A Concise Guide to cDNA Microarray Analysis
|
Priti
Hegde, Rong Qi, Kristie Abernathy, Cheryl Gay, Sonia Dharap, Renee Gaspard,
Julie Earle-Hughes, Erik Snesrud, Norman Lee, and John Quackenbush
The
Institute for Genomic Research, Rockville, MD 20850
|
|
Introduction
Recently, a variety of techniques including SAGE
(12), differential display (5), oligonucleotide arrays (6), and cDNA microarrays
(8), have been developed that allow mRNA expression to be assessed on a
global scale, allowing the parallel assessment of gene expression for hundreds
or thousands of genes in a single experiment. The most common use of these
is for the determination of patterns of differential gene expression, comparing
differences in mRNA expression levels between identical cells subjected
to different stimuli or between different cellular phenotypes or developmental
stages. Microarray expression analysis (8) has a number of features that
have made it the most widely used method for profiling mRNA expression.
DNA segments representing the collection of genes to be assayed are amplified
by PCR and mechanically spotted at high density on glass microscope slides
using relatively simple x-y-z stage robotic systems, creating a
microarray containing thousands of elements. Microarrays containing the
entire set of genes from a microbial genome or tens of thousands of eukaryotic
cDNA clones can be easily constructed. The microarrays are queried in a
co-hybridization assay using two or more fluorescently labeled probes prepared
from messenger RNA from the cellular phenotypes of interest (10). The kinetics
of hybridization allows relative expression levels to be determined based
on the ratio with which each probe hybridizes to an individual array element.
Hybridization is assayed using a confocal laser scanner to measure fluorescence
intensities, allowing simultaneous determination of the relative expression
levels all the genes represented in the array.
Efficient expression
analysis using microarrays requires the development and successful implementation
of a variety of laboratory protocols and strategies for fluorescence intensity
normalization. The process of expression analysis can be broadly divided
into three stages:
I. Array Fabrication
II. Probe Preparation and Hybridization
III. Data Collection, Normalization and Analysis
Below we present protocols
that we have standardized and that have been used regularly in our laboratory
for microarray analysis. The procedures described in this article have
been tested and refined over the past year and have been optimized using
hybridization of RNA derived from cell lines to give reproducible and consistent
results. It should be noted that a number of alternative protocols have
been published (4) or are available via the World Wide Web (see for example,
Table 1), but the system that we describe here has a number of advantages
over these. In particular, the combination of printing, labeling, and hybridization
conditions that we have derived have allowed a significant reduction in
the quantity of starting total RNA required for analysis.
|
|||||||||||||||||||||||||||||||||
I. Array FabricationMicroarrays are constructed by arraying PCR amplified
cDNA clones or genes at high density on derivatized glass microscope slides.
For the analysis of expression in most eukaryotes, expressed sequence tag
(EST) data represent the most extensive data for gene identification. ESTs
are single-pass, partial sequences of cDNA clones, and they have been used
extensively for gene discovery and mapping in humans and other organisms.
The EST approach has been widely adopted; more than 71% of all GenBank
entries and 40% of the individual nucleotides in the database are derived
EST sequences (9).
Generally, cDNA clones are selected to represent
as many unique transcripts as possible. There are a number of analyses
of these data that attempt to identify unique human transcripts within
the EST data, the two most widely used are UniGene (1) (<http://www.ncbi.nlm.nih.gov/UniGene/>)
at the National Center for Biotechnology Information (NCBI) and the TIGR
Human Gene Index (7) (HGI; <http://www.tigr.org/tbd/hgi/hgi.html>)
at The Institute for Genomic Research (TIGR). While both UniGene and HGI
are based on EST clustering, the TIGR protocol assembles the ESTs within
the clusters, producing Tentative Human Consensus (THC) sequences.
We selected cDNA clones for array construction using
the TIGR HGI as part of a program to assemble a 30,000 gene clone set.
THCs were chosen for representation in the clone set with preference given
to those containing known genes or those with mapped positions; additional
THCs were selected to represent as yet uncharacterized transcripts. For
each target THC, a single cDNA clone was identified based on the EST content
of the THC assembly.
I.A. PCR Amplification and Clone PreparationcDNA clone inserts can be amplified by PCR from
plasmid miniprep DNA or directly from clones in culture. In high-throughput
applications, amplification of clones from culture has the advantage of
being both more cost efficient and less labor intensive with lower cross-contamination
rates than amplification from plasmid DNA. Our amplification success rate
from culture is equivalent to that we have achieved using plasmid templates.
For more than 30,000 clones, our success rate for single-band amplification
is approximately 87.5%; 6.3% of reactions yield multiple or weak bands
and 6.2% fail to amplify. Results from a typical amplification using the
protocol described below are shown in Figure 1.
