Global gene expression analysis reveals reduced abundance of putative microRNA targets in human prostate tumours.

Abstract

Recently, microRNAs (miRNAs) have taken centre stage in the field of human molecular oncology. Several studies have shown that miRNA profiling analyses offer new possibilities in cancer classification, diagnosis and prognosis. However, the function of miRNAs that are dysregulated in tumours remains largely a mystery. Global analysis of miRNA-target gene expression has helped illuminate the role of miRNAs in developmental gene expression programs, but such an approach has not been reported in cancer transcriptomics.
In this study, we globally analysed the expression patterns of miRNA target genes in prostate cancer by using several public microarray datasets. Intriguingly, we found that, in contrast to global mRNA transcript levels, putative miRNA targets showed a reduced abundance in prostate tumours relative to benign prostate tissue. Additionally, the down-regulation of these miRNA targets positively correlated with the number of types of miRNA target-sites in the 3' untranslated regions of these targets. Further investigation revealed that the globally low expression was mainly driven by the targets of 36 specific miRNAs that were reported to be up-regulated in prostate cancer by a miRNA expression profiling study. We also found that the transcript levels of miRNA targets were lower in androgen-independent prostate cancer than in androgen-dependent prostate cancer. Moreover, when the global analysis was extended to four other cancers, significant differences in transcript levels between miRNA targets and total mRNA backgrounds were found.
Global gene expression analysis, along with further investigation, suggests that miRNA targets have a significantly reduced transcript abundance in prostate cancer, when compared with the combined pool of all mRNAs. The abnormal expression pattern of miRNA targets in human cancer could be a common feature of the human cancer transcriptome. Our study may help to shed new light on the functional roles of miRNAs in cancer transcriptomics.

Full text

Page 1

BioMed Central
Page 1 of 14
(page number not for citation purposes)
BMC Genomics
Open Access
Research article
Global gene expression analysis reveals reduced abundance of
putative microRNA targets in human prostate tumours
Ruping Sun†, Xuping Fu†, Yao Li*, Yi Xie and Yumin Mao*
Address: State Key Laboratory of Genetic Engineering, Institute of Genetics, School of Life Sciences, Fudan University, Shanghai, PR China
Email: Ruping Sun - regularhand@gmail.com; Xuping Fu - xupingfu@fudan.edu.cn; Yao Li* - yaoli@fudan.edu.cn; Yi Xie - yixie@fudan.edu.cn;
Yumin Mao* - ymmao@fudan.edu.cn
* Corresponding authors †Equal contributors
Abstract
Background: Recently, microRNAs (miRNAs) have taken centre stage in the field of human
molecular oncology. Several studies have shown that miRNA profiling analyses offer new
possibilities in cancer classification, diagnosis and prognosis. However, the function of miRNAs that
are dysregulated in tumours remains largely a mystery. Global analysis of miRNA-target gene
expression has helped illuminate the role of miRNAs in developmental gene expression programs,
but such an approach has not been reported in cancer transcriptomics.
Results: In this study, we globally analysed the expression patterns of miRNA target genes in
prostate cancer by using several public microarray datasets. Intriguingly, we found that, in contrast
to global mRNA transcript levels, putative miRNA targets showed a reduced abundance in prostate
tumours relative to benign prostate tissue. Additionally, the down-regulation of these miRNA
targets positively correlated with the number of types of miRNA target-sites in the 3' untranslated
regions of these targets. Further investigation revealed that the globally low expression was mainly
driven by the targets of 36 specific miRNAs that were reported to be up-regulated in prostate
cancer by a miRNA expression profiling study. We also found that the transcript levels of miRNA
targets were lower in androgen-independent prostate cancer than in androgen-dependent prostate
cancer. Moreover, when the global analysis was extended to four other cancers, significant
differences in transcript levels between miRNA targets and total mRNA backgrounds were found.
Conclusion: Global gene expression analysis, along with further investigation, suggests that
miRNA targets have a significantly reduced transcript abundance in prostate cancer, when
compared with the combined pool of all mRNAs. The abnormal expression pattern of miRNA
targets in human cancer could be a common feature of the human cancer transcriptome. Our study
may help to shed new light on the functional roles of miRNAs in cancer transcriptomics.
Background
MicroRNAs are endogenous, approximately 22 nt single-
stranded non-coding RNAs that negatively regulate pro-
tein expression by binding to the 3' untranslated regions
(3' UTR) of messenger RNAs (mRNAs) and inhibiting
translation or inducing mRNA degradation or deadenyla-
tion [1]. MiRNA genes are expressed as large precursor
RNAs, called pri-mRNAs, which may encode multiple
miRNAs in a polycistronic arrangement. These precursors
are converted into mature miRNAs of 19 to 25 nucleotides
by the RNase III enzymes, Drosha (nuclear) and Dicer
(cytosolic). MiRNAs have been identified in the genomes
Published: 26 February 2009
BMC Genomics 2009, 10:93 doi:10.1186/1471-2164-10-93
Received: 30 July 2008
Accepted: 26 February 2009
This article is available from: http://www.biomedcentral.com/1471-2164/10/93
© 2009 Sun et al; licensee BioMed Central Ltd.
This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/2.0),
which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

