前言
致谢
编者简介
撰稿人
第1章 MicroMining:通过计算方式发现未知的microRNA Adam Grundhoff
第2章 动物microRNA基因预测 Ola Snφve,Pal S*trom
第3章 研究microRNA存在与功能的一系列资源 Praveen Sethupathy,Molly Megraw, Artemis
G. Hatzigeorgiou
第4章 大肠杆菌Hfq结合小RNA对mRNA稳定性及翻译的调控 Hiroji Aiba
第5章 动物细胞巾microRNA调控基因表达的机制 Yang Yu,Timothy W. Nilsen
第6章 秀丽隐杆线虫microRNA Mona J. Nolde,Frank J. Slack
第7章 秀丽隐杆线虫小RNA的分离及鉴定 Chisato Ushida, Yusuke Hokii
第8章 MicroRNA与果蝇发育 Utpal Bhadra,Sunit KumarSingh,Singh,S. N. C. V.
L. Pushpavalli,Praveensingh B. Hajeri,Manika Pal-Bhadra
第9章 斑马鱼RNA干扰与microRNA Alex S. Flynt,Elizabeth J. Thatcher,James G.
Patton
第10章 植物microRNA的产生和功能 Zoltan Havelda
第11章 拟南芥内源小RNA途径 Manu Agarwal,Julien Curaba,Xuemei Chen
第12章 如何评价microRNA表达——技术指导 Mirco Castoldi,Vladimir Benes,Martina U.
Muckenthaler
第13章 MicroRNA基因表达定量的方法 Lori A. Neely
第14章 MicroRNA介导的可变剪切调控 Rajesh K. Gaur
第15章 RNA聚合酶Ⅱ介导的内含子microRNA表达系统研究进展 Shi-Lung Lin,Shao-Yao Ying
第16章 基于microRNA的RNA聚合酶Ⅱ表达载体在动物细胞RNA干扰中的应用 Anne B. Vojtek,Kwan-Ho
Chung,Paresh D. Patel,David L. Turner
第17章 转基因RNA干扰技术——一种用于哺乳动物反向遗传学研究的快速低成本方法 Linghua Qiu,Zuoshang
Xu
第18章 AIDS交响曲——基于microRNA的治疗方法 Yoichi R. Fujii
第19章 MicroRNA与癌症——连点成线 Sumedha D. Jayasena
第20章 哺乳动物巾小RNA介导的转录水平基因沉默 Daniel H. Kim, John J. Rossi
第21章 由RNA介导的转录水平基因沉默控制的基因表达调控 Kevin V. Morris
索引
內容試閱：
1 MicroMining
Computational Approaches
to microRNA Discovery
Adam Grundhoff
Overview....................................................................................
............................1
1.1
Introduction.......................................................................................................2
1.2 When Is a Small RNA an
miRNA?...................................................................2
1.3 Advantages and Disadvantages of Experimental versus
Computational
miRNA
Identification........................................................................................3
1.4 Computational Prediction of
miRNAs..............................................................5
1.4.1 Getting Started: Upstream
Filtering......................................................7
1.4.2 Following Through: Structure Prediction and
Scoring....................... 12
1.4.3 Wrapping It Up: Experimental
Validation........................................... 14
1.5 Viral
miRNAs.................................................................................................
15
1.6
Conclusions......................................................................................................
16
References.................................................................................................................
16
Overview
The recent past has seen the rapid identification of thousands
of microRNAs
miRNAs encoded by various metazoan organisms as well as some
viruses, and it
is very likely that many more still await discovery. Most of the
hitherto-known miRNAs
have been identified via the cloning and sequencing of small
RNAs. While very
powerful, this approach is not without its limitations:
especially those miRNAs that
are of low abundance, or which are only expressed in certain
cell types or only during
brief periods of organismal development, or are easily missed in
cloning-based
screens. Thus, alternative means of miRNA discovery are
needed.
Given that the signal that marks the miRNA precursor for the
cellular processing
machinery appears to be a relatively simple one i.e., a hairpin
structure, and
considering the rapidly increasing availability of large-scale
genomic sequencing
data for many organisms, computational methods appear ideally
suited for the comprehensive
identification of hitherto-unknown miRNAs. This chapter
discusses the
general principles of computational miRNA identification
methods, examines their
advantages and disadvantages as compared to the cloning method,
and takes a look
at the various miRNA prediction algorithms that have been
developed recently.
1.1 I ntroduction
miRNAs are small ～22 nt RNA molecules that are able to
regulate the expression of
fully or partially complementary mRNA transcripts. As described
in greater detail
elsewhere in this book, they are initially transcribed as part
of hairpin structures
within much larger precursor transcripts the so-called primary
RNAs or pri-miRNAs.
