Alessandra Rogato · Valeria Zazzu

Mario Guarracino Editors

Dynamics of
Mathematical
Models in
Biology
Bringing Mathematics to Life
Dynamics of Mathematical Models in Biology
Alessandra Rogato • Valeria Zazzu
Mario Guarracino
Editors

Dynamics of Mathematical
Models in Biology
Bringing Mathematics to Life

123
Editors
Alessandra Rogato Valeria Zazzu
National Research Council National Research Council
Institute of Biosciences and Bioresources Institute of Genetics and Biophysics, ABT
Naples, Italy Naples, Italy
Integrative Marine Ecology
Stazione Zoologica Anton Dohrn
Napoli, Italy

Mario Guarracino
Laboratory for Genomics, Transcriptomics
and Proteomics (Lab-GTP)
High Performance Computing
and Networking Institute (ICAR)
National Research Council (CNR)
Naples, Italy

ISBN 978-3-319-45722-2 ISBN 978-3-319-45723-9 (eBook)

DOI 10.1007/978-3-319-45723-9

Library of Congress Control Number: 2016955240

Mathematics Subject Classification (2010): 00B15 92B05

© Springer International Publishing Switzerland 2016

This work is subject to copyright. All rights are reserved by the Publisher, whether the whole or part of
the material is concerned, specifically the rights of translation, reprinting, reuse of illustrations, recitation,
broadcasting, reproduction on microfilms or in any other physical way, and transmission or information
storage and retrieval, electronic adaptation, computer software, or by similar or dissimilar methodology
now known or hereafter developed.
The use of general descriptive names, registered names, trademarks, service marks, etc. in this publication
does not imply, even in the absence of a specific statement, that such names are exempt from the relevant
protective laws and regulations and therefore free for general use.
The publisher, the authors and the editors are safe to assume that the advice and information in this book
are believed to be true and accurate at the date of publication. Neither the publisher nor the authors or
the editors give a warranty, express or implied, with respect to the material contained herein or for any
errors or omissions that may have been made.

Printed on acid-free paper

This Springer imprint is published by Springer Nature

The registered company is Springer International Publishing AG
The registered company address is: Gewerbestrasse 11, 6330 Cham, Switzerland
Preface

Despite Galileo’s claim that mathematics is the language of nature, the two
disciplines of mathematics and life sciences had been considered two planets
belonging to two very far galaxies which would never meet. The two communities
were vastly different and it seemed impossible for them to collaborate. Only recently
when life scientists began producing experimental data at an unprecedentedly high
pace, did it become clear that mathematical models were necessary to interpret
such data and to structure them, with the ultimate goal of unveiling biological
mechanisms, to making new discoveries, and to making predictions.
There are very few examples of events that bring the two communities together
to discuss research questions. For this reason we decided to create a series of annual
workshops to gather a multidisciplinary and international community. “Bringing
Maths to Life” enabled the two communities of life scientists and mathematicians
to exchange a bidirectional flow of ideas. The broad community of mathematician
enabled life scientists to introduce new algorithms, methods, and software that may
be useful to model life. Biologists enabled scientists to pose new challenges for
mathematicians, thereby bringing to life novel opportunities for mathematicians to
explore interesting problems. From this workshop many ideas and collaboration
began. In the second year of the workshop, the leitmotiv had surrounded the concepts
of time and dynamicity of nature. In the necessary simplifications applied during
the modeling process, time is sometimes not accounted for in an attempt to avoid
exponential complexity in computations. Nevertheless time imposed a different
thought paradigm, which in turn created more elegant mathematical models.
The second workshop, held during October 19–21, 2015, in Naples (Italy)
featured three main sessions. “Dynamics of genomes and genetic variation” was the
topic of the first session. In this session, we discussed the molecular mechanisms
and evolutionary processes that shape the structure and function of genomes and
that govern genome dynamics. The session on “Dynamics of motifs” provided an
overview of current methods for motif searching in DNA, RNA, and proteins, a
key process to discover emergent properties of cells, tissues, and organisms. The
third session was dedicated to the “Dynamics of biological networks.” Networks
representing complex biological functions and activities are useful to interpret

v
vi Preface

processes in the cell, and several mathematical models and algorithms are now
available for their integration, analysis, and characterization. As mentioned above,
in the necessary simplifications applied during the modeling process, time was often
not accounted for in an effort to avoid exponential complexity in computations.
In this volume we collect many of the important ideas that derived from the
workshop which are representative of the research questions that can be posed
within such multidisciplinary applications. In the first chapter, Verena Thormann
and colleagues describe the transcriptional regulation (when 1 C 1 ¤ 2). In the chap-
ter “Differential Network Analysis and Graph Classification: A Glocal Approach”,
a glocal approach to differential network analysis and graph classification is
introduced by Giuseppe Jurmal and colleagues. Maria Pia Saccomanni and Karl
Thomaseth discuss the identifiability of differential equation models that are used
in systems biology. In the chapter “Boolean Dynamics of Regulatory Compound
Circuits”, Elisabeth Remy et al. discuss the regulatory circuits and their dynamics.
Target genes of homologous transcription factors are differentially analyzed by
Elijah K. Lowe and colleagues. In the chapter “Reconstructing a Genetic Network
from Gene Perturbations in Secretory Pathway of Cancer Cell Lines”, a pipeline
for gene regulatory networks reconstruction is proposed by Marina Piccirillo
et al., and in the chapter “Dissecting the Functions of the Secretory Pathway
by Transcriptional Profiling” the functions of secretory pathways are analyzed
starting from transcriptional profiling by Sonali Gopichand Chavan and colleagues.
Saraunas Germanas et al. propose a Beta-Binomial model to detect rare mutations
in NGS experiments. In the chapter “An Overview of Genotyping by Sequencing
in Crop Species and Its Application in Pepper”, pepper genotyping by sequencing
is discussed by Francesca Taranto et al. Irma Terracciano and colleagues describe
in the chapter “Hybridization-Based Enrichment and Next Generation Sequencing
to Explore Genetic Diversity in Plants” how to explore genetic diversity in plants.
Lastly, in the chapter “DecontaMiner: A Pipeline for the Detection and Analysis
of Contaminating Sequences in Human NGS Sequencing Data”, Ilaria Granata
and colleagues describe a pipeline for the detection of sequences belonging to
contaminating organisms in human NGS sequencing data.
We would like to acknowledge the work and support we have received for
realizing this volume.
The workshop has been organized by Alessandra Rogato (Institute of Bio-
sciences and Bioresources), Valeria Zazzu and Enza Colonna (Institute of Genetics
and Biophysics “Adriano Buzzati-Traverso”), and Mario Guarracino (High Per-
formance Computing and Networking Institute and Institute for Higher Math-
ematics “F. Saveri”) from the Italian National Research Council (CNR), Italy.
Gerardo Toraldo from the Department of Mathematics and Applications “Renato
Caccioppoli,” University of Naples Federico II, contributed to the organization.
The initiative has been supported by the Italian National Research Council (CNR),
the Institute for High Mathematics “F. Saveri” (INDAM), the High Performance
Preface vii

Computing and Networking Institute (ICAR), the Institute of Biosciences and

Bioresources (IBBR), the Institute of Genetics and Biophysics “Adriano Buzzati-
Traverso” (IGB-ABT), the LABGTP, and the University of Naples Federico II.

Naples, Italy Alessandra Rogato

Valeria Zazzu
Mario Guarracino
Contents

Transcriptional Regulation: When 1 C 1 ¤ 2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1

Verena Thormann, Marina Borschiwer, and Sebastiaan H. Meijsing
Differential Network Analysis and Graph Classification:
A Glocal Approach . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
Giuseppe Jurman, Michele Filosi, Samantha Riccadonna,
Roberto Visintainer, and Cesare Furlanello
Structural vs Practical Identifiability of Nonlinear Differential
Equation Models in Systems Biology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
Maria Pia Saccomani and Karl Thomaseth
Boolean Dynamics of Compound Regulatory circuits . . . . . . . . . . . . . . . . . . . . . . . 43
Elisabeth Remy, Brigitte Mossé, and Denis Thieffry
A Differential Transcriptomic Approach to Compare Target
Genes of Homologous Transcription Factors in Echinoderm Species . . . . . . 55
Elijah K. Lowe, Claudia Cuomo, and Maria I. Arnone
Reconstructing a Genetic Network from Gene Perturbations
in Secretory Pathway of Cancer Cell Lines . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
Marina Piccirillo, Kumar Parijat Tripathi, Sonali Gopichand
Chavan, Seetharaman Parashuraman, Alessandra Varavallo,
and Mario Guarracino
Dissecting the Functions of the Secretory Pathway
by Transcriptional Profiling. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
Sonali Gopichand Chavan, Kumar Parijat Tripathi,
Marina Piccirilo, Prathyush Deepth Roy, Mario Guarracino,
Alberto Luini, and Seetharaman Parashuraman

ix
x Contents

Detection of Rare Mutations Using Beta-Binomial

and Empirical Quantile Models in Next-Generation
Sequencing Experiments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89
Sarunas Germanas, Audrone Jakaitiene, and Mario Guarracino
An Overview of Genotyping by Sequencing in Crop Species
and Its Application in Pepper . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101
Francesca Taranto, Nunzio D’Agostino, and Pasquale Tripodi
Hybridization-Based Enrichment and Next Generation
Sequencing to Explore Genetic Diversity in Plants . . . . . . . . . . . . . . . . . . . . . . . . . . . 117
Irma Terracciano, Concita Cantarella, and Nunzio D’Agostino
DecontaMiner: A Pipeline for the Detection and Analysis of
Contaminating Sequences in Human NGS Sequencing Data . . . . . . . . . . . . . . . 137
Ilaria Granata , Mara Sangiovanni , and Mario Guarracino
Contributors

Maria I. Arnone Stazione Zoologica Anton Dohrn, Naples, Italy

Marina Borschiwer Max Planck Institute for Molecular Genetics, Berlin,
Germany
Concita Cantarella Consiglio per la ricerca in agricoltura e l’analisi dell’economia
agraria (CREA) - Centro di ricerca per l’orticoltura, Pontecagnano Faiano (SA),
Italy
Sonali Gopichand Chavan Institute of Protein Biochemistry, National Research
Council (CNR-IBP), Naples, Italy
Claudia Cuomo Stazione Zoologica Anton Dohrn, Naples, Italy
Nunzio D’Agostino Consiglio per la ricerca in agricoltura e l’analisi
dell’economia agraria (CREA) – Centro di ricerca per l’orticoltura, Pontecagnano
Faiano (SA), Italy
Michele Filosi Fondazione Bruno Kessler, Trento, Italy
Cesare Furlanello Fondazione Bruno Kessler, Trento, Italy
Sarunas Germanas Institute of Mathematics and Informatics, Vilnius University,
Vilnius, Lithuania
Ilaria Granata ICAR-CNR, Naples, Italy
Mario Guarracino Laboratory for Genomics, Transcriptomics and Proteomics
(LAB-GTP), High Performance Computing and Networking Institute (ICAR),
National Research Council (CNR), Naples, Italy
Audrone Jakaitiene Department of Human and Medical Genetics, Faculty of
Medicine, Vilnius University, Vilnius, Lithuania
Giuseppe Jurman Fondazione Bruno Kessler, Trento, Italy
Elijah K. Lowe Stazione Zoologica Anton Dohrn, Naples, Italy

xi
xii Contributors

Beacon Center for Evolution in Action, Michigan State University, East Lansing,
MI, USA
Alberto Luini Institute of Protein Biochemistry at National Research Council
(CNR), Naples, Italy
Istituto di Ricovero e Cura a Carattere Scienti co SDN, Naples, Italy
Sebastiaan H. Meijsing Max Planck Institute for Molecular Genetics, Berlin,
Germany
Brigitte Mossé Aix Marseille Université, CNRS, Centrale Marseille, I2M, UMR
7373, Marseille, France
Seetharaman Parashuraman Institute of Protein Biochemistry, National
Research Council (CNR-IBP), Naples, Italy
Marina Piccirilo Laboratory for Genomics, Transcriptomics and Proteomics
(LAB-GTP), High Performance Computing and Networking Institute (ICAR) at
National Research Council (CNR), Naples, Italy
Elisabeth Remy Aix Marseille Université, CNRS, Centrale Marseille, I2M, UMR
7373, Marseille, France
Samantha Riccadonna Centro Ricerca e Innovazione, Fondazione Edmund Mach,
San Michele all’Adige, Italy
Prathyush Deepth Roy Institute of Protein Biochemistry at National Research
Council (CNR), Naples, Italy
Maria Pia Saccomani Department of Information Engineering, University of
Padova, Padova, Italy
Mara Sangiovanni ICAR-CNR, Naples, Italy
Francesca Taranto Consiglio per la ricerca in agricoltura e l’analisi dell’economia
agraria (CREA) – Centro di ricerca per l’orticoltura, Pontecagnano Faiano (SA),
Italy
Irma Terracciano Consiglio per la ricerca in agricoltura e l’analisi dell’economia
agraria (CREA) - Centro di ricerca per l’orticoltura, Pontecagnano Faiano (SA),
Italy
Denis Thieffry Computational Systems Biology team, Institut de Biologie de
l’Ecole Normale Supérieure (IBENS), CNRS UMR8197, INSERM U1024, Ecole
Normale Supérieure, PSL Research University, Paris, France
Karl Thomaseth Institute of Electronics, Computer and Telecommunication Engi-
neering (IEIIT-CNR) c/o DEI, Padova, Italy
Verena Thormann Max Planck Institute for Molecular Genetics, Berlin, Germany
Contributors xiii

Kumar Parijat Tripathi Laboratory for Genomics, Transcriptomics and Pro-

teomics (LAB-GTP), High Performance Computing and Networking Institute
(ICAR) at National Research Council (CNR), Naples, Italy
Pasquale Tripodi Consiglio per la ricerca in agricoltura e l’analisi dell’economia
agraria (CREA) – Centro di ricerca per l’orticoltura, Pontecagnano Faiano (SA),
Italy
Alessandra Varavallo Institute of Protein Biochemistry, National Research
Council (CNR-IBP), Naples, Italy
Roberto Visintainer Fondazione Bruno Kessler, Trento, Italy
Transcriptional Regulation: When 1 C 1 ¤ 2

Verena Thormann, Marina Borschiwer, and Sebastiaan H. Meijsing

Abstract One of the fascinating questions in biology is to understand how an

identical genome can give rise to distinct tissues with different functions, for
example, brain and muscle. A key role in selectively decoding the genome is played
by transcription factors (TFs), which bind to specific DNA sequences to help specify
if and how much of a gene is expressed in a particular tissue. In a simple scenario,
binding of TFs near a gene would result in activation of gene expression whereas
in the absence of binding the gene would not be expressed. One of the objectives
of computational biology is to use the genomic sequence to predict where TFs bind
and to both qualitatively and quantitatively predict which genes it regulates. In this
chapter, we will discuss how the information encoded in the genome in the form
of DNA can serve as a discreet code where combinations of As, Ts, Cs, and Gs
specify which TFs can bind. Further, structural features of DNA can be read by
proteins to influence their structure and fine-tune their activity towards target genes.
In practice, predicting genome-wide binding patterns of TFs based on sequence is
problematic and even when we know where TFs bind, all bets appear to be off
regarding the effect of TF binding on the regulation of genes. At the moment it
sometimes seems as if 1 C 1 ¤ 2 when studying gene regulation. However, this
mostly reflects our lack of understanding of the signaling inputs that specify if a
gene is activated and at which level it is expressed. For example, in this chapter
we will discuss how taking the three-dimensional organization of the genome and
the chromatin context in which these binding sites are embedded into account can
improve the link between binding of TFs and the regulation of genes. Eventually,
by adding more and more pieces of the puzzle, we hope to identify what is missing
in our current equations to model gene expression.

Keywords Transcriptional regulation • Transcription factor • Glucocorticoid

receptor • Computational modeling • Hi-C • Chromatin

V. Thormann • M. Borschiwer • S.H. Meijsing ()

Max Planck Institute for Molecular Genetics, Ihnestrasse 63-73, 14195 Berlin, Germany
e-mail: [email protected]

© Springer International Publishing Switzerland 2016 1

A. Rogato et al. (eds.), Dynamics of Mathematical Models in Biology,
DOI 10.1007/978-3-319-45723-9_1
2 V. Thormann et al.

1 Introduction

How can muscle cells have a distinct phenotype compared to blood cells although
both share the same genetic information? One answer to this fundamental question is
that different sets of genes are expressed in different cell types. Therefore, a detailed
understanding of the mechanisms that control the expression of genes is needed
to better understand how cells adopt and change their identity. Two key players
in regulating the expression of genes are cis- and trans-acting elements. The cis
elements are DNA sequences encoded in the genome that can be bound by trans-
acting transcription factors (TFs), which in turn can influence the recruitment or
activity of the RNA polymerase to influence the expression of genes. Notably, only
about 1 % of the genome codes for proteins, which leaves a large fraction of the
genome available for potential regulatory functions.
Activation of the right genes at the right place and at the right time is critical,
as the misexpression of genes can have pathological consequences. For example,
Sonic hedgehog is an essential gene involved in embryonic development of the limbs
and a failure to express this gene results in severe limb malformations, e.g., hands
with only one digit [1]. Similarly, expressing a gene in the wrong place can have
detrimental effects as was shown in the fruit fly Drosophila melanogaster where
misexpression of the Antp gene in the head leads to the growth of legs instead of
antennas [2]. In addition to expressing the right genes at the right place, proper
development requires genes to be expressed at the right level and a failure to
express genes at the right dosage can lead to impaired development and disease.
One well-known example is Down syndrome, where an extra copy of chromosome
21 and the resulting increased gene-dosage results in several severe developmental
defects. Similarly, expressing too little of the tumor suppressor gene p53 results in
an increased chance to develop cancer [3].
To regulate the expression of genes, TFs are recruited to specific regulatory
sequences, encoded in the genome (Fig. 1) [4]. These transcription factor binding
sites (TFBS) are specific DNA recognition sequences located in regulatory regions.
Typically, TFBSs for different TFs are found in clusters that can be referred to
as enhancers. These enhancers act on the promoter of genes to influence the
recruitment or activity of RNA polymerase and ultimately influence if and how
much of a gene is expressed (Fig. 1). Enhancers can be located proximal to the
promoter or at a large distance from the transcriptional start site (TSS) of genes [5],
which raises the question how enhancers that are remote from the promoter in linear
space can influence events at the promoter of genes. One explanation is the fact
that looping of the DNA and its three-dimensional organization in the nucleus can
bring together sequences that are remote in linear space [6]. Other levels of genome
organization that influence the functioning of enhancers include the fact that the
DNA in the nucleus is wrapped around histone proteins to form nucleosomes. The
tails of these histones can be post-translationally modified and specific modifications
were shown to correlate with the activity of enhancer elements [7].
Transcriptional Regulation: When 1 C 1 ¤ 2 3

Fig. 1 Signaling pathway of the glucocorticoid receptor. Unbound glucocorticoid receptor (GR)
resides in the cytoplasm and upon binding to its cognate steroid hormone (dark red) translocates
to the nucleus where it interacts with GR binding sites (GBS) in the promoter and/or in
enhancer regions. Genome-bound GR, together with cofactors and other transcription factors
bound to transcription factor binding sites (TFBS), influences the recruitment and activity of RNA
polymerase II to ultimately regulate the expression of its target gene

In this chapter we describe efforts to predict TF binding based on sequence and

TF-dependent gene regulation based on their genome-wide binding pattern. As a
model TF, we will often refer to findings for the glucocorticoid receptor (GR), a
member of the nuclear steroid hormone receptor family. The reason for using this
TF as a model is that GR’s activity is strictly hormone-dependent. In the absence
of hormone, GR resides in the cytoplasm, whereas upon hormone activation, GR
translocates to the nucleus and binds to specific DNA sequences to regulate the
expression of genes (Fig. 1). This hormonal on/off switch allows the relatively
simple identification of putative target genes, by comparing the expression of genes
between cells treated with hormone with untreated cells. Notably, depending on the
response element bound, GR can either activate or repress the expression of genes
[8]. Importantly, fundamental insights derived from studies using GR likely also
apply to other TFs.
The aim of this chapter is to give the reader insight into mechanisms that specify
where TFs bind in the genome and once bound, how they may, or may not, influence
the expression of genes. First we will discuss how genomic TF binding can be
predicted based on DNA sequence, but also depict the limitations of sequence-based
4 V. Thormann et al.

predictions. Second, we will discuss attempts to link the binding of TFs to the
regulation of genes, the role of the three-dimensional organization of the genome
in the nucleus, and how the sequence of TFBSs influence how much of a target gene
is expressed. Finally, we will present an outlook of how newly developed methods
can contribute to our understanding of the role of TFs and TFBSs in orchestrating
the expression of genes.

2 When 1 C 1 ¤ 2: The Prediction of Genomic Transcription

Factor Binding Sites Based on DNA Sequence

One of the crucial steps in the regulation of gene expression is the binding of
sequence-specific TFs to regulatory DNA sequences associated with their target
genes. In principle, the binding of TFs can be predicted from sequence. In practice,
however, sequence alone is a poor predictor of TF binding. This is in part a
consequence of the fact that TFBSs are typically short and degenerate and thus
potential binding sites are ubiquitously present in the genome and only a minority
of these potential binding sites is actually bound by TFs. Furthermore, TF binding is
often highly cell-type specific despite the fact that these cell types harbor the same
genome.
Experimentally, in vivo genome-wide binding of TFs can be determined by
chromatin immunoprecipitation (ChIP)-based techniques (Fig. 2a). As a first step
of the ChIP procedure, formaldehyde is used to covalently cross-link TFs to their
genomic binding sites. Subsequently, the cross-linked DNA is sheared into smaller
fragments of approximately 200–300 base pairs in length and the resulting protein–
DNA complexes are co-precipitated using an antibody specific for the TF of interest.
Finally, either qPCR-based methods or DNA sequencing (for ChIP-seq) identifies
the enrichment of DNA sequences that are occupied by a given TF (Fig. 2b). In the
past decade, the advent of next generation sequencing methods resulted in a wealth
of available genome-wide ChIP-seq data for different TFs from a wide variety of
different cell types, tissues, and model organisms. From this data, the recognition
sequence of a TF of interest can be derived using computational methods. These
methods can uncover sequences that are over-represented in regions bound by a
specific TF and can be used to generate a consensus motif. The consensus motif can
be graphically displayed as a sequence logo to represent the position weight matrix
(PWM) which describes the nucleotide preference at each nucleotide position within
the motif (Fig. 2c) [9].
Conceivably, the PWM could now be used to directly predict TF binding to a
given DNA sequence or even an entire genome. However, prediction of genome-
wide binding based on the PWM typically fails for several reasons. First, not all
DNA sequences that are bound in vivo match the consensus motif. This could
be due to the fact that some TFs can bind to highly degenerate sequences, in
which up to several base pairs can differ from its consensus sequence [10].
Transcriptional Regulation: When 1 C 1 ¤ 2 5

2
PWM for GR

bits
1

10
11
12
13
14
15
16
1
2
3
4
5
6
7
8
9
5’ 3’
position

Fig. 2 (a) Chromatin immunoprecipitation (ChIP)-sequencing for the identification of in vivo

TFBSs. For ChIP-seq, protein-DNA interactions are fixed by the addition of formaldehyde. Next,
the fixed chromatin is sheared into smaller fragments by sonication. DNA-fragments occupied by
the TF of interest, here GR, are enriched by immunoprecipitation using a GR-specific antibody. (b)
Generation of ChIP-seq tracks. The genome-wide location of TFBSs is analyzed by mapping and
quantifying DNA-sequences obtained from ChIP-seq. (c) Generation of a position weight matrix
(PWM). TFBSs can be represented by a PWM, describing the binding preferences of a given TF
(here depicted for GR). To generate a PWM, DNA sequences obtained from ChIP-seq (or by other
experimental approaches such as SELEX) are aligned and screened for TF binding motifs

Cooperative binding with other TFs can turn such degenerate sequences into high
affinity binding sites. For example, GR was reported to bind together with AP1 at
composite regulatory sequences to cooperatively regulate Notch4 gene expression
[11]. Moreover, some TFs can bind without direct contact to the DNA by binding
to other proteins, a mechanism referred to as DNA tethering. Hence, at tethered
regions computational prediction of TFBSs using the PWM would miss indirect
interactions mediated by other proteins. For example, studies using human cell lines
have shown that GR binds at promoter regions of genes involved in mediating
the immune-modulating actions of glucocorticoids that contain no obvious GR
consensus motif. At these regions, GR-tethering to NFkB was shown to be an
important mechanism responsible for GR-mediated gene regulation [12]. A second
reason why PWMs fail to accurately predict genome-wide TF binding patterns is
that not all computationally predicted DNA sequences are bound in vivo. In fact,
only a minor fraction of all possible sequences matching the consensus motif of a TF
are actually bound in vivo. For GR, the vast majority of genomic GR binding sites
6 V. Thormann et al.

are located in the so-called open chromatin [13], arguing that chromatin accessibility
(as assayed by DNase-I hypersensitivity assays) is a key player in specifying which
of the potential binding sites encoded in the genome can be bound. Changes in
chromatin accessibility, which can occur in response to environmental signals and
during cellular differentiation [14, 15], can thus explain why TF occupancy can be
highly cell-type [16, 17] and cell-stage specific [18].
Together, the computational prediction of genomic TFBSs suffers from two
critical issues. First, false-negative predictions, when TFBSs are missed due to TF
tethering by other DNA-binding factors or by binding to degenerate sequences.
Notably, comparison of different computational models for the prediction of TF
binding specificity showed that most often the best performing motifs were those
with the highest nucleotide degeneracy [19]. Second, computational prediction of
TFBSs may result in false-positive predictions for TFBSs that match the consensus
but are not available for TF binding in vivo, e.g., due to their location in closed
chromatin.

3 When 1 C 1 ¤ 2: The Prediction of Regulatory Activity

Based on Genomic TF-DNA-Binding

TF binding and the regulation of nearby genes are clearly connected. However,
this link is typically statistical rather than deterministic. For example, scanning a
window of 300 kb around the TSS of genes showed that for all genes with a GR
binding site in this window, only a fraction actually change their expression in
response to GR binding (Meijsing lab unpublished results). Although the fraction
of regulated genes is higher when only TFBSs in close proximity to the TSS are
considered, the link between promoter-proximal GR binding and gene regulation
remains far from deterministic. Similarly, ChIP-seq experiments typically uncover
several thousands of peaks for an individual TF, whereas TF perturbations usually
result in only a small number of affected genes [20, 21]. Consequently, TF binding is
a poor predictor of gene regulation and understanding what distinguishes productive
TF binding events (resulting in the regulation of a gene) from non-productive
binding events remains a key challenge.
One additional signal that may help distinguish productive from non-productive
binding events is the post-translational modification state of the histones located at
the enhancer regions harboring TFBSs. For example, actively transcribed promoters
and active enhancers show elevated levels of histone H3 lysine 27 acetylation
(H3K27ac) [22, 23]. Thus, one possibility to computationally predict productive TF
binding events is to combine information regarding TF binding with the occurrence
of specific histone modifications. Such computational strategies were shown to
be quite successful, especially when additional information such as sequence
conservation, DNA accessibility, or gene expression data were also taken into
Transcriptional Regulation: When 1 C 1 ¤ 2 7

account [24]. Testing if predicted enhancers are indeed capable of regulating the
expression of genes is traditionally done using reporter gene assays. To test their
activity, the predicted regulatory region is cloned in front of a minimal promoter
sequence that drives the expression of a reporter gene, e.g., the expression of
the luciferase gene (Fig. 3b). Next, the regulatory activity of a given TFBS can
be analyzed in a heterologous context by measuring the amount of reporter gene
activity. The presence of specific histone modifications at the enhancer region can
serve as a good indicator of in vivo regulatory activity as detected by reporter gene
assays [23]. However, the accuracy of the prediction is limited. For example, a high-
throughput functional screen of enhancers computationally predicted based on their
pattern of histone modifications, showed that only about one-fourth of all tested

Fig. 3 (a) Endogenous regulation of gene expression by enhancers. In vivo, bound TFBSs are
mainly located in open chromatin regions, where they either bind directly to DNA or indirectly by
tethering to other DNA-binding TFs. Productive TF binding can be influenced by the presence of
associated chromatin marks or the occurrence of other co-factors. TFBSs can be located several
thousands of kilo bases away from their target genes and can regulate gene expression by DNA-
looping. To regulate gene expression, bound TFs influence the recruitment and activity of RNA
polymerase II. (b) Reporter gene assays. To test the regulatory activity of a TFBS in reporter
gene assays, the candidate regulatory region is cloned in front of a minimal promoter that drives
the expression of a reporter gene. Upon transfection of the reporter plasmid into living cells, its
regulatory activity can be analyzed by measuring the amount of generated gene product. In the
depicted example, the regulatory activity of the tested regulatory region correlates with the level
of luciferase activity. (c) Tab.1. Features influencing the regulation of gene expression in vivo in
comparison to reporter gene assays
8 V. Thormann et al.

sequences was indeed active in reporter gene assays. Especially the classification
into strong and weak enhancers based on their level of histone modifications did
not have a great predictive value [25]. This could either mean that the predictions
are wrong or that the reporter setting fails to recapitulate the complexity of gene
regulation in the endogenous context. Regarding the latter, reporter genes differ
from the endogenous genomic setting at which gene regulation takes place in a
number of ways (Fig. 3). These differences include the fact that enhancers and
TFBS are typically tested using a heterologous promoter and that reporters fail
to recapitulate the endogenous sequence context or the chromatin environment of
the investigated TFBS. Therefore, a regulatory sequence that is unable to drive
reporter gene expression must not necessarily be inert in its natural genomic context.
Conversely, the ability of an enhancer to activate the reporter gene does not proof
that an enhancer region is capable of doing the same in the endogenous genomic
context.
Notably, even when the function of putative enhancers is tested in their endoge-
nous genomic context the results might be hard to interpret. For example, studies
in Drosophila showed that the deletion of two enhancers linked to the expression
of an important developmental gene resulted in only minor developmental defects
when cultured under standard laboratory conditions. In contrast, at high or low
temperatures the deletion of these obviously non-functional enhancers resulted
in pronounced developmental defects [26]. This shows that the importance of
enhancers might be context-dependent and only become apparent under specific
environmental conditions. Furthermore, functional redundancy among enhancers
might mask the functional importance of a specific enhancer when they are mutated
individually [27].
In summary, although on a global scale the binding of TFs and the regulation of
genes are clearly connected, if and how TF binding and gene regulation are linked
at individual genes is typically unknown. Thus, unraveling the operating principles
that specify which binding events are productive remains a major challenge.

