Briefings in Bioinformatics, 21(4), 2020, 1313–1326

doi: 10.1093/bib/bbz079
Advance Access Publication Date: 28 August 2019
Review article

Circulating microRNA trafficking and regulation:

computational principles and practice
Juan Cui and Jiang Shu
Correspondence author. Juan Cui, Systems Biology and Biomedical Informatics (SBBI) Laboratory, Department of Computer Science and Engineering,
University of Nebraska-Lincoln, Lincoln, NE 68588, USA. Tel: +1 402 472 5023; Fax: +1 402 472 7767; E-mail: [email protected].

Abstract
Rapid advances in genomics discovery tools and a growing realization of microRNA’s implication in intercellular
communication have led to a proliferation of studies of circulating microRNA sorting and regulation across cells and
different species. Although sometimes, reaching controversial scientific discoveries and conclusions, these studies have
yielded new insights in the functional roles of circulating microRNA and a plethora of analytical methods and tools. Here,
we consider this body of work in light of key computational principles underpinning discovery of circulating microRNAs in
terms of their sorting and targeting, with the goal of providing practical guidance for applications that is focused on the
design and analysis of circulating microRNAs and their context-dependent regulation. We survey a broad range of
informatics methods and tools that are available to the researcher, discuss their key features, applications and various
unsolved problems and close this review with prospects and broader implication of this field.

Key words: circulating microRNA; exosomal microRNA; microRNA sorting; microRNA target; RNA interactome; RNA motif;
gene regulation network; network analysis

Introduction epithelial cells to hepatocytes [7–9]. In addition to the endoge-

Since microRNAs (miRNAs) were discovered, studies of their nously synthesized miRNAs, humans and mice can absorb
roles in post-transcriptional gene silencing in eukaryotes have a quantitatively meaningful amount of miRNAs from diets
revealed significant evidence that links miRNA-controlled gene of plant source (e.g. honeysuckle and rice [10, 11]) or animal
regulation with various aspects of cellular growth and response, source (e.g. milk and egg [12, 13]). While some discoveries
immunity, nutrient homeostasis and complex diseases [1–5]. on the biogenesis and function of exogenous miRNA are
In human, it is estimated that over 2000 human miRNAs can contradicted [14–16], it has been documented that humans
regulate up to 60% of human genes [3]. This class of small can absorb dietary exosomes that deliver miRNA cargos to
non-coding RNAs have been conventionally considered as various peripheral tissues, possibly regulating human genes
critical endogenous regulators of intracellular processes until [17, 18]. Similar evidence of miRNA cross-species transfer
very recently accumulating evidence shows some miRNAs and regulation were also observed between eukaryotes and
can be transferred across cells and tissues [6]. For example, prokaryotes [6, 19–22]. In all these scenarios, exogenous miRNA
extracellular miRNAs have been studied as new intercellular confounders can hypothetically target host genes and influence
messengers in cell–cell communication between dendritic the host’s action, which leads to a set of fundamental questions
cells, hepatocellular carcinoma cells and adipocytes, from T about which and how miRNAs can be transferred, exert
cells to antigen-presenting cells, stem cells to endothelial functions and possibly contribute to disease process. Centered
cells and fibroblasts, macrophages to breast cancer cells and on these questions, recent miRNA researches have made

Juan Cui is an assistant professor at the Department of Computer Science and Engineering in the University of Nebraska-Lincoln and an associated member
of Nebraska Center for the Prevention of Obesity Diseases
Shu Jiang was a research assistant in the University of Nebraska- Lincoln during 2014–2018.
Submitted: 18 March 2019; Received (in revised form): 7 June 2019

© The Author(s) 2019. Published by Oxford University Press. All rights reserved. For Permissions, please email: [email protected]

1313
 1314 Cui and Shu

Figure 1. Schematic diagram of miRNA transfer between cells and competitive miRNA binding (in recipient cell as an illustration). MVB: multivesicular body

compelling progress toward mechanistic understanding of very similar. It is also difficult to affirm whether an interaction
circulating miRNAs sorting and function, but also revealed occurred or not although the identical exogenous miRNAs can
challenging complexity underlying each of those processes in presumably bind to the targets of their host homologs.
the meantime. Circulating miRNAs, once entering into recipient cells, may
regulate host genes under the same binding principle of endoge-
nous cellular miRNAs. Therefore, the precise understanding of
Mechanistic evidence and challenges their function is also largely depending on reliable identification
Current knowledge of miRNA secretion and circulation is lim- of the mRNA targets. In this regard, recent genomic approach
ited to its carrier-mediated transport, which is realized by its using AGO crosslinking immunoprecipitation sequencing (CLIP-
encapsulation in microvesicles and exosomes or protein-bound seq) [49] and cross-linking ligation and sequencing of hybrids
complexes, e.g. associated with Argonaute [23] family proteins (CLASH) [50] can pinpoint locations of the direct linkage between
and lipoprotein [24–29] (Figure 1). In either form, miRNAs need an AGO-bound miRNA and target mRNAs while there is a lack of
to bind to other molecules, either exosome sorting proteins [30] information about the downregulation as a result of binding
or other RNA-binding proteins (RBPs), in which the sequence or in the meantime. Following these biochemical innovations,
structure features of miRNA are very likely to affect the binding many computational studies focused on the assessment of
affinity and thus the secretion that is largely unknown. Recent binding likelihood between a pair of miRNA and mRNA that are
new evidence from exosome study also shows that cells can conditionally independent [23, 51–53]. Since one possible
selectively package certain types of miRNA into exosomes and downside of these approaches is that only a small fraction of
export them to recipient cells [31], while it is unclear what it distinct interactions were examined in contrast to the large
contributes to the selection. number of interaction types that can possibly occur between
Exosomes, as important cell-to-cell communication media- miRNA and their targets, prediction models built based on such
tors that deliver a variety of cargos (proteins, RNAs and lipids) insufficiently representative collection is often biased. Moreover,
from donor to recipient cells, contain the majority of circulating miRNA–mRNA interactions can vary widely across different
miRNAs that regulate key biological and pathophysiological phenotypic conditions [54–56]. It has been revealed that miRNAs
functions in complex human diseases [1, 2, 32–36]. For example, regulate target genes in a dynamic and conditional manner
exosomal miRNAs from cancer cell promote tumor growth in where the dramatic complexity can be explained by competitive
distant tissues, e.g. miR-105 from breast cancer to lung, brain RNA binding and multifaceted gene regulation [52, 54–57]. As
and other tissues [37–42] and adipose tissue-specific exosomal shown in Figure 1, one miRNA can simultaneously bind to
miRNAs regulate metabolic function in distant tissues [21, 43]. various target mRNAs, long non-coding RNAs and circular RNAs,
In addition, miRNAs transferred by exosome from dietary while one gene can be regulated by multiple miRNAs [54–57].
species can also be health-related, e.g. those from cow’s milk can Competition among different miRNAs also takes place when
enter human circulatory system and regulate gene expressions they potentially target the same transcript at the same or close
related to bone development [12], while plant exosomal miRNAs binding regions, while collaboration of miRNAs exists when
regulate immune system in humans and animals [44]. All these they bind to different non-overlapping regions or different
evidence serves as basis for researchers to investigate the copies of the same target transcript [54–57]. In addition, other
regulatory function of exosome-mediated miRNAs. To this end, factors such as genetic mutations, competition with other RBPs
next-generation sequencing (NGS) has been widely applied to and the conditional availability of miRNA and mRNA can also
analyze RNA expression profile of exosomes from different affect the equilibrium of miRNA–mRNA interactions. Each of
sources [45–47]. A novel NGS analytical pipeline for the cross- the aforementioned mechanisms stresses the dynamic property
species miRNA detection was developed and new strategies of miRNA regulation from a different perspective. Obviously,
for multi-species sequence comparison and sensitive variation most existing computational methods focusing on static miRNA
detection were presented [48]. Despite all those innovative target prediction are not sufficient in this regard.
technical advances, challenges remain when differentiating With the goals to characterize miRNA sorting process and
endogenous and exogenous sequences that are identical or the mechanistic link of miRNA-gene regulation and function,
 Circulating miRNA trafficking and regulation 1315

past efforts exploring novel bioinformatics and genomic G + C content [71–73], folding energy and stem length [74–
approaches to achieve those goals have made great progress. 80]. Using a supervised learning approach, discriminative
In this review, we organize the exiting methods and tools sequence features associated with circulating miRNAs were
into three research topics including miRNA sorting, targeting examined [81]. In the same study, an ensemble model with
and regulation, based on the evidence, rationale and technical manifold ranking was developed to predict transferable dietary
designs. Note that the majority of computational methods miRNAs from 14 common food species. This type of study
on targeting and regulation were originally designed for represents one of the first efforts to design and utilize effective
endogenous miRNA study, which however can be directly computational method for circulating miRNAs discovery. On
applied to the emerging field of circulating miRNAs without top of developing new computational techniques, it formulates
specification. and solves the biological puzzle as a computational classifi-
cation problem. An in-depth investigation of the mechanism
underlying exosome-mediated miRNA trafficking and kinetics
Bioinformatics methodologies and tools in is, however, beyond the scope of this kind of computational
circulating miRNA study work.

