Content uploaded by Jean Hausser
Author content
All content in this area was uploaded by Jean Hausser on Dec 16, 2015
Content may be subject to copyright.
MicroRNAs (mi RNAs) are 21–22‑nucleotide RNAs that
mediate post‑transcriptional gene silencing by guid‑
ing Argonaute (AGO) proteins to RNA targets. The first
report of a miRNA, lin‑4, that regulates the development
of Caenorhabditis elegans dates back 20years
1,2
; however,
the discovery of the let‑7 miRNA
3,4
and its conservation
from worms to humans first underscored the prevalence
and functional importance of miRNA‑dependent gene
regulation. Both lin‑4 and let‑7 were found to bind to the
3ʹ untranslated regions (UTRs) and repress translation
of their respective target mRNAs, lin‑14 and lin‑41, dur‑
ing crucial transitions in worm development
1,4
. Studies
from many groups have subsequently unravelled vari‑
ous mechanisms that link the recruitment of the miRNA-
induced silencing complex (miRISC) to mRNA targets to
reduce the output of the target proteins (reviewed in
REF.5) (FIG.1).
The lin‑14 and lin‑41 transcripts carry several
regions of extensive complementarity to their regu‑
latory mi RNAs
4,6
; however, this phenomenon is the
exception rather than the rule. Many metazoan miRNA
targets show clear complementarity to only 6–8 nucleo‑
tides at the 5ʹ end of the miRNA
7–9
. This region (that is,
nucleotides 2–7 of the miRNA) is thought to nucleate
the interaction of the miRNA with the target mRNA
and has become known as the miRNA ‘seed’ sequence
10
.
Sites in target mRNAs that are complementary to the
miRNA seed region are known as ‘canonical’ sites.
These sites show strong evidence of evolutionary selec‑
tion
11
and typically guide miRNA‑dependent mRNA
degradation
12
. The best evolutionary conservation sig‑
nal has been reported for sites that are 7–8 nucleotides
long
11,13,14
. In this Review and elsewhere
15
, canonical sites
are hence defined as sites that match the seed sequence
but that also pair with position 8 of the miRNA or
have an adenine opposite position 1 of the miRNA,
or both. Canonical sites are the main focus of compu‑
tational methods for miRNA target prediction, such as
TargetScan
16
, ElMMo
13
and others that are included in
miRBase
17
. Six‑nucleotide sites that pair strictly with the
miRNA seed region typically have a smaller effect on
target mRNA expression and have been termed marginal
sites
15
. Recently, high‑throughput studies have revealed
many sites that seem to be bound by mi RNAs, yet
their complementarity to the cognate mi RNAs and their
location within transcripts do not conform to current
paradigms
18–23
. Transcripts with these ‘non‑canonical’
sites undergo, on average, more modest changes when
the miRNA expression is perturbed than those that con‑
tain canonical sites
9,22
. The actual prevalence of these
non‑canonical sites and whether they have functions other
than mediating mRNA repression are open questions.
Although mi RNAs have been found to be necessary
for processes that range from cell differentiation
24–27
to
the development of individual tissues and organs
28–31
,
it remains challenging to conceptualize how these
phenotypes emerge from the effects of mi RNAs at the
molecular level
5
. Recent studies
32–35
have uncovered new
mechanisms that may be involved beyond the induction
of mRNA degradation and the inhibition of translation
1
Department of Molecular
Cell Biology, Weizmann
Institute of Science,
Herzl Street 234,
76100 Rehovot, Israel.
2
Biozentrum, University of
Basel and Swiss Institute
of Bioinformatics,
Klingelbergstrasse 50–70,
4156 Basel, Switzerland.
Correspondence to M.Z.
e-mail: mihaela.zavolan@
unibas.ch
doi:10.1038/nrg3765
Published online 15 July 2014
Argonaute
(AGO). A protein that, together
with a microRNA (miRNA),
forms a minimal
miRNA-induced silencing
complex (miRISC). Although
the number of AGO proteins
varies across species, four
paralogues (which are thought
to have overlapping activities)
are known in humans and mice.
Identification and consequences of
miRNA–target interactions — beyond
repression of gene expression
Jean Hausser
1
and Mihaela Zavolan
2
Abstract | Comparative genomics analyses and high-throughput experimental studies
indicate that a microRNA (miRNA) binds to hundreds of sites across the transcriptome.
Although the knockout of components of the miRNA biogenesis pathway has profound
phenotypic consequences, most predicted miRNA targets undergo small changes at
the mRNA and protein levels when the expression of the miRNA is perturbed. Alternatively,
mi RNAs can establish thresholds in and increase the coherence of the expression of their
target genes, as well as reduce the cell-to-cell variability in target gene expression. Here,
we review the recent progress in identifying miRNA targets and the emerging paradigms
of how mi RNAs shape the dynamics of target gene expression.
NON-CODING RNA
Nature Reviews Genetics
|
AOP, published online 15 July 2014; doi:10.1038/nrg3765
REVIEWS
NATURE REVIEWS
|
GENETICS ADVANCE ONLINE PUBLICATION
|
1
© 2014 Macmillan Publishers Limited. All rights reserved
miRNA-induced silencing
complex
(miRISC). A ribonucleoprotein
complex that includes
Argonaute (AGO), a microRNA
(miRNA) and additional
proteins such as Dicer and
trinucleotide repeat-containing
gene 6A protein (TNRC6A).
The miRNA in this complex
guides the AGO protein to
targets.
for which mi RNAs are best known. For example, it has
been proposed that mi RNAs counteract ‘leaky’ tran‑
scription by establishing thresholds in gene expression
levels and induce correlations in the expression of their
targets. Moreover, as a result of recent technological
advances, parameters that determine miRNA‑mediated
gene regulation (such as AGO–target association and
dissociation rates) and readouts (such as mRNA abun‑
dance) can now be measured with high resolution
36,37
.
These developments have provided the opportunity to
quantitatively study, using computational models, how
the interplay of regulators and targets gives rise to the
observed gene expression patterns.
Here, we provide an overview of the main experimen‑
tal approaches for miRNA target identification, as well
as the modulators and consequences of miRNA–target
interactions. Furthermore, we discuss the role of compu‑
tational modelling in synthesizing the extensive experi‑
mental evidence into a conceptual understanding of the
function of mi RNAs within gene regulatory networks.
A complementary view on the use of high‑throughput
experimental methods for characterizing the reciprocal
relationship between mi RNAs and their mRNA targets
has been covered in another recent review
38
.
Experimental identification of miRNA targets
Many methods have been proposed for miRNA target
identification
39
. Experimental approaches have ini‑
tially focused on the effects of miRNA–target inter‑
actions at levels that range from broad phenotypes
to changes in the abundance of mRNAs and proteins
(FIG.1). More recently, methods to directly capture
AGO‑bound RNAs have become available (FIG.1).
Here, we outline the main features of these meth‑
ods in increasing order of their resolution — from
broad phenotypes to individual miRNA‑binding
sites in mRNAs — in target mRNA identification. Of
note, consensus as to which experimental approach
is the most accurate is still lacking, partly because it
remains unclear what the most appropriate readout of
miRNA–target interactions shouldbe.
Genetic screening. A general approach to characterize
gene function is genetic screening. As mi RNAs repress
the expression of their targets, the loss of miRNA
function should lead to increased target expression.
Therefore, genes that when mutated rescue a miRNA
loss‑of‑function phenotype are good candidates as
direct miRNA targets. The first miRNA–target interac‑
tions described in the worm — lin‑4 with lin‑14 (REF.2)
and let‑7 with lin‑41 (REF.4) — emerged from such an
approach, as did the bantam miRNA, which controls
tissue growth in Drosophila melanogaster
40
. Genetic
screens have the advantage that the identified targets are
already linked to a phenotype. This is not the case when
targets are identified on the basis of their expression
changes in response to mi RNAs or of their association
with AGO proteins (see below).
There are also some disadvantages; for example,
genetic screens can yield both direct and indirect
Figure 1 | Methods for miRNA target identification. The steps in the
microRNA (miRNA) targeting cascade probed by each method are shown.
Methods such as RNA immunoprecipitation (RIP), crosslinking and immu-
noprecipitation (CLIP), and crosslinking, immunoprecipitation and
sequencing of hybrids (CLASH) identify RNAs (particularly mRNAs) and sites
therein that are bound by Argonaute (AGO) proteins. Binding of
miRNA-induced silencing complex (miRISC) to mRNA targets results in
translational repression that can be measured using ribosome profiling or
quantitative proteomics coupled with mRNA expression analysis.
Ultimately, the mRNA is decapped, deadenylated and degraded. mRNA
expression analysis using microarrays or mRNA sequencing probes this
specific outcome of miRNA–target interaction. AUG and UGA indicate
translation start and stop codons, respectively. miRNA-loaded AGO
typically binds to the 3ʹ untranslated region of mRNAs. eIFs, eukaryotic
translation initiation factors; m
7
G, 7-methyl-guanosine cap; PABPC,
cytoplasmic poly(A)-binding protein.
Nature Reviews | Genetics
AUG
eIFs
AUG
AGOAGO
AAA
PABPC
AUG
UGA
AAAAAAAA
AAAAAAAA
UGA
UGA
UGA
AGO
AUG
PABPC
m
7
G
Crosslinking and immunoprecipitation
RNA fragmentation
m
7
G
• RIP
• CLIP
• CLASH
• Ribosome
profiling
• Proteomics
PABPC
AAAAAAAA
m
7
G
miRISC
binding
Translation complex
disassembly and
translational
repression
eIFs eIFs
mRNA
expression
profiling
Decapping,
deadenylation
and mRNA
degradation
CLASH
CLIP
AGO RNA ligation
REVIEWS
2
|
ADVANCE ONLINE PUBLICATION www.nature.com/reviews/genetics
© 2014 Macmillan Publishers Limited. All rights reserved
CRISPR–Cas
(Clustered regularly interspaced
short palindromic repeat–
CRISPR-associated). A system
that mediates the RNA-based
immune defence of bacteria
and archea to viruses and
plasmids. It is composed of a
genomic CRISPR array (which
contains virus- or plasmid-
complementary sequences that
are interspersed with repeat
elements) and a Cas protein
(which carries the
endonuclease activity).
Stable isotope labelling by
amino acids in cell culture
(SILAC). A mass
spectrometry-based
experimental technique that
is used to compare protein
abundance in different
experimental conditions. Cells
of one sample are grown in a
medium containing amino
acids labelled with light
isotopes, and cells of another
sample are grown in medium
with heavy-isotope-labelled
amino acids. The samples are
then mixed, and changes in
protein abundance are
determined from the ratio
between the signal from the
light and heavy isotopes.
targets, and are difficult to carry out in mammals.
These problems may soon be mitigated by the avail‑
ability of methods for genome editing with few off‑
target effects, such as those based on the CRISPR–Cas
system
41,42
.
However, studies of worm
43
and mouse
44
miRNA
knockouts indicate that most mi RNAs are not indi‑
vidually essential for viability or development, possibly
because other mi RNAs or mechanisms act in a com‑
pensatory manner. Thus, other approaches have to be
used to accurately and comprehensively identify miRNA
targets.
Quantification of gene expression changes follow-
ing miRNA transfection. Measurements of mRNA
expression by gene expression microarrays or RNA
sequencing (RNA‑seq) have revealed mRNAs with
decay rates that increased in the presence of individual
mi RNAs
12,45
. Similarly, stable isotope labelling by amino
acids in cell culture (SILAC) approaches have enabled the
identification of proteins that are affected by changes in
miRNA expression
46,47
. Changes in mRNA translation
rates have been estimated by combining measurements
of mRNA expression with either ribosome profiling
48
or
SILAC‑based measurements of protein levels
49
. These
methods provide a quantitative view of the regulatory
effect of mi RNAs at different stages of gene expres‑
sion but, similarly to genetic screening, they yield both
direct and indirect miRNA targets, and do not reveal
the precise location of miRNA‑bindingsites.