4 When 1 C 1 ¤ 2: The Prediction of Target Genes

by Incorporating the Three-Dimensional Genome
Organization

In a classical view of transcriptional regulation, TFs bind to TFBSs located proximal

to the promoter region of their target genes. Subsequently, bound TFBSs can
serve as a binding platform to recruit other co-factors and RNA polymerase II to
ultimately regulate gene expression. This classical view of gene regulation justifies
approaches where target genes of a specific TFBS are predicted simply by assigning
them to the gene whose TSS is closest. In support of this strategy, computational
correlation of enhancer activity in reporter assays with gene expression profiles
indicated that the majority of enhancers indeed act on the nearest gene. Nevertheless,
Transcriptional Regulation: When 1 C 1 ¤ 2 9

up to 21 % of all enhancer candidates appeared to regulate more distal genes,

suggesting that long-range regulation is a common phenomenon contributing to
the complexity of transcriptional gene regulation [18]. Moreover, for many TFs,
including GR, the majority of TFBSs identified by ChIP-seq are located distal from
TSSs, suggesting that long-range enhancer–promoter interactions play a role in GR-
mediated gene regulation [28].
For TFBSs that are localized at great linear distances from the TSS of genes,
simply assigning them to the closest gene might be conceptually flawed due to the
fact that based on their distance in linear space, the promoter and TFBS could in
principle be located at opposite ends of the nucleus. In this case it would be unlikely
that the promoter and the TFBS are functionally connected. In three-dimensional
space, however, TFBSs that are remote in linear space might be in close proximity
to promoters by looping of the flexible DNA polymer. Accordingly, imaging
approaches have shown that remote enhancers can be in close proximity to the
promoter of their target genes in three-dimensional space [29]. Furthermore, studies
of the mammalian “-globin locus have uncovered that remote enhancers can regulate
the expression of the closest gene, but also of other genes that are located further
away in linear space [30, 31]. In three-dimensional space, however, these enhancers
appear to be in close proximity to the promoters of several regulated genes within
the cluster as assayed by chromosome conformation capture (3C), which maps the
spatial organization of the genome in the nucleus [32, 33]. (3C)-based techniques
such as 4C, 5C, or Hi-C (see [34] for a detailed overview of the different techniques)
rely on the ability of formaldehyde to cross-link DNA-protein complexes and the
assumption that loci in close spatial proximity have a higher probability to become
part of the same cross-linked DNA-protein complex. The fixed chromatin is cut
with a restriction enzyme and a subsequent ligation step joins DNA molecules that
are in the same DNA-protein complex resulting in unique chimeric DNA-hybrids.
Finally, the mapping of sequences obtained from DNA sequencing identifies pairs of
loci that interact with a higher frequency than expected from random collision [34].
Genome-wide analysis of long-range interactions revealed that enhancer–promoter
interactions predominantly take place within chromosomal units up to several
megabases in size, referred to as topologically associating domains (TAD) [35].
How these chromosomal domains are established and maintained is largely unclear
but two proteins, CTCF and cohesin, appear to be important for both DNA-looping
and TAD establishment [36]. The resolution of 3C-based techniques, such as 4C,
currently lies in the range of tens of kilobases (kbs). Of note, 3C-based methods
generally suffer from a high background for regions close to the viewpoint and
can therefore only give reliable information for long-distance enhancer–promoter
interactions. Furthermore, it is important to keep in mind that 3C-based methods
identify all physical interactions that are present in a given cell population [34]. In
fact, evidence from single-cell Hi-C showed that the overall domain organization
of TADs at megabase scale remains relatively stable among single cells. However,
some individual chromosomal contacts could vary quite dramatically from cell to
cell [37]. In addition, a high relative interaction frequency, as revealed by 3C, does
10 V. Thormann et al.

not necessarily reflect a functional promoter–enhancer interaction but could also

simply be a consequence of how the DNA is packaged in the nucleus or its co-
localization to a distinct sub-nuclear structure [34].
Despite its limitations, 3C-based experiments have yielded important insights
into the three-dimensional organization of the genome in the nucleus and its role in
gene regulation. One of these studies, investigating Hi-C data across nine different
cell types, revealed that the vast majority of DNA-looping interactions were highly
conserved among cell types and even between different species. In addition to
these invariant interactions, the study reported the occurrence of a relatively small
number of cell-type specific enhancer–promoter interactions that correlated with
distinct cell-type specific gene expression patterns [38]. In this context, emerging
evidence suggests that the establishment of such cell-type specific DNA-loops
depends on the expression of cell-type specific TFs [39, 40]. Conceivably, these TFs
might either recruit other co-factors required for DNA-looping or self-assembly to
efficiently bridge DNA interactions and promote cell-type specific DNA-looping
[41]. In support of this hypothesis, several studies showed that a knock-out of
tissue-specific TFs destabilized cell-type specific enhancer–promoter interactions
[40, 42]. Another intriguing question is if, and how, these enhancer–promoter
contacts might change in response to different environmental stimuli. Surprisingly,
Hi-C data from human cells after TNF-alpha treatment showed no significant
changes for the vast majority of DNA-looping contacts. Similarly, exposure to other
stimuli such as IFN-gamma or estradiol resulted in only few changes in looping
contacts. This finding suggests that most contacts are already preformed even in
the absence of an activated signaling cascade [43] and that TFs that act upon these
stimuli mostly use pre-established enhancer–promoter interactions. The prediction
of target genes is further complicated by the possibility that promoters might
interact with multiple enhancers. Given that genomic TFBSs by far outnumber the
number of genes, it is indeed reasonable to assume that most genes are regulated
by several enhancers. Indeed, 3C-based approaches revealed that TSS-viewpoints
most often show contacts with multiple enhancer regions [38, 43]. Supporting
this hypothesis, the effect of individual TF knock down on gene expression levels
showed a negative correlation with the number of interacting TFBSs, suggesting
that redundant TFs could rescue transcriptional outcome by integration of signals
from several enhancers [21]. Furthermore, single nucleotide polymorphisms (SNP)
in TFBSs at the population level rarely result in dramatic gene expression changes
or disease phenotypes [44] and if effects were observed it required the presence of
simultaneous SNPs in multiple enhancer regions [45]. Hence, cooperative regulation
of gene expression by multiple TFBSs and the creation of regulatory hubs might
be a common mechanism to ensure regulatory robustness by integrating regulatory
signals from both remote contacts and promoter-proximal regions.
Notably, a recent study integrated information regarding DNA-looping from Hi-
C data with genome-wide TF binding based on ChIP-seq experiments to predict
TF-dependent gene regulation [43]. This study showed that genes associated with
TFBSs looping to their promoters are more likely to be regulated than their
counterparts that have TFBSs at the same distance without looping contacts.
Transcriptional Regulation: When 1 C 1 ¤ 2 11

Furthermore, this study found that the magnitude of gene expression changes
increased with an increasing number of TFBSs that show long-range interactions
[43]. This suggests that data from 3C-based approaches can help identify, or at least
enrich, for productive TF binding events that result in the regulation of associated
target genes. However, although the ability to predict changes in gene expression
improves when taking long-range interactions into account, the connection is still
far from deterministic. This might in part be due to the limited resolution of the
Hi-C experiments (5–10 kb range), which could result in false-positive enhancer–
promoter contacts and might thus improve further if technological advances improve
the resolution of 3C-based methods.

5 When 1 C 1 ¤ 2: The Prediction of Gene Expression Level

Based on Transcription Factor Binding Strength

So far we have discussed the regulatory activity of TFBSs and their associated target
gene expression as an all-or-nothing event where genes are either regulated by a TF
or not. However, in addition to expressing the right genes at the right time, getting
the dosage of individual genes right is important for development and homeostasis.
This fine-tuning of gene expression is a consequence of the integration of several
signaling inputs that impinge on a gene. These inputs include the combinatorial
interactions of TFs at response elements, post-translational modifications of DNA,
RNA, and proteins, and processes that influence the stability of RNA once produced.
Here, we will focus on one signaling input that can influence the level of expression
of genes: the sequence of the TFBS and its sequence environment. One mechanism
by which the sequence of a TFBS can influence transcriptional output is through
differences in TF affinity, with high affinity binding sites resulting in more TF
recruitment and consequently higher expression levels of associated target genes.
However, in addition to affinity-driven differences in activity, TFBS sequence
variants may also modulate transcriptional output by acting as allosteric ligands
that influence the structure and activity of associated TFs towards their target genes.
The sequences of individual TFBSs bound by a TF typically differs between
genomic loci and depending on the sequence, TFs can have higher or lower
affinities for individual binding sites. In vitro, systematic evolution of ligands by
exponential enrichment (SELEX) can be used to identify DNA sequences with
the highest binding affinity for a specific TF. SELEX starts with a large initial
library of random DNA oligonucleotides. From this library, high affinity binding
sites are enriched by repeated cycles of TF binding followed by isolation and PCR
amplification of bound sequences. The resulting pool of enriched DNA sequences
can then be sequenced to identify sequences bound by the TF of interest [46] and
to calculate relative TF affinities from the level of sequence enrichment [47]. In
vitro approaches, such as SELEX, showed that the intrinsic DNA-binding affinity
for a TF is in part determined by the base readout of the TF binding sequence
as represented by its consensus motif. However, the base readout is not the only
12 V. Thormann et al.

variable that contributes to the overall binding preference of a given TF. Evidence
from structural biology showed that TF binding affinities are also influenced by the
sequence-specific higher order conformation of DNA, resulting in specific bending
of the DNA structure and altered protein-DNA interactions [48]. The consensus
recognition motif derived from SELEX experiments captures which sequences are
bound at high affinity by the TF investigated. In vivo, however, high affinity binding
sites are not necessarily responsible for the biological consequences of TF signaling.
In fact, the biological significance of low-affinity binding sites was confirmed for
several TFs [49, 50]. For instance, it was shown that low-affinity binding sites of a
Hox TF are responsible for the regulation of target genes in vivo. In addition, these
low-affinity binding sites safeguard that only specific members of the HOX family
of TFs can bind and activate transcription from these binding sites. Thus, low-
affinity binding sites provide specificity among paralogous Hox TFs that was lost
when these binding sites were changed to high affinity binding sites [50]. In support
of the importance of low-affinity binding sites in gene regulation, computational
modeling of enhancer evolution predicted that regulation by multiple low-affinity
binding sites might be favored by evolutionary selection. A possible reason for
this could be that multiple low-affinity binding sites offer more possibilities for the
regulation of gene expression by changing multiple weak sites rather than one high
affinity TFBS [51]. Furthermore, the usage of multiple low-affinity binding sites
was suggested to enable efficient fine-tuning of gene expression in response to the
integration of several signaling inputs [52]. Finally, enhancers containing multiple
low-affinity binding sites for the same TF could maintain genetic redundancy and
confer regulatory robustness [50].
If affinity is a major driver of transcriptional output levels, these levels can be
calculated based on TFBS affinity [53, 54]. However, this occupancy hypothesis
has recently been challenged by several studies showing that high affinity binding
sites are not necessarily those with the highest activity [50, 55–57]. For example,
the affinity of GR for different GR binding site variants determined in vitro does not
correlate with in vivo transcriptional output as determined by reporter gene assays
[56]. An alternative explanation for the binding site-specific activities could be that
sequence variants induce distinct subtle structural changes in associated TFs which
in turn influence their activity towards target genes [56, 58]. Although studying the
role of the TFBS sequence on transcriptional output in isolation, where all other
variables are kept the same, simplifies interpretation of the results, in reality, TFBSs
are not an isolated linear stretch of DNA, but are embedded in a binding-site-specific
context. Consequently, in vivo, additional factors contribute to the overall binding
affinity and activity of a TFBS. For instance, the conformation of DNA is not only
influenced by the core TFBS sequence but also by nucleotides flanking these sites
[59]. Further, interactions between TFs binding at regions with multiple TFBSs
modulate their interaction with the genome by direct physical interactions [60].
These interactions between TFs bound at regulatory regions can either be additive,
synergistic, or antagonistic which can influence the level of transcriptional output.
To complicate things even further, depending on the composition of the proteins
binding at a single TFBS, GR can either act synergistically or antagonistically with
these proteins [61].
Transcriptional Regulation: When 1 C 1 ¤ 2 13

Together, the multitude of mechanisms and signaling inputs that influence the
expression level of genes provides the cell with a variety of mechanisms to fine-tune
the expression of genes within individual cells or tissues. The effects of individual
signaling inputs on gene expression may be context-specific and consequently,
predicting expression levels from a limited number of features, for example, the
affinity of a TF for its TFBS, is unlikely to achieve great levels of accuracy.

6 Conclusions and Future Directions

Efforts to predict TF binding based on sequence and to link TF binding to the

expression of genes have failed to accurately describe the in vivo situation. This
could be due to a lack of fundamental knowledge about the processes that specify
where TFs bind and what determines if these binding events are productive in terms
of resulting in gene expression changes. Recent advances have shown that adding
additional knowledge can greatly advance our understanding. For example, adding
information regarding chromatin accessibility helps to explain why TFs bind to only
a subset of potential TFBSs, namely those that are accessible [62]. Similarly, adding
knowledge regarding the three-dimensional organization of the genome improves
the correlation between TF binding and gene regulation [43], which might further
improve with increased resolution of the assays used to catalog the 3D genome
organization. Collectively, 3C-based approaches have yielded important insights
into the organization of enhancer–promoter interactions and for the prediction of
candidate TFBS target genes. However, only few loci have been comprehensively
investigated with respect to the relative contribution of individual enhancers to the
overall expression of its target gene.
High-throughput functional testing of enhancer sequences is further facilitated
by the development of several massive parallel enhancer analysis methods. For
example, the STARR-seq method (Self Transcribing Regulatory Regions) [63] in
which active enhancers drive their own expression. Next generation sequencing
can subsequently reveal the sequence identity of active enhancers and quantitative
information about their activity. This method can also be used to study how
the combinatorial action of several TFBSs influences transcriptional output or
the interplay between different core promoters and enhancers [64, 65]. Gener-
ally though, reporter gene assays differ in several fundamental ways from gene
regulation in the endogenous genomic setting (Fig. 3c) and thus have a limited
ability to uncover how gene regulation is orchestrated in vivo. This problem might
be circumvented using genome-editing tools to study the role of enhancers in
their endogenous genomic context. For example, the CRISPR/Cas9 system can
be used for targeted disruption of enhancers to study their role in the regulation
of genes [66]. This can help uncover if the effects of individual enhancers are
additive, or if they act in a mutually redundant fashion. Further, by studying large
numbers of enhancers, general principles that distinguish productive from non-
productive binding events might become clear. Another interesting application of the
CRISPR/Cas9 methodology is that usage of enzymatically inactive Cas9 enzymes
14 V. Thormann et al.

fused to activators or inhibitors enables in vivo manipulation of enhancer activity

[67]. Moreover, inactive Cas9 enzymes fused to chromatin modifying enzymes can
be used to study the role of specific chromatin marks at individual loci [68].
Together, the combination of in vivo, in vitro, and in silico methods may
ultimately provide the variables that are missing in our current equations and explain
why currently it seems as if 1 C 1 ¤ 2 when we try to quantitatively and qualitatively
model gene regulation. A greater understanding of the gene regulatory landscape
could also be of therapeutic relevance as increasing evidence suggests that sequence
variations in non-coding regions are a cause for several diseases including cancer,
developmental, metabolic, immune, and neuropsychiatric disorders [69, 70].

References

1. Chiang, C., et al.: Manifestation of the limb prepattern: limb development in the absence of
sonic hedgehog function. Dev. Biol. 236(2), 421–435 (2001)
2. Struhl, G.: A homoeotic mutation transforming leg to antenna in Drosophila. Nature 292(5824),
635–638 (1981)
3. Donehower, L.A., et al.: Mice deficient for p53 are developmentally normal but susceptible to
spontaneous tumours. Nature 356(6366), 215–221 (1992)
4. Consortium, E.P.: An integrated encyclopedia of DNA elements in the human genome. Nature
489(7414), 57–74 (2012)
5. Bulger, M., Groudine, M.: Functional and mechanistic diversity of distal transcription
enhancers. Cell 144(3), 327–339 (2011)
6. de Laat, W., Duboule, D.: Topology of mammalian developmental enhancers and their
regulatory landscapes. Nature 502(7472), 499–506 (2013)
7. Calo, E., Wysocka, J.: Modification of enhancer chromatin: what, how, and why? Mol. Cell
49(5), 825–837 (2013)
8. Meijsing, S.H.: Mechanisms of glucocorticoid-regulated gene transcription. Adv. Exp. Med.
Biol. 872, 59–81 (2015)
9. Zhang, Z., et al.: Evolutionary optimization of transcription factor binding motif detection.
Adv. Exp. Med. Biol. 827, 261–274 (2015)
10. Zhang, C., et al.: A clustering property of highly-degenerate transcription factor binding sites
in the mammalian genome. Nucleic Acids Res. 34(8), 2238–2246 (2006)
11. Wu, J., Bresnick, E.H.: Glucocorticoid and growth factor synergism requirement for Notch4
chromatin domain activation. Mol. Cell Biol. 27(6), 2411–2422 (2007)
12. Rao, N.A., et al.: Coactivation of GR and NFKB alters the repertoire of their binding sites and
target genes. Genome Res. 21(9), 1404–1416 (2011)
13. Biddie, S.C., et al.: Transcription factor AP1 potentiates chromatin accessibility and glucocor-
ticoid receptor binding. Mol. Cell 43(1), 145–155 (2011)
14. West, J.A., et al.: Nucleosomal occupancy changes locally over key regulatory regions during
cell differentiation and reprogramming. Nat. Commun. 5, 4719 (2014)
15. He, H.H., et al.: Differential DNase I hypersensitivity reveals factor-dependent chromatin
dynamics. Genome Res. 22(6), 1015–1025 (2012)
16. Gertz, J., et al.: Distinct properties of cell-type-specific and shared transcription factor binding
sites. Mol. Cell 52(1), 25–36 (2013)
17. Morikawa, M., et al.: ChIP-seq reveals cell type-specific binding patterns of BMP-specific
Smads and a novel binding motif. Nucleic Acids Res. 39(20), 8712–8727 (2011)
18. Kvon, E.Z., et al.: Genome-scale functional characterization of Drosophila developmental
enhancers in vivo. Nature 512(7512), 91–95 (2014)
Transcriptional Regulation: When 1 C 1 ¤ 2 15

19. Weirauch, M.T., et al.: Evaluation of methods for modeling transcription factor sequence
specificity. Nat. Biotechnol. 31(2), 126–134 (2013)
20. Cusanovich, D.A., et al.: The functional consequences of variation in transcription factor
binding. PLoS Genet. 10(3), e1004226 (2014)
21. Gitter, A., et al.: Backup in gene regulatory networks explains differences between binding and
knockout results. Mol. Syst. Biol. 5, 276 (2009)
22. Creyghton, M.P., et al.: Histone H3K27ac separates active from poised enhancers and predicts
developmental state. Proc. Natl. Acad. Sci. U. S. A. 107(50), 21931–21936 (2010)
23. Heintzman, N.D., et al.: Distinct and predictive chromatin signatures of transcriptional
promoters and enhancers in the human genome. Nat. Genet. 39(3), 311–318 (2007)
24. Hardison, R.C., Taylor, J.: Genomic approaches towards finding cis-regulatory modules in
animals. Nat. Rev. Genet. 13(7), 469–483 (2012)
25. Kwasnieski, J.C., et al.: High-throughput functional testing of ENCODE segmentation predic-
tions. Genome Res. 24(10), 1595–1602 (2014)
26. Frankel, N., et al.: Phenotypic robustness conferred by apparently redundant transcriptional
enhancers. Nature 466(7305), 490–493 (2010)
27. Spivakov, M.: Spurious transcription factor binding: non-functional or genetically redundant?
Bioessays 36(8), 798–806 (2014)
28. So, A.Y., et al.: Determinants of cell- and gene-specific transcriptional regulation by the
glucocorticoid receptor. PLoS Genet. 3(6), e94 (2007)
29. Amano, T., et al.: Chromosomal dynamics at the Shh locus: limb bud-specific differential
regulation of competence and active transcription. Dev. Cell 16(1), 47–57 (2009)
30. Levings, P.P., Bungert, J.: The human beta-globin locus control region. Eur. J. Biochem. 269(6),
1589–1599 (2002)
31. Hilton, I.B., et al.: Epigenome editing by a CRISPR-Cas9-based acetyltransferase activates
genes from promoters and enhancers. Nat. Biotechnol. 33(5), 510–517 (2015)
32. Tolhuis, B., et al.: Looping and interaction between hypersensitive sites in the active beta-
globin locus. Mol. Cell 10(6), 1453–1465 (2002)
33. Dekker, J., et al.: Capturing chromosome conformation. Science 295(5558), 1306–1311 (2002)
34. Dekker, J., Marti-Renom, M.A., Mirny, L.A.: Exploring the three-dimensional organization of
genomes: interpreting chromatin interaction data. Nat. Rev. Genet. 14(6), 390–403 (2013)
35. Dixon, J.R., et al.: Topological domains in mammalian genomes identified by analysis of
chromatin interactions. Nature 485(7398), 376–380 (2012)
36. Zuin, J., et al.: Cohesin and CTCF differentially affect chromatin architecture and gene
expression in human cells. Proc. Natl. Acad. Sci. 111(3), 996–1001 (2014)
37. Nagano, T., et al.: Single-cell Hi-C reveals cell-to-cell variability in chromosome structure.
Nature 502(7469), 59–64 (2013)
38. Rao, S.S., et al.: A 3D map of the human genome at kilobase resolution reveals principles of
chromatin looping. Cell 159(7), 1665–1680 (2014)
39. Fang, F., et al.: Coactivators p300 and CBP maintain the identity of mouse embryonic stem
cells by mediating long-range chromatin structure. Stem Cells 32(7), 1805–1816 (2014)
40. Drissen, R., et al.: The active spatial organization of the beta-globin locus requires the
transcription factor EKLF. Genes Dev. 18(20), 2485–2490 (2004)
41. Bouwman, B.A., de Laat, W.: Getting the genome in shape: the formation of loops, domains
and compartments. Genome Biol. 16, 154 (2015)
42. Vakoc, C.R., et al.: Proximity among distant regulatory elements at the beta-globin locus
requires GATA-1 and FOG-1. Mol. Cell 17(3), 453–462 (2005)
43. Jin, F., et al.: A high-resolution map of the three-dimensional chromatin interactome in human
cells. Nature 503(7475), 290–294 (2013)
44. Kilpinen, H., et al.: Coordinated effects of sequence variation on DNA binding, chromatin
structure, and transcription. Science 342(6159), 744–747 (2013)
45. Corradin, O., et al.: Combinatorial effects of multiple enhancer variants in linkage disequilib-
rium dictate levels of gene expression to confer susceptibility to common traits. Genome Res.
24(1), 1–13 (2014)
16 V. Thormann et al.

46. Wang, J., et al.: In vitro DNA-binding profile of transcription factors: methods and new insights.
J. Endocrinol. 210(1), 15–27 (2011)
47. Slattery, M., et al.: Cofactor binding evokes latent differences in DNA binding specificity
between Hox proteins. Cell 147(6), 1270–1282 (2011)
48. Stella, S., Cascio, D., Johnson, R.C.: The shape of the DNA minor groove directs binding by
the DNA-bending protein Fis. Genes Dev. 24(8), 814–826 (2010)
49. Ramos, A.I., Barolo, S.: Low-affinity transcription factor binding sites shape morphogen
responses and enhancer evolution. Philos. Trans. R. Soc. Lond. B Biol. Sci. 368(1632),
20130018 (2013)
50. Crocker, J., et al.: Low affinity binding site clusters confer hox specificity and regulatory
robustness. Cell 160(1-2), 191–203 (2015)
51. He, X., Duque, T.S., Sinha, S.: Evolutionary origins of transcription factor binding site clusters.
Mol. Biol. Evol. 29(3), 1059–1070 (2012)
52. Gao, R., Stock, A.M.: Temporal hierarchy of gene expression mediated by transcription factor
binding affinity and activation dynamics. mBio 6(3), e00686-15 (2015)
53. Bain, D.L., et al.: Glucocorticoid receptor-DNA interactions: binding energetics are the
primary determinant of sequence-specific transcriptional activity. J. Mol. Biol. 422(1), 18–32
(2012)
54. Segal, E., et al.: Predicting expression patterns from regulatory sequence in Drosophila
segmentation. Nature 451(7178), 535–540 (2008)
55. Garcia, H.G., et al.: Operator sequence alters gene expression independently of transcription
factor occupancy in bacteria. Cell Rep. 2(1), 150–161 (2012)
56. Meijsing, S.H., et al.: DNA binding site sequence directs glucocorticoid receptor structure and
activity. Science 324(5925), 407–410 (2009)
57. Hammar, P., et al.: Direct measurement of transcription factor dissociation excludes a simple
operator occupancy model for gene regulation. Nat. Genet. 46(4), 405–408 (2014)
58. Zhang, J., et al.: DNA binding alters coactivator interaction surfaces of the intact VDR-RXR
complex. Nat. Struct. Mol. Biol. 18(5), 556–563 (2011)
59. Rohs, R., et al.: Nuance in the double-helix and its role in protein-DNA recognition. Curr.
Opin. Struct. Biol. 19(2), 171–177 (2009)
60. Meyer, M.B., Benkusky, N.A., Pike, J.W.: Selective distal enhancer control of the Mmp13 gene
identified through clustered regularly interspaced short palindromic repeat (CRISPR) genomic
deletions. J. Biol. Chem. 290(17), 11093–11107 (2015)
61. Diamond, M.I., et al.: Transcription factor interactions: selectors of positive or negative
regulation from a single DNA element. Science 249(4974), 1266–1272 (1990)
62. John, S., et al.: Chromatin accessibility pre-determines glucocorticoid receptor binding
patterns. Nat. Genet. 43(3), 264–268 (2011)
63. Arnold, C.D., et al.: Genome-wide quantitative enhancer activity maps identified by STARR-
seq. Science 339(6123), 1074–1077 (2013)
64. Zabidi, M.A., et al.: Enhancer-core-promoter specificity separates developmental and house-
keeping gene regulation. Nature 518(7540), 556–559 (2015)
65. Dupin, C., et al.: Treatment of head and neck paragangliomas with external beam radiation
therapy. Int. J. Radiat. Oncol. Biol. Phys. 89(2), 353–359 (2014)
66. Korkmaz, G., et al.: Functional genetic screens for enhancer elements in the human genome
using CRISPR-Cas9. Nat. Biotechnol. 34(2), 192–198 (2016)
67. Maeder, M.L., et al.: CRISPR RNA-guided activation of endogenous human genes. Nat
Methods 10(10), 977–979 (2013)
68. Mendenhall, E.M., et al.: Locus-specific editing of histone modifications at endogenous
enhancers. Nat. Biotechnol. 31(12), 1133–1136 (2013)
69. Lupianez, D.G., et al.: Disruptions of topological chromatin domains cause pathogenic
rewiring of gene-enhancer interactions. Cell 161(5), 1012–1025 (2015)
70. Zhang, X., et al.: Identification of focally amplified lineage-specific super-enhancers in human
epithelial cancers. Nat. Genet. 48(2), 176–182 (2016)
Differential Network Analysis and Graph
Classification: A Glocal Approach

Giuseppe Jurman, Michele Filosi, Samantha Riccadonna,

Roberto Visintainer, and Cesare Furlanello

Abstract Based on the glocal HIM metric and its induced graph kernel, we propose
a novel solution in differential network analysis that integrates network comparison
and classification tasks. The HIM distance is defined as the one-parameter family
of product metrics linearly combining the normalised Hamming distance H and
the normalised Ipsen–Mikhailov spectral distance IM. The combination of the
two components within a single metric allows overcoming their drawbacks and
obtaining a measure that is simultaneously global and local. Furthermore, plugging
the HIM kernel into a Support Vector Machine gives us a classification algorithm
based on the HIM distance. First, we outline the theory underlying the metric
construction. We introduce two diverse applications of the HIM distance and the
HIM kernel to biological datasets. This versatility supports the adoption of the HIM
family as a general tool for information extraction, quantifying difference among
diverse instances of a complex system. An Open Source implementation of the HIM
metrics is provided by the R package nettools and in its web interface ReNette.

Keywords Differential network • Network distance

1 Introduction

The paradigm shift towards complex systems science [3], stimulated by its recent
theoretical and computational advances [4, 15], has paved the way for a parallel leap
in computational biology by moving the focus from the differential gene expression
analysis to differential network analysis (NetDA) [16, 25]. Due to the heterogeneity
in the NetDA process and potential ill-posedness of some of the involved functional
operations [1, 5, 38], a number of alternative approaches have appeared in the
literature, with different strategies and aims [6, 7, 10, 16, 22, 23, 25, 41, 45, 50, 51].

