Enhanced microarray performance using low complexity representations of the transcriptome.

Abstract

Low abundance mRNAs are more difficult to examine using microarrays than high abundance mRNAs due to the effect of concentration on hybridization kinetics and signal-to-noise ratios. This report describes the use of low complexity representations (LCRs) of mRNA as the targets for cDNA microarrays. Individual sequences in LCRs are more highly represented than in the mRNA populations from which they are derived, leading to favorable hybridization kinetics. LCR targets permit the measurement of abundance changes that are difficult to measure using oligo(dT) priming for target synthesis. An oligo(dT)-primed target and three LCRs detect twice as many differentially regulated genes as could be detected by the oligo(dT)-primed target alone, in an experiment in which serum-starved fibroblasts responded to the reintroduction of serum. Thus, this target preparation strategy considerably increases the sensitivity of cDNA microarrays.

Full text

Page 1

Enhanced microarray performance using low
complexity representations of the transcriptome
Gae ¨lle Rondeau, Michael McClelland, Toan Nguyen, Rosana Risques1,
Yipeng Wang, Martin Judex2, Ann H. Cho and John Welsh*
Sidney Kimmel Cancer Center, 10835 Altman Row, San Diego, CA 92121, USA,1Department of Pathology,
University of Washington, Box 357705, 1959 NE Pacific Ave HSB K-081, Seattle, WA 98195, USA and2Klinikum
rechts der Isar, III. Medizinische Klinik, Ismaningerstrasse 22, 81675 Mu ¨nchen, Germany
Received February 16, 2005; Revised May 10, 2005; Accepted May 29, 2005
ABSTRACT
Low abundance mRNAs are more difficult to examine
using microarrays than high abundance mRNAs
due to the effect of concentration on hybridization
kinetics and signal-to-noise ratios. This report des-
cribes the use of low complexity representations
(LCRs) of mRNA as the targets for cDNA microarrays.
Individual sequences in LCRs are more highly rep-
resented than in the mRNA populations from which
they are derived, leading to favorable hybridization
kinetics. LCR targets permit the measurement of
abundance changes that are difficult to measure
using oligo(dT) priming for target synthesis. An
oligo(dT)-primed target and three LCRs detect twice
as many differentially regulated genes as could be
detected by the oligo(dT)-primed target alone, in an
experiment in which serum-starved fibroblasts res-
ponded to the reintroduction of serum. Thus, this tar-
get preparation strategy considerably increases the
sensitivity of cDNA microarrays.
INTRODUCTION
cDNA and oligonucleotide microarrays are convenient for
identifying changes in mRNA abundances (1–3). The propor-
tion of transcripts that can be detected and measured and the
accuracy of measurements of changes in transcript abundance
determine the kinds of problems that can be addressed using
microarrays. The abundance of a transcript can fall well below
one copy per cell, on average, such as in cases where a tran-
script is rare but biologically active, in cases where a message
has a brief transcription window, such as during cell cycle, or
in complex clinical samples where cells with high expression
