REPRODUCTION A guide to issues in microarray analysis: application to endometrial biology

A guide to issues in microarray analysis: application
to endometrial biology
Christine A White1,2 and Lois A Salamonsen1
Prince Henry’s Institute of Medical Research, PO Box 5152, Clayton, Victoria, 3168, Australia and 2Dept of
Obstetrics & Gynaecology, Monash University, Clayton, Victoria, 3168, Australia
Correspondence should be addressed to C A White; Email: [email protected]
Within the last decade, the development of DNA microarray technology has enabled the simultaneous measurement of thousands of gene transcripts in a biological sample. Conducting a microarray study is a multi-step process; starting with a welldefined biological question, moving through experimental design, target RNA preparation, microarray hybridisation, image
acquisition and data analysis – finishing with a biological interpretation requiring further study. Advances continue to be
made in microarray quality and methods of statistical analysis, improving the reliability and therefore appeal of microarray
analysis for a wide range of biological questions. The purpose of this review is to provide both an introduction to microarray
methodology, as well as a practical guide to the use of microarrays for gene expression analysis, using endometrial biology as
an example of the applications of this technology. While recommendations are based on previous experience in our laboratory, this review also summarises the methods currently considered to be best practice in the field.
Reproduction (2005) 130 1–13
Principles of microarray analysis
As for many other techniques used in molecular biology,
microarrays rely on the complementarity of the DNA
duplex, i.e. that the two strands will always reassemble
with base pairing A to T and C to G. In addition, singlestranded DNA will bind strongly to a solid support, where
it is available for hybridisation with complementary DNA
(cDNA). In a DNA microarray experiment, many genespecific ‘probes’ are immobilised on a solid support
(usually nylon membrane or glass) and the array is
exposed to labelled cDNA ‘targets’ derived from one or
more biological samples (Schena et al. 1995). Nylon
membrane arrays are normally hybridised with a single
cDNA population, labelled with a radioactive or chemiluminescent tag, so that the intensity of the signal generated
by each bound probe indicates the abundance of that
transcript in the sample. Transcript abundance in different
samples (e.g. treatment and control) can then be compared across serial or parallel array hybridisations. The
advantage of the glass DNA microarray is that cDNA from
two or more biological samples can be labelled with
different fluorescent dyes and competitively hybridised, so
that the relative abundance of gene transcripts can be
determined by the fluorescent signal obtained (Figure 1).
This review will focus on the techniques and data analysis
associated with the two most common cDNA microarray
q 2005 Society for Reproduction and Fertility
ISSN 1470–1626 (paper) 1741–7899 (online)
platforms; dual colour fluorescence on glass and radioactively labelled nylon membranes. A summary of some of
the key issues involved in microarray analysis is provided
in Table 1.
Choice of microarray
As most laboratories are not equipped with robotic printers, the investigator must usually obtain microarrays from
commercial or other sources. There are a range of mouse,
rat and human microarrays available, and they fall into
two broad categories; cDNA and high-density synthetic
oligonucleotide (reviewed in Barrett & Kawasaki 2003).
As described above, cDNA microarrays can be based on
either nylon membrane or dual colour fluorescence on
glass. Both formats of cDNA microarray involve the deposition of purified and/or PCR-amplified DNA in solution
onto the solid support at defined locations using a robotic
printer (Figure 1). The quality of the microarray is therefore dependent on the performance of the print tips,
which must deliver reproducible volumes and uniform
spot sizes to enable effective data analysis. The main
advantages of cDNA microarrays are their relatively low
cost (, US $150 per slide) and greater flexibility in terms
of producing and spotting custom-made clone sets.
Manufacturers of high-density synthetic oligonucleotide
microarrays, such as Affymetrix (Santa Clara, CA, USA;
DOI: 10.1530/rep.1.00685
Online version via
C A White and L A Salamonsen
Table 1 Some key issues involved in microarray analysis
† Consider the biological question(s)
and the ability to achieve
statistical significance
† Seek expert statistical advice during the
early planning stages
† Microarray experiments have multiple
sources of variation and must be
carefully controlled
† Biological and technical replication
are essential
† Sample pooling should be avoided if
accurate sample synchronisation is not
† Microarray analysis of purified cells will
only reveal genes expressed by these cells,
but removal from the in vivo microenvironment may alter gene expression
† There are limitations in the use of both
whole tissue and purified cells, which
may necessitate the use of microdissection
and RNA amplification techniques
† When using clinical samples, detailed
patient history and tissue histopathology
are critical to the interpretation of gene
expression profiles
Target RNA preparation † The quality of the target RNA is one of the
most important factors in the success or
failure of a microarray experiment
Data analysis
† While critical to the outcome of a
microarray experiment, statistical analysis
of microarray data is not well understood
by many biologists and expert advice
should be sought
Data validation
† The biomedical research community
does not yet accept that microarray data
can stand alone without independent
† The investigator must decide which genes
to examine further, and those with
larger fold changes and statistical
significance are often the best candidates
† To describe a biological event or system,
gene expression data obtained by
microarray analysis must be extended to
the study of protein products
Experimental design, use photolithography and
solid-phase DNA synthesis to generate synthetic 25 base
polynucleotides (25mers) directly on the glass surface
(Lipschutz et al. 1999). Each gene is represented by 11 to
20 different 25 mers, in ‘perfect match’ or ‘mismatch’
sequence pairs. Probes can be generated representing a
unique part of a gene transcript, enabling discrimination
between closely related genes or splice variants and the
‘mismatch’ sequences provide an internal control for
every gene. Longer oligonucleotide (50 to 100 mers)
microarrays are also available, which provide even greater
hybridisation specificity (Barrett & Kawasaki 2003). Oligonucleotide microarrays are hybridised with a single fluorescently labelled sample and gene expression in different
samples compared across multiple microarrays. The main
disadvantage of these microarrays is their high cost (up to
Reproduction (2005) 130 1–13
US$800 per slide), so their appeal is likely to increase as
they become more economical.
Commercially-available microarrays are printed or synthesised with a particular clone set (reviewed in Bowtell
1999) and these differ in the proportion of known genes
and expressed sequence tags (ESTs). Some ESTs correspond
to a segment of a known gene, but most represent partially
sequenced novel genes. A number of groups including
those at Merck, Washington University, the IMAGE (Integrated Analysis of Genomes and their Expression) Consortium and the Cancer Genome Anatomy Project (CGAP)
have been responsible for sequencing over one million
human ESTs (Bowtell 1999). The online database dbEST (a
division of GenBank;
index.html) houses all these EST sequences and the automated process known as UniGene assigns overlapping
sequences to a single cluster, which may or may not have a
known identity (
index.html). As the human genome sequence is effectively
complete (Venter et al. 2001) it is expected that all ESTs
will progressively be assigned an identity. Until then, careful consideration should be given to whether identifying
differentially expressed ESTs is a priority in any particular
microarray experiment. If not, then using a more tailored
array specific to a cellular process or pathway may be
more appropriate and cost-effective.
Experimental design
The most important considerations in microarray experimental design are the biological question under study and
the ability to achieve statistical significance (reviewed in
Churchill 2002, Kerr & Churchill 2001, Smyth et al. 2003,
Yang & Speed 2002). Answering the biological question
may require identification of downstream target genes, and
may also involve a time course. Designing a microarray
experiment with sufficient statistical power requires input
from a statistician or bioinformatician with experience in
microarray technology. The right statistical advice during
early planning can save vast amounts of time and money.