Methods based on AGO crosslinking and immunoprecip-
itation. To determine direct miRNA targets, approaches
that rely on the immunoprecipitation of AGO proteins
to identify the AGO‑bound mRNAs have been proposed
(FIG.1). Initially, AGO–RNA complexes were captured by
RNA immunoprecipitation (RIP), and the RNAs were
identified either by microarrays
50,51
or through RNA‑
seq
52
. Further refinements of the method led to the
identification of individual miRNA target sites up to single‑
nucleotide resolution. For example, in AGO crosslinking
and immunoprecipitation (CLIP), ultraviolet (UV) light is
used to induce protein–RNA crosslinks. AGO is then
immunoprecipitated using a specific antibody, bring‑
ing with it both guide mi RNAs and their targets, which
are prepared together for sequencing
18,53
. Incorporation
of the photoreactive 4‑thiouridine into RNAs, fol‑
lowed by crosslinking using UVA light (365 nm),
in the photoactivatable ribonucleoside-enhanced CLIP
(PAR‑CLIP) method
19
has been shown to further increase
the efficiency of miRNA target capture
54
.
To identify individual AGO‑binding sites from the
reads obtained in CLIP experiments, specific data analy‑
sis methods are needed. A generally applicable approach
is to estimate the enrichment of reads obtained with
CLIP in relation to the number expected on the basis of
relative mRNA abundance. This enrichment has been
used as a measure of affinity of the miRNA‑guided AGO
for individual sites
18,53,54
. The number of reads expected
on the basis of the relative abundance of the mRNAs
with putative AGO‑binding sites can be estimated either
from RNA‑seq data
54,55
or by resampling the transcrip‑
tome that was previously quantified using micro arrays
18
.
The result of this analysis is a set of AGO‑binding
sites. Other methods have to be used to identify the
miRNA that guided the AGO protein to each individual
site (BOX1).
The uridine ribonucleotides that have undergone
crosslinking in PAR‑CLIP can be identified through
their distinctive mutations to deoxycytidines in the
sequenced reads
19,54
. These mutations, which are intro‑
duced during sample preparation at the locations where
RNA–protein crosslinks occurred, have been termed
crosslinking‑diagnostic mutations. Mutations, particu‑
larly deletions, also occur when crosslinking is carried
out with 254 nm UVC light in the absence of photoreac‑
tive nucleotides, as in high-throughput sequencing of RNAs
isolated by CLIP (HITS‑CLIP)
54,56
. The approach that was
specifically designed for nucleotide‑level resolution in
binding site identification is individual-nucleotide resolution
CLIP (iCLIP)
57
. This method takes advantage of the pro‑
pensity of the reverse transcriptase to stop polymerizing
when it reaches a block, such as the crosslinked nucleo‑
tide with a protein stub attached to it. AGO iCLIP data
sets have started to emerge very recently
58
, and the accu‑
racy of iCLIP relative to the other approaches for miRNA
target site identification can thus be evaluated.
Ideally, the guiding mi RNAs should be captured
together with the target sites. This has been achieved
using the crosslinking, immunoprecipitation and sequencing
of hybrids (CLASH) method, which is similar to CLIP
but includes an additional step in which the miRNA is
Box 1 | Computational models for miRNA target identification from CLIP data
Crosslinking and immunoprecipitation (CLIP) enables the identification of Argonaute
(AGO)-binding sites with high resolution. However, the specific microRNA (miRNA)
that guides the interaction of AGO with each target site remains to be inferred
computationally.
microMUMMIE
Features that are known to be relevant for the interaction between the
miRNA-induced silencing complex (miRISC) and miRNA targets can be evaluated for
each miRNA and each putative target site. For example, microMUMMIE
129
combines
evolutionary conservation, the type of miRNA ‘seed’ match, its spatial positioning
within the peaks of CLIP reads and the sequence composition of the peak into a
multivariate Markov model.
PARma
Photoactivatable ribonucleoside-enhanced CLIP miRNA assignment (PARma)
130
attempts to provide an explanation for the clusters of CLIP reads on the basis of the
processing signature of the nuclease that was used in preparing the sample for
sequencing, as well as on the basis of the identity and position of sequence motifs that
correspond to miRNA seed matches relative to the crosslinked nucleotides.
MIRZA
Rather than assuming specific modes of interaction between mi RNAs and their target
sites, as the above-mentioned approaches do, the CLIP data can be used to infer how
mi RNAs interact with target sites and then to predict which miRNA is most likely to
have guided the interaction with each target site. This approach has been used in the
construction of MIRZA — a biophysical model of miRNA–target interaction
9
.
Application of this model revealed that ~25% of the sites that are reproducibly
obtained by CLIP are ‘non-canonical’. The degree of destabilization of transcripts
with MIRZA-predicted non-canonical sites upon transfection of the cognate miRNA
correlates with their MIRZA score, which indicates that this method reliably predicts
non-canonical sites.
REVIEWS
NATURE REVIEWS
|
GENETICS ADVANCE ONLINE PUBLICATION
|
3
© 2014 Macmillan Publishers Limited. All rights reserved
Ribosome profiling
An experimental method to
quantify the translation
efficiency of individual mRNAs.
Sucrose gradients are used to
separate mRNAs that are
actively translated and that
are therefore associated with
multiple ribosomes.
Quantification can be done
either at the level of the whole
transcript (as in polysome
profiling) or of the precise
regions in the mRNA that are
bound by ribosomes, which
are protected from nuclease
digestion (as in ribosome
profiling or footprinting).
Crosslinking and
immunoprecipitation
(CLIP). An experimental
method to map the binding
sites of RNA-binding proteins
across the transcriptome.
Proteins are crosslinked to
RNA using ultraviolet light,
and an antibody is used to
specifically isolate the
RNA-binding protein of interest
together with its RNA
interaction partners, which are
subjected to sequencing.
Photoactivatable
ribonucleoside-enhanced
CLIP
(PAR-CLIP). A variant of
CLIP in which photoreactive
ribonucleoside analogues
such as 4-thiouridine are
incorporated into RNAs before
crosslinking with ultraviolet A
light (the wavelength of which
is 365 nm).
High-throughput
sequencing of RNAs
isolated by CLIP
(HITS-CLIP). A variant of CLIP
in which the crosslinking is
achieved using ultraviolet C
light (the wavelength of which
is 254 nm).
Individual-nucleotide
resolution CLIP
(iCLIP). A variant of CLIP that,
in contrast to the other
methods that rely on the
reverse transcriptase to
polymerize beyond the
crosslinked nucleotides, aims
to sequence the cDNAs at
which the reverse transcriptase
stopped at the crosslinked
nucleotides, thereby achieving
individual-nucleotide resolution
in the mapping of sites of
RNA-binding proteins.
ligated to its target site within the miRISC complex
22
.
The miRNA–target site chimaeras are sequenced, and
computational methods are further used to infer the
structure of the hybrid that the miRNA presumably
formed with the target site invivo. CLASH seems to
reveal miRNA‑specific models of interaction with tar‑
gets and holds promise for improved miRNA target pre‑
diction. However, the efficiency of the method is low
22
,
and the targets identified so far respond only weakly to
perturbations in miRNA expression
22
. This observation
suggests that further improvements of the method are
needed before it can be used to comprehensively map
and model miRNA–target interactions.
With the current availability of various CLIP‑based
methods, a question that arises concerns their rela‑
tive efficacy in capturing miRNA targets. Answering
this question is not entirely trivial because it remains
unclear what the most accurate readout of miRNA–
target interaction should be. An approach that has been
generally used is to measure the changes that are induced
by perturbations in miRNA levels in the targets identi‑
fied by a given method. The implicit assumption is that
‘stronger’ miRNA–target interactions would lead to
larger responses of the target to miRNA perturbations.
As described below, the change in target mRNA level
may not be the appropriate readout of all miRNA–target
interactions.
CLIP‑based methods are not without drawbacks.
First, it is not entirely clear how to best identify the
highest‑affinity sites from CLIP data. Amino acids dif‑
fer in their propensity to crosslink to RNA: aromatic
and positively charged amino acids are more reactive
to UV radiation
59–61
than other amino acids. Thus, the
efficiency with which a protein crosslinks to a binding
site depends, to some extent, on the configuration of the
protein–binding site complex. Second, the enrichment
of a target in CLIP is a function not only of its affinity
for miRISC but also of other parameters. To illustrate
this point, if the rate of target dissociation from miRISC
is faster than the rate at which the target decays, then
the CLIP data should reflect the affinity of AGO for the
target. However, if miRNA‑induced target degradation
is faster than the target dissociation from miRISC, then
increasing the affinity of miRNA–target interaction
beyond some threshold will make little difference to
the abundance of the target in the CLIP data. Third, as
is the case with other high‑throughput methods, low‑
abundance targets may not be captured at all regardless
of their affinity of interaction with AGO. Furthermore,
it remains unclear which AGO–mRNA interactions have
physiologically relevant consequences. Finally, on the
practical side, the experiments require highly specific
antibodies that bind to the protein in the protein–RNA
complex.
Anatomy of miRNA–target interactions
The miRNA seed region. The seed region of the miRNA
has long been known to be important for target recog‑
nition
11,62
. Structural studies of AGO in complex with
small RNAs revealed that the 5ʹ end of the small RNA
(positions 2–10) is stacked within the AGO protein, and
the Watson–Crick edges of bases 2–6 are positioned for
nucleating the interaction with the mRNA target
63,64
.
Isothermal titration calorimetry experiments suggested
that the helical conformation induced by AGO in the seed
region of the small RNA reduces the entropic cost of organ‑
izing the small RNA for interaction with the mRNA tar‑
get
65
, which increases the affinity of the small RNA for its
target by ~300‑fold relative to that expected in the absence
of AGO. Generally, the more extensive the complemen‑
tarity is to the seed region, the stronger is the response
of the target mRNA level to miRNA expression changes
8
.
However, how much ‘imperfection’ is tolerated in the
interaction of the miRNA 5ʹ end with its target?
Early studies suggested that G∙U base pairs that
involve the miRNA seed region are tolerated
66
and
that pairing of the miRNA 3ʹ end could compensate for
reduced complementarity in the seed region
7
. However,
answers have only started to emerge from recent studies
on the contribution of the AGO protein to the thermo‑
dynamics of miRISC–target binding
37,65
. Measurements
of the rates of miRNA‑guided binding and target cleav‑
age by AGO2 yield relatively similar values for targets
that only pair with the miRNA seed region and for
those with additional complementarity to the miRNA 3ʹ
end
37
, which suggests that the miRNA 3ʹ end has limited
contribution to miRNA–target binding.
Alternatively, the precise location of the comple‑
mentary nucleotides may be important. An analysis
of small‑RNA–target site hybrids with two‑nucleotide
symmetrical loops at different positions in the hybrid
indicates that, in the context of miRISC, the free energy
of interaction increases by up to 3 kcal mol
–1
when
the mismatches are in the seed region (particularly at
nucleo tides 5 and 6 of the small RNA) and by <1 kcal mol
–1
when the mismatches occur in the 3ʹ half of the small
RNA
37
. Thus, one or two defects in seed pairing might be
tolerated, particularly when the pairing in the 3ʹ region
is extensive. However, the type of defect and its precise
position within the seed sequence are important
37,65
,
and sites with <5 nucleotides of complementarity in the
seed region are unlikely to be bound by miRISC, as ini‑
tially suggested
7
. Nonetheless, CLASH data suggest that
some mi RNAs, such as miR‑92a, primarily interact with
targets through their 3ʹ end rather than the 5ʹ end
22
.