PCR amplification
The cDNA clones that are widely available through
the IMAGE consortium distributors – The American Type Culture Collection
(ATCC), Research Genetics, and Genome Systems – have been cloned into a
variety of vectors. While the majority have both M13(-21) and M13REV priming
sites. However, many have point mutations in either of these two “universal”
priming sites. We have designed alternative M13 primers that avoid these
point mutations and that have allowed amplification of clone inserts from
all of the vectors we have encountered to date. These new universal amplification
primers are:
M13 FWD: 5' GTT TTC CCA GTC ACG ACG TTG 3' M13 REV: 5' TGA
GCG GAT AAC AAT TTC ACA CAG 3'
Clone inserts are amplified using the following
protocol:
1.Selected clones are inoculated into 96 well deep-well blocks (Qiagen; Cat # 19573) containing1.2 ml of LB/Ampicillin (50 mg/ml) and incubated for 16 hours at 37ºC and 200 rpm in a shaking incubator. A 100 ml aliquot of each is archived for future use in microtiter plates containing 10% glycerol at -80°C. 2.Following overnight growth, 5ml of culture suspension are transferred into a 96 well Falcon U-bottom plate (BD Biosciences; Cat # 353077) containing 95ml of MilliQ water. 3.Microtiter plates containing diluted culture are heated to 95oC for 10 minutes in a laboratory oven to lyse the cells and release the plasmid clones. 4.Prior to PCR, cellular debris is removed by centrifugation at 1200´g for 3 minutes in a centrifuge equipped with microtiter plate carriers. 5.Clone inserts are amplified in 50ml PCR reactions in 96 well reaction plates (Perkin-Elmer Applied Biosystems Cat # N801-0560). A reaction master mix is prepared for each reaction plate:
6.For
each clone, add 48ml of master
mix to 2ml of culture supernatant
in 96 well PCR plate.
7.Reactions
are amplified in an thermocycler (MJ Research; PTC-225 Tetrad) using the
following cycling protocol:
Reaction clean-up
For efficient binding of the amplified clone inserts
to the slides, it is essential to remove unincorporated nucleotides and
primers from the reaction products. While there are a variety of techniques
that can be used, we have found filtration using 96 well multiscreen filter
plates (Millipore; Cat # MANU 03050) to give excellent DNA product recovery
without any significant contamination at relatively low cost.
PCR products are cleaned using the following filtration
protocol:
1.Transfer
PCR product (50ml) to the Millipore
filter plate.
2.Place
the filter plate on a vacuum manifold filtration system (Qiagen,
Cat # 19504 or Millipore Cat # MAVM0960R) and filter at a pressure of 15in
(380 mm) Hg for 10 minutes or until the plate is dry. 3.Add
30ml
of MilliQ water to each well and filter at 15mm (380
mm)Hg for 5-10 minutes or until the plate is dry. 4.Repeat
step 3. 5.Remove
plate from the manifold filtration system. Add 60ml
of MilliQ water to each well and place on a shaker. Shake vigorously for
10 minutes to resuspend the DNA. 6.Manually
pipet the purified product to a new 96 well plate. 7.Plates
containing the purified PCR products are then sealed using a cap mat (VWR;
Cat # 40002-002) and stored at 4oC for future arraying. I.B. Array PrintingMicroarrays are prepared by printing PCR amplicons
suspended in either a high salt or other denaturing buffer onto poly-L-lysine
or aminosilane coated glass microscope slides using a high-speed robotic
system. This process was originally described by Patrick Brown and collaborators
(8) at Stanford University and they provide plans so that others can replicate
their arraying robot (<http://cmgm.stanford.edu/pbrown/mguide/index.html>).
However, there are a number of companies that are selling robotic systems
for microarraying and these are listed in Table 2. We use a microarray
robot built by Intelligent Automation Systems (IAS) of Cambridge, Massachusetts.
Based on a high precision, four-axis Seiko robotic arm, the IAS arrayer
uses a 12-tip print head to array DNA samples from either 96- or 384-well
microtiter plates onto as many as 100 silanized glass microscope slides.
With an average spot size of 130 µm and the capability to adjust
the spot-to-spot spacing, the IAS arrayer can spot 19,200 elements (the
contents of 200 microtiter plates) or more onto a single slide.
Both
the slide surface and the spotting buffer are critical components for reproducible,
high-fidelity micorarray analysis. Most published reports have used high
salt buffers such as 3´SSC
to print DNA on poly-L-lysine coated slides (10). Our analysis suggests,
however, that aminosilane offers a more consistent surface with lower background
fluorescence. There are a number of commercial vendors for aminosilane
coated slides (see Table 2), but Corning CMT-GAPS™ aminosilane coated glass
microscope slides have been the most consistent. Using our protocol in
number of side-by-side comparisons, the CMT-GAPS™ slides produce approximately
half the background fluorescence of poly-L-lysine slides while yielding
signal intensities that are consistently higher (data not shown). In addition,
the spot morphology on CMT-GAPS™ slides is much more uniform, with fewer
“doughnuts” than on any of the alternatives we have investigated.