Page 2

BMC Genomics 2009, 10:93http://www.biomedcentral.com/1471-2164/10/93
Page 2 of 14
(page number not for citation purposes)
of plants, animals and viruses. Human miRNAs have been
implicated in a variety of biological processes, and it is
estimated that 30% of protein coding genes are regulated
by miRNAs [2].
Recently, the potential role of miRNAs in human cancers
has been indicated by several studies, which suggest that
aberrations in miRNA expression in cancers may be
involved in tumour genesis and progression [3]. Abnor-
mal expression of miRNAs has been found in various can-
cers, and profiling of miRNAs has been shown to be a
more accurate method of classifying cancer subtypes than
profiling protein-coding genes. In the case of prostate can-
cer, a highly prevalent disease in the western world, sev-
eral miRNA expression profiles have been reported [4-8].
Although some of these studies are not consistent, poten-
tially due to differences in samples or chip platforms, all
of them confirmed the widespread dysregulation of miR-
NAs in prostate cancer. However, the disruption of
miRNA expression observed in human cancers needs to be
understood by analysing the causes and consequences of
the miRNA alterations. Thus far, the causes of miRNA dys-
regulation are partially known to be the results of three
mechanisms: (1) miRNA genes tend to locate in cancer
related genomic regions; (2) miRNA expression is epige-
netically regulated in cancers; and (3) miRNA processing
genes such as Drosha and Dicer are de-regulated in cancers.
However, little is known about the consequences of
improper regulation of miRNA expression. More work
needs to be done to show whether these miRNAs have a
direct function in cancer progression or are simply differ-
entially modulated in tumours.
Identifying the genes targeted by miRNA is crucial for
understanding the functions of the miRNAs. Based on the
conservation of 3' UTRs, which are complementary to the
"seed" region (nucleotides 2–7 from the 5' end) of miR-
NAs, several computational methods have been devel-
oped to search for miRNA targets. Some of these methods
have been biologically validated and proven to be accu-
rate [2,9]. According to single miRNA-mRNA target pair
information, miRNAs are thought to function as either
tumour suppressors or oncogenes. For example, the let-7
family has been shown to target the RAS gene [10]. Down-
regulation of the let-7 has been found in lung cancer, and
this finding is correlated with a poor prognosis. Thus,
reduced expression of let-7 is predicted to promote lung
cancer progression. However, the situation is complicated
by the fact that miRNAs regulate multiple genes, and a sin-
gle mRNA can be targeted by several different miRNAs.
Therefore, the impact of changes in miRNA expression in
cancers is likely to be dependent on the cellular context.
Evidence is emerging that miRNAs can not only repress
translation of mRNAs but can also induce their degrada-
tion, even if the mRNA target sites have only partial com-
plementarity to the miRNAs. For example, several studies
have revealed a correlation between the expression of
miRNAs and that of their targets through analysis of
mRNA target gene expression profiles and in situ hybridi-
sation [11-13]. A recent publication has reported the
impact of miRNAs on global mRNA and protein expres-
sion and showed that the regulation of protein-coding
genes by miRNAs is quite similar at both the transcript
and protein levels [14]. Moreover, Wu et al. have reported
that miRNAs increase target mRNA decay rates by promot-
ing rapid deadenylation [1]. Assuming that the miRNAs
dysregulated in prostate cancer have an influence on the
expression of their targets, analysis of target gene expres-
sion may provide clues to the functions of miRNAs in
prostate cancer.
In this study, we computationally explored the global
expression patterns of miRNA targets in human prostate
cancer using several published microarray datasets. Inter-
estingly, in contrast to all the mRNAs with altered expres-
sion in prostate cancer relative to benign prostate tissue,
the transcript levels of miRNA targets were significantly
lower. Closer examination revealed a positive correlation
between the reduced abundance of miRNA targets and the
number of target-site-types in the 3' UTRs of the target
mRNAs. Remarkably, we found that the predicted targets
of the up-regulated miRNAs in prostate cancer, reported
by a miRNA profiling study, were significantly more likely
to be down-regulated in prostate tumours than the pre-
dicted targets of all other miRNAs. We also found that the
transcript levels of miRNA targets were lower in androgen-
independent prostate cancer than in androgen-dependent
prostate cancer. Furthermore, the abnormal expression
pattern of miRNA targets could be a common feature of
the human cancer transcriptome.
Results
Globally reduced transcript levels of miRNA targets
relative to total mRNAs in prostate tumours
Using the gene expression atlas published by three inde-
pendent groups (Table 1) [15-17], and the conserved
miRNA target predictions from PicTar [9] and TargetScanS
[2], we generated a global view of the transcript levels of
miRNA targets in prostate cancer. First, the expression val-
ues of each mRNA were compared between localised pros-
tate cancer and benign prostate tissue in each dataset, to
determine if the mRNA had a higher or lower expression
level in the cancer tissue. After sorting total mRNAs into
three groups (high expression, low expression and
unchanged), we calculated the Rmir, Rtotal and RR values
(see Methods). The RR value is a surrogate for an increased
or decreased abundance of miRNA targets relative to total
mRNAs. Notably, for all three datasets, RR values were less
than 1 when comparing localised prostate cancer with
benign prostate tissue (Figure 1A and [see Additional file
1]). Resampling statistical tests indicated that the differ-

