26-30 Microarray March17_RZ
18.03.2004
12:21 Uhr
Seite 26
Evaluation of Procedures for Amplification of
Small-Size Samples for Hybridization on Microarrays
Sandra Klur1, Karen Toy2, Mickey P. Williams2, and Ulrich Certa1,*
1Roche Center for Medical Genomics, F. Hoffmann–La Roche, Ltd. Basel, Switzerland
2Department of Molecular Biology, Genentech, Inc., San Francisco, USA
*Corresponding author: ulrich.certa@roche.com
Microarrays
Various approaches have been developed for the preparation of samples for
gene expression monitoring. For Affymetrix chips, a standard protocol is
widely used; however, this is inefficient for small samples such as laser
capture microdissections. Several amplification procedures for such samples
Sandra Klur
already exist, and our goal was to test two of them: the first is based on
random PCR amplification, and the second, linear amplification, involves
performing the standard protocol twice. We analyzed a dilution of a commercially available
mouse brain total RNA preparation and microdissections from mouse hippocampus and striatum.
We evaluated the quality of microarray data by analyzing several chip parameters and performing
multiple comparisons. At the biological level, brain microdissections prepared with either method
gave similar expression results. At the technical level, analysis of the commercial sample showed
that random PCR amplification is more reproducible, requires smaller RNA input, and generates
cRNA of higher quality than linear amplification.
Reprinted as a shortened version from Genomics 83 (2004), Klur S et al. ”Evaluation of procedures for amplification of small-size samples for
hybridization on microarrays”, p. 508– 517 with permission from Elsevier. [...] designates where text has been shortened.
[...] DNA microarrays are widely used in biology [3–5] and
have applications in various fields such as tumor profiling
[6, 7], analysis of complex diseases [8], or determination
of genes involved in development [9]. Thanks to the
development of techniques such as laser capture
microdissection [10–12], cell flow sorting, or needle biopsies, one is now able to analyze complex organic structures more accurately. For example, microdissection is
revolutionizing neurobiology: brain, one of the most complex biological structures, is made of around 500 different
RNA isolation
5 g total RNA
10–50 ng total RNA
Standard procedure
PCR amplification
Linear amplification
cDNA synthesis
cDNA synthesis
cDNA synthesis
In vitro transcription
Random PCR amplification
cDNA synthesis
In vitro transcription
Hybridization on array
Figure 1: Strategies used to generate biotinylated targets for
gene expression profiling
26
cell types, with a total of 1012 cells [13,14]. Laser microdissection allows for the excision of very precise small tissue pieces or even single cells for the analysis of site-specific functions (i.e., spatial learning in the hippocampus) or
study of diseases (i.e., tumor cells disseminated in a network of normal cells). As standard protocols for sample
preparation are inefficient with total RNA amounts smaller
than 5 µg, it is evident that an amplification step is necessary for analyzing minute amounts of total RNA. Several
strategies have been developed for target preparation and
amplification in combination with Affymetrix chips. In this
report, we will focus on procedures in which a doublestranded cDNA (ds cDNA) is produced from total RNA
and then used as template to generate a labeled cRNA
during an in vitro transcription reaction [15,16]. We compared microarray data from samples amplified through
two different processes derived from this standard protocol. The first technique, random PCR amplification,
includes a PCR amplification step at the ds cDNA level
(Microarray Target Amplification Kit, Roche Applied
Science). The second method, known as linear amplification, consists of two cycles of cDNA synthesis followed by
in vitro transcription (GeneChip Eukaryotic Small Sample
Target Labeling, Affymetrix) and is an adaptation of the
method first described by Eberwine [15]. [...] To evaluate
the performance of both amplification methods and to
compare them to the standard protocol, we studied samples from two sources. [...]