We
also investigated a number of different spotting chemistries to determine
which provides the best results in subsequent hybridization assays. Figure
2 shows the results of a comparison between 50% dimethyl-sulfoxide (DMSO)
and 3´SSC
as a spotting buffer. We have found that PCR products printed onto CMT-GAPS™
aminosilane coated glass microscope slides (Corning, Cat# 2550) using 50%
DMSO as a printing buffer provides the best substrate for hybridization,
giving the greatest hybridization intensities. Using 50% DMSO as a printing
solution has a number of additional advantages. DMSO denatures the DNA
allowing better binding to the slide and providing more single-stranded
targets for hybridization. Further, DMSO is hygroscopic and has a low vapor
pressure, allowing DNA prepared for arraying to be stored for long periods
of time without significant evaporation.
The
print head on our arrayer and most others use “quill” pens that use capillary
action to draw fluid into the spotting pens and surface tension interactions
to dispense solution onto the slide. The
Arrayit ChipMaker3ä
microspotting pins (TeleChem International Inc.) are very durable and can
reproducibly generate high-quality spots with good precision; all array
images shown were printed with the same set of ChipMakerä
3 over more than six months.A
variety of parameters such as the robot arm acceleration, temperature,
and humidity control both spot morphology and size. We have found printing
to be optimal at approximately 45% relative humidity and a constant temperature
of 72oF (22oC). Changes in humidity and temperature
have a significant impact on the size and morphology of spots, as well
as the efficiency of DNA binding to the slides and these must be carefully
controlled to provide the consistent spotting. Figure 3 shows the effects
of varying humidity and temperature on spot morphology and DNA retention.
DNA samples were spotted onto the slides as described above while temperature
and humidity levels were recorded on a chart recorder. During the printing,
temperature and humidity levels were allowed to vary continuously from
72ºF (22.2ºC) and 45-50% to a low of 62ºF (16.7ºC)
and 40-45%and a high of 80ºF (26.7ºC) and 80-85% respectively.
Following hybridization with a vector specific probe, we were able to reconstruct
the optimal printing conditions by using the chart recorder data to assign
temperature and humidity values to the spots. Arraying 1.Add equal volumes of purified PCR product to DMSO in a 96 well V-bottom plate (Corning; Cat # 3897). Typically, 5ml of each are used to prepare spotting plates that can be used to print 100 or more slides. 2.Slides to be printed are marked with a diamond-tipped pen, dust is removed by blowing the slides with high-pressure nitrogen gas, and the slides are placed in the arrayer. Care must be taken not to touch the surface of the slides as oils adversely affect the ability of the slide surface to bind DNA. 3.Microtiter spotting plates are loaded into the arrayer and PCR products are spotted onto the slides at 72oF and 45% relative humidity. 4.Following printing, the slides are allowed to dry and spotted DNA is bound to slide by UV-crosslinking at 90 mJ using a Stratalinkerä (Stratagene, Cat# 400071) and baking at 80oC for two hours. 5.Printed
slides are stored in a light-tight box in a bench-top dessicator at room
temperature until they are to be used for hybridization.
|
|||||||||||||||||||||||||||||||||
II. Probe Preparation and HybridizationMicroarrays assay differential gene expression
by co-hybridization of fluorescently labeled probes prepared from different
RNA sources. As with many other RNA-based assays, the purity and quality
of the starting RNA has a significant effect on the results of the assay.
Further, the products of the labeling reactions must be cleaned to remove
unincorporated labeled nucleotides that can produce a significant background
on the slides following hybridization. Finally, hybridization conditions
and wash must be optimized to provide high specificity to minimize cross-hybridization.
We have developed probe preparation and hybridization protocols using RNA
derived from human carcinoma cell lines as a model system; variations of
this protocol have been applied to the study of expression in rat and other
systems.
II.A. RNA ExtractionImpurities in RNA preparations
can have an adverse effect both on labeling efficiency and the stability
of the fluorescent labels that are used for microarray expression analysis.
We have found that Trizolä
(Life Technologies; Cat# 15596-014) gives consistently high quality RNA
from cell culture and many tissue samples, although additional steps must
be taken to remove polysaccharides when extracting RNA from some tissues.
Trizol extraction is quick and produces a high yield of total RNA.
RNA Extraction1.Aspirate
media from the cells and wash once with Phosphate Buffered Saline (PBS).
2.Add
5ml PBS and scrape cells from the plate.
3.Transfer
cell suspension to a 50ml polypropylene conical-bottom tube (Falcon Cat#
352070).
4.Wash
the plate with an additional 1ml PBS and add the suspension to the tube.
5.Pellet
the cells by centrifugation at 2300 rpm (900´g)
for 3 minutes at 4oC and discard the supernatant.