The reason for their complexity is that microarray experiments have multiple sources of variation, each of which
must be considered in the experimental design (reviewed in
Churchill 2002, Chen et al. 2004). Firstly, there is biological
variation between animals or patients. Secondly, technical
variation arises from the RNA extraction, reverse transcription, label incorporation and hybridisation steps. Thirdly,
measurement errors occur due to differences in hybridisation efficiencies between spots, between different print-tip
groups across the array and between slides in the same and
different print runs. Technical variation and measurement
error can also interact. For example, the scanning properties
of the fluorescent dyes can vary with the spot intensity and
spatial position on the slide (Smyth & Speed 2003).
A conventional power analysis requires prior knowledge
of the variance of individual measurements, the
Guide to issues in microarray analysis
Figure 1 Dual colour fluorescence cDNA microarray analysis. Clones of interest (probes) are amplified by PCR and printed onto treated glass
slides using a robotic printer. Total RNA samples extracted from treated and control cells/tissues (targets) are reverse transcribed and labelled
with either Cy3 (green) or Cy5 (red). The samples are combined and competitively hybridised to the microarray under stringent conditions.
Following washing to remove non-hybridised target, laser excitation is applied and the emissions measured in each colour channel. Specialised
software is used to attach gene names, fluorescence intensity values and intensity ratios to each spot, which are then exported for advanced
statistical analysis to identify differentially expressed genes.
magnitude of the effect to be detected, the acceptable
false-positive rate and the desired ‘power’ of the calculation; that is, the probability of detecting an effect of the
specified or greater magnitude (Yang & Speed 2002, Yang
et al. 2003, Chen et al. 2004). In a microarray experiment,
two of these components are unknown; both the variance
of the expression ratio measurements and the magnitude
of the effects of interest will be different for every gene
on the microarray. To overcome this, power calculations
can be performed using the median variance across all of
the genes in a previous microarray hybridisation (Yang &
Speed 2002).
It is critical that microarray experiments are carefully
controlled, particularly when using dual colour fluorescence microarrays in which the endpoint is a ratio of
expression between two or more samples. As in any
experiment, treatment controls must be carefully incorporated into the study design. To ensure that there is only one
source of experimental variation, consistency must also be
applied to tissue collection, processing and RNA extraction, as well as the microarray hybridisations. Even with a
single variable, such as a differentiation stimulus, it is
possible to end up comparing cells or tissues in completely different physiological states. In this situation, differentially expressed genes will likely be the consequence,
rather than the cause, of the differences in phenotype.
This problem can be minimised by using carefully controlled inducible systems and examining early rather than
later time points.
When an experiment involves comparisons across
multiple dual colour fluorescence microarrays, there are
a number of possible design matrices (Figure 2). Hybridising an appropriate reference sample to each microarray
(Figure 2A) can provide a consistent control across multiple slides. The ideal reference contains all possible
mRNA transcripts present in the experimental samples, so
that fluorescence ratio measurements less than zero cannot occur. This is usually achieved by generating a pool of
RNA from multiple samples of the tissue or cell type
under study. Another approach is to create a reference
mixture of all the PCR products spotted on the microarray
(Sterrenburg et al. 2002). Although there are some advantages to using a reference sample, it is more precise and
economical to make the critical comparisons directly on
the same microarray (Figure 2B; Kerr & Churchill 2001,
Churchill 2002). As a reference design requires more complicated statistical analysis, it should only be used for a
well-defined purpose. It may also be useful to use a common reference if a large number of experimental samples
are to be collected and analysed over a long period of
time. A saturated design (Figure 2C) may be used when an
experiment has more than two treatments, and all comparisons are of interest in answering the biological question. The efficiency of time course experiments can be
maximised by a loop design (Figure 2D). Regardless of the
design matrix, care should be taken that all pairings are
biologically relevant and controlled. For example, pairs
could be wild type and knockout littermates, or isolated
cells from the same endometrial biopsy with and without
Replication is critical in a microarray experiment, as it
enables the data to be effectively analysed using formal
Reproduction (2005) 130 1–13
C A White and L A Salamonsen
Time 0
Time 1
Time 2
Time 3
Figure 2 Microarray experimental designs (adapted from Churchill 2002, Smyth et al. 2003). Letters refer to different treatments/genotypes and
subscripts indicate biological replicates. Each arrow represents one microarray, with the arrow pointing away from the Cy3 (green) labelled
sample and towards the Cy5 (red) labelled sample. Double arrows indicate dye-swap pairs. A, Indirect comparison with a common reference;
B, Direct comparison; C, Saturated design and D, Time course experiment loop design.
statistical methods, and the results of that analysis to be
broadly applicable to the sampled population. Given the
sources of variation described above, replication should
be both biological and technical. Biological replication is
the use of RNA samples from multiple animals or patients.
Technical replication (i.e. repeated measures) includes the
presence of duplicate DNA probes on the microarray and
hybridisation of the same RNA sample on multiple microarrays. As biological variability is usually greater than
technical variability, the hybridisation of independent
RNA samples should be prioritised. Replication allows the
investigator to identify and remove false-positives and
false-negatives, so only reproducible data are considered
for further analysis. A single hybridisation may be justified
if the aim is to generate hypotheses for further testing, but
advanced statistical analysis will be more productive on 3
or more replicates per treatment (reviewed in Sasik et al.
2004). In the case of dual colour fluorescence microarrays, dye swap replicates should also be performed to
account for unequal dye incorporation and quenching
(Figure 2). To avoid or minimise bias, the assignment of
dye labels should be randomised.
Reproduction (2005) 130 1–13
Pooling samples is often considered when RNA is in
limited supply, or to minimise the effects of biological
variation. In addition, it has been demonstrated that pooling RNA from an increased number of subjects can reduce
the number of microarrays required, without any loss of
precision (Kendziorski et al. 2003, Peng et al. 2003).
Sample pooling should be avoided when it is not possible
to accurately synchronise the samples (Sasik et al. 2004).
For example, when using pseudopregnant mice on the
same day after vaginal plug detection, there is likely to
be variation in the time at which mating occurred and
therefore in the physiological state of the uterus. In this
situation, it is better to maintain sample independence
and find common gene expression features at the data
analysis stage.
Careful consideration should be given to the use of
whole tissue or purified homogeneous cell populations in
a microarray study. The advantage of using whole tissue is
that there is a greater amount of RNA available for technical replicates and subsequent validation studies (see ‘Data
validation’ below). It is important to consider, however,
that tissues contain a range of different cell types. Whole
endometrial biopsies, for example, contain luminal
Guide to issues in microarray analysis
and glandular epithelium, non-decidualised and decidualised stroma, endothelial cells, smooth muscle cells and
leukocytes. If the cell type of interest makes up a small
proportion of the total tissue, whole tissue gene expression
data may not be particularly informative. Microarray analysis of purified cells will only reveal genes expressed by
those cells, but they will also have been removed from
their in vivo microenvironment and cultured under conditions which are likely to alter gene expression. The limitations of both approaches need to be balanced against
the aims of the experiment and perhaps additional technologies such as laser capture microdissection (EmmertBuck et al. 1996) considered. New RNA amplification
methods (see ‘Target RNA preparation’ below) have
improved the feasibility of using microdissected tissue for
gene expression studies and this approach has the advantage of maintaining close to in vivo cellular context.
The parallel measurement of other biological parameters
can be used to assist the interpretation of microarray data.
For example, variation in prolactin secretion levels from
different preparations of decidualising human endometrial
stromal cells may correlate with variations in the expression
levels of other genes. When using clinical samples, patient
history and tissue histopathology are also critical in the
final interpretation of gene expression profiles.
