|
|
Automated analysis
Quantitative
expression analysis in tissue has a long and checkered history.
Pathologists have devised numerous "semi-quantitative" grading
systems that have waxed and waned in popularity. As the information
capacity of computers increased, morphometric quantitative
analysis became possible. Like the cDNA microarrays, the tissue
microarray format lends itself to more quantitative analysis.
However, tissue microarrays present some special problems that
require dedicated readers, or at least dedicated software.
An automated analysis protocol must not only be able to select
the region of interest, but also normalize it so that the expression
level read from any given disk can be compared with other disks.
A related problem is that of subcellular localization. Comparisons
of nuclear or membranous staining are quite different than
total cytoplasmic staining. Although there are now a number
of automated devices for reading tissue microarrays (BLISS,
Chromavision ACIS, Biogenex, Applied Imaging) in our lab we
use AQUA, a dedicated, automated TMA analysis software, written
by Dr. Robert Camp. This system is completely described in
a paper in Nature Medicine (Camp et al, 2002).
Briefly, the concept behind AQUA is to use molecular, rather
than feature based compartmentalization. Subcellular compartments
are defined by molecular interactions using one set of fluorophores,
then the protein of interest is quantified using another fluorophore
within the previously defined compartments. Often Cy5, a fluorophore
in the far red, is used since there is minimal tissue auto-fluorescence
at this emission wavelength. The analysis is based on two algorithms,
one for co-localization and one to compensate for section thickness.
The co-localization algorithm defines and normalizes for area.
The exponential subtraction algorithm is required since the
thickness of the tissue sections results in overlap of the
subcellular compartments. The result is an AQUA score that
is directly proportional to the number of molecules per unit
area. The images shown in Figure 1 are of a breast cancer tissue
microarray core immunofluorescently stained with a rabbit pan-cytokeratin
antibody (Figure 1A), DAPI (Figure 1B) and an estrogen receptor
antibody (Figure 1C) allowing for differential fluorescent
tagging of each. In this example, keratin defines a tumor mask,
DAPI defines a nuclear compartment and estrogen receptor is
measured quantitatively within the pixels in the keratin mask
within the DAPI compartment. This objective and continuous
scoring technology has revealed numerous associations with
outcome not previously discernable to pathologists using nominal "by-eye" scoring
methods, as illustrated in Figure 2.
 |
Figure 1. Immunofluorescent Images Used in Automated
Quantitative Analysis of Tissue Microarrays
The antibodies used for immunofluorescence were rabbit
pan-cytokeratin antibody from DAKO (Glostrup, Denmark)
Estrogen Receptor antibody (mAb clone 1D5, DAKO) and
DAPI, allowing for differential fluorescent tagging of
each. A. Cytokeratin staining (Cy2, green) of the breast
cancer TMA core shows strong staining of epithelial tissue,
which is used to define a binary mask for the tumor region
to separate it from the surrounding stroma. B. Top right:
DAPI (blue) stains all nuclei in the specimen within
both tumor and stromal regions. This is used to define
the subcellular compartment of 'nuclei'. C. Estrogen
Receptor (ER) staining (Cy5, red) shows nuclear staining.
Cy5 is used as for the staining of the target of interest
since it is outside the auto-fluorescence spectrum of
tissue. D. This three-color overlay image illustrates
the separation of epithelial tumor (green regions) from
the stroma, which stained only with DAPI. The overlay
of the ER staining onto the cytokeratin and DAPI images
shows that ER stains nuclei only within the breast tumor
region and not the stromal nuclei, resulting in a magenta
color. |
 |
Figure 2. Automated quantitative analysis with subcellular
localization of β-catenin identifies prognostic classes
in colon carcinoma not discernable by pathologist-based
analysis.
A cohort of 310 colon carcinomas was analyzed for the
relative amount of nuclear-localized β-catenin. A. Analysis
of nuclear β-catenin levels in a pathologist-based analysis
using a 4-point scale (0-3+) fails to find a significant
survival difference when comparing tumors with the highest
levels of nuclear localized β-catenin (3+, 19% of cases)
versus the remaining cases (P = 0.2354). B.-C. In contrast,
automated analysis of tumor subsets with progressively
higher levels of β-catenin show increasingly poorer survival,
with increasing significance: B, top 10%, P = 0.0309;
C, top 6.7%, P = 0.0028). Insets show the frequency distribution
of intensity scores for each analysis with the selected
subset in black (the x-axes for the insets are not shown
but represent the AQUA score and extend linearly from
1 to 1000). |
|
|