6.Add
2ml Trizolä
per ~2´106
cells (approximately one 150mm plate of fibroblasts) to the pellet and
pass the suspension through an 18 gauge syringe several times to disrupt
the pellet.
7.Incubate
the sample at room temperature for 5 minutes. 8.Add
0.4ml of chloroform (0.2ml/1ml Trizol) and shake vigorously for 1 minute 9.Incubate
at room temperature for 2 minutes 30 seconds. 10.Remove
cellular debris by centrifugation at 4000rpm (2700´g)
for 15min at 4oC. 11.Transfer
the supernatant to 1.2 ml microfuge tubes (0.5ml/tube) and an equal volume
of isopropanol to precipitate the RNA. 12.Incubate
at room temperature for 15 minutes. 13.Centrifuge
at 15,000rpm (21,000´g)
for 15 minutes to pellet the RNA. 14.Discard
the supernatant and resuspend the pellet in 70% ethanol. The RNA can be
stored in 70% ethanol at –20°C
until use. 15.Prior
to use, centrifuge at 15,000 rpm (21,000´g)
for 15 minutes at 4oC and discard supernatant. 16.Resuspend
the pellet in diethylpyrocarbonate (DEPC) treated water or RNase-free TE
buffer for labeling. II.B.RNA
Labeling The
ability to label small quantities of starting material is an important
consideration for the study of expression in rare patient samples and consequently,
we have focused on decreasing the quantity of starting material required.
Probes for microarray analysis are prepared from RNA templates by incorporation
of fluorescently labeled deoxyribonucleotides during first strand cDNA
synthesis. Either total or poly(A+) RNA can be used in the reverse
transcription reaction. Oligo(dT) labeling of total RNA provides consistently
high-quality probes from smaller quantities of starting RNA and without
the expense of poly(A+) purification. Figure 4 shows the results
of microarray hybridizations using labeled total or poly(A+)
RNA prepared from the same cell lines. An analysis of the fluorescence
intensities for the elements in arrays hybridized with probes prepared
from 1.5mg
of poly(A+) RNA (the equivalent 50-100mg
of starting total RNA) and 4mg
of total RNA indicate that total RNA labeling provides comparable probe
activity without any increase in background fluorescence. Typically,
we prepare labeled probes using Cy3- and Cy5-dUTP (Amersham Cat#s PA53022,
PA55022), although Cy-labeled dCTP (Amersham Cat#s PA53021, PA55021) can
be used with an appropriate change in the concentrations of unlabeled dNTPs
in the reaction. We have investigated a number of reverse transcriptases,
including AMV and MMLV and have found that Superscriptä
II RT (LifeTechnologies; Cat# 18064-014) generates probes with significantly
greater activity (data not shown). It
should be noted that both Cy3 and Cy5 are photosensitive and care should
be taken to minimize exposure to light during the labeling, hybridization,
washing, and scanning processes. Upon receipt, Cy-labeled nucleotides should
be aliquotted into single-use light- proof tubes and stored at –20°C
until needed. All reactions should be carried out in foil-wrapped tubes
and all hybridizations and washes in foil-wrapped containers. Probe Labeling and Purification1.Prepare
a labeling reaction master mix containing 500mM
dCTP, 500mM
dATP, 500mM
dGTP, 100mM
dTTP, Cy 3-dUTP/Cy 5-dUTP, 400U Superscriptä
II RT, 1mM dithiothreitol (DTT) and 1´
RT buffer. We typically prepare sufficient quantities for 20 labeling reactions
and store the unused solution at -20°C:
2.To
10mg
of total RNA (or 2mg
poly(A+)) in a microfuge tube, add 2mg
of oligo(dT) (18-20mer; Life Technologies Cat# Y012120) and DEPC-treated
WATER to a total volume of 10ml.
3.Incubate
the reaction mixture at 70oC for 10 minutes, and chill on ice
for one minute.
4.To
the RNA, add:
5.Mix
thoroughly and incubate at 42oC for 2 hours.
6.Briefly
centrifuge the reaction and add 1.5ml
of 20mM EDTA to stop the reaction.
7.Add
1.5ml
of 500mM NaOH and heat at 70oC
for 10 minutes to degrade the RNA
8.Neutralize
the reaction by adding 1.5ml
of 500mM HCl
9.Unincorporated
fluorescent nucleotides are removed by glass fiber filtration using GFX
columns (Pharmacia Cat# 27-9602-01) and the instructions provided by the
manufacturer.
10.Elute
the purified products using 50ml
of TE pH 8.0 and dry the probe to completion in a speedvac.
11.Resuspend
the probe in 10ml
of DEPC treated WATER. II.C. HybridizationThe
goal in any hybridization is to obtain high specificity while minimizing
background. We have developed protocols that give reproducible, high-quality
hybridization results while maximizing the measured fluorescence from the
array.