G. Jurman () • M. Filosi • R. Visintainer • C. Furlanello

Fondazione Bruno Kessler, via Sommarive 18 Povo, I-38122 Trento, Italy
e-mail: [email protected]; [email protected]; [email protected]; [email protected]
S. Riccadonna
Centro Ricerca e Innovazione, Fondazione Edmund Mach, I-38010 San Michele all’Adige, Italy
e-mail: [email protected]

© Springer International Publishing Switzerland 2016 17

A. Rogato et al. (eds.), Dynamics of Mathematical Models in Biology,
DOI 10.1007/978-3-319-45723-9_2
18 G. Jurman et al.

Fig. 1 A pair of similar subgraphs from a comparison of D. melanogaster and S. cerevisiae

protein–protein interaction network as shown in [12]. Blue links are present only in the S. cerevisiae
subnet

For example, NetDA can be used to compare networks corresponding to different

organisms, phenotypes or conditions. The subgraph of the protein–protein interac-
tion network shown in Fig. 1 (from [12]) is the same in terms of shared nodes for
the fruit fly and the budding yeast. A group of links is shared by both instances
of the subgraph, but the budding yeast network includes nine additional edges.
Clearly, when graphs to compare have a more complex structure, more sophisticated
quantitative indicators are needed also to ensure a reproducible analysis [26]. In
general, the two key applications of NetDA are network comparison and network
classification. Both can be framed in terms of similarity between graphs, which
is best dealt with by defining a distance. However, non-metric alternatives can be
used [17, 49], and even combinations of metric and statistical approaches [35, 43].
Here we propose to use the Hamming–Ipsen–Mikhailov (HIM) distance [31, 32]
first as the underlying metric for the NetDA framework, and also to induce a kernel
for classification purposes. The HIM metric linearly combines two distances, the
Hamming [18, 24, 28, 40, 48] and the Ipsen–Mikhailov [27]; the first is an edit dis-
tance, while the latter is a spectral measure. These are the two most relevant families
of graph distances: the edit distances are based on functions of insertion and deletion
of matching links between the compared graphs, while the spectral measures are
functions of the eigenvalues of one of the graph connectivity matrices. The Ipsen–
Mikhailov distance was chosen after a comparative review [30], while Hamming
was selected as the simplest member of the edit family. As a characterising feature,
HIM is a glocal distance that overcomes the drawbacks of local (edit) and global
(spectral) metrics when separately considered. In fact, local functions disregard the
overall network structure, while spectral measures cannot distinguish isospectral
graphs. Superiority of using the HIM distance over H or IM separately in practical
applications is shown in the literature: NetDA based on the HIM distance has
been used in metagenomics [52], MEG neuroimaging [21], liver high-throughput
Differential Network Analysis and Graph Classification: A Glocal Approach 19

oncogenomics [20] and oncoimmunology [39]. In all cases, the findings derived
by NetDA have been validated by matching the obtained quantitative outcomes
with the qualitative biological knowledge reported in the literature. Moreover,
the same method has found applicability also out of computational biology, e.g.,
socioeconomics [32] or even in multiplex network theory [29]. Here we present,
after a brief summary of the main definitions, two novel application examples, in
neurogenomics and in developmental functional genomics. In the first example, we
highlight and quantify weighted network dissimilarities among gene expression of
brain tissues with different phenotypes (location, sex and health status), while in the
latter we describe the trajectory of the binary developmental gene network in fruit
fly across its different life stages.
Finally, we describe the CRAN R package nettools and the web framework
ReNette [19], which are available to implement NetDA projects.

2 The HIM Distance and Kernel

We recap hereafter the main definitions and results about the HIM metric and kernel.
The synthesis is based on the notations of Table 1: a fully detailed description,
including mathematical proofs, goes beyond the scope of the present chapter, and it
is included in [31]. The (normalised) Hamming distance [18, 24, 28, 40, 48] is the
(local) simplest edit metric, counting the presence/absence of matching links:

Hamming.N1 ; N2 / 1 X .1/ .2/

H.N1 ; N2 / D D jAij  Aij j :
Hamming.EN ; FN / N.N  1/
1i6DjN

By definition, H ranges between 0 and 1, where

H D 0 for A.1/ D A.2/ and H D 1 for A.1/ C A.2/ D 1N  IN :

Note that, for H, all links are equivalent regardless of their position within the
network: for instance, in Fig. 2, both networks B1 and B2 differ from A for just
one link, and thus H.A; B1 / D H.A; B2 /, although B1 is connected as A while B2
is not.The Ipsen–Mikhailov distance [27] is the (global) L2 integrated difference of
the Laplacian spectral densities:
sZ
1
IM.N1 ; N2 / D ŒN1 .!;  /  N2 .!;  /2 d! :
0

The definition of IM follows the dynamical interpretation of an N nodes network

as an N molecules system connected by identical elastic strings, where the pattern
of connections is defined by the adjacency matrix A of the corresponding network.
The dynamics of the system is described by the set of N differential equations
20 G. Jurman et al.

Table 1 Notation and list of symbols

N1 ; N2 Simple networks on N nodes fzi gniD1
.1/ .2/ .1/ .2/
A ;A Corresponding adjacency matrices, with aij ; aij 2 F
F Field F2 D f0; 1g (unweighted case) or Œ0; 1  R (weighted case)
 1 0  0 
IN Identity matrix 0 1  0
0 0  1
1 1  1

1 1  1
1N Unitary matrix 
1 1  1
0 0  0

0 0  0
0N Zero matrix 
0 0  0
EN Empty network (adjacency matrix 0N )
FN Clique (adjacency matrix 1N  IN )
PN
@g Degree of node zg , @g D @.zg / D jD1 Agj
!
@1 0  0
0 @2  0
D Degree matrix 
0 0  @n

L Laplacian matrix D  A, positive and semidefinite [11]

SpecL Laplacian spectrum f0; 1 ; 2 ; : : : ; N g, with 1  : : :  N eigenvalues
p PN
!i Vibrational frequencies i , solution of the ODE system xR i C jD1 Aij
.xi xj / D 0 [27]
PN1
 Spectral density as sum of Lorentz distributions .!; / D K iD1

.!!i /2 C 2
Z 1
K Normalisation constant defined by .!; /d! D 1
0
 Half-width at half-maximum
Z 1
 2
 Unique solution of EN .!; /  FN .!; / d! D 1 [31]
0

B1 A B2

Fig. 2 Link equivalence for Hamming metric: H.A; B1 / D H.A; B2 / although B1 is connected
while B2 consists of two connected components

X
N
xR i C Aij .xi  xj / D 0 for i D 0;    ; N  1 :
jD1

The vibrational frequencies !i for this network model are given by the square root of
the eigenvalues of the Laplacian matrix of the network: i D !i2 , with 0 D !0 D 0.
The spectral density for a graph as the sum of Lorentz distributions is defined as
Differential Network Analysis and Graph Classification: A Glocal Approach 21

X
N1

.!;  / D K ;
iD1
.!  !i /2 C  2

where  isZ the common width and K is the normalisation constant defined by the
1
condition .!;  /d! D 1, and thus
0

1
KD :
XZ 1
N1
d!

iD1 0 .!  !i /2 C  2

The scale parameter  specifies the half-width at half-maximum, which is equal to

half the interquartile range. Then the spectral distance  between two graphs N1
and N2 on N nodes with densities N1 .!;  / and N2 .!;  / can then be defined as
sZ
1
 .N1 ; N2 / D ŒN1 .!;  /  N2 .!;  /2 d! :
0

The highest value of  is reached, for each N, when evaluating the distance between
EN and FN . Denote then by  the unique solution of

 .EN ; FN / D 1 :

Thus, by definition, IM too ranges between 0 and 1, where

IM D 0 for spec.L.1/ / D spec.L.2/ / and IM D 1 for fN1 ; N2 g D fEN ; FN g:

In fact, being a spectral measure, IM cannot distinguish isospectral (non-

isomorphic) networks.

To overcome the drawbacks of both H and IM, we define their normalised

cartesian product, the Hamming–Ipsen–Mikhailov distance:
q
1
HIM .N1 ; N2 / D p H2 .N1 ; N2 / C   IM2 .N1 ; N2 /;
1C

for  2 Œ0; C1/.

22 G. Jurman et al.

When  is not close to the bounds f0; C1g (and one of the factors becomes
dominant), the impact of  is minimal, and in general more relevant when HIM is
used as a kernel [21]. Hereafter  D 1 will be assumed, and the subscript  omitted.
Again, HIM is bounded between 0 and 1, with

HIM D 0 for A.1/ D A.2/ and HIM D 1 for fN1 ; N2 g D fEN ; FN g:

The HIM distance can be naturally extended to directed networks, by transforming

it into an undirected bipartite graph through the procedure shown in [36].

The HIM distance naturally induces a kernel via Gaussian (Radial Basis
Function) map [9, 13] to be used standalone or in a Multi-Kernel Learning
framework to increase performance and enhance interpretability [33]:
2
K.N1 ; N2 / D eHIM .N1 ;N2 / ;

for a positive real number  .

Although the HIM kernel is not positively defined in general for all  2 RC 0 ,
by results in [44] it can be used in Support Vector Machines or other algorithms
whenever K is positively defined for the given training data, which is the case for
all the examples shown in what follows. In general, the range of suitable values
for  can be computed by imposing positiveness to all eigenvalues of the matrix
2
eHIM .xi ;xj / for xi ; xj in the training set.

3 Application to -omic Studies

3.1 The UKBEC Dataset

The United Kingdom Brain Expression Consortium (UKBEC) hybridised on a

Affymetrix Human Exon 1.0 ST Array (transcript version) 1213 human brain
samples from ten diverse regions. Samples originated from 134 neurologically
and neuropathologically normal individuals and were used in three studies aimed
at better understanding gene expression differences [42, 46, 47]. Data details
about sample stratification according to sex and tissue location are listed in
Table 2(a). Here, this dataset1 is used to build the absolute Pearson coexpression
networks corresponding to different region/gender/age group defined on the 50

1
Available as GEO46706 at http://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE46706.
Differential Network Analysis and Graph Classification: A Glocal Approach 23

Table 2 Sample size of the UKBEC human brain dataset stratified by gender and tissue location
(a) and by gender and age group (b)
(a) (b)
Region Abbr. M F Region Abbr. M F Age M F Age M F
Cerebellar CB 95 35 Frontal cortex FCX 93 34 < 32 86 39 58–62 117 20
cortex
Hippo campus HC 92 30 Medulla Med 88 31 32–44 130 19 62–68 72 29
Occipital cortex OCC 94 35 Putamen PUT 96 33 44–48 74 24 68–76 82 39
Substantia nigra SN 73 28 Temporal cortex TCX 86 33 48–53 109 27 76–83 66 56
Thalamus Thal 91 33 White matter WM 97 34 53–58 101 20  83 68 53
Region: the tissue location. Abbr.: abbreviation as in Fig. 3, M: number of samples from male
individuals, F: number of samples from female individuals. a  b means a < x  b

genes belonging to the BRAIN_DEVELOPMENT (GO:0007420, GSEA M7203)

pathway,2 corresponding to 1012 probes on the Affymetrix Human Exon 1.0 ST
Array platform.3 In detail, each network has 1012 nodes (one for each probe) and the
weight of a link between two nodes is the absolute value of the Pearson correlation
between the vectors collecting the expression levels of the corresponding probes for
the samples belonging to the considered region/gender/age group.
First, we consider planar projections of all the mutual HIM distances between
networks with shared nodes based on the metric multidimensional scaling
(mMDS) [14, 37] in Fig. 3. The mMDS plot shows the mutual HIM distances
with networks stratified for both sex and tissue location. Citing the authors, the
study in [46] ‘provides unequivocal evidence that sex-biased gene expression in the
adult human brain is widespread in terms of both the number of genes and range
of brain regions involved’. In our analysis, the result is numerically confirmed by
the major effect emerging at the gene coexpression level (Fig. 3): male and female
networks can be linearly separated in the mMDS space, with large HIM distances
between both inter- and intragender tissue locations. In particular, intragender HIM
distances among different tissue regions are larger for the female samples (range
[0.112,0.232], median 0.146) than for the male (range [0.077,0.200], median 0.118),
with statistical significance (t-test p-value 1:9  104 ).
In Fig. 4, we show instead the mMDS projections for the mutual HIM distances
of the coexpression networks built separately for male and female subjects, parti-
tioned in ten age groups: the sample size for each network is listed in Table 2(b).
While the plot for the females does not show any global pattern, for males the first
five groups (age < 58 y) have small mutual HIM distances and they result clustered
together. On the other hand, the five older male groups are both mutually distant and
distant from the younger subjects cluster, too. In this dataset, the small sample size
in the female subgroup may be a relevant source of noise for some of the age groups,

2
Available at http://software.broadinstitute.org/gsea/msigdb/cards/BRAIN_DEVELOPMENT.
3
The platform has no probes for the 51st gene of the pathway, VCX3A.
24 G. Jurman et al.

Med M
Thal M
CB M
0.04

WM M

PUT
HC M
0.02

TCX M
SN M

OCC M
FCX M
0.00

WM F OCC F
−0.02

CB F
FCX F PUT F
Thal F
TCX F
−0.04

HC F
SN F
Med F

−0.10 −0.05 0.00 0.05

CB : Cerebellar Cortex Med : Medulla SN : Substantia Nigra WM : White Matter

FCX : Frontal Cortex OCC : Cccipital Cortex TCX : Temporal Cortex F: Female
HC : Hippocampus PUT : Putamen Thal : Thalamus M: Male

Fig. 3 Metric multidimensional scaling projection on two dimensions of all 190 mutual HIM
distances between gene coexpression brain development networks stratified by gender and tissue
locations

e.g., the 32–44. Our results are consistent with findings obtained with different data
and methodology by Berchtold and colleagues in [8], suggesting the existence of a
global pattern of gene expression change associated with brain aging, more evident
from the sixth decade onward, with different evolutions between males and females,
with larger variations in male subjects. Biologically, this is due to a wider global
decrease in males in the catabolic and anabolic capacity with aging, mainly in genes
linked to energy production and protein synthesis and transport [8].
Differential Network Analysis and Graph Classification: A Glocal Approach 25

(a) (b)

0.08
< 32 76~83 32~44
0.06

32~44
44~48

0.06
48~53
0.04

0.04
0.02

58~62

0.02
48~53 44~48 >=83 62~68 48~53 68~76
0.00

−0.02 0.00
58~62
< 32 32~44 76~83 >=83 < 32
53~58
−0.02

62~68

44~48
−0.04

68~76 53~58
−0.06
−0.04 −0.02 0.00 0.02 0.04 0.06 −0.05 0.00 0.05

Fig. 4 Metric multidimensional scaling projection on two dimensions of all 45 mutual HIM
distances between gene coexpression brain development networks stratified by age groups,
separately for the male (a) and female (b) subjects. a  b means a < x  b

3.2 The D. melanogaster Development Dataset

In [34], Kolar and colleagues applied the Keller algorithm to infer the gene regu-
latory networks of Drosophila melanogaster from a time series of gene expression
data measured during its full life cycle, originally published in [2]. They followed
the dynamics of 588 development genes along 66 time points spanning through
four different stages (Embryonic—time points 1–30, Larval—t.p. 31–40, Pupal—
t.p. 41–58, Adult—t.p. 59–66), constructing a time series of inferred networks Ni ,4
where a link between two nodes exists whenever the Keller algorithm detects a
mutual inference between the corresponding genes at the given time point: in Fig. 5a
we show four instances of the Ni networks, at different timing.
As a first step in the quantitative NetDA of this dataset, we measure the HIM
distance between each Ni and the initial network N1 : the resulting distance time
series is shown in Fig. 5b. The largest variations, both between consecutive terms
and with respect to the initial network N1 , occur in the Embryonal stage (E). In
particular, the HIM distance grows until time point 23; next networks get closer
again to N1 , showing that the interactions of the selected 588 genes in the adult
stage are more similar to the corresponding net of interaction in the Embryonal
stage, rather than in the other two stages, consistently with the findings reported
in the original reference [34]. Moreover, while the Hamming component ranges
between 0 and 0:0223, the Ipsen–Mikhailov distance has 0:0851 as its maximum,

4
Publicly available at http://cogito-b.ml.cmu.edu/keller/downloads.html.
26 G. Jurman et al.

(a)
t=1

(b)

0.08 E L P A

0.06
Distance

t=20
0.04

0.02

H
IM
0.00 HIM

1 10 20 30 40 50 60 66
Time step

t=35

(c)

t=66

Fig. 5 D. melanogaster development network dataset. (a) Keller interaction network Ni for the D.
melanogaster development genes at the time points i D 1; 20; 35; 66. (b) Evolution of H (cyan),
IM (magenta) and HIM (golden red) distances network time series across 66 time points in the four
stages Embryonic (E), Larval (L), Pupal (P) and Adult (A). (c) Metric multidimensional scaling
planar projection of the mutual HIM distances between the 66 networks Ni , coloured according to
the developmental stage Embryonic (blue), Larval (red), Pupal (green) and Adult (orange)
Differential Network Analysis and Graph Classification: A Glocal Approach 27

indicating an higher variability of the networks in terms of structure rather than

matching links: in this case, in fact, the HIM distance is driven by the evolution of
the IM component.
Then we computed all 2145 HIM distances HIM.Ni ; Nj /, and we projected them
on a 2D mMDS representation, shown in Fig. 5c. Interestingly, the networks for the
Embryonal stage split into two clusters (before and after time points 17), and the
Embryonal and Pupal stages are orthogonal in this representation.
Moreover, the Adult stage networks form a cluster well separated from the other
nets, with the Larval stage graphs mixing with the Pupal and late Embryonal stages.
Finally, a Support Vector Machine classifier with HIM kernel was developed with
the kernlab package in R, with a five-fold cross validation with  D 103 and
C D 1. The classifier reached accuracy 0.97 in discriminating Embryonic and Adult
networks from Larval and Pupal. Similarly, in the same setup, perfect separation is
reached between Embryonic and Adult stages for all values of  > 103 .

4 Conclusion

The interest of the HIM metric is its global/local approach: by combining edit
and spectral distance types, we overcome the drawbacks of the two distance
components. The two presented applications in functional high-throughput -omics
support the effectiveness of the approach. The strategy of a NetDA based on the HIM
distance offers a reproducible method: the metric gives a completely quantitative
assessment of the differences among networks (on shared nodes) as well as a scalar
product for kernel learning machines.
Operatively, we provide an Open Source implementation of the HIM distance
with the R package nettools available on CRAN and GitHub,5 and in the web
interface ReNette [19].6 In particular, ReNette includes a complete pipeline for
NetDA, integrating a comprehensive collection of tools for network inference, net-
work comparison and network stability analysis [20] (a methodology for assessing
the robustness of an inferred network w.r.t. data subsampling) through queue-based
submission system and asynchronous task management. The software is already
configured for usage on multicore workstations, on high performance computing
clusters and on a cloud-based cluster, to deal with the extraction of the Laplacian
spectrum, which represents the computational bottleneck of the algorithm.

5
https://github.com/MPBA/nettools.git.
6
http://renette.fbk.eu.
28 G. Jurman et al.

References

1. Angulo, M., Moreno, J., Barabási, A.L., Liu, Y.Y.: Fundamental limitations of network
reconstruction (2015). arXiv:1508.03559
2. Arbeitman, M., Furlong, E., Imam, F., Johnson, E., Null, B., Baker, B., Krasnow, M., Scott,
M., Davis, R., White, K.: Gene expression during the life cycle of Drosophila melanogaster.
Science 297(5590), 2270–2275. Erratum in Science 298(5596), 1172 (2002)
3. Barabási, A.L.: The network takeover. Nat. Phys. 8, 14–16 (2012)
4. Barabási, A.L.: Network science. Philos. Trans. R. Soc. A 371(1987), 20120375 (2013)
5. Baralla, A., Mentzen, W., de la Fuente, A.: Inferring gene networks: dream or nightmare? Ann.
N. Y. Acad. Sci. 1158, 246–256 (2009)
6. Barla, A., Jurman, G., Visintainer, R., Squillario, M., Filosi, M., Riccadonna, S., Furlanello,
C.: A machine learning pipeline for discriminant pathways identification. In: Biganzoli,
E., Vellido, A., Ambrogi, F., Tagliaferri, R. (eds.) Computational Intelligence Methods for
Bioinformatics and Biostatistics. Lecture Notes in Computer Science, vol. 7548, pp. 36–48.
Springer, Berlin (2012)
7. Barla, A., Jurman, G., Visintainer, R., Squillario, M., Filosi, M., Riccadonna, S., Furlanello,
C.: A Machine learning pipeline for discriminant pathways identification. In: Kasabov, N. (ed.)
Springer Handbook of Bio-/Neuroinformatics, Chap. 53, p. 1200. Springer, Berlin (2013)
8. Berchtold, N., Cribbs, D., Coleman, P., Rogers, J., Head, E., Kim, R., Beach, T., Miller, C.,
Troncoso, J., Trojanowski, J., Zielke, H., Cotman, C.: Gene expression changes in the course
of normal brain aging are sexually dimorphic. Proc. Natl. Acad. Sci. U. S. A. 105(40), 15605–
15610 (2008)
9. Bolla, M.: Spectral Clustering and Biclustering: Learning Large Graphs and Contingency
Tables. Wiley, New York (2013)
10. Chuang, H.Y., Lee, E., Liu, Y.T., Lee, D., Ideker, T.: Network-based classification of breast
cancer metastasis. Mol. Syst. Biol. 3, 140 (2007)
11. Chung, F.: Spectral Graph Theory. CBMS Regional Conference Series in Mathematics, vol. 92.
American Mathematical Society, Philadelphia (1997)
12. Cootes, A., Muggleton, S., Sternberg, M.: The identification of similarities between biological
networks: application to the metabolome and interactome. J. Mol. Biol. 369, 1126–1139 (2007)
13. Cortes, C., Haffner, P., Mohri, M.: Positive definite rational kernels. In: Learning Theory and
Kernel Machines. Proceedings of COLT 2003. Lecture Notes on Computer Science, vol. 2777,
pp. 41–56. Springer, Berlin (2003)
14. Cox, T., Cox, M.: Multidimensional Scaling. Chapman and Hall, Boca Raton (2001)
15. Csermely, P., Korcsmáros, T., Kiss, H., London, G., Nussinov, R.: Structure and dynamics of
biological networks: a novel paradigm of drug discovery. A comprehensive review. Pharmacol.
Ther. 138, 333–408 (2013)
16. de la Fuente, A.: From ‘differential expression’ to ‘differential networking’ - identification of
dysfunctional regulatory networks in diseases. Trends Genet. 26(7), 326–333 (2010)
17. Dehmer, M., Mowshowitz, A.: The discrimination power of structural superindices. PLoS ONE
8(7), e70551 (2013)
18. Dougherty, E.: Validation of gene regulatory networks: scientific and inferential. Brief.
Bioinform. 12(3), 245–252 (2010)
19. Filosi, M., Droghetti, S., Arbitrio, E., Visintainer, R., Riccadonna, S., Jurman, G., Furlanello,
C.: ReNette: a web-infrastructure for reproducible network analysis (2014). bioRxiv-doi:10.
1101/008433
20. Filosi, M., Visintainer, R., Riccadonna, S., Jurman, G., Furlanello, C.: Stability indicators in
network reconstruction. PLoS ONE 9(2), e89815 (2014)
21. Furlanello, T., Cristoforetti, M., Furlanello, C., Jurman, G.: Sparse predictive structure of
deconvolved functional brain networks. High-Dimensional Statistical Inference in the Brain,
NIPS 2013 Workshop (2013). arXiv:1310.6547[q-bio.NC]
Differential Network Analysis and Graph Classification: A Glocal Approach 29

22. Gill, R., Datta, S., Datta, S.: A statistical framework for differential network analysis from
microarray data. BMC Bioinf. 11(1), 1–10 (2010)
23. Ha, M., Baladandayuthapani, V., Do, K.A.: DINGO: differential network analysis in genomics.
Bioinformatics 31(21), 3413–3420 (2015)
24. Hamming, R.: Error detecting and error correcting codes. Bell Syst. Tech. J. 29(2), 147–160
(1950)
25. Ideker, T., Krogan, N.: Differential network biology. Mol. Syst. Biol. 8, 565 (2012)
26. Ioannidis, J., Allison, D., Ball, C., Coulibaly, I., Cui, X., Culhane, A.C., Falchi, M., Furlanello,
C., Game, L., Jurman, G., Mehta, T., Mangion, J., Nitzberg, M., Page, G., Petretto, E., van
Noort, V.: Repeatability of published microarray gene expression analyses. Nat. Genet. 41(2),
499–505 (2009)
27. Ipsen, M., Mikhailov, A.: Evolutionary reconstruction of networks. Phys. Rev. E 66, 046109
(2002). Erratum in Phys. Rev. E 67, 039901 (2003)
28. Iwayama, K., Hirata, Y., Takahashi, K., Watanabe, K., Aihara, K., Suzuki, H.: Characterizing
global evolutions of complex systems via intermediate network representations. Sci. Rep. 2,
423 (2012)
29. Jurman, G.: Metric projections for dynamic multiplex networks (2016). arXiv:1601.01940
30. Jurman, G., Visintainer, R., Furlanello, C.: An introduction to spectral distances in networks.
In: Apolloni, B., Bassis, S. (eds.) Proceedings of WIRN10, Frontiers in Artificial Intelligence
and Applications, vol. 226, pp. 227–234. IOS Press, Amsterdam (2011)
31. Jurman, G., Visintainer, R., Riccadonna, S., Filosi, M., Furlanello, C.: The HIM glocal metric
and kernel for network comparison and classification (2014). arXiv:1201.2931v3
32. Jurman, G., Visintainer, R., Filosi, M., Riccadonna, S., Furlanello, C.: The HIM glocal
metric and kernel for network comparison and classification. In: Proceedings IEEE DSAA’15,
vol. 36678, pp. 1–10. IEEE, New York (2015)
33. Kloft, M., Brefeld, U., Sonnenburg, S., Zien, A.: `p -norm multiple kernel learning. J. Mach.
Learn. Res. 12, 953–997 (2011)
34. Kolar, M., Song, L., Ahmed, A., Xing, E.: Estimating time-varying networks. Ann. Appl. Stat.
4(1), 94–123 (2010)
35. Koutra, D., Vogelstein, J., Faloutsos, C.: DELTACON: a principled massive-graph similarity
function. In: Proceedings of the 13th SIAM International Conference on Data Mining (SDM),
pp. 162–170. SIAM, New York (2013)
36. Liu, Y.Y., Slotine, J.J., Barabási, A.L.: Controllability of complex networks. Nature 473(7346),
167–173 (2011)
37. Mardia, K.: Some properties of classical multidimensional scaling. Commun. Stat. Theory
Meth. A7, 1233–1241 (1978)
38. Meyer, P., Alexopoulos, L., Bonk, T., Califano, A., Cho, C., de la Fuente, A., de Graaf, D.,
Hartemink, A., Hoeng, J., Ivanov, N., Koeppl, H., Linding, R., Marbach, D., Norel, R., Peitsch,
M., Rice, J., Royyuru, A., Schacherer, F., Sprengel, J., Stolle, K., Vitkup, D., Stolovitzky,
G.: Verification of systems biology research in the age of collaborative competition. Nat.
Biotechnol. 29(9), 811–815 (2011)
39. Mina, M., Boldrini, R., Citti, A., Romania, P., D’Alicandro, V., De ioris, M., Castellano,
A., Furlanello, C., Locatelli, F., Fruci, D.: Tumor-infiltrating T lymphocytes improve clinical
outcome of therapy-resistant neuroblastoma. Oncoimmunology 4(9), e1019981 (2015)
40. Morris, M., Handcock, M., Hunter, D.: Specification of exponential-family random graph
models: terms and computational aspects. J. Stat. Softw. 24(4), 1–24 (2008)
41. Pavlopoulos, G., Secrier, M., Moschopoulos, C., Soldatos, T., Kossida, S., Aerts, J., Schneider,
R., Bagos, P.: Using graph theory to analyze biological networks. BioData Min. 4(1), 10 (2011)
42. Ramasamyi, A., Trabzuni, D., Guelfi, S., Varghese, V., Smith, C., Walker, R., De, T.,
United Kingdom Brain Expression Consortium (UKBEC), North American Brain Expression
Consortium, Coin, L., de Silva, R., Cookson, M., Singleton, A., Hardy, J., Ryten, M., Weale,
M.: Genetic variability in the regulation of gene expression in ten regions of the human brain.
Nat. Neurosci. 17(10), 1418–1428 (2014)
30 G. Jurman et al.

