SEMIQUANTITATIVE AND MULTIRESOLUTION-BASED HISTOLOGICAL ANALYSIS OF GERM LAYER COMPONENTS IN TERATOMAS DERIVED FROM HUMAN, NON-HUMAN PRIMATE AND MOUSE EMBRYONIC STEM CELLS.

1
SEMIQUANTITATIVE AND MULTIRESOLUTION-BASED
HISTOLOGICAL ANALYSIS OF GERM LAYER COMPONENTS IN
TERATOMAS DERIVED FROM HUMAN, NON-HUMAN PRIMATE
AND MOUSE EMBRYONIC STEM CELLS. John A. Ozolek1,
Carlos A. Castro2, Garrett Jenkinson3,4, Amina
Chebira3, Jelena Kovacevic3,4, Christopher S.
Navara2, Meena Sukhwani2, Kyle E. Orwig2, Ahmi
Ben-Yehudah2, Gerald Schatten2 1Department of
Pathology, Children's Hospital of Pittsburgh,
University of Pittsburgh, Pittsburgh, PA,
USA 2Department of Obstetrics and Gynecology,
Magee Womens Research Institute and Foundation,
University of Pittsburgh, Pittsburgh, PA,
USA 3Department of Biomedical Engineering,
Carnegie Mellon University, Pittsburgh, PA, USA
4Department of Electrical and Computer
Engineering, Carnegie Mellon University,
Pittsburgh, PA, USA
ABSTRACT Background The capability of cells
derived as embryonic stem cells (ES) to produce
tissue components comprising the three
developmental germ layers (teratomas) is the
single most important test of pluripotency.
Within the pathology literature, teratomas have
been classified according to the complexity of
tissue organization such that in diagnostic
terms, lesions with two of the three germ layers
are considered teratomas and lesions with high
order arrangement of tissues resembling an embryo
or fetus are considered by some to be teratomas.
Little is known about the volume of distinct
tissue types produced, how tissue types are
organized, variables that may influence tissue
differentiation, and species differences within
teratomas. We hypothesize teratomas derived from
ES cells of different mammalian species will
exhibit species specific three dimensional tissue
distribution and volumes within
teratomas. Methods Testes of SCID mice were
injected with putative ES cells derived from
mouse (MES) and non-human primate (nhpES). A
human ES line (H7) was also used. Animals were
sacrificed when visible lesions were identified.
The entire teratoma was extracted, fixed in
formalin, serially sectioned and processed by
routine histological techniques. For each lesion,
the amount of each representative germ layer
(ectoderm (EC neuroglial, skin), mesoderm (ME
mesenchyme, bone, cartilage), endoderm (EN
gastrointestinal, bronchial, pancreas)) was
semiquantified according to the following scale
1-0-20, 2-21-40, 3-41-60, 4-61-80, 5-81-100.
germ layer is given as mode of percentage.
Incubation times are given in days with standard
deviation in parenthesis. Statistical analyses
were done with ANOVA and t-test for continuous
variables and Wilcoxon and Mann-Whitney tests for
non-parametric variables. Results are expressed
as (human vs nhp vs mouse) where values are
given. Results Days of ES cell incubation after
mice injection were not statistically significant
between groups (71 vs 78 vs 68). However, human
ES derived teratomas were larger than nhpES
teratomas, but not larger than MES teratomas (2.6
cm vs 1.8 cm vs 1.9 cm). Teratomas derived from
MES and nhpES showed significantly higher amounts
of EC (5 vs 1) than human ES teratomas while
human derived teratomas demonstrated higher
amounts of ME than nhp or mouse (4 vs 1). EN
amount did not differ between groups. Within
species nhpES and MES derived teratomas had
greater amounts of EC than ME or EN while human
derived teratomas had greater amounts of ME than
EC or EN. Conclusions We conclude that species
differences exist by amounts of the various germ
layers produced in teratomas and may not be
related to incubation time or tumor size. We
speculate that this may reflect basic
developmental programming differences between the
species. Further sophisticated bioimaging
analysis and three-dimensional reconstruction of
teratomas will further elucidate these
differences.