Aminosilane
coated slides bind DNA with high efficiency. Prior to hybridization, the
free amine groups on the slide must be blocked or inactivated, otherwise
nonspecific binding of labeled cDNA to the slide can deplete the probe
and produce high background. Although the slides can be blocked chemically,
we have found a simple prehybridization in a solution containing 1% bovine
serum albumin to be extremely effective in eliminating nonspecific binding
of the probe to the slide.
Prehybridization
has the additional advantage of washing unbound DNA from the slide prior
to the addition of the probe. Any DNA that washes from the surface during
hybridization competes with DNA bound to the slide. As the kinetics of
solution hybridization is much more favorable than surface hybridization,
this can dramatically decrease the measured fluorescence signal from the
microarray. All prehybridization and hybridization washes are carried out
in microscope slide staining trays (VWR Cat# 25461-003).
Prehybridization
1.Prepare
prehybridization buffer containing 5´SSC,
0.1% SDS and 1% bovine serum albumin (BSA; Sigma
Cat# A-9418).
2.Prepare
2´
hybridization buffer containing 50% formamide, 10´SSC,
and 0.2% SDS.
3.Place
slides to be analyzed into a Coplin jar (VWR Cat# 25457-200), fill with
prehybridization buffer, and incubate at 42oC for 45 minutes.
4.Wash
the slides by dipping five times in room temperature MilliQ water.
5.Dip
the slides in room temperature isopropanol and air dry.
Slides
should be used immediately following prehybridization. We have found that
hybridization efficiency decreases rapidly if the slides are allowed to
dry for more than one hour. Hybridization1.Combine
10ml
each of purified Cy3- and Cy5-labeled probes, mix well and add
to
block nonspecific hybridization.
2.Heat
the probe mixture at 95oC for 3 minutes to denature.
3.Centrifuge
the probe in a microfuge set at maximum angular velocity for 1 minute.
4.Combine
the probe with an equal volume of 2´
hybridization buffer that has been heated to 42oC.
5.Apply
the labeled probe to a prehybridized microarray slide and cover with a
22mm´60mm
polyethylene hydrophobic coverslip (PGC Scientific
Cat# 62-6504-06).
6.Place
the slide in a sealed hybridization chamber (Corning Costar Cat #2551),
add 20ml
of water to the chamber at the end of the slide.
7.Place
the sealed chamber in a 42oC water bath and incubate for 16-20
hours.
8.Remove
the array from the hybridization chamber, taking care not to disturb the
coverslip. 9.Place
the slide in a staining dish containing low-stringency wash buffer containing
1´SSC
and 0.2% SDS at 42oC. 10.Gently
remove the coverslip while the slide is in solution and agitate for 4 minutes. 11.Wash
the slide at high-stringency in a staining dish containing 0.1´SSC
and 0.2% SDS at room temperature, agitating for 4 minutes. 12.Wash
the slide in 0.1´SSC,
agitating for 4 minutes. 13.Allow
the slides to air dry.
|
|||||||||||||||||||||||||||||||||
III. Data Collection, Normalization, and AnalysisDifferential gene expression is assessed by scanning
the hybridized arrays using a confocal laser scanner capable of interrogating
both the Cy3- and Cy5-labeled probes and producing separate TIFF images
for each. As is the case with arraying robots, there are a number of manufacturers
that produce scanners capable of detecting Cy3 and Cy5 (see Table 4) and
most are planning to release instruments capable of detecting additional
dyes.
Slide Scanning
We
are currently using the
ScanArray 3000 produced by GSI Lumonics. This scanner uses red and green
Helium-Neon lasers operating at 633nm
and 543nm to excite Cy5 and Cy3,
respectively. Hybridized slides are scanned first in the Cy5 channel, and
then the Cy3 channel, as Cy5 is more susceptible to photodegradation than
Cy3. Data from each fluorescence channel is collected and stored as a separate
16-bit TIFF image. These images are analyzed to calculate the relative
expression levels of each gene and to identify differentially expressed
genes. The analysis process can be divided into two steps – image processing
and data analysis. Figure 5 shows a typical hybridization image produced
when things work well. The contrast in this image has been adjusted to
allow faint spots to be easily visualized. Important aspects of the hybridization
to note are the low level, uniform background and the good signal-to-noise
Image ProcessingImage processing involves three stages. First,
the spots representing the arrayed genes must be identified and distinguished
from spurious signals that can arise due to precipitated probe or other
hybridization artifacts or contaminants such as dust on the surface of
the slide. This task is simplified to a certain extent because the robotic
arraying systems used to construct the arrays produce a regular arrangement
of the spotted DNA fragments. However, variable intensities and uneven
slide backgrounds as well as some irregularities in the gridded arrays
complicate the problem slightly. Generally, problem of grid spot location
is coupled with estimation of the fluorescence background. For microarrays,
it is important the background be calculated locally for each spot, rather
than globally for the entire image as uneven background can often arise
during the hybridization process. The second step in analysis of the array
images is the estimation of background.