Following excision of the stem-loops by the RNase III?like
enzyme Drosha,
the isolated hairpins called precursor miRNAs or pre-miRNAs
are exported to
the cytoplasm and further processed by the Dicer complex to
produce the mature,
single-stranded miRNA molecule. Recent evidence suggests that
plants and animals
encode a multitude of miRNAs, many of which are evolutionarily
conserved. As of
this writing, it is still true that the majority of known miRNAs
have been identified
experimentally, that is, by cloning of small RNAs. However, this
method has certain
limitations, and alternative means for the prediction of novel
miRNAs are therefore
increasingly sought.
The observation that pre-miRNAs form characteristic stem-loops
has spurred the
development of a number of computational approaches designed to
identify novel
miRNA candidates based on the prediction and analysis of
secondary structures.
Given the already complete or near-complete sequencing of whole
genomes from
many species, such approaches hold great promise for identifying
the full complement
of miRNAs encoded by a given organism. However, because the
precise set of
structural features that differentiate a pre-miRNA stem-loop
from the large number
of hairpins in the genome is not known, additional filters have
to be employed to
reduce the number of false-positive predictions, and
experimental confirmation of
the remaining candidates is required. In this chapter, I will
compare the benefits
and disadvantages of computational miRNA prediction methods in
comparison to
the cloning method, review principles of the existing miRNA
prediction algorithms,
discuss the general challenges and pitfalls of in silico miRNA
identification, and
provide an outlook of what might be expected from these
approaches in the future.
Finally, I will consider a special application of the miRNA
prediction problem: the
identification of miRNAs in viral genomes.
1.2 W hen is a small RNA an miRNA ?
In order to devise approaches designed to identify miRNAs, be
they experimental
or computational, it is important to clearly define what an
miRNA is. In a biological
sense, such a definition is quite straightforward: an miRNA is
simply a small,
single-stranded regulatory RNA molecule that is generated from
its precursor molecules
via successive processing by Drosha and Dicer. It is much more
difficult,
however, to define practicable criteria that are readily
testable on an experimental
or computational basis and that can unequivocally identify a
candidate sequence as
a genuine miRNA. Following the realization that miRNAs represent
abundant molecules
expressed in a wide variety of organisms, a consortium of
researchers agreed
on a set of criteria that have to be fulfilled before a
candidate can be called a bona
fide miRNA.1 According to these guidelines, it is necessary to
provide evidence that
1 the candidate sequence is expressed as an appropriately
sized RNA molecule in
living cells and, furthermore, does not stem from random
degradation Expression
criteria, and 2 that the maturation of the candidate involves
processing by Drosha
and Dicer Biogenesis criteria. The expression criteria are
preferentially satisfied by
detection of a distinct band of approximately 22 nt on a
Northern blot. Alternatively,
the ability to detect the molecule in a library of cloned,
size-selected RNAs is considered
sufficient evidence, especially if the library contains high
copy numbers of
the particular candidate sequence.
To satisfy the biogenesis criteria, the guidelines by Ambros et
al.1 call for experimental
proof of Dicer processing by demonstrating that increased levels
of the precursor
accumulate in cells with decreased Dicer expression. In
contrast, experimental
proof of Drosha processing is generally not required; instead,
it is sufficient to show
that the putative precursor transcript has the capacity to adopt
a secondary structure
that is likely to be amenable to Drosha processing. Of course,
given the incomplete
knowledge of the rules governing recognition of target mRNAs by
Drosha, it is not
known what exactly makes a given RNA structure amenable to
Drosha processing,
and as will be discussed later this complicates the
computational prediction
of miRNA candidates considerably. Based on the characteristics
of known miRNA
precursor structures, however, it is generally agreed that the
minimal requirements
are 1 the adopted structure is a hairpin that does not contain
many or large internal
bulges, and 2 the mature miRNA is to be found within the stem
not the loop part
of the hairpin.
Evolutionary conservation serves as a third biogenesis
criterion: As miRNAs
are often conserved in closely related and sometimes even in
distant species,
phylogenetic conservation of the miRNA sequence itself as well
as its fold-back
structure is considered strong evidence that the candidate
sequence represents a
genuine miRNA. An ideal miRNA candidate would meet all of the
preceding criteria;
however, it is generally considered sufficient to provide
convincing evidence
for at least one criterion out of the two categories. Indeed,
because Dicer knockout
cells are not readily available for most organisms, and
effective knockdown of
Dicer is technically challenging, positive experimental proof of
Dicer processing
is rarely shown.