Individual requirements of specific AGO proteins
have also been reported. For example, Ago2 in flies
requires more extensive miRNA–target complementarity
than mouse AGO2, and the outcome of the interaction
is generally mRNA cleavage. By contrast, mouse AGO2
requires much less stringent base‑pairing of the miRNA
3ʹ end than fly Ago2, and it has similar rates of target
dissociation and cleavage
37
. Of note, although the idea
that the extent of pairing to the 3ʹ end of the miRNA cor‑
relates with the extent of gene regulation by the miRNA
might be attractive, it has not gathered experimental
support. Targets that extensively pair contiguously with
the 3ʹ end of a miRNA are very rare
67
. Furthermore,
experimental measurements of the contribution of the
miRNA 3ʹ end to the energy of interaction, which could
be correlated with measurements of changes in target
expression, have not been available until very recently
37
.
REVIEWS
4
|
ADVANCE ONLINE PUBLICATION www.nature.com/reviews/genetics
© 2014 Macmillan Publishers Limited. All rights reserved
Crosslinking,
immunoprecipitation and
sequencing of hybrids
(CLASH). An experimental
method to isolate RNAs that
interact by hybridization
in a ribonucleoprotein
complex. The complex is
immunoprecipitated with a
specific antibody, the RNA is
partially fragmented,
interacting RNAs are ligated
and the resulting chimeric
products are sequenced.
The degree of evolutionary conservation of the seed
match remains the single strongest indicator of the
functionality of miRNA target sites
11
, and methods that
simply evaluate the selection pressure on seed comple‑
mentary sites repeatedly showed the best performance
in comparative analyses
15,68,69
. As mentioned above, the
strongest conservation signal has been reported for
sites that are 7–8 nucleotides long
11,13,14
. However, can
the small change that is induced by a miRNA in the
expression of a typical mRNA target confer a sufficient
selective advantage on the organism so that the target
site remains conserved through evolution? This question
remains unanswered, and there is a wealth of discussions
around this apparent paradox and the possible alternative
functions of mi RNAs
70
.
miRNA seed families. The importance of the miRNA
seed region for target selection is further emphasized
by the observation that most of the mRNA expression
changes that are induced by small interfering RNAs
(siRNAs) can be explained by siRNA seed‑dependent
targeting of mRNAs
71
. Consistently, mi RNAs with almost
identical sequences at their 5ʹ ends, which form miRNA
seed families, share targets. For example, within the cell‑
cycle‑related gene expression networks, miR‑15a and
miR‑16 regulate the G0/G1‑to‑S phase transition
45
;
miR‑17, miR‑20 and miR‑106 target the cyclin‑dependent
kinase inhibitor 1A (CDKN1A; also known as p21 or
CIP); and miR‑221 and miR‑222 target CDKN1B (also
known as p27Kip1) and CDKN1C (also known as
p57Kip2)
72
. Beyond these examples, 64% of the human
mi RNAs that are currently listed in miRBase
73
are part
of multimember seed families, which raises the follow‑
ing question: what function would so much redundancy
serve? The answer is unlikely to be that co‑expression
of seed‑related mi RNAs induces a stronger down‑
regulation of their common targets because a similar
level of target downregulation could presumably be
achieved through the increased transcription of a sin‑
gle miRNA gene. Some seed‑related mi RNAs have dis‑
tinct domains of expression and thereby repress their
common targets in different tissues. For example, in
mice, the retinoblastoma‑like 2 (RBL2) protein, which
has a crucial role in cell division, is regulated by the
miR‑290~295 cluster in embryonic stem cells (ESCs)
74,75
,
and by the miR‑17~92a cluster during adipogenesis
76
.
The miR‑290~295 cluster, which is also known as an early
embryonic miRNA cluster, is evolutionarily conserved
in placental mammals
77
. Its homologue in humans is the
miR‑371~373 cluster. Interestingly, the AAGUGC seed
sequence in the miR‑290~295 cluster is also present in
many other mi RNAs that are expressed in embryonic tis‑
sues in various species and that seem to have independ‑
ent evolutionary origins. Among these are the miR‑302
cluster in humans
78
, the miR‑430 cluster in zebrafish
79,80
and miR‑427 in frogs
81
. The AAGUGC sequence that
occurs at positions 2–7 of the miR‑290~295 cluster in
mice occurs at positions 3–8 in other miRNAs, such as
miR‑427 in frogs and miR‑17 and miR‑20a in mam‑
mals. An alternative hypothesis for the high preponder‑
ance of seed‑related mi RNAs is that differences in the 3ʹ
ends of these mi RNAs impart subtle differences in their
sets of targets or in the dynamics of target regulation.
It remains to be determined whether the seed‑related
mi RNAs have identical target specificities — in which
their differential expression would perhaps lead to dif‑
ferent sets of mRNAs being targeted in different tissues
— or different target specificities that are mediated by
distinct 3ʹ ends. An additional layer of complexity arises
from the fact that miRNA genes are frequently organized
in polycistronic clusters
82
(BOX2) from which up to tens
of mi RNAs are co‑expressed
79,83,84
. The evolution of these
clusters is likely to have involved gene duplications, the
result of which is that clustered mi RNAs frequently form
seed families
78–80,85,86
. Clustered mi RNAs seem to evolve
more rapidly than individual mi RNAs
87
, which may be
due to each of the related mi RNAs that are co‑expressed
from a cluster undergoing reduced negative selection.
Alternatively, the seed‑related mi RNAs might be sub‑
jected to positive selection to specialize towards distinct
subsets of targets and thereby provide a ‘fine‑tuned’ level
of regulation.
Beyond the seed: type and prevalence of non-canonical
sites. Of the ~18,000 PAR‑CLIP sites reported
19
, only
~15% had a perfect match to nucleotides 2–8 of one of
the 100 most abundant mi RNAs. Interestingly, a simi‑
larly low prevalence (20%) of canonical miRNA–mRNA
interactions was inferred from CLASH data
22
. The rest
of the sites may have resulted from the nonspecific
interaction of AGO with mRNAs, from low‑abundance
guiding mi RNAs and non‑canonical miRNA–target
interactions, or from binding sites of other proteins
that have been inadvertently captured in the experi‑
ment. Non‑canonical binding modes — that is, those
with bulges or single‑nucleotide loops in the miRNA
seed region (FIG.2) — were significantly enriched in
the PAR‑CLIP data
19
and constituted 6–7% of the cap‑
tured sites. Most frequently, the fifth nucleotide of the
miRNA was predicted to be looped out (FIG.2b). HITS‑
CLIP data obtained from neurons
20
and lymphocytes
21
revealed another frequent non‑canonical binding mode,
in which the mRNA nucleotide located between those
that paired with the fifth and sixth miRNA nucleotides
was bulged out (FIG.2c). This represented 15–40% of the
sites captured by CLIP, depending on the data set. It was
hypothesized that the bulged nucleotide contributes to
nucleating the interaction between the target site and
nucleotides 2–6 of the miRNA, which are exposed to the
solvent. Subsequently, base pairing propagates to the rest
of the duplex, and the ‘pivot’ nucleotide can loopout.
Evidence that the prevalence of non‑canonical bind‑
ing is related to the relative abundance of the miRNA
— in which more abundantly expressed mi RNAs have a
larger proportion of non‑canonical sites — was provided
by one study
9
, which further found that >25% of AGO2
CLIP sites are non‑canonical. Of note, in this study
9
the 6‑nucleotide marginal sites
15
, which only pair with
nucleotides 2–7 of the miRNA, were counted as non‑
canonical. Thus, among the 1,095 AGO‑binding
non‑canonical sites predicted, three types of base‑
pairing in the miRNA seed region were observed: 273
REVIEWS
NATURE REVIEWS
|
GENETICS ADVANCE ONLINE PUBLICATION
|
5
© 2014 Macmillan Publishers Limited. All rights reserved
sites (25%) showed base‑pairing only over miRNA
nucleotides 2–7, 307 sites (28%) formed G·U wobble
base pairs, and 515 sites (47%) formed hybrids in which
nucleotides were looped out in the mRNA, the miRNA
or both. Other examples of non‑canonical binding sites
have been reported
26,88
, some of which even showed
base‑pairing with the central region of the miRNA
89
.
Consistent with their lower predicted affinity for
miRISC, non‑canonical sites seem to be overall less
effective in mRNA destabilization than canonical sites
9
.
However, as the accuracy of the methods for isolating
miRNA‑binding sites and miRNA–target hybrids is still
limited, it remains unclear what fraction of the non‑
canonical interactions that were predicted on the basis
of CLIP studies do indeed bind to mi RNAs as opposed
to being isolated in a nonspecific manner. As various
functionally relevant non‑canonical targets are known
— most notably, the initially described lin‑14 (REFS1,6)
and lin‑41 (REF.4) in the worm, which are essential for
development — the prevalence and the regulatory
mechanisms of non‑canonical sites remain of interest
and should be resolved in future studies.
Box 2 | Clustered mi RNAs target functionally related genes
As clustered microRNAs (mi RNAs) are co-expressed, one might expect that they jointly regulate molecular pathways either
by co-targeting individual genes or by targeting different components of the same pathway. Indeed, in Drosophila
melanogaster it has been shown that mi RNAs that are expressed from the same cluster have a small but significant
tendency to co-target the same genes
131
. Whether clustered mi RNAs target distinct components of the same molecular
pathway has been less studied, although some evidence suggests this to be the case in humans. For example, one of the
most studied mammalian miRNA clusters is the broadly expressed oncogenic miR-17~92a cluster
132
, which contains six
mi RNAs with four distinct ‘seed’ sequences (see the figure, parts a and b; seed sequences are highlighted in red). We
determined the 50 highest scoring sites for each of the unique seed sequence of mi RNAs of the miR-17~92a cluster (that is,
miR-17, miR-18a, miR-19a and miR-92a-1) from the predictions made in a Argonaute 2 (AGO2) crosslinking and immuno-
precipitation (CLIP) study
9
and used the search tool for the retrieval of interacting genes/proteins (STRING) database
133
to
identify functional interactions among the corresponding genes. The main connected component of the graph of
functional interactions between targets of the four mi RNAs mentioned above is shown (see the figure, part c).
The miRNA that targets each gene is reflected in the colour of the circle representing the gene. Red arrows indicate the
genes that are co-targeted by multiple mi RNAs of this cluster. Considering that we started with <1% of the human genes,
the large cluster of functionally related genes identified by STRING is surprising. The multipronged targeting of cell
proliferation by the miR-17~92a cluster is immediately apparent. Moreover, as in flies
131
, these mi RNAs have a small
tendency to co-target the same genes.
Nature Reviews | Genetics
miR-18a
miR-19a
miR-92a-1
CAAAGUGCUUACAGUGCAGGUAG
UAAGGUGCAUCUAGUGCAGAUAG
UGUGCAAAUCUAUGCAAAACUGA
UAUUGCACUUGUCCCGGCCUGU
miR-17
ba
17
18a
19b-119a
20a 92a-1
CDC27
RAB5A
ARF6
MCOLN2
PIP5K1C
PTEN
ANKFY1
TPR
ARID4B
NOP58
EIF2S1
RBM25
EIF4G2
ESCO2
BTF3L4
NIP7
CCT6A
SIN3B
CCNT2
PARD6B
TTK
CCND2
POLQ
AURKA
DYRK2
ZNF217
NR3C1
TSC22D3
TNRC6B
PHF12
MDM2
TBRG1
TRIP13
CEP350
MCL1
NAA50
NUP43
SMARCA2
USP14
E2F1
NCOA3
DICER1
c
TGIF1
REVIEWS
6
|
ADVANCE ONLINE PUBLICATION www.nature.com/reviews/genetics
© 2014 Macmillan Publishers Limited. All rights reserved
The effect of sequence context on target site functionality.