43. Ruan, D., Young, A., Montana, G.: Differential analysis of biological networks. BMC Bioinf.
16, 327 (2015)
44. Schölkopf, B.: Support Vector Learning. Oldenbourg, Munchen (1997)
45. Sharan, R., Ideker, T.: Modeling cellular machinery through biological network comparison.
Nat. Biotechnol. 24(4), 427–433 (2006)
46. Trabzuni, D., Ramasamy, A., Imran, S., Walker, R., Smith, C., Weale, M., Hardy, J., Ryten, M.,
North American brain expression consortium. Widespread sex differences in gene expression
and splicing in the adult human brain. Nat. Commun. 4, 2771 (2013)
47. Trabzuni, D.: United Kingdom Brain Expression Consortium (UKBEC), Thomson, P.: Anal-
ysis of gene expression data using a linear mixed model/finite mixture model approach:
application to regional differences in the human brain. Bioinformatics 30(11), 1555–1561
(2014)
48. Tun, K., Dhar, P., Palumbo, M., Giuliani, A.: Metabolic pathways variability and
sequence/networks comparisons. BMC Bioinf. 7(1), 24 (2006)
49. Xiao, Y., Dong, H., Wu, W., Xiong, M., Wang, W., Shi, B.: Structure-based graph distance
measures of high degree of precision. Pattern Recogn. 41(12), 3547–3561 (2008)
50. Yang, B., Zhang, J., Yin, Y., Zhang, Y.: Network-based inference framework for identifying
cancer genes from gene expression data. BioMed. Res. Int. 2013, 12pp. (2013). Article ID
401649
51. Yoon, B.J., Qian, X., Sahraeian, S.: Comparative analysis of biological networks. IEEE Signal
Process. Mag. 29(1), 22–34 (2012)
52. Zandoná, A., Chierici, M., Jurman, G., Furlanello, C., Cucchiara, S., Del Chierico, F.,
Putignani, L.: A metagenomic pipeline integrating predictive profiling methods and complex
networks for the analysis of NGS microbiome data. NIPS Workshop - Machine Learning in
Computational Biology (2014)
Structural vs Practical Identifiability
of Nonlinear Differential Equation Models
in Systems Biology

Maria Pia Saccomani and Karl Thomaseth

Abstract This paper reappraises two different viewpoints adopted for testing
identifiability of nonlinear differential equation models. The aim is to take advantage
through their joint use of the complementary information provided. The common
objective is to assess whether model parameters can be estimated from specific
input/output (I/O) experiments. The structural identifiability analysis investigates
whether unknown model parameters can be identified uniquely, at all, with a
particular I/O configuration. This is investigated using differential algebra, e.g.,
as implemented in the software DAISY (Differential Algebra for Identifiability of
SYstems). In contrast, practical identifiability analysis is a data-based approach
to assess the precision of parameter estimates obtainable from experimental data.
It is based on simulated model outputs and their sensitivities with respect to
parameters. The relevant novelty of using both methodologies together is that
structural identifiability analysis allows a clearer understanding of the practical
identifiability results. This result is shown in the identifiability analysis of a much
quoted biological model describing the erythropoietin(Epo)-induced activation of
the JAK-STAT signaling pathway, which is known to play a role in the regulation
of cell proliferation, differentiation, chemotaxis, and apoptosis and is important for
hematopoiesis, and immune development. This study shows that some results on
practical identifiability tests can be proven in an analytical way by a differential
algebra test and that this test can provide additional information helpful for the
experiment design.

Keywords Biological systems • Identifiability • Parameter estimation

M.P. Saccomani ()

Department of Information Engineering, University of Padova, Via G Gradenigo 6a, 35131
Padova, Italy
e-mail: [email protected]
K. Thomaseth
Institute of Electronics, Computer and Telecommunication Engineering (IEIIT-CNR) c/o DEI,
Via G Gradenigo 6b, 35131 Padova, Italy
e-mail: [email protected]

© Springer International Publishing Switzerland 2016 31

A. Rogato et al. (eds.), Dynamics of Mathematical Models in Biology,
DOI 10.1007/978-3-319-45723-9_3
32 M.P. Saccomani and K. Thomaseth

1 Introduction

Mathematical modeling has become ubiquitous in quantitative molecular biology

and biotechnology with applications ranging from metabolic engineering to cancer
therapy. A large number of publications present mathematical models to investigate
complex, dynamic, nonlinear interaction mechanisms in cellular processes like
signal transduction pathways and metabolic networks. Typically these mechanisms
are modeled according to physicochemical laws, such as mass or molar balance,
and mathematical equations are introduced to describe the rates of reactions or
transformations between different molecules. Different mathematical frameworks
are available depending on the alternative approximations adopted to represent the
biological system under study. Ordinary nonlinear differential equations (ODE)
involving parameters such as reaction rates are commonly used. For example,
the Michaelis–Menten equation is frequently used to describe the internal law
governing the biochemistry of a system, assuming that diffusion is fast compared
to reaction rates. The unknown parameters of ODE models contain key information
that can be gathered in general only indirectly, as it is usually not possible to measure
directly the dynamics of every portion of the system. The recovery of parameter
values can then be only approached as an estimation problem of internal parameters
to fit external input/output (I/O) measurements, where input represents external
perturbations of a system and output the (measured) system response to this input.
In this context, the first relevant question is whether the model parameters can
be (uniquely) determined, at least for suitable input functions, assuming that all
observable variables are error free. This is a mathematical property of the model
called a priori or structural identifiability [1, 4, 5, 8, 9] that can, and should, in
principle be checked before collecting experimental data, or at the latest before
trying to fit the model to data.
Concerning uniqueness, it is important to distinguish between global, i.e.,
one parameter solution, and local, i.e., a finite number of parameter solutions,
identifiability [11]. It has been observed by several authors, see, e.g., [5, 10], that
the weaker property of local identifiability about a parameter value may often be
sufficient in practice. However, in biomedical applications that rely upon model
parameter values to discriminate between classes, e.g., healthy versus pathological
states, global identifiability should be a necessary request [6]. If the postulated
model is neither globally nor locally identifiable, the parameter estimates that
could still be obtained by some numerical optimization algorithms could be totally
unreliable.
From a theoretical point of view only proven structural identifiability can
guarantee, either globally or locally, the uniqueness of the parameter solution, which
is a prerequisite for the parameter estimation problem to be well-posed. Obviously,
although necessary, structural identifiability is not sufficient to guarantee an accurate
identification of the model parameters from real I/O data.
Conversely, many currently studied models in systems biology are rather large
networks containing many states and parameters such that checking structural
Structural vs Practical Identifiability of Nonlinear Differential Equation Models 33

identifiability may become prohibitively complex. Situations of this kind can

be approached by semi-empirical techniques, which are essentially based on
simulations and on the study of the level curves of a cost function which, once
minimized, should yield parameter estimates that are at least unique around an
optimal parameter value. This is called practical identifiability in the literature
[10, 12]. Checking practical identifiability can be done on more realistic models,
which may explicitly involve noise in the measurements. It should be kept in
mind, however, that, since they are data-based (or simulation-based), practical
identifiability methods cannot provide a mathematically rigorous answer to the
uniqueness problem.
In the following sections we provide first the theoretical background for struc-
tural and practical identifiability analysis and then we use a typical benchmark
biological model to show how the joint use of the two approaches appears the most
promising because of the complementary information provided.

2 Structural vs Practical Identifiability

Consider a nonlinear dynamic system described in state-space form as

xP .t/ D f.x.t/; u.t/; / (1)

y.t/ D h.x.t/; u.t/; / (2)

with state x.t/ 2 Rn , input u.t/ 2 Rq ranging on some vector space of differentiable
functions, output y.t/ 2 Rm , and the constant unknown parameter vector 
belonging to some open subset ‚  Rp . Whenever initial conditions are specified,
the relevant equation x.0/ D x0 is added to the system. The functions f and h are
vectors of rational functions in x.

2.1 Structural Identifiability Analysis

We adopt the definition of identifiability proposed in [13]. Let y D x0 .; u/ be the

I/O map of the system (1), (2) started at the initial state x0 . The I/O map allows to
calculate the output function y in terms of the input function u, being u and y both
known from the I/O configuration of system (1), (2). In particular, when starting
with a system described by a state-space representation, as (1), (2), it is possible to
calculate the solution x of the differential equations (1) and to define the I/O map
in terms of the unknown parameters  only. The identifiability definitions are thus
given on the basis of this I/O map.
34 M.P. Saccomani and K. Thomaseth

Definition 1. The system (1), (2) is (a priori) globally (or uniquely) identifiable
from I/O data if, for at least a generic set of points   2 ‚, there exists (at least)
one input function of time, u.t/, such that the equation

x0 .; u/ D x0 . ; u/ (3)

has only one solution  D   for almost all initial states x0 2 X  Rn .

Structural identifiability analysis addresses whether the system model (1), (2) is
globally identifiable and, if not, which subsets of parameters are either globally
(only one solution), locally (a finite number of solutions), or nonidentifiable (an
infinite number of solutions).
For testing structural identifiability, different methods have been proposed to
check models described by linear and nonlinear differential equations [4, 5, 7–9]. In
this paper we focus on a structural identifiability test based on differential algebra
and on the software DAISY (Differential Algebra for Identifiability of SYstems) [2].
The reader is referred to [1, 13] for a detailed documentation of the theory behind
DAISY.
Briefly, this algorithm permits to find the I/O relations as a set of polynomial
differential equations involving only the variables .u.t/; y.t// and their first and
higher order time derivatives. The coefficients of these I/O relations are polynomials
of the unknown parameter  that form the exhaustive summary of the model.
Identifiability is tested by checking injectivity of the exhaustive summary on
parameter . This is achieved by applying Buchberger’s algorithm [3] to compute
a Gröbner basis of the system. In particular, if (3) has one and only one solution ,
the Gröbner basis is of the following form:
 
G.;   / D f 1   1 ; : : : ;  p   p g (4)

showing that the model (1), (2) is globally identifiable. In any case, the Gröbner
basis provides the unique parameterization of the model and allows to count the
number of solutions, i.e., the number of distinct values of the unknown parameter 
that solve the system of equations implied by (3).
In contrast, (3) has infinite many solutions if the basis G.;   / has less
components than the number of estimated parameters. This occurs either if one
or more parameters disappear from the Gröbner basis or if the parameters satisfy
a number of algebraic relations less than p. This means that the I/O map will be
identical, thus non-distinguishable, for all values of the hidden parameters, and/or
for specific, analytically known, combinations of parameters.

2.2 Practical Identifiability Analysis

Practical or data-based identifiability aims at assessing the (statistical) confidence

with which model parameters are estimated from noisy measurements, typically
to judge the reliability of results and to support consequent interpretations. For this
Structural vs Practical Identifiability of Nonlinear Differential Equation Models 35

purpose the model output equations (2) are normally revised by adding measurement
noise, such as

y.t/ D h.x.t/; u.t/; / C ".t/ WD yO .t; / C ".t/ (5)

Prior assumptions on the statistical properties of the random noise ".t/ 2 Rm ,

or actual knowledge of prediction error residuals after model fit to data, greatly
influence calculations of confidence intervals. These latter are most often obtained
using asymptotic results with local quadratic approximation of estimation criteria,
e.g., log-likelihood, around point estimates or nominal parameter values, and first
and second order statistics of measurement noise. While posterior reconstruction
of confidence intervals of model parameter estimates, e.g., obtained by the profile
likelihood approach, is more accurate than quadratic approximations, these latter
are applicable even in an a priori setting and are often fully equivalent in practice,
including as concerns their ability to detect structurally nonidentifiable parameters
characterized by a completely flat profile likelihood function [10].
Assuming that a finite set of N > m input–output measurements is available, the
weighted sum of squared prediction errors is

1X
N
VN ./ WD Œy.tk /  yO .tk ; /> Qk ; Œy.tk /  yO .tk ; / (6)
2 kD1

where Qk are positive semidefinite weights usually taken as the inverse of measure-
ment noise variance, but without loss of generality, assumed in the following equal
to the identity matrix, Qk D I. Finally, with parameter estimates obtained as

O D arg min VN ./ : (7)



The model (1), (2), or parameter , can be defined practically identifiable if the
minimum of VN ./ is well characterized in terms of necessary and sufficient
conditions for a local minimum, i.e., a vanishing gradient: r  VN ./O D 0, and
convexity in the neighborhood of ,O i.e., with a positive definite Hessian matrix:
O > 0.
r 2 2 VN ./
Straightforward calculations under the simplifying a priori assumption of
expected zero measurement noise yield the simplified Hessian matrix:

X
N X
m
O
r 2 VN ./ D O >
r  yj .tk ; /r O
 yj .tk ; / D S./ S./ ;
T
(8)
kD1 jD1

where r  yj .tk ; / 2 Rp 1 is the sensitivity with respect to all estimated param-

eters of the j-th output component measured at time tk ; S./ 2 RmN p , with
S./T D ŒS1 ./T ; S2 ./T ; : : : ; Sm ./T  is the sensitivity matrix formed by all
 of measured model outputs, defined as Sj ./ D
T
individual
 sensitivity matrices
r  yj .t1 ; /; : : : ; r  yj .tN ; / .
36 M.P. Saccomani and K. Thomaseth

It is reminded that positive semi-definite matrices, in contrast to positive definite

ones, can be singular with one, or more, zero eigenvalues. Their corresponding
eigenvectors provide, at the minimum, directions along which the cost function
remains constant. This is the prerequisite for nonidentifiability. A more exhaustive
O e.g., its relationship with the Fisher informa-
statistical interpretation of r 2 VN ./,
tion matrix and the covariance matrix of parameter estimates, goes beyond the scope
of this paper but can be found in standard textbooks, e.g., [15].
Parameter estimation based on prediction error minimization is therefore a well-
posed problem that leads to (local) minima of the cost function (6) if the sensitivity
matrix S./, calculated at some point , has full rank. This must hold almost
everywhere in the admissible parameter space.
Finally, the most dependable approach to assess the rank of a matrix is based on
Singular Value Decomposition which provides the following factorization:

S./ D U†VT (9)

where U 2 RmN mN and V 2 Rp p are the orthonormal eigenvector matrices of

S./S./T and S./T S./, respectively, and † 2 RmN p is diagonal (referring
to the top p  p submatrix) with sorted singular values 1  2 ; : : :  p  0,
which are also the square roots of the eigenvalues of the positive semidefinite matrix
S./T S./. The theoretical (vs practical) rank of S./ is defined as the smallest
r  p at which rC1 D 0 (vs N > rC1 , with N being a user-defined threshold) [16].

3 A Model of JAK-STAT Signaling Pathway

In this section we consider a dynamic model published in several journals [10, 14].
The aim of the model is to investigate the Epo-induced activation of the JAK-STAT
signaling pathway that primarily consists of the cytoplasmic tyrosine kinase JAK
and the latent transcription factor STAT. This pathway is known to play a role in
the regulation of cell proliferation, differentiation, chemotaxis, and apoptosis and is
important for hematopoiesis and immune development. The biochemical reactions
of the JAK-STAT pathway are described by the following nonlinear ODE system:
8
ˆ
ˆ xP 1 .t/ D k1 u.t/ x1 .t/ C 2 k4 x4 .t  /
ˆ
ˆ
ˆ
ˆ xP 2 .t/ D k2 x2 2 .t/ C k1 u.t/ x1 .t/
ˆ
<
xP 3 .t/ D k3 x3 .t/ C k2 x2 2 .t/=2
(10)
ˆ xP 4 .t/ D k4 x4 .t/ C k3 x3 .t/
ˆ
ˆ
ˆ
ˆ
ˆ y1 .t/ D s1 .x2 .t/ C 2 x3 .t//
:̂
y2 .t/ D s2 .x1 .t/ C x2 .t/ C 2 x3 .t//

where xi .t/ i D 1; : : : ; 4 denote the four involved molecular compounds, u.t/

is the input function, ki i D 1; : : : ; 4 the kinetic rate constants, is a “delay
Structural vs Practical Identifiability of Nonlinear Differential Equation Models 37

reaction” which is obtained using a linear chain approximation with intermediate

steps assumed equal to ( D 10=k4 ), s1 ; s2 the scaling parameters, and y1 .t/; y2 .t/
the experimentally observable quantities. The initial conditions are assumed to be
zero, except for x1 .0/ D ic1 which needs to be estimated from the experimental
data.
In the literature, practical identifiability of the model (10) has been analyzed
using statistical criteria. In particular, in [10] the profile likelihood approach is used
based on the idea of detecting flatness of the likelihood function VN ./ by exploring
the parameter space in the direction of least increase of the objective function for
each parameter component. This allows to experimentally observe the behavior of
the function around a nominal parameter value.
In this way, the authors establish that parameters k2 ; s1 ; s2 together with the initial
condition ic1 cannot be identifiable while the others are found to be (practically)
identifiable except for one (k3 ) where the minimum is so flat to be declared
“practically nonidentifiable.” They conclude that this structural nonidentifiability
is a result of missing information about absolute concentration in the experimental
setup. Thus, to get an identifiable model, they enrich the experiment and add a new
measurement. In the more recent paper [14], the authors include the input into the
parameters estimation process. In this case, to get the identifiability of the model,
they fix the initial condition x1 .0/.

3.1 Identifiability Analysis of the JAK-STAT Model

Here we check structural identifiability of model (10) by using the software DAISY
and practical identifiability at nominal parameter values by determining the rank of
the sensitivity matrix. This comparative analysis seems to be done for the first time.
DAISY is applied initially to check the uniqueness of parameter estimates in
the entire parameter space. Results, supported by the Gröbner bases computed
analytically by the algorithm, show that parameters k2 ; s1 ; s2 ; ic1 are linked by
algebraic constraints with one degree of freedom. This actually indicates that these
four parameters are structurally nonidentifiable and that it is sufficient to know
just one of them (not necessarily the scale factor parameter) to make the model
structurally globally identifiable. This “flexibility” for recovering identifiability is
an important issue because it allows for different choices in the design of the
experiment, where many constraints exist especially in a biological experimental
setup. The structural test also guarantees that the practical nonidentifiability of k3
is actually only due to data problems and that it is sufficient to include only one
constraint equation on the initial conditions to retrieve the structural identifiability
of the model. Hence the structural identifiability analysis has essentially replicated
by an analytical approach without experimental data nor assumptions on the initial
conditions values, the results obtained about a nominal point in [10].
In order to integrate the analytical results provided by DAISY with the practical
identifiability approach, the Gröbner basis determined for the JAK-STAT model (10)
38 M.P. Saccomani and K. Thomaseth

Table 1 Gröber basis and Jacobian matrix

G.;   / k1 k2 k3 k4 s1 s2 ic1
20 k1  39 20 0 0 0 0 0 0
95 k2  11 s2 0 95 0 0 0 11 0
25 k3  17 0 0 25 0 0 0 0
100 k4  149 0 0 0 100 0 0 0
19 s1  25 s2 0 0 0 0 19 25 0
20 s2 ic1  19 0 0 0 0 0 20 19

is recalculated for the same nominal parameter values used to assess practical
identifiability [14]. Nominal parameter values are reported here as decimal as
well as rational numbers, in parenthesis, being the latter used by DAISY for
carrying out calculations on a ring with infinite precision: k1 D 1:95.39=20/,
k2 D 0:11.11=100/, k3 D 0:68.17=25/, k4 D 1:49.149=100/, s1 D 1:25.5=4/,
s2 D 0:95.19=20/, ic1 D 1.1/ where k3 was fixed to twice the lower limit.
The Gröbner basis reported in Table 1 (first column), and its Jacobian matrix
formed by the partial derivatives of the Gröber basis with respect to the parameters,
shown for an easier inspection in the right-hand side of Table 1, confirm the
structural results already discussed. In particular, the first, third, and fourth rows
and columns depend each upon one parameter only, and define thus uniquely the
values of the structurally globally identifiable parameters k1 , k3 , and k4 . The second,
fifth, and sixth rows involve the remaining nonidentifiable parameters, namely k2 , s1 ,
s2 , and ic1 , that are thus linked by algebraic constraints with one degree of freedom,
i.e., three Gröbner basis equations in four unknowns.
In order to check practical identifiability, the model variables y1 .t/ and y2 .t/
are simulated between 0 and 60 min, together with all sensitivity equations, using
the nominal parameter values mentioned previously. The model input, u.t/, was
calculated between 0 and 30 min, as the positive half-cycle of the sine wave with
total period 60 min. In Fig. 1 the time course of y1 .t/ and y2 .t/ are shown.
Virtually identical profiles (not shown) are obtained by changing the model
nonidentifiable parameters according to the above Gröbner basis (Table 1).
In particular, the Gröbner basis polynomials equated to zero provide a globally
identifiable parameterization of the model. It is easy to see that the most straightfor-
ward approach to reach it is to fix s2 to an arbitrary numerical value and calculate
the remaining parameters from the other equations: k1 D 39=20, k2 D 11 s2 =95,
k3 D 17=25, k4 D 149=100, s1 D 25 s2 =19, ic1 D 19=.20 s2 /, where fixed
parameters are reported for completeness. By varying the value of s2 one could
observe that the trajectory of y.t/ is not affected.
Alternatively, as considered in [14], one can fix the initial condition ic1 , and
calculate from Table 1, s2 = 19/(20 ic1 ), s1 = 5/(4 ic1 ), k2 = 209/(1900 ic1 ).
Obviously, assignment of any one of the parameters k2 , s1 , s2 , and ic1 and
recalculation of the other parameters is possible.
A final remark regards a geometric interpretation of the relationship between
Gröbner basis and Jacobian matrix reported in Table 1, and the eigenvectors of
Table 2 that form a basis for expressing output variations as functions of parameter
Structural vs Practical Identifiability of Nonlinear Differential Equation Models 39

1.0
0.8
0.6
y1
y
y2
0.4
0.2
0.0

0 10 20 30 40 50 60
t (min)

Fig. 1 Time course of model outputs using as forcing input u.t/ D max.0; sin.2
t=60//

Table 2 Singular values  of sensitivity matrix S./ and relative right eigenvectors V
 23.94 11.29 4.734 0.8312 0.5376 0.313 0
k1 0.00103 0.003642 0.05622 0.06273 0.8462 0.5261 0
k2 0.9356 0.2611 0.2281 0.02152 0.02083 0.01021 0.05899
k3 0.0228 0.000688 0.02984 0.2179 0.5047 0.8345 0
k4 0.04448 0.04458 0.01768 0.9721 0.167 0.151 0
s1 0.1928 0.07973 0.7103 0.01419 0.01845 0.04546 0.6704
s2 0.1296 0.7311 0.4328 0.03032 0.02316 0.01742 0.5095
ic1 0.2612 0.6236 0.5018 0.04418 0.003354 0.03915 0.5363

variations: ıy.t/ D S./ı. In particular, the last column of Table 2, V7 , defines the
null space for parameter perturbations, i.e., ıy.t/ D 0 if ı / V7 , because 7 D 0.
Interestingly, it can be verified that V7 generates also the null space for the Jacobian
matrix in Table 1, i.e., rG.;   /V7 D 0 (up to roundoff errors). This result may
be unexpected but not surprising since G.;   / D 0 defines, for a fixed   , the
values of  that produce identical output trajectories. This is consistent with the fact
that parameter variations, which do not modify output trajectories, do not change
G.  C ı;   /  rG.  ;   /ı.

4 Conclusions

In this study we propose a unified viewpoint of the two identifiability analysis

approaches, namely structural and practical identifiability, by motivating their joint
use. These methodologies are traditionally regarded as disjoint because they are
40 M.P. Saccomani and K. Thomaseth

based, in turn, on differential algebraic manipulations or numerical simulation of

systems equations. The former does not require experimental data and can be
tested using differential algebra software, such as DAISY, on the model equations
without assuming prior knowledge on parameter values, whereas the method based
on sensitivity analysis requires “nominal” parameter values, obtainable from a
simulated or a real experiment, and consists of procedures based on the analysis
of the minima of a likelihood-type function depending on these data. These minima
correspond to numerical parameter estimates. Thus an important difference consists
in the inability of structural identifiability analysis to infer on the precision of
parameter estimation, whereas practical identifiability analysis relies on simulated
data that depend on assumed parameter values and on subjective thresholds to define
a discrimination line between identifiable and nonidentifiable parameters. The
relevant novelty of using both methodologies is that, in case of nonidentifiability,
the structural analysis allows to integrate the practical identifiability results: one can
calculate the analytical relations among the nonidentifiable parameters, described
by the Gröbner basis of the model, at the “nominal” parameter values.
Furthermore, if the parameter turns out to be practically nonidentifiable from
simulated or real data, without having performed the structural identifiability, it
may be difficult to assess the causes from a practical identifiability test. Apparent
nonidentifiability may be due either to structural nonidentifiability or to the paucity
of information in the data or to an imprecise reconstruction, due to noise, of the
level sets or of the minima of the function VN ./. This instead may be revealed,
in analytic terms, by structural methods. By knowing for example that the model
is structurally globally identifiable, the investigator knows that the problem is
related to the simulated or real experimental data, for example, to the scarceness of
measurement samples. Thus, the joint use of the two identifiability approaches can
provide guidelines to avoid unfruitful studies and simulations of modified model
structures.

References

1. Audoly, S., Bellu, G., D’Angiò, L., Saccomani, M.P., Cobelli, C.: Global identifiability of
nonlinear models of biological systems. IEEE Trans. Biomed. Eng. 48(1), 55–65 (2001)
2. Bellu, G., Saccomani, M.P., Audoly, S., D’Angiò, L.: DAISY: a new software tool to test global
identifiability of biological and physiological systems. Comput. Methods Prog. Biomed. 88,
52–61 (2007)
3. Buchberger, B.: Ph.D. thesis 1965: An algorithm for finding the basis elements of the residue
class ring of a zero dimensional polynomial ideal. J. Symb. Comput. 41(3), 475–511 (2006)
4. Chapman, M.J., Godfrey, K.R., Chappell, M.J., Evans, N.D.: Structural identifiability of non-
linear systems using linear/non-linear splitting. Int. J. Control 76(3), 209–216 (2003)
5. Chis, O., Banga, J.R., Balso-Canto, E.: Structural identifiability of systems biology models: a
critical comparison of methods. PloS ONE 6(11), e27755 (2011)
6. Cobelli, C., Saccomani, M.P.: Unappreciation of a priori identifiability in software packages
causes ambiguities in numerical estimates. Letter to the editor. Am. J. Physiol. 21, E1058–
E1059 (1990)
Structural vs Practical Identifiability of Nonlinear Differential Equation Models 41

7. Joly-Blanchard, G., Denis-Vidal, L.: Some remarks about identifiability of controlled and
uncontrolled nonlinear systems. Automatica 34, 1151–1152 (1998)
8. Ljung, L., Glad, S.T.: On global identifiability for arbitrary model parameterizations.
Automatica 30(2), 265–276 (1994)
9. Ollivier, F.: Le problème de l’identifiabilité structurelle globale: étude théorique, méthodes
effectives et bornes de complexité. Thèse de Doctorat en Science, École Polytéchnique, Paris
(1990)
10. Raue, A., Kreutz, C., Maiwald, T., Bachmann, J., Shilling, M., Klingmüller, U., Timmer, J.:
Structural and practical identifiability analysis of partially observed dynamical models by
exploiting the profile likelihood. Bioinformatics 25, 1923–1929 (2009)
11. Raue, A., Karlsson, J., Saccomani, M.P., Jirstrand, M.M., Timmer, J.: Comparison of
approaches for parameter identifiability analysis of biological systems. Bioinformatics 30(10),
1440–1448 (2014)
12. Rodriguez-Fernandez, M., Rehberg, M., Kremling, A., Banga, J.R.: Simultaneous model
discrimination and parameter estimation in dynamic models of cellular systems. BMC Syst.
Biol. 7, 76 (2013)
13. Saccomani, M.P., Audoly, S., D’Angiò, L.: Parameter identifiability of nonlinear systems: the
role of initial conditions. Automatica 39, 619–632 (2004)
14. Schelker, M., Raue, A., Timmer, J., Kreutz, C.: Comprehensive estimation of input signals and
dynamics in biochemical reaction networks. Bioinformatics, ECCB 28, i529–i534 (2012)
15. Seber, G.A., Wild, C.J.: Nonlinear Regression. Wiley, New York (1989)
16. Thomaseth, K., Batzel, J.J., Bachar, M., Furlan, R.: Parameter estimation of a model for
Baroreflex control of unstressed volume. In: Mathematical Modeling and Validation in
Physiology, 215–246. Springer, Berlin (2012)
Boolean Dynamics of Compound Regulatory
circuits

Elisabeth Remy, Brigitte Mossé, and Denis Thieffry

Abstract In biological regulatory networks represented in terms of signed, directed

graphs, topological motifs such as circuits are known to play key dynamical roles.
After reviewing established results on the roles of simple motifs, we present novel
results on the dynamical impact of the addition of a short-cut in a regulatory
circuit. More precisely, based on a Boolean formalisation of regulatory graphs, we
provide complete descriptions of the discrete dynamics of particular motifs, under
the synchronous and asynchronous updating schemes. These motifs are made of
a circuit of arbitrary length, combining positive and negative interactions in any
sequence, and are including a short-cut, and hence a smaller embedded circuit.

Keywords Regulatory motifs • Boolean dynamics

1 Introduction

1.1 Motivations

Biological regulatory networks are often represented in terms of signed, directed

graphs. In these graphs, topological motifs, such as elementary directed and signed
circuits, also often called ‘feedback loops’, are known to play significant dynamical
roles [19]. In particular, positive regulatory circuits (involving an even number of
negative interactions) have been associated with multistability, and more generally
with the occurrence of multiple attractors, which may account for biological
differentiation phenomena. On the other hand, negative circuits have been associated
with sustained periodic behaviour and/or homeostasis [18]. Necessary conditions

E. Remy () • B. Mossé

Aix Marseille Université, CNRS, Centrale Marseille, I2M, UMR 7373, 13453 Marseille, France
e-mail: [email protected]; [email protected]
D. Thieffry
Computational Systems Biology team, Institut de Biologie de l’Ecole Normale Supérieure
(IBENS), CNRS UMR8197, INSERM U1024, Ecole Normale Supérieure, PSL Research
University, F-75005 Paris, France
e-mail: [email protected]

© Springer International Publishing Switzerland 2016 43

A. Rogato et al. (eds.), Dynamics of Mathematical Models in Biology,
DOI 10.1007/978-3-319-45723-9_4
44 E. Remy et al.

relating the occurrence of such circuits with the corresponding dynamical properties
have been defined and properly demonstrated in continuous and discrete frameworks
[13, 14, 16, 17]. However, the dynamical properties of more complex regulatory
motifs made of intertwined circuits still need to be clarified [6]. In this article, we
rely on a Boolean modelling framework (introduced in Sect. 1.2) to review recent
achievements associating simple or more complex regulatory motifs with specific
dynamical properties, i.e. in terms of the number and type of attractors (Sect. 1.3).
Next, we report novel results regarding the dynamical properties of chorded circuits,
made of an elementary (positive or negative) circuit with a chord (Sect. 2). For sake
of brevity, we introduce our main results here, leaving the details (theorems and
proofs) for a forthcoming publication.