BIOCHEMICA · NO. 2 · 2004
26-30 Microarray March17_RZ
18.03.2004
12:22 Uhr
Seite 27
Table 1: Samples analyzed and hybridization result
Input
Method
R1
Commercial
Mouse brain total RNA
Mouse brain total RNA
Mouse brain total RNA
Mouse brain total RNA
Mouse brain total RNA
Microdissections Hippocampus—indpdt MD
Hippocampus—pooled MD
Striatum
Hippocampus—indpdt MD
Hippocampus—pooled MD
Striatum
Yield (µg)
R2 R3 R4
3′/5′ ratio
Mean STDEV
% present calls
Mean STDEV
5 µg
50 ng
10 ng
50 ng
10 ng
Standard
PCR
PCR
2x Amp
2x Amp
73.6
52.6
63.2
42.2
51.6
56.0
57.6
59.2
38.2
53.4
58.9
58.0
63.7
55.0
45.6
76.2
56.2
65.7
57.3
45.1
1.1
1.0
1.0
3.9
5.7
0.1
0.1
0.1
0.1
0.8
50.2
51.3
50.8
40.7
33.7
2.9
1.4
1.4
5.0
7.0
50 ng
50 ng
50 ng
50 ng
50 ng
50 ng
PCR
PCR
PCR
2x Amp
2x Amp
2x Amp
98.8
82.8
88.4
48.8
71.1
75.5
77.6
73.6
72.0
76.4
74.0
60.8
64.0
73.6
82.4
83.7
73.4
58.8
70.0
72.0
75.2
81.2
80.6
74.2
1.1
1.1
1.2
4.7
5.8
7.7
0.2
0.0
0.1
0.9
0.8
2.0
38.8
37.2
35.7
37.9
37.8
33.7
3.5
3.3
2.3
3.7
1.0
3.1
Source, source of RNA; Input, amount of total RNA used for target synthesis, the amount indicated for microdissections is an evaluation; Method, method used for
generation of labeled cRNA; Yield, yield of cRNA from the four replicates of each condition; 3′/5′ ratio, ratio of hybridization signals obtained from GeneChip probe
pairs directed at the 3′ or 5′ end of GAPDH and β-actin transcripts; % present calls, percentage of genes detected as present by MAS 5.0; Mean and STDEV, mean and
corresponding standard deviation of ratio and present calls of the four replicates of each condition.
Results
Amplification strategy
The aim of our study was to evaluate the sensitivity,
reproducibility, and reliability of two amplification protocols for gene expression analysis of small-size samples.
Both amplification techniques are derived from a standard protocol [16]. Figure 1 shows the different sample
preparation strategies we have tested in our analysis.
The first step of classical sample preparation consists of
reverse transcription of total RNA with an oligo(dT)
primer containing a T7 promoter for generation of a
ds cDNA. Then, the second strand of cDNA is synthesized through RNA nicking and priming. Finally, in vitro
transcription (IVT) of ds cDNA with a T7 RNA polymerase and biotinylated nucleotides produces a labeled
cRNA, which is hybridized to the chip after fragmentation. The random PCR protocol relies on amplification of
the cDNA before IVT. For this purpose, the primers used
for cDNA synthesis are slightly modified: they contain an
additional sequence (target amplification sequence,
TAS) which is used for PCR amplification. The linear
amplification protocol consists of two cycles of cDNA
synthesis followed by IVT: the first cycle provides initial
amplification of total RNA, resulting in unlabeled cRNA;
the second cycle achieves further amplification and
incorporation of biotinylated nucleotides for labeling.
Samples analyzed
To evaluate the performance of both amplification protocols, we analyzed a commercial mouse brain total RNA
solution: we amplified 50- and 10-ng aliquots through
random PCR or by linear amplification (four replicates
for each quantity and each condition) and we compared
the chip results to a ‘‘gold standard’’ reference comprising a 5-µg aliquot processed with the standard protocol
(four replicates).
To see if we could also obtain reliable data from biological samples, we included laser microdissections from
different mouse brain regions in our study: two groups of
samples that were constituted of independent microdissections from either hippocampus or striatum and a
third group, which was a pool of hippocampus microdissections. The reason for preparing a pool from microdissected samples was to test the effect of amplification on
homogeneous biological samples, and the independent
microdissections served to pinpoint differences resulting
from biological variability.
Microarrays
Source
Gene expression profiles were obtained from all these
samples using the murine genome U74Av2 microarray
from Affymetrix, allowing analysis of ≈ 6000 functionally
characterized Mouse Unigene Sequences (Build 74) and
≈ 6000 EST clusters. For each condition, we prepared
four samples in parallel that were independently
hybridized on chips. Resulting chip data were normalized before multiple comparison was performed. [...]
Amplification method performance
The yield of labeled cRNA is a critical issue because sufficient material has to be prepared for hybridization onto
a microarray. Generally, a minimum of 10 µg of labeled
cRNA is loaded on a chip.