Whether a transcript will respond to changes in miRNA
concentration cannot be predicted solely on the basis
of the degree of complementarity between the miRNA
and the putative target sites in the transcript, which indi‑
cates that additional factors contribute to functionality.
One such factor is the location of miRNA comple‑
mentary sites within transcripts. miRNA‑binding sites
are fairly depleted in coding regions and up to ~15
nucleotides downstream of the stop codon, presumably
owing to the translating ribosomes that hinder stable
miRNA–transcript association
8
. Additionally, miRNA
target sites that are under strong evolutionary selection
are preferentially located at the beginning or the end of
long 3ʹUTRs
8,13,90
. It has been proposed that the sequence
around the miRNA‑binding sites that emerged early in
evolution (that is, those in short 3ʹUTRs) is optimized to
enable efficient targeting by mi RNAs
13
. Thus, new sites
that emerged in this sequence environment may have
had a larger functional impact and have therefore been
preferentially selected as the length of 3ʹUTRs increased
in evolution. Interactions between regulatory complexes
or subunits may also impose constraints on the spatial
positioning of various regulatory sites. For example,
a recent study found that the recruitment of the RNA
helicase eIF4A2 at structured 5ʹUTRs is necessary for
miRNA‑induced translational repression
91
.
In spite of the unfavourable effect of translating
ribosomes on AGO binding
92
, CLIP studies identified
similar absolute numbers of miRNA‑binding sites,
albeit in lower density, in coding regions compared with
3ʹUTRs
18,19,53,54
. miRNA‑binding sites in coding regions
have a smaller impact on mRNA stability than those in
3ʹUTRs
55,93,94
, but they seem to be effective in inhibit‑
ing target translation
55
and contribute significantly,
along with the miRNA‑binding sites in 3ʹUTRs, to the
repression of target genes
55,93
.
Another important determinant of the response
of mRNAs to mi RNAs
95
and siRNAs
96
is the energy
required for the target site to acquire a single‑stranded
conformation (that is, the accessibility of the target site).
As expected, target site accessibility affects the binding
of miRISC to mRNAs but not the rate of target degrada‑
tion
69
. By contrast, the degree of mRNA destabilization
seems to be influenced by the adenine or uracil content
around the target sites
8,69,97
: A‑rich motifs are associated
with stabilization
98
and U‑rich motifs with destabiliza‑
tion
69,98
of mRNAs upon miRNA transfection. Proteins
that bind to A‑rich or U‑rich sequence elements have
been found to modulate specific miRNA–target inter‑
actions. For example, deadend protein homologue1
(DND1) hinders the association of mi RNAs with tar‑
get sites
99
, ELAV‑like protein 1 (ELAVL1; also known
as HuR) releases the SLC7A1 transcript (which encodes
cationic amino acid transporter 1) from miRNA‑
mediated repression
100
, and Pumilio homologue1
(PUM1) induces a structural change that promotes
the association of the p27 transcript with mi RNAs and
thereby its repression
101
. Intriguingly, AGO itself has
recently been reported to have a preference for A‑rich
sequence elements
102
.
Nature Reviews | Genetics
A
UGAG GUAG GGUUGUAUAG
UUU
U
G
UUUA
5′
Frequent symmetrical loop sequences
miRNA
mRNA
Seed
miRNA
mRNA
Paired
pivot
miRNA
mRNA
Nucleation
let-7
a
b
c
lin-41
ACUC CAUC CCAACAUAUU
G
U
U
A
3′
lin-41 CDS
Propagation
of base pairing
U G U G
or or or
5′
3′
5′
3′
5′
3′
5′
3′
5′
3′
3′
5′
5′
3′
Seed
Bulged pivot
Symmetrical
loop
Figure 2 | Examples of ‘non-canonical’ miRNA–
target interactions. a| A let‑7‑binding site in the
lin‑41 mRNA is shown
4
. b| Binding sites that are
predicted to lack Watson–Crick pairing of the fifth
microRNA (miRNA) nucleotide were found to be
enriched in photoactivatable ribonucleoside-enhanced
crosslinking and immunoprecipitation (PAR-CLIP) data
19
.
The nucleotides that are commonly found in this
symmetrical loop are also shown. c| The ‘pivot’
nucleotide in the mRNA is involved in the nucleation of
the miRNA–target interaction by contributing to the
contiguous complementarity to nucleotides 2–6 of
the miRNA, and it subsequently bulges out between the
mRNA nucleotides that pair with the fifth and sixth
nucleotides of the miRNA
20
. miRNA nucleotides are
shown as filled circles and mRNA nucleotides as open
circles. The nucleotides that are involved in base-pairing
interactions in the ‘seed’ region are shown in blue, and
those that do not form Watson–Crick base pairs in the
final miRNA–target site hybrid are shown in red. CDS,
coding DNA sequence.
REVIEWS
NATURE REVIEWS
|
GENETICS ADVANCE ONLINE PUBLICATION
|
7
© 2014 Macmillan Publishers Limited. All rights reserved
Various target prediction programs that combine
the determinants described above have been devel‑
oped (see REF.68 for a recent assessment of algorithm
performance and REF.103 for additional online tools).
Consequences of miRNA–target interactions
Post-transcriptional repression of gene expression.
Destabilization of mRNAs and translational repres‑
sion are the best‑studied outcomes of miRNA–mRNA
interaction, and both lead to a decrease in the levels of
the encoded proteins (FIG.3A). The underlying molecu‑
lar mechanisms have been extensively reviewed else‑
where
5
and are not further discussed here. Notably,
it remains a challenge to relate the targets identified
using various methods to the behaviour and survival
of cells and of the entire organism. As a class, predicted
miRNAs targets show significant changes in response
to knockdown or overexpression of cognate mi RNAs.
However, the change of any individual target is typi‑
cally small, and the magnitude of repression is rarely
higher than 2–3 fold (FIG.3A), at least in miRNA trans‑
fection experiments. Such small changes are difficult
to reconcile with the phenotypes of mi RNAs, at least
in some contexts, examples of which include embry‑
onic lethality in nematodes
1
, mice
24
and sea urchins
104
;
reduced size in flies
40
; and abnormal morphogenesis
and brain development in zebrafish
80
.
A possibility to consider is that various factors
may obscure the effect of mi RNAs on the mRNA
and protein levels of their targets within the time‑
frame of the experiment. For example, as the rate of
small RNA loading into the AGO protein seems to be
~10hours
49,105
and the median half‑life of mammalian
proteins is ~48hours
106
, the magnitude of protein‑level
changes that can be observed among miRNA targets
depends on the design and timing of the experiment
49
.
Feedback mechanisms that counteract the effect
of mi RNAs may also be at work. A well‑documented
example is that of Dicer, which is a key enzyme
in miRNA biogenesis and is under the control of the
let‑7 miRNA
46
. Another feedback mechanism involves
the trinucleotide repeat‑containing gene 6A (TNRC6A)
protein, which is part of the miRNA effector pathway
and seems to be regulated by the miR‑30 family of
mi RNAs
69
. Feedback loops that involve expression
of transcriptional regulators in a tissue‑specific man‑
ner may also provide an explanation for differences
in miRNA targeting between tissues, as does the pro‑
duction of target isoforms that have differential sus‑
ceptibility to miRNA regulation through the use of
tissue‑specific polyadenylation sites
107
.
More recent hypotheses propose that mi RNAs
affect gene expression in alternative ways other than
by downregulating protein abundance (see below).
Increasing the precision of target gene expression. It
has long been recognized that transcription factors are
the preferred targets of mi RNAs
108
. In gene expression
regulatory networks, mi RNAs and transcription factors
often form feed‑forward loops (FFLs), in which a com‑
mon target is regulated by both a transcription factor
Nature Reviews | Genetics
ORF
mRNA
Pseudogenes
Competing mRNA
AAAAA
ORF
AAAAA
AAAAA
circRNA
miRNA
miRNA
lncRNA
Fold regulation by miRNA
0.1 0.2 0.5 2.01.0
A
Wrinkled
Vha68-1
HMGA2
RECK
miRNA transfections
miR-223 knockout
MZdicer
Protein
mRNA
Transcription rateTime
Bb Threshold-linear response
in target expression
+ miRNA– miRNA
Ba Reduction of target expression noise
Target gene expression
Target gene expression
C
miRNAs
Figure 3 | Consequences of miRNA–target interactions. A| Target mRNA‑ and
protein-level responses to changes in microRNA (miRNA) expression are shown; bars
cover the extent of regulation of 90% of genes with a 7- or 8-nucleotide match to the
5ʹ end of the targeting miRNA. The regulation of Wrinkled (also known as hid) by bantam
and of Vha68‑1 by both miR-9b and miR-279 was measured in miRNA-expressing
Drosophila melanogaster S2 cells
136
. The regulation of HGMA2 (high mobility group
AT-hook 2) by let-7 and of RECK (reversion-inducing cysteine-rich protein with kazal
motifs) by miR-21 was similarly measured in a mammalian system
137
. mRNA- and
protein-level changes were measured using microarrays and mass spectrometry
following the transfection of let-7, miR-155, miR-1, miR-16 or miR-30 (REF.46), and
similar quantifications were carried out following the knockout of miR-223 in mouse
neutrophils
47
. Gene expression differences between wild-type zebrafish and MZdicer
animals (which are deficient in miRNA biogenesis) were quantified using RNA
sequencing and ribosome fragment-protected sequencing
138
. Here, the blue bar
represents the translational output rather than abundance of the protein. B| Other
consequences of miRNA interaction with their targets include reduction in the
cell‑to‑cell variability (that is, ‘noise’) of target gene expression (part Ba) and
introduction of thresholds in the expression of miRNA targets as a function of the target
transcription rates (part Bb). C| miRNA ‘sponging’ is shown. Pseudogenes, linear long
non-coding RNAs (lncRNAs), circular RNAs (circRNAs) or other mRNAs can function as
miRNA decoys to sequester mi RNAs from their target mRNAs. ORF, open reading frame.
REVIEWS
8
|
ADVANCE ONLINE PUBLICATION www.nature.com/reviews/genetics
© 2014 Macmillan Publishers Limited. All rights reserved
and a miRNA, and the transcription of the miRNA
itself is regulated by the same transcription factor
109–111
.
In ‘coherent’ FFLs, transcription factors regulate their
targets both directly and indirectly (through mi RNAs)
in a consistent manner. Through this mechanism,
mi RNAs are thought to counteract ‘leaky’ transcription
and to enforce spatial domains of target gene expres‑
sion that are initially defined by transcription factors
110
.
In ‘incoherent’ FFLs, the direct effects of transcription
factors on target genes are counter‑balanced by the
indirect effects of the transcription factor‑induced
mi RNAs. Such networks are thought to be effective in
‘noise’ buffering
110
. A study of the dynamics of inco‑
herent FFLs recently found that, in contrast to regu‑
latory circuits composed only of transcription factors
and protein‑coding genes, FFLs buffer fluctuations
in target gene expression that are induced by fluc‑
tuations in upstream regulators
34
(FIG.3Ba). Moreover,
this model predicted that an optimal reduction
in fluctuations is accompanied by a small reduction in
target gene expression. A study of miRNA–transcrip‑
tion factor FFLs identified hundreds of such networks
in humans
112
; thus, buffering transcriptional noise
through FFL dynamics may indeed be the main activity
of mi RNAs on a subset of their targets. However, this
network motif does not explain why the response of the
targets remains small when expression of the miRNA,
but not that of the transcription factor, is perturbed.