are mixed with cells with low expression or no expression.
However, sensitivities in the range of one or a few transcripts
per mammalian cell are difficult to achieve routinely, and
experimental noise at the lower limits of sensitivity complic-
ates the quantitative assessment of changes ingene expression.
While the measurement of changes in relatively abundant
transcripts is appropriate for certain goals, such as in the clas-
sification of cancer types (4–9), greater sensitivity and accur-
acy is often desirable, if not necessary, such as in surveys for
changes in individual transcript abundances that are important
indiseases, or when analysisishampered by missing data (10).
This report describes a strategy for improving microarray
performance by using subsets, or low complexity representa-
tions(LCRs),ofthe transcriptomeasmicroarraytargets.There
are several methods for producing LCRs (11–14). Here, we
use arbitrarily primed PCR applied to oligo(dT)-primed first
strand cDNA to generate LCRs. In contrast to a random pri-
mer, most of the positions in an arbitrary primer are specified,
but its sequence need not be chosen on the basis of homology,
as would be the case with a specific PCR primer. Arbitrarily
primed PCR amplifies the sequences between sites in a DNA
template where an arbitrary primer or a pair of arbitrary pri-
mers find approximate matches on opposite strands in close
proximity. The complex class of transcripts participates in this
reaction more often than the less complex class of abundant
transcripts due to these requirements. As a result, arbitrary sets
of rare transcripts become highly represented in the reaction
product. The sequence of the arbitrary primers, the character-
isticsofarbitraryprimingsites, their distance from oneanother
and the characteristics of the sequences that they flank deter-
mine the sequences that are amplified and the extent of their
amplification. Different primers result in the amplification of
different subsets of the original mRNA sequence space,
including different transcripts and different parts of mRNA
isoforms. Sequences amplify reproducibly such that, when
two different mRNA populations are compared, differences
in expression can be detected (15). Lower complexity, over-
representation of sequences from the class of rare transcripts,
*To whom correspondence should be addressed. Tel: +1 858 450 5990; Fax: +1 858 450 3251; Email: jwelsh@skcc.org
? The Author 2005. Published by Oxford University Press. All rights reserved.
The online version of this article has been published under an open access model. Users are entitled to use, reproduce, disseminate, or display the open access
version of this article for non-commercial purposes provided that: the original authorship is properly and fully attributed; the Journal and Oxford University Press
areattributedastheoriginalplaceofpublicationwiththecorrectcitationdetailsgiven;ifanarticleissubsequentlyreproducedordisseminatednotinitsentiretybut
only in part or as a derivative work this must be clearly indicated. For commercial re-use, please contact journals.permissions@oupjournals.org
Nucleic Acids Research, 2005, Vol. 33, No. 11e100
doi:10.1093/nar/gni095