Page 3

BMC Genomics 2009, 10:93http://www.biomedcentral.com/1471-2164/10/93
Page 3 of 14
(page number not for citation purposes)
ences were significant (P < 0.05). To confirm this observa-
tion, we used the Hyper-Geometric distribution to
evaluate the enrichment levels of miRNA targets in the
three mRNA pools and found a unique and significant
enrichment in the low expression pool across all three
datasets (Enrichment P < 0.05) [see Additional file 1]. Fur-
thermore, the average expression ratio of miRNA targets
(0.997) was significantly lower than that of the non-
miRNA-target genes (1.021, P < 10-100, t test, dataset 1).
These findings reveal that miRNA targets have a high pro-
pensity to fall in the low expression mRNA group in local-
ised prostate cancer, an observation that is robust across
different datasets. Similar results were observed when
comparing metastatic prostate tumours or all prostate
tumours with benign tissues [see Additional file 1].
Through the rest of this study, we focused on the compar-
ison between localised prostate tumour and benign pros-
tate. To determine whether the low expression of miRNA
targets occurred in benign tissues, we randomly divided
the benign samples from dataset 1 into two classes and
undertook a benign-benign comparison. In this case,
there was no significant difference between miRNA targets
and total mRNAs (P > 0.05). The absolute Rmir and Rtotal
value variation in the three datasets might result from dif-
ferences in the chip platforms or from intrinsic differences
in the samples. Nonetheless, the fact that RR ratios were
always less than 1 for tumour when compared with
benign prostate indicates that, in contrast to total mRNA,
the transcript abundances of miRNA targets are signifi-
cantly lower in prostate cancer than in benign prostate tis-
sue.
Given the considerable noise of the gene expression data
and miRNA target prediction, we ensured the validity of
the above observations using three approaches. Firstly,
because microarray results may vary depending on quality
of the samples, we collected the gene expression profiles
that contained a relatively large number of samples and
classified the tissues adjacent to the tumours as benign
prostate tissues for all three datasets. Secondly, to rule out
the influence of different methods on determining the
high, low and unchanged expression groups, two different
methods, a and b, were adopted for datasets 1 and 2,
where the measurements of matched samples were pro-
vided. We also adopted three different cut-off values to
determine the mRNAs with altered expression. As shown
in Figure 1A, the global low expression of miRNA targets
did not vanish when we changed methods or cut-offs, and
the RR values actually decreased with the raised cut-offs.
For example, in dataset 1, RR = 0.79, 0.76, and 0.71 for
low, medium and high cut-offs, respectively (Method a,
PicTar targets). Finally, to determine if this observation is
robust for different miRNA target predictions, we carried
out the same analysis using two different target predic-
tions, which are the leading programs in this field, and
obtained similar results. Figure 1C shows that the RR val-
ues obtained using the two predictions were highly corre-
lated (Pearson correlation = 0.9691). Additionally, since
Table 1: Gene expression datasets used in this global analysis
Number of samples
Cancer
Type
Dataset Refs.Array type Probe
sets
BenignLocalised Metastatic
Prostate cancer
1 [15] Affy†
37690 5866 25
2 [16] cDNA26260 4161 9
3[17]cDNA998419 1420
Correlation analysis[19]Affy†
22277 1010
AD and AI analysis [21] cDNA13314317 AD/8 AI
Breast cancer
4[33]Affy†
470007 40
Lung adeno-carcinoma
5[34]Affy†
71291086
Acute myeloid leukaemia
6[35]cDNA213706‡
23§
Liver
cancer
7[36]cDNA2307576 104
†: Affymetrix oligo nucleotide microarray. ‡: Normal Bone Marrow. §: Acute Myeloid Leukaemia. Correlation analysis: The dataset used in the
correlation analysis between the transcript levels of individual miRNAs and those of their putative targets. AD and AI analysis: The dataset used in
the analysis of the transcript levels of miRNA targets in androgen-dependent and androgen-independent prostate cancer.