Another more general mechanism through which
small RNAs may act to buffer stochastic fluctuations
in and to increase the precision of the expression of
their targets was initially described in a study of small‑
RNA‑mediated gene regulation in bacteria
32
. An essen‑
tial aspect of this mechanism is that the binding of the
small RNA to the target does not sequester the target
from degradation. When the target is transcribed at a
low rate or as a result of transcriptional noise, the small
number of resulting mRNA molecules are bound by the
small RNA and degraded. By contrast, when the tran‑
scription rate is sufficiently high, the mRNA can ‘escape’
from the small‑RNA‑induced degradation and begin
to accumulate in proportion to the transcription rate
(FIG.3Bb). This behaviour, which is known as the ‘thresh‑
old‑linear’ response of the target to its transcriptional
induction, has also been reported for miRNA‑sensitive
reporters in mammalian systems
33
. The transition
between the two regimes is more abrupt when the tar‑
gets have high affinity for the miRNA and when the
miRNA concentration is high. Furthermore, the loca‑
tion of the threshold depends on the presence of other
miRNA targets.
Although the miRNA‑dependent regulatory net‑
work is complex, it is becoming possible to study its
dynamics by combining measurements of the miRNA
and target mRNA expression in single cells
36
with
measurements of their interaction affinities and of
the rates of various miRNA‑dependent responses. The
affinities of interaction between mi RNAs and their
targets can be measured invitro
37
or predicted using
biophysical models, such as that inferred from CLIP
data
9
(BOX3). Overall, the hypothesis that miRNAs
have a general function in increasing the precision of
target gene expression by introducing nonlinearity in
the response of targets to changes in transcription rate
seems compelling.
Box 3 | A model of miRNA-dependent regulation of mRNA levels
A general model describing the interaction of microRNA-induced silencing complex
(miRISC) with mRNA targets and the consequences of this interaction at the
mRNA level can be formalized as follows
32,123–125
. For a selected target (T) of a microRNA
(miRNA), let α be the rate of transcription, β be the rate of association to miRNA-loaded
Argonaute (AGO), ρ be the rate of dissociation from AGO, μ be the rate of decay when it
is not bound to AGO and mμ be the rate of decay when it is bound to AGO. Let us
further consider the population M of all of the targets (except T) of the miRNA, and
denote the overall rates of transcription, association with and dissociation from AGO
of these other targets as αʹ, bβ and rρ, respectively. Assuming that the total amount of
miRNA-loaded AGO (A
Σ
) does not change and denoting the AGO-associated selected
target and other targets by A
T
and A
M
, respectively, the dynamics of this system can be
described by the following differential equations:
(1)= α – μT – β(A
Σ
– A
T
– A
M
) T + ρA
T
dT
dt
(2)= β(A
Σ
– A
T
– A
M
) T – ρA
T
– mμA
T
dA
T
dt
(3)= αʹ – μM – bβ(A
Σ
– A
T
– A
M
) M + rρA
T
dM
dt
(4)= bβ(A
Σ
– A
T
– A
M
) M – rρA
M
– mμA
M
dA
M
dt
We can determine the steady state of this system numerically by determining the
values of the variables at which the time derivatives for all variables are zero. Although
the system may have up to four steady states, enforcing positive molecular counts for all
four species of the model allows the selection of the biologically relevant solution.
Most of the parameters of this model can be estimated from available experimental
data (see the table).
Parameter Value Ref
α
Variable NA
μ
0.1 per hour 106
β
0.024–0.24 per molecule per hour* 37
A
Σ
100,000 molecules for all mi RNAs
‡
134
ρ ~2.16 per hour
§
37
m
1.55
||
47
αʹ 1 molecule per hour
¶
36
b 1
#
37
r 1 when M carries a ‘canonical’ site, and 40 when M
carries a ‘non‑canonical’ site
37
NA, not applicable. *With k
on
~ 2 × 10
7
M
–1
s
–1
and assuming that the volume of a HeLa cell is
5,000 μm
3
, β = 0.024 per molecule per hour. Assuming that the effective volume is only the
volume of the endoplasmic reticulum
135
, which represents 10% of the cell volume, β = 0.24 per
molecule per hour.
‡
The total number of AGO molecules per cell is ~100,000 (REF.134). The
number of these molecules that are loaded with an individual miRNA must vary depending on
the relative abundance of the miRNA. Very abundant mi RNAs occur at tens of thousands of
copies per cell. mi RNAs with frequency in the range of 1% of the total miRNA population
would have an A
Σ
~ 1,000 molecules.
§
Based on a k
off
~ 6 × 10
−4
s
–1
(REF.37).
||
This value can be
obtained by modelling the change in expression of miR-223 targets upon miR-223 knockout
47
,
assuming that miR-223 represents 5% of the mi RNAs in neutrophils and that the miRNA has
~1,000 targets that occur at 10 copies per cell on average and are upregulated by 37% upon
knockdown of the miRNA.
¶
The average abundance of an mRNA varies between cell types from
~1 copy per cell in embryonic stem cells to ~10 copies per cell in mouse embryonic fibroblasts.
#
Similar rates of association of different miRNA targets with AGO are assumed.
REVIEWS
NATURE REVIEWS
|
GENETICS ADVANCE ONLINE PUBLICATION
|
9
© 2014 Macmillan Publishers Limited. All rights reserved
Long non-coding RNA
(lncRNA). An RNA molecule
that is generally longer than
structural RNAs (such as
tRNAs, small nuclear RNAs and
small nucleolar RNAs) and that
does not encode proteins.
Further subcategories are
distinguished depending on
the type of genomic regions
from which they derive. One
example is the long intergenic
non-coding RNAs (lincRNAs)
that are transcribed from
genomic regions between
protein-coding gene loci.
Circular RNAs
(circRNAs). Very stable RNAs
with circular structures that
result from the ligation of
the 3ʹ ends to the 5ʹ ends, for
example, of exons. The circRNA
CDR1 antisense RNA (CDR1as)
has recently been found to
function as a miRNA ‘sponge’.
miRNA ‘sponging’. We have so far discussed the conse‑
quences of miRNA–mRNA interactions on the mRNA.
However, mi RNAs can themselves be sequestered
and neutralized by the targets with which they inter‑
act (FIG.3C). A study provided computational support
to this idea by showing that the ability of mi RNAs to
repress their targets decreases with the number and
the abundance of their targets
113
. The possibility then
emerges that miRNA activity is regulated by specific
‘competing’ RNAs, the main function of which is to
‘sponge’ mi RNAs away from their protein‑coding tar‑
gets
114
. This mechanism is exploited, for example, by
herpesvirus saimiri, which encodes U‑rich RNAs that
induce degradation of the host miR‑27, thereby affect‑
ing the expression of miR‑27 targets
115
. Degradation of
mouse miR‑27a and miR‑27b is induced by the m169
transcript of murine cytomegalovirus
116
. Pseudogenes
such as PTENP1 (phosphatase and tensin homologue
pseudogene 1) and KRASP1 also seem to have evolved
this function and are able to relieve their protein‑coding
homologues from miRNA‑mediated repression
35
.
Similarly, it has been reported that the long non-coding
RNA (lncRNA) LINC‑ROR sponges out miR‑145, thereby
contributing to the self‑renewal of human ESCs
117
, and
that the lncRNA H19 sponges out let‑7 to promote
muscle differentiation
118
. Although they interact with
the mi RNAs through non‑canonical binding sites, both
of these lncRNAs are expressed at much higher levels
than the average of mRNA expression. For example,
whereas mRNAs are expressed with a median of 2 copies
per ESC
36
, LINC‑ROR occurs at ~100 copies per cell
117
.
The clearance of miR‑145 seems to be important for
ESC self‑renewal, even though this miRNA is present
at only 10–20 copies per ESC
117
. For comparison, the
hepatocyte‑specific miR‑122 carries out its function at
10
4
‑fold higher expression levels (1.20–1.35 × 10
5
cop‑
ies per cell in human primary hepatocytes
119,120
), and
comparable concentrations of a competitor RNA are
necessary to relieve miR‑122 targets from miRNA‑
mediated repression. Interestingly, a recent study found
that the derepression of miR‑122 targets was observed at
similar concentrations of competing RNAs for miR‑122
concentrations that vary over a 100‑fold range
120
.
Owing to their high stability, circular RNAs (circRNAs)
that are derived from back splicing of 3ʹ exons onto
5ʹ exons seem to be especially suited for the sponging
function. An extreme case was discovered in CDR1
antisense RNA (CDR1as), which is a circRNA encoded
by the opposite strand of the cerebellar degeneration‑
related protein 1 (CDR) gene. This circRNA contains a
large number of evolutionarily conserved miR‑7‑binding
sites (63 in total) and has been shown to suppress miR‑7
activity in the brain of mice and zebrafish
121,122
. Whether
other circRNAs have such a miRNA sponging function
remains to be determined.
Thus, whereas mi RNAs might set thresholds that
ultimately increase the precision of target gene expres‑
sion, some miRNA targets evolved to ensure rapid
and efficient sequestration of specific mi RNAs during
particular transitions between cell states. To obtain a
quantitative understanding of this complex regulatory
network and its dynamics, computational models need
to be developed.
Modelling crosstalks in miRNA–target networks
Several recent studies have investigated how the
altered expression of individual transcripts might affect
the expression of other transcripts by miRNA spong‑
ing
123,124
. Work on small‑RNA‑based regulation in bac‑
teria
32
and molecular titration
125
suggests that the highest
impact of an mRNA on the expression of another mRNA
would occur when the concentrations of mi RNAs and
target transcripts are comparable. It remains unclear
whether this is the typical regime in which mi RNAs
operate invivo, although a recent study suggested that
target sites typically outnumber the cognate mi RNAs
120
.
Additionally, targets with disparate affinities for miRISC
are predicted to have asymmetrical effects on each
other
123
: a target with high affinity to miRISC, which
would be bound at low miRISC concentration, could
influence the expression of both high‑ and low‑affinity
targets, whereas a low‑affinity target would only start to
bind to miRISC at miRISC concentrations at which the
higher‑affinity targets are already bound and would not
be expected to influence their expression.
We summarized these hypotheses using a generic
model of gene expression in which a miRNA binds to
mRNA targets and to a lncRNA competitor, both of
which decay faster when bound to the miRNA (BOX3).
We considered three parameter regimes that corre‑
spond to concentrations of miRNA‑primed miRISC
that are similar to, or tenfold or hundred‑fold higher
than the concentration of the mRNA targets (FIG.4a). If
the miRNA‑primed miRISC is initially in high excess
compared with the mRNA targets, then expression of the
competing lncRNA can strongly upregulate the mRNA
targets. However, if the initial concentration of the
miRNA‑primed miRISC is comparable to the concen‑
tration of its mRNA targets, then the lncRNA will have
fairly little influence on the expression of these mRNA
targets. Not shown is the dynamics expected when
the miRISC abundance is already lower than the target
mRNA abundance before the lncRNA is induced. In
this case, the upregulation of the mRNA upon lncRNA
induction is marginal (8% when the RISC:mRNA
molecule ratio is 1:10). A competitor that contains
low‑affinity binding sites for the miRNA is predicted
to have a similarly low impact (FIG.4a). The threshold‑
linear response of the competitor as a function of its
transcription rate is illustrated in FIG.4b. As its tran‑
scription rate increases, the lncRNA starts to associate
with miRISC and displaces the mRNAs. The transition
between a state in which miRISC mostly associates with
mRNAs and a state in which it is mostly associated
with lncRNA molecules occurs when the lncRNA
reaches a concentration comparable to that of the
miRNA‑loaded miRISC (which was initially in excess of
the concentration of mRNA targets). After the miRNA‑
loaded miRISC is saturated with lncRNAs, the mRNAs
are relieved of suppression and reach a steady‑state level
that depends on their own transcription rate and on
their miRNA‑independent rate ofdecay.