Following spot identification and local background
determination, the background-subtracted hybridization intensities for
each spot must be calculated. There are currently two schools of thought
regarding the calculation of intensities – the use of the median or the
mean intensity for each spot. As array analysis generally uses ratios of
measured Cy3 to Cy5 intensities to identify differentially expressed genes,
the mean and the integrated intensities are operationally equivalent. In
comparisons of intensities measured for normalization controls spiked into
the labeling reactions, we have found mean intensities to give more consistent
results and consequently we use these in subsequent calculations (V. Sharov
and J. Quackenbush, in preparation).
A
number of image processing software packages are available and are listed
in Table 5. We have developed a software package called TIGR_Spotfinder
for image processing (<http://www.tigr.org/softlab/>). TIGR_Spotfinder
uses a thresholding algorithm that separates spots from the background,
allowing a grid to be laid across the spots. Having found a grid, spots
are found within each grid element, local background is calculated, and
background-subtracted, integrated intensities are calculated in
both the Cy3 and Cy5 channels. Measured intensities are entered into the
Molecular Analysis of Gene Expression (MAGE) database, a Sybase relational
database specifically designed to capture gene expression data.
Data Normalization and AnalysisFollowing
image processing, the data generated for the arrayed genes must be further
analyzed before differentially expressed genes can be identified. The first
step in this process is the normalization of the relative fluorescence
intensities in each of the two scanned channels. Normalization is necessary
to adjust for differences in labeling and detection efficiencies for the
fluorescent labels and for differences in the quantity of starting RNA
from the two samples examined in the assay. These problems can cause a
shift in the average ratio of Cy5 to Cy3 and the intensities must be rescaled
before an experiment can be properly analyzed.
The normalization strategies that can be used
are based on some underlying assumptions regarding the data and the strategies
used for each experiment should be adjusted to reflect both the system
under study and the experimental design. The primary assumption is that
for either the entire collection of arrayed genes or some subset of the
genes such as housekeeping genes, or for some added set of controls, the
ratio of measured expression averaged over the set should be one.
Depending
on the experimental design, there are three useful approaches for calculating
normalization factors. The first simply uses total measured fluorescence
intensity. The assumption underlying this approach is that the total mass
of RNA labeled with either Cy3 or Cy5 is equal. While the intensity for
any one spot may be higher in one channel than the other, when averaged
over thousands of spots in the array, these fluctuations should average
out. Consequently, the total integrated intensity across all the spots
in the array should be equal for both channels. Alternatively, one could
add a number of controls in increasing but equimolar concentrations to
both the labeling reactions and the sum of the intensities for these spots
should be equal. A second approach uses linear regression analysis. For
closely related samples, one would expect many of the genes to be expressed
at nearly constant levels. Consequently, a scatterplot of the measured
Cy5 versus Cy3 intensities should have a slope of one. Measured intensities
for added equimolar controls should behave similarly. Under this assumption,
one can use regression analysis techniques to calculate the slope. This
is then used to rescale the data and adjust the slope to one. A third approach
has been described by Chen et al. (2). They assume that some subset
of housekeeping genes exists and that for these, the distribution of transcription
levels should have some mean value ? and standard deviation ? independent
of the sample. In this case, the ratio of measured Cy5 to Cy3 ratios for
these genes can be modeled and the mean of the ratio adjusted to 1. Chen
and collaborators describe an iterative procedure to achieve this normalization
and we have implemented their algorithm and a variation of it that uses
the entire data set, as well the total intensity and linear regression
normalization, into a data visualization and analysis tool called TIGR
ArrayViewer. TIGR ArrayViewer is freely available and can be obtained through <http://www.tigr.org/softlab/>.
In any normalization approach, care must be taken in handling genes expressed
at low levels. Statistical fluccuations in the measured levels can cause
a significant variation in the ratios that are calculated and inefficiencies
in labeling for either of the two dyes can cause these low intensity genes
to disappear from the arrays. Typically, we only use spots in the final
analysis where the intensities in both channels are two standard deviations
above background.
Following
normalization, data are typically analyzed to identify genes that are differentially
expressed. Most published studies have used a post-normalization cutoff
of two-fold up- or down-regulation to define differential expression; the
approach defined by Chen et al. (2) provides confidence intervals
that can be used to identify differentially expressed genes. In order to
separate genes that are truly differentially expressed from stochastic
changes, we typically conduct three independent microarray assays starting
from independent mRNA isolations and define differential expression based
on their consensus. Conclusion The
examination of gene expression using microarrays holds tremendous promise
for the identification of candidate genes involved in a variety of processes.
