5 -
—fi
0 15
20
25
30
0
35
FIG. 2. Comparison of S-phase fraction from paired frozen and paraffinembedded tissue specimens without debris subtraction (closed squares
= diploid cases; open squares = aneuploid cases).
algorithm showed no improvement in SPF correlation
between frozen and paraffin-embedded tissue compared
with the correlation without debris subtraction. The exponential algorithm did not accommodate the debris plateau or the sharp decrease in debris as the G0G1 peak is
traversed (Fig. IE). This resulted in under subtraction of
debris below GOG 1 and over subtraction of debris above
G0G1 in the S-phase region. The over subtraction produced a large number (n = 25) of cases in which the cal-
r=0.67
y=0.61x+2.38
n=81
%S-phase
(frozen)
•
SEA
0
5
10
10
15
1
20
H
25
H
30
1
1
35
40
%S-phase (paraffin embedded)
%S-phase (paraffin embedded)
35 - -
5
-+-
•+-
•+-
•+-
15
20
25
30
35
40
%S-phase (paraffin embedded)
FIG. 3. Comparison of S-phase fraction from paired frozen and paraffinembedded tissue specimens with debris subtracted using an exponential
debris algorithm. Note the large number of cases (n = 25) on the y-axis
where the calculated S-phase fraction was zero for paraffin-embedded
tissue due to debris over subtraction (closed squares = diploid cases;
open squares = aneuploid cases).
FIG. 4. Comparison of S-phase fraction from paired frozen and paraffinembedded tissue specimens with debris subtracted using a multicut nucleus debris algorithm for frozen tissue and a single-cut nucleus debris
algorithm for paraffin-embedded tissue (closed squares = dilpoid cases;
open squares = aneuploid cases).
culated SPF from paraffin-embedded tissue was zero,
which adversely affected the SPF correlation.
Two new debris subtraction algorithms, referred to as
single-cut and multicut, have been developed.3 Significant
improvement in SPF correlation between frozen and paraffin-embedded tissue was demonstrated when these algorithms were used. Both algorithms assume that debris
is created by cut nuclei and that the quantity and distribution of debris are dependent on the magnitude and
position of histogram peaks. The single-cut debris subtraction algorithm for paraffin-embedded tissue assumes
that the major component of debris is partial nuclei that
are created when some of the whole nuclei are transected
by the microtome blade. The single-cut algorithm has a
debris curve that better approximates the plateau observed
in histograms from paraffin-embedded tissue (Fig. \F).
The multicut algorithm calculates debris curves as if nuclei
were randomly cut multiple times, approximating the
shattering of some nuclei that occurs in nuclear suspensions from fresh and frozen tissue. The resulting debris
curve is similar in shape to an exponential curve (Fig. 1C)
but is dependent on peak positions and magnitudes as is
the single-cut model.
Calculated SPFs were generally higher from paraffinembedded tissue than from frozen tissue, despite that the
SPF correlation between paired samples was high when
debris was subtracted with the use of the single-cut and
multicut algorithms. This was attributed to incomplete
debris subtraction in the S-phase compartment. There are
two possible explanations. Small-particle, low fluorescent
debris is left in the supernatant during centrifugation and
never appears in the FCM histogram; loss of this low flu-
582
A.J.C.P. • November 1990
WEAVER ET AL.
Table 4. Correlation and Regression Curves for Paired Frozen and Paraffin-Embedded Tissue Specimens Using
Various Combinations of Debris and S-Phase Models
Frozen Debris
Paraffin Debris
S-Phase
Correlation
Coefficient
Slope
Intercept
None
Exponential
Exponential
Multicut
Multicut
Multicut
Multicut
Multicut
None
Exponential
Single-cut
Single-cut
Single and multicut
Single-cut and exponential
Single-cut
Single-cut
Rectangle
Rectangle
Rectangle
Rectangle
Rectangle
Rectangle
Trapezoid
Polynomial
0.67
0.67
0.59
0.80*
0.78f
0.64
0.81J
-0.33
0.54
0.61
0.42
0.64
0.54
0.57
0.68
-0.06
0.82
2.38
-0.06
0.06
1.03
4.27
-0.38
13.35
P values are compared with no debris subtraction.