MiRNA sorting and trafficking

Knowing exosome packaging of miRNA is not a random
MiRNA targeting
process but involving selective machineries that favor some Both cellular and circulating miRNAs can bind to various
miRNAs over others [58] and are possibly cell specific [59], transcripts; the way they bind and the interaction environment
increasing efforts were made to identify discriminative molec- of the binding site determine whether it triggers mRNA
ular features having impacts on protein binding and miRNA degradation or translation inhibition [82, 83]. In humans, it is
secretion [60]. For example, bioinformatics analyses of exosomal known that miRNAs bind to their targets through imperfect
miRNA sequences have revealed a few sequence motifs to pairing in the seed region, while in some cases additional pairing
be essential for miRNA packaging [30]. A short motif pattern region occurs adjacent to the seed region. Multiple sequencing
[AGU]G [AG] G was found highly enriched in human T-cell- analysis using CLASH, CLEAR (covalent ligation of endogenous
derived exosomes using two well-established motif discovery Argonaute-bound RNAs)-CLIP, high-throughput sequencing of
tools (COSMO [61] and MEME [30]). Follow-up structural analysis RNAs isolated by crosslinking immunoprecipitation (HITS-CLIP)
has confirmed the predicted motif is associated with interaction and photoactivatable ribonucleoside-enhanced (PAR)-CLIP and
sites between the transport protein hnRNPA2B1 (heterogeneous ligation of miRNA-target chimeras on the Argonaute protein
nuclear ribonucleoprotein A2B1) and its bound small RNAs [23] have identified large numbers of such interaction patterns
[30]. Furthermore, in vitro test shows mutant on sequence despite of the aforementioned limitations. In addition to those
GGAG suppressed the transportation of associated miRNAs into listed in Table 1, starBase database archived over 100 sets
exosomes, which provides further evidence that the constituent of CLIP-seq data generated from various tissues or cell lines
4-mer sequence of the predicted motif is important for miRNA under different treatments [84]. Other than archiving miRNA–
sorting. Similar observation was made on another 6 bp-long mRNA interactions, this database also includes information
motif pattern, [GAU][GUA][GAU][CAG][UA][GC] that is associated about interactions among miRNA, ncRNA and RBP. Variations
with exosomal packaging in murine hepatocyte 3A cells [63], across different datasets are possibly caused by the biological
identified by using Improbizer [64]. In general, although many variability among various cell types and species or the technical
motif discovery tools are available, as reviewed in [65], common variability among the platforms.
problems of these tools initially designed for DNA regulatory MiRNA target prediction, used to be the most popular topic
element prediction lie in poor sensitivity of detecting short in miRNA bioinformatics research, has been broadly reviewed
motifs, e.g. MEME [66], tends to predict motifs longer than 6 bps in several articles where numerous software and tools were
(not a problem for DNA regulatory elements). Some programs extensively discussed [85–87]. Existing methods usually utilizes
require a significant amount of negative control information (rule-based) machine learning (ML) techniques and sequence
(otherwise leading to high false discovery), e.g. DREME [67], a and structural features of the experimentally validated binding
modified version of MEME. To tackle these issues, a new motif- sites to identify new miRNA targets (representative methods
finding method was applied to identify cell-specific sequence listed in Table 2). For examples, TargetScan, one of the most
motifs for all exosomal RNAs available in the public domain, popular tools, applies multiple linear regressions based on site-
which was documented with an online tool entitled MDS2 and a context features about seed matching and conserved sites [88],
database in [68]. Compared to other graph-based algorithms, e.g. while miRanda replied on conservation and expression profiles
MotifCut [69] and MotifClick [70], that search for maximum- and support vector regression [89]. Similarly, DIANA-microT-CDS
density subgraph based on highly similar k-mers, MDS2 applied linear regression on base paring and folding energy [90].
evaluate significant k-mers exhaustively based on automatically MIRZA-G [91] and miRTar2GO [53] are all rule-based learning
generated background control and search for motifs with high models, while mBSTAR represents a classification model based
coverage and explicitness through combinatorial optimization. on random forest [92]. Human-crafted features are critical for
The unique strength of such approach is the capability of building a reliable prediction model using these (rule-based) ML
detecting significant short motifs without requiring a quality- approaches. Most existing studies have applied a large number
negative dataset. of molecular features related to the binding sites such as count
In addition to motif analysis, data mining approaches of perfect Watson–Crick base pairing in the seed region, AU-rich
were adopted to identify other types of molecular features nucleotide composition near the binding site, thermodynamic
contributing to miRNA sorting or targeting process. A recent stability of the binding sites, evolutionary conservation, target
study examined over 1000 sequence and structural features site accessibility and the flanking sequence of the target site
related to mature and precursor miRNAs, such as nucleotide on mRNA and the localization of the binding sites in the 3’
compositions on seed region (positions 2–8 from the 5 -end), UTR. RNAfolding and RNAsubopt were often used to evaluate
 1316 Cui and Shu

Table 1. Selected CLIP-seq data on AGO associated miRNA–mRNA interactions

GEO Accession Experiment Samples Findings Ref

GSE46039 CLASH-AGO1 Human kidney HEK293 More than 18 000 [134]

high-confidence miRNA–mRNA
interactions
GSE90089 CLEAR-CLIP Bovine kidney 170 000 unique miRNA–target [135]
(MDBK) cells interactions
GSE97060; AGO-CLIP miR-122 KO; mouse liver and miR-122 interacted mRNAs [136]
GSE97058; human hepatocellular
GSE97056 carcinoma (HCC) tissue
GSE64679; HITS-CLIP miR-122 KO Huh7.5 cells and miR-122 targets and changes [137]
GSE64678; controls with Hepatitis C virus infection
GSE64676;
GSE64675
GSE41285 HITS-CLIP Mouse miR-155 KO CD4+ T cell miR-155-dependent Argonaute [138]
with control binding targets
GSE33584 Ago1 and Ago2-CLIP Human cytomegalovirus Novel HCMV miRNA and human [139]
(HCMV)-infected fibroblast cells miRNA targets
and control
GSE73059 CLEAR-CLIP Mouse brain and human 130 000 endogenous [140]
hepatoma cells miRNA-target interactions in
mouse brain and ∼40 000 in
human hepatoma cells
GSE16338 AGO-CLIP Mouse brain miRNA target mRNAs [49]
GSE56180 AGO-CLIP Caenorhabditis elegans ∼17 000 miRNA–target site [141]
interactions

