|
Introduction:
The microarray is a popular gene expression-profiling tool
allowing the analysis of many genes in a single experiment. Upon choosing and
starting to use a microarray platform, several experimental parameters need
to be optimized before performing an experiment. Microarray manufacturers
optimize appropriate conditions specific to their platform such as
hybridization time, temperature, and washing stringency. However, the amount
of labeled RNA material used in the experiment depends not only on the array
itself but also on the genes of interest in the study. This article discusses
some considerations to make when deciding how much labeled sample to
hybridize with a microarray using the Oligo GEArray® from SABiosciences as an
example.
Generation of a Microarray Signal:
In a microarray experiment, the relative levels of gene expression are
directly proportional to the intensity of a hybridization signal. High
hybridization intensity at a given microarray spot indicates that the
original sample contains a relatively high abundance of the corresponding
transcript. Low hybridization intensity means the transcript is relatively
rare. Two different steps in the microarray protocol contribute to the
intensity of each signal on a microarray: the hybridization of the labeled
sample to the microarray element and the detection of the labeled sample on
the array.
Microarray manufacturers pre-optimize any chemical reactions involved in
detection, and standard imaging devices (such as a fluorescent laser scanner,
chemiluminescent CCD camera, or X-ray film) are specifically designed to
capture and record the array images. Therefore, careful control of exposure
times as well as instrument or software parameters easily optimizes the
detection of the signal without requiring a repeat experiment. Optimizing
hybridization conditions proves more difficult because determining the
results of changing one condition requires a complete repetition of the
microarray experiment. Therefore, it is often helpful to perform a pilot
experiment by hybridizing replicate microarrays with different amounts of
labeled sample.
The hybridization reaction consists of two substrates: the nucleic acid on
the array and the labeled nucleic acid in solution. The amount and
distribution of the nucleic acid present on the array is fixed by the array
platform. The researcher controls the amount of labeled sample incubated with
the array. The amount of labeled sample used in a microarray experiment
depends on balancing two opposing considerations.
-
On one hand, the hybridization of more labeled sample
to the microarray maximizes the detection of less abundantly expressed
genes, increases the present call, and decreases the number of false
negatives.
-
On the other hand, increasing amount of the labeled
sample causes signal saturation, decreases the ability to resolve changes
in the expression of more abundantly expressed genes, and increases the
number of false positives.
To illustrate how microarray results change with
increasing amounts of labeled sample, the data from identical microarrays
hybridized with different amounts of sample are directly compared to one
another in the following experiments. Hundreds to thousands of hybridization
reactions are measured simultaneously on the same microarray surface. Every
RNA sample contains thousands of transcripts ranging from low to high copy
number. Thus, a wide range of hybridization signals intensities are observed
on any given microarray. To determine how well the results agree, the
intensity values are simply plotted against one another. If the data in these
types of plots agree, the data can be fit to a straight line. Incidentally
when comparing microarray data from two different experimental samples,
points deviating from a line of with a slope of one indicate changes in gene
expression.
Example One: Higher Density Array with Smaller Spot Sizes Figure
1 displays results from a higher density array (480 genes) with smaller array
spots. When the data from a microarray hybridized with a small amount of
sample (2 micrograms) are compared with larger amounts (4, 6, or 8
micrograms), the signal intensities increase for each gene with increasing
sample as expected. At very high amounts of sample (particularly 6 or 8
micrograms), the intensity of some array spots becomes saturated; that is,
their intensity reaches a maximum and does not increase any further.
Therefore, the curve fit deviates from a straight line at higher intensity
values and levels off instead. Saturated microarray signals confound the
interpretation of microarray data. They mask differences in the expression of
the corresponding genes between experimental conditions. Saturated signals
also skew the expression profiles of the other genes on the microarray when
using the median value to normalize or standardize the microarray data.
Choose an amount of labeled sample that minimizes the number of saturated
signals.
 |
Figure 1: More labeled sample saturates the signals
from more genes.