REVIEWS
10
|
ADVANCE ONLINE PUBLICATION www.nature.com/reviews/genetics
© 2014 Macmillan Publishers Limited. All rights reserved
The application of this analysis to the various sys‑
tems in which the expression of a miRNA target has been
found to influence the expression of other targets could
reveal novel aspects of the miRNA interaction network
in finer quantitative detail. In particular, the fact that a
significant upregulation (by ~40% upon transfection of
LINC‑ROR) of miR‑145 targets has been observed when
the miRNA is only present at 10–20 copies per cell
117
could indicate that the population of miRNA targets is
also in this range of expression levels, which is smaller
than that expected on the basis of current computational
predictions. Similarly, the model predicts that the com‑
peting RNA concentration at which the derepression of
endogenous targets starts to be observed is fairly insensi‑
tive to miRISC abundance, as long as the miRISC is not
in large excess compared with the targets. This is consist‑
ent with recent experimental observations
120
. The model
further predicts that as the concentration of the com‑
peting RNA increases, the derepression of endogenous
targets approaches a limit that depends on and that could
thereby inform about both the miRISC:endogenous tar‑
get abundance ratio and the miRNA‑dependent rate of
target degradation.
Conclusions and perspectives
New mechanisms have come to light, which can relate
the increased rate of miRNA‑induced mRNA degra‑
dation to phenotypes that are observed at cellular and
organismal levels. Specifically, it has been reported
that mi RNAs establish thresholds in the response of
their targets to transcriptional induction
32,33
, reduce the
cell‑to‑cell variability of target gene expression
34
and
induce correlations between the expression of various
targets within individual cells
35
. Which of these mecha‑
nisms is relevant in a particular context is an essential
yet difficult question to answer because the underlying
interaction networks are large, complex and only par‑
tially known. Thus, much of the current debate in the
field oscillates between defining (and redefining) what a
miRNA target is and determining the appropriate read‑
out of miRNA–target interactions, while taking into
account that the impact of a miRNA on individual targets
depends on many dynamic factors. Among these factors
are the cellular localization of miRNAs and their targets,
their relative concentrations and the context‑specific
effects of other regulators, including transcription
factors and RNA‑binding proteins
99–101,107,113,126–128
.
With the availability of recently developed high‑
throughput approaches for miRNA target identifica‑
tion, our knowledge of the size and connectivity of
the miRNA‑dependent regulatory network continues
to increase. The apparent complexity of this network
raises various fundamental questions. First, what aspect
of mRNA expression do mi RNAs regulate and how
is this achieved? Although mi RNAs were originally
thought to regulate the translation rate of a restricted
number of targets, it soon became clear that mi RNAs
increase the decay rates of a substantial proportion of
transcripts. However, as individual targets typically
show small changes in expression levels, it remains
Figure 4 | Crosstalk between miRNA targets. Sequestration of microRNAs (mi RNAs) by competing long non-coding
RNAs (lncRNAs) releases mRNAs from miRNA-mediated repression. a| The graph shows upregulation of miRNA target
expression as a function of the transcription rate of the lncRNA relative to the situation in which no lncRNA is present.
The transcription rate of the lncRNA increases from a value that would yield 1 copy per cell to a value that would yield
10,000 copies per cell if the miRNA were not expressed. The parameter settings realistically correspond to the
LINC‑ROR system (BOX3), in which there are 10 mRNA targets each occurring at a concentration of 1 copy per cell
36
.
Concentrations of miRNA-induced silencing complex (miRISC) vary between 10 complexes per cell (black; as
measured for miR-145 in embryonic cells
117
), 100 complexes per cell (red) and 1,000 complexes per cell (blue). The
response of mRNA targets when the lncRNA carries a ‘non‑canonical’ site (dashed lines), the off‑rate of which is 40
times higher than the average off-rate of the other targets, is also depicted. b| The graph shows detailed dynamics of
the molecular species in the model for the parameter settings that yield the red solid curve in part a (that is, 100
miRISCs targeting mRNAs and lncRNAs with canonical binding sites). AGO, Argonaute.
Nature Reviews | Genetics
0.1
1
10
100
1,000
1
2
5
10
20
50
100
Fold change in target mRNA level
Molecular abundance (number of copies)
200
Limiting miRISC
Intermediate miRISC
Excess miRISC
Average-affinity
competitor
Low-affinity
competitor
Free mRNA
AGO–mRNA complex
Free lncRNA
AGO–lncRNA complex
lncRNA transcription rate (copies per hour) lncRNA transcription rate (copies per hour)
1e–01 1e+01 1e+03
0.1
0.5 5.0 50.0
ba
REVIEWS
NATURE REVIEWS
|
GENETICS ADVANCE ONLINE PUBLICATION
|
11
© 2014 Macmillan Publishers Limited. All rights reserved
1. Wightman,B., Ha,I. & Ruvkun,G. Posttranscriptional
regulation of the heterochronic gene lin-14 by lin-4
mediates temporal pattern formation in C.elegans.
Cell 75, 855–862 (1993).
2. Lee,R.C., Feinbaum,R.L. & Ambros,V.
The C.elegans heterochronic gene lin-4 encodes
small RNAs with antisense complementarity to lin-14.
Cell 75, 843–854 (1993).
3. Pasquinelli,A.E. etal. Conservation of the
sequence and temporal expression of let-7
heterochronic regulatory RNA. Nature 408, 86–89
(2000).
4. Reinhart,B.J. etal. The 21-nucleotide let-7 RNA
regulates developmental timing in Caenorhabditis
elegans. Nature 403, 901–906 (2000).
5. Huntzinger,E. & Izaurralde,E. Gene silencing by
microRNAs: contributions of translational repression
and mRNA decay. Nature Rev. Genet. 12, 99–110
(2011).
6. Ha,I., Wightman,B. & Ruvkun,G. A bulged lin-4/
lin-14 RNA duplex is sufficient for Caenorhabditis
elegans lin-14 temporal gradient formation. Genes
Dev. 10, 3041–3050 (1996).
7. Brennecke,J., Stark,A., Russell,R.B. & Cohen,S.M.
Principles of microRNA–target recognition. PLoS Biol.
3, e85 (2005).
8. Grimson,A. etal. MicroRNA targeting specificity
in mammals: determinants beyond seed pairing.
Mol. Cell 27, 91–105 (2007).
9. Khorshid,M., Hausser,J., Zavolan,M. &
van Nimwegen,E. A biophysical miRNA–mRNA
interaction model infers canonical and
noncanonical targets. Nature Methods 10,
253–255 (2013).
10. Lewis,B.P., Shih,I.H., Jones-Rhoades,M.W.,
Bartel,D.P. & Burge,C.B. Prediction of mammalian
microRNA targets. Cell 115 , 787–798 (2003).
11. Lewis,B.P., Burge,C.B. & Bartel,D.P.
Conserved seed pairing, often flanked by adenosines,
indicates that thousands of human genes are
microRNA targets. Cell 120, 15–20 (2005).
12. Lim,L.P. etal. Microarray analysis shows that some
microRNAs downregulate large numbers of target
mRNAs. Nature 433, 769–773 (2005).
13. Gaidatzis,D., van Nimwegen,E., Hausser,J. &
Zavolan,M. Inference of miRNA targets using
evolutionary conservation and pathway analysis.
BMC Bioinformatics 8, 69 (2007).
14. Krek,A. etal. Combinatorial microRNA
target predictions. Nature Genet. 37, 495–500
(2005).
15. Bartel,D.P. MicroRNAs: target recognition and
regulatory functions. Cell 136, 215–233 (2009).
16. Friedman,R.C., Farh,K.K., Burge,C.B. &
Bartel,D.P. Most mammalian mRNAs are conserved
targets of microRNAs. Genome Res. 19, 92–105
(2009).
17. Kozomara,A. & Griffiths-Jones,S. miRBase:
annotating high confidence microRNAs using deep
sequencing data. Nucleic Acids Res. 42, D68–D73
(2014).
18. Chi,S.W., Zang,J.B., Mele,A. & Darnell,R.B.
Argonaute HITS-CLIP decodes microRNA–mRNA
interaction maps. Nature 460, 479–486 (2009).
19. Hafner,M. etal. Transcriptome-wide identification of
RNA-binding protein and microRNA target sites by
PAR-CLIP. Cell 141, 129–141 (2010).
20. Chi,S.W., Hannon,G.J. & Darnell,R.B.
An alternative mode of microRNA target
recognition. Nature Struct. Mol. Biol. 19, 321–327
(2012).
This study identifies a prevalent type of
non-canonical miRNA-binding sites — namely,
the pivot sites.
21. Loeb,G.B. etal. Transcriptome-wide miR-155
binding map reveals widespread noncanonical
microRNA targeting. Mol. Cell 48, 760–770
(2012).
22. Helwak,A., Kudla,G., Dudnakova,T. & Tollervey,D.
Mapping the human miRNA interactome by CLASH
reveals frequent noncanonical binding. Cell 153,
654–665 (2013).
23. Betel,D., Koppal,A., Agius,P., Sander,C. & Leslie,C.
Comprehensive modeling of microRNA targets
predicts functional non-conserved and non-canonical
sites. Genome Biol. 11 , R90 (2010).
24. Kanellopoulou,C. etal. Dicer-deficient mouse
embryonic stem cells are defective in differentiation
and centromeric silencing. Genes Dev. 19, 489–501
(2005).
25. Yi,R., Poy,M.N., Stoffel,M. & Fuchs,E. A skin
microRNA promotes differentiation by repressing
‘stemness’. Nature 452, 225–229 (2008).
26. Tay,Y., Zhang,J., Thomson,A.M., Lim,B. &
Rigoutsos,I. MicroRNAs to Nanog, Oct4 and Sox2
coding regions modulate embryonic stem cell
differentiation. Nature 455, 1124–1128 (2008).
27. Cheung,T.H. etal. Maintenance of muscle stem-cell
quiescence by microRNA-489. Nature 482, 524–528
(2012).
28. Muljo,S.A. etal. Aberrant Tcell differentiation in the
absence of Dicer. J.Exp. Med. 202, 261–269 (2005).
29. Koralov,S.B. etal. Dicer ablation affects antibody
diversity and cell survival in the B lymphocyte lineage.
Cell 132, 860–874 (2008).
30. Chen,J.F. etal. Targeted deletion of Dicer in the heart
leads to dilated cardiomyopathy and heart failure.
Proc. Natl Acad. Sci. USA 105, 2111–2116 (2008).
31. Poy,M.N. etal. miR-375 maintains normal pancreatic
α- and β-cell mass. Proc. Natl Acad. Sci. USA 106,
5813–5818 (2009).
unclear what fraction of the miRNA‑induced changes
in target mRNA levels have functional consequences.
It may indeed be the case that this fraction is small and
that there are few crucial targets that respond strongly to
the miRNA or that influence phenotypes very sensitively.
Validating this hypothesis would require a substantial
amount of work, including perturbing the expression of
mi RNAs invivo, detecting the phenotypes — which may
be apparent only in specific experimental conditions —
and linking these phenotypes to individual miRNA tar‑
gets. An alternative approach that may be within reach is
to carry out miRNA loss‑ or gain‑of‑function screening
in cell culture systems and to link the observed pheno‑
types to miRNA targets by modelling the phenotypes as
the combined effect of predicted miRNA targets.