the optional secondary structure formed between miRNA and same miRNA, those predictions could be biased. To this end,
the fragment of the 3’UTR. In addition, probabilistic approaches inclusion of the experimental miRNA–mRNA interatomic infor-
include Hidden Markov models (e.g. PicTar [55]) and a Dirichlet mation derived from aforementioned sequencing innovations
Process Gaussian Mixture Model (DPGMM) where the output of and the assessment of conditional binding certainly ensures
a posterior probability can indicate binding potential, and the an improved performance in several new or recently updated
prediction was made on the transcript level instead of gene methods, e.g. miRTar2GO, TargetScan V7.0 [88] and deep learning
level [51]. models [94–96]. Furthermore, new research efforts lead to hybrid
Worthwhile mentioning is that early studies of miRNA target pipelines that are based on the correlated regulation observed
prediction used very few experimental data on miRNA–mRNA from expression data and sequence-based targeting prediction.
interactions for model training, which is often insufficiently to For example, in [97], novel functional non-canonical interactions
capture the complex patterns in miRNA–mRNA complementary were identified using pairwise AGO-CLIP and mRNA sequencing
paring. As a result, the performances of static target prediction data where false discovery was controlled by stringent statis-
methods usually are held back by high false prediction rate. tics. It is also proved in [98] that when estimating the regu-
Very often, a single miRNA can bind to hundreds and thou- latory potential, sequencing-identified interactions subject to
sands of target genes, introducing numerous binding sites when careful statistical filtering process can eliminate false interac-
searching against the entire transcriptome. To this end, a com- tions. In addition to integrating more rules and data into the
mon practice was to integrate prediction results from multiple prediction, new strategies studying functional miRNA targeting
different tools, e.g. using the prediction intersections, to filter are desired to have following properties: allowing easy data
high confidence prediction [93]. Given the availability of a large integration, minimal requirement of the determination of var-
set of experimental interactions from sequencing, recent work ious cut-offs and predictions capturing various types of known
also demonstrates the success in using deep learning for miRNA interactions [99].
target prediction, either using features or sequence only [94–96]. However, in contrast to the extensive review on the function,
These models consider seed region as the common interacting theory and application of existing software and tools [85–87],
sites, along with complementary pairing in other regions. They there is a lack of systematic performance comparison. Clearly
all show promising results evaluated by using reserved experi- quality benchmark data that includes a large number of true
mental interactions as testing set. Based on the CLIP-seq data, interactions to validate the predictions are currently missing.
some de novo methods focus on motif-finding analysis of the Note that sequencing identified interactions, e.g. those archived
experimental binding sites, e.g. miRBShunter [23], and identify in miRTarbase [100], tend to be cell specific and highly discordant
sequence patterns for new target prediction. across datasets, making it less reliable for general validation
Early sequence-based methods for target prediction exam- purpose. Small-scale comparison may result in questionable
ine the likelihood of each binding site of a given miRNA indi- conclusions and very often integrated predictions across widely-
vidually regardless of the competitive and cooperative events accepted tools were recommended. For example, a previous
(Figure 1). Considering the relatively low abundance of miRNA analysis showed improved performance when integrating pre-
in a real condition and the competition among all potential diction results from TargetScan [88], miRanda [89], Pita [101]
targets with different abundance and binding efficacy to the and RNA22 [102] compared to individual prediction, which was
 Circulating miRNA trafficking and regulation 1317

Table 2. Selected software tools for miRNA target prediction

Tool Method Link Ref Key properties

TargetScan v7.0 Multiple linear http://www.targetscan.org/ [88] • Focus on canonical binding sites within
regression 3 -UTR regions
• Species-specific prediction
• Both web-server and standalone program
are available
• Stepwise variable selection based on
Akaike Information Criterion

miRanda Support vector http://34.236.212.39/microrna/ [89] • Assess thermodynamic folding energy of

regression home.do [142] miRNA:UTR duplex
• Conservation of the binding sites
• Good mirSVR score

miRecords Integration http://c1.accurascience.com/ [93] • Integration of 11 established miRNA target

miRecords/ prediction programs
• Archiving experimentally validated miRNA
targets

RNA22 v2 Motif pattern-based https://cm.jefferson.edu/rna22/ [102] • Based on conserved sequence features of

Interactive/ mature miRNA
• Base pairing and folding energy
• miRbase 21/Ensembl 78

DIANA-microT-CDS General linear model http://www.microrna.gr/ [90] • Searches canonical binding sites in CDS
V5.0 webServer and 3 -UTR regions
• miRbase 18 and Ensembl 69

PicTar Hidden Markov http://pictar.mdc-berlin.de/ [55] • Considered combinatorial miRNA

Probabilistic model regulations

miRCript DPGMM http://sbbi-panda.unl.edu/ [51] • Target prediction on transcripts

miRCript/ • Output interaction probability

PITA Parameter-free model https://genie.weizmann.ac.il/ [101] • Evaluated the structural accessibility of a

pubs/mir07/ binding site based on free energy
• Seed and sequence conservation
• Predicted targets in worm, mouse, fly and
human

MIRZA-G General linear model https://tools4mirs.org/software/ [91] • Binding affinity assessed by the MIRZA
target_prediction/mirza-g/ biophysical model
• Included proteomics datasets for training
and testing

mBSTAR Random forest https://www.isical.ac.in/&#x007 [92] • Multiple instance learning was used on
classification E;bioinfo_miu/MBStar30.htm experimental validated binding sites
• Randomly constructed negative dataset

miRTar2GO Rule-based model http://www.mirtar2go.org/ [53] • Rule-based

• Predict cell-type specific target

miRBShunter De novo motif finding https://github.com/ [23] • Based on AGO-CLIP data

TrabucchiLab/
MiRTDL convolutional neural http://nclab.hit.edu.cn/ccrm [94] • Analyze matrices of miRNA binding site
networks feature

Continued.
 1318 Cui and Shu

Table 2. Continued.

Tool Method Link Ref Key properties

DeepTarget recurrent neural http://data.snu.ac.kr/pub/ [95] • Identification of canonical sites

network deepTarget • A limited small data set at the training
phase

miRAW deep neural network https://bitbucket.org/account/ [96] • Leaning based on sequence only, followed
user/bipous/projects/MIRAW by feature-based posteriori filtering

validated by subsequent in vivo and in vitro experiments [86]. previous studies such as Sumazin et al. [62] and Bosia et al.
Specifically, it showed that the union, rather than intersection [116] have shown that the observed interactions regulate miRNA
of the aforementioned tools, is the best strategy for maximizing targets without affecting miRNA expression; miRNA competition
performance while maintaining the minimal cost of compu- can only be observed in particular expression ranges. Learning
tation time and resources. Similar but larger scale efforts are such dependencies appears to be out of the scope in the existing
desired given the fact that most tools suffer from high false approaches listed in Table 3.
prediction rate, and there is a lack of consensus about the
optimum strategy to select and use the prediction results. Note
that in circulating miRNA research, target analysis was also Dynamic aspect of miRNA regulation
largely depending on those aforementioned computational tools
or repositories, including predicted or/and experimental miRNA To further tackle the challenges in dynamic modeling of
targets [103–105]. conditional and modulated miRNA–mRNA interaction, con-
tinued efforts on multi-level genomic data integration have
been invested in the past decade. Table 4 listed a few such
Context-sensitive approach to infer miRNA regulatory
integrative methods that take into consideration both competing
interactions
and cooperative miRNA regulation, as well as other types of
Once binding targets are determined computationally or exper- gene expression regulation mechanisms in human cancers, by
imentally, miRNA function can be inferred through pathway using different types of genomic profiles. Similar to context-
analysis in general. Context-dependent regulation, however, can dependent miRNA targeting prediction, many early approaches
be only assessed when taking into consideration data specific used simple linear regression models combined with partial
to certain conditions. To this end, gene expression profiles have correlation [117–120], a norm for expressed-based prediction
been largely used in correlation calculation and network infer- of miRNA interaction since a decade ago. Chen et al. used
ence in many bioinformatics analyses. Specifically, negative cor- Bayesian network on expression data from multiple types of
relation between expressions of miRNA and target was com- cancer to identify common miRNA interactions associated with
monly used as a strong indicator to validate predicted interac- cancer [121]. Although integration of miRNA–gene and gene–
tions [106–110], where a threshold normally is needed. Infor- gene regulation into the same network represents non-trivial
mation fusion in this case has transformed the previous static task and demands better ensemble approach, comparative
interaction prediction into a semi-conditional manner, lead- analysis in such case has the advantage to capture strongly
ing to more practically useful results if not ideal. As listed in recurrent interactions. Still, other authors used random forest
Table 3, many of these studies focused on making discoveries in with predicted interactions and expression data sets in human
a specific case study and therefore didn’t make their software liver cancer [122]. Note that most of these models can be
and tools available to the public. Several linear LASSO regres- intuitively more powerful than correlation-based approaches,
sion models were developed based on gene expression profiles however, none of them explicitly discuss modulated regulation.
along with mixed use of predicted and experimental interactions In 2017, Shu et al. built a meta-Lasso regression model based
[52, 111–113]. GenMIR++ employs a generative model in which on a comprehensive set of genomic profiles, including both
miRNA expression profiles are used to explain the expression gene and miRNA expression in various cancer conditions,
of putative targets inferred from sequence analysis [114]. Fur- genomic copy number variations (CNVs) and DNA methylation
thermore, recent work has applied network analysis on top of (DM) and sequencing-detected Ttanscription factor (TF) [123]
general predicted miRNA-target networks, in order to search for binding sites and miRNA-mRNA interactions [98]. The model
specific sub-networks associated with the condition of interest, considers gene expression altered via regulation by miRNAs
e.g. human cancers [62, 107, 115]. In Bayesian network, inclusion and TFs. Based on the conditional miRNA–gene interactions
of a prior derived from sequence-related features in the miRNA– derived from each model, modularized miRNA regulation was
mRNA regulation model may improve the structure learning assessed based on its involvement in fundamental human
stage, which, however, is still challenging to capture the complex pathways [124]. By integrating cancer genomic data from
dynamics in miRNA regulation where cooperation and comple- The Cancer Genome Atlas (TCGA), it has identified novel
tion is involved. Early success of such research includes a com- regulatory modules where participating miRNAs jointly bind
putational inference of miRNA-mediated interaction networks to functionally related genes in different types of cancer [52].
in complex process such as glioblastoma [62]. It has been found This work demonstrates how promising information fusion and
that those interactions regulate thousands of genes, including modeling can facilitate the discovery of miRNA competitive
oncogenes and tumor suppressors, enabling crosstalk between binding, as well as modulated miRNA regulation, in human
established oncogenic pathways. Of particular interest is that cancers.
 Circulating miRNA trafficking and regulation 1319