1.2 Boolean Formalism

The Boolean formalisation of (biological) regulatory networks relies on the

delineation of two types of graphs, called regulatory graphs and state transition
graphs 1 [5].
In a regulatory graph, each vertex represents a regulatory component and
each arc (oriented, signed edge) represents a regulatory interaction (activation or
inhibition) between two components. Here, each component is associated with a
Boolean variable, meaning that it can take two possible levels, 0 or 1, denoting
the absence/inactivation or the presence/activity of the modelled entity. A logical
rule associated with each component specifies its target value depending on the
presence/absence of its regulators. The dynamical behaviour of the resulting model
can then be computed starting from any initial state, step by step, updating the
current state according to the logical formulae (logical simulations).
The dynamics of a logical model can be represented in terms of a state transition
graph (STG), in which vertices denote different states of the system (represented
by a Boolean vector encompassing the values of all the components), whereas arcs
represent enabled transitions between pairs of states.
In this work, we have considered the two main updating policies for the
generation of STGs. According to the synchronous policy, all components are
updated simultaneously at each step; consequently, each state has at most one

1
Classical terms of graph theory can be found in [3]. Moreover, we use here the following
terminology:
Isolated (elementary) circuit: a connected directed graph with every vertex of in-degree and
out-degree equal to 1;
Circuit: a subgraph of a regulatory graph amounting to an isolated circuit;
Flower-graph: group of circuits sharing one single vertex;
Chorded circuit: circuit with a chord, possibly a self-loop;
Cycle: a subgraph of a state transition graph amounting to an isolated circuit.
Boolean Dynamics of Compound Regulatory circuits 45

successor. In contrast, according to the asynchronous policy, only one variable can
be updated at each step and all the possible successors of a state are considered
(non-deterministic, branching dynamics).
Of particular interest are the sets of states forming attractors, i.e., minimal
groups of states from which the system cannot escape, which represent potential
asymptotical behaviours. Attractors can be ranged into two main classes: stable
states, corresponding here to fixed states (i.e. without successors), and cyclic
attractors, corresponding here to terminal cycles or to more complex terminal
strongly connected components comprising several intertwined cycles.
Several methods have been proposed to efficiently identify all stable states (see,
e.g., [10]). However, other means are needed to assess the reachability of the stable
states from specific initial states, or yet to identify cyclic attractors (see, e.g., [8]).
Proper dynamical analyses often rely on the computation of the STG. As the size of
the model increases, the size of the STG increases exponentially. To cope with this
problem, one can reduce directly the model before simulation (model reduction),
and/or compress the resulting STG into a hierarchical transition graph (HTG) [4].
Strongly connected components (SCCs) form a partition of the STG. They are
trivial (constituted by a unique state) or complex (containing at least two states).
The compression of an STG into a HTG is achieved by clustering the states of
the complex SCCs, and gathering the trivial SCCs leading to the same complex
SCC and attractors. The components grouping trivial SCCs are called irreversible
components.
The HTG displays all the reachable attractors, and the other clusters of states
leading to one single attractor or to specific subsets of attractors. HTG computation
is done on the fly, without having to store the whole STG, which often enables
strong memory and CPU usage shrinking [4]. Furthermore, this functionality
eases the identification of the key commutations (change of component levels)
underlying irreversible choices between the different reachable attractors. The HTG
representation is very compact and very informative regarding the organisation of
the original STG.

1.3 Asymptotical Properties of Simple Motifs

Figures 1 and 2 provide examples of different classes of motifs, each endowed

with specific dynamical properties. The two first rows of Fig. 1 correspond to
the two main classes of regulatory circuits, also often called feedback circuits
or feedback loops, namely positive and negative circuits [19]. The third and
fourth rows correspond to the two main (coherent versus incoherent) classes of
feedforward motifs, also often called—somewhat improperly—feedforward loops.
The dynamical properties for these simple motifs have been extensively documented
[1, 13].
46 E. Remy et al.

Type of motif Definition/Topology Properties (Boolean case)

Positive circuits Circular sequence of Multistability

A B signed interactions involving an even Differentiation
number of negative interactions

A Negative circuits Circular sequence of Periodic or homeostatic

signed interactions involving an odd num- behaviour
C B ber of negative interactions Biological cycles

A
Coherent FFM Direct and indirect (via Filtering of transient signal
B B) interactions from input A onto output
C, with coherent (positive or negative) ef-
C fects on output

A
Incoherent FFM Direct and indirect (via Generation of pulses
B B) interactions from input A onto output
C, with incoherent effects on output
C

Fig. 1 Boolean dynamics of simple regulatory motifs: summary of previous results (notation
FFM: feedforward motif)

Over the last years, a series of results has been obtained regarding the dynamical
properties of compound regulatory circuits, in particular sets of circuits sharing one
single vertex (‘hub vertex’). In such cases, one can infer the dynamics of the whole
system based on that of the hub vertex, as the hub vertex fully determines (directly
or indirectly) the behaviour of the other vertices. Hence, these flower-graphs (as
they are called in [7]) can give rise to 0, 1 or 2 stable states. Figure 2 gives six
examples of such motifs; they together illustrate all possible situations in terms
of attractors, i.e., regarding the potential occurrence of multiple stable states or of
cyclic attractors. The first motif is associated with bistability (coexistence of two
mirroring stable states), in the absence of cyclic attractor. The second motif has a
unique, cyclic attractor. The fifth and sixth motifs have a unique stable state and no
cyclic attractor. The third and fourth motifs correspond to a variety of dynamical
situations depending on the logical rules associated with the hub vertex (AND, OR
or XOR).
Note that each of these cases represents a large class of networks, encompassing
potentially more vertices and interactions, but which can be formally reduced to
these prototypic motifs without fundamental impact on the dynamics (i.e. regarding
the number and types of attractors, see [11]). This suggests that the association
of specific dynamical behaviours with the motifs listed in Figs. 1 and 2 could be
extended to larger classes of motifs.
Boolean Dynamics of Compound Regulatory circuits 47

Type of motif Definition/Topology Properties

Composition of positive circuits sharing Multistability -Two stable

C A B one component (hub) states

Composition of negative circuits sharing Oscillatory behavior - No

C A B one component (hub) stable state

C A B Composition of circuits including one At least one stable state

negative circuit and at least one positive (depending on the logical rule
D circuit, all sharing one component (hub) associated with the hub)

C A B Composition of circuits including one At most one stable state

positive circuit and at least one negative (depending on the logical rule
D circuit, all sharing one component (hub) associated with the hub)

Composition of one negative and one One stable state - No attracting

C A B positive circuits sharing one component cycle
(hub)

Composition of a two-component One stable state - No attracting

A B negative circuit with one positive cycle
autoregulation

Fig. 2 Boolean dynamics of simple regulatory motifs: summary of previous results

2 Boolean Dynamics of Circuits and Chorded Circuits

Most of the works on the regulatory motif listed in Fig. 2 focus only on their
asymptotical behaviours (attractors, and even often only stable states). In the line
of our previous study devoted to isolated circuits [13], we describe the whole
synchronous and asynchronous STGs of regulatory motifs made of an isolated
circuit with a unique chord—possibly a self-loop—(chorded circuits), and compare
their dynamical properties to those of isolated circuits. Using combinatorics on
specific abacus and analysis of recurrent sequences (not shown here), we emphasise
that whatever the chosen updating rule, the STG depends on a small number of
parameters.
We recall the structural properties of the synchronous and asynchronous STGs
of isolated circuits in Sect. 2.1. Then, we present an outline of our new results
concerning the synchronous and asynchronous dynamical structures of chorded
circuits (Sect. 2.2).
48 E. Remy et al.

2.1 Boolean Dynamics of Isolated Circuits

Whatever the updating policy, the STG of an isolated circuit is hierarchically

organised in different levels, each encompassing states with identical numbers of
updating calls (i.e. the number of genes called to change their expression level). At
each level, the number of updating calls is always even in the case of a positive
circuit, while it is odd in the case of a negative circuit.
In the synchronous case, the STG encompasses vertex-disjoint cycles involving
states with the same number of updating calls.
In the asynchronous case, the STG of isolated circuits is connected. In the case
of a positive isolated circuit, the STG is characterised by two stable states (no cyclic
attractor), while in the case of a negative isolated circuit it is characterised by a
single cyclic attractor (the STG generated by a 4-component positive circuit under
asynchronous updating is shown in Figure 3(I), centre).

2.2 Boolean Dynamics of Chorded Circuits

Chorded circuits are made of a long circuit with a chord (additional short-cut
interaction) between two components of the circuit (or amounting to a self-loop),
thereby creating a small circuit (see Fig. 3(II), (III) and (IV) left). The chorded
circuit is coherent if the signs of the two embedded circuits are identical; otherwise,
the chorded circuit is incoherent. We compared the dynamics of chorded circuits
with the dynamics of the long circuit. In any case, part of the states keep the same
updating calls, while other states are sensitive to the presence of the short-cut, and
called therefore hereafter sensitive states. Three cases for the logical rule have been
considered, using the logical operators OR, AND and XOR. Note that using XOR
amounts to define two dual interactions (i.e. with context sensitive signs) converging
on a single vertex. The dynamics obtained with OR and AND rules are symmetrical:
one can transform one of the resulting STGs into the other one by switching (ON
or OFF) all component values. The topology of the STG, and thus the dynamical
properties depend on the sign of the long circuit, and if it is a coherent chorded
circuit or not. In contrast, the topology of the STG and the dynamics obtained with
the XOR rule depends only on the number of genes involved, not on the signs of the
two circuits.

2.2.1 Attractors of Chorded Circuits for the Synchronous Updating

In the cases of the OR and AND logical rules, the synchronous STG contains terminal
cycles.
• If the long circuit is positive, these terminal cycles are found in the synchronous
STG of the long circuit. If the chorded circuit is further coherent (positive small
circuit), there are two stable states; if it is incoherent (negative small circuit),
there is only one stable state.
 (I)
1010 0101
1101
0001
i# 2
0100 1100
1001
A 0111
1000
cc# 12
D B 0010 1011

C
0110 0011
1110 0000 1111

0000 1111

(II)
1010 0101
1101
0001
1010
0100 1100
A 1001
0111
1000
D B cc# 12
0010 1011
C
0100
C if A AND B 0110 0011
1110 0000 1111

0000 1111

(III)
1010 0101
1101
0001
i# 3
0100 1100
A 1001
0111
1000
D B cc# 12
0010 1011
C
C if Ā AND B 0110 0011
1110 0000

0000 1111

(IV)
1010 0101
1101
0001
0100 1100 i# 7
1001
A 0111
1000
D B 0010 1011 i# 7
C
A if A AND D 0110 0011
1110 0000 1111

0000 1111

Fig. 3 Description of the asynchronous dynamics of: a 4-components isolated circuit (I); a
coherent chorded circuit (II); an incoherent chorded circuit (III); a circuit with a coherent
self-regulation (IV). From left to right: regulatory graph, state transition graph (STG) and its
compression into a hierarchical transition graph (HTG). In the later, ‘cc’ and ‘i’ stand for cyclic
and irreversible components, respectively, while the number written after ‘#’ corresponds to the
number of states encompassed by the component
50 E. Remy et al.

• If the long circuit is negative, the terminal cycles differ from those obtained for
the long circuit. If the chorded circuit is incoherent (positive small circuit), there
is only one stable state; if it is coherent (negative small circuit), there is no stable
state.
Accordingly, in the cases OR and AND, a coherent chorded circuit and its corre-
sponding long circuit have the same number of stable states.
In the case of the XOR logical rule, the synchronous STG is constituted of vertex-
disjoint cycles. It contains only one stable state and cycles with pseudo-random
sequence of states, whatever the signs of the circuits.

2.2.2 Attractors of Chorded Circuits for the Asynchronous Updating

When the small circuit is not a self-loop, the asynchronous STG of the chorded
circuit is obtained from that of the long circuit by changing the direction of edges
between pairs of sensitive states that differ by the coordinate of the target component
of the short-cut. When the small circuit consists in a self-loop, these edges are
suppressed or created.
In the cases of the OR and AND logical rules, compare Fig. 3(II) and (III)
with Fig. 3(I) if the small circuit is not a self-loop, and Fig. 3(IV) and (I) in the
case of a self-loop. It can be demonstrated that a coherent chorded circuit and its
corresponding long circuit have the same number and type of attractors, and in
particular the same number of stable states. When the chorded circuit is incoherent,
there is a unique attractor: a stable state. Moreover, the STG of an isolated circuit
is always symmetrical by the transformation switching the component values (cf.
Fig. 3(I) centre: the structure of the STG is conserved when switching all Os to 1 and
vice-versa), and encompasses pairs of such symmetrical states at each level (a level
is characterised by a constant number of updating calls) [13]. The introduction of a
short-cut skews the dynamics. For example, in the case where both long and small
circuits are positive, the basin of attraction of one of the stable states is increased at
the expense of the other one (compare Fig. 3(II) with Fig. 3(I), right).
In the case of the XOR logical rule, the asynchronous STG of a chorded circuit
encompasses a unique stable state as sole attractor. As using an XOR rule amounts to
introduce dual regulations, this could be considered as a particular case of incoherent
chorded motif.

3 Conclusion and Prospects

Figure 4 summarises our novel results regarding the dynamics of chorded circuits,
focusing on the Boolean framework and the asynchronous updating scheme, and
considering three different rules (AND, OR and XOR) for the vertex targeted by
two regulations. These results can be generalised to a wide range of regulatory
Boolean Dynamics of Compound Regulatory circuits 51

Isolated circuits AND / OR chorded circuits XOR chorded circuits

Vertex-disjoint cycles Terminal cycles Vertex-disjoint cycles

In each cycle, states with the same Long pseudo-random cycles
number of updating calls One stable state
Synchronous STG

Positive circuits Positive long circuits

Even numbers of transitions Terminal cycles from the
Two stable states synchronous STG of the long circuit
• Coherent chorded circuit
Two stable states
• Incoherent chorded circuit
One stable state

Negative circuits Negative long circuits

Odd numbers of transitions • Coherent chorded circuit
No stable state
No stable state
• Incoherent chorded circuit
One stable state

Connected level structure Deduced from the asynchronous Deduced from the asynchronous
Levels form the SCCs (except STG of the long circuit STG of the long circuit
perhaps for the two extremal levels), → deleting or creating edges if the → deleting or creating edges if the
and gather states with the same short-circuit is a self-loop small circuit is a self-loop
number of successors → inverting edges otherwise → inverting edges otherwise
Asynchronous STG

Positive circuits Positive long circuits One stable state

Even numbers of transitions • Coherent chorded circuit
Two stable states Two stable states
• Incoherent chorded circuit
One stable state

Negative circuits: Negative long circuits

Odd numbers of transitions • Coherent chorded circuit
One terminal SCC One terminal SCC
• Incoherent chorded circuit
One stable state

Fig. 4 Boolean asynchronous dynamics of chorded circuits, compared to that of isolated circuits

motifs, e.g., involving longer short-cut paths, with the help of the reduction method
described in [11]. However, simple and compound regulatory motifs are usually
embedded in large, intricated networks. In this respect, it can be shown that motifs
embedded in more complex networks may still display the associated properties in
specific conditions, called ‘context of functionality’ in [6].
Noteworthy, recent developments in synthetic biology recurrently refer to reg-
ulatory motifs corresponding to the classes considered in this study, thereby
demonstrating the potential practical impact of studies aiming at fully characterise
the dynamics of simple regulatory motifs (for recent reviews on synthetic biological
circuits, see [9, 12, 20]).
When facing a large and complex network, the enumeration and analysis of its
constitutive motifs can lead to interesting insights about the network dynamics. For
52 E. Remy et al.

example, in the Boolean case, a bound on the number of attractors can be computed
based on the number of positive regulatory circuits, taking into account potential
(indirect) cross-interactions between them [2, 15]. Such results could be refined by
considering recent results on the Boolean dynamics of more complex motifs, such
as the flower-graphs [7], or yet the chorded circuits reported here.
More prospectively, the results obtained in the Boolean framework could serve
as a guide to extend them to the multilevel logical framework, or even to transpose
them into the differential framework, as it was the case with the delineation
of theorems linking elementary positive and negative regulatory circuits with
multistability and cyclic properties (see, e.g., [17, 19]).

References

1. Alon, U.: Network motifs: theory and experimental approaches. Nat. Rev. Genet. 8(6), 450–
461 (2007)
2. Aracena, J., Demongeot, J., Goles, E.: Positive and negative circuits in discrete neural
networks. IEEE Trans. Neural Netw. 15(1), 77–83 (2004)
3. Bang-Jensen, J., Gutin, G.: Digraphs, Theory, Algorithms, Applications. Springer, Berlin
(2008)
4. Bérenguier, D., Chaouiya, C., Monteiro, P.T., Naldi, A., Remy, E., Thieffry, D., Tichit, L.:
Dynamical modeling and analysis of large cellular regulatory networks. Chaos (Woodbury
N.Y.) 23(2), 025114 (2013)
5. Chaouiya, C., Remy, E., Mossé, B., Thieffry, D.: Qualitative analysis of regulatory graphs :
a computational tool based on a discrete formal framework. In: Lecture Notes in Control and
Information Science, vol. 294, pp. 119–26. Springer, Berlin (2003)
6. Comet, J.-P., Noual, M., Richard, A., Aracena, J., Calzone, L., Demongeot, J., Kaufman, M.,
Naldi, A., Snoussi, E.H., Thieffry, D.: On circuit functionality in boolean networks. Bull.
Math. Biol. 75(6), 906–919 (2013)
7. Didier, G., Remy, E.: Relations between gene regulatory networks and cell dynamics in
Boolean models. Discret. Appl. Math. 160(15), 2147–2157 (2012)
8. Garg, A., Dicara, A., Xenarios, I., Mendoza, L., De Micheli, G.: Synchronous vs. Asyn-
chronous Modeling of Gene Regulatory Networks, Bioinformatics (Oxford, England) 24(17),
1917–1925
9. Khalil, A.S., Collins, J.J.: Synthetic biology: applications come of age. Nat. Rev. Genet. 11(5),
367–379 (2010)
10. Naldi, A., Thieffry, D., Chaouiya, C.: Decision diagrams for the representation and analysis of
logical models of genetic networks. In: Computational Methods in Systems Biology. Lecture
Notes in Computer Science, vol. 4695, pp. 233–47. Springer, Berlin (2007)
11. Naldi, A., Remy, E., Thieffry, D., Chaouiya, C.: Dynamically consistent reduction of logical
regulatory graphs. Theor. Comput. Sci. 412(21), 2207–2218 (2011)
12. Purcell, O., Savery, N.J., Grierson, Claire, S., di Bernardo, M.: A comparative analysis of
synthetic genetic oscillators. J. R. Soc. Interface/R. Soc. 7(52), 1503–1524 (2010)
13. Remy, E., Mossé, B., Chaouiya, C., Thieffry, D.: A description of dynamical graphs associated
to elementary regulatory circuits. Bioinformatics (Oxford, England) 19(Suppl. 2), 172–178
(2003)
14. Remy, E., Ruet, P., Thieffry, D.: Graphic requirements for multistability and attractive cycles
in a Boolean dynamical framework. Adv. Appl. Math. 41(3), 335–350 (2008)
15. Richard, A.: Positive circuits and maximal number of fixed points in discrete dynamical
systems. Appl. Math. 157(15), 3281–3288 (2009)
Boolean Dynamics of Compound Regulatory circuits 53

16. Richard, A., Comet, J.-P.: Necessary conditions for multistationarity in discrete dynamical
systems. Discret. Appl. Math. 155(18), 2403–2413 (2007)
17. Soulé, C.: Graphic requirements for multistationarity. Complexus 1, 123–133 (2003)
18. Thomas, R.: On the relation between the logical structure of systems and their ability to
generate multiple steady states or sustained oscillations. In: Numerical Methods in the Study
of Critical Phenomena. Springer Series in Synergetics 9, 180–193 (1981)
19. Thomas, R., D’Ari, R.: Biological Feedback. CRC Press, Boca Raton (1990)
20. Weber, W., Fussenegger, M.: Synthetic gene networks in mammalian cells. Curr. Opin.
Biotechnol. 21(5), 690–696 (2010)
A Differential Transcriptomic Approach
to Compare Target Genes of Homologous
Transcription Factors in Echinoderm Species

Elijah K. Lowe, Claudia Cuomo, and Maria I. Arnone

Abstract Embryonic development is controlled by differential gene expression

throughout developmental time. The ParaHox genes, Cdx and Xlox have been
shown to be involved in the formation of the properly functioning gut in the sea
urchin Strongylocentrotus purpuratus and the sea star Patiria miniata. Several genes
involved in the gene regulatory network (GRN) are known, however, the network is
still incomplete. With the current state of sequencing technology, we are now able
to expand the network and gain further insight into the process of gut development
on a more global scale. Through the use of high-throughput sequencing technology
and knockdown experiments we have further characterized the effects of Cdx and
Xlox on the GRN involved in gut development at different developmental stages.
Additionally, we have conducted a cross-species comparison to identify genes
that are more likely to be evolutionarily important for the development of the
echinoderm gut. Within those genes we found a number of transcription factors
that could potentially have important roles in the formation of the echinoderm gut.
Using both RNA-seq and gene homology, we have set the foundation for further
studies of echinoderm gut and the ParaHox GRN downstream of Xlox and Cdx.

Keywords Differential transcriptomics • Gene regulatory network

1 Introduction

The developmental program of an organism and its phenotypic features are encoded
into its DNA. The binding of transcription factors to specific DNA, which controls
the expression of genes and ultimately the development of the embryo, is known
as a gene regulatory network (GRN). Evolutionary conservation has provided us

E.K. Lowe ()

Stazione Zoologica Anton Dohrn, Naples, Italy
Beacon Center for Evolution in Action, Michigan State University, East Lansing, MI 48823, USA
e-mail: [email protected]
C. Cuomo • M.I. Arnone
Stazione Zoologica Anton Dohrn, Naples, Italy

© Springer International Publishing Switzerland 2016 55

A. Rogato et al. (eds.), Dynamics of Mathematical Models in Biology,
DOI 10.1007/978-3-319-45723-9_5
56 E.K. Lowe et al.

with a good tool to study the origins of phenotypic features and their developmental
programs. With the advances in sequencing technology and the continued drop in
prices, it has become more common to sequence an organism’s transcriptome. This
has facilitated the ability to examine organism on a genomic scale, allowing the
study of all genes expressed at a giving time point in development, as well as for
wild type versus experimental conditions. With transcriptomics we are able to better
understand the complicity of evolution and increasing studies are taking advantage
of this fact [1, 2].
In bilateria, homeobox-containing genes are important for the patterning of the
anterior–posterior axis and mediate much of the embryonic development, with one
of the most studied families being the Hox genes [3, 4]. Another important family
of homeobox genes is the ParaHox family—Gsx, Pdx (Xlox in echinoderms), and
Cdx, which are thought to be the ancient sister group to Hox genes and to have
emerged from the ProtoHox cluster [5]. The ParaHox genes have been shown to
be involved in gut development in vertebrates [6, 7] and also in the echinoderms
[8, 9]. It appeared from the examination of the sea urchins Strongylocentrotus
purpuratus that echinoderm had lost some chordate-like features in their function
of Xlox and Cdx [10]. However, through the use of another echinoderm, the bat
star Patiria miniata, it was discovered that these features appear to only have been
lost in echinoids, while being retained in asteroids [11]. This shows the necessity
to continue to study new organisms in order to gain a more complete evolutionary
picture. The embryonic guts of both S. purpuratus and P. miniata first form a tube
like structure with no sections known as the archenteron, then later divide into three
sections, the foregut, the midgut, and the hindgut, which become in the larva the
esophagus, the stomach, and the intestine, respectively.
Portions of the GRN for gut development in echinoderms have already been
formed, but the network downstream of Xlox and Cdx has yet to be assembled.
In S. purpuratus, Xlox morpholino antisense oligo (MASO) RNA-seq experiments
have been conducted looking at known genes in the network [9], but have not been
studied in-depth. Here we present the groundwork for reconstructing the GRN for
gut development downstream of Xlox and Cdx in both S. purpuratus and P. miniata.
Through the analysis of these MASO RNA-seq experiments we will identify direct
and indirect targets of Xlox and Cdx in both species. Secondly, looking at the
overlap in these two networks at homologous stages, and will better define the genes
needed for the developing gut to form and properly section.

2 Results and Discussion

2.1 Gene Orthology

Prior to understanding or reconstructing the gut GRN for S. purpuratus and

P. miniata, we must first understand the homology relationship between the two
species. Proteomes for S. purpuratus, P. miniata, and Xenopus tropicalis were used
A Differential Transcriptomic Approach to Compare Target Genes. . . 57

Fig. 1 Gene ortholog relationship between S. purpuratus (SPU), P. miniata (PMI), and X. tropi-
calis (XEN). Each circle represents one of the species and their overlap represents the orthologous
groups that are in common. The numbers in the larger black print are the total number of
orthologous groups and in the smaller blue print are the number of single copy orthologous groups

to construct orthologous groups and examine the gut GRN on an evolutionary

scale. There were 29,805, 29,129, and 22,718 protein sequences in each proteome,
respectively. Five sequences were removed from the P. miniata’s proteome during
the filtering process, they were all eight base pairs or less in length. The three
proteomes clustered into 14,066 homologous groups, being composed of 22,576
S. purpuratus proteins in 10,480 groups, 22,252 P. miniata proteins in 10,724
groups, and 20,813 X. tropicalis proteins in 8386 groups. Of these orthologous
groups there were 6034 conserved amongst all three organisms, with 1217 (20 %)
being single copy orthologs. The echinoderms had the largest number of orthologs
as expected with 2545 orthologous groups and 45 % of the groups being single copy
orthologs (Fig. 1).

2.2 Differential Expression Analysis

To identify genes downstream of Xlox and Cdx we analyzed both S. purpuratus

and P. miniata embryos that were separately injected with MASO designed to
block translation of Xlox or Cdx in each species. Time points for transcriptomic
sequencing were selected based on QPCR expression trends from earlier studies
[9, 11]; the midpoint of expression was chosen for each gene in their respective
species. In S. purpuratus the time points selected are 48 hpf (late gastrula) and 72 hpf
(pluteus larva) for Xlox and 66 hpf (prism) for Cdx. In P. miniata the time points are
66 hpf (late gastrula) for Xlox and 90 hpf (early bipinnaria larva) for Cdx, which
are homologous stages to those of S. purpuratus. Looking at the morphology of
the embryos, the sea urchin and the sea star late gastrula represent the stages when
the gut is only an elongated tube without any constrictions, instead the sea urchin
58 E.K. Lowe et al.

Table 1 Homology of differential expressed transcripts

Splox 48 h Splox 72 h Spcdx 66 h Pmlox 66 h Pmcdx 90 h
(294) (2384) (723) (108) (693)
Orthologous groups 183 1457 470 70 404
Proteins in core 97 (33 %) 929 (39 %) 289 (40 %) 39 (36 %) 270 (39 %)
SCO in all 16 (5 %) 145 (6 %) 39 (5 %) 5 (5 %) 34 (5 %)
SCO in echino 23 (8 %) 150 (6 %) 48 (7 %) 10 (9 %) 31 (4 %)
Total proteins 207 (70 %) 1659 (70 %) 529 (73 %) 78 (72 %) 450 (65 %)
In parenthesis is the number of differential expressed genes for the given MASO RNA-seq
experiment. Orthologous groups refer to the number of groups the total number of proteins were
clustered into, while “Total proteins” refers to the total number of proteins that were clustered into
orthologous groups. Proteins in core are the number of proteins found in S. purpuratus, P. miniata,
and X. tropicalis. SCO in all are the number of single copy orthologous found in S. purpuratus,
P. miniata, and X. tropicalis, while SCO in echino are the number of single copy orthologs found
in only S. purpuratus and P. miniata

prism and the sea star early bipinnaria larva have already a tripartite shaped gut. The
pluteus larva is an extra time point we chose for the sea urchin in which the gut is
now complete with its cardiac and pyloric sphincters visible.
Differential expressed transcripts were identified using DESeq2 with a threshold
of log2 fc > ˙0.5 and adjusted p-value of <0.05. In S. purpuratus, as time progressed,
the knockdowns had a larger effect of more transcripts. There were only a couple
of hundreds (294) transcripts affected by the Sp-Lox MASO at 48 hpf, compared to
723 transcripts effected by Sp-Cdx at 66 h, and 2384 at 72 h. Fifty-seven percent
(167) of transcripts affected by the Sp-Lox MASO at 48 hpf were also affected at
72 hpf, showing similarities in the GRN as gut transitions from a tube like structure
to a trisectioned gut.
When examining the P. miniata Xlox MASO RNA-seq we did not find a large
number of transcripts to be differentially expressed at the late gastrula stage, with
there being only 109 transcripts differentially expressed. However, when examining
Pm-Cdx MASO RNA-seq at the early larva stage we observed many more genes
being affected, 693, 450 (65 %) of which had a homologous relation to S. purpuratus
and/or X. tropicalis.
Across all species at least 65 % of the transcripts were clustered into homologous
groups, meaning that 30–35 % of the transcripts from each experiment were species
specific or fell below our threshold (Table 1). Further analysis including phylogenic
trees is necessary to better understand the relationship of these two species, but is
currently out of the scope of this paper.