Another factor to consider is that the regulatory net‑
work of a miRNA is probably dynamic. As each indi‑
vidual cell expresses only a subset of genes and transcript
isoforms, only a proportion of the miRNA complemen‑
tary sites that are annotated transcriptome wide will
be present and relevant in any given cell. Furthermore,
the various miRNA‑binding sites are likely to differ in
their affinity for the miRNA‑loaded AGO protein, and
different sites are likely to be saturated at different con‑
centrations of miRISC. Finally, RNA‑binding proteins
modulate the accessibility of individual sites in a tissue‑
specific manner. Experiments that take advantage of
rapidly emerging technologies for studying gene expres‑
sion in single cells will help to quantify the effective
number of miRNA‑binding sites in individual cells and
bring new insights into its dynamics.
Initially derived from computational studies of small‑
RNA‑mediated gene regulation in bacteria, the hypothesis
that mi RNAs set thresholds on the expression of their
targets has gained experimental support in mammalian
systems. In this model, mi RNAs impose a burden on the
expression of their targets and promote their degradation.
By ensuring that their targets are expressed only upon sus‑
tained transcriptional activation, mi RNAs may reduce the
metabolic costs of propagating spurious transcriptional
events through the intracellular network.
Accurate prediction of miRNA targets, especially
non‑canonical ones, remains of interest not only
because computational models enable generalizations
to systems that cannot be easily studied experimentally
but also because a predictive model indicates a good
understanding of the underlying system. In what direc‑
tions can miRNA target prediction further develop?
In recent years, substantial progress has been made in
obtaining quantitative data concerning the affinities
of small‑RNA–target interactions. High‑throughput
miRNA‑binding‑site data and biochemical measure‑
ments of miRNA–target interaction affinity may soon
make it possible to develop miRNA‑specific models of
such interactions.
Furthermore, decay rates of mRNAs in the presence
and absence of mi RNAs have been measured, and the
expression levels of mi RNAs and mRNAs in individ‑
ual cell types, up to the level of single cells, have been
obtained. These data could be used to parameterize
models of miRNA‑dependent gene regulation to study
the response of the entire network to the perturbations
that have been described in the literature. The incor‑
poration of computational modelling into the study of
individual experimental systems has not only become
possible but is also likely to facilitate the progress in
understanding the dynamics of the complex regulatory
network involving mi RNAs and their RNA targets.
REVIEWS
12
|
ADVANCE ONLINE PUBLICATION www.nature.com/reviews/genetics
© 2014 Macmillan Publishers Limited. All rights reserved
32. Levine,E., Zhang,Z., Kuhlman,T. & Hwa,T.
Quantitative characteristics of gene regulation by
small RNA. PLoS Biol. 5, e229 (2007).
This paper introduces the threshold-linear
response of small RNA targets to their
transcriptional induction.
33. Mukherji,S. etal. MicroRNAs can generate
thresholds in target gene expression. Nature Genet.
43, 854–859 (2011).
This paper shows the threshold-linear response
with a miRNA target reporter construct in
mammalian cells.
34. Osella,M., Bosia,C., Cora,D. & Caselle,M. The role
of incoherent microRNA-mediated feedforward loops
in noise buffering. PLoS Comput. Biol. 7, e1001101
(2011).
This study computationally explores the
consequences of a gene being regulated only at the
transcriptional level or by both a transcription
factor and a miRNA that collectively form a FFL.
35. Poliseno,L. etal. A coding-independent function of
gene and pseudogene mRNAs regulates tumour
biology. Nature 465, 1033–1038 (2010).
36. Islam,S. etal. Characterization of the single-cell
transcriptional landscape by highly multiplex
RNA-seq. Genome Res. 21, 1160–1167 (2011).
37. Wee,L.M., Flores-Jasso,C.F., Salomon,W.E. &
Zamore,P.D. Argonaute divides its RNA guide into
domains with distinct functions and RNA-binding
properties. Cell 151, 1055–1067 (2012).
This study assesses the importance of different
miRNA positions for the interaction with an RNA
target in the context of miRISC and provides a
substantial number of measurements of rate
constants of miRNA–target interaction.
38. Pasquinelli,A.E. MicroRNAs and their targets:
recognition, regulation and an emerging reciprocal
relationship. Nature Rev. Genet. 13, 271–282 (2012).
39. Thomson,D.W., Bracken,C.P. & Goodall,G.J.
Experimental strategies for microRNA target
identification. Nucleic Acids Res. 39, 6845–6853
(2011).
40. Hipfner,D.R., Weigmann,K. & Cohen,S.M.
The bantam gene regulates Drosophila growth.
Genetics 161, 1527–1537 (2002).
41. Jinek,M. etal. A programmable dual-RNA-guided
DNA endonuclease in adaptive bacterial immunity.
Science 337, 816–821 (2012).
42. Wang,T., Wei,J.J., Sabatini,D.M. & Lander,E.S.
Genetic screens in human cells using the CRISPR–
Cas9 system. Science 343, 80–84 (2014).
43. Miska,E.A. etal. Most Caenorhabditis elegans
microRNAs are individually not essential for
development or viability. PLoS Genet. 3, e215
(2007).
44. Park,C.Y. etal. A resource for the conditional ablation
of microRNAs in the mouse. Cell Rep. 1, 385–391
(2012).
45. Linsley,P.S. etal. Transcripts targeted by the
microRNA-16 family cooperatively regulate cell cycle
progression. Mol. Cell. Biol. 27, 2240–2252 (2007).
46. Selbach,M. etal. Widespread changes in protein
synthesis induced by microRNAs. Nature 455, 58–63
(2008).
47. Baek,D. etal. The impact of microRNAs on protein
output. Nature 455, 64–71 (2008).
48. Guo,H., Ingolia,N.T., Weissman,J.S. & Bartel,D.P.
Mammalian microRNAs predominantly act to
decrease target mRNA levels. Nature 466, 835–840
(2010).
49. Hausser,J. etal. Timescales and bottlenecks in
miRNA-dependent gene regulation. Mol. Syst. Biol. 9,
711 (2013).
This study integrates a range of data sets to
estimate kinetic parameters for miRNA-dependent
target regulation.
50. Karginov,F.V. etal. A biochemical approach to
identifying microRNA targets. Proc. Natl Acad. Sci.
USA 104, 19291–19296 (2007).
51. Zhang,L. etal. Systematic identification of C.elegans
miRISC proteins, mi RNAs, and mRNA targets by their
interactions with GW182 proteins AIN-1 and AIN-2.
Mol. Cell 28, 598–613 (2007).
52. Meier,J. etal. Genome-wide identification of
translationally inhibited and degraded miR-155
targets using RNA-interacting protein-IP. RNA Biol.
10, 1018–1029 (2013).
53. Zisoulis,D.G. etal. Comprehensive discovery of
endogenous Argonaute binding sites in
Caenorhabditis elegans. Nature Struct. Mol. Biol. 17,
173–179 (2010).
54. Kishore,S. etal. A quantitative analysis of CLIP
methods for identifying binding sites of RNA-binding
proteins. Nature Methods 8, 559–564 (2011).
55. Hausser,J., Syed,A.P., Bilen,B. & Zavolan,M.
Analysis of CDS-located miRNA target sites suggests
that they can effectively inhibit translation. Genome
Res. 23, 604–615 (2013).
56. Zhang,C. & Darnell,R.B. Mapping invivo protein-
RNA interactions at single-nucleotide resolution from
HITS-CLIP data. Nature Biotech. 29, 607–614 (2011).
57. Konig,J. etal. iCLIP — transcriptome-wide mapping of
protein–RNA interactions with individual nucleotide
resolution. J.Vis. Exp. 50, e2638 (2011).
58. Broughton,J.P. & Pasquinelli,A.E. Identifying
Argonaute binding sites in Caenorhabditis elegans
using iCLIP. Methods 63, 119–125 (2013).
59. Smith,K.C. Photochemical addition of amino acids
to
14
C-uracil. Biochem. Biophys. Res. Commun. 34,
354–357 (1969).
60. Shetlar,M.D., Carbone,J., Steady,E. & Hom,K.
Photochemical addition of amino acids and peptides
to polyuridylic acid. Photochem. Photobiol. 39,
141–144 (1984).
61. Hockensmith,J.W., Kubasek,W.L., Vorachek,W.R. &
von Hippel,P.H. Laser cross-linking of nucleic acids to
proteins. Methodology and first applications to the
phage T4 DNA replication system. J.Biol. Chem. 261,
3512–3518 (1986).
62. Lai,E.C. MicroRNAs are complementary to 3ʹ UTR
sequence motifs that mediate negative post-
transcriptional regulation. Nature Genet. 30,
363–364 (2002).
63. Wang,Y., Sheng,G., Juranek,S., Tuschl,T. & Patel,D.J.
Structure of the guide-strand-containing argonaute
silencing complex. Nature 456, 209–213 (2008).
64. Wang,Y. etal. Nucleation, propagation and cleavage
of target RNAs in Ago silencing complexes. Nature
461, 754–761 (2009).
65. Parker,J.S., Parizotto,E.A., Wang,M., Roe,S.M. &
Barford,D. Enhancement of the seed–target
recognition step in RNA silencing by a PIWI/MID
domain protein. Mol. Cell 33, 204–214 (2009).
This paper evaluates the influence of a PIWI/MID
domain on the strength of interaction between a
small RNA and a target, which reveals that the
organization of the small RNA seed sequence by
AGO reduces the entropy penalty for the
interaction with the target.
66. Doench,J.G. & Sharp,P.A. Specificity of microRNA
target selection in translational repression. Genes Dev.
18, 504–511 (2004).
67. Yekta,S., Shih,I.H. & Bartel,D.P. MicroRNA-directed
cleavage of HOXB8 mRNA. Science 304, 594–596
(2004).
68. Alexiou,P., Maragkakis,M., Papadopoulos,G.L.,
Reczko,M. & Hatzigeorgiou,A.G. Lost in translation:
an assessment and perspective for computational
microRNA target identification. Bioinformatics 25,
3049–3055 (2009).
69. Hausser,J., Landthaler,M., Jaskiewicz,L.,
Gaidatzis,D. & Zavolan,M. Relative contribution of
sequence and structure features to the mRNA binding
of Argonaute/EIF2C–miRNA complexes and the
degradation of miRNA targets. Genome Res. 19,
2009–2020 (2009).
70. Ebert,M.S. & Sharp,P.A. Roles for microRNAs in
conferring robustness to biological processes. Cell
149, 515–524 (2012).
71. Jackson,A.L. etal. Widespread siRNA “off-target”
transcript silencing mediated by seed region sequence
complementarity. RNA 12, 1179–1187 (2006).
72. Kim,Y.K. etal. Functional links between clustered
microRNAs: suppression of cell-cycle inhibitors by
microRNA clusters in gastric cancer. Nucleic Acids
Res. 37, 1672–1681 (2009).
73. Kozomara,A. & Griffiths-Jones,S. miRBase: integrating
microRNA annotation and deep-sequencing data.
Nucleic Acids Res. 39, D152–D157 (2011).
74. Benetti,R. etal. A mammalian microRNA cluster
controls DNA methylation and telomere
recombination via Rbl2-dependent regulation of DNA
methyltransferases. Nature Struct. Mol. Biol. 15,
268–279 (2008).
75. Sinkkonen,L. etal. MicroRNAs control denovo DNA
methylation through regulation of transcriptional
repressors in mouse embryonic stem cells. Nature
Struct. Mol. Biol. 15, 259–267 (2008).
76. Wang,Q. etal. miR-17–92 cluster accelerates
adipocyte differentiation by negatively regulating
tumor-suppressor Rb2/p130. Proc. Natl Acad. Sci.
USA 105, 2889–2894 (2008).
77. Houbaviy,H.B., Dennis,L., Jaenisch,R. & Sharp,P.A.
Characterization of a highly variable eutherian
microRNA gene. RNA 11, 1245–1257 (2005).