Table 3. Selected methods for context-dependent miRNA–gene interaction

Study Method Data type Key properties Ref

Liu et al. (2009) Bipartition graph Predicted • Identify functional miRNA-mRNA [106]
with correlation interactions and regulatory module
calculation expression profiles • PCC > 0.75 & P < 0.05

Betel et al. (2010) Unified probabilistic Expression profiles • Regression model on sequence and [114]
of miRNA and contextual features from predicted
mRNA target sites
• Ranking target sites by a
down-regulation score

Lu et al. (2011) LASSO regression Predicted targets by • To identify reliable miRNA–mRNA [111]
TargetScan and targeting relationships that
PicTar; and combines sequence-based
cell-specific prediction, co-regulation, RISC
expression availability and expression data.
• Software download discontinued

Muniategui et LASSO regression Mixed predicted and • Quantification of miRNA–mRNA [112]

al. (2012) validated targets interactions
from multiple • Software download discontinued
databases and
cell-specific
expression
Quitadamo et al. eQTL and Bayesian Expression profiles • Takes account of miRNA expression [111]
(2015) network of miRNA and quantitative trait loci (eQTL)
mRNA associations, miRNAs and their
targets, protein-protein interactions,
co-expressions among miRNAs and
genes
• Case study on ovarian cancer

Sadeghi et al. Pearson Predicted target list • PCC > 0.4 & P < 0.05 [115]
(2016) correlation-based and expression • Initial interactions from databases
network analysis profiles • Focused on differentially expression
miRNAs and genes in metastatic
prostate tumor

Huang et al 2011 Ensembled Sequence and • Combine static and dynamic [110]
networks expression networks using
• User-defined association measure

Suzuki et al. Pearson Expression profiles • Identify miRNAs associated with [107]
(2017) correlation-based cancer hallmarks
network analysis
Ding et al. (2015) Pearson Expression profiles • Focused on differentially expression [108]
correlation-based of miRNA and miRNAs and genes in hepatic cancer
network analysis mRNA stem cell

Liu et al. (2016) Network community Predicted target list • Search for sub-networks in the [143]
detection and expression general predicted miRNA–target
profiles networks

Baglioni et al. Text mining and Predicted target list • Search for disease associated genes [144]
(2015) network analysis and expression • Correlation validated interactions
profiles • Sub-network extraction

Continued.
 1320 Cui and Shu

Table 3. Continued.

Study Method Data type Key properties Ref

Koo et al. (2018) Artificial neural Sequence and • Context-dependent prediction of the [145]
network expression data presence and strength of
interactions
• Tiresias available at https://
bitbucket.org/cellsandmachines/

Sumazin et al. Multivariate Expression profiles • Differentiate sponger modulators [62]

(2011) analysis of miRNA and where competitive binding affects
mRNA miR availability and other miR
activity modulators
• Method Hermes identifies candidate
miR-activity modulators by finding
genes whose expression is
associated with changes in mutual
information between regulators and
targets
• Identify dense subgraph revealing
regulations in gliomagenesis

Given the complexity of such analysis, particularly with inte- be used for this purpose. It has been shown that the purified
gration of various heterogeneous data, many of these studies exosome-like vesicles secreted by multivesicular bodies (MVBs)
didn’t provide or discontinued the web use of their software are enriched in the components of RNA-induced silencing com-
tools and only a few provide standalone packages are avail- plex [127], e.g. GW182 and AGO2 and miRNA-repressible mRNA
able, which introduces difficulties for in-depth evaluation. Most [29]. Congregation of miRNA with these associated proteins into
methods were validated only in their original study; again, there the sorting complex could be key to subsequent procedure of
is no consensus use of common datasets for validating condi- loading. Among those over 2000 exosomal proteins [126], one
tional interactions, which has hindered the broad use of some can separate RBPs from the rest based on the Uniprot annotation
methods. More sophisticated models capable of capturing mul- and/or RBP prediction. Subsequently, protein–RNA docking can
tifaceted gene regulation based on fusion of information from be performed, e.g. through HADDOCK2.2 [128] to identify loading
multiple reliable sources will always be more appealing to a new protein candidates by examining the protein–miRNA interactive
application, which is true to the circulating miRNA functional sites. Such findings could not only shed light on fundamental
study. mechanism of molecular interaction during the sorting process
but also provide alternative approaches to biomarker discovery.
For example, some miRNAs are preferentially sorted into
Discussions and future prospects exosomes and widely distributed in tumor cells and circulation
Topics on cross-species bioavailability, transfer and regulation [58, 129–131], which can become highly promising cancer
of circulating miRNA, as we reviewed previously, have offered biomarker for clinical application. Note two databases, miRan-
novel directions in the emerging research fields, such as cell– dola [132] and ExoCarta [126], have archived such extracellular
cell communication and nutrigenomics, and have presented miRNAs.
exceptional challenges and opportunities that warrant contin- As discussed earlier, effective prediction of miRNA–mRNA
ued research attention. The computational methodologies dis- interactions has encountered new challenges in the current
cussed in this article, either dedicated to circulating miRNAs genomic era where integration of multi-level information
or not, can facilitate future related research. Specifically, in becomes extremely important. New target prediction tools are
future study of miRNA sorting, researches can analyze known designed to use more molecular features and genomics profiles,
exosomal miRNAs (e.g. available in EVpedia [59], Vesiclepedia obtain accurate results, decrease the false detection, provide
[125] and ExoCarta [126]) using existing well-established motif easier interface and requires less configuration. However, on
tools such as MEME [66], DREME [67] and MDS2 [68]. Considering the other side, the limited knowledge of the rules that govern
different cells show distinct miRNA content in their exosomes, the mechanism of mRNA targeting and degradation hinders
possibly indicating a different loading mechanism, one may the validation of predictions. There is a lack of research on
carry out such analysis for each cell type, as illustrated in [68]. RBPs that act as determinants of miRNA targeting and the
Prediction results can be compared and integrated across dif- understanding of mechanisms related to targeting proficiencies.
ferent tools and generated consensus prediction. Note that in In this regard, it is particularly essential to weigh the biological
addition to sequence motif, structural information can be also aspects derived from data. Integration of data from various
important especially when handling RNA sequence longer than sources including expression profiles, RNA interactome and
miRNAs. other gene regulatory mechanisms can contribute immensely
Another profound application of the motif information is to construction of a close-to-reality model that reflects the
to reversely screen for miRNA loading proteins that are mech- complexity of (circulating) miRNA regulation.
anistically important but poorly studied. Archived exosomal As a result, the computational focus of miRNA targeting
data, including miRNA expression and proteomics profiles can research evolves from recognition of sequence-guided static
 Circulating miRNA trafficking and regulation 1321

Table 4. Selected methods and tools on dynamic miRNA regulation analysis in human disease study

Study Identification Integrated regulatory Data type Key properties Ref

model factors other than miRNA

CNV DM TF

Chen et al. Empirical Bayes  • Interactions based on • To identify common [121]

(2013) method negative correlation miRNA–mRNA
• Cell-specific expression interactions associated
with different types of
cancer
• Standalone package:
MCMG