Page 4

BMC Genomics 2009, 10:93http://www.biomedcentral.com/1471-2164/10/93
Page 4 of 14
(page number not for citation purposes)
repression by miRNAs also result in decreased translation,
we analysed the protein levels of several miRNA targets
using a small proteomic dataset. In this dataset, 64 pro-
teins were reported to be dysregulated in prostate cancer,
relative to benign prostate, by employing a high-through-
put immunoblot approach [18]. After converting the
names of these proteins into RefSeq mRNA IDs (some
proteins have more than one RefSeq ID), we mapped all
miRNA targets to them. As shown in Figure 1B, the
miRNA targets showed a propensity to fall in the
decreased protein group (RR < 1), suggesting a reduced
protein level of miRNA targets in prostate tumour. Taken
together, our data indicate that miRNA target genes are
expressed at lower levels in prostate cancer than normal
prostate tissue.
Association between transcript abundance of miRNAs and
their target mRNAs in prostate cancer
To further investigate the globally low expression of all
miRNA targets, we calculated the RR values of the targets
of 120 individual miRNAs. These miRNAs had more than
50 PicTar targets with altered expression (with an average
of 125 targets per miRNA) in comparing localised prostate
tumours with benign tissues (Method a, Medium cut-off,
Dataset 1). If we were to randomly select a set of mRNAs
same as the number of targets of each miRNA found in
Comparison of transcript levels of miRNA targets between in prostate cancer and in benign tissues
Figure 1
Comparison of transcript levels of miRNA targets between in prostate cancer and in benign tissues. (A) The
transcript levels of miRNA targets in prostate tumours were compared with those of total mRNAs using Rmir, Rtotal and RR val-
ues (see methods). The R values of all miRNA targets (Rmir) and total altered mRNAs (Rtotal) are represented as black and
white bars, respectively. Also plotted are the RR (Rmir/Rtotal) values (colored dots, right axis). The RR value is a surrogate for an
increased or decreased abundance of miRNA targets relative to total mRNAs. Blue color denotes RR < 1 and red denotes RR
> 1. Three gene expression microarray datasets (Dataset 1, 2 and 3) were used for this comparison. Three different cut-off val-
ues (L: Low; M: Medium; and H: High) and two methods a and b (for dataset 1 and 2) were chosen to rule out the bias of single
method. Asterisk represents the statistical significance of each comparison (resampling statistical test, see methods). One
asterisk means P < 0.05; two asterisks, P < 0.01; three asterisks, P < 0.0002. The complete data are reported [see Additional
file 1]. (B) Analysis of the protein levels of miRNA targets using a small proteomic dataset. Black and green cycles represent the
numbers of increased and decreased proteins in prostate tumours relative to benign prostate tissues, respectively. Small cycles
represent the number of miRNA targets mapped into these two protein groups. Areas of the cycles are scaled to the proteins
number. (C) Correlation between the RR values obtained using two different miRNA target predictions.