The total RNA amount contained in the microdissections
has been deduced from the number of PCR cycles necessary for amplification of 10- and 50-ng mouse brain
total RNA aliquots and of 50 ng of control total RNA
(provided in the RAS Microarray Target Amplification Kit).
We carried out 25 PCR cycles for all samples and for the
positive control. From this observation, we deduced that
there was around 10 to 50 ng of total RNA in a microdissected sample (≈ 1000 cells). The quantity of labeled
cRNA obtained after amplification was measured by
BIOCHEMICA · NO. 2 · 2004
27
18.03.2004
12:22 Uhr
Seite 28
absorbance at 260 nm. The values are reported in Table 1.
Both amplification methods afforded the production of
enough material for hybridization to microarray: for all
samples we obtained at least 40 µg cRNA. This is comparable to the values obtained for a 5-µg sample prepared with the classical protocol. Slightly more material
was yielded from random PCR amplification.
We first examined the present call percentage. This value
corresponds to the percentage of probe sets that have a
detectable and specific hybridization signal, thus reflecting
directly the percentage of genes that are expressed in a
sample. At first sight, present call percentage is higher for
mouse brain samples than for microdissections (Table 1).
As this parameter is related to sample complexity, this
result is not surprising: in a mouse brain total RNA extract
(whole organ), much more genes are expressed than in a
brain microdissection containing only around 1000 cells
(small piece of organ). For mouse brain total RNA, the
highest value, around 50%, is obtained when preparing
5 µg aliquots with the standard protocol or amplifying 50and 10-ng aliquots through random PCR. For linear amplification of 50- and 10-ng aliquots, only 40% present calls
or even less was detected, indicating a clear loss of information (10–15% of present calls are not detected) with this
amplification method. With regard to microdissections, all
samples had similar present call percentages, around
28
a
An additional important issue is the degree of reproducibility within individual replicates and different conditions. We thus compared the methods using scatter plots
of data from mouse brain total RNA (Figure 2). For the
standard protocol (Figure 2A), the point distribution
between two chips looks like the classical scatter-plot
shape typical for Affymetrix chips. When scatter plots of
chips prepared from PCR- or linear-amplified samples (2B
and 2C) are examined, the data display also looks very
similar to the classical pattern. However, comparing amplified to nonamplified samples, the point distribution is more
widely spread. For random PCR amplification, we observe
a moderate bias for all signal intensities, whereas for linear
amplification, the scatter plot has a banana shape: those
genes expressed at either a low (intensity smaller than 50)
or high (intensity higher than 5000) level are amplified to a
greater extent than those expressed at a medium level.
Furthermore, we examined the correlation factors (r2)
between chips, and the values reflected the results of
scatter plots (Table 2). Within conditions, the correlation
was very good, at around 0.98 for nonamplified samples
and around 0.97 for random PCR-amplified samples. For
linear-amplified samples, correlation factors were not as
good, but still acceptable. As expected, comparing amplified samples to nonamplified ones, the correlation factors
drastically decreased. Actually, linear-amplified samples
showed a better correlation to nonamplified samples
c
10000
1000
100
10
1
1
10
100
1000
10000 100000
Signal intensity - PCR amplified
b
100000
Signal intensity - non amplified
100000
10000
1000
100
10
1
1
Signal intensity - non amplified
10
100
1000
10000 100000
Signal intensity - PCR amplified
e
d
100000
10000
1000
100
10
1
1
10
100
1000
10000 100000
Signal intensity - non amplified
Signal intensity - linear amplified
Figure 2: Scatter plots: signal
intensities in a log10 scale. Samples
are mouse brain total RNA. (a)
Standard protocol: 5 µg against
5 µg. (b) PCR amplification:
50 ng against 50 ng. (c) Linear
amplification: 50 ng against 50 ng.
(d) PCR amplification on 50 ng
against standard protocol on 5 µg.
(e) Linear amplification on 50 ng
against standard protocol on 5 µg.
(a to c) Signal intensities of one
chip were plotted against signal
intensities of another chip from the
same condition. (d and e) Mean of
signal intensities of the four chips
from one condition plotted against
mean of signal intensities of the
four chips from another condition.
Signal intensity - PCR amplified
Microarrays
Array quality parameters
To evaluate whether the two amplification protocols
allow for preparation of high-quality samples for
microarray analysis, we examined various chip quality
parameters calculated after analysis with the RACE-A
software, and the data resulting from the different
experimental conditions were compared.