2.3 Evolutionary Conserved Elements in S. purpuratus

and P. miniata Gut GRNs

Through the use of our orthology analysis and our MASO differential expression
analysis we are able to discover conserved components in the downstream networks
of both Xlox and Cdx in S. purpuratus and P. miniata. In the Cdx MASO RNA-seq
there was the largest overlap between the species, 129 transcripts were found
A Differential Transcriptomic Approach to Compare Target Genes. . . 59

in both networks. Ninety-one out of these 129 genes were found in the “core”
orthology group, meaning that at least one gene from S. purpuratus, P. miniata,
and X. tropicalis was present in the orthologous group, and 11 (9 %) of those genes
were identified as transcription factors that belong to the bzip, bHLH, C2H2, hmg,
p53, and zf-C4 families. Late gastrula in S. purpuratus and P. miniata occurs at
48 hpf and 66 hpf, respectively, with 15 genes shared in their network, 67 % (10)
of which were transcription factors. Although 48 hpf in S. purpuratus and 66 hpf
in P. miniata are more morphologically similar, the overlaps in affected genes were
stronger at 72 hpf in S. purpuratus and 66 hpf in P. miniata, with an additional
10 genes (25 in total) compared to the earlier stage, which also included the same
group of transcription factors. Without the use of ChIP or other technologies such as
ATAC-seq we are not able to determine the connectivity of these GRNs. Although
we are not able to distinguish direct versus indirect targets in this study, identifying
key components in the way of transcription factors is essential and will provide a
foundation for future studies.

3 Conclusion

Here we present the foundation for studying the downstream GRN for gut devel-
opment in S. purpuratus and P. miniata through the use of a MASO RNA-seq
analysis. Seeing that RNA-seq can yield hundreds to thousands of potential genes
we used the correlation between S. purpuratus and P. miniata to identify a subset
of genes to be examined in future studies. Moreover the genes identified in our
study as transcription factors will be the starting points for ATAC-seq and ChIP
analyses. This study provides evidence that a genome-wide approach to study GRNs
in development and evolution is feasible in echinoderms.

4 Methods

All computational analyses were conducted on the high performance-computing

cluster at Michigan State University. Scripts for all the performed analyses are
readily available for use and can be found in the following github repository https://
github.com/elijahlowe/paraHox_analysis. Snippets of code were generated with the
help of biostar and seq-answer online forums [12, 13]. RNA-seq reads will be stored
in EBI database.

4.1 Animal Handling and Microinjection Procedures

Adults S. purpuratus and P. miniata have been obtained from Patrick Leahy
(Kerchoff Marine Laboratory, California Institute of Technology, Pasadena,
CA, USA), housed in circulating seawater aquaria in the Stazione Zoologica
Anton Dohrn of Naples and kept in large tanks of seawater at 15–16 ı C.
60 E.K. Lowe et al.

Microinjection was performed as described in Annunziata and Arnone [9], for

sea urchin, and in Cheatle Jarvela and Hinman [14], for sea star. MASOs were
obtained from Gene Tools (Corvallis) and injected at the following concentration:
150 M, for Sp-Lox and Sp-Cdx translation MASOs (sequences as reported
in [8] and [9]); 700 M, for the Pm-Lox translation MASO (sequence 50 -
CCAGGGTCATCATGTTCATGTTGGT-30 ), and for the Pm-Cdx splicing MASO
(sequence 50 -TTGACCTGTAGTTGAAATATGAGAA-30 ). For each experiment
and for each MASO, 600 zygotes were injected in sea urchin and 50 zygotes in
sea star and each experiment was repeated three times with different batches of
embryos to obtain three independent biological replicas. As a negative control, the
same number of eggs was injected with 100 M of the standard control morpholino
(Gene Tools) and compared side-by-side with uninjected and MASO-injected
embryos.

4.2 Embryos Collection, RNA Extraction, and Sequencing

Injected and uninjected sea urchin and sea star fertilized eggs have been allowed to
develop until the desired stage at 15 ı C in filtered seawater and then collected for
the RNA extraction. The embryos have been collected in a tube and centrifuged
at 3000 rpm for 2–3 min to remove all the seawater. RNA extraction has been
carried out using the RNAqueous-Micro Kit (Ambion). Integrity and quantification
of RNA has been checked before the sequencing using the Agilent Bioanalyzer
2100 with the RNA 6000 Pico kit for total eukaryote RNA. cDNA libraries have
been prepared with 1 g of starting total RNA and using the Illumina TruSeq RNA
Sample Preparation Kit (Illumina), according to TruSeq protocol. Each library has
been diluted to 2 nM and denaturated; 8 pM of each library has been loaded onto
cBot (Illumina) for cluster generation with cBot Paired End Cluster Generation Kit
(Illumina) and sequenced using the Illumina HiSeq 1500 with 100 bp paired-end
reads in triplicate, obtaining 31–38 million reads for replicate. The sequencing
service has been provided by the Laboratory of Molecular Medicine and Genomics
(http://www.labmedmolge.unisa.it) at the University of Salerno, Italy.

4.3 Quality Control, Mapping, and Differential Expression

Reads were first trimmed using Trimmomatic (v0.33) with the scripts trim_pm.qsub,
trim_spcdx.qsub, and trim_splox.qsub [15]. The parameters for trimming were
chosen to efficiently remove erroneous reads while maximizing the information
within the reads [16]. S. purpuratus reads were mapped to Genome sequence
(V3.1) [17] and P. miniata reads were mapped to the genome sequence (V1.0)
Scaffolds [18] using Bowtie2 (2.2.6) and Tophat (2.0.8b) [19, 20]. After mapping,
reads were sorted using SamTools (v1.2) [21] and counts were extracted using
A Differential Transcriptomic Approach to Compare Target Genes. . . 61

HTSeq (v0.6.1) [22]. The gff3 from Build 7 was used for generating exon-based
transcript counts for S. purpuratus which is more informative seeing than DESeq2
does not use length-based count normalization [23, 24]. The following scripts were
used sp_cdx.qsub, sp_lox48.qsub, and sp_lox72.qsub for Sp, while pm_cdx.qsub
and pm_lox.qsub were used for Pm.
Differentially expressed genes were identified using DESeq2 [23], transcripts not
meeting the threshold of 10 counts for at least one of the samples were removed.
DESeq2 provides two methods of hypothesis testing: Wald test and likelihood
ratio test (LRT). To account for the batch effect across different animals we
used LRT, with the full model being batch C condition and the reduced model
being batch. After, the differentially expressed genes using extracted information
from Echinobase [18] for both species, which are in the data/ directory, using
annot_sp.py and annot_pm.py scripts.

4.4 Identification and Clustering of Orthologs

The proteomes for S. purpuratus (SPU_peptide sequence) and P. miniata

(PMI_protein sequence) were downloaded from echinobase (http://www.
echinobase.org/Echinobase/SpDownloads and http://www.echinobase.org/ Echi-
nobase/PmDownload) while Xenopus tropicalis proteome (release 83) was
downloaded from Ensembl (ftp://ftp.ensembl.org/pub/release-83/fasta/xenopus_
tropicalis/pep/) in fasta format [18, 25]. Orthology was determined using orthoMCL
[26]. Sequences of the three proteomes were concatenated into one file and
transformed into orthoMCL format, so an all-vs-all protein blast search was
conducted using the blastp program in the BLASTC (v2.2.30) suite [27]. Prior to
the blast search, sequences with stop codons and of a length shorter than 20 amino
acids were removed. A protein blast (blastp) was performed using the concatenated
fasta as the query and database. Blast results were then parsed, loaded into a MySQL
database, and then proteins with at least 50 % similarity were clustered through the
use of orthoMCL programs. The steps for orthoMCL are in the script ortho.qsub.

4.5 Transcription Factor Identification

Both the S. purpuratus and P. miniata proteomes were searched against the Pfam
database [28] using HMMER/3.1b2 hmmscan [29]. These commands were executed
using the following scripts hmmer_pm_tf.qsub and hmmer_spur_tf.qsub. The grep
program was then used to search for the following term homeobox, Pax, bzip,
hmg, sox, hlh, PF00104.25 (nuclear receptor), t-box, mh2 (smad), b-box, f-box,
fork_head, ets, phd-finger, zf-C2H2 within –tblout output. Additionally, Pfam ids
were extracted from the DBD Transcription Factor prediction database [30] and
62 E.K. Lowe et al.

then grep against the –tblout output, combined filtered for redundancy. The list of
Pfam ids can be found in the data directory in the github repository along with the
TF we identified for S. purpuratus and P. miniata.

Acknowledgements This work was supported in part by Michigan State University through
computational resources provided by the Institute for Cyber-Enabled Research and by MIUR
(premiale PANTRAC to MIA). C.C. has been supported by a SZN PhD fellowship.

References

1. Wang, Z., Dai, M., Wang, Y., Cooper, K.L., et al.: Unique expression patterns of multiple key
genes associated with the evolution of mammalian flight. Proc. Biol. Sci. 281(1783), 20133133
(2014)
2. Lmanna, F., Kirschbaum, F., Waurick, I., Dieterich, C., Tiedemann, R.: Cross-tissue
and cross-species analysis of gene expression in skeletal muscle and electric organ of
African weakly-electric fish (Teleostei; Mormyridae). BMC Genomics 16, 668 (2015).
doi:10.1186/s12864-015-1858-9
3. Finnerty, J.R.: The origins of axial patterning in the metazoa: how old is bilateral symmetry?
Int. J. Dev. Biol. 47(7–8), 523–529 (2003)
4. Mallo, M., Alonso, C.R.: The regulation of hox gene expression during animal. Development
140(19), 3951–3963 (2013)
5. Brooke, N.M., Garcia-Fernandez, J., Holland, P.W.: The ParaHox gene cluster is an evolution-
ary sister of the Hox gene cluster. Nature 392, 920–922 (1998)
6. Wright, C.V., Cho, K.W., Oliver, G., De Robertis, E.M.: Vertebrate homeodomain proteins:
families of region-specific transcription factors. Trends Biochem. Sci. 14, 52–56 (1989)
7. Young, T., Deschamps, J.: Hox, Cdx, and anteroposterior patterning in the mouse embryo.
Curr. Top. Dev. Biol. 88, 235–255 (2009)
8. Cole, A.G., Rizzo, F., Martinez, P., Fernandez-Serra, M., Arnone, M.I.: Two ParaHox genes,
SpLox and SpCdx, interact to partition the posterior endoderm in the formation of a functional
gut. Development 136, 541–549 (2009)
9. Annunziata, R., Arnone, M.I.: A dynamic regulatory network explains ParaHox gene
control of gut patterning in the sea urchin. Development 141(12), 2462–2472 (2014).
doi:10.1242/dev.105775
10. Arnone, M.I., Rizzo, F., Annunciata, R., Cameron, R.A., Peterson, K.J., Martínez, P.: Genetic
organization and embryonic expression of the ParaHox genes in the sea urchin S. purpuratus:
insights into the relationship between clustering and collinearity. Dev. Biol. 300, 63–73 (2006)
11. Annunziata, R., Martinez, P., Arnone, M.I.: Intact cluster and chordate-like expression of
ParaHox genes in a sea star. BMC Biol. 11, 68 (2013). http://www.biomedcentral.com/1741-
7007/11/68
12. Parnell, L.D., Lindenbaum, P., Shameer, K., Dall’Olio, G.M., Swan, D.C., et al.: BioStar: an
online question & answer resource for the bioinformatics community. PLoS Comput. Biol.
7(10), e1002216 (2011)
13. Li, J.W., Schmieder, R., Ward, R.M., Delenick, J., Olivares, E.C., Mittelman, D.: SEQanswers:
an open access community for collaboratively decoding genomes. Bioinformatics 28(9), 1272–
1273 (2012)
14. Cheatle Jarvela, A.M., Hinman, V.: A method for microinjection of Patiria miniata zygotes. J.
Vis. Exp. (91), e51913 (2014). doi:10.3791/51913
15. Bolger, A.M., Lohse, M., Usadel, B.: Trimmomatic: a flexible trimmer for Illumina sequence
data. Bioinformatics 30, 2114–2120 (2014)
A Differential Transcriptomic Approach to Compare Target Genes. . . 63

16. MacManes, M.D.: On the optimal trimming of high-throughput mRNAseq data. bioRxiv
(2014). doi: 10.1101/000422
17. Sodergren, E., Weinstock, G.M., Davidson, E.H., Cameron, R.A., Gibbs, R.A., Angerer,
R.C., Coffman, J.A.: The genome of the sea urchin Strongylocentrotus purpuratus. Science
314(5801), 941–952 (2006)
18. Cameron, R.A., Samanta, M., Yuan, A., He, D., Davidson, E.: SpBase: the sea urchin genome
database and web site. Nucleic Acids Res. 37, D750–D754 (2009)
19. Langmead, B., Salzberg, S.L.: Fast gapped-read alignment with Bowtie 2. Nat. Methods 9(4),
357–359 (2012)
20. Kim, D., Pertea, G., Trapnell, C., Pimentel, H., Kelley, R., Salzberg, S.L.: TopHat2: accurate
alignment of transcriptomes in the presence of insertions, deletions and gene fusions. Genome
Biol. 14(4), R36 (2013)
21. Li, H., Handsaker, B., Wysoker, A., Fennell, T., Ruan, J., Homer, N., Marth, G., Abecasis, G.,
Durbin, R., 1000 Genome Project Data Processing Subgroup: The Sequence alignment/map
(SAM) format and SAMtools. Bioinformatics 25, 2078–2079 (2009)
22. Anders, S., Pyl, P.T., Huber, W.: HTSeq — a Python framework to work with high-throughput
sequencing data. Bioinformatics 31, 166–169 (2014). doi:10.1093/bioinformatics/btu638
23. Love, M.I., Huber, W., Anders, S.: Moderated estimation of fold change and dispersion for
RNA-seq data with DESeq2. Genome Biol. 15, 550 (2014). doi:10.1186/s13059-014-0550-8
24. Zhao, S., Xi, L., Zhang, B.: Union exon based approach for RNA-seq gene quantification: to
be or not to be? PLOS One (2015). doi:10.1371/journal.pone.0141910
25. Cunningham, F., Amode, M.R., Barrell, D., Beal, K., Billis, K., Brent, S., Carvalho-Silva,
D., Clapham, P., Coates, G., Fitzgerald, S., Gil, L., Girón, C.G., Gordon, L., Hourlier, T.,
Hunt, S.E., Janacek, S.H., Johnson, N., Juettemann, T., Kähäri, A.K., Keenan, S., Martin,
F.J., Maurel, T., McLaren, W., Murphy, D.N., Nag, R., Overduin, B., Parker, A., Patricio,
M., Perry, E., Pignatelli, M., Riat, H.S., Sheppard, D., Taylor, K., Thormann, A., Vullo, A.,
Wilder, S.P., Zadissa, A., Aken, B.L., Birney, E., Harrow, J., Kinsella, R., Muffato, M., Ruffier,
M., Searle, S.M.J., Spudich, G., Trevanion, S.J., Yates, A., Zerbino, D.R., Flicek, P.: Ensembl
2015. Nucleic Acids Res. 43(Database issue), D662–D669 (2015). doi:10.1093/nar/gku1010
26. Fischer, S., Brunk, B.P., Chen, F., Gao, X., Harb, O.S., Iodice, J.B., Shanmugam, D., Roos,
D.S., Stoeckert Jr., C.J.: Using OrthoMCL to assign proteins to OrthoMCL-DB groups or
to cluster proteomes into new ortholog groups. Curr. Protoc. Bioinformatics. Chapter 6:Unit
6.12.1–19 (2011)
27. Camacho, C., Coulouris, G., Avagyan, V., Ma, N., Papadopoulos, J., Bealer, K., Madden, T.L.:
BLASTC: architecture and applications. BMC Bioinformatics 10, 421 (2008)
28. Finn, R.D.: Pfam: the protein families database. Encyclopedia of Genetics, Genomics,
Proteomics and Bioinformatics (2012)
29. Durbin, R., Eddy, S.R., Krogh, A., Mitchison, G.: Biological Sequence Analysis: Probabilistic
Models of Proteins and Nucleic Acids. Cambridge University Press (1998). ISBN 0-521-
62971-3
30. Wilson, D., Charoensawan, V., Kummerfeld, S.K., Teichmann, S.A.: DBD - taxonomically
broad transcription factor predictions: new content and functionality. Nucleic Acids Res. 36,
D88–D92 (2008). doi:10.1093/nar/gkm964
Reconstructing a Genetic Network from Gene
Perturbations in Secretory Pathway of Cancer
Cell Lines

Marina Piccirillo, Kumar Parijat Tripathi, Sonali Gopichand Chavan,

Seetharaman Parashuraman, Alessandra Varavallo, and Mario Guarracino

Abstract Gene perturbation studies play an important role in the reconstruction of

genetic networks and in determining the influence of genes on each other activities.
According to this hypothesis, we planned to develop new analysis methods, based
on novel algorithms, to reconstruct genetic networks by incorporating gene expres-
sion datasets, containing profiles of cell lines that have been exposed to genetic
perturbations. In the present work, we focus on a list of genes, localized in secretory
pathway. These genes and their products are responsible for the delivery of different
kind of proteins from their site of synthesis to their proper cellular location and they
are essential for cellular functions and multicellular development. Using data from
high-throughput experiments, gene expression profiles are collected from 33 genes
perturbations (knockdown and over-expressed) experiments in four cancer cell lines.
Data have been downloaded from the Library of Integrated Network-Based Cellular
Signatures. We characterized gene regulatory networks of secretory pathway, and
we provided some empirical results of the network modular organization. The
interesting observation is that all these regulatory genes are also connected with each
other through hub nodes. It means that interactions do not have a separate entity and
are not regulated by independent behavior of perturbed genes, but probably, there is
a global effect of all these perturbations on all subnetworks present in an interaction
network.

Keywords Perturbation • Genetic network • Algorithm • Secretory pathway

M. Piccirillo () • K.P. Tripathi • M. Guarracino

Laboratory for Genomics, Transcriptomics and Proteomics (Lab-GTP), High Performance
Computing and Networking Institute (ICAR), National Research Council (CNR),
Via Pietro Castellino 111, Naples, Italy
e-mail: [email protected]
S.G. Chavan • S. Parashuraman • A. Varavallo
Institute of Protein Biochemistry, National Research Council (CNR-IBP), Via Pietro Castellino
111, Naples, Italy

© Springer International Publishing Switzerland 2016 65

A. Rogato et al. (eds.), Dynamics of Mathematical Models in Biology,
DOI 10.1007/978-3-319-45723-9_6
66 M. Piccirillo et al.

1 Introduction

Genetic perturbations are experimental alterations of gene activity, by manipulating

either the gene itself or its products. Such perturbations include point mutations,
gene deletions, over-expression, or any other interference with the activity of the
genes or their product. They can be used in conjunction with a reverse engineering
algorithm to reconstruct and reveal the architecture of a gene regulatory network
(GRN), by analyzing the steady-state changes in gene expression of a particular
node in the network [9]. GRNs are the most important abstract organizational
level in the cell, because they symbolize the signals that cells receive, in terms
of activation and inhibition of genes [1], as shown in Fig. 1 (image extracted 14
May 2007 from the http://genomics.energy.gov website from the U.S. Department
of Energy Genome Programs). GRNs are represented as graphs, in which nodes
are genes, proteins, or metabolites and edges are the relations between the nodes;
therefore it is possible to understand the molecular mechanism of each gene by
identifying their interactions within the GRNs [6]. Our aim is to study GRNs from
gene expression data, to identify direct and indirect interactions among genes,
and to reconstruct the characteristics of the secretory pathway [2]. We chose to
study secretory pathway because experimental evidence indicates that endoplasmic
reticulum (ER) and Golgi apparatus can activate both survival mechanisms and
cell suicide programs if the stress-signaling threshold is exceeded. Furthermore it
is possible that the fragile balance of protein trafficking between various subcellular
compartments provides a good therapeutic opportunity [11]. Several techniques
have been proposed to analyze large data sets from whole-genome networks, such
as cluster analysis and enrichment analysis, but they typically provide only indirect
information about network structure [7].

Fig. 1 A gene regulatory network

Reconstructing a Genetic Network from Gene Perturbations in Secretory. . . 67

2 Material and Methods

2.1 Data Retrieval

The secretory pathway has evolved to facilitate the transfer of cargo molecules
to internal and cell-surface membranes [8]. Its study and characterization are
a challenge, that the use of high-throughput experiments and network analysis
tools have enabled to outdo. In this work, we try to reconstruct the regulatory
networks of secretory pathways starting from 476,251 signatures and 22,268 probes
present in the LINCS website (http://www.lincscloud.org/); selecting the gene
expression profile data related to 33 gene perturbation experiments carried out in
four cancer cell lines (A549, HA1E, HEPG2, and PC3). In Mitocheck database
(http://www.mitocheck.org/), these latter are classified as mild or strong inhibitors of
secretory cargo proteins from ER to plasma membrane and they are involved in the
morphological alterations of COPII and/or COPI vesicular coat complexes. Then,
for each expression profile we collected two or more technical replicates at different
time points, considering profiles of differentially expressed genes computed by
robust z-scores for each profile relative to population control. The genes are shown
in Fig. 2, where we see the 33 perturbations with all the functions in which they are
involved.

2.2 Reconstruction of Regulatory Interactions

To reconstruct regulatory interactions in gene networks from gene perturbation

experiments, we developed a Python computational pipeline, which is divided into
four steps:
1. For each cell line and for each perturbation, we computed the mean among the
technical replicates of each biological replicate.
Let ci 2 C a set of cell lines, with i D 1; 2; 3; 4. Perturbation experiments
ij
pij 2 P with j D p1; p2; : : : p33 have biological replicate represented as bl 2 B
ijl
with l D 1; : : : Lij . In turn, each biological replicate has technical replicate tk
with k D 1; ::kijl . The mean m for each perturbation p for a given cell line i is
calculated as:
ij Pk ijl
– bl D kD1 tk =kpijl , where k is the number of technical replicates for
perturbation experiments.
2. In the second step, we created a matrix M; whose columns are the biological
replicates, and then we calculated the first principal component pcij , which is the
linear combination of x-variables along the direction of maximum variance.
68 M. Piccirillo et al.

condensation followed by decon

metaphase aligment problems

accumulation of gfp-rnf168 on
retention of sh4(yes)-mcherry
reduction in ir-induced 53bp1
reduction in ir-induced 53bp1
condensation without mitosis

strong inhibition of secretion

retention of sh4(haspb)-gfp
altered gm130 morphology
mild inhibition of secretion
nuclei stay close together

failure in decondensation

altered copii morphology

altered copi morphology
strange nuclear shape

increased proliferation
segregation problems

inhibition of secretion
prometaphase delay

migration (distance)

enhanced secretion
dynamic changes

migration (speed)
metaphase delay

pulsating nuclei
small nucleus

large nucleus
cell migration

mitotic delay
cell death

polylobed
binuclear

grape
large
ACSL1
ACTR3
AR
BUB1B
C1S
CALM3
CIRBP
COPB2
DNM1
EML3
ESYT1
GPR142
GPR35
GSTP1
HMBOX1
IFIH1
LRP4
MMP14
MXD4
NT5C2
PHKG2
PML
PROCR
RAP1GDS1
RUVBL1
RXRA
SIRT2
SLC25A4
SLC25A46
SOS2
TBX20
ZFP36
ZFP36L2

Fig. 2 List of 33 secretion regulators

3. For each cell line and for each perturbation, we determined the biological
replicate bN l among bl with maximum correlation, and then we constructed a
ij ij

matrix with them. The matrix Ai D bN ij represents the influence of perturbation

on the expression values of all the probes in the experiments.
4. For each Ai and for each perturbation, we selected only those probes, for which
we observed a fold change greater than 4 in case of over expression genes and
decrease more than 4 in under expression genes, in at least one bN ij .
Reconstructing a Genetic Network from Gene Perturbations in Secretory. . . 69

2.3 Networks Reconstructions and Enrichment Analysis

The molecular interaction networks, for each cell line, were studied using the
network visualization software Cytoscape [5]. To reconstruct the characteristics of
the secretory pathway, we performed an enrichment analysis using the Molecu-
lar Signature Database (MsigDB) (http://www.broadinstitute.org/gsea/downloads.
jsp).1 We used Kyoto Encyclopedia of Genes and Genomes (KEGG) and BIO-
CARTA pathway gene set to study the enriched pathways in the networks. These
tools reduce the complexity of analysis by grouping long lists of individual genes
into smaller sets of related genes or proteins, that are involved in the same biological
processes, components, or structures [4].

3 Results

3.1 Networks Reconstructions

From the reconstruction described in the previous section, we obtained four

networks as shown in Figs. 3, 4, 5, and 6.
In A375 cancer cell line network there are 679 nodes and 1094 edges, while A549
cancer cell line contains 523 nodes and 978 edges. Instead we can distinguish 1140
nodes and 1609 edges in HA1E and 397 nodes with 665 edges in HEPG2 network,
respectively. The obtained networks portray the direct and indirect interactions
among genes, as well as the regulatory effects that perturbations have with other
interaction partner genes. As we can see some perturbations, don’t have independent
behavior, but a combinatorial effect on transcriptional regulation.

3.2 Role of Secretion Regulators in Different Cellular

Processes

Over the last years with the new high-throughput imaging-based methods, and more
recently, with RNA interference (RNAi)-mediated gene knockdown experiments, a
significant number of regulators associated with the secretory pathway have been
revealed. In this study, we used a computational approach to try to find these
regulators of secretory pathway. Comparing our networks with a list of secretion
inhibitors involved in cell death, cell division, and motility; which we selected
from a precedent study [8], we found that some perturbations regulate several
of these inhibitors in each network as shown in Tables 1 and 2. An enrichment

1
http://software.broadinstitute.org/gsea/msigdb/annotate.jsp.
70 M. Piccirillo et al.

Fig. 3 Regulatory network in A375 cancer cell line. In red are depicted the 33 perturbations

analysis of the genes showed that more than 70 % of them participate in fundamental
cellular processes such as transcription, explaining how their knockdown, causes
cell death and therefore the transport inhibition. In particular, we put our focus
on the membrane traffic regulators, such as COPB2, which encode for subunit
beta of the Golgi coatomer complex and whose depletion causes cell death, thus
underlining the importance of the secretory pathway for general cell health. This is
a perturbation that is indirectly down-regulated from other perturbations in A375,
A549, and HA1E networks. For example, in A549 cancer cell line, COPB2 is
indirectly connected with PML and EML3 perturbations. See Fig. 7.
A crucial observation is that in HA1E network, COPB2 is connected directly
with RUVBL1, which is located in Golgi apparatus and vesicles, and both regulate
some common genes. Membrane traffic pathways are also regulated through the
activities of kinases and phosphatases, and some of these are involved in ER–Golgi
recycling. Overlapping, our regulatory interactions with a list of 48 genes, which are
scored as secretion inhibitors in the study of Farhan et al. [3], we found that, in each
network, there are some of these genes that are regulated from the perturbations
(see Table 3). With respect to HEPG2 and A549 networks, two perturbations are
connected indirectly through EPBHB2 of ER receptor tyrosine kinase, involved in
axon guidance.
Reconstructing a Genetic Network from Gene Perturbations in Secretory. . . 71

Fig. 4 Regulatory network in A549 cancer cell line

4 Discussions and Conclusions

GRNs represent a combination of diverse regulation and interaction mechanisms

operating in different conditions and time scales. We integrated such data to try to
describe with a computational approach the characteristics of secretory pathway.
We applied here the proposed algorithm to gene expression’s data of 33 perturbed
genes in four cancer cell lines, and we reconstructed GRNs, providing also better
understanding of cellular response towards chemical and genetic perturbations. A
72 M. Piccirillo et al.

Fig. 5 Regulatory network HA1E cancer cell line

main result of the proposed procedure is that all these regulatory genes are also con-
nected with each other through hub nodes. It means that the transcriptional response
with respect to each perturbations does not have independent behavior, but somehow
these perturbations put a combinatorial effect on transcriptional regulation, perhaps
there is a global effect of all these perturbations on all subnetworks present in
an interaction network. Our analyses indicate that the perturbations control some
genes, which are involved in several processes of secretory pathway regulations.
For example, depletion of the Golgi coatomer complex, COPB2 which is essential
for Golgi budding and vesicular trafficking, caused cell death. Furthermore, we
Reconstructing a Genetic Network from Gene Perturbations in Secretory. . . 73

Fig. 6 Regulatory network HEPG2 cancer cell line

also found kinases and phosphatases, that can regulate membrane traffic pathways.
In particular we found that the secretion inhibitors EPHB2 indirectly connect two
perturbations in HEPG2 and A549 network. Together, these results imply that the
mammalian cells have a highly sophisticated signaling and feedback system that
allows them to modulate their secretory activity in response to external signals and
their local environment. So our algorithm may help answer a multitude of questions
about the genetic architecture of organisms. What is the structure of genetic
networks? How do patterns of interactions genes change in different developmental
stages, in different physiological states, in different environmental conditions, or
in different cell types? Are there many genes that do not affect the activity of
other genes? This approach could be useful also in determining the potential drug
targets in case of aggressive human tumors. Furthermore, it is possible to cluster
the mechanisms of action rather than the gene expression pattern; and moreover our
algorithm is very scalable in term of the number of experiments used to a given
network. For the future work by taking into account of existing algorithms and
inference methods for reverse engineering of gene networks from large scale gene
expression data, we plan to deepen the study of these techniques and overcome their
limits; indeed our next goal will be to implement Wagner’s algorithm [10], which is
important to find short cycles and loops in the network.
74 M. Piccirillo et al.

