VIRTUAL FLOW CYTOMETRY OF IMMUNOSTAINED LYMPHOCYTES ON MICROSCOPIC TISSUE SLIDES: Hernani D. Cualing MD, Lynn Moscinski MD*.

1

• VIRTUAL FLOW CYTOMETRY OF IMMUNOSTAINED
  LYMPHOCYTES ON MICROSCOPIC TISSUE SLIDES Hernani
  D. Cualing MD, Lynn Moscinski MD.
• H Lee Moffitt Cancer Center Research
  Institute, University of South Florida, Tampa,
  FL, USA.
• Acknowledgements Eric Zhong, Mantle Cell
  Lymphoma Task Force(J Tao, E Sotomayor, S
  Dessureault), H Molina, J Balasi, G Shaheen
  HISTOLOGY FLOW CYTOMETRY, MOFFITT)

ABSTRACT BACKGROUND A method and approach was
developed for fully automated measurements of
immunostained lymphocytes in tissue sections by
means of digital color microscopy and patent
pending advanced cell analysis. The validation
data for population statistic measurements of
immunostained lymphocytes in tissue sections
using tissue cytometry is presented. DESIGN
Segmentation of a 512 x 474 RGB image and display
of statistical results table took 12 to 15
seconds using proprietary developed algorithms.
We used a panel of 7 antibodies for validation on
14 cases of mantle cell lymphoma (MC) giving
percentage positive, total lymphocytes, density
staining. An total of 2,027 image frames with
810,800 cell objects were evaluated. Antibodies
to CD3, CD4, CD8, Bcl-1, Ki-67, CD20, CD5 were
subjected to virtual flow cytometry on tissue.
The results of tissue cytometry were compared
with manual counts of expert observers and with
the results of flow cytometry immunophenotyping
of the same specimen. RESULTS The correlation
coefficient and 95 confidence interval by
linear regression analysis yielded a high
concordance between manual human results (M),
flow cytometry results(FC), and tissue cytometry
(TC) results per antibody, (r 0.9365 manual vs
TC, r 0.9537 FC vs TC). We noted a lower CD8
tumor infiltrating lymphocytes in blastic form
than in non-blastic form of mantle cell and a
higher Ki67 in the former subtype than in the
latter. CONCLUSION These results suggest the
new technology of Tissue Cytometry could be a
clinically valid surrogate for both manual and
flow cytometry analysis when only tissue
immunohistochemistry(IHC) is available for
diagnosis and prognosis. The application for
cancer diagnosis, monitoring, and prognosis is
for objective, rapid, automated counting of
immunostained cells in tissues with percentage
results. These results could be also be used in
standardization of cut off percentage used in
diagnosis for exemplary tool for leukemic
blasts(CD34) gt20 for acute leukemia, for
determining automated 30 cut off for positivity
for prognostic markers in diffuse large B-cell
lymphoma for bcl-2, bcl-6, Mum-1, and
CD10. INTRODUCTION We developed a method and
apparatus that perform image analysis of
immunohistochemically stained tissue. The
objective was to recreate the functionality of a
flow cytometer, but instead of using the
requisite cell suspension specimen, was modified
to be applied directly on stained microscopic
slides with tissue sections. The
hematopathologists standard of clinical practice
in most of patients reports is to estimate the
percentage of immunohistochemically stained cells
and report the visual or manual estimate. This
practice is subjective and often gives wide range
of results that depends on the level of the
microscopists skill. This is due to the
difficulty in counting positive cells accurately
because of overlapped stained nuclei, variability
of staining, and the limitation of our visual
system. A significant number of hematology cases
are fixed and embedded in paraffin and are not
amenable to flow cytometry analysis which
requires fresh cell suspension of tissue.