Page 2

and differential selection of isoforms and family members
suggested that LCRs may be useful for measuring changes
in the abundances of rare transcripts that are difficult to
measure accurately using cDNA microarrays. In previous
work, LCRs made using arbitrary priming methods (11,13)
allowed the measurement of abundance changes in transcripts
that were difficult to detect using oligo(dT)-primed reverse-
transcribed targets applied to nylon membrane cDNA arrays
(16,17). Here, this approach is adapted to glass slide microar-
rays. Individual LCRs can detect one-third to one-half of all
transcripts, and three different LCRs used in combination with
an oligo(dT)-primed target can detect 80% of all genes rep-
resented on a cDNA microarray. The number of differentially
regulated genes that can be detected and measured using three
LCRs together with oligo(dT)-primed targets is ?2-fold
higher than can be detected and measured using oligo(dT)-
primed targets alone.
MATERIALS AND METHODS
Cell lines and RNA preparation
Human fibroblast from ATCC (CRL 2091) were grown
to ?80% confluence in 150 cm dishes in DMEM with 10%
fetal bovine serum (heat inactivated at 56?C for 30 min,
Omega scientific), and with 200 U/ml penicillin and
200 mg/ml streptomycin. For serum starvation, cells were
grown in media containing 0.01% serum for 48 h as described
previously (3), and then were treated with 10% serum for 0
(i.e. no serum), 1 and 4 h. Cells were washed with ice-cold
phosphate-buffered saline, and total RNA was prepared using
an RNeasy Mini Kit (Qiagen, Valencia, CA). RNA concen-
tration was determined spectroscopically, and integrity was
assessed qualitatively by agarose gel electrophoresis.
LCR preparation
LCRs were prepared using RNA arbitrarily primed PCR
(11,16,17). Reverse transcription was performed on 5 mg
total RNA using an oligo(dT)20-VN primer (Genosys Biotech-
nologies, The Woodlands, TX). The reactions contained
50 mM Tris–HCl, pH 8.3, 75 mM KCl, 3 mM MgCl2,
20 mM DTT, 0.2 mM each dNTP, 0.5 mM primer and 20 U
M-MuLV reverse transcriptase (Promega, Madison, WI), in a
final volume of 200 ml. Reverse transcription was performed
at room temperature for 15 min, followed by 37?C for 1 h.
Reactions were stopped by heating for 5 min in boiling water
and cooling on ice. cDNA was diluted to 3 ng/ml with distilled
water prior to the PCRs.
Primers synthesized by Genosys Biotechnologies used to
generate LCRs were (in 50–30orientation) pm13 (CAGTGG-
GAG + AGTGAGCAC), pm14 (ACGAAGAAG + AGGGC-
ACCAC), pm19 (RRRGACAGTG), pm20 (RRRCTGCGCT),
pm21 (CAGAGGTRRR), pm22 (AACGGCGRRR), pm23
(AACGGCGACR), pm24 (GGGTGTGTAR), pm25 (GGT-
GAACGRR), pm28 (RTCCCCGCGA), pm29 (RRATGC-
CACT), pm30 (RRTTCGGAAG), pm31 (TCCGATGCTG),
pm32 (TGACGTCCGATGCTG), pm33 (GTGACAGACA),
pm34 (AACTGGTGACAGACA), pm35 (TGCGAAGGG-
GCACCA), pm36 (AACTGGAACTAGGGGCACCA), pm37
(AGGGGCACCA) and pm38 (TGCGAAGGGGCACCA).
Diluted cDNA (25 ml) was mixed with an equal volume of
2· PCR mixture containing 20 mM Tris–HCl, pH 8.3, 20 mM