Page 5

BMC Genomics 2009, 10:93http://www.biomedcentral.com/1471-2164/10/93
Page 5 of 14
(page number not for citation purposes)
dataset 1, the distribution of RR values generated by the
random sets would approximate a background distribu-
tion. The distribution of individual miRNAs clearly dif-
fered from the background distribution: the RR values
obtained for individual miRNAs clustered in the range of
0.4 to 0.8 (Figure 2A), suggesting that these miRNAs may
down-regulate their predicted targets in prostate tumours.
We also found that the distribution of mean expression
values of the target genes of individual miRNAs clearly dif-
fered from the mean expression value of total mRNAs
(Figure 2B). Among the 120 individual miRNAs studied,
miR-194, miR-193 and miR-29a showed the lowest RR val-
ues (0.414, 0.423 and 0.435). It should be mentioned
that a few target groups preferentially show a relatively
high expression level. For example, miR-133a, which was
recently reported to be down-regulated in prostate
tumours [19], had an RR value > 1 (1.194). Similarly, miR-
125b, which was validated to be down-regulated in pros-
tate cancer by quantitative RT-PCR assays [8], had a rela-
tively high RR value (0.958).
We next asked whether there was an association between
the low expression of general targets and the dysregula-
tion of miRNAs in prostate cancer. Currently, the litera-
ture shows that the expression of miRNAs and their targets
are expected to be inversely correlated [12]. Namely, the
low expression of miRNA targets might imply a concur-
rently high expression of these miRNAs in prostate cancer.
This trend was seen in a miRNA expression profiling
study, which showed a significant up-regulation of many
miRNAs in prostate cancer [4]. As expected, when relating
the expression of miRNA targets to that of the miRNAs
using this miRNA expression profile, we found a weak
negative correlation between the differential expression
scores of individual miRNAs and the RR values of their
targets (Spearman correlation, Rs = -0.342, P = 0.025, N =
43) [see Additional file 2]. In contrast to the studies using
over-expression (or knockdown) of a single miRNA to
detect correlation between the expression of the miRNA
and its targets, cancer cells show alterations of many miR-
NAs with overlapping targets, and thus, it is hard to judge
the contributions of individual miRNAs to the low expres-
sion of their targets. To facilitate a more global view, we
divided all PicTar targets into two groups. Group I con-
tained 5514 targets of 39 significantly up-regulated miR-
NAs in prostate cancer reported by Volinia et al. (36 of the
39 had conserved PicTar targets). Group II contained all
other predicted miRNA targets (2663 targets). After map-
ping these two groups to the mRNAs with altered expres-
sion (Medium cut-off, Method a), we found that the
predicted targets of the 36 up-regulated miRNAs were sig-
nificantly more likely to be down-regulated in prostate
tumours than the predicted targets of all other miRNAs
(Figure 3, P = 0.001, 0.007, 0.013 for Dataset 1–3, respec-
tively). In all three datasets, there was no difference in the
relative transcript abundance between mRNA from group
II and total mRNAs (P > 0.05). The RR values for the puta-
tive targets of the 36 up-regulated miRNAs (group I tar-
gets) were significantly lower than the RR values for the
group II targets in these datasets, indicating a selective
down-regulation of group I targets in prostate tumours for
all three datasets. We then constructed a gene set contain-
Transcript levels of the target groups of individual miRNAs in prostate cancer
Figure 2
Transcript levels of the target groups of individual
miRNAs in prostate cancer. (A) Count distribution of RR
values of the targets of 120 individual miRNAs is shown
above. Count distribution of RR values of random mRNA
sets is shown below. Blue color denotes RR < 1 and red
denote RR > 1. (B) Density distribution (green line) of mean
expression values (log ratio) of the targets of individual miR-
NAs clearly departed from the mean expression value of
total mRNAs (blue line) and that of non target mRNAs (red
line).