Routinely in pathology practice, a panel of 5 to
15 antibodies in average are applied to the slide
based tissue sections to create a differential
matrix to rule in or out a diagnosis based on the
tumor associated markers. The role of these
markers extend beyond ancillary and often takes
center stage in accurate diagnosis. The use of
IHC may shift the diagnostic probability from 75
to 100. This is especially true in
hemato-pathology diagnosis where an enhanced
diagnostic accuracy is reported if the
immunologic results are included ( Blood,
Armitage Int Lymphoma study, 1997). The enhanced
accuracy was across the board and was noted from
5 to 35 of the cases. Yet currently, there is
yet no system commercially available that will do
flow cytometry results on paraffin embedded
tissue immunohistochemically stained sections
hence the motivation to tackle this difficult
challenge. We applied a novel, useful, and
accurate algorithm based on population or
cell-based approach instead of the more common
pixel or area-based approach used by many current
research imaging algorithms to come up with
results similar to flow cytometry dot plot
histograms. These results, along with a table,
are familiar and provide an objective percent
positive and negative counts of tissue
immunohistochemistry to the diagnostic
pathologists and hematologist-oncologists and
enhance accuracy of diagnosis, disease
monitoring, and prognostication.
CORRELATED IMMUNOSTAINED IMAGES AND CORRESPONDING
DOT PLOT TISSUE CYTOMETRY RESULTS (CELL SIZE IN
MICRONS VS STAINING DENSITY)
Color image frame of CD8 immunostained tumor
infiltrating cell response and mantle cell
lymphoma(Brown DAB and Blue Hematoxylin, 20x).
Segmented brown stained CD8 lymphocytes along
with the unstained blue nuclei of non-CD8 tumor
MC lymphocytes
CD3
24
99/422
Bcl-1
RESULTS OF TISSUE CYTOMETRY ON CD8 STAINING Table
showing Staining Density, Total Cell Population,
Positive Cells, Percentage Positive Cells. (1 run
in 1case table output formatted as a batch mode )
85
Positive brown cells segmented below (The
numerator)
FLOW CYTOMETRY SIZE(FSC) VS CD8 FLUORESCENCE
INTENSITY(LINEAR )
6 CD8
311/367
Ki-67
28
Negative cells below are added to positive cells
to give the total cells ( The denominator )
TISSUE CYTOMETRY SIZE VS STAINING DENSITY
(LINEAR) Cell Size in microns in Y and the
Staining Density in X( 0 to 255, where blue stain
is close to zero and brown stain is towards 255)
DESIGN MATERIALS Paraffin embedded material of
lymphocyte rich tissue was used for this study.
We used 7 monoclonal antibodies for validation on
14 cases with tissue biopsy of the excised lymph
nodes obtained for diagnosis of lymphoma. A
total of 2,027 image frames with 810,800 cell
objects were evaluated. Membrane reactive
antibodies to T cell associated markers CD3, CD4,
CD5, and CD8 were used. In addition, B cell
reactive monoclonal antibody to CD20 ( L26,
mature B cells) was used. These antibodies were
run in flow cytometry and correlated with tissue
immunoreactivity using virtual flow cytometry on
tissue. Nuclear reactive antibodies to
Bcl-1(cyclin D1, mantle cell lymphoma and others)
and Ki-67 (Mib-1, proliferative marker) were
analysed by tissue cytometry only. METHODS Flow
cytometry analysis was done using logical gating
on dual CD20 and CD5 positive cells counting
5,000 events. Becton Dickinson FacsCalibur with
CellQuest software version 3.2.1 four color
acquisition and analysis. Membrane reactive
antibodies to T cell associated markers/clone
names with fluorophores are CD3 PerCP-Cy5.5 (
SK7), CD4 FITC ( SK3), CD5 APC (L17F12), and CD8
APC (SK1) were used. CD20 APC ( L27, mature B
cells) and CD71 PE (YDJ1.2.2 ) was used. The
manual estimates of immunoreactivity were
compared with the results of cell suspension
flow cytometry results and with the results of
virtual flow cytometry on each of the
corresponding series of tissue obtained for
diagnosis. Both the positively stained cells and
the unstained nuclei of the same type of cells
are sequentially extracted to yield a numerator
and denominator to calculate percent positive and
total cells. Stain density is also obtained per
cell and correlated with size. The pixel size was
converted to microns and the density spread from
0 to 255 where negative cell events are plotted
blue and positive cell events plotted red.