KCl, 6 mM MgCl2, 0.35 mM each dNTP, 2 mM each primer,
2 mCi [a-32P]dCTP (ICN, Irvine, CA) and 0.5 U/ml AmpliTaq
DNA polymerase Stoffel fragment (Applied Biosystems,
Foster City, CA). Thermocycling used 3 min at 94?C followed
by 35 cycles of 94?C for 1 min, 35?C for 1 min and 72?C for
2 min. Product was purified using a QIAquick PCR Puri-
fication Kit (Qiagen) and examined for repeatability on
sequencing-style polyacrylamide gels. For fluorescent dye
labeling, 1 mg of RNA arbitrarily primed PCR product was
mixed with 12 mg of random hexamer and boiled for 5 min at
95?C. Reactions were performed at 37?C overnight in 50 ml
of buffer containing 10 mM Tris–HCl, pH 7.4, 5 mM MgCl2,
7.5 mM DTT, 0.025 mM dATP, dCTP, dGTP and 0.009 mM
dTTP, 0.014 mM Cy3- or Cy5-linked dUTP (Amersham,
Arlington Heights, IL) and 10 U of exonuclease-free Klenow
(New England Biolabs Inc., Beverly, MA). The targets were
purified with QIAquick PCR Purification Kit (Qiagen).
Labeling was checked by spectrophotometry at 550 and
650 nm for Cy3 and Cy5, respectively.
Microarray preparation
Human cDNA clones (I.M.A.G.E) (Research Genetics/
Invitrogen) were grown overnight in 96 well plates, inserts
were amplified using vector-specific primers and purified
using multiscreen filter plates (Millipore, Billerica, MA).
EachamplifiedcDNAwascombined1:1withdimethylsulfox-
ide for arraying as described previously (18) and printed onto
Ultra-GAPS coated glass microscope slides (Corning Inc.
LifeSciences, Acton, MA) using an OmniGrid 100 printer
(GenomicSolutionsAnnArbor,MI)at40–60%relativehumid-
ity.PrintedslideswereUVcross-linked(250mJ),bakedfor3h
at 80?C and stored at room temperature in a desiccator.
Hybridization and washes
After incorporation of fluorescent nucleotides, the LCR and
oligo(dT)-primed targets were lyophilized and brought to a
final volume of 45 ml in 25% formamide, 5· SSC, 0.1% SDS
and blocking agent [poly(A)15, yeast tRNA and COT-1 DNA].
The target was heated for 5 min at 95?C, centrifuged briefly,
and immediately applied to the slide in a hybridization cham-
ber. The chambers were submerged in a 42?C water bath
overnight. The slides were washed for 30 min at 42?C in a
2· SSC, 0.1% SDS solution and two times for 30 min at room
temperature in 0.1· SSC, 0.1% SDS and in 0.1· SSC solu-
tions, sequentially. Fluorescence intensities were measured
using a ScanArray5000 (Hewlett Packard) laser scanner.
Real-time RT–PCR
Real-time PCRs were performed in a solution containing
SybrGreen I, 0.35 mM 6-ROX (Molecular Probes, Eugene
OR), 0.2 mM dNTP, 1· PCR buffer (Qiagen), 4 mM MgCl2,
5 mM each primer and 0.025 U of HotStartTaq DNA poly-
merase (Qiagen), using an ABI PRISM 7900HT Sequence
Detector. Thermocycling was performed with an initial
10 min incubation at 95?C followed by 50 cycles of 95?C
for15s,60?Cfor1minand72?Cfor30s.Thiscyclingreaction
wasfollowedwith2minat95?C,15sat60?Cand15sat95?C.
A standard curve for each gene was prepared with a four point
e100Nucleic Acids Research, 2005, Vol. 33, No. 11
PAGE 2 OF 7