Jacobsen et al. Multivariate   - • Predicted interactions • To infer recurrent [117]

(2013) linear regression using miRanda and cancer-associated
model TargetScan miRNA–target
• Cell-specific expression • No software available

Marcel et al. Unified  • Temporal gene expression • Discover miRNA [120]

(2013) probabilistic involvement in
model idiopathic pulmonary
fibrosis
• mirDREM

Balwierz et al. Simple linear - -  • Predicted interactions • To identify miRNA (and [118]
(2014) model using TargetScan TF) regulator based on
• Non-specific predicted regulatory
sites
• Webserver
discontinued

Li et al. (2013) Two-stage    • Predicted interactions • To model gene [119]

regression model using TargetScan with regulation in Acute
number of binding sites Myeloid Leukemia
• Cell-specific expression • Standalone package:
RACER

Kazan et al. Random forests    • Predicted interactions • To model gene [122]

(2016) using TargetScan regulation in liver
• Cell-specific expression hepatocellular
carcinoma
• No software available

Setty et al. Regression model  -  • Predicted based on 7-mer • To infer miRNA (and [113]
(2012) with seed matches TF) mediated gene
group-LASSO • Cell-specific expression regulation in
glioblastoma

Shu et al. Meta-Lasso    • Experimentally validated • To identify dynamic [52]

(2017) regression models interactions miRNA regulation in
with Frisch– • Cell-specific expression different cancers and
Waugh–Lovell and binding sites stages
method • Integration of
competitive regulation
• Standalone package:
miRDR, along with an
online database

Continued.
 1322 Cui and Shu

Table 4. Continued.

Study Identification Integrated regulatory Data type Key properties Ref

model factors other than miRNA

CNV DM TF

Li et al. (2018) Ensembled  • Predicted and experimental • To construct [146]

network model interactions cancer-specific miRNA
• Annotated cancer related and TF co-regulation
interactions netoworks
• Expression data from TCGA • Manual curation of
cancer-related
interactions
• Incorporate expression
data

CNV: copy number variation; DM: DNA methylation; TF: transcription factor.

binding site to the identification of expression-guided context a powerful analytical tool to understand complex multifaceted
dependent interactions and now is moving toward more holistic regulatory processes in human disease with more holistic and
network analysis. Realization of the dynamic features of the systems views.
networks that involve both competitive and cooperative miRNA
regulation is rather challenging and only a handful of tools and
methods are currently available. Integrated network analyses
come with challenges in heterogeneous data and model
Key Points
integration and the best practices are still far from established. • Major computational studies focused on microRNA
However, with more reliable and comprehensive genomic data circulation, targeting and regulation and have been
becoming available and novel ideas emerging, research on reviewed in this article along with discussion on various
this topic will continue to be of great interest to the research bioinformatics techniques and tools.
community. Further refined models will allow for a better • MicroRNAs, including circulating miRNAs, can regulate
understanding of regulatory mechanisms involving miRNAs and genes in a dynamic fashion where competitive and
facilitate the identification of intervention targets for disease cooperative binding has been involved. Computational
study. prediction of microRNA regulation has to take into
Note that when studying circulating miRNA regulation in consideration such mechanism.
distant cells or organism, new challenges arise. For examples, • Circulating microRNAs play important roles in cell
dietary miRNAs (bovine miR-200c and 29b [12, 13] and plant miR- communication and exogenous microRNAs can be
168 and 2511 [10, 11]), as exogenous confounders incorporated taken from diet. Both have strong implication in health-
into human system, may have identical or similar sequence with related processes.
human homologs, making it extremely difficult to distinguish • MicroRNA cross-species transfer as an emerging topic
the endogenous and exogenous sources. Plants have distinct in the microRNA field has begun to draw research
miRNA sequences from animal, and the different miRNA–gene attentions and a handful studies on microRNA sorting
interaction mechanisms often lead to different gene regulation. mechanisms have been discussed.
In this regard, new experimental protocols have been proposed • Further research focused on discovery of molecular
to differentiate the source [133]. Comparative sequencing analy- mechanism of sorting and elucidation of microRNA
sis for exogenous miRNA detection represents a non-trivial task function through reliable modeling of conditional
that involves large-scale cross-species comparison where new microRNA regulation is emerging. New insights about
strategies were discussed [48]. For functional study, given the future applications of those tools are presented.
unprecedented availability of genomic profiles including expres-
sions and interactions that has not yet been broadly applicable in
the nutrigenomics field, the fact that no systematic assessment
is available in cross-species miRNA binding reveals the need for Acknowledgements
new approaches.
We thank all members of Systems Biology and Biomedical
At last but not the least, given the challenges of performing
Informatics group for their helpful discussion and com-
large-scale experimental validation on predicted miRNA–mRNA
ments.
interactions, especially in a dynamic and conditional manner,
and the unavailable quality benchmark data for tool evaluation,
it poses significant challenges to the users. Apart from a clear Funding
need for dedicated efforts on this issue, a general understand-
ing of the properties and rationales, strength and weakness This work was supported by the National Institutes of
of each class of methodologies is expected in order to make Health (NIH) funded COBRE grant (1P20GM104320), NIH [1R01
right selection of the analysis. Again, the rule of thumb is that DK107264]/NIFA (2016-67001-06314), University of Nebraska
a unified model that combines condition-specific (ideally time (UNL) Food for Health seed grant and Layman grant, and the
series) high-throughput data in various types can provide us Tobacco Settlement Fund as part of J.C.’s startup grant.
 Circulating miRNA trafficking and regulation 1323