1.3 A dvantages and Disadvantages of Experimental
versus Computational miRNA Identification
The “traditional” approach to identifying miRNAs consists of
cloning of small RNA
moieties. Although several protocols for the efficient cloning
of such molecules have
been devised, they all rely on the common principle of ligating
linkers to size-fractionated
RNAs, followed by cDNA synthesis and typically PCR
amplification. The
obtained products are then either cloned often after
concatamerization to increase
the information obtained in a single-sequence read and
sequenced, or subjected
directly to massive parallel sequencing approaches “deep
sequencing”. According
to the guidelines described earlier, these candidates are then
further evaluated to
ensure that the putative pre-miRNA sequence adopts an
appropriate hairpin structure
around the candidate. If this is the case, the candidate can
generally be considered a
bona fide miRNA, since the recovery of the clone from a small
RNA library already
satisfies the expression criterion nevertheless, Northern blots
are often performed to
allow for proper quantification of the miRNA.
The cloning approach has been extremely successful, and although
increasing
numbers of miRNAs are being identified via computational means,
the majority
of confirmed miRNAs currently listed in the miRNA database
miRBase, http:
microrna.sanger.ac.uk still have been identified via this
method. One of the great
advantages of the cloning protocol is that it provides the
precise sequence of the
mature miRNA molecule. Therefore, in contrast to
hybridization-based methods,
even closely related miRNAs that differ in only one nucleotide
position can be distinguished.
Also, the currently available computational prediction tools
generally only
allow identification of miRNA precursors but do not reliably
predict the location of
Drosha and Dicer cleavage sites. In contrast, cloning identifies
the precise 5′ and 3′
termini of the mature miRNA molecule.
As it appears that nucleotides 2 to 8 of the miRNA the
so-called seed region
are especially important for target recognition, knowledge of
the precise ends and
particularly the 5′ terminus is a distinct advantage if a
computational prediction
of target transcripts is to be performed. As might be expected,
the frequency with
which a given miRNA is cloned often is approximately equivalent
to its abundance
although this frequency may also be affected by other factors;
see the following text
and therefore provides a rough estimate of its expression
levels. Thus, abundantly
expressed miRNAs are usually readily identified. However, it can
be challenging
to achieve a saturated screen that also captures rare miRNAs.
Furthermore, even if
such miRNAs are contained within the library, one can never be
entirely certain that
enough clones have been sequenced to identify all of them.
In addition to these constraints, the scope of a cloning screen
is also limited by
its source material; naturally, only miRNAs that are expressed
in the cells from
which the RNA material was derived can be identified. Many
miRNAs, however,
are expressed in a tissue-dependent manner, or are only
expressed at certain developmental
stages. This limitation can be partially overcome for relatively
simple organisms,
where the RNA can be prepared from whole animals e.g., mixed
larvae stages
and adults from worms or insects.
In organisms with higher complexity such as vertebrates,
however, the situation is
more difficult: RNA from different embryonic or adult tissues
can be mixed, but the
sensitivity of the screen will dramatically decrease with the
complexity of the source
material, and it is very unlikely that nonabundant miRNAs could
be identified in
such screens. While these problems could be theoretically solved
by massive screening
efforts, that is, performing separate screens with material
prepared from every
individual tissue at each developmental stage, the cloning
approach also appears
limited in a more fundamental way. Several observations suggest
that some miRNAs
are more readily cloned than others owing to intrinsic
properties such as sequence
composition, the presence of certain nucleotides at their
termini, or posttranscriptional
modifications such as methylation or RNA editing.2?6
Computational approaches to miRNA discovery are not subject to
many of the
limitations that apply to the cloning method. Certainly, one of
the biggest advantages
of computational miRNA identification is the universal scope of
the analysis; as the
prediction does not require experimental material, it can
potentially discover all of
the miRNAs encoded by a given organism, even those that are
expressed only at
very low levels, in rare cells, or during brief periods of
development. However, this
advantage is partially annulled by the insufficient precision of
the presently available
algorithms: as the programs to varying degrees produce large
numbers of falsepositive
predictions, experimental verification is still a necessity.
Northern blotting is
frequently performed to investigate the expression of the
computationally predicted
candidates, or the predicted sequences are amplified from small
RNA libraries.
These procedures are not particularly compatible with
high-throughput screening,
and since many computational methods produce large numbers of
candidates, only a
small contingent of the predictions is usually subjected to
experimental verification,
whereas the majority remains untested. More importantly, the
experimental validation
methods are subject to many of the same limitations that hamper
the cloning
approach. Thus, even if an experimental verification is
attempted and fails, it is often
impossible to decide whether the failure was due to a
false-positive prediction, insufficient
sensitivity of the experimental detection method, or lack of
expression in the
tested tissue or cell line.