Page 3

dilution series. Each measurement was made in duplicate.
Measurements were normalized to an internal glyceraldehyde-
3-phosphate dehydrogenase RNA. Oligonucleotide primers
used are described in the Supplementary Table 1.
RESULTS
LCRs select and amplify subsets of mRNA sequences
With nylon membrane arrays, it had been shown that LCRs
select and amplify subsets of the mRNA represented in an
oligo(dT)-primed target (16,17). To demonstrate this effect
using microarrays, total RNA was isolated from growing
humanfibroblasts, followed bytargetsynthesisusinganchored
oligo(dT) priming or RNA arbitrarily primed PCR to make
an LCR. These were compared by hybridization to a glass
slide cDNA microarray containing several thousand human
gene sequences. The two target preparation methods resulted
in different sequence-specific signal intensities (Figure 1). In
many cases, the LCR-derived signal exceeded the oligo(dT)-
target-derived signal and vice versa. This suggested that LCRs
could be used to enhance signals for some transcripts on
microarrays, consistent with previous studies using LCRs as
targets for cDNA arrays printed on nylon membranes (16,17).
LCRs can detect differential gene regulation
The differential amplification of sequences in an LCR, when
compared with an oligo(dT)-primed target or with different
LCRs, suggested that LCRs couldbe used to detect differential
gene expression in cases where the signal from an oligo(dT)-
primed target is too weak to be useful. However, first, it was
necessary to demonstrate that LCRs could, indeed, be used as
targets for microarrays to measure differential gene expres-
sion. Transcripts having altered abundances 4 h after the addi-
tion of serum to serum-starved fibroblasts were measured by
microarray analysis of an LCR target (LCR pm22), and these
measurements were compared with real-time RT–PCR meas-
urements. Each microarray contained three identical subarrays
of 4621 spotted probes, 3770 of which were human gene
sequences, and the rest of which were controls of various
sorts, including 96 Salmonella sequences that served as neg-
ative controls for non-specific signals. Analyses were per-
formed using the limma package in BioConductor and the
R programming environment (19–21). Fluorescence intensity
measurements from human and control gene sequences were
adjusted using background subtraction, print-tip loess normal-
ization and scaling between reciprocal dye-swap pairs of
chips. MA plots were constructed [M = log2R ? log2G;
A = (log2R + log2G)/2] on the background subtracted, nor-
malized and scaled channel intensities for visual inspection of
thedata(SupplementaryTable2andSupplementaryFigure1a).
Real-time RT–PCR was performed for 17 transcripts that
showed an apparent change in abundance in the microarray
experiment, and had a modified t-statistic with P < 0.05 (21).
Primers for real-time RT–PCR spanned splice junctions to
avoid amplification from unspliced mRNA or from possible
genomic contamination. Agreement between the two quant-
itation methods was observed, with Pearson’s correlation
r = 0.90 (Figure 2). This indicated that LCRs can be used
as targets for microarrays to discover mRNA abundance
changes. Data have been deposited in the NCBI Gene Expres-
sion Omnibus(GEO)(http://www.ncbi.nlm.nih.gov/geo)
(GEO GSE2655).
LCRs detect differential gene regulation not
detected by oligo(dT)-primed targets
The enrichment of certain sequences in LCRs and the
depletion of others suggested that LCRs might be used to
detect differential gene regulation for genes that are normally
difficult to study using only an oligo(dT)-primed target, due
Figure 1. LCRs enhance subsets of mRNA sequences. Comparison between
oligo(dT) and an LCR target on a microarray. The LCM target (red) displays
different sequence representation than the oligo(dT) target (green), which is
seen as different ratios of red and green false coloring of fluorescence signal
intensity.
Figure 2. LCRtargetscandetectdifferentialgeneregulation.LCRpm22target
was prepared from RNA from serum-starved and starved-refed fibroblasts 4 h
after refeeding, andanalyzed usingmicroarrays. Differentially regulatedgenes
detected by the LCR were quantified using real-time RT–PCR. This figure
showsthatthelog2ratios(M)oftranscriptsfromthetwomethodscorrelatewell
and confirms that LCRs reliably report differential transcript abundance. Axis
label‘M.LCR’istheaveragelog2ratioofthenormalizedchannelintensities,I,
for the 4 and 0 h treatments, i.e. M.LCR = log2(It=4/It=0), reported by the LCR
target, and ‘M.Real-time RT–PCR’ is the log2of the corresponding ratio
reported by real-time RT–PCR. Pearson’s correlation, r, is shown.
PAGE 3 OF 7
Nucleic Acids Research, 2005, Vol. 33, No. 11e100