References 20. Hoy AM, Lundie RJ, Ivens A, et al. Parasite-derived microR-
NAs in host serum as novel biomarkers of helminth infec-
1. Lee J, Padhye A, Sharma A, et al. A pathway involving
tion. PLoS Negl Trop Dis 2014;8:e2701.
farnesoid X receptor and small heterodimer partner pos-
21. Zhang T, Zhao YL, Zhao JH, et al. Cotton plants export
itively regulates hepatic sirtuin 1 levels via microRNA-34a
microRNAs to inhibit virulence gene expression in a fungal
inhibition. J Biol Chem 2010;285:12604–11.
2. Choi SE, Fu T, Seok S, et al. Elevated microRNA-34a in pathogen. Nat Plants 2016;2:16153.
obesity reduces NAD+ levels and SIRT1 activity by directly 22. Liu S, da AP, Rezende RM, et al. The host shapes the gut
targeting NAMPT. Aging Cell 2013;12:1062–72. microbiota via fecal microRNA. Cell Host Microbe 2016;19:
3. Friedman RC, Farh KK, Burge CB, et al. Most mammalian 32–43.
mRNAs are conserved targets of microRNAs. Genome Res 23. Bottini S, Hamouda-Tekaya N, Tanasa B, et al. From bench-
2009;19:92–105. marking HITS-CLIP peak detection programs to a new
4. Shukla GC, Singh J, Barik S. MicroRNAs: processing, matu- method for identification of miRNA-binding sites from
ration, target recognition and regulatory functions. Mol Cell Ago2-CLIP data. Nucleic Acids Res 2017;45:e71.
Pharmacol 2011;3:83–92. 24. Valadi H, Ekstrom K, Bossios A, et al. Exosome-mediated
5. Selbach M, Schwanhausser B, Thierfelder N, et al. transfer of mRNAs and microRNAs is a novel mechanism of
Widespread changes in protein synthesis induced by genetic exchange between cells. Nat Cell Biol 2007;9:654–9.
microRNAs. Nature 2008;455:58–63. 25. Hunter MP, Ismail N, Zhang X, et al. Detection of microRNA
6. Zhou G, Zhou Y, Chen X. New insight into inter-kingdom expression in human peripheral blood microvesicles. PLoS
communication: horizontal transfer of mobile small RNAs. One 2008;3: e3694.
Front Microbiol 2017;8:768. 26. Diehl P, Fricke A, Sander L, et al. Microparticles: major
7. Simons M, Raposo G. Exosomes–vesicular carriers for transport vehicles for distinct microRNAs in circulation.
intercellular communication. Curr Opin Cell Biol 2009; Cardiovasc Res 2012;93:633–44.
21:575–81. 27. Turchinovich A, Weiz L, Langheinz A, et al. Characteriza-
8. Thery C. Exosomes: secreted vesicles and intercellular tion of extracellular circulating microRNA. Nucleic Acids Res
communications. F1000 Biol Rep 2011;3:15. 2011;39:7223–33.
9. Vickers KC, Remaley AT. Lipid-based carriers of microR- 28. Arroyo JD, Chevillet JR, Kroh EM, et al. Argonaute2 com-
NAs and intercellular communication. Curr Opin Lipidol plexes carry a population of circulating microRNAs inde-
2012;23:91–7. pendent of vesicles in human plasma. Proc Natl Acad Sci USA
10. Zhou Z, Li X, Liu J, et al. Honeysuckle-encoded atypical 2011;108:5003–8.
microRNA2911 directly targets influenza a viruses. Cell Res 29. Gibbings DJ, Ciaudo C, Erhardt M, et al. Multivesicular bodies
2015;25:39–49. associate with components of miRNA effector complexes
11. Zhang L, Hou D, Chen X, et al. Exogenous plant MIR168a and modulate miRNA activity. Nat Cell Biol 2009;11:1143–9.
specifically targets mammalian LDLRAP1: evidence of 30. Villarroya-Beltri C, Gutierrez-Vazquez C, Sanchez-Cabo F, et
cross-kingdom regulation by microRNA. Cell Res 2012;22: al. Sumoylated hnRNPA2B1 controls the sorting of miRNAs
107–26. into exosomes through binding to specific motifs. Nat Com-
12. Baier SR, Nguyen C, Xie F, et al. MicroRNAs are absorbed mun 2013;4:2980.
in biologically meaningful amounts from nutritionally rel- 31. Zhang Y, Liu D, Chen X, et al. Secreted monocytic miR-
evant doses of cow milk and affect gene expression in 150 enhances targeted endothelial cell migration. Mol Cell
peripheral blood mononuclear cells, HEK-293 kidney cell 2010;39:133–44.
cultures, and mouse livers. J Nutr 2014;144:1495–500. 32. Schneider A, Simons M. Exosomes: vesicular carriers for
13. Wang K, Li H, Yuan Y, et al. The complex exogenous RNA intercellular communication in neurodegenerative disor-
spectra in human plasma: an interface with human gut ders. Cell Tissue Res 2013;352:33–47.
biota? PLoS One 2012;7(12):e51009. 33. Mathivanan S, Ji H, Simpson RJ. Exosomes: extracellular
14. Kang W, Bang-Berthelsen CH, Holm A, et al. Survey of organelles important in intercellular communication. J Pro-
800+ data sets from human tissue and body fluid reveals teomics 2010;73:1907–20.
xenomiRs are likely artifacts. RNA 2017;23:433–45. 34. Taylor DD, Gercel-Taylor C. MicroRNA signatures of tumor-
15. Title AC, Denzler R, Stoffel M. Uptake and function stud- derived exosomes as diagnostic biomarkers of ovarian can-
ies of maternal milk-derived MicroRNAs. J Biol Chem 2015; cer. Gynecol Oncol 2008;110:13–21.
290:23680–91. 35. Li J, Chen X, Yi J, et al. Identification and characteriza-
16. Auerbach A, Vyas G, Li A, et al. Uptake of dietary milk tion of 293T cell-derived exosomes by profiling the pro-
miRNAs by adult humans: a validation study. F1000Res tein, mRNA and microRNA components. PLoS One 2016;11:
2016;5:721. e0163043.
17. Wolf T, Baier SR, Zempleni J. The intestinal trans- 36. Shurtleff MJ, Temoche-Diaz MM, Karfilis KV, et al. Y-box
port of bovine milk exosomes is mediated by endocy- protein 1 is required to sort microRNAs into exosomes in
tosis in human colon carcinoma Caco-2 cells and rat cells and in a cell-free reaction. Elife 2016;5.
small intestinal IEC-6 cells. J Nutr 2015;145(10):2201–6. 37. Zhou W, Fong MY, Min Y, et al. Cancer-secreted miR-105
doi: 10.3945/jn.115.218586. destroys vascular endothelial barriers to promote metas-
18. Kusuma RJ, Manca S, Friemel T, et al. Human vascular tasis. Cancer Cell 2014;25:501–15.
endothelial cells transport foreign exosomes from cow’s 38. Yang M, Chen J, Su F, et al. Microvesicles secreted
milk by endocytosis. Am J Physiol Cell Physiol 2016;310: by macrophages shuttle invasion-potentiating microRNAs
C800–7. into breast cancer cells. Mol Cancer 2011;10:117.
19. Li J, Zhang Y, Li D, et al. Small non-coding RNAs transfer 39. Ogata-Kawata H, Izumiya M, Kurioka D, et al. Circulating
through mammalian placenta and directly regulate fetal exosomal microRNAs as biomarkers of colon cancer. PLoS
gene expression. Protein Cell 2015;6:391–6. One 2014;9: e92921.
 1324 Cui and Shu