XpressRef™ Human Universal Reference Total RNA (Catalog Number
GA-004) was converted to labeled cRNA target using the TrueLabeling-AMP™
Linear RNA Amplification Kit (Catalog Number GA-010). Different
amounts (2, 4, 6, and 8 µg) of biotinylated cRNA were hybridized with
separate Oligo GEArray Human Hematology / Immunology Microarrays
(Catalog Number OHS-801). The signal intensities from each spot on
each microarray are plotted versus the signal intensities for the
microarray hybridized with 2 µg of labeled cRNA target. |
On the other hand, the higher amount of sample increases
the number of genes determined to be expressed (or present) from 53.5 to 79
percent, and decreases the number determined to be not expressed (or absent)
from 46.5 to 21 percent as defined and seen in Table 1. Absent calls cannot
be interpreted because their expression could either be non-existent or could
lie below the limit of detection of the method. Microarray users refer to the
latter situation as a false negative. Using larger amounts of labeled sample
allows the detection of less abundant messages and reduces false negatives.
However, microarray experiments often generate non-specific label.
Hybridizing larger amounts of labeled sample also increases the exposure of
the microarray to this non-specific label. Microarray users call any
contribution of signal intensity to microarray spots by non-specific label as
false positives. Therefore, using more labeled sample may also increase the
rate of false positive signals. Choose an amount of labeled sample that
maximizes the present call (minimizes false negatives) while trying not to
introduce too many false positives.
Table 1: But more labeled sample also increases the
present call.
The maximum, total, average, and median signal intensities for each
microarray result in Figure 1 are displayed. The average background and
standard deviation in the background value are also displayed. To define the
number and percent of present (expressed) genes and absent (unexpressed)
genes, a threshold was defined as the average background plus three standard
deviations. Signals above this threshold are present calls; signals below,
absent calls.
|
2 ug |
4 ug |
6 ug |
8 ug |
| Maximum intensity |
52546.00 |
49122.00 |
51301.00 |
51729.00 |
| Total Intensity |
961229.00 |
1351152.00 |
1833587.00 |
2110141.00 |
| Average intensity |
2002.56 |
2814.90 |
3819.97 |
4396.13 |
| Median intensity |
396.00 |
564.50 |
761.00 |
894.50 |
| Background |
270.88 |
370.38 |
437.63 |
504.63 |
| stdev of background |
35.56 |
29.14 |
44.91 |
27.08 |
| Cutoff |
377.55 |
457.79 |
572.36 |
585.85 |
| P calls |
257 |
295 |
325 |
378 |
| A calls |
223 |
185 |
155 |
102 |
| %P |
53.54% |
61.46% |
67.71% |
78.75% |
| %A |
46.46% |
38.54% |
32.29% |
21.25% |
Example Two: Lower Density Array with Larger Spot Sizes
The amount of labeled material required for the experiment also depends on
the array itself. The optimal amount of sample for one array may not be the
same for another array. To illustrate this point, Figure 2 displays the
results from a lower density array (112 genes) with larger array spots. The
spots on these microarrays contain more nucleic acid distributed over a
larger area allowing the spot to bind more labeled sample. As a result, no
obvious signal saturation occurs for the more abundantly expressed genes at
high levels of labeled sample. The curve fit does not deviate form a straight
line as dramatically as the higher density, smaller spot array despite the
significantly larger amounts of labeled sample (6 to 20 micrograms) used in
the hybridization. This array platform seems to tolerate larger amounts of
sample without causing undesirable effects such as signal saturation
providing the researcher more latitude in choosing the optimal amount.
Figure 2: More labeled sample does not saturate larger spot sizes as
easily.
XpressRef™ Mouse Universal Reference Total RNA (Catalog Number GA-005) was
converted to labeled cRNA target using the TrueLabeling-AMP™ Linear RNA
Amplification Kit (Catalog Number GA-010). Different amounts (6, 10, 15 and
20 µg) of biotinylated cRNA were hybridized with separate Oligo GEArray
Mouse Cell Cycle Microarrays (OMM-020). The signal intensities from each spot
on each microarrays are plotted versus the signal intensities for the
microarray hybridized with 6 µg of cRNA target.