Table 1 List of secretion inhibitors involved in cell death, cell division, and motility in A375 and
A549 cells
A375 A549
Gene Cell division Gene Cell division
symbol Ensembl ID phenotypes symbol Ensembl ID phenotypes
ACADVL ENSG00000072778 Mitosis, cell AKT1 ENSG00000142208 Other
death phenotypes
BMPR1A ENSG00000107779 Mitosis, Other BUB1B ENSG00000156970 Mitosis
phenotypes
BUB1B ENSG00000156970 Mitosis CDC23 ENSG00000094880 Mitosis
CDK4 ENSG00000135446 Other CDK4 ENSG00000135446 Other
phenotypes phenotypes
CHN1 ENSG00000128656 Other COPB2 ENSG00000184432 Cell death
phenotypes
CLIC4 ENSG00000169504 Other EML3 ENSG00000149499 Mitosis
phenotypes
COPB2 ENSG00000184432 Cell death GSTP1 ENSG00000084207 Cell death
DNAL4 ENSG00000100246 Other IDH1 ENSG00000138413 Mitosis
phenotypes
EML3 ENSG00000149499 Mitosis MXD4 ENSG00000123933 Other
phenotypes
GJB3 ENSG00000188910 Mitosis NOL3 ENSG00000140939 Mitosis
GRWD1 ENSG00000105447 Cell death PML ENSG00000140464 Mitosis
GSTP1 ENSG00000084207 Cell death SAMD ENSG00000020577 Other
4A phenotypes
KIF2C ENSG00000142945 Mitosis SIRT2 ENSG00000068903 Mitosis
MXD4 ENSG00000123933 Other ST6GAL ENSG00000122912 Other
phenotypes NAC2 phenotypes
NFKBIE ENSG00000146232 Mitosis TBCA ENSG00000171530 Cell death
NUP93 ENSG00000102900 Cell death TRIB3 ENSG00000101255 Migration
PLCB2 ENSG00000137841 Mitosis TXND ENSG00000115514 Mitosis
C9
PML ENSG00000140464 Mitosis TYMS ENSG00000176890 Cell death
PPOX ENSG00000143224 Other
phenotypes
PRSS23 ENSG00000150687 Other
phenotypes
RPA1 ENSG00000132383 Mitosis, cell
death
SCYL3 ENSG00000000457 Other
phenotypes
SIRT2 ENSG00000068903 Mitosis
SLC16A3 ENSG00000141526 Other
phenotypes
SLC25A ENSG00000122912 Other
16 phenotypes
ST6GAL ENSG00000122912 Other
NAC2 phenotypes
TXNDC9 ENSG00000115514 Mitosis
Reconstructing a Genetic Network from Gene Perturbations in Secretory. . . 75

Table 2 List of secretion inhibitors involved in cell death, cell division, and motility in HA1E and
HEPG2 cells
HA1E HEPG2
Gene Cell division Gene Cell division
symbol Ensembl ID phenotypes symbol
Ensembl ID phenotypes
AMHR2 ENSG00000135409 Other phenotypes ACAD ENSG00000072778 Mitosis, cell
VL death
BUB1B ENSG00000156970 Mitosis ALDOA
ENSG00000149925 Mitosis
C10orf68 ENSG00000150076 Other phenotypes BUB1BENSG00000156970 Mitosis
CDK5R1 ENSG00000176749 Mitosis, migration, CLIC4
ENSG00000169504 Other
other phenotypes phenotypes
CER1 ENSG00000147869 Cell death COPB2
ENSG00000184432 Cell death
CLCNKB ENSG00000184908 Mitosis EML3ENSG00000149499 Mitosis
CLIC4 ENSG00000169504 Other phenotypes GSTP1 ENSG00000084207 Cell death
COPB2 ENSG00000184432 Cell death IDH1ENSG00000138413 Mitosis
ECD ENSG00000122882 Mitosis ITGB5
ENSG00000082781 Migration,
other
phenotypes
EEF1E1 ENSG00000124802 Migration MXD4 ENSG00000123933 Other
phenotypes
EML3 ENSG00000149499 Mitosis MYL6B ENSG00000196465 Mitosis
GRWD1 ENSG00000105447 Cell death PML ENSG00000140464 Mitosis
GSTP1 ENSG00000084207 Cell death SAMD ENSG00000020577 Other
4A phenotypes
KCNQ4 ENSG00000117013 Other phenotypes SIRT2 ENSG00000068903 Mitosis
MXD4 ENSG00000123933 Other phenotypes ST6GAL ENSG00000122912 Other
NAC2 phenotypes
NBR1 ENSG00000188554 Cell death TAGLN ENSG00000149591 Cell death
OGG1 ENSG00000114026 Mitosis TYMS ENSG00000176890 Cell death
PLA2G3 ENSG00000138308 Cell death USP1 ENSG00000162607 Mitosis,
migration
PML ENSG00000140464 Mitosis
PPP2R1A ENSG00000105568 Mitosis
ROS1 ENSG00000047936 Other phenotypes
RRM1 ENSG00000167325 Other phenotypes
SAMD4AENSG00000020577 Other phenotypes
SCN5A ENSG00000183873 Mitosis
SIRT2 ENSG00000068903 Mitosis
ST6GAL ENSG00000122912
NAC2
TAGLN ENSG00000149591 Cell death
TRIB3 ENSG00000101255 Migration
TXNDC9 ENSG00000115514 Mitosis
TYMS ENSG00000176890 Cell death
76 M. Piccirillo et al.

IGKC

FABPS
up GOLT1B
dw dw
dw
PML
COPB2
dw
dw dw
dw LGMN
RPS4Y1 dw
up
up TMSB4X
LGALS1

dw
IKBKE

EML3 NPRL2

Fig. 7 Sub-network of A549 cancer cell line, in which we can see all the first neighbors nodes of
COPB2 and its indirect interactions with other perturbations depicted in red

Table 3 List of kinases and phosphatases which are regulated from our perturbations in each
network
Farhan et al. Class Gene symbol Gene bank Description
A375
Golgi ABL1 NM005157 v-Abl Abelson murine leukemia viral
oncogene homolog 1
Golgi AURKB NM004217 Aurora kinase B
Golgi CDK4 NM000075 Cyclin-dependent kinase 4
A549
Golgi ABL1 NM005157 v-Abl Abelson murine leukemia viral
oncogene homolog 1,
Golgi AURKB NM004217 Aurora kinase B
Golgi CDK4 NM000075 Cyclin-dependent kinase 4
ER EPHB2 NM017449 EPH receptor B2
Golgi KIT NM000222 v-Kit Hardy-Zuckerman 4 feline sarcoma
viral oncogene homolog
HA1E
Golgi ABL1 NM005157 v-Abl Abelson murine leukemia viral
oncogene homolog 1
ER EGFR NG007726 Epidermal growth factor receptor
ER IKBKB NM001556 Inhibitor of kappa light polypeptide gene
enhancer in B-cells, kinase beta
Golgi KIT NM000222 v-Kit Hardy-Zuckerman 4 feline sarcoma
viral oncogene homolog
HEPG2
ER EPHB2 NM017449 EPH receptor B2
Golgi KIT NM000222 v-Kit Hardy-Zuckerman 4 feline sarcoma
viral oncogene homolog
Reconstructing a Genetic Network from Gene Perturbations in Secretory. . . 77

References

1. Crombach, A., Hogeweg, P.: Evolution of evolvability in gene regulatory networks. PLoS
Comput. Biol. 4, e1000112 (2008)
2. Farhan, H., Rabouille, C.: Signalling to and from the secretory pathway J. Cell Sci. 124,
171–180 (2011)
3. Farhan, H., et al.: MAPK signaling to the early secretory pathway revealed by
kinase/phosphatase functional screening. J. Cell Biol. 189, 997–1011 (2010)
4. Khatri, P., Sirota M., Butte A.J.: Ten years of pathway analysis: current approaches and
outstanding challenges. PLoS Comput. Biol. 8, e1002375 (2012)
5. Kohl, M., Wiese, S., Warscheid, B.: Cytoscape: software for visualization and analysis of
biological networks. Methods Mol. Biol. 696, 291–303 (2011)
6. Liu, L.-Z., Wu, F.-X., Zhang, W.-J.: Reverse engineering of gene regulatory networks from
biological data. WIREs Data Min. Knowl. Discovery 2, 365–385 (2012)
7. Marbach, D., Costello, J.C., Küffner, R., et al.: Wisdom of crowds for robust gene network
inference. Nat. Methods 9, 796–804 (2012)
8. Simpson, J.C., Joggerst, B., Laketa, V., et al.: Genome-wide RNAi screening identifies human
proteins with a regulatory function in the early secretory pathway. Nat. Cell Biol. 14, 764–774
(2012)
9. Tegner, J., Yeung, M.K., Hasty, J., Collins, J.J.: Reverse engineering gene networks: integrating
genetic perturbations with dynamical modeling. Proc. Natl. Acad. Sci. U S A 100, 5944–5949
(2003)
10. Wagner, A.: How to reconstruct a large genetic network from n gene perturbations in fewer
than n2 easy steps. Bioinformatics 17, 1183–1197 (2001)
11. Wlodkowic, D., Skommer, J., McGuinness, D., Hillier, C., Darzynkiewicz, Z.: ER-Golgi
network–a future target for anti-cancer therapy. Leuk. Res. 33, 1440–1447 (2009)
Dissecting the Functions of the Secretory
Pathway by Transcriptional Profiling

Sonali Gopichand Chavan, Kumar Parijat Tripathi, Marina Piccirilo,

Prathyush Deepth Roy, Mario Guarracino,
Alberto Luini, and Seetharaman Parashuraman

Abstract The secretory pathway is responsible for biosynthesis, processing, sort-

ing, and delivery of variety of proteins encoded in the human genome to their proper
cellular location through a series of controlled events. Each step along the secretory
pathway is controlled by regulatory modules that maintain the homeostasis of the
system. Impairment in the functioning of the secretory pathway forms the basis
of several pathologies. Nevertheless, the modules that interact with the secretory
pathway and the underlying molecular circuits remain under explored, especially
in the mammalian system. In order to identify and characterize these circuits we
are implementing an approach based on the deconvolution of the transcriptional
profiles resulting from the perturbation of the secretory pathway. Such analysis
will help to detect the cellular modules that interact with secretory pathway and
thus provide insights into the regulatory pathways coordinating their activities.
Preliminary observations from these analyses indicate an interaction between the
secretory pathway and the DNA replication/repair module, an interaction that can
have potential implications for cancer.

Keywords Secretory pathway • DNA repair • GSEA

S.G. Chavan • P.D. Roy • S. Parashuraman ()

Institute of Protein Biochemistry at National Research Council (CNR), Naples, Italy
e-mail: [email protected]
K.P. Tripathi • M. Piccirilo • M. Guarracino
Laboratory for Genomics, Transcriptomics and Proteomics (LAB-GTP), High Performance
Computing and Networking Institute (ICAR) at National Research Council (CNR), Naples, Italy
A. Luini
Institute of Protein Biochemistry at National Research Council (CNR), Naples, Italy
Istituto di Ricovero e Cura a Carattere Scienti co SDN, Via Emanuele Gianturco, 113, 80143
Naples, Italy

© Springer International Publishing Switzerland 2016 79

A. Rogato et al. (eds.), Dynamics of Mathematical Models in Biology,
DOI 10.1007/978-3-319-45723-9_7
80 S.G. Chavan et al.

1 Introduction

The secretory pathway is responsible for delivery of a large variety of proteins

to their proper cellular location and is essential for cellular function and multi-
cellular development [1]. It is composed of a series of compartments that include
endoplasmic reticulum (ER), Golgi apparatus, and trans Golgi network (TGN),
through which the cargoes (protein or lipid) are transported in an orderly fashion
starting from the ER where the biosynthesis of cargoes is initiated. This is followed
by processing of cargoes at the Golgi apparatus by addition of glycan groups and
they are then sorted to their appropriate sites at the TGN [2]. At all these levels,
each step including the anterograde transport (transport from ER to the PM via
Golgi) and retrograde transport (transport in the reverse direction, but here refers
mainly to the transport from Golgi to ER) is controlled by regulatory modules that
maintain the homeostasis of the system [3]. Like every other module of the cell [4],
the secretory pathway does not work in isolation but interacts with other cellular
modules. Co-ordination circuits regulate the activities of these interacting modules
so as to maintain homeostasis of the cell.
The functions and interactions of the secretory pathway have been studied
by genome-wide RNA-mediated interference screens in Drosophila cell lines [5],
in cultured human cells [6], and also in Saccharomyces cerevisiae [7]. Altered
functioning of the secretory pathway has been associated with several pathologies,
like the involvement of GOLPH3 in cancer [8] or the involvement of secretory
pathway localized proteins in genetic diseases [9]. Nevertheless, the modules
that interact with the secretory pathway and the underlying molecular circuits
remain under explored, especially in the mammalian system. In order to identify
and characterize these circuits we are implementing an approach based on the
deconvolution of the transcriptional profiles resulting from the perturbation [10] of
the secretory pathway. Such a study would help identify modules that interact with
secretory pathway module and also provide insights into the regulatory pathways
that coordinate their activities.

2 Materials and Methods

2.1 Strategy

To identify the modules that interact with the secretory pathway, we have per-
turbed its functioning by knocking down secretory pathway localized genes (using
shRNAs), followed by an analysis of the changes in gene expression. Pathways or
functions that were modulated under these conditions were identified using gene
set enrichment analysis (GSEA). Following this, the transcription factors (TF) that
might potentially regulate these pathways or functions were identified. The TFs can
then be used to predict upstream signaling pathways that respond to the original
perturbation (knockdown of secretory pathway localized genes). This analysis
Dissecting the Functions of the Secretory Pathway by Transcriptional Profiling 81

Fig. 1 Strategy to identify interacting modules of secretory pathway and the underling molecular
circuit. Gene expression profiles obtained from cells, where the secretory pathway is perturbed by
shRNA mediated silencing of secretory pathway localized genes, will be analyzed using GSEA
to obtain pathways that are modulated by the perturbation. Then putative upstream transcription
factors that can regulate the genes associated with these pathways will be predicted and validated.
Then, literature mining coupled to experimental validation will be used to dissect the signaling
pathways that modulate the TF activity under these conditions so to build the molecular circuit
connecting perturbation to the gene expression changes

will help to map the molecular pathway connecting the perturbation of secretory
pathway function to the modulation of other specific functions of the cells. This
connection between cause and effect would help reveal the underlying molecular
circuit regulating the interacting module. This general strategy is represented in
Fig. 1.

2.2 Data Collection and Processing

The micro-array profiles following knockdown of secretory pathway genes were

obtained from Library of Integrated Cellular Signatures (LINCS) which is an
NIH program (http://www.lincscloud.org/perturbagens/) that funds the generation
of perturbation profiles across multiple cell and perturbation types. Details about
the cell types used in the study can be found here (http://www.lincscloud.org/
cell-types/). It comprises of gene expression signatures produced by using L1000
technology which is a bead-based, high-throughput gene expression array. Only
1000 landmarks genes were experimentally measured and rest were computationally
inferred. The transcriptional response was studied following the genetic and chemi-
cal perturbations. This data is computationally processed; wherein raw fluorescence
82 S.G. Chavan et al.

intensity is converted to differential gene expression signatures. Data at different

level of processing is available. Level 4 data was used in this study which represents
signatures with differentially expressed genes computed by robust z-scores for each
profile relative to population control. LINCS provides gene expression profiles for
every perturbation obtained from different cell lines with multiple biological and
technical replicates. The profiles for each perturbation of interest were subjected
to pre-processing steps wherein the cell line dependent effect was removed by
converting expression values to rank and then merging all of them into a single
Prototype Ranked List (PRL) for each perturbation. This conversion and merging
of data into PRL was done using “Gene Expression Signature Package” from
Bioconductor in R [11]. This R package uses built-in “krubor” function to carry out
rank merging process. It comprises of two steps (1) a distance is measured between
two ranked list using Spearman’s foot rule and two or more ranked lists are merged
using Borda Merging method; (2) a single ranked list is obtained in a hierarchical
way using Kruskal algorithm. Finally, all the PRLs representing the individual state
(perturbation experiments) were generated as one input for the downstream analysis.
In this study, only small set of perturbation profiles were considered in order
to standardize the system. This test set represents expression profiles following
knockdown of a selected set of secretory pathway localized genes—ARF1, COPA,
COPB2, COPZ1, COG2, COG4, COG7, M6PR, BLFZ1, GOLGA5, PLA2G4A,
YKT6, RAB1B, SAR1B, TMED7, TMED9, TMED10, SEC24B, SEC24C,
SEC24D, and BNIP1. The localization of these proteins across secretory pathway
is represented in Fig. 2.

Plasma membrane

M6PR

ARF1
COPB2 BLZF1/Golgin-45
COPZ1 GOLGA5/Golgin-84
COPA PLA2G4A
COG2 Ykt6
COG4
COG7
AKAP9 Golgi apparatus
Rab1B
TMED7
TMED9
Sar1B TMED10
Sec24B
Sec24C
Sec24D BNIP1

Endoplasmic Reticulum (ER)

Fig. 2 Localization of the genes, whose expression was perturbed, to the compartments of the
secretory pathway is represented
Dissecting the Functions of the Secretory Pathway by Transcriptional Profiling 83

2.3 Gene Set Enrichment Analysis

The PRLs generated from gene expression profiles were subjected to GSEA using
a java desktop application available at Molecular Signature Database (MsigDB;
http://www.broadinstitute.org/gsea/down-loads.jsp). Given a set of a priori anno-
tated set of genes (based on Gene ontology classifications, KEGG (Kyoto Encyclo-
pedia of Genes and Genomes) pathways, or others), GSEA determines whether this
set of genes shows statistically significant differences between two biological states
viz. perturbation vs control that are being analyzed [12]. MsigDB has a collection
of annotated gene sets (curated gene set, motif gene set, GO gene set, oncogenic
signature, immunologic signature, etc.) for use with GSEA software. In order to
study for the enrichment of the pathways, we used KEGG pathway gene set. Using
GSEA, a number of enriched pathways were predicted across all the 22 PRLs. Only
those pathways with the False Discovery Rate (FDR) cutoff 0.25 were taken into
account. It has been suggested that given the lack of coherence in most expression
datasets and the relatively small number of gene sets being analyzed, a FDR cutoff
0.25 is appropriate for the purposes of hypothesis generation [12]. We noted that
many predicted pathways had significantly overlapping set of genes. So in order to
streamline the results, the enriched pathways were consolidated into one group if
they have more than 50 % of the genes overlapping.

2.4 Transcription Factor Prediction

The upstream transcription factors that can potentially regulate the expression of the
genes belonging to the enriched pathways were predicted using the online resources
TransFind (http://transfind.sys-bio.net/) and Locamo Finder (https://sysimm.ifrec.
osakau.ac.jp/tfbs/locamo/) and HTRIDB (http://www.lbbc.ibb.unesp.br/htri). Trans-
Find and Locamo Finder predict the TFs based on their affinity towards the putative
promoters of the genes on interest. These affinities have been pre-calculated based
on the available positional frequency matrices for the transcription factors [13]. On
the other hand, prediction by HTRIDB is based on experimentally verified human
transcriptional regulation interactions. Among the TFs obtained, only those that
were commonly predicted by all these tools were selected for further analysis.

3 Result and Discussion

The gene expression profiles following shRNA mediated knockdown (KD) of

selected secretory pathway genes were downloaded from LINCS database and
processed to obtain PRLs (see methods). These PRLs were then subjected to
GSEA analysis to obtain enriched pathways/functions that were modulated under
84 S.G. Chavan et al.

Fig. 3 The perturbed genes were grouped based on the common enriched pathways. The color
code refers to downregulated pathways (in red) and upregulated pathways (in blue). Module
unrelated to secretory pathway is marked by asterisk (orange)

the perturbation conditions. The perturbations were then grouped on the basis of
the pathways that were modulated in common (Fig. 3). Most of these groups were
related to secretory pathway module viz. glycosaminoglycan (GAG) biosynthesis,
protein export, ribosome, Pantothenate and CoA biosynthesis, aminoacyl tRNA
biosynthesis, and Phe, Tyr, His biosynthesis pathway, as expected. However, the
group of COPZ1, COG4, and COG7 gene KDs was associated with the downregu-
lation of DNA repair and replication pathway, which is a function not known to be
related to the secretory pathway module. Thus GSEA analysis reveals both expected
and unexpected modules that are modulated in response to a perturbation of the
secretory pathway. COPZ1, COG4, and COG7 share a common known function
of retrograde transport from Golgi to ER. This association suggests a possible
interaction between the Golgi retrograde transport and the DNA repair response.
We then experimentally tested whether the DNA repair pathway is indeed regu-
lated by the secretory pathway localized genes. To this end, we have downregulated
COPZ1, COG4, or COG7 using siRNAs in HeLa cells, and then measured the
increase of the DNA damage by studying the changes in the levels of phospho
histone H3, a marker of the sites of DNA double strand breaks. Only the downregu-
lation of COPZ1 showed increased levels of DNA damage as shown by an increase
in the levels of phospho histone H3. Moreover, knockdown of coatomer proteins
(COPA, COPZ1) has already been showed to increase DNA damage [14]. These
findings suggest that the interaction between the modules of the secretory pathway
Dissecting the Functions of the Secretory Pathway by Transcriptional Profiling 85

Table 1 Transcription factors predicted for DNA repair cluster

Cluster Gene KD Transcription factors
DNA repair/replication COPZ1, COG4, COG7 E2F TF’s (E2F1 and E2F4)
HIF1A (Hypoxia inducible factor 1 alpha)
NRF (Nuclear respiratory factor 1)
RFX1 (MHC class II regulatory factor RFX1)
Nkx2-5 (Homeobox protein Nkx-2-5)

Table 2 Position of predicted TF for DNA

repair cluster
TF’s Probe ids Position in profile
E2F1 2028_s_at 18146
204947_at 22001
E2F4 202248_at 19056
HIF1A 200989_at 3507
NRF1 204651_at 19224
204652_s_at 21124
211279_at 16940
211280_s_at 10201
RFX1 206321_at 3081

and DNA repair that we identified is probably a true interaction and moreover
validates our strategy for identification of modules interacting with the secretory
pathway.
We then analyzed the genes belonging to the DNA repair pathway that is
modulated by the perturbation of Golgi retrograde transport (COPZ1, COG4, or
COG7 KD), to identify the putative TFs that can regulate their expression. The
enriched TFs obtained for this DNA repair group are listed in Table 1. Among these,
E2F1, E2F4, NRF1, and RFX1 are known to be involved in regulation of DNA repair
pathway genes of which E2F4 and RFX1 act as repressors [15, 16].
Since transcription factors are usually co-expressed along with their target genes,
their position across the PRL (rank in the PRL) associated with COPZ1 KD was
analyzed (Table 2), in order to restrict the TFs to those that are more likely to
be the true effectors under our perturbation conditions. This analysis revealed
that transcription factors E2F1, NRF1, and E2F4 are probably downregulated
and HIF1A and RFX1 are probably upregulated. However, only the behavior
of E2F1 (activator), NRF1 (activator), and RFX1 (repressor) are in concordant
with the observed effect of the target genes, i.e., downregulation of DNA repair
pathways (Fig. 4). This TF information can be used to map the upstream signaling
pathways that connect DNA repair pathway to perturbation of COPZ1 expression
(or impaired retrograde transport) by further analysis using online resources as well
as experimental validation studies.
86 S.G. Chavan et al.

Fig. 4 Hypothetical model of Golgi retrograde transport mediated by COPZ1 possibly regulating
the DNA repair pathway. The predicted TFs that might be involved in this regulation are indicated
and the direction of their modulation (up- or downregulation) under conditions of COPZ1 KD is
indicated by colored arrows. The known activity of the TF as a transcriptional activator or repressor
is indicated by the color coding of the text. (Refer to the key for details)

3.1 Conclusion

The transcriptional profiling following the perturbation of genes localized to the

secretory pathway shows a modulation in the levels of genes associated with path-
ways involved in the secretory transport as well as those unrelated to the secretory
pathway. This study revealed an interesting possibility wherein the perturbation
of secretory pathway function, particularly Golgi retrograde transport, might lead
to the downregulation of DNA repair pathway. However, these predictions need
to be experimentally validated and characterized. Since DNA repair pathways
are associated with genome stability, understanding such interactions would be
important for cancer studies. Moreover, we wish to extend this strategy of analysis
for all the genes of secretory pathway in order to identify their novel functions.

References

1. Zhong, W.: Golgi during development. Cold Spring Harb. Perspect. Biol. 3(9), a005363 (2011)
2. Kelly, R.B.: Pathways of protein secretion in eukaryotes. Science 230(4721), 25–32 (1985)
3. Luini, A., Mavelli, G., Jung, J., Cancino, J.: Control systems and coordination protocols of the
secretory pathway. F1000prime reports 6 (2014)
4. Costanzo, M., Baryshnikova, A., Bellay, J., Kim, Y., Spear, E.D., Sevier, C.S., Ding, H., Koh,
J.L., Toufighi, K., Mostafavi, S., Prinz, J.: The genetic landscape of a cell. Science 327(5964),
425–431 (2010)
Dissecting the Functions of the Secretory Pathway by Transcriptional Profiling 87

5. Bard, F., Casano, L., Mallabiabarrena, A., Wallace, E., Saito, K., Kitayama, H., Guizzunti, G.,
Hu, Y., Wendler, F., DasGupta, R., Perrimon, N.: Functional genomics reveals genes involved
in protein secretion and Golgi organization. Nature 439(7076), 604–607 (2006)
6. Simpson, J.C., Joggerst, B., Laketa, V., Verissimo, F., Cetin, C., Erfle, H., Bexiga, M.G.,
Singan, V.R., Hrich, J.K., Neumann, B., Mateos, A.: Genome-wide RNAi screening identifies
human proteins with a regulatory function in the early secretory pathway. Nat. Cell Biol. 14(7),
764–774 (2012)
7. Jonikas, M.C., Collins, S.R., Denic, V., Oh, E., Quan, E.M., Schmid, V., Weibezahn, J.,
Schwappach, B., Walter, P., Weissman, J.S., Schuldiner, M.: Comprehensive characterization
of genes required for protein folding in the endoplasmic reticulum. Science 323(5922), 1693–
1697 (2009)
8. Scott, K.L., Kabbarah, O., Liang, M.C., Ivanova, E., Anagnostou, V., Wu, J., Dhakal, S., Wu,
M., Chen, S., Feinberg, T., Huang, J.: GOLPH3 modulates mTOR signalling and rapamycin
sensitivity in cancer. Nature 459(7250), 1085–1090 (2009)
9. De Matteis, M.A., Luini, A.: Mendelian disorders of membrane trafficking. N. Engl. J. Med.
365(10), 927–938 (2011)
10. Ideker, T., Thorsson, V., Ranish, J.A., Christmas, R., Buhler, J., Eng, J.K., Bumgarner, R.,
Goodlett, D.R., Aebersold, R., Hood, L.: Integrated genomic and proteomic analyses of a
systematically perturbed metabolic network. Science 292(5518), 929–934 (2001)
11. Li, F., Cao, Y., Han, L., Cui, X., Xie, D., Wang, S., Bo, X.: GeneExpressionSignature: an
R package for discovering functional connections using gene expression signatures. OMICS
17(2), 116–118 (2013)
12. Subramanian, A., Tamayo, P., Mootha, V.K., Mukherjee, S., Ebert, B.L., Gillette, M.A.,
Paulovich, A., Pomeroy, S.L., Golub, T.R., Lander, E.S., Mesirov, J.P.: Gene set enrichment
analysis: a knowledge-based approach for interpreting genome-wide expression profiles. Proc.
Natl. Acad. Sci. U. S. A. 102(43), 15545–15550 (2005)
13. Roider, H.G., Kanhere, A., Manke, T., Vingron, M.: Predicting transcription factor affinities to
DNA from a biophysical model. Bioinformatics 23(2), 134–141 (2007)
14. Paulsen, R.D., Soni, D.V., Wollman, R., Hahn, A.T., Yee, M.C., Guan, A., Hesley, J.A., Miller,
S.C., Cromwell, E.F., Solow-Cordero, D.E., Meyer, T.: A genome-wide siRNA screen reveals
diverse cellular processes and pathways that mediate genome stability. Mol. Cell 35(2), 228–
239 (2009)
15. Verona, R., Moberg, K., Estes, S., Starz, M., Vernon, J.P., Lees, J.A.: E2F activity is regulated
by cell cycle-dependent changes in subcellular localization. Mol. Cell. Biol. 17(12), 7268–7282
(1997)
16. Lubelsky, Y., Reuven, N., Shaul, Y.: Autorepression of rfx1 gene expression: functional
conservation from yeast to humans in response to DNA replication arrest. Mol. Cell. Biol.
25(23), 10665–10673 (2005)
Detection of Rare Mutations Using
Beta-Binomial and Empirical Quantile Models
in Next-Generation Sequencing Experiments

Sarunas Germanas, Audrone Jakaitiene, and Mario Guarracino

Abstract Next-generation sequencing is often used to identify genetic variants.

The probability of variant detection also depends on the variant caller. Pooled data
could be used to lower the sequencing cost. However identifying variants from
pooled data is more challenging and demands more sophisticated mathematical
methods. In this article we propose two novel SNP calling approaches: modification
of Beta-binomial model as proposed in Flaherty et al. (2011) using posterior Beta
distribution and empirical quantile method. Both offered methods and original Beta-
binomial model were applied to pooled exome data of patients with neuromuscular
diseases. The results showed that Beta-binomial model and modification of it were
highly specific, however, with lower sensitivity compared to empirical quantile
model. The positions could be identified as mutated using empirical quantile model
much faster rather Beta-binomial models.