40. Que R, Ding G, Chen J, et al. Analysis of serum exosomal transferred into human circulation. PLoS One 2015;10:
microRNAs and clinicopathologic features of patients with e0140587.
pancreatic adenocarcinoma. World J Surg Oncol 2013;11:219. 61. Bembom O, Keles S, van der MJ. Supervised detection of
41. Leidinger P, Backes C, Dahmke IN, et al. What makes a blood conserved motifs in DNA sequences with cosmo. Stat Appl
cell based miRNA expression pattern disease specific?–a Genet Mol Biol 2007;6(Article8).
miRNome analysis of blood cell subsets in lung cancer 62. Sumazin P, Yang X, Chiu HS, et al. An extensive microRNA-
patients and healthy controls. Oncotarget 2014;5:9484–97. mediated network of RNA-RNA interactions regulates
42. Falcone G, Felsani A, D’Agnano I. Signaling by exosomal established oncogenic pathways in glioblastoma. Cell
microRNAs in cancer. J Exp Clin Cancer Res 2015;34:32. 2011;147:370–81.
43. Ferrante SC, Nadler EP, Pillai DK, et al. Adipocyte-derived 63. Santangelo L, Giurato G, Cicchini C, et al. The RNA-
exosomal miRNAs: a novel mechanism for obesity-related binding protein SYNCRIP is a component of the hepatocyte
disease. Pediatr Res 2015;77:447–54. Exosomal machinery controlling microRNA sorting. Cell Rep
44. Cavalieri D, Rizzetto L, Tocci N, et al. Plant microRNAs as 2016;17:799–808.
novel immunomodulatory agents. Sci Rep 2016;6:25761. 64. Ao W, Gaudet J, Kent WJ, et al. Environmentally induced
45. Cheng L, Sun X, Scicluna BJ, et al. Characterization and foregut remodeling by PHA-4/FoxA and DAF-12/NHR. Sci-
deep sequencing analysis of exosomal and non-exosomal ence 2004;305:1743–6.
miRNA in human urine. Kidney Int 2014;86:433–44. 65. Wei W, Yu XD. Comparative analysis of regulatory motif
46. Tsang EK, Abell NS, Li X, et al. Small RNA sequencing in cells discovery tools for transcription factor binding sites.
and exosomes identifies eQTLs and 14q32 as a region of Genomics Proteomics Bioinformatics 2007;5:131–42.
active export. G3 (Bethesda) 2017;7:31–9. 66. Bailey TL, Elkan C. Fitting a mixture model by expectation
47. Langevin S, Kuhnell D, Parry T, et al. Comprehensive maximization to discover motifs in biopolymers. Proc Int
microRNA-sequencing of exosomes derived from head and Conf Intell Syst Mol Biol 1994;2:28–36.
neck carcinoma cells in vitro reveals common secretion 67. Bailey TL. DREME: motif discovery in transcription factor
profiles and potential utility as salivary biomarkers. Onco- ChIP-seq data. Bioinformatics 2011;27:1653–9.
target 2017;8:82459–74. 68. Gao T, Shu J, Cui J. A systematic approach to RNA-
48. Zhang H, Vieira Resende ESB, Cui J. miRDis: a web tool for associated motif discovery. BMC Genomics 2018;19:146.
endogenous and exogenous microRNA discovery based on 69. Fratkin E, Naughton BT, Brutlag DL, et al. MotifCut: regu-
deep-sequencing data analysis. Brief Bioinform 2018;19:415– latory motifs finding with maximum density subgraphs.
24. doi: 10.1093/bib/bbw140. Bioinformatics 2006;22:e150–7.
49. Chi SW, Zang JB, Mele A, et al. Argonaute HITS-CLIP 70. Zhang S, Li S, Niu M, et al. MotifClick: prediction of
decodes microRNA-mRNA interaction maps. Nature cis-regulatory binding sites via merging cliques. BMC Bioin-
2009;460:479–86. formatics 2011;12:238.
50. Majoros WH, Lekprasert P, Mukherjee N, et al. MicroRNA tar- 71. Bartel DP. MicroRNAs: target recognition and regulatory
get site identification by integrating sequence and binding functions. Cell 2009;136:215–33.
information. Nat Methods 2013;10:630–3. 72. Rinck A, Preusse M, Laggerbauer B, et al. The human tran-
51. Hakguder Z, Shu J, Liao C, et al. Genome-scale MicroRNA scriptome is enriched for miRNA-binding sites located
target prediction through clustering with Dirichlet process in cooperativity-permitting distance. RNA Biol 2013;10:
mixture model. BMC Genomics 2018;19:658. 1125–35.
52. Shu J, Silva B, Gao T, et al. Dynamic and modularized 73. Kandeel M, Al-Taher A, Nakashima R, et al. Bioenergetics
MicroRNA regulation and its implication in human cancers. and gene silencing approaches for unraveling nucleotide
Sci Rep 2017;7:13356. recognition by the human EIF2C2/Ago2 PAZ domain. PLoS
53. Ahadi A, Sablok G, Hutvagner G. miRTar2GO: a novel One 2014;9: e94538.
rule-based model learning method for cell line specific 74. Zhou J, Cheng Y, Yin M, et al. Identification of novel miRNAs
microRNA target prediction that integrates Ago2 CLIP-seq and miRNA expression profiling in wheat hybrid necrosis.
and validated microRNA-target interaction data. Nucleic PLoS One 2015;10: e0117507.
Acids Res 2017;45:e42. 75. Jones-Rhoades MW, Bartel DP. Computational identifica-
54. Bartel DP. MicroRNAs: genomics, biogenesis, mechanism, tion of plant microRNAs and their targets, including a
and function. Cell 2004;116:281–97. stress-induced miRNA. Mol Cell 2004;14:787–99.
55. Krek A, Grun D, Poy MN, et al. Combinatorial microRNA 76. Zhan S, Lukens L. Identification of novel miRNAs and
target predictions. Nat Genet 2005;37:495–500. miRNA dependent developmental shifts of gene expression
56. Seitz H. Redefining microRNA targets. Curr Biol 2009; in Arabidopsis thaliana. PLoS One 2010;5: e10157.
19:870–3. 77. Campo S, Peris-Peris C, Sire C, et al. Identification of
57. Cannell IG, Kong YW, Bushell M. How do microRNAs a novel microRNA (miRNA) from rice that targets an
regulate gene expression? Biochem Soc Trans 2008;36: alternatively spliced transcript of the Nramp6 (natu-
1224–31. ral resistance-associated macrophage protein 6) gene
58. Squadrito ML, Baer C, Burdet F, et al. Endogenous RNAs involved in pathogen resistance. New Phytol 2013;199:
modulate microRNA sorting to exosomes and transfer to 212–27.
acceptor cells. Cell Rep 2014;8:1432–46. 78. Maragkakis M, Vergoulis T, Alexiou P, et al. DIANA-microT
59. Ohshima K, Inoue K, Fujiwara A, et al. Let-7 microRNA web server upgrade supports fly and worm miRNA target
family is selectively secreted into the extracellular environ- prediction and bibliographic miRNA to disease association.
ment via exosomes in a metastatic gastric cancer cell line. Nucleic Acids Res 2011;39:W145–8.
PLoS One 2010;5: e13247. 79. Mitra R, Bandyopadhyay S. MultiMiTar: a novel multi objec-
60. Shu J, Chiang K, Zempleni J, et al. Computational tive optimization based miRNA-target prediction method.
characterization of exogenous MicroRNAs that can be PLoS One 2011;6: e24583.
 Circulating miRNA trafficking and regulation 1325

80. Oulas A, Karathanasis N, Louloupi A, et al. A new microRNA 99. Kim D, Chang HR, Baek D. Rules for functional microRNA
target prediction tool identifies a novel interaction of a targeting. BMB Rep 2017;50:554–9.
putative miRNA with CCND2. RNA Biol 2012;9:1196–207. 100. Chou CH, Chang NW, Shrestha S, et al. miRTarBase
81. Shu J, Chiang K, Zhao DY et al. Human absorbable dietary 2016: updates to the experimentally validated miRNA-
microRNAs prediction based on an ensemble manifold target interactions database. Nucleic Acids Res 2016;44:
ranking model. IEEE International Conference on Bioinformatics D239–47.
and Biomedicine (IEEE BIBM 2015). IEEE Computer Society, 101. Kertesz M, Iovino N, Unnerstall U, et al. The role of site
Washington, DC, USA. accessibility in microRNA target recognition. Nat Genet
82. Bagga S, Bracht J, Hunter S, et al. Regulation by let-7 and 2007;39:1278–84.
lin-4 miRNAs results in target mRNA degradation. Cell 102. Miranda KC, Huynh T, Tay Y, et al. A pattern-based
2005;122:553–63. method for the identification of MicroRNA binding sites
83. Hu W, Coller J. What comes first: translational repression or and their corresponding heteroduplexes. Cell 2006;126:
mRNA degradation? The deepening mystery of microRNA 1203–17.
function. Cell Res 2012;22:1322–4. 103. Jiang Z, Ma J, Wang Q, et al. Circulating microRNA expres-
84. Li JH, Liu S, Zhou H, et al. starBase v2.0: decoding miRNA- sion and their target genes in deep vein thrombosis: a
ceRNA, miRNA-ncRNA and protein-RNA interaction net- systematic review and bioinformatics analysis. Medicine
works from large-scale CLIP-seq data. Nucleic Acids Res (Baltimore) 2017;96: e9330.
2014;42:D92–7. 104. Fan Y, Habib M, Xia J. Xeno-miRNet: a comprehensive
85. Riffo-Campos AL, Riquelme I, Brebi-Mieville P. Tools for database and analytics platform to explore xeno-miRNAs
sequence-based miRNA target prediction: what to choose? and their potential targets. PeerJ 2018;6:e5650.
Int J Mol Sci 2016;17(12):e1987. 105. Wang SR, Chen X, Ling S, et al. MicroRNA expression, tar-
86. Oliveira AC, Bovolenta LA, Nachtigall PG, et al. Com- geting, release dynamics and early-warning biomarkers in
bining results from distinct MicroRNA target prediction acute cardiotoxicity induced by triptolide in rats. Biomed
tools enhances the performance of analyses. Front Genet Pharmacother 2019;111:1467–77.
2017;8:59. 106. Liu B, Li J, Tsykin A. Discovery of functional miRNA-mRNA
87. Chen L, Heikkinen L, Wang C, et al. Trends in the develop- regulatory modules with computational methods. J Biomed
ment of miRNA bioinformatics tools. Brief Bioinform 2018. Inform 2009;42:685–91.
doi: 10.1093/bib/bby054. 107. Suzuki HI, Young RA, Sharp PA. Super-enhancer-mediated
88. Agarwal V, Bell GW, Nam JW, et al. Predicting effective RNA processing revealed by integrative MicroRNA network
microRNA target sites in mammalian mRNAs. Elife 2015;4. analysis. Cell 2017;168:1000–1014 e1015.
do: 10.7554/eLife.05005. 108. Ding M, Li J, Yu Y, et al. Integrated analysis of miRNA, gene,
89. Betel D, Wilson M, Gabow A, et al. The microRNA. and pathway regulatory networks in hepatic cancer stem
org resource: targets and expression. Nucleic Acids Res cells. J Transl Med 2015;13:259.
2008;36:D149–53. 109. Quitadamo A, Tian L, Hall B, et al. An integrated network
90. Paraskevopoulou MD, Georgakilas G, Kostoulas N, et al. of microRNA and gene expression in ovarian cancer. BMC
DIANA-microT web server v5.0: service integration into Bioinformatics 2015;16(Suppl 5):S5.
miRNA functional analysis workflows. Nucleic Acids Res 110. Huang GT, Athanassiou C, Benos PV. mirConnX: condition-
2013;41:W169–73. specific mRNA-microRNA network integrator. Nucleic Acids
91. Gumienny R, Zavolan M. Accurate transcriptome-wide Res 2011;39:W416–23.
prediction of microRNA targets and small interfering RNA 111. Lu Y, Zhou Y, Qu W, et al. A Lasso regression model for
off-targets with MIRZA-G. Nucleic Acids Res 2015;43: the construction of microRNA-target regulatory networks.
1380–91. Bioinformatics 2011;27:2406–13.
92. Bandyopadhyay S, Ghosh D, Mitra R, et al. MBSTAR: multiple 112. Muniategui A, Nogales-Cadenas R, Vazquez M, et al. Quan-
instance learning for predicting specific functional binding tification of miRNA-mRNA interactions. PLoS One 2012;7:
sites in microRNA targets. Sci Rep 2015;5:8004. e30766.
93. Xiao F, Zuo Z, Cai G, et al. miRecords: an integrated 113. Setty M, Helmy K, Khan AA, et al. Inferring transcriptional
resource for microRNA-target interactions. Nucleic Acids Res and microRNA-mediated regulatory programs in glioblas-
2009;37:D105–10. toma. Mol Syst Biol 2012;8:605.
94. Shuang C, Maozu G, Chunyu W, et al. MiRTDL: a deep 114. Betel D, Koppal A, Agius P, et al. Comprehensive
learning approach for miRNA target prediction. IEEE/ACM modeling of microRNA targets predicts functional
Trans Comput Biol Bioinform 2016;13:1161–9. non-conserved and non-canonical sites. Genome Biol
95. Lee B, Baek J, Park S et al. deepTarget: end-to-end learn- 2010;11:R90.
ing framework for microRNA target prediction using deep 115. Sadeghi M, Ranjbar B, Ganjalikhany MR, et al. MicroRNA
recurrent neural networks. In: The 7th ACM International and transcription factor gene regulatory network
Conference on Bioinformatics, Computational Biology, and Health analysis reveals key regulatory elements associated
Informatics. Seattle, WA, USA:ACM, 2016, p. 434–42. with prostate cancer progression. PLoS One 2016;11:
96. Pla A, Zhong X, Rayner S. miRAW: a deep learning-based e0168760.
approach to predict microRNA targets by analyzing whole 116. Bosia C, Sgro F, Conti L, et al. RNAs competing for microR-
microRNA transcripts. PLoS Comput Biol 2018;14: e1006185. NAs mutually influence their fluctuations in a highly non-
97. Kim D, Sung YM, Park J, et al. General rules for functional linear microRNA-dependent manner in single cells. Genome
microRNA targeting. Nat Genet 2016;48:1517–26. Biol 2017;18:37.
98. Li Q, Wang S, Huang CC, et al. Meta-analysis based vari- 117. Jacobsen A, Silber J, Harinath G, et al. Analysis of microRNA-
able selection for gene expression data. Biometrics 2014; target interactions across diverse cancer types. Nat Struct
70:872–80. Mol Biol 2013;20:1325–32.
 1326 Cui and Shu