Indeed, the experiments that have been described to date have confirmed
known patterns of expression and provided information on genes of unknown
function. However, most applications have
to date only allowed the identification of genes differentially expressed
at significant levels. The true challenge, and the promise of this technique,
will be to use it to identify genes that are consistently up- or down-regulated
by 10 or 20% yet play significant roles in the development and progression
of disease. This will require the analysis of data from multiple experiments
and the correlation of patterns of gene expression with additional experimental
and clinical information. Recently a variety of techniques including hierarchical
clustering (3) and self-organizing maps (11)
have been applied to the analysis of microarray expression data across
multiple experiments. However, each of these depends on having reliable
and reproducible data from each microarray assay. The laboratory techniques
outlined here have allowed reproducible hybridization results such as those
shown in Figure 5. Although these protocols will likely continue to evolve,
we believe that they represent a reliable starting point for those beginning
microarray experimentation. |
|||||||||||||||||||||||||||||||||
|
Acknowledgements
This work was supported with funding
from the National Cancer Institute’s Cancer Genome Anatomy Project (R01
CA77049-01; PI: J. Quackenbush). The authors wish to thank
V. Sharov, A. Saeed, R.T. Cline, and S. Peterson for valuable comments
and contributions.
|
|||||||||||||||||||||||||||||||||
|
References 1.
Boguski, M.S., and Schuler, G.D. 1995. ESTablishing a human transcript
map. Nature Genet. 10:369-371. 2.
Chen, Y., Dougherty, E.R., and Bittner, M.L. 1997. Ratio-based decisions
and the quantitative analysis of cDNA microarray images. J. Biomed.
Optics 24:364-374 3.
Eisen, M.B., Spellman, P.T., Brown, P.O., Botstein, D. 1998. Cluster analysis
and display of genome-wide expression patterns. Proc. Natl. Acad. Sci.
USA 95:14863-14868. 4.
Eisen, M.B. and Brown, P.O. 1999. DNA arrays for analysis of gene expression.
Methods
Enzymol. 303:179-205. 5.
Liang, P., and Pardee, A.B. 1992. Differential display of eukaryotic messenger
RNA by means of the polymerase chain reaction. Science 257:967-971. 6.
Lockhart, D.J., Dong, H., Byrne, M.C., Follettie, M.T., Gallo, M.V., Chee,
M.S., Mittmann, M., Wang, C., Kobayashi, M., Horton, H., Brown,
E.L. 1996. Expression monitoring by hybridization to high-density oligonucleotide
arrays. Nature Biotechnol. 14:1675-1680. 7.
Quackenbush, J., Liang, F., Holt, I. Pertea, G., and Upton, J. 2000. The
TIGR Gene Indices: Reconstruction and Representation of Expressed Gene
Sequences. Nucleic Acids Res. 28:141-145. 8.
Schena, M., Shalon, D., Davis, R.W. and Brown, P.O. 1995. Quantitative
monitoring of gene expression patterns with complementary DNA microarray.
Science270:467-470. 9.
Schuler, G.D. 1997. Pieces of the puzzle: expressed sequence tags and the
catalog of human genes. J. Mol. Med. 75:694-698. 10.
Shalon, D., Smith, S.J. and Brown, P.O. 1996. A DNA microarray system for
analyzing complex DNA samples using two-color fluorescent probe hybridization.
Genome
Res. 6:639-645. 11. Tamayo, P., Slonim, D., Mesirov, J., Zhu, Q., Kitareewan, S., Dmitrovsky, E. Lander, E.S., and Golub, T.R. 1999. Interpreting patterns of gene expression with self-organzing maps: methods and application to hematopoietic differentiation. Proc. Natl. Acad. Sci. USA 96:2907-2912. 12.