* P = 0.07.
t /• = 0.14.
%P< 0.05.
orescent debris decreases baseline debris (fluorescence below GOG 1). Because baseline debris is used by subtraction
algorithms to extrapolate debris in higher channels, less
debris would be subtracted and increased SPFs would be
calculated. The other explanation is that small partial nuclei (lowfluorescentdebris) adhere to whole cells, shifting
a percentage of GOG 1 cells into the S-phase compartment
while decreasing lowfluorescentdebris used as a baseline
in debris subtraction algorithms. This apparent decrease
in low fluorescent debris and the artificial increase in
number of cells in the S-phase compartment creates two
additive errors. The effect is noticed more in paraffin histograms than in frozen histograms because debris levels
are inherently higher in paraffin-embedded tissue preparations. Preliminary studies in our laboratory suggest a
combination of both hypotheses contribute to the higher
paraffin SPF; however, the latter appears to be the major
contributor.
Other investigators have attempted to solve the debris
problem by gating out debris using forward versus side
scatter or setting the discriminator very close to the GOG 1
peak on the red (DNA) fluorescence histogram. Delib-
Table 5. Comparison of S-Phase Fraction (SPF)
Correlation for Aneuploid Cases from Frozen and
Paraffin-Embedded Tissue
SPF Aneuploid*
Aneuploid cases
(n = 22)
Diploid cases
(n = 59)
All cases
(n = 81)
erately eliminating debris below the GOGl peak alters
baseline debris, which is essential for proper application
of debris subtraction routines. As the discriminator is
progressively moved closer to the GOGl peak, the calculated SPF systematically increases (unpublished data).
Gating out debris is not recommended if subtraction routines will subsequently be used on the histograms.
The trapezoid S-phase model produced an initially
higher coefficient of correlation with these data. A significant portion of G2M was inappropriately included in the
SPF in two cases with the use of the trapezoid model.
When this was corrected, the trapezoid and rectangle Sphase models produced equivalent correlations. It is not
clear which model, if either, is superior based on this study.
The disadvantage of the trapezoid model is that calculated
SPF can increase if either the GOG 1 or G2M peak is
%Aneuploid
(frozen)
SPF Total*
r
Slope
Intercept
r
Slope
Intercept
0.68
0.53
1.18
0.56
0.65
1.05
0.54
0.42
1.19
0.54
0.42
1.19
0.80
0.64
0.06
0.53
0.53
1.78
r « correlation coefficient.
* SPF is calculated as a percentage of the aneuploid population (SPF aneuploid) or as a percentage
of the combined aneuploid and diploid population (SPF total).
0+
0
1
1
r-
10
20
30
40
50
60
70
80
90
100
%Aneuploid (paraffin embedded)
FIG. 5. Comparison of percent aneuploid populations from paired
frozen and paraffin-embedded tissue specimens. Paraffin-embedded tissue
preparations contained smaller populations of aneuploid cells than frozen
tissue preparations; note that the majority of cases lie above the diagonal
perfect correlation line.
Vol. 94 • No. 5
S-PHASE FROM PARAFFIN-EMBEDDED TISSUE
skewed toward the S-phase compartment or an aggregate
peak is present at either end of the S-phase compartment.
The rectangle S-phase model is minimally affected by these
variations in histogram shape and calculates a more conservative value that may be better suited to clinical applications.