Positivity criteria was determined by novel
automated thresholding-visual concordance
subroutine. The slides were examined using a
Leica brightfield microscope, objective 20x
equipped with Insight color CCD(Diagnostic
Instruments, Inc,Westlar Germany), version 3.4
for Windows NT 2000. The camera chip
photoreceptive field (1060 x 1020 pixels) was by
software-mode trimmed to 512 x 474 pixels for
dimensionality reduction, object size optimum
visualization, and for computational efficiency.
These images were stored as either a JPEG or PICS
file and analysed using an previously developed
advanced cell imaging software. No manual or
interactive labeling or shading or color
correction was performed. The light intensity
rheostat was set to 7.0 of 12.0. The light source
was 30 watt 12 v incandescent bulb with a
condenser blue filter 80a Tiffen, with condenser
aperture set a 0.5 ph, under Kohler illumination,
and using a 20x nplan 0.4 NA objective. The
camera has a single CCD with interpolated R, G,
B filters overlying each substrate pixel was
used. The pixel size for the 20x image frame of
512 x 474 pixels corresponded to 1.5 pixels per
micron. A lymphocyte nuclei averaged 12-15
microns in nuclear diameter but since these
nuclei were tangentially cut in a random fashion
by the microtome, the range is noted from 5 to 35
microns in tangent diameters. The image was
manually focused, captured, and saved. Each image
file was 711 Kbytes. Each image frame took
from12-15 seconds from start of analysis to
statistical table and correlated dot plot
histogram result generation.
133/480
CD5
55
 
Correlation of flow cytometry(flow) cell
suspension fluorescent analysis with manual
estimate of immunostaining results and with the
computerized Tissue Cytometer (TC)
357/644
CONCLUSION We applied a recursive thresholding
algorithm to segment intensity histograms from
red and blue channels of image in RGB as a
preprocessing step in segmentation of
chromogen-stained cells. We found that this
approach along with cellular logic is able to
accurately segment optimally stained cells with
appropriate counterstain( for brown
DAB-chromogen-immunostained cells and blue
hematoxylin-stained non-immunoreactive
cells). Given an array of parameters that could
be used to automate counting of positively
stained cells, we found this approach of using
density and color and size only information is
able do population statistics and convert the
tissue immunostaining image to computerized data
as a table of results and dot plot
two-dimensional histogram results with reliable,
robust, accurate reporting similar to that of a
flow cytometer. We differ from the current
methods of image analysis systems by focusing on
cell-based population statistics instead of
pixel-area data. The correlation between each
case run in flow cytometry and estimated by
experts and by the advanced image Tissue
Cytometry is high and suggest a valid approach to
objectively quantifying immunostaining of lymph
nodes in tissue by a defined set of monoclonal
antibodies useful in lymphoma diagnosis,
monitoring and prognosis. REFERENCES Cualing H.
Automated Analysis in Flow Cytometry. Cytometry,
2000 , 42p.110-113. Young IT, Quantitative
Microscopy. IEEE Engineering in Medicine and
Biology, 1996, 15(1) p.59-66. Ridler TW ,Calvard
S. Picture Thresholding Using Iterative Selection
Method. IEEE Trans. On Systems, Man, and
Cybernetics, 1978. SMC-8(8)p 630-632. Cualing H.
Kothari R, Balachander T. Immunophenotypic
Diagnosis of Acute Leukemia Using Decision Tree
Induction. Lab Investigation, 1999,
79p.205-212.
r 0.95
r 0.93
RESULTS When compared with manual and flow
cytometry results, the concordance of Tissue
Cytometry is high. The results of tissue
cytometry were compared with manual counts of
expert observers and with the results of flow
cytometry immunophenotyping of the same specimen.
The correlation coefficient and 95 confidence
interval by linear regression analysis yielded a
high concordance between manual human results
(M), flow cytometry results(FC), and tissue
cytometry (TC) results per antibody, (r 0.9365
manual vs TC, r 0.9537 FC vs TC). We noted a
lower CD8 tumor infiltrating lymphocytes in
blastic form than in non-blastic form of mantle
cell and a higher Ki67 in the latter subtype than
in the former. Technically oriented images,
histograms, image analysis results are presented
on the right panels including the regression
analysis figures.