Target RNA preparation
The target in microarray analysis is a labelled population
of cDNAs representing the mRNA repertoire of a cell type
or tissue. The purity of the extracted total RNA is one of
the most critical factors in the success or failure of a
microarray experiment. Residual chloroform, phenol or
ethanol from the extraction process can interfere in the
efficiency of reverse transcription to cDNA, and other
contaminants such as cellular protein, lipids and carbohydrates can cause non-specific binding of fluorescent
cDNAs to the glass surface (Duggan et al. 1999). For most
tissues, including human endometrium, optimal quality
total RNA can be extracted using TRIzol Reagent (Invitrogen, Carlsbad, CA, USA), followed by column purification
using the RNeasy Minikit (Qiagen, Hilden, Germany).
Highly pure total RNA can be extracted from isolated
cells using the RNeasy kit alone. Both of these methods
should be followed by removal of genomic DNA. It is recommended that a number of different RNA extraction
methods be compared for the tissue or cell type of interest, to maximise RNA quality and yield.
Particularly in microarray experiments, it is critical to
accurately measure RNA quality and quantity, to minimise
variation and therefore improve labelling and hybridisation consistency. The standard UV spectrophotometer is
useful for an initial estimate of RNA quality and quantity.
Optical density (OD) can be measured at 230, 260 and
280 nm and RNA purity considered acceptable at values
of OD260/OD280 1.8 –2.0 and OD230/OD280 , 1.0. Agarose gel electrophoresis may then be used to further
confirm RNA integrity. However, other systems such as
the RiboGreen Assay (Molecular Probes, Eugene, OR,
USA) or the Agilent 2100 Bioanalyser (Agilent Technologies UK Ltd, Cheadle, Cheshire, UK) are more sensitive
and accurate. The Agilent Bioanalyser requires only 50 –
500 ng of total RNA and produces a detailed electrophoretogram which will reveal any RNA degradation or genomic DNA contamination.
The amount of RNA required per hybridisation is the
greatest limitation to the use of this technology, particularly when the tissue of interest is in limited supply, or
when using isolated cell populations. Until very recently,
it was recommended that 50 –200 mg total RNA per
sample be used for each hybridisation to generate a sufficient signal (Duggan et al. 1999). However, improved
RNA purification, fluorescent labelling methods and
hybridisation conditions have reduced this requirement to
5 – 10 mg for both glass and nylon membrane arrays. As
this amount of RNA may still be difficult to obtain in some
systems, a number of different RNA amplification
approaches have been developed. The most commonly
used method is T7 polymerase in vitro transcription (IVT;
van Gelder et al. 1990). While this method can significantly reduce the RNA requirement for each hybridisation,
it is also expensive, time-consuming and labour-intensive.
In addition, multiple rounds of amplification may be
required, which decreases the linearity of the amplification and may result in a cDNA target which is no longer
representative of the original sample (Petalidis et al.
2003). Newly developed PCR-based cDNA amplification
techniques can decrease the amount of starting total RNA
required to 200 ng, while maintaining amplification linearity (Petalidis et al. 2003). This method vastly improves the
feasibility of glass microarray studies on clinical samples.
Microarray hybridisation
Effective hybridisation of the target to the microarray is
essential in obtaining high quality data. Coverslips with
raised Teflon edging (Lifter Slips; Erie Scientific, Portsmouth, NH, USA) are a useful way of ensuring the full
hybridisation volume maintains contact with the arrayed
probes. Hybridisation chambers, such as those available
from Corning (Corning, NY, USA), allow the microarrays
to be submerged in water of a set temperature, greatly
improving hybridisation consistency across the slide.
Commercially available buffers such as ExpressHyb
Hybridisation Solution (Clontech, Palo Alto, CA, USA) are
very effective for nylon membrane microarray hybridisation (Evans et al. 2003). The use of non-specific blocking
agents including Cot-1 DNA, polyadenylic acid and salmon sperm DNA ensures that the signal detected from
each spot is specific to the particular probe sequence and
background is minimised. Most microarrays include
positive and negative (printing buffer alone) hybridisation
controls, as well as spiked controls and RT efficacy controls. Internal control spots can be useful for assessing
Reproduction (2005) 130 1–13
C A White and L A Salamonsen
data quality, but it is not recommended that they be used
to standardise the data, as a housekeeping gene would be
used in quantitative RT-PCR (Churchill 2002). Optimal target quality and hybridisation conditions will ensure that
the maximum microarray sensitivity is achieved, allowing
even very low abundance genes to be detected and differential expression determined.
Image acquisition
Nylon membrane cDNA microarrays hybridised with a
radioactive 33P-labelled probe are scanned with a phosphorimager screen. Commercially available software such
as Imagene (BioDiscovery, CA, USA) is then used to align
the specific grid of arrayed DNA spots and quantify the
signal intensity at each location. Scanners and software
used for fluorescent image acquisition from glass cDNA
microarrays are more complex, and there are many different systems available (see
The Axon GenePix 4000B scanner has two lasers which
simultaneously excite a small region of the glass surface
(, 100 mm2), at a focal plane pre-set by the user (2 50 mm
to þ200 mm relative to the slide). The entire image is
obtained by moving the laser lens across the glass slide.
Light emitted at the wavelengths of the fluorescent labels
(532 nm for Cy3 and 635 nm for Cy5) is converted to an
electrical signal with a photomultiplier tube (PMT). These
signals are then displayed as a 16-bit tagged image file format (TIFF) image (Cy5 coloured red and Cy3 coloured
green) and given numerical values. Scanning and processing images from glass microarrays requires the investigator to perform a number of manual tasks, each of which
demands a high level of technical knowledge. The PMT of
the GenePix scanner needs to be adjusted to maximise its
dynamic range (0 to 65 536 pixels), to prevent signal saturation and balance the intensities of the two excitation
wavelengths (Forster et al. 2003). Both red and green foreground and background intensities are measured for each
spot. The foreground intensity for the spot is given as the
mean intensity of all foreground pixels and this is assumed
to be proportional to the number of complementary
mRNA molecules present in the sample. The method of
background calculation differs between software
packages, but local background correction is preferred
over global methods (Kim et al. 2002). The default method
used by GenePix software is local background subtraction,
in which a different background value is computed for
each individual spot and the median background pixel
intensity is used for correction purposes.
Following image acquisition, the user must align an
appropriate grid containing spot identities to the image, as
well as identify artefacts of the hybridisation process so
that they can be removed from subsequent analyses.
As the settings used for background calculation,
background thresholds and ratio calculation can greatly
influence data quality, the investigator should be aware of
the implications of using each of the different methods.
Reproduction (2005) 130 1–13
Importantly, the methods used for image acquisition can
be optimised from slide to slide, but those used for image
quantification should be identical for all slides in the
experiment (Forster et al. 2003).
Data analysis
Following image acquisition and conversion of the image
into spot intensity and intensity ratio measures, this large
body of data must be stored in spreadsheet form for
further analysis. Depending on the number of microarrays
processed, a high capacity system such as an SQL server
may be required. Data analysis is perhaps the most critical
aspect of a microarray experiment and the least understood by the majority of biologists. Many different analytical approaches have been developed to achieve sensitivity
in detecting gene changes while also providing a measure
of statistical significance and likelihood of error. Measures
of statistical significance can be made for a single microarray, or across multiple biological and technical replicate
microarrays. The complexity of the data and statistical
analysis requires the use of sophisticated visualisation and
analysis software.
Graphical displays are useful in determining the overall
success of a microarray experiment (Smyth et al. 2003).