• METHODS
• Semiquantitative Analysis
• Testes of NOD-SCID mice (Jackson Laboratories,
  Bar Harbor, Maine) were injected with putative ES
  cells derived from mouse (MES), non-human primate
  (nhpES 60,000-120,000cells/testes), and a human
  (hESH7 3,000-4,500 cells/testes) source. A
  total of 8, 3, and 2 teratomas were derived from
  non-human primate, murine, and human ES cells
  respectively.
• Mice were sacrificed when visible lesions were
  identified.
• The entire teratoma was carefully dissected and
  removed in its entirety and fixed in 10
  phosphate buffered formalin (3.6 formaldehyde).
• After fixation, lesions were measured, serially
  sectioned and processed by routine histological
  methods.
• For each lesion, the amount of each
  representative germ layer (i.e. ectoderm,
  mesoderm, and endoderm) was estimated on each
  slide of the serially sectioned teratoma using
  the following scale 1-0-20, 2-21-40,
  3-41-60, 4-61-80, and 5-81-100. Tissue
  components of each germ layer were identified
  according to the following table (Table 1).
• Size (greatest dimension) of lesions is given in
  centimeters. Incubation times are given in days.
  The percentage of germ layer present is given as
  median of percentage. Statistical analyses were
  done with ANOVA and t-test for continuous
  variables and Wilcoxon and Mann-Whitney tests for
  non-parametric variables.

Figure 6
Figure 2 Multiresolution classifier where an
image is decomposed (see Figure 3) and the
resultant subbands subjected to the generic
classifier (Figure 1) until assigned class label
(global decision) is achieved.
SWT transformation provides highest accuracy of
identifying specific tissue types compared to
using DWT or the generic classifier alone.

• CONCLUSIONS
• In general, ectoderm and mesoderm predominate
  within teratomas derived from ES cells regardless
  of the species.
• Teratomas derived from hES cells tend to be
  larger than those derived from nhpES or MES
  cells.
• Non-human primate and mouse teratomas show a
  greater percentage of ectoderm derived tissue
  than human teratomas. Human teratomas show a
  greater percentage of mesoderm than non-human
  primate or mouse. For all species endoderm
  derived tissues are present in the least amount.
• Using the multiresolution classifier with texture
  features only computed on the multiresolution-deco
  mposed digital images of tissue types within
  teratomas, we obtain accuracy of 83.
• SPECULATIONS
• We speculate that developmental programs and/or
  timing differ between species such that mammals
  with shorter gestational ages show greater
  prevalence of ectoderm derived tissues
  (particularly neural tissue which is first to
  develop) even in a seemingly disorganized
  conglomerate of tissues comprising the teratoma.
• The ability to recognize and quantify tissue
  types using digital imaging analysis tools
  including MR transforms and then reconstruct
  these lesions in three dimensions will allow us
  to understand the spatial relationships of tissue
  types and correlate with high-resolution imaging
  studies.
• Higher accuracy of tissue typing using these and
  other MR transforms can be achieved using color,
  shape, and location.

• RESULTS
• Semiquantitative Analysis
• No differences were seen for incubation days
  between teratomas derived from nhpES, MES, or
  HES. The human teratomas sampled were
  significantly larger than either nhp or mouse
  derived lesions (Table 2).
• Both nhp and mouse derived teratomas demonstrated
  higher median percentage of ectoderm derived
  tissue present in their teratomas compared to hES
  derived teratomas. However, hES cell derived
  teratomas demonstrated higher percentages of
  mesoderm derived tissues than nhpES or MES
  derived teratomas (Figure 5). No differences
  were seen for percentage of endoderm derived
  tissues between nhp and mouse ES teratomas.
  Significant differences at a p-value of 0.02 were
  seen between endoderm derived tissues from nhpES
  and MES compared to hES teratomas (Table 2).
• Classification with and without Multiresolution
• For tissue types selected for analysis
  (mesenchyme, skin, myenteric plexus, bone,
  necrosis, and striated muscle), the
  multiresolution classifier improved accuracy of
  detecting a particular tissue type over the use
  of a generic classifier. Stationary wavelet
  transform produced a mean of 83 accuracy
  compared to 75 and 68, for discrete wavelet
  transform and no multiresolution respectively
  (Figure 6 Table 3).