The tendency toward lower proportions of aneuploid
cells in paraffin-embedded tissue preparations relative to
paired frozen tissue preparations has been noted by others.14,20 This tendency significantly affected the SPF correlation between frozen and paraffin-embedded tissue
when total S-phase was used for aneuploid cases (Table
5). Aneuploid S-phase showed better correlation, but the
correlation was also affected by large differences in aneuploid populations between paired specimens. It is not clear
why paraffin-embedded tissue preparations have a lower
proportion of aneuploid cells. Malignant nuclei are large
and have complex nuclear membranes, features that may
make the nuclei fragile and more susceptible to enzymatic
digestion. This would lead to selective loss of aneuploid
cells. On the other hand, there may be selective enhancement of aneuploid cells in frozen tissue preparations. Fresh
and frozen tissue protocols use mechanical dissociation
(forceps and wire mesh), which tends to exclude connective tissue stroma (benign, diploid nuclei) from the nuclear
preparations. An important point here is that reliable SPFs
cannot be calculated when the aneuploid population is a
small percentage of the total cells analyzed.
Malignant diploid nuclei undoubtedly are affected by
fixation and paraffin embedding in the same manner as
aneuploid nuclei. Unfortunately, there is no reliable way
to separate benign from malignant nuclei in diploid neoplasms. SPF must be calculated as a percentage of the
entire cell population. This probably contributed to the
poorer correlation for diploid cases than for aneuploid
cases (Table 5).
A variety of tissues, including normal tissue, benign
neoplasms, and malignant neoplasms, was chosen for
analysis to generate a heterogeneous database. The overall
exclusion rate was high (39 of 120) because of the multiple
criteria needed to compare SPF in the paired histograms.
The method of mincing tissues may have resulted in
higher-quality paraffin-embedded histograms than are
generally achieved with routine paraffin-embedded tissue
blocks as a result of more rapid and thorough formalin
fixation of the small pieces of tissue. The histograms in
the current study were not perceived to be significantly
different from routine paraffin-embedded material analyzed in our laboratory. The exclusion rate for poor-quality histograms (11.6%) was similar to that of other studies
performed on paraffin-embedded tissue.
Cases with fewer than 15% aneuploid cells in the paraffin-embedded tissue preparation were excluded. The low
583
signal to noise ratio in this situation makes SPF difficult
to distinguish from debris. Small variations in calculated
S-phase events have dramatic effects on percentage Sphase when dealing with such small cell populations.
Frozen tissue histograms were used as the reference
histogram in the current study. For this reason, frozen
tissue histograms with greater than 30% debris were excluded to minimize inaccurate SPFs, which might bias
the comparison. Cases with peaks (aggregates) in the Sphase region that could not be modeled and subtracted
from the histogram were also excluded to eliminate bias.
Conclusions
The validity of calculating SPFs from paraffin-embedded tissue is supported by this study. Improved procedural
techniques may result in decreased debris, but, until these
techniques are available, mathematical subtraction offers
the best alternative. It is recommended that debris not be
gated out of the histogram if mathematical subtraction is
to be used. The single-cut debris algorithm appeared to
be better than exponential subtraction because it virtually
eliminated SPFs of zero. Additional support for the algorithms was evidenced by an improved correlation coefficient. SPFs showed better correlation between frozen
and paraffin-embedded tissue when SPF was calculated
as a percentage of the aneuploid population. Both aneuploid SPF and total SPF should probably be evaluated in
survival studies to determine which is clinically more relevant. No separation of tumor and nontumor populations
is possible in diploid specimens, and it may be necessary
to establish different high and low S-phase cut-off values
for diploid and aneuploid tumors in studies examining
the prognostic significance of S-phase.
Acknowledgments. The authors thank Dr. Edmund J. Lovett, III and
Dr. Kenneth A. Ault for their editorial comments.
References
1. Askensten UG, von Rosen AK, Nilsson RS, Auer GU. Intratumoral
variations in DNA distribution patterns in mammary adenocarcinomas. Cytometry 1989;10:326-333.
2. Bagwell CB. Theory and application of DNA histogram analysis.
University of Miami Doctoral Thesis, ERIC/EDRS ED 299 130,
1979.