The red-green image produced during scanning can detect
any problems with colour balance, hybridisation, spatial
effects, spot quality or artefacts such as scratches and
dust. The original red-green image can also be used to
check differential expression of a particular gene. Before
they are plotted or analysed, the raw intensity data are
always log-transformed (log2) to spread the values more
evenly across the scale from 0 to 65 535 pixels. If any
negative values for red (R) or green (G) foreground intensity have arisen due to high spot background, these will
be removed from the analysis on a log scale. Using these
log-transformed values, an informative visualisation tool is
the MA-plot (Dudoit et al. 2002b). This scatterplot has
M-values (R/G ratio log-transformed to M ¼ log2R/G) on
the vertical axis and A-values (spot intensity expressed as
A ¼ log2 ðR £ GÞ) on the horizontal axis. Particularly
when using large microarrays with thousands of probes,
the majority of genes should not be differentially
expressed. An MA-plot of good quality microarray data
should therefore have an elongated comet shape centred
around M ¼ 0 (i.e. equal red and green intensities over a
wide intensity range). As well as helping to identify spot
artefacts and intensity-dependent patterns, MA-plots can
be used to display the effects of normalisation on the data
(Smyth et al. 2003).
Normalisation is essential in microarray experiments to
adjust the data for systematic non-biological effects arising
from technical variation and measurement error (see
‘Experimental design’ above). The aim of normalisation is
to remove the effect of this ‘noise’ from the data, while
Guide to issues in microarray analysis
still maintaining the ability to detect significantly differentially expressed genes. When using multiple nylon membrane or other single sample microarrays, each of the
arrays must be paired with another and normalised or
‘scaled’ to its pair (reviewed in Evans et al. 2003). Dual
colour fluorescence microarrays require normalisation to
account for differences between microarrays, print-tips
groups and fluorescent dye channels (reviewed in Smyth
& Speed 2003). There is no universally accepted method
of microarray data normalisation, and a description and
comparison of all available methods is beyond the scope
of this review. Overall, the literature supports the use of
intensity-dependent normalisation methods, such as printtip loess (local weighted regression) normalisation (Dudoit
et al. 2002b, Yang et al. 2002, Park et al. 2003, Smyth &
Speed 2003). This method is capable of removing biases
without altering the structure of the data. Essentially, printtip loess normalisation corrects the M-values (log2R/G
ratios) for non-biological spatial and intensity effects.
Statistical analysis
Clustering was one of the first methods used to impose
order on microarray data (Eisen et al. 1998). This method
involves grouping genes on the basis of similar expression
patterns, with the assumption that each cluster of genes is
co-ordinately regulated, perhaps as part of the same signaling pathway. Clustering can be useful in assigning
potential functions to unidentified genes and ESTs, which
can then be tested in further studies. Related methods
such as supervised clustering, principle component analysis, self-organising maps and linear discriminant analysis
are also widely used to discover patterns of gene
expression common to a particular physiological state.
The aim of a microarray experiment is usually to identify differentially expressed genes, with a measure of statistical significance (reviewed in Dudoit et al. 2002b, Cui
& Churchill 2003). Most microarray experiments are
designed with only one categorical factor (eg. treatment
or genotype), so the statistical analysis is based on the
paired t-test. Experiments with multiple categorical factors
(eg. genotype and time) require methods based on the
analysis of variance (ANOVA). Once the data are appropriately normalised, it is common practice to consider a
univariate testing problem for each gene and calculate
t-statistics (Dudoit et al. 2002b). The t-statistic tests the
null hypothesis of equal mean expression levels in the two
samples (e.g. treatment and control). Another useful indicator of differential expression is the B-statistic (Lonnstedt
& Speed 2002), which is an estimate of the odds that the
gene is differentially expressed. The challenge in assigning
statistical significance to a differentially expressed gene is
that the often thousands of genes on a microarray result in
a high level of multiple testing. Determining the false
discovery rate is the most powerful method of controlling
for multiple testing (Tusher et al. 2001), but this can also
be achieved using adjusted P values (Dudoit et al. 2002b).
Time course experiments require even more specialised
statistical analysis (Cui & Churchill 2003) and should only
be conducted if the primary biological question is one of
time dependence.
Just as diagnostic MA-plots can be invaluable for visualising trends in raw and normalised data, plots of values
obtained during statistical analysis are also useful. Both
fold change (difference in gene abundance between two
samples) and significance measures can be represented
graphically in a ‘volcano plot’ (Cui & Churchill 2003),
with the log odds of differential expression on the vertical
axis and the mean M-value (log2R/G ratio) on the horizontal axis. Genes with statistically significant differential
expression will appear above a horizontal threshold line
and those with large fold changes (up- or downregulated)
will lie to the far left or right. Differentially expressed
genes identified by the B-statistic will appear in the upper
left or right quadrants.
There are many different software packages available
for performing normalisation, statistical analysis and visualisation with single and dual sample microarrays. Some
of the more widely used packages include Cyber-T (Baldi
& Long 2001), SAM (Tusher et al. 2001), BRB-ArrayTools
(, QVALUE (Storey
& Tibshirani 2003) and Focus (Cole et al. 2003). The statistical language R (Ihaka & Gentleman 1996, http://www. has also been used successfully for the analysis of microarray data (Dudoit et al. 2002a) and indeed
many of the other packages are based on R commands.
Bioconductor ( provides a
more user-friendly interface for the R statistical language.
Although they are often easier for biologists to use, care
should be exercised in the choice of commercially available software packages. Some are excellent for data visualisation and normalisation, but cannot assign measures of
statistical significance within and across multiple microarrays, or do not handle time course data.
Data validation
The biomedical research community does not yet accept
that microarray data can stand alone, without independent
validation (reviewed in Rockett & Hellmann 2004). There
are a number of reasons for this caution, including the
relatively recent development of the technology, the lack
of standard operating procedures and the potential for
errors to exist in the data (Knight 2001, Kothapalli et al.
2002). Hybridisation errors may occur due to crosshybridisation between transcripts of high homology and
data may be misleading if mis-annotation of probe
sequences has occurred. Even in well-maintained clone
sets, it is estimated that 1 –5% of clones do not contain
the correct sequence (Knight 2001). In addition, as the
statistical tests used for microarray data analysis are yet to
be standardised, often several methods are used and the
resulting data requires further validation. Best practice in
microarray analysis can achieve less than 5 –10%
Reproduction (2005) 130 1–13
C A White and L A Salamonsen
variation in signal intensity from replicate probes on the
same microarray, and around 10 –30% variation between
corresponding probe signal intensities on different microarrays (reviewed in Stears et al. 2003). Despite this, there
is an expectation that an additional mRNA quantification
method will be used to confirm the differential expression
of the genes of interest (Firestein & Pisetsky 2002).
The first important task for the investigator is to decide
which genes to investigate further. From experience, genes
displaying a large fold change (. 2) and statistical significance are the best candidates for validation. Before
embarking on additional studies, it is good practice to
review the primary red-green image data to confirm differential expression of these genes, and the spotted DNA
sequence may also be checked for correct annotation.
Comparing data with that obtained from other microarray
studies on the same system can also provide ‘in silico’
validation and increase confidence in the data set as a
whole (Chuaqui et al. 2002).
Quantitative real-time RT-PCR is commonly used to
confirm mRNA levels, as it has higher sensitivity and
lower RNA requirements than Northern blot. Previous
studies have demonstrated that genes with relatively high
expression and at least 2-fold regulation are likely to be
validated using real-time RT-PCR (Rajeevan et al. 2001).