37%, whether they were amplified through random PCR or
with the linear protocol. In that case, no clear difference
between PCR- and linear-amplified samples was
observed. Nevertheless, the data from mouse brain total
RNA indicate a better sensitivity of PCR amplification
(10 ng) compared to linear amplification (50 ng).
100000
10000
1000
100
10
1
1
10
100
1000
10000 100000
Signal intensity - non amplified
BIOCHEMICA · NO. 2 · 2004
Signal intensity - linear amplified
26-30 Microarray March17_RZ
100000
10000
1000
100
10
1
1
10
100
1000
10000 100000
Signal intensity - linear amplified
26-30 Microarray March17_RZ
18.03.2004
12:22 Uhr
Seite 29
Table 2: Chip correlation factors (r2)
None, 5 µg
PCR, 50 ng
PCR, 10 ng
Linear, 50 ng
Linear, 10 ng
(B) Amplification
Standard
5 µg
Random PCR
50 ng
10 ng
0.977
—
—
—
—
0.593
0.974
—
—
—
Random PCR
Independent
Pooled
hippocampus hippocampus
Linear amplification
50 ng
0.591
0.959
0.968
—
—
Striatum
0.744
0.499
0.495
0.955
—
10 ng
0.570
0.386
0.385
0.875
0.925
Linear amplification
Independent
Pooled
Striatum
hippocampus hippocampus
PCR
Independent hippocampus
Pooled hippocampus
Striatum
0.919
—
—
0.945
0.980
—
0.811
0.840
0.881
0.465
0.513
0.459
0.504
0.524
0.465
0.422
0.444
0.472
Linear
Independent hippocampus
Pooled hippocampus
Striatum
—
—
—
—
—
—
—
—
—
0.923
—
—
0.951
0.983
—
0.868
0.887
0.944
Correlation factors were calculated with RACE-A for all possible pair-wise comparisons of chips analyzed in our study. These factors were then merged within
each condition and reported in (A) for mouse brain total RNA and (B) for mouse brain microdissections.
(r2 = 0.74) than to PCR-amplified (r2 = 0.6), probably
because the same enzymatic reactions are applied in both
methods. We also asked whether it was possible to detect
differences in correlation factors within and between the
three groups of microdissections. The results were very
interesting: correlation factors for pooled hippocampus
replicates were better than those for independent hippocampus, and striatum microdissections correlated better with other striatum microdissections than with hippocampus microdissections (pooled or not).
Next, we wanted to examine the quality of amplified
cRNA and therefore we looked at the 3′/5′ ratio. This
quality parameter is automatically calculated by the
analysis software, after scanning and data normalization.
The value corresponds to the hybridization signal intensity of probe sets specific for the 3′ and 5′ ends of certain
transcripts. The optimal value is 1 and indicates that the
mRNA sequences are full length, neither degraded during total RNA extraction nor shortened during cDNA
synthesis. In our experiment, ratios for nonamplified
mouse brain total RNA were very close to 1 (Table 1). All
PCR-amplified samples (mouse brain total RNA and
microdissections) also had ratios of around 1. For linearamplified samples the ratios were higher, between 3.9
and 5.7 for mouse brain total RNA, and microdissections
also showed very high ratios. This reflects the fact that
linear amplification yields truncated transcripts without
5′ ends. This was somehow expected because random
primers are used for the second cDNA synthesis cycle.
Since most of the probe sets present on the microarray
represent the 3′ end of each transcript, this shortening
should not have a major impact on data quality.
Microarrays
(A) Amplification
We concluded from all these technical parameters that
both methods can be used for amplification of samples
for microarray analysis, with a small advantage for PCR
amplification: samples amplified through PCR are of
higher quality and the method offers increased sensitivity. PCR amplification has also some technical advantages: the protocol is easier to set up and less time consuming. Finally, an important observation is that the
same preparation protocol should be used for all samples within a set of experiments; if one uses different
protocols in a given study, there is a risk of adding variability due to technical manipulation and therefore of
disturbing overall gene expression level within replicates
of a condition and ultimately, the results will not be as
accurate. However, this can be a drawback if it is desired
to compare results from amplified samples to previous
results from nonamplified samples.