Page 4

to their low representation in the mRNA population. Total
RNA was harvested from serum-starved fibroblasts before
and 1 h after the reintroduction of serum. Two replicate
biological experiments and the corresponding reciprocal
dye-swap experiments were performed for each probe type.
Every scanner ‘channel’ corresponded to an independently
synthesized LCR, such that four microarrays represent results
from eight independent LCRs. This design was chosen so that
variance in LCR preparation could be explored, but in usual
practice, the dye-swap replicates would comprise technical
reciprocal labeling, as is commonly done. LCRs were made
using the arbitrary sequence oligonucleotide primers pm19,
pm22 and pm28. Oligo(dT)-primed and LCR targets were
labeled with Cye dyes and hybridized to microarrays to
detect genes that were differentially regulated between the
two biological conditions. LCRs made using RAP–PCR are
reproducible, as shown in Figure 3. Intensities for each
gene were determined from each microarray, with print-tip
loess normalization and scaling between chips. Two intensity
vectors were generated by averaging, gene-by-gene, one array
from each biological replicate and its reciprocal from the other
biological replicate,andthelog2ofthe resulting averages were
plotted as scatter plots. Figure 3 shows that correlations of
r > 0.94 were achieved for all four target types. Similar ana-
lysis of any of the three subarrays gave correlations r > 0.95
for all target types. Analysis of single biological replicates (i.e.
from a single pair of chips) gave correlations of r > 0.95 for
all but pm28, which gave r = 0.93 and r = 0.91 for the two
biological replicates, respectively. Background subtraction
increased scatter due to variance in the background, itself,
particularly for lower signals, but had only a minor effect
on the overall correlation (r > 0.93) (data not shown). If
RAP–PCR amplified sequences only according to the
occurrence of a partial match between the arbitrary primer
and its target, and no other efficiency terms were involved,
one may expect that subsets of mRNA sequences would be
sampled, but that their representation in the final product
would remain unchanged relative to their representation in
an oligo(dT)-primed target. This is not the case, however.
Scatter plots between different LCRs are not concentrated
along the diagonal, and correlations are below r = 0.65 for
all pairwise comparisons between LCRs, and between LCRs
and the oligo(dT)-primed targets (Supplementary Figure 2).
Combined, these observations indicate that the large differ-
ences between the oligo(dT)-primed targets and the LCR tar-
gets result from reproducible differences in the sequence
representations contained within the LCR targets, and not
from variance intrinsic to the LCRs. Thus, LCR synthesis
can be simple and robust, and can detect differential gene
regulation when hybridized to microarrays.
LCR targets can detect changes in transcript levels that are
missed by oligo(dT)-primed targets. The modified t-statistics
and associated P-values calculated for the four target types,
using limma, BioConductor and R (19–21) as described above
(Supplementary Table 2), were used to assess differential gene
regulation in response to introduction of serum to serum-
starved fibroblasts (see Table 1). Using P < 0.05 as a thresh-
old, the oligo(dT) target detected changes in 325 transcripts
out of the 3770 represented on the chip, and 213 of these were
unique to the oligo(dT) target, while the remaining 112 were
also detected as changes by one or more of the LCR targets.
The three LCRs, combined, detected changes in 416 tran-
scripts, 304 of which were missed by the oligo(dT) target.
LCRs from pm19, pm22 and pm28 contributed 123, 149
and 41 of these 304 changes, respectively, with some overlap.
Figure 4 shows the correlation between those mRNA abund-
ance changes detected only by the LCRs and the same
changes measured by quantitative RT–PCR. High Pearson’s
correlation (r = 0.79) indicates that LCRs are able to detect
differential gene expression that is largely invisible to
oligo(dT)-derived targets. When the changes in these tran-
scripts measured using oligo(dT)-derived targets were com-
pared with the real-time RT–PCR measurements, correlation
was lower (r = 0.55), as would be expected from their higher
P-values. Individual gene results, accession numbers and
descriptions are available in Supplementary Tables 3 and 4.
Further confirmation that LCRs reliably report differential
gene regulation can be seen in the comparison of changes in
expression reported by LCR targets with those reported by
oligo(dT) targets for those cases where both target types
reported changes. Figure 5 shows the correlation between
M-values calculated for these transcripts. Recall that M is
the log2of the ratio of normalized channel intensities, such
Figure 3. Reproducibility. Scatter plots of average intensities from technical
and biological replicates. Each average was calculated from loess adjusted,
normalized intensities from dye-swap replicate microarrays, one from each
biological replicate. (a) oligo(dT) target, (b) LCR pm19, (c) LCR pm22,
(d) LCR pm28.
Table 1. Detection of differential regulation by oligo(dT) and LCR targetsa
P-value threshold
All changes detected
Detected with more than one target
Detected by oligo(dT) target and possibly by LCR targets
Detected only by oligo(dT) target
Detected by LCR targets and possibly by oligo(dT) target
Detected by LCR targets only
Detected by oligo(dT) target and one or more LCR targets
0.05
629
121
325
213
416
304
112
aDifferential gene expression detected by an oligo(dT) target and three LCR
targets,comparingserum-starvedfibroblastsbeforeand1hafterreintroduction
of serum.
e100 Nucleic Acids Research, 2005, Vol. 33, No. 11
PAGE 4 OF 7