The advantage of Northern blot and RNase protection
assay is that they provide a quantitative measure as well
as reveal the number and size of transcripts detected by
the particular spotted DNA sequence. Quantitative data
obtained with microarray and Northern blot are comparable, with Northern blot slightly more sensitive in
detecting differential expression compared with microarray (Taniguchi et al. 2001). In complex tissues such as the
endometrium, defining the cellular localisation of mRNA
expression using in situ hybridisation can provide important functional information. As it is almost impossible to
differentiate between primary and secondary gene
expression effects in microarray data, further testing may
be required to define the molecular interactions
While mRNA reflects the functional state of the cell, it
is the proteins which ultimately carry out the instructions
of the genome. Translation of mRNA into protein may be
controlled independently of transcription and proteins
may undergo post-translational modifications that alter
their function. To describe a biological event or system,
therefore, gene expression data obtained by microarray
analysis must be extended to the study of protein products. Particularly if target RNA has been prepared from
whole tissue, characterising the cellular distribution of the
corresponding protein by immunostaining or tissue array
is critical to understanding the function of a gene. Protein
quantification by Western blot or ELISA will indicate
whether transcription and translation are co-ordinately
Defining the functions of differentially expressed genes
may be considered the ultimate validation of microarray
Reproduction (2005) 130 1–13
data. Functional studies may include in vitro experiments
using dominant-negative mutants or RNA interference, or
in vivo experiments using antisense morpholino oligonucleotides, knockout or conditional knockout technologies.
Though the experiments may be carried out some time
later, each level of data validation (mRNA, protein and
function) should be considered at the microarray experimental design stage, to allow additional controlled
samples to be obtained.
Endometrial gene expression analysis
In the last few years, both cDNA and oligonucleotide
microarray technology have been successfully applied to
the study of endometrial gene expression (reviewed in
Giudice 2003). The endometrium is a uniquely dynamic
tissue, with the capacity to undergo dramatic remodelling
in response to cyclic variations in steroid hormones and
local autocrine and paracrine factors. The mRNA complement of each of its different cell types is altered during
different phases of the menstrual cycle, with the onset of
decidualisation and in response to an implanting embryo,
as well as in pathological conditions such as endometriosis, abnormal bleeding, infection or cancer. Gene
expression profiling has the capacity to identify new targets for the manipulation of fertility and the diagnosis and
treatment of endometrial abnormalities.
A number of endometrial gene expression studies have
been discussed in recent reviews (Giudice 2003, 2004,
Horcajadas et al. 2004), so rather than providing a
detailed description of their findings, the experimental
design features of these studies have been summarised in
Table 2. With only three exceptions (Popovici et al. 2000,
Martin et al. 2002, Okada et al. 2003), all of these studies
included validation of a small number of differentially
expressed genes (usually less than 10) by an independent
mRNA quantification method (Northern blot, semi-quantitative or quantitative RT-PCR). Less than half also included
cellular localisation studies (in situ hybridisation and/or
The molecular events of endometrial stromal cell
decidualisation are still not well defined, so microarray
analysis is ideally suited to identify genes with important
regulatory roles in this process. A number of factors
including combined oestrogen and progesterone or
cAMP can be used to induce decidualisation in vitro
and the comparison of decidualised with non-decidualised cells has revealed many of the genes which are
likely to contribute to this process (Popovici et al. 2000,
Brar et al. 2001, Tierney et al. 2003). The two studies
investigating the window of implantation (Kao et al.
2002, Borthwick et al. 2003) were identical in experimental design except that the microarrays used in the
first study were hybridised with RNA samples from individual endometrial biopsies, whereas the second study
used pooled samples. It is important to note that there is
no consensus on the use of pooled or individual samples
Guide to issues in microarray analysis
Table 2 Summary of studies using DNA microarray analysis to investigate gene expression in the endometrium during normal processes or in
response to stimuli
Cell/tissue type
Isolated endometrial stromal cells
Endometrial biopsies
Pre-receptive vs receptive
Popovici et al. 2000
Brar et al. 2001
Tierney et al. 2003
Okada et al. 2003
Carson et al. 2002
Martin et al. 2002
Dominguez et al. 2003
Riesewijk et al. 2003
Horcajadas et al. 2004
Kao et al. 2002
Borthwick et al. 2003
Ponnampalam et al. 2004
Mirkin et al. 2004
Horcajadas et al. 2005
Mutter et al. 2001
Risinger et al. 2003
Moreno-Bueno et al. 2003
Cao et al. 2004
Saidi et al. 2004
Ferguson et al. 2004
Ferguson et al. 2005
Eyster et al. 2002
Lebovic et al. 2002
Kao et al. 2003
Arimoto et al. 2003
Matsuzaki et al. 2004
Yanaihara et al. 2005
Catalano et al. 2003
Punyadeera et al. 2005
Tan et al. 2003
Yoshioka et al. 2000
Reese et al. 2001
Cheon et al. 2002
Curtis-Hewitt et al. 2003
Ho Hong et al. 2004
Watanabe et al. 2003
Yao et al. 2003
Poggi et al. 2003
Naciff et al. 2002
Wu et al. 2003
Ace & Okulicz, 2004
Tynan et al. 2005
Ishiwata et al. 2003
Window of implantation vs late
proliferative phase
Across normal menstrual cycle
Normal vs IVF cycles
Endometrial cancer
Laser capture microdissection
Endometrial explants
Epithelial vs stromal cells
Progestin antagonist (RU486)
Menstrual vs late proliferative phase; oestrogen
Across normal oestrous cycle
Pre-vs post-implantation
Implantation vs interimplantation sites
Progestin antagonist (RU486)
Whole uterus
Dissected decidua
Whole uterus
Hoxa-10 deficiency
Ovariectomy and oestrogen treatment
Rhesus monkey
Cynomolgus monkey
Endometrial biopsies
Endometrial biopsies
Endometrial biopsies
Pre-receptive vs receptive
Progestin antagonist (RU486)
Pregnant vs non-pregnant
(see ‘Experimental design’ above). Many of the differentially expressed genes identified by Borthwick et al.
(2003) were the same as those reported by Kao et al.
(2002), suggesting that different experimental designs
may be equally valid. Similarly, there was some consensus between the study by Kao et al. (2002) and that of
Carson et al. (2002) examining the transition of the
endometrium into receptivity. The study by Riesewijk
et al. (2003) was slightly different from that of Carson
et al. (2002) in that consecutive endometrial biopsies
were taken from the same patient (rather than from
different patients) and the timing of the biopsies was
more precise. There was good agreement in the data
obtained by Riesewijk et al. (2003) and Kao et al.
(2002), but less between the two more similar studies
(Carson et al. 2002, Riesewijk et al. 2003). Discrepancies in data sets are not unexpected, considering the
many sources of variation in microarray experiments,
and the differences in methodology used by different
laboratories. Any genes which show consistent differential expression under similar experimental conditions can
therefore be considered almost certain to play an important biological role.
Genes involved in endometrial receptivity and implantation have also been examined using the progesterone
receptor antagonist RU486 (Cheon et al. 2002, Catalano
et al. 2003, Tynan et al. 2005). As RU486 is known to
inhibit implantation in mice and humans, its downstream
target genes are likely to be involved in normal
implantation, and have been identified in the whole
mouse uterus (Cheon et al. 2002), in human endometrial
explants (Catalano et al. 2003) and in cynomolgus
monkey endometrial biopsies (Tynan et al. 2005). While
no genes were found to be regulated by RU486 in all
Reproduction (2005) 130 1–13
C A White and L A Salamonsen
three species, some were identified as downregulated
with both RU486 treatment (Cheon et al. 2002) and in
post-implantation compared with pre-implantation mouse
uterus (Yoshioka et al. 2000). Reese et al. (2001) examined
genes involved in mouse implantation using a combined
approach of implantation versus interimplantation sites
and activated versus delayed implantation. Interestingly,
many of the genes regulated in both models were associated with the maternal immune response. Genes found to
be regulated in both mouse and human endometrium
with the onset of decidualisation and/or receptivity are
attractive targets for the manipulation of implantation
mechanisms conserved across species.
Microarrays have also been utilised to identify potential markers of endometrial pathologies. Studies exploring differential gene expression in endometriotic lesions
versus eutopic endometrium (Eyster et al. 2002, Lebovic
et al. 2002, Arimoto et al. 2003) have revealed dysregulation of a number of genes in endometriotic tissue
(reviewed in Giudice 2003), which may prove to have
functional roles in this disease. Interestingly, a comparison of gene expression in eutopic endometrium from
women with and without endometriosis (Kao et al.