• INTRODUCTION
• The ability to form lesions that recapitulate the
  three germ layers ectoderm, mesoderm, endoderm)
  during development is one of the assays
  (considered a gold standard) for determining if
  potential embryonic stem cell (ES cells)
  candidates are pluripotent.
• Within the pathology literature, human teratomas
  are classified according to the presence of
  immature and/or malignant tissue elements as
  these have prognostic significance in the
  pediatric and adult populations.
• While at first glance, most teratomas derived
  from ES cells appear as disorganized tissue
  masses with recognizable germ layer elements,
  little is known about the contribution of each
  germ layer to the lesion, the spatial
  organization of germ layer elements to one
  another, three-dimensional hierarchy of germ
  layer contribution and whether the final
  constitution of the teratoma is time and/or
  species dependent reflecting attempts to follow a
  developmental program.
• The ability to accurately detect and quantify
  specific tissue types will begin to allow the
  ability to detect species specific differences in
  developmental programming and enable accurate
  three-dimensional reconstruction of teratomas and
  comparison to high-resolution magnetic resonance
  imaging.

Discrete wavelet transform (DWT) with two levels
of decomposition and reconstrcution. g and h are
orthogonal lowpass and highpass filters

• METHODS
• Multiresolution Classification
• Multiresolution techniques have been developed
  over 20 years ago.
• Multiresolution classification new---First
  attempt to apply it to this type of data.
• If feasible, will allow accurate classification
  and quantification of tissue types throughout the
  entire teratoma.
• Results can be correlated with three-dimensional
  high-resolution magnetic resonance image
  renderings.
• Generic classifier Typically feature extraction
  (numerical) followed by classification (Figure
  1).
• Large multi-class images are separated into small
  single class images.
• Texture features are used in neural net
  classifier using 10-fold cross validation.
• Classification of tissue type achieved through
  multiresolution classification (Figure 2). It
  uses multiresolution decomposition---discrete
  wavelet transformation (DWT) or stationary
  wavelet transformation (SWT) (Figures 3, 4),
  followed by texture feature classification as
  well as weighting to combine local decisions into
  a global one.

TABLE 1 Tissue components of germ layers
ECTODERM MESODERM ENDODERM
Central nervous system Retina Cranial, sensory,enteric ganglia and nerves Epidermis Hair Skeletal muscle Bones Dermis Connective tissues Urogenital system Heart Hematopoietic Stomach Colon Liver Pancreas Epithelium of Trachea Lungs Pharynx Thyroid Intestine.
TABLE 2 of tissue types in species-specific
teratomas

• AIMS
• Compare using a semiquantitative approach the
  contribution of each germ layer to teratoma
  formation within species and between species.
• Determine whether germ layer contribution is
  based on the size of the teratoma or incubation
  time.
• Determine the accuracy of multiresolution based
  imaging analysis techniques in identifying
  specific tissue types derived from each germ
  layer.

NHP MOUSE HUMAN
Incubation (d) 77.8 (13.8) 68.3 (26.3) 70.5 (6.4)
Size (cm) 1.8 (0.3) 1.9 (0.6) 2.6 (0.1)
EC (median) 3 5 2
ME (median) 2 1 4
EN (median) 1 1 1

• CORRESPONDENCE
• John A. Ozolek, M.D.
• Assistant Professor of Pathology
• Children's Hospital of Pittsburgh
• 3705 Fifth Avenue
• Pittsburgh, PA 15213
• 412-692-5641/412-251-2248 (office/cell)
• 412-692-5650 (Department)
• 412-692-6550 (fax)
• ozolekja_at_upmc.edu
• Carlos A. Castro, D.M.D., M.D.
• Research Associate
• Department of Obstetrics, Gynecology and
  Reproductive Medicine
• Magee Womens Research Institute
• 204 Craft Avenue
• Pittsburgh, PA 15213
• 412-641-6086/412-310-3091 (office/cell)
• 412-641-2410 (fax)

SD in ( ), -plt0.01, See text in Results section
for statistical analysis of median percentage of
germ layers
TABLE 3 Accuracy of tissue classification using a
multiresolution classifier
ACCURACY NO MR DWT SWT
MEAN 68.0 74.7 83.2
SD 4.1 2.1 1.2
MAX 73.3 77.4 84.9
MIN 59.2 71.6 81.7
Neural network based generic classifier where
image features (texture, shape, color) are
subjected to neural network until output matches
desired class label