118. Balwierz PJ, Pachkov M, Arnold P, et al. ISMARA: automated 133. Wang L, Sadri M, Giraud D, et al. RNase H2-dependent
modeling of genomic signals as a democracy of regulatory polymerase chain reaction and elimination of confounders
motifs. Genome Res 2014;24:869–84. in sample collection, storage, and analysis strengthen evi-
119. Li Y, Liang M, Zhang Z. Regression analysis of combined dence that microRNAs in bovine Milk are bioavailable in
gene expression regulation in acute myeloid leukemia. PLoS humans. J Nutr 2018;148:153–9.
Comput Biol 2014;10: e1003908. 134. Helwak A, Kudla G, Dudnakova T, et al. Mapping the human
120. Schulz MH, Pandit KV, Lino Cardenas CL, et al. Reconstruct- miRNA interactome by CLASH reveals frequent noncanon-
ing dynamic microRNA-regulated interaction networks. ical binding. Cell 2013;153:654–65.
Proc Natl Acad Sci USA 2013;110:15686–91. 135. Scheel TKH, Moore MJ, Luna JM, et al. Global mapping of
121. Chen X, Slack FJ, Zhao H. Joint analysis of expres- miRNA-target interactions in cattle (Bos taurus). Sci Rep
sion profiles from multiple cancers improves the iden- 2017;7:8190.
tification of microRNA-gene interactions. Bioinformatics 136. Luna JM, Barajas JM, Teng KY, et al. Argonaute CLIP defines
2013;29:2137–45. a deregulated miR-122-bound transcriptome that corre-
122. Kazan H. Modeling gene regulation in liver hepatocellular lates with patient survival in human liver cancer. Mol Cell
carcinoma with random forests. Biomed Res Int 2016;2016: 2017;67:400–410 e407.
1035945. 137. Luna JM, Scheel TK, Danino T, et al. Hepatitis C virus
123. Durbin RM, Altshuler DL, GaR A, et al. A map of RNA functionally sequesters miR-122. Cell 2015;160:
human genome variation from population-scale sequenc- 1099–110.
ing. Nature 2010;467:1061–73. 138. Loeb GB, Khan AA, Canner D, et al. Transcriptome-wide
124. Ogata H, Goto S, Sato K, et al. KEGG: Kyoto encyclopedia of miR-155 binding map reveals widespread noncanonical
genes and genomes. Nucleic Acids Res 1999;27:29–34. microRNA targeting. Mol Cell 2012;48:760–70.
125. Simpson RJ, Kalra H, Mathivanan S. ExoCarta as a resource 139. Stark TJ, Arnold JD, Spector DH, et al. High-resolution
for exosomal research. J Extracell Vesicles 2012;1. doi: profiling and analysis of viral and host small RNAs
10.3402/jev.v1i0.18374. eCollection 2012. during human cytomegalovirus infection. J Virol 2012;
126. Mathivanan S, Simpson RJ. ExoCarta: a compendium of 86:226–35.
exosomal proteins and RNA. Proteomics 2009;9:4997–5000. 140. Moore MJ, Scheel TK, Luna JM, et al. miRNA-target chimeras
127. Yanez-Mo M, Siljander PR, Andreu Z, et al. Biological prop- reveal miRNA 3 -end pairing as a major determinant of
erties of extracellular vesicles and their physiological func- Argonaute target specificity. Nat Commun 2015;6:8864.
tions. J Extracell Vesicles 2015;4:27066. 141. Grosswendt S, Filipchyk A, Manzano M, et al. Unambiguous
128. van GC, Rodrigues JP, Trellet M, et al. The HADDOCK2.2 web identification of miRNA:target site interactions by different
server: user-friendly integrative modeling of biomolecular types of ligation reactions. Mol Cell 2014;54:1042–54.
complexes. J Mol Biol 2016;428:720–5. 142. John B, Enright AJ, Aravin A, et al. Human MicroRNA targets.
129. Skog J, Wurdinger T, van Rijn S, et al. Glioblastoma PLoS Biol 2004;2:e363.
microvesicles transport RNA and proteins that promote 143. Liu Z, Guo Y, Pu X, et al. Dissecting the regulation rules of
tumour growth and provide diagnostic biomarkers. Nat Cell cancer-related miRNAs based on network analysis. Sci Rep
Biol 2008;10:1470–6. 2016;6:34172.
130. Guduric-Fuchs J, O’Connor A, Camp B, et al. Selective 144. Baglioni M, Russo F, Geraci F, et al. A new method for discov-
extracellular vesicle-mediated export of an overlapping ering disease-specific MiRNA-target regulatory networks.
set of microRNAs from multiple cell types. BMC Genomics PLoS One 2015;10: e0122473.
2012;13:357. 145. Koo J, Zhang J, Chaterji S. Tiresias: context-sensitive
131. Rana S, Malinowska K, Zoller M. Exosomal tumor approach to decipher the presence and strength of
microRNA modulates premetastatic organ cells. Neoplasia MicroRNA regulatory interactions. Theranostics 2018;8:
2013;15:281–95. 277–91.
132. Russo F, Di S, Vannini F, et al. miRandola 2017: a curated 146. Li R, Chen H, Jiang S, et al. CMTCN: a web tool for inves-
knowledge base of non-invasive biomarkers. Nucleic Acids tigating cancer-specific microRNA and transcription factor
Res 2018;46:D354–9. co-regulatory networks. PeerJ 2018;6:e5951.