2003) has shown that the endometrium of women with
endometriosis has an altered transcriptional profile to
that of women without the disease. As endometriosis is
often associated with infertility, genes with altered
expression in endometriosis patients may be involved in
endometrial receptivity and embryo implantation
(reviewed in Giudice et al. 2002). Similarly, genes
found by microarray to be differentially expressed in
endometrial tumours compared with normal endometrium (Mutter et al. 2001, Saidi et al. 2004) are likely
to provide diagnostic markers and treatment targets in
the future (reviewed in Giudice 2003). Another powerful application of microarray technology is the classification of tumour types by their gene expression profiles
and a number of studies have successfully utilised this
approach in endometrial cancer (Moreno-Bueno et al.
2003, Risinger et al. 2003, Cao et al. 2004, Ferguson
et al. 2004, 2005). Molecular classification of tumours
using microarray technology has the potential to greatly
enhance patient management and improve treatment
and prognosis.
Conducting a microarray experiment requires extensive
planning and a high level of technical ability. Each of
the steps involved, from experimental design through to
data analysis, requires careful consideration and greatly
benefits from collaboration with a bioinformatician. As
the volume of microarray data continues to expand,
there is an increasing need for accepted standard operating procedures and coordinated data deposition. The
Microarray Gene Expression Data (MGED) Society exists
to encourage MIAME (Minimum Information About a
Reproduction (2005) 130 1–13
Microarray Experiment) compliance, so that microarray
data from different laboratories and platforms can be
unambiguously interpreted and independently verified.
Indeed publication in a number of leading journals,
including Nature, is now dependent upon meeting
MIAME standards (Rockett & Hellmann 2004). Standardisation of the field could also be achieved by
including a Standard Gene Set (SGS) on all microarray
platforms (Fryer et al. 2002). Publicly-available databases are also required to enable large-scale data
mining. For example, mRNA levels measured in many
human tissues using the Affymetrix system are now
available online (HuGE Index; http://www.hugeindex.
org) and an endometrial database containing gene
expression data from many of the studies mentioned
above can be accessed at http://endometrium.bcm.tmc.
Despite many recent advances, microarray analysis
should not be considered the end-point of an investigation, but rather as a tool to assist in the formulation
of hypotheses. With improved microarray quality, standardised data analysis methods and integration with
proteomic approaches, gene expression profiling will be
an extremely effective tool towards understanding the
biology of the reproductive system and in developing
diagnostic tests and therapeutic strategies for reproductive abnormalities.
The authors would like to thank Dr Garry Myers, Dr Gordon
Smyth, Prof Terry Speed and Dr Andrew Sharkey for training
and helpful discussions and Sue Panckridge for assistance in
the preparation of Figure 1. LAS is supported by the National
Health and Medical Research Council of Australia (grants
#241000 and #143798) and CAW by an Australian Postgraduate Award. The authors declare that there is no conflict of
interest that would prejudice the impartiality of this scientific
Ace CI & Okulicz WC 2004 Microarray profiling of progesteroneregulated endometrial genes during the rhesus monkey secretory
phase. Reproductive Biology & Endocrinology 2 54.
Arimoto T, Katagiri T, Oda K, Tsunoda T, Yasugi T, Osuga Y,
Yoshikawa H, Nishii O, Yano T, Taketani Y & Nakamura Y 2003
Genome-wide cDNA microarray analysis of gene-expression
profiles involved in ovarian endometriosis. International Journal of
Oncology 22 551 –560.
Baldi P & Long A 2001 A Bayesian framework for the analysis of
microarray expression data: regularized t-test and statistical
inferences of gene changes. Bioinformatics 17 509–519.
Barrett JC & Kawasaki ES 2003 Microarrays: the use of oligonucleotides and cDNA for the analysis of gene expression. Drug
Discovery Today 8 134– 141.
Borthwick JM, Charnock-Jones DS, Tom BD, Hull ML, Teirney R,
Phillips SC & Smith SK 2003 Determination of the transcript
profile of human endometrium. Molecular Human Reproduction
9 19–33.
Guide to issues in microarray analysis
Bowtell DDL 1999 Options available - from start to finish - for
obtaining expression data by microarray. Nature Genetics 21 Suppl
25– 32.
Brar AK, Handwerger S, Kessler CA & Aronow BJ 2001 Gene induction and categorical reprogramming during in vitro human endometrial fibroblast decidualization. Physiological Genomics
7 135– 148.
Cao QJ, Belbin T, Socci N, Balan R, Prystowsky MB, Childs G &
Jones JG 2004 Distinctive gene expression profiles by cDNA
microarrays in endometrioid and serous carcinomas of the endometrium. International Journal of Gynecological Pathology
23 321–329.
Carson DD, Lagow E, Thathiah A, Al-Shami R, Farach-Carson MC,
Vernon M, Yuan L, Fritz MA & Lessey B 2002 Changes in gene
expression during the early to mid-luteal (receptive phase)
transition in human endometrium detected by high-density
microarray screening. Molecular Human Reproduction 8
Catalano RD, Yanaihara A, Evans AL, Rocha D, Prentice A, Saidi S,
Print CG, Charnock-Jones DS, Sharkey AM & Smith SK 2003
The effect of RU486 on the gene expression profile in an endometrial explant model. Molecular Human Reproduction 9
Chen JJ, Delongchamp RR, Tsai C-A, Hsueh H-M, Sistare F,
Thompson KL, Desai VG & Fuscoe JC 2004 Analysis of variance
components in gene expression data. Bioinformatics 20
Cheon Y-P, Li Q, Xu X, DeMayo FJ, Bagchi IC & Bagchi MK 2002 A
genomic approach to identify novel progesterone receptor regulated pathways in the uterus during implantation. Molecular Endocrinology 16 2853– 2871.
Chuaqui RF, Bonner RF, Best CJ, Gillespie JW, Flaig MJ, Hewitt SM,
Phillips JL, Krizman DB, Tangrea MA & Ahram M et al. 2002 Postanalysis follow-up and validation of microarray experiments.
Nature Genetics 32 Suppl 509–514.
Churchill GA 2002 Fundamentals of experimental design for cDNA
microarrays. Nature Genetics 32 Suppl 490–495.
Cole SW, Galic Z & Zack JA 2003 Controlling false negative errors in
microarray differential expression analysis: a PRIM approach.
Bioinformatics 19 1808–1816.
Cui X & Churchill GA 2003 Statistical tests for differential expression
in cDNA microarray experiments. Genome Biology 4 210.
Curtis-Hewitt S, Deroo BJ, Hansen K, Collins J, Grissom S, Afshari
CA & Korach KS 2003 Estrogen receptor-dependent genomic responses in the uterus mirror the biphasic physiological response to
estrogen. Molecular Endocrinology 17 2070– 2083.
Dominguez F, Avila S, Cervero A, Martin J, Pellicer A, Castrillo JL &
Simon C 2003 A combined approach for gene discovery identifies
insulin-like growth factor-binding protein-related protein 1 as a
new gene implicated in human endometrial receptivity. Journal of
Clinical Endocrinology & Metabolism 88 1849–1857.
Dudoit S, Yang YH & Bolstad B 2002a Using R for the analysis of
DNA microarray data. R News 2 24– 32.
Dudoit S, Yang YH, Callow MJ & Speed TP 2002b Statistical
methods for identifying differentially expressed genes in replicated
cDNA microarray experiments. Statistica Sinica 12 111–139.
Duggan DJ, Bittner M, Chen Y, Meltzer P & Trent JM 1999
Expression profiling using cDNA microarrays. Nature Genetics 21
Suppl 10– 14.