As seen in Table 2, the present call still increases (from 47 to 77 percent)
and the absent call still decreases (from 53 to 23 percent) with an
increasing amount of sample hybridized to the lower density, larger spot
microarray. At the same time, the number of false negatives should also
increase, but the number of false positives would also still increase as
suggested for the higher density, smaller spot microarray. This analysis
cannot tell whether the lower density microarray produces more or less false
negatives and false positives than the higher density microarray. At the very
least, these results suggest that different array platforms may require a
different balance of these considerations.
Table 2: But more labeled sample still increases the present call on
larger spot sizes.
The maximum, total, average, and median signal intensities for each
microarray result in Figure 2 are displayed. The average background and
standard deviation in the background value are also displayed. To define the
number and percent of present (expressed) genes and absent (unexpressed)
genes, a threshold was defined as the average background plus three standard
deviations. Signals above this threshold are present calls; signals below,
absent calls.
|
6 ug |
10 ug |
15 ug |
20 ug |
| Maximum intensity |
47059.00 |
46656.00 |
47109.00 |
47676.00 |
| Total Intensity |
504099.00 |
635073.00 |
701454.00 |
821343.00 |
| Average intensity |
3938.27 |
4961.51 |
5480.11 |
6416.74 |
| Median intensity |
686.00 |
1203.50 |
1356.50 |
2076.50 |
| Average Background |
617.71 |
811.71 |
1044.57 |
1549.29 |
| Standard Deviation |
32.11 |
64.69 |
32.56 |
42.64 |
| Cutoff |
714.05 |
1005.78 |
1142.24 |
1677.22 |
| Present calls |
60 |
89 |
83 |
99 |
| Absent calls |
68 |
39 |
45 |
29 |
| Percent Present |
46.88% |
69.53% |
64.84% |
77.34% |
| Percent Absent |
53.13% |
30.47% |
35.16% |
22.66% |
Summary:
The amount of labeled material used for microarray hybridization strikes a
balance between minimizing the number of saturated signals and the maximizing
present call without introducing false positives. When deciding on the amount
of labeled target to be used, perform a preliminary experiment such as the
one described in this article. Start with the recommended range of amounts
provided by the manufacturer of the microarray platform, but also take into
account the nature of the genes of interest. Genes expressed at low level
require more sample to yield a signal above the limit of detection. On the
other hand, genes expressed at high level require a smaller amount of sample
to avoid saturation of their signals.
The results of such an experiment easily identify the presence of saturated
signals and measure the percentage of present and absent calls. However, only
methods that verify microarray data such as RT-PCR can identify the frequency
of false negatives and false positives. Their precise numbers cannot be
predicted by the experimental analysis described here. RT-PCR may be
performed for a subset of the genes on the microarray to completely optimize
the amount of labeled sample for your experiment. However, such work
generally exceeds the scope of a preliminary optimization of a microarray. A
good balance between saturation as well as positive and negative calls
usually maximizes the number of true positives and true negatives at the same
time. Just remember to always verify any interesting microarray with a more
rigorous gene-specific assay before continuing with your study or submitting
the results for publication.
Related Products:
Oligo GEArray® Microarrays for Human, Mouse, and
Rat
Oligo GEArray focused DNA microarrays are carefully designed to provide gene
expression information relevant to biological or disease pathways quickly and
simply at a cost every laboratory can afford. Because of the focused design,
data handling is straightforward and your research project can progress more
rapidly with information from well-characterized genes.
TrueLabeling-AMP™Linear RNA
Amplification Kit
Our TrueLabeling-AMP™Linear RNA Amplification Kit employs an extremely
convenient one-tube protocol that saves a day over conventional oligo array
labeling protocols. In addition to being fast and convenient, the
TrueLabeling-AMP™ Linear Amplification Kit provides the accuracy and
sensitivity you need for reliable gene expression profile analysis.
XpressRef™ Universal Reference RNA from
Human, Mouse, and Rat
XpressRef Universal Reference Total RNA is a standardized sample of RNA
designed to help streamline and optimize your gene expression studies using
microarrays or RT-PCR. The high quality of the RNA insures the successful
synthesis of microarray probe or PCR template every time. The broad
representation of genes in these RNA samples makes them useful for studying
nearly every gene in the human, mouse, or rat genome.
|