78. Houbaviy,H.B., Murray,M.F. & Sharp,P.A.
Embryonic stem cell-specific microRNAs. Dev. Cell 5,
351–358 (2003).
79. Chen,P.Y. etal. The developmental miRNA profiles of
zebrafish as determined by small RNA cloning. Genes
Dev. 19, 1288–1293 (2005).
80. Giraldez,A.J. etal. MicroRNAs regulate brain
morphogenesis in zebrafish. Science 308, 833–838
(2005).
81. Tang,G.Q. & Maxwell,E.S. Xenopus microRNA genes
are predominantly located within introns and are
differentially expressed in adult frog tissues via post-
transcriptional regulation. Genome Res. 18, 104–112
(2008).
82. Altuvia,Y. etal. Clustering and conservation patterns
of human microRNAs. Nucleic Acids Res. 33,
2697–2706 (2005).
83. Aravin,A.A. etal. The small RNA profile during
Drosophila melanogaster development. Dev. Cell 5,
337–350 (2003).
84. Landgraf,P. etal. A mammalian microRNA expression
atlas based on small RNA library sequencing. Cell
129, 1401–1414 (2007).
85. Suh,M.R. etal. Human embryonic stem cells express
a unique set of microRNAs. Dev. Biol. 270, 488–498
(2004).
86. Bentwich,I. etal. Identification of hundreds of
conserved and nonconserved human microRNAs.
Nature Genet. 37, 766–770 (2005).
87. Mohammed,J., Flynt,A.S., Siepel,A. & Lai,E.C.
The impact of age, biogenesis, and genomic clustering
on Drosophila microRNA evolution. RNA 19,
1295–1308 (2013).
88. Lal,A. etal. miR-24 Inhibits cell proliferation by
targeting E2F2, MYC, and other cell-cycle genes via
binding to “seedless” 3ʹUTR microRNA recognition
elements. Mol. Cell 35, 610–625 (2009).
89. Shin,C. etal. Expanding the microRNA targeting
code: functional sites with centered pairing. Mol. Cell
38, 789–802 (2010).
90. Majoros,W.H. & Ohler,U. Spatial preferences of
microRNA targets in 3ʹ untranslated regions.
BMC Genomics 8, 152 (2007).
91. Meijer,H.A. etal. Translational repression and
eIF4A2 activity are critical for microRNA-mediated
gene regulation. Science 340, 82–85 (2013).
92. Gu,S., Jin,L., Zhang,F., Sarnow,P. & Kay,M.A.
Biological basis for restriction of microRNA targets to
the 3ʹ untranslated region in mammalian mRNAs.
Nature Struct. Mol. Biol. 16, 144–150 (2009).
93. Fang,Z. & Rajewsky,N. The impact of miRNA target
sites in coding sequences and in 3ʹUTRs. PLoS ONE 6,
e18067 (2011).
94. Reczko,M., Maragkakis,M., Alexiou,P., Grosse,I. &
Hatzigeorgiou,A.G. Functional microRNA targets
in protein coding sequences. Bioinformatics 28,
771–776 (2012).
95. Kertesz,M., Iovino,N., Unnerstall,U., Gaul,U. &
Segal,E. The role of site accessibility in microRNA
target recognition. Nature Genet. 39, 1278–1284
(2007).
96. Tafer,H. etal. The impact of target site accessibility
on the design of effective siRNAs. Nature Biotech. 26,
578–583 (2008).
97. Nielsen,C.B. etal. Determinants of targeting by
endogenous and exogenous microRNAs and siRNAs.
RNA 13, 1894–1910 (2007).
98. Jacobsen,A., Wen,J., Marks,D.S. & Krogh,A.
Signatures of RNA binding proteins globally coupled
to effective microRNA target sites. Genome Res. 20,
1010–1019 (2010).
99. Kedde,M. etal. RNA-binding protein Dnd1 inhibits
microRNA access to target mRNA. Cell 131,
1273–1286 (2007).
100. Bhattacharyya,S.N., Habermacher,R., Martine,U.,
Closs,E.I. & Filipowicz,W. Relief of microRNA-
mediated translational repression in human cells
subjected to stress. Cell 125, 1111–1124 (2006).
101. Kedde,M. etal. A Pumilio-induced RNA structure
switch in p27-3ʹ UTR controls miR-221 and miR-222
accessibility. Nature Cell Biol. 12, 1014–1020
(2010).
102. Li,J. etal. Identifying mRNA sequence elements for
target recognition by human Argonaute proteins.
Genome Res. 24, 775–785 (2014).
103. Vlachos,I.S. & Hatzigeorgiou,A.G. Online resources
for miRNA analysis. Clin. Biochem. 46, 879–900
(2013).
REVIEWS
NATURE REVIEWS
|
GENETICS ADVANCE ONLINE PUBLICATION
|
13
© 2014 Macmillan Publishers Limited. All rights reserved
104. Song,J.L. etal. Select microRNAs are essential for
early development in the sea urchin. Dev. Biol. 362,
104–113 (2012).
105. Ohrt,T. etal. Fluorescence correlation spectroscopy
and fluorescence cross-correlation spectroscopy
reveal the cytoplasmic origination of loaded nuclear
RISC invivo in human cells. Nucleic Acids Res. 36,
6439–6449 (2008).
The data obtained in this study enable the
estimation of the rates of AGO loading and
unloading with small RNAs.
106. Schwanhausser,B. etal. Global quantification of
mammalian gene expression control. Nature 473,
337–342 (2011).
107. Nam,J.W. etal. Global analyses of the effect of
different cellular contexts on microRNA targeting.
Mol. Cell 53, 1031–1043 (2014).
108. Enright,A.J. etal. MicroRNA targets in Drosophila.
Genome Biol. 5, R1 (2003).
109. Shalgi,R., Lieber,D., Oren,M. & Pilpel,Y.
Global and local architecture of the mammalian
microRNA–transcription factor regulatory network.
PLoS Comput. Biol. 3, e131 (2007).
110. Hornstein,E. & Shomron,N. Canalization of
development by microRNAs. Nature Genet 38,
S20–S24 (2006).
111. Marson,A. etal. Connecting microRNA genes to the
core transcriptional regulatory circuitry of embryonic
stem cells. Cell 134, 521–533 (2008).
112. Re,A., Cora,D., Taverna,D. & Caselle,M.
Genome-wide survey of microRNA–transcription
factor feed-forward regulatory circuits in human.
Mol. Biosyst 5, 854–867 (2009).
113. Arvey,A., Larsson,E., Sander,C., Leslie,C.S. &
Marks,D.S. Target mRNA abundance dilutes
microRNA and siRNA activity. Mol. Syst. Biol. 6, 363
(2010).
114. Ebert,M.S., Neilson,J.R. & Sharp,P.A. MicroRNA
sponges: competitive inhibitors of small RNAs in
mammalian cells. Nature Methods 4, 721–726
(2007).
115. Cazalla,D., Yario,T. & Steitz,J.A. Down-regulation
of a host microRNA by a Herpesvirus saimiri
noncoding RNA. Science 328, 1563–1566
(2010).
116. Marcinowski,L. etal. Degradation of cellular miR-27
by a novel, highly abundant viral transcript is
important for efficient virus replication invivo. PLoS
Pathog. 8, e1002510 (2012).
117. Wang,Y. etal. Endogenous miRNA sponge lincRNA-
RoR regulates Oct4, Nanog, and Sox2 in human
embryonic stem cell self-renewal. Dev. Cell 25, 69–80
(2013).
This is a study of crosstalk between a lncRNA and
mRNA targets of mi RNAs, in which the abundance
of the involved molecular species have been
measured.
118. Kallen,A.N. etal. The imprinted H19 lncRNA
antagonizes let-7 microRNAs. Mol. Cell 52, 101–112
(2013).
119. Chang,J. etal. miR-122, a mammalian liver-specific
microRNA, is processed from hcr mRNA and may
downregulate the high affinity cationic amino acid
transporter CAT-1. RNA Biol. 1, 106–113 (2004).
120. Denzler,R., Agarwal,V., Stefano,J., Bartel,D.P. &
Stoffel,M. Assessing the ceRNA hypothesis with
quantitative measurements of miRNA and target
abundance. Mol. Cell 54, 766–776 (2014).
121. Hansen,T.B. etal. Natural RNA circles function as
efficient microRNA sponges. Nature 495, 384–388
(2013).
122. Memczak,S. etal. Circular RNAs are a large class of
animal RNAs with regulatory potency. Nature 495,
333–338 (2013).
123. Bosia,C., Pagnani,A. & Zecchina,R. Modelling
competing endogenous RNA networks. PLoS ONE 8,
e66609 (2013).
124. Figliuzzi,M., Marinari,E. & De Martino,A.
MicroRNAs as a selective channel of communication
between competing RNAs: a steady-state theory.
Biophys. J. 104, 1203–1213 (2013).
125. Buchler,N.E. & Louis,M. Molecular titration and
ultrasensitivity in regulatory networks. J.Mol. Biol.
384, 1106–1119 (2008).
126. Larsson,E., Sander,C. & Marks,D. mRNA turnover
rate limits siRNA and microRNA efficacy. Mol. Syst.
Biol. 6, 433 (2010).
127. Kim,H.H. etal. HuR recruits let-7/RISC to repress
c-Myc expression. Genes Dev. 23, 1743–1748
(2009).
128. Erhard,F. etal. Widespread context-dependency of
microRNA-mediated regulation. Genome Res. 24,
906–919 (2014).
129. Majoros,W.H. etal. MicroRNA target site
identification by integrating sequence and binding
information. Nature Methods 10, 630–633 (2013).
130. Erhard,F., Dolken,L., Jaskiewicz,L. & Zimmer,R.
PARma: identification of microRNA target sites in
AGO-PAR-CLIP data. Genome Biol. 14, R79 (2013).
131. Grun,D., Wang,Y.L., Langenberger,D. & Gunsalus,K. C.
& Rajewsky,N. microRNA target predictions
across seven Drosophila species and comparison to
mammalian targets. PLoS Comput. Biol. 1, e13 (2005).
132. He,L. etal. A microRNA polycistron as a potential
human oncogene. Nature 435, 828–833 (2005).
133. Franceschini,A. etal. STRING v9.1: protein–protein
interaction networks, with increased coverage and
integration. Nucleic Acids Res. 41, D808–D815
(2013).
134. Wang,D. etal. Quantitative functions of Argonaute
proteins in mammalian development. Genes Dev. 26,
693–704 (2012).
135. Stalder,L. etal. The rough endoplasmatic reticulum is
a central nucleation site of siRNA-mediated RNA
silencing. EMBO J. 32, 1115–1127 (2013).
136. Djuranovic,S., Nahvi,A. & Green,R. miRNA-mediated
gene silencing by translational repression followed
by mRNA deadenylation and decay. Science 336,
237–240 (2012).
137. Bethune,J., Artus-Revel,C.G. & Filipowicz,W.
Kinetic analysis reveals successive steps leading to
miRNA-mediated silencing in mammalian cells. EMBO
Rep. 13, 716–723 (2012).
138. Bazzini,A.A., Lee,M.T. & Giraldez,A.J. Ribosome
profiling shows that miR-430 reduces translation
before causing mRNA decay in zebrafish. Science 336,
233–237 (2012).
Acknowledgements
The authors thank P. Pemberton-Ross for help with the figure
in Box 2 and members of M.Z.’s laboratory for comments on
the manuscript.
Competing interests statement
The authors declare no competing interests.
REVIEWS
14
|
ADVANCE ONLINE PUBLICATION www.nature.com/reviews/genetics
© 2014 Macmillan Publishers Limited. All rights reserved