Eisen MB, Spellman PT, Brown PO & Botstein D 1998 Cluster analysis and display of genome-wide expression patterns. PNAS
95 14863–14868.
Emmert-Buck MR, Bonner RF, Smith PD, Chuaqui RF, Zhuang Z,
Goldstein SR, Weiss RA & Liotta LA 1996 Laser capture microdissection. Science 274 998–1001.
Evans AL, Sharkey AS, Saidi SA, Print CG, Catalano RD, Smith SK &
Charnock-Jones DS 2003 Generation and use of a tailored gene
array to investigate vascular biology. Angiogenesis 6 93–104.
Eyster KM, Boles AL, Brannian JD & Hansen KA 2002 DNA microarray analysis of gene expression markers of endometriosis. Fertility & Sterility 77 38–42.
Ferguson SE, Olshen AB, Viale A, Awtrey CS, Barakat RR & Boyd J
2004 Gene expression profiling of tamoxifen-associated uterine
cancers: evidence for two molecular classes of endometrial carcinoma. Gynecological Oncology 92 719– 725.
Ferguson SE, Olshen AB, Viale A, Barakat RR & Boyd J 2005 Stratification of intermediate-risk endometrial cancer patients into groups
at high risk or low risk for recurrence based on tumor gene
expression profiles. Clinical Cancer Research 11 2252–2257.
Firestein GS & Pisetsky DS 2002 DNA microarrays: boundless technology or bound by technology? Guidelines for studies using
microarray technology. Arthritis & Rheumatism 46 859–861.
Forster T, Roy D & Ghazal P 2003 Experiments using microarray
technology: limitations and standard operating procedures. Journal
of Endocrinology 178 195 –204.
Fryer RM, Randall J, Yoshida T, Hsaio L-L, Blumenstock J, Jensen KE,
Dimofte T, Jensen RV & Gullans SR 2002 Global analysis of gene
expression: methods, interpretation, and pitfalls. Experimental
Nephrology 10 64–74.
Giudice LC, Telles TL, Lobo S & Kao LC 2002 The molecular basis
for implantation failure in endometriosis: on the road to discovery.
Annals of the New York Academy of Sciences 955 252–264.
Giudice LC 2003 Elucidating endometrial function in the post-genomic era. Human Reproduction Update 9 223 –235.
Giudice LC 2004 Microarray expression profiling reveals candidate
genes for human uterine receptivity. American Journal of Pharmacogenomics 4 299–312.
Ho Hong S, Young Nah H, Yoon Lee J, Chan Gye M, Hoon Kim C &
Kyoo Kim M 2004 Analysis of estrogen-regulated genes in mouse
uterus using cDNA microarray and laser capture microdissection.
Journal of Endocrinology 181 157–167.
Horcajadas JA, Riesewijk A, Martin J, Cervero A, Mosselman S,
Pellicer A & Simon C 2004 Global gene expression profiling of
human endometrial receptivity. Journal of Reproductive Immunology 63 41–49.
Horcajadas JA, Riesewijk A, Polman J, van Os R, Pellicer A,
Mosselman S & Simon C 2005 Effect of controlled ovarian hyperstimulation in IVF on endometrial gene expression profiles. Molecular Human Reproduction 11 195– 205.
Ihaka R & Gentleman R 1996 R: a language for data analysis and
graphics. Journal of Computational & Graphical Statistics
5 299–314.
Ishiwata H, Katsuma S, Kizaki K, Patel OV, Nakano H, Takahashi T,
Imai K, Hirasawa A, Shiojima S, Ikawa H, Suzuki Y, Tsujimoto G,
Izaike Y, Todoroki J & Hashizume K 2003 Characterization of
gene expression profiles in early bovine pregnancy using a custom
cDNA microarray. Molecular Reproduction & Development
65 9 –18.
Kao LC, Germeyer A, Tulac S, Lobo S, Yang JP, Taylor RN, Osteen K,
Lessey BA & Giudice LC 2003 Expression profiling of endometrium from women with endometriosis reveals candidate genes for
disease-based implantation failure and infertility. Endocrinology
144 2870– 2881.
Kao LC, Tulac S, Lobo S, Imani B, Yang JP, Germeyer A, Osteen K,
Taylor RN, Lessey BA & Giudice LC 2002 Global gene profiling in
human endometrium during the window of implantation. Endocrinology 143 2119–2138.
Kendziorski CM, Zhang Y, Lan H & Attie AD 2003 The efficiency of
pooling mRNA in microarray experiments. Biostatistics
4 465–477.
Kerr MK & Churchill GA 2001 Statistical design and the analysis
of gene expression microarray data. Genetics Research
77 123–128.
Kim JH, Shin DM & Lee YS 2002 Effect of local background
intensities in the normalization of cDNA microarray data with a
skewed expression profiles. Experimental Molecular Medicine
34 224–232.
Reproduction (2005) 130 1–13
C A White and L A Salamonsen
Knight J 2001 When the chips are down. Nature 410 860–861.
Kothapalli R, Yoder SJ, Mane S & Loughran TP Jr 2002 Microarray
results: how accurate are they? BMC Bioinformatics 3 22.
Lebovic DI, Baldocchi RA, Mueller MD & Taylor RN 2002
Altered expression of a cell-cycle suppressor gene, Tob-1, in
endometriotic cells by cDNA array analyses. Fertility & Sterility
78 849–854.
Lipschutz RJ, Fodor SPA, Gingeras TR & Lockhart DJ 1999 High density synthetic oligonucleotide arrays. Nature Genetics 21 Suppl
Lonnstedt I & Speed TP 2002 Replicated microarray data. Statistica
Sinica 12 31–46.
Matsuzaki S, Canis M, Vaurs-Barriere C, Pouly JL, Boespflug-Tanguy O, Penault-Llorca F, Dechelotte P, Dastugue B,
Okamura K & Mage G 2004 DNA microarray analysis of gene
expression profiles in deep endometriosis using laser capture
microdissection. Molecular Human Reproduction 10 719–728.
Martin J, Dominguez F, Avila S, Castrillo JL, Remohi J, Pellicer A &
Simon C 2002 Human endometrial receptivity: gene regulation.
Journal of Reproductive Immunology 55 131 –139.
Mirkin S, Nikas G, Hsiu J-G, Diaz J & Oehninger S 2004 Gene expression profiles and structural/functional features of the peri-implantation endometrium in natural and gonadotropin-stimulated
cycles. Journal of Clinical Endocrinology & Metabolism
89 5742–5752.
Moreno-Bueno G, Sanchez-Estevez C, Cassia R, Rodriguez-Perales S,
Diaz-Uriarte R, Dominguez O, Hardisson D, Andujar M, Prat J,
Matias-Guiu X, Cigudosa JC & Palacios J 2003 Differential gene
expression profile in endometrioid and nonendometrioid endometrial carcinoma: STK15 is frequently overexpressed and amplified in nonendometrioid carcinomas. Cancer Research 63
Mutter GL, Baak JPA, Fitzgerald JT, Gray R, Neuberg D, Kust GA,
Gentleman R, Gullans SR, Wei LJ & Wilcox G 2001 Global expression changes of constitutive and hormonally regulated genes
during endometrial neoplastic transformation. Gynecological
Oncology 83 177 –185.
Naciff JM, Jump ML, Torontali SM, Carr GJ, Tiesman JP, Overmann
GJ & Daston GP 2002 Gene expression profile induced by
17alpha-ethynyl estradiol, bisphenol A, and genistein in the developing female reproductive system of the rat. Toxicological Sciences 68 184–199.