It is thus perhaps not surprising that the expression criterion
has not been satisfied
for most computationally predicted miRNA candidates. While some
groups have
attempted to reconcile these difficulties by developing
expression analysis tools that
are, for example, more sensitive or allow high-throughput
screening, there is also
tremendous effort to increase the reliability of computational
prediction methods
such that experimental confirmation is becoming less
important.
1.4 Computational Prediction of miRNA s
A plethora of computational approaches aimed at the prediction
of miRNAs have
been devised, and although nearly all of them use the evaluation
of features that are
thought to be characteristic for miRNAs in order to identify
novel candidates, they
vary significantly in scope, complexity, and level of
sophistication of the underlying
algorithms. Some approaches strive to identify the totality of
miRNAs encoded by a
given organism, whereas others aim to identify only miRNAs that
represent closely
related ortho- or paralogs of those that are already known. Some
programs investigate
some of the largest genomes, those of mammals, whereas others
consider only
some of the smallest, those of viruses.
Despite these differences, most of the approaches function
according to a common
scheme that might be abstracted as follows. First, a pool of
input sequences
usually representing the complete genome of a given organism
is filtered in order to
limit the number of candidates that have to be evaluated by
downstream algorithms.
I will refer to this process as upstream filtering in the
following. The filtered pool is
then subjected to a structure prediction. The obtained
structures are then compared
to those of known pre-miRNAs, and a score calculation is
performed, depending on
the degree of similarity. Finally, experimental validation is
attempted, usually for a
selection of the highest-scoring candidates.
There are considerable differences in the degree to which
structural features are
investigated during the scoring step; sometimes the filter might
simply ensure that the
candidate forms a hairpin structure, whereas in other cases it
might investigate the
candidate’s structure down to the minutest detail. The level of
sophistication, in large
part, will depend on the design of the upstream filter and the
efficiency with which
this filter preselects a set of candidates enriched for genuine
miRNAs. For example,
phylogenetic conservation is the most widely used upstream
filter and at least presently,
it is also appears to be the most efficient. Indeed, if the
sequence of a known
mature miRNA is perfectly conserved in a closely or even
distantly related species,
a relatively simple structural analysis that shows that the
ability of the surrounding
sequences to adopt a fold-back structure is conserved as well
might suffice.
In contrast, an ab initio prediction method in which the
upstream filter is minimal
will require a much more detailed structural analysis during the
downstream scoring
step. Thus, a highly efficient upstream filter requires a less
elaborate downstream
structure evaluation, and vice versa. The cloning method might
be considered a special
case of this scheme in which the upstream filtering is based on
an experimental
procedure; since this method produces only little background,
the subsequent structural
investigation can be minimal.
All of the available computational approaches are subject to the
production
of false-positive i.e., candidates that pass the filters but do
not represent genuine
miRNAs and false-negative predictions i.e., bona fide miRNAs
that are rejected
during the upstream filtering or the downstream scoring step.
The ratio with which
true-positive versus false-positive predictions are made will
determine the algorithm’s
accuracy, while the ratio of true-positive versus false-negative
predictions
will determine its sensitivity. Such rates are frequently
estimated in order to judge an
algorithm’s performance.
Estimating the rate of false-negative predictions is a
relatively straightforward
process. Often, only a limited number of the contingent of known
miRNAs is used to
establish the parameters of the filtering and scoring
algorithms. The remaining miRNAs
are then subjected to the prediction procedure, and the number
of rejected versus
retained miRNAs is determined. Alternatively, the full
complement of miRNAs
is repeatedly passed through the filters, and the method
parameters are adjusted
until an acceptable ratio between rejected and accepted miRNAs
is achieved what
exactly an acceptable ratio is will greatly vary with the
overall design and scope of
the method.
The estimation of false-positive prediction rates is a more
complicated matter:
in order to measure such numbers with high reliability, one
ideally would
have a set of sequences that assuredly does not contain any
miRNAs at all, or a
set in which all of the genuine miRNAs are known beforehand. In
theory, such
a set can be created artificially from randomly generated
sequences, or by shuffling
naturally occurring ones, but since biological sequences are
nonrandom, such
a reference set would be hardly representative of the
experimental sequence set.
Alternatively, one might select genetic elements that have known
functions and
are thus unlikely to additionally represent miRNAs, but this
would reduce the
complexity of the reference set so drastically that the gained
information would be
close to meaningless.
In reality, the rate of false-positive predictions is often
estimated on an experimental
basis. For this purpose, a representative subset of the
predictions or all of
購物車/收銀台(0) | 在線留言板
| 付款方式
| 運費計算
| 聯絡我們
| 幫助中心 |
加入書簽
HOME
新書上架