Velculescu, V.E., Zhang, L., Vogelstein, B., and Kinzler, K.W. 1995. Serial
analysis of gene expression. Science270:484-487. |
|||||||||||||||||||||||||||||||||
|
National
Human Genome
Research
Institute
|
<http://www.nhgri.nih.gov/DIR/LCG/15K/HTML/protocol.html>
|
|
Stanford
University
|
<http://cmgm.stanford.edu/pbrown/protocols/index.html>
|
|
Telechem
|
<http://arrayit.com/DNA-Microarray-Protocols/>
|
|
University
of Pennsylvania
|
<http://genomics.med.upenn.edu/vcheung/protocols.htm>
|
|
The
Institute for Genomic Research
|
<http://www.tigr.org/tdb/microarray>
|
Table
1. Protocols for microarray analysis.
|
Beecher
Instruments
|
<http://www.beecherinstruments.com>
|
|
BioRobotics
|
<http://www.BioRobotics.com/>
|
|
Cartesian
Technologies
|
<http://www.cartesiantech.com/>
|
|
Engineering
Services
|
<http://www.ESIT.com/>
|
|
Genetic
Microsystems
|
<http://www.geneticmicro.com>
|
|
Genetix
|
<http://www.genetix.co.uk/>
|
|
Gene
Machines
|
<http://www.genemachines.com>
|
|
Genomic
Solutions
|
<http://www.genomicsolutions.com/>
|
|
Intelligent
Automation Systems
|
<http://www.ias.com>
|
|
Packard
|
<http://www.packardinst.com/>
|
Table
2. Manufacturers of microarray spotting robots.
|
Amersham
Pharmacia Biotech
|
<http://www.apbiotech.com>
|
|
Corning
Costar
|
<http://www.cmt.corning.com>
|
|
Telechem
|
<http://www.arrayit.com/>
|
|
Axon
|
<http://www.axon.com>
|
|
Beecher
Instruments
|
<http://www.beecherinstruments.com>
|
|
GSI
Lumonics
|
<http://www.gsilumonics.com>
|
|
Genetic
Microsystems
|
<http://www.geneticmicro.com>
|
|
Genomic
Solutions
|
<http://www.genomicsolutions.com/>
|
|
Molecular
Dynamics
|
<http://www.mdyn.com>
|
|
Virtek
|
<http://www.virtek.ca/>
|
Table
4. Manufacturers of microarray scanners.
|
BioDiscovery
|
<http://www.biodiscovery.com/>
|
|
Imaging
Research
|
<http://imaging.brocku.ca/Arrayvision.html>
|
|
National
Human Genome
Research
Institute
|
<http://www.nhgri.nih.gov/DIR/LCG/15K/HTML/img_analysis.html>
|
|
Stanford
University
|
<http://rana.Stanford.EDU/software/>
|
|
The
Institute for Genomic Research
|
<http://www.tigr.org/softlab/>
|
Table
5. Microarray image processing software sources.
|
BioDiscovery
|
<http://www.biodiscovery.com/>
|
|
Silicon
Genetics
|
<http://www.sigenetics.com/>
|
|
Spotfire
|
<http://www.spotfire.com/>
|
|
Stanford
University
|
<http://rana.Stanford.EDU/software/>
|
|
TIGR
|
<http://www.tigr.org/softlab/>
|
Table
6. Microarray data analysis sources.
Figure 2. Effects
of various spotting buffers and DNA clean-up protocols on DNA binding and
hybridization using Corning CMT-GAPS™ aminosilane coated slides. This
false color image was generated by spotting identical samples in adjacent
rows and hybridizing with a labeled mRNA probe; red lines separate paired
rows. Paired rows 1-3 and 5-7 contain samples spotted using either 50%
DMSO or 3´SSC
as a spotting buffer. Comparing spots vertically adjacent to each other,
it is clear that spotting with DMSO allows hybridization with significantly
higher affinity than does spotting with SSC. In our evaluation, DMSO consistently
gives 1.5-fold or greater hybridization intensities with 10% fewer “drop
out” spots. Paired rows 4 and 8 show the effects of different clean-up
protocols on DNA binding and hybridization. The glass-filter method described
in the text gives visibly better results than does simple ethanol precipitation.
Figure 3. Effect
of temperature and humidity on slide morphology
Printing on this slide
began at 72ºF (22.2ºC) and 45-50% relative humidity, which we
had determined to be optimal. Under these conditions, the spots have a
uniform appearance. As printing progressed, the temperature was reduced
to approximately 62ºF (16.7ºC), resulting in smaller, less distinct
spots. Temperature and humidity were then increased. The rows of large
spots in the center of the slide (rows 13-15) were printed at 80ºF
(26.7ºC) and 80-85% relative humidity. As the temperature and humidity
were decreased once more, optimal conditions were again achieved. By correlating
data from a chart recorder with the spot number, we were able to determine
the conditions that subsequently gave the best hybridization performance.
Representative temperature and humidity levels are shown. The cDNA clones
in this array were hybridized with a Cy3 labeled vector-specific probe.
Figure 4. Comparison of hybridization probes
made using total and poly(A+) RNA.
Corresponding areas of cDNA microarrays containing
7,200 elements, each hybridized with labeled probe prepared from 20?g of
total RNA, 4?g of total RNA, and 1.5?g of poly(A+) selected
RNA. Note that the relative hybridization intensities are similar for each
of the total RNA samples and that both are slightly greater than for the
poly(A+) hybridized sample. In our experience, 4-10?g of total
RNA gives consistently high quality hybridization results. All images were
obtained using the same laser and PMT settings during scanning and are
displayed using the same parameters.
Figure 5. Hybridization of mRNA to a portion of a 19,200 element array. Using the protocols described in this manuscript, 10mg of RNA extracted from related human colon carcinoma test and reference cell-lines (KM12L4A and KM12C respectively) was reverse transcribed and labeled with Cy-5 dUTP and Cy-3 dUTP, respectively. These were then hybridized to a microarray containing 19,200 distinct human cDNA clones. The contrast on the image has been adjusted to allow the majority of the spots in the array to be easily visualized. The protocols outlined in this manuscript consistently provide hybridization results similar to this.