Detection of genes differentially expressed
in two brain regions
We also wanted to assess if it was possible to detect
genes differentially expressed in different brain areas
and if the same genes were identified by each amplification method. For this purpose, we compared the microarray data from hippocampus and striatum microdissections (Figure 3).
We first filtered for genes specific for hippocampus and
having a change factor higher than 2: we identified
50 genes by PCR amplification and 41 by linear amplification. We then checked if genes that were detected as
‘‘upregulated’’ by one method were similarly detected by
the other. No expression contradiction between the two
BIOCHEMICA · NO. 2 · 2004
29
26-30 Microarray March17_RZ
18.03.2004
a
PCR
amplification
b
PCR
amplification
12:22 Uhr
Seite 30
Genes specific for hippocampus
4
60
5
Linear
amplification
Genes specific for striatum
4
49
3
Linear
amplification
Microarrays
Figure 3: Overview of the number of genes that are
differentially expressed in hippocampus and
striatum. Numbers within the circles indicate genes
identified by the corresponding method and
numbers within the intersections indicate number
of genes commonly detected by both methods.
amplification methods was observed: all genes detected
as induced in hippocampus after PCR amplification were
also detected as induced in hippocampus after linear
amplification and few genes had no change in one
method. Ultimately, 60 genes specific for hippocampus
were successfully identified by both methods, with
4 genes being detected only by PCR amplification and
5 only by linear amplification. To complete the technical
and statistical issues of this comparison, we then evaluated the biological meaning of our data. We briefly examined the function of genes induced in hippocampus compared to striatum. Most of the genes identified and for
which enough data about function and tissue specificity
were known were brain-specific. Moreover, we successfully detected hippocampus-specific genes like hippocalcin, transcription factor 4, and eph receptor a7. Genes
described as specific for other tissues and overexpressed
in hippocampus were not necessarily false positives; the
data indicated only that expression in hippocampus was
higher than in striatum, rather than implying that these
genes have a biological function in hippocampus.
We performed a similar analysis for striatum-specific
genes. Forty-eight genes were identified through PCR
amplification and 19 through linear amplification. As
before, we did not notice any expression contradiction,
and very few genes had no change in one amplification
method. Finally, 49 genes specific for striatum were successfully detected by both methods, with 4 detected
only by PCR amplification and 3 only by linear amplification. We did not identify genes known as being specific
for striatum but most of the genes whose expression
was higher in striatum than in hippocampus were brainspecific. Our microarray results demonstrated that random PCR amplification and linear amplification are both
30
reliable for sample analysis on oligonucleotide chips.
Even if some genes have not been identified after both
amplification processes, they represent only a minority
(16 of 125 genes, around 13%). An important finding is
that direction of change (induction or repression) was
identical for the two protocols, and also change factor
values were very similar. In addition, the data were biologically significant and moreover, they correlated with
another expression study done using SAGE on whole
hippocampi, without sample amplification [23]. This is
clearly a further validation for the use of amplification in
functional genomics studies.
Discussion
We have shown that two unrelated sample amplification
protocols give good quality microarray data, with a small
advantage for random PCR amplification, which provides the best 3′/5′ ratios and cRNA yield. Comparing
expression profiles obtained by both methods, we found
that the direction of change was essentially identical for
both amplification techniques. Moreover, our results are
biologically relevant and show that amplification can be
used for gene expression analysis because no major distortion is introduced into the sample, thus allowing for
reliable study of very small samples. Another important
fact is that PCR, often described as introducing bias to
samples [5,15], gave results comparable to those of
other methods: when performing a reasonable number
of PCR cycles (around 25), almost no bias was detected.
Whatever tendency may be introduced is compensated
for since all samples are prepared with the same protocol and so are submitted to the same bias. The relative
expression level between samples is preserved, even
though the absolute expression level may be altered. [...]
Global gene expression analysis has a wide range of
applications in biological research and has already provided interesting results. The combination of DNA
microarray technology with microdissection and sample
amplification is promising and allows access to analysis
of discrete cell populations [24–26].
Please refer to the publisher for Materials and Methods and for
References.
Product
Pack Size
Cat. No.
Microarray
Target
Amplification Kit
1 kit
(10 reactions)
3 310 191
Microarray RNA Target
Synthesis Kit (T7)
1 kit
(25 reactions)
3 266 877
Microarray Target
Purification Kit
1 kit
3 266 885
(50 purifications)
BIOCHEMICA · NO. 2 · 2004