3. Bagwell CB, Mayo SW, Whetstone SD, et al. DNA histogram debris
theory and compensation. Cytometry (in press).
4. Bauer KD. Clinical flow cytometry course. Part I: Applications and
methods. Hershey Medical Center, Hershey, PA, October 9-14,
1988.
5. Bauer KD, Lincoln ST, Vera-Roman JM, et al. Prognostic implications of proliferative activity and DNA aneuploidy in colonic
adenocarcinomas. Lab Invest 1987;57:329-335.
6. Bauer KD, Merkel DE, Winter JN, et al. Prognostic implications of
ploidy and proliferative activity in diffuse large cell lymphomas.
Cancer Res 1986;46:3173-3178.
584
WEAVER ET AL.
7. Bevington PR. Least squares fit to an arbitrary function. In: Data
reduction and error analysis for the physical sciences. New York:
McGraw-Hill, 1969:235-243.
8. Clark GM, Dressier LG, Owens MA, Pounds G, Oldaker T, McGuire
WL. Prediction of relapse or survival in patients with node-negative breast cancer by DNA flow cytometry. N Engl J Med
1989;320:627-633.
9. Haag D, Feichter G, Goerttler K, Kaufmann M. Influence of systematic errors on the evaluation of the S-phase portions from
DNA distributions of solid tumors as shown for 328 breast carcinomas. Cytometry 1987;8:377-385.
10. Hedley DW. Flow cytometry using paraffin-embedded tissue: five
years on. Cytometry 1989;10:229-241.
11. Hedley DW, Friedlander MI, Taylor IW, Rugg CA, Musgrove EA.
Method for analysis of cellular DNA content of paraffin-embedded
pathological material using flow cytometry. J Histochem Cytochem 1983;31:1333-1335.
12. Jacobsen AB, Thorud E, Fossa SD, et al. DNA flow cytometry in
metastases and a recurrency of malignant melanomas: a comparison of results from fresh and paraffin embedded material.
Virchows Arch [B] 1988;54:273-277.
13. Kallioniemi O-P. Comparison of fresh and paraffin-embedded tissue
as starting material for DNA flow cytometry and evaluation of
intratumor heterogeneity. Cytometry 1988;9:164-169.
A.J.C.P. • November 1990
14. Koss LG, Czerniak B, Herz F, Wersto RP. Flow cytometric measurements of DNA and other cell components in human tumors:
a critical appraisal. Hum Pathol 1989;20:528-548.
15. Mclntire TL, Goldey SH, Benson NA, Braylan RC. Flow cytometric
analysis of DNA in cells obtained from deparaffinized formalinfixed lymphoid tissues. Cytometry 1987;8:474-478.
16. Meyer JS, Coplin MD. Thymidine labeling index, flow cytometric
S-phase measurement, and DNA index in human tumors. Am J
Clin Pathol 1988;89:586-595.
17. Schutte B, Reynders MMJ, Bosman FT, Blijham GH. Flow cytometric determination of DNA ploidy level in nuclei isolated from
paraffin-embedded tissue. Cytometry 1985;6:26-30.
18. Seckinger D, Sugarbaker E, Frankfurt O. DNA content in human
cancer: application in pathology and clinical medicine. Arch Pathol Lab Med 1989; 113:619-626.
19. Sledge GW, Eble JN, Roth BJ, Wuhrman BP, Fineberg N, Einhorn
LH. Relation of proliferative activity to survival in patients with
advanced germ cell cancer. Cancer Res 1988;48:3864-3868.
20. Stephenson RA, Gay H, Fair WR, Malamed MR. Effect of section
thickness on quality of flow cytometric DNA content determinations in paraffin-embedded tissues. Cytometry 1986;7:41-44.
21. Vindelov LL, Christensen IJ, Nissen NI. A detergent-trypsin method
for the preparation of nuclei for flow cytometric DNA analysis.
Cytometry 1983;3:323-327.