Okada H, Nakajima T, Yoshimura T, Yasuda K & Kanzaki H 2003
Microarray analysis of genes controlled by progesterone in human
endometrial stromal cells in vitro. Gynecological Endocrinology
17 271–280.
Park T, Yi S-G, Kang S-H, Lee S-Y, Lee Y-S & Simon R 2003 Evaluation of normalization methods for microarray data. BMC Bioinformatics 4 33.
Peng X, Wood CL, Blalock EM, Chen KC, Landfield PW & Stromberg
AJ 2003 Statistical implications of pooling RNA samples for microarray experiments. BMC Bioinformatics 4 26.
Petalidis L, Bhattacharyya S, Morris GA, Collins VP, Freeman TC &
Lyons PA 2003 Global amplification of mRNA by template-switching PCR: linearity and application to microarray analysis. Nucleic
Acids Research 31 e142.
Poggi SH, Goodwin KM, Hill JM, Brenneman DE, Tendi E, Schninelli
S & Spong CY 2003 Differential expression of c-fos in a mouse
model of fetal alcohol syndrome. American Journal of Obstetrics &
Gynecology 189 786–789.
Ponnampalam AP, Weston GC, Trajstman AC, Susil B & Rogers PA
2004 Molecular classification of human endometrial cycle stages
by transcriptional profiling. Molecular Human Reproduction 10
Popovici RM, Kao LC & Giudice LC 2000 Discovery of new inducible genes in in vitro decidualized human endometrial stromal cells
using microarray technology. Endocrinology 141 3510–3513.
Punyadeera C, Dassen H, Klomp J, Dunselman G, Kamps R,
Dijcks F, Ederveen A, de Goeij A & Groothuis P 2005
Reproduction (2005) 130 1–13
Oestrogen-modulated gene expression in the human endometrium. Cellular & Molecular Life Sciences 62 239–250.
Rajeevan MS, Vernon SD, Taysavang N & Unger ER 2001 Validation
of array-based gene expression profiles by real-time (kinetic) RTPCR. Journal of Molecular Diagnostics 3 26–31.
Reese J, Das SK, Paria BC, Lim H, Song H, Matsumoto H, Knudtson
KL, DuBois RN & Dey SK 2001 Global gene expression analysis
to identify molecular markers of uterine receptivity and
embryo implantation. Journal of Biological Chemistry 276
Riesewijk A, Martin J, van Os R, Horcajadas JA, Polman J, Pellicer A,
Mosselman S & Simon C 2003 Gene expression profiling of human
endometrial receptivity on days LH þ 2 versus LH þ 7 by microarray technology. Molecular Human Reproduction 9 253 –264.
Risinger JI, Maxwell GL, Chandramouli GV, Jazaeri A, Aprelikova O,
Patterson T, Berchuck A & Barrett JC 2003 Microarray analysis
reveals distinct gene expression profiles among different histologic
types of endometrial cancer. Cancer Research 63 6–11.
Rockett JC & Hellmann GM 2004 Confirming microarray data - is it
really necessary? Genomics 83 541– 549.
Saidi SA, Holland CM, Kreil DP, MacKay DJ, Charnock-Jones DS,
Print CG & Smith SK 2004 Independent component analysis of
microarray data in the study of endometrial cancer. Oncogene
23 6677–6683.
Sasik R, Woelk CH & Corbeil J 2004 Microarray truths and consequences. Journal of Molecular Endocrinology 33 1–9.
Schena M, Shalon D, Davis RW & Brown PO 1995 Quantitative
monitoring of gene expression patterns with a complementary
DNA microarray. Science 270 467– 470.
Smyth GK & Speed T 2003 Normalization of cDNA microarray data.
Methods 31 265 –273.
Smyth GK, Yang YH & Speed T 2003 Statistical issues in cDNA
microarray data analysis. Methods in Molecular Biology
224 111– 136.
Stears RL, Martinsky T & Schena M 2003 Trends in microarray analysis. Nature Medicine 9 140 –145.
Sterrenburg E, Turk R, Boer JM, van Ommen GB & den Dunnen JT
2002 A common reference for cDNA microarray hybridizations.
Nucleic Acids Research 30 e116.
Storey JD & Tibshirani R 2003 Statistical significance for genomewide studies. PNAS 100 9440–9445.
Tan YF, Li FX, Piao YS, Sun XY & Wang YL 2003 Global gene profiling analysis of mouse uterus during the oestrous cycle. Reproduction 126 171–182.
Taniguchi M, Miura K, Iwao H & Yamanaka S 2001 Quantitative
assessment of DNA microarrays - comparison with Northern blot
analyses. Genomics 71 34–39.
Tierney EP, Tulac S, Huang S-TJ & Giudice LC 2003 Activation of the
protein kinase A pathway in human endometrial stromal cells
reveals sequential categorical gene regulation. Physiological
Genomics 16 47–66.
Tusher VG, Tibshirani R & Chu G 2001 Significance analysis of
microarrays applied to the ionizing radiation response. PNAS 98
Tynan S, Pacia E, Haynes-Johnson D, Lawrence D, D’Andrea MR,
Guo JZ, Lundeen S & Allan G 2005 The putative tumor suppressor
deleted in malignant brain tumors 1 (DMBT1) is an estrogen-regulated gene in rodent and primate endometrial epithelium. Endocrinology 146 1066–1073.
van Gelder RN, von Zastrow ME, Yool A, Dement WC, Barchas JD
& Eberwine JH 1990 Amplified RNA synthesized from limited
quantities of heterogeneous cDNA. PNAS 87 1663–1667.
Venter JC, Adams MD, Myers EW, Li PW, Mural RJ, Sutton GG,
Smith HO, Yandell M, Evans CA & Holt RA et al. 2001 The
sequence of the human genome. Science 291 1304– 1351.
Watanabe H, Suzuki A, Kobayashi M, Takahashi E, Itamoto M,
Lubahn DB, Handa H & Iguchi T 2003 Analysis of temporal
changes in the expression of estrogen-regulated genes in the
uterus. Journal of Molecular Endocrinology 30 347–358.
Guide to issues in microarray analysis
Wu X, Pang ST, Sahlin L, Blanck A, Norstedt G & Flores-Morales A
2003 Gene expression profiling of the effects of castration and
estrogen treatment in the rat uterus. Biology of Reproduction
69 1308–1317.
Yanaihara A, Otsuka Y, Iwasaki S, Aida T, Tachikawa T, Irie T &
Okai T 2005 Differences in gene expression in the proliferative
human endometrium. Fertility & Sterility 83 1206–1215.
Yang MCK, Yang JJ, McIndoe RA & She JX 2003 Microarray
experimental design: power and sample size considerations.
Physiological Genomics 16 24–28.
Yang YH & Speed T 2002 Design issues for cDNA microarray
experiments. Nature Reviews Genetics 3 579–588.
Yang YH, Dudoit S, Luu P, Lin DM, Peng V, Ngai J & Speed TP 2002
Normalization for cDNA microarray data: a robust composite
method addressing single and multiple slide systematic variation.
Nucleic Acids Research 30 e15.
Yao MW, Lim H, Schust DJ, Choe SE, Farago A, Ding Y, Michaud S,
Church GM & Maas RL 2003 Gene expression profiling reveals
progesterone-mediated cell cycle and immunoregulatory roles of
Hoxa-10 in the preimplantation uterus. Molecular Endocrinology
17 610 –627.
Yoshioka K, Matsuda F, Takakura K, Noda Y, Imakawa K & Sakai S
2000 Determination of genes involved in the process of
implantation: application of GeneChip to scan 6500 genes.
Biochemical & Biophysical Research Communications 272
531 –538.
Received 10 February 2005
First decision 12 April 2005
Revised manuscript received 27 April 2005
Accepted 3 May 2005
Reproduction (2005) 130 1–13