Page 5

that M = 1 corresponds to a 2-fold change, and so forth. A
correlation of r = 0.79 was obtained when both measurements
had P < 0.05, and using lower P-values tended to make the
correlation better (e.g. r = 0.84 for both measurements having
P < 0.02; data not shown). In this experiment, up-regulated
genes outnumber down-regulated genes, which might be
expected, given that down-regulation must be accompanied
by mRNA decay before it can be detected by microarray
hybridization. This result indicated that LCRs are able to
detect many of the same differentially regulated genes that
can be detected by an oligo(dT)-primed target and agrees with
the findings reported in Figures 2 and 4, where real-time RT–
PCR was used to confirm that changes in gene expression can
be detected using LCRs.
Fraction of genes for which LCRs and oligo(dT) targets
were able to detect differential expression
A point of interest is the number of array probes that had
intensities large enough relative to background that a change,
haditoccurred,wouldhave beenobserved.Thedatausedwere
that described above for fibroblast serum starvation and
refeeding, involving targets from oligo(dT), pm19, pm22
and pm28. For a randomly chosen set of 300 genes, the back-
ground was subtracted from the intensity values, followed by
divisionofthesignalsfromonechannel(i.e.the1htimepoint)
by small factors to artificially reduce the mean intensity value
fromthatchannel, mimicking down-regulation.Sincevariance
may not scale with the mean, other probe intensities were
searched to find the one with a mean channel intensity closest
to the artificial values, and the channel intensities from these
were substituted in for each of the 300 randomly chosen genes.
The other channel (i.e. the 0 h time point) was left unchanged.
Those differentially regulated transcripts that were normally
detected without the artificial change were excluded. Modified
t-tests were performed using limma after these artificial
changes, as described earlier, and P < 0.05 was used as the
criterion for the detection of a change; the results are shown
in Table 2. The columns labeled ‘4-fold’ and ‘2-fold’ show the
number and percent of the 300 genes that were detected as
changed at P < 0.05. This experiment was performed several
times with different random sets of 300 and gave similar res-
ults (data not shown). One limitation of this procedure is that
background determined from the area of the chip surrounding
a spot does not necessarily reflect the variance within the spot
due to other nuisance factors, such as cross-hybridization.
However, 81% of the genes had A-values (i.e. average of
Figure 4. LCR targets detect differential expression that is missed by
oligo(dT)-primed targets. Differential gene expression detected by LCRs but
not by oligo(dT) targets are confirmed by quantitative RT–PCR. This result
indicatesthatLCRscanbeusedtodetectchangesingeneexpressionthatcannot
be detected using an oligo(dT) target on these microarrays. Limma output for
these genes is in Supplementary Table 3, while accession numbers, current
Unigene designations, and descriptions are in Supplementary Table 4. Axis
label‘M.LCRs’istheaveragelog2ratioofthenormalizedchannelintensities,I,
forthe1and0htreatmentsreportedbytheLCRtargets,i.e.M.LCRs=log2(It=1/
It=0),and‘M.Real-timeRT–PCR’isthelog2ofthecorrespondingratioreported
by real-time RT–PCR. Pearson’s correlation, r, is shown.
Figure 5. LCR targets and oligo(dT)-primed targets report similar changes
where overlap occurs. LCR targets and oligo(dT) targets report similar log
ratios of differential expression, in those cases where both methods detect a
change.Thisgraphshowsdifferentialgeneexpressiondiscoveredusingmicro-
arrays and oligo(dT) priming for target synthesis, compared with the corre-
sponding measurement from the LCRs. Only those genes detected as changed
withP < 0.05inanLCRareincluded.Axislabel‘M.LCRs’istheaveragelog2
ratio of the normalized channel intensities, I, for the 1 and 0 h treatments, i.e.
M.LCRs = log2(It=1/It=0), and ‘M.oligo dT’ is the log2of the ratio for the same
gene reported by the oligo(dT) targets. Pearson’s correlation, r, is shown.
Table 2. Simulated down-regulation for 300 random genesa
4-fold 2-fold
Oligo(dT)
pm19
pm22
pm28
Combined
166 (55%)
191 (64%)
119 (40%)
71 (24%)
256 (85%)
130 (43%)
49 (16%)
92 (31%)
27 (9%)
165 (55%)
aSignal intensities for 300 randomly chosen genes from refed fibroblasts were
divided by 2 and 4. These channel intensity values were then replaced with the
channel intensities from the probe with a mean nearest these divided means,
thereby mimicking 2- and 4-fold down-regulation, and including the appro-
priate variance. This table contains the number of genes that were scored as
differentially regulated by limma with P < 0.05 (% of 300 is indicated in
parentheses).
PAGE 5 OF 7
Nucleic Acids Research, 2005, Vol. 33, No. 11e100