Keywords Variant calling • Pooling • NGS • Empirical quantile • Beta-

binomial

1 Introduction

New generation sequencing (NGS) approach has revolutionized our possibilities to

do a genetic and genomic research [7]. This approach provides cheap and fast way
to sequence genetic data. NGS is used for de novo sequencing, for disease mapping,

S. Germanas
Institute of Mathematics and Informatics, Vilnius university, LT-08663 Vilnius, Lithuania
e-mail: [email protected]
A. Jakaitiene ()
Department of Human and Medical Genetics, Faculty of Medicine, Vilnius University,
LT-01513 Vilnius, Lithuania
e-mail: [email protected]
M. Guarracino
Laboratory for Genomics, Transcriptomics and Proteomics (Lab-GTP), High Performance
Computing and Networking Institute (ICAR), National Research Council (CNR),
Via Pietro Castellino 111, Naples, Italy
e-mail: [email protected]

© Springer International Publishing Switzerland 2016 89

A. Rogato et al. (eds.), Dynamics of Mathematical Models in Biology,
DOI 10.1007/978-3-319-45723-9_8
90 S. Germanas et al.

for variant calling, and for diagnostics [4, 8]. NGS suffers from high error rates
which come from several error sources including base-calling and alignment.
During the single nucleotide polymorphism (SNP) calling process using NGS
the variable sites of genome could be identified. Sophisticated SNP calling math-
ematical models could be used to reduce and quantify the uncertainty of variant
calling process caused by high error rates. Another way could be target-sequencing
of certain genetic region with higher sequencing rate (20x and more). However
increasing demand of large samples suggests that high-depth sequencing is too
expensive in time and cost. In such large sample cases alternative way of sequencing
could be grouping of patients to pools and sequencing them together. This strategy
gives a possibility to sequence more effective in terms of time and money. Although,
this strategy also has a drawback—the allele frequency of certain individual from the
pool cannot be estimated directly. The same applies to SNP and genotype calling.
Therefore even more sophisticated mathematical methods and pooling strategies
must be used in order to take advantage of pooled NGS data.
There are many SNP calling methods for pooled NGS data [1–3, 5, 6, 11, 13, 15].
One common property of these methods is that the non-referent allele frequency riv
is modeled, where i D 1; : : : ; N is genetic position, v D 1; : : : ; J is the index of a
pool. In [5, 6] hierarchical Beta-binomial model is offered and applied to synthetic
pooled genetic data of virus. The quantity riv is assumed to have Beta-binomial
distribution. In [3, 11, 15] random variable riv is assumed to have hierarchical
binomial–binomial distribution. In models of [1, 2] statistically significant differ-
ences of riv between pools are searched to identify genetic variants. In [9, 14]
sequencing error riv is modeled through genetic positions. This gives a benefit
depending on the number of genetic positions, which often is very large. Although
the sequencing error rate () which is assumed constant in these models could be
very different in different genetic loci. Applications of the methods mentioned above
to real data with high minor allele frequency (>5 %) show that methods are quite
sensitive and specific, but the rare event case is not clear or not sufficient [10, 12].
We offer two novel approaches for detecting SNPs in pooled NGS data. The first
model is modification of Beta-binomial model [5], when the posterior distribution of
riv is considered. In the second method we model the function of random variables
riv
niv
with different v and use empirical quantile to detect the SNP calling threshold.
Also, we use binomial approximation of riv to model the data in order to choose
significant value of empirical quantile.
We use pooled NGS data from 128 patients with diagnosis of neuromuscular
disease for SNP identification. Results show that the modification of Beta-binomial
model detects variants with almost 100 % specificity. However the empirical
quantile model has better sensitivity and is much faster compared to Beta-binomial
models.
Detection of Rare Mutations Using Beta-Binomial and Empirical Quantile. . . 91

2 Materials and Methods

2.1 Modification of Beta-Binomial Model

We propose the first SNP calling model which is a modification of Beta-binomial

model. The original Beta-binomial model was proposed in [5]:

ri;kj .s/ ji;kj .s/ Binomial.i;kj .s/ ; ni;kj .s/ /; (1)
i;kj .s/ Beta. i ; /; (2)

where ri;kj .s/ is non-referent allele frequency (observed), ni;kj .s/ is read depth
(observed), i;kj .s/ is error rate parameter at position i D 1; : : : ; N in pool k.s/ D
k1 .s/; : : : ; kJ .s/, i (expected value of Beta distribution) and  (precision of Beta
distribution) are hyperparameters of the model; where kj D kj .s/ is a function which
maps from the model set-up s to the kj -th reference pool, j D 1; : : : ; J. Parameter
i;kj .s/ and hyperparameters i ;  are estimated using Expectation-Maximization
algorithm for the whole likelihood function:

Y
N Y
J
L.ri;kj .s/ ; i;kj .s/ j i ; / D Pr.ri;kj .s/ ji;kj .s/ ; ni;kj .s/ /Pr.i;kj .s/ j i ; / (3)
iD1 jD1

l.ri;kj .s/ ; i;kj .s/ j i ; / D ln L.ri;kj .s/ ; i;kj .s/ j i ; /; (4)

Distribution of ri;kj .s/ which is assumed to be Beta-binomial is approximated with

normal distribution with estimate of i and standard deviation i D i .1 i/
.1 C
ni0 1 1 P J
ni0

1
/; ni0 D J jD1 ni;kj .s/ , and distribution of reference data is modeled. Z-test is
applied for a main pool data.
We propose modification of Beta-binomial model. We use posterior expectation
of i;kj .s/ instead of prior:
P
i  C JjD1 ri;kj .s/
E.i;kj .s/ jri;kj .s/ / D post;i D P : (5)
 C JjD1 ni;kj .s/

Therefore we use more information from the data and expect to get more accurate
estimates of i and i;kj .s/ . Estimate of standard deviation i remains the same, but
also the posterior standard deviation could be used. Z-test is applied for the case
data.
We apply original Beta-binomial model and modification of it for set-ups of
pooled data described in Sect. 2.3. Significance value for Z-test is chosen ˛ D 106 .
92 S. Germanas et al.

2.2 Empirical Quantile Model

In this section we present another SNP calling method as empirical quantile method.
The idea of this method is do not use any theoretical distribution when predicting
mutated positions and to use the data across all positions as it was applied in [9, 14].
We introduce the function f W Œ0; 1JC1 ! ŒJ; 1 (J is number of reference
pools):

X
J
ysi D f .Mi;l.s/ ; Ri;kj .s/ / D Mi;l.s/  Ri;kj .s/ ; (6)
jD1

ri;l.s/
where Mi;l.s/ D ni;l.s/
is relative frequency of non-referent allele of main pool and
ri;kj .s/
Ri;kj .s/ D ni;kj .s/
is relative frequency of non-referent allele of reference pool in i-th
position, kj .s/-th pool and s-th data set-up for a model; function l D l.s/ maps from
model set-up s to the l-th main pool.
We expect that the value ysi is higher for mutated positions and lower for non-
mutated positions. Therefore we use the empirical quantile q˛ :

q˛ W Pr.ysi > q˛ / D ˛: (7)

Identification of mutated positions will depend on selection of ˛. For ˛ estimation

we model ysi as a sum of Bernoulli random variables. We assume that a pool is
contributed by patients equally as Bernoulli random variables with the constant
success probability which differs only between the main and reference pools.

X
Q
Mi;l.s/ D c D WpM Binomial.Q; pM /;
BM (8)
cD1

X
J X
J X
Q
Ri;kj .s/ D BRc D WpR Binomial.QJ; pR /; (9)
jD1 jD1 cD1

1 PS PN ri;l.s/
where BM c Bernoulli.pM /,Bc Bernoulli.pR /, pM D SN
R
sD1 iD1 ni;l.s/ and
1 PS PN PJ ri;kj .s/
pR D SJN sD1 iD1 jD1 ni;k .s/ are, respectively, estimated error rate of the main
j
and reference pools, c D 1; : : : ; Q is the number of a patient in a pool, Q is assumed
to be constant, and S is a number of data set-ups.
We do not use any prior information about mutation status of the positions and
model distribution of ysi in three cases:
1. There is no information about position i (neither confirmed as mutated nor
confirmed as non-mutated). ysi distribution using convolution formula is
Detection of Rare Mutations Using Beta-Binomial and Empirical Quantile. . . 93

X
Q
P.ysi D tji is general/ D PWRgen .h  t/PWMgen .h/; (10)
hD1

gen gen gen gen gen gen

where WR Binomial.QJ; pR /, WM Binomial.QJ; pM /, pR and pR are,
respectively, error rates of main and reference pools estimated from all data.
2. Position i is mutated. For this case one pair of individuals has at least one
mutation. We assume that this mutation is heterozygous and is present in both
individuals. Therefore Mi;l.s/ D 1 and 1  Ri;kj .s/ Binomial.QJ  Q C 2; ptest
R /.

P.ysi D tji is mutated/ D PWRtest .t  1/; (11)

where WRtest Binomial.QJ  Q C 2; ptest R /, pR is error rate of main pools

test

estimated from positions for which method is tested.

3. Position i is not mutated. For this case the pair of individuals has no mutation
and Mi;l.s/ D 0.

P.ysi D tji is not mutated/ D PWRtest .t/: (12)

In the second and the third case we use quantity QJQC2 instead of QJ because we
assume that part of main pool which is present in reference pools was canceled out
(see Table 1). Modeled sensitivity and specificity are computed using (11) and (12)
accordingly. Having calculated modeled sensitivity and specificity for different ysi ,
we determine the value of ysi and compute ˛ from general distribution expressed
in (10). Finally, for the assessment of the model, we calculate sensitivity and
specificity using positions checked with Sanger.

2.3 Data Organization

We use pooled data from 128 patients with neuromuscular disease to identify
mutated variants. Target exome regions were sequenced using Illumina sequencing
platform. The target region consists of approximately 13,000 position with relative
frequency of non-referent allele Mi;l.s/ varying from 0.01 to 0.06. Data consists of 8
original pools where each pool has 16 patients and 8 replicated pools which consist
from the same 128 patients but with different pool composition (Table 1). For the
models described above we used different organization of main and replicated pools
as it is presented in Table 2.
Every pool from the original pool group was taken as main pool together with
7 pools from the replicated pool group as reference pools in such a way that every
pair of patients from the main pool was not present in the reference pools. Every
such combination of one main pool and seven reference pools we denote s, where
s D 1; : : : ; 64, and call data set-up for the model in the paper. Positions in every data
set-up were filtered according to main pool—positions with Mi;l.s/ < 0:011 where
94 S. Germanas et al.

Table 1 Organization of pools

Original pools (1–8)
Pool P9 P9 P10 P10 P11 P11 P12 P12 P13 P13 P14 P14 P15 P15 P16 P16
P1 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16
P2 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32
P3 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48
P4 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64
P5 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80
P6 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96
P7 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112
P8 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128
Replicated pools (9–16)
Pool P1 P1 P2 P2 P3 P3 P4 P4 P5 P5 P6 P6 P7 P7 P8 P8
P9 1 2 17 18 33 34 49 50 65 66 81 82 97 98 113 114
P10 3 4 19 20 35 36 51 52 67 68 83 84 99 100 115 116
P11 5 6 21 22 37 38 53 54 69 70 85 86 101 102 117 118
P12 7 8 23 24 39 40 55 56 71 72 87 88 103 104 119 120
P13 9 10 25 26 41 42 57 58 73 74 89 90 105 106 121 122
P14 11 12 27 28 43 44 59 60 75 76 91 92 107 108 123 124
P15 13 14 29 30 45 46 61 62 77 78 93 94 109 110 125 126
P16 15 16 31 32 47 48 63 64 79 80 95 96 111 112 127 128

not considered, because of the reasoning in [4]: when Mi;l.s/ < 0:011 there cannot
be any mutation because of finite number (16) of individuals in pool.
We have a list of mutated and non-mutated positions confirmed using Sanger
sequencing which gives the possibility of golden standard for calculation of
sensitivity and specificity.

3 Results

We apply Beta-binomial, modified Beta-binomial, and empirical quantile methods

to pooled data from patients with neuromuscular diseases. For each model we
use data set-up as in Table 2. For Beta-binomial model we model distribution of
reference pools and apply Z-test for main pool for every model set-up.
An illustration of empirical quantile method is in Fig. 1 where values of ysi for
all sequenced, mutated, and non-mutated positions are plotted. As we expected, the
value ysi is positive for mutated positions and negative for non-mutated positions.
This indicates that if we observe mutation in specific position, the error rate of
the main pool is larger as error rate sum of the same position in reference pools.
Therefore for the detection of mutated position, we need to find a threshold above
which all positions would be detected as mutated only. For the determination
Detection of Rare Mutations Using Beta-Binomial and Empirical Quantile. . . 95

Table 2 Organization of pools: for every original pool eight combinations of replicated pools
Model Main Model Main
set-up pools, set-up pools,
s l.s/ Reference pools, k.s/ Patient s l.s/ Reference pools, k.s/ Patient
1 10, 11, 12, 13, 14, 15, 16 1 2 33 10, 11, 12, 13, 14, 15, 16 65 66
2 9, 11, 12, 13, 14, 15, 16 3 4 34 9, 11, 12, 13, 14, 15, 16 67 68
3 9, 10, 12, 13, 14, 15, 16 5 6 35 9, 10, 12, 13, 14, 15, 16 69 70
4 9, 10, 11, 13, 14, 15, 16 7 8 36 9, 10, 11, 13, 14, 15, 16 71 72
1 5
5 9, 10, 11, 12, 14, 15, 16 9 10 37 9, 10, 11, 12, 14, 15, 16 73 74
6 9, 10, 11, 12, 13, 15, 16 11 12 38 9, 10, 11, 12, 13, 15, 16 75 76
7 9, 10, 11, 12, 13, 14, 16 13 14 39 9, 10, 11, 12, 13, 14, 16 77 78
8 9, 10, 11, 12, 13, 14, 15 15 16 40 9, 10, 11, 12, 13, 14, 16 79 80
9 10, 11, 12, 13, 14, 15, 16 17 18 41 10, 11, 12, 13, 14, 15, 16 81 82
10 9, 11, 12, 13, 14, 15, 16 19 20 42 9, 11, 12, 13, 14, 15, 16 83 84
11 9, 10, 12, 13, 14, 15, 16 21 22 43 9, 10, 12, 13, 14, 15, 16 85 86
12 9, 10, 11, 13, 14, 15, 16 23 24 44 9, 10, 11, 13, 14, 15, 16 87 88
2 6
13 9, 10, 11, 12, 14, 15, 16 25 26 45 9, 10, 11, 12, 14, 15, 16 89 90
14 9, 10, 11, 12, 13, 15, 16 27 28 46 9, 10, 11, 12, 13, 15, 16 91 92
15 9, 10, 11, 12, 13, 14, 16 29 30 47 9, 10, 11, 12, 13, 14, 16 93 94
16 9, 10, 11, 12, 13, 14, 15 31 32 48 9, 10, 11, 12, 13, 14, 16 95 96
17 10, 11, 12, 13, 14, 15, 16 33 34 49 10, 11, 12, 13, 14, 15, 16 97 98
18 9, 11, 12, 13, 14, 15, 16 35 36 50 9, 11, 12, 13, 14, 15, 16 99 100
19 9, 10, 12, 13, 14, 15, 16 37 38 51 9, 10, 12, 13, 14, 15, 16 101 102
20 9, 10, 11, 13, 14, 15, 16 39 40 52 9, 10, 11, 13, 14, 15, 16 103 104
3 7
21 9, 10, 11, 12, 14, 15, 16 41 42 53 9, 10, 11, 12, 14, 15, 16 105 106
22 9, 10, 11, 12, 13, 15, 16 43 44 54 9, 10, 11, 12, 13, 15, 16 107 108
23 9, 10, 11, 12, 13, 14, 16 45 46 55 9, 10, 11, 12, 13, 14, 16 109 110
24 9, 10, 11, 12, 13, 14, 15 47 48 56 9, 10, 11, 12, 13, 14, 16 111 112
25 10, 11, 12, 13, 14, 15, 16 49 50 57 10, 11, 12, 13, 14, 15, 16 113 114
26 9, 11, 12, 13, 14, 15, 16 51 52 58 9, 11, 12, 13, 14, 15, 16 115 116
27 9, 10, 12, 13, 14, 15, 16 53 54 59 9, 10, 12, 13, 14, 15, 16 117 118
28 9, 10, 11, 13, 14, 15, 16 55 56 60 9, 10, 11, 13, 14, 15, 16 119 120
4 8
29 9, 10, 11, 12, 14, 15, 16 57 58 61 9, 10, 11, 12, 14, 15, 16 121 122
30 9, 10, 11, 12, 13, 15, 16 59 60 62 9, 10, 11, 12, 13, 15, 16 123 124
31 9, 10, 11, 12, 13, 14, 16 61 62 63 9, 10, 11, 12, 13, 14, 16 125 126
32 9, 10, 11, 12, 13, 14, 15 63 64 64 9, 10, 11, 12, 13, 14, 16 127 128

of the significance value for empirical quantile method, we approximate riv by

gen gen test
binomial distribution and compute Mi;l.s/ ; pM ; ptest
M for main and Ri;k.s/ ; pR ; pR
s
for reference pools. We model distributions of yi for general, mutated, and non-
mutated positions. The latter distributions for all data set-ups are plotted in Fig. 2.
We consider ysi0 at integer values fJQ; JQ C 1; : : : ; 0; 1; 2; : : : Qg as possible
threshold points. Specifically, we compute ˛ D P.ysi  ysi0 ji is general), modeled
96 S. Germanas et al.

0.06
General positions
Mutated positions
Nonmutated positions
0.04 SNP Calling treshold

0.02

−0.02

−0.04

−0.06

−0.08

−0.1
0 0.5 1 1.5 2 2.5
x 108

Fig. 1 In X axis all sequenced positions are represented, in Y axis values of ysi for model set-up
s D 1 are represented
1.0

General position distribution

Mutated position distribution
Nonmutated position distribution
Threshold point
0.8
0.6
0.4
0.2
0.0

−8 −7 −6 −5 −4 −3 −2 −1 0 1 2 3

Fig. 2 In X axis values of ysi are represented, in Y axis modeled probabilities of ysi values for
general, mutated, and non-mutated positions
Detection of Rare Mutations Using Beta-Binomial and Empirical Quantile. . . 97

Table 3 Sensitivity and specificity of the methods at different significance levels

˛ D 106 ˛ D 103 ˛ D 0:03
Sensitivity Specificity Sensitivity Specificity Sensitivity Specificity
Models (%) (%) (%) (%) (%) (%)
Beta-binomial 4.2 100 20.3 100 46.3 97.1
Modified 5.9 100 35.6 100 69.5 97.1
Beta-binomial
Empirical quantile 0 100 48.3 100 95.7 82.3

sensitivityD P.ysi  ysi0 ji is mutated), modeled specificityD P.ysi < ysi0 ji is mutated)
for every value ysi0 . Our objective is to find ysi0 for which both modeled specificity
and sensitivity would be as close to 1. We obtained that only at ysi0 D 1 both modeled
sensitivity and specificity were larger than 0.7, i.e., 0.73, 1 and ˛  0:03. We
use computed value ˛ D 0:03 in (7) to compute empirical quantile q˛ . q˛ is the
threshold to distinguish between mutated and non-mutated positions.
For the model performance evaluation and comparison, we calculated sensitivity
and specificity of Beta-binomial model, modification of Beta-binomial model, and
empirical quantile method using positions checked with Sanger sequencing. As
there is some methodological differences in significance value selection (˛ is
selected for Beta-binomial models and estimated for empirical quantile method),
we present sensitivity and specificity results at three different levels (see Table 3).
˛ D 106 was used in [5] to account for of multiple testing and ˛ D 0:03 was
estimated in empirical quantile method. Empirical quantile method gives better
sensitivity for ˛ equal 103 and 0:03. Additional advantage of empirical quantile
method is speed. It takes approximately 4–5 s estimate mutated positions of all
individuals. While the time for the implementation of Beta-binomial model is
approximately 1 week. Therefore, we can conclude that empirical quantile method
is applicable for detection of mutated positions in pooled NGS experiments.
However, empirical quantile method might be extended in several ways, as
there are some strong assumptions made: (1) model parameters are not position
dependent; (2) contributions of individuals into pools are equal; (3) sequencing error
is not modeled; (4) read errors are independent between pools and positions; (5)
selection of ˛ must be done by researcher; (6) the method depends on the experiment
structure; and (7) it was not considered in the model that observed frequencies in
pools which have leastwise one common individual are statistically dependent.
Mentioned assumptions could be relaxed when, for example, Poisson-Binomial
distribution instead of Binomial would be considered, dependence between pools
and positions would be taken into account, weighted sums instead of sums would
be calculated, selection of ˛ would be automated.
There are several articles in which error rate across positions of target region
is modeled. In [9] empirical quantile is computed with prescribed ˛ and Poisson
distribution is used to compute probability of SNP. In [14] also Poisson distribution
is assumed as read error distribution, parameter of Poisson distribution is calculated
98 S. Germanas et al.

from average error rate across positions, and threshold equal to 0.001 is used for
SNP calling. These methods have limitations that distribution of read errors is
assumed, parameter of Poisson distribution is assumed constant across positions,
and SNP calling threshold is chosen without knowledge about analyzed data.
Therefore the ability for selecting SNP calling threshold is advantage of proposed
empirical quantile method.

4 Concluding Remarks

Pooled NGS experiments demand non-standard tools to accurately discover SNPs,

therefore empirical quantile method could be appropriate, as it detects mutations
with high sensitivity and specificity very fast. This method could be extended in
many ways, for example, Poisson-Binomial distribution instead of Binomial could
be used or dependence between pools and positions could be considered. These
extensions could make the method even more efficient.

References

1. Altmann, A., Weber, P., Quast, C., Rex-Haffner, M., Binder, E.B., Müller-Myhsok, B.: vipR:
variant identification in pooled DNA using R. Bioinformatics 27, 77–84 (2011)
2. Bansal, V.: A statistical method for the detection of variants from next-generation resequencing
of DNA pools. Bioinformatics 26, 318–324 (2010)
3. Chen, Q., Sun, F.: A unified approach for allele frequency estimation, SNP detection and
association studies based on pooled sequencing data using EM algorithms. BMC Genomics
14, 1–14 (2013)
4. Ferraro, M.B., Savarese, M., di Fruscio, G., Nigro, V., Guarracino, M.R.: Prediction of
rare single-nucleotide causative mutations for muscular diseases in pooled next-generation
sequencing experiments. J. Comput. Biol. 21, 665–675 (2014)
5. Flaherty, P., Natsoulis, G., Muralidharan, O., Winters, M., Buenrostro, J., Bell, J., Brown,
S., Holodniy, M., Zhang, N., Ji, H.P.: Ultrasensitive detection of rare mutations using next-
generation targeted resequencing. Nucleic Acids Res. 40, 861–872 (2011)
6. He, Y., Zhang, F., Flaherty, P.: RVD2: an ultra-sensitive variant detection model for low-depth
heterogeneous next-generation sequencing data. Bioinformatics 31–17, 2785–2793 (2015)
7. Mardis, E.R.: A decade’s perspective on DNA sequencing technology. Nature 470, 198–203
(2011)
8. Nielsen, R., Paul, J.S., Albrechtsen, A., Song, Y.S.: Genotype and SNP calling from next-
generation sequencing data. Nat. Rev. Genet. 12, 443–451 (2011)
9. Out, A.A., van Minderhout, I.J.H.M., Goeman, J.J., Ariyurek, Y., Ossowski, S., Schneeberger,
K., Weigel, D., van Galen, M., Taschner, P.E.M., Tops, C.M.J., Breuning, M.H., van Ommen,
G.-J.B., den Dunnen, J.T., Devilee, P., Hes, F.J.: Deep sequencing to reveal new variants in
pooled DNA samples. Hum. Mutat. 9, 1703–1712 (2009)
10. Pabinger, S., Dander, A., Fischer, M., Snajder, R., Sperk, M., Efremova, M., Krabichler, B.,
Speicher, M.R., Zschocke, J., Trajanoski, Z.: A survey of tools for variant analysis of next-
generation genome sequencing data. Brief. Bioinform. 15–2, 256–278 (2012)
Detection of Rare Mutations Using Beta-Binomial and Empirical Quantile. . . 99

11. Raineri, E., Ferretti, L., Esteve-Codina, A., Nevado, B., Heath, S.: SNP calling by sequencing
pooled samples. BMC Bioinf. 13, 239–246 (2012)
12. Spencer, D.H., Tyagi, M.M., Vallania, F., Bredemeyer, A.J., Pfeifer, J.D., Mitra, R.D.,
Duncavage, E.J.: Performance of common analysis methods for detecting low-frequency single
nucleotide variants in targeted next-generation sequence data. J. Mol. Diagn. 16, 75–88 (2014)
13. Vallania, F.L.M., Druley, T.E., Ramos, E., Wang, J., Borecki, I., Province, M., Mitra, R.D.:
Quantification of rare allelic variants from pooled genomic DNA. Nat. Methods 6, 263–265
(2009)
14. Wang, C., Mitsuya, Y., Gharizadeh, B., Ronaghi, M., Shafer, R.W.: Characterization of muta-
tion spectra with ultra-deep pyrosequencing: application to HIV-1 drug resistance. Genome
Res. 17, 1195–1201 (2007)
15. Wei, Z., Wang, W., Hu, P., Lyon, G.J., Hakonarson, H.: SNVer: a statistical tool for variant
calling in analysis of pooled or individual next-generation sequencing data. Nucleic Acids
Res. 39, 132–144 (2011)
An Overview of Genotyping by Sequencing
in Crop Species and Its Application in Pepper

Francesca Taranto, Nunzio D’Agostino, and Pasquale Tripodi

Abstract The exploitation of genetic variation in crops is essential to establish

innovative breeding programs in the frame of global population increase and the
sustainable intensification of agriculture. The advent of next generation sequencing
technologies and the availability of complete or draft genome sequences of many
crops allowed the development of several methods for SNP discovery. Genotyping
by sequencing (GBS) has recently emerged as a promising approach to simulta-
neously allow SNP identification and genotyping. GBS provides a rapid, highly
informative, high-throughput and cost-effective tool for exploring plant genetic
diversity on a genome-wide scale and does not require any a priori knowledge on
the genome of the species of interest. The features of GBS make it an attractive
technology for (1) the assessment of population structure of germplasm collections;
(2) the development of high density linkage maps and (3) genetic mapping studies.
Herein, we present an overview of the GBS method and describe the main protocols
in use, the principal methods for genetic diversity analysis and potential applications
of the results in crop breeding programs. Finally, we illustrate the strategy we
adopted to investigate the genetic diversity in cultivated pepper (Capsicum annuum).

Keywords Genotyping by sequencing • SNP markers • Genetic diversity • Pop-

ulation structure • Pepper

1 Introduction

The accessibility and use of natural genetic variation in plant breeding is currently
restricted due to gaps in the genetic information that limit the comparison of
germplasm accessions of different crops. The generation of novel varieties and
the establishment of innovative breeding programs play a crucial role in food
security and nutrition. In the last century, breeding programs have led to the
selection of a small number of cultivars carrying genes for resistance to diseases

F. Taranto • N. D’Agostino • P. Tripodi ()

Consiglio per la ricerca in agricoltura e l’analisi dell’economia agraria (CREA) – Centro di
ricerca per l’orticoltura, Via dei Cavalleggeri 25, Pontecagnano Faiano (SA), Italy
e-mail: [email protected]

© Springer International Publishing Switzerland 2016 101

A. Rogato et al. (eds.), Dynamics of Mathematical Models in Biology,
DOI 10.1007/978-3-319-45723-9_9
102 F. Taranto et al.

and pests with higher and more uniform yield [1], implicating a reduction of
genetic diversity. In order to contrast this trend, international efforts are focusing
on the recovery, protection and assessment of biodiversity and on the promotion
of the sustainable use of plant genetic resources. Plant collections are constituted
over time using both locally ecotypes, selected on the basis of a recognizable
phenotype, and well-adapted crops selected for their fitness in climate change-
affected production systems. The use of wild relatives and under-utilized varieties
are instead challenging due to their unexplored genetic potentiality.
Crop improvement programs reaped the benefits from cutting-edge technolo-
gies in biological science, particularly in form of molecular markers, which in
combination with conventional phenotype-based selection, define modern plant
breeding practices. Molecular markers are extremely useful in plants to characterize
germplasm collections and improve the conventional plant breeding schemes
through marked-assisted selection (MAS). Different molecular markers have been
successfully applied for genetic mapping [2], to infer phylogenetic relationships
[3, 4], for the development of mapped genetic resources [5, 6] and comparative
studies [7, 8].
Among various types of markers in use, single nucleotide polymorphisms (SNPs)
are abundant in plant genomes; however, before the advent of next generation
sequencing (NGS) technologies, they were considered costly for application in plant
breeding [9, 10]. NGS is used for both whole genome sequencing and re-sequencing
projects, leading to the discovery of a large number of SNPs useful to explore inter-
and intra-species nucleotide diversity. As a consequence, SNPs have become the
primary choice for many genetic studies thanks to their flexibility, speed and cost-
effectiveness [11] inducing plant breeders to use them in their programs.
Genotyping by sequencing (GBS) has recently emerged as an innovative genomic
approach for exploring plant genetic diversity on a genome-wide scale [12, 13].
GBS is based on genome reduction with restriction enzymes; it does not require a
reference genome for SNP discovery and provides a rapid, high-throughput and
cost-effective tool for the investigation of genetic variability in model and non-
model species. Herewith it is provided an overview of the GBS method through the
description of the main protocols in use and their applications in plants. In addition,
research activity on the investigation of the genetic diversity in cultivated pepper
(Capsicum annuum) is illustrated.

2 Why Genotyping by sequencing?

As mentioned above, a deep assessment of the available genetic variability within

a crop is a necessary condition for a plant geneticist prior to plan a genetic
improvement program. Moreover, the association between genetic variation and
phenotypes of interest is the basis for MAS. The possibility to combine the
processes of marker discovery and genotyping with a high-throughput and low-cost
technology is the main achievement of GBS.
An Overview of Genotyping by Sequencing in Crop Species and Its Application. . . 103

GBS was first introduced in plant science by Elshire et al. [13], and to date is one
of the most powerful applications in the field of plant breeding. The information
derived from GBS experiments have been widely used in genomic diversity studies
and molecular marker discovery, genome-wide association studies (GWAS), genetic
linkage analysis and genomic selection [9]. GBS can be performed through either a
reduced-representation or a whole genome re-sequencing approach [12] generating
a large number of genome-wide SNP data. It does not require any a priori knowledge
on the genome of the species of interest, though several studies have been mainly
carried out in species with reference genomes because SNP genotyping is much
easier when a reference genome is available. Furthermore, GBS typically shows
good results when it is applied to an inbred diploid species with a well-established
reference genome as in the case of barley, maize, sorghum and brassica [9, 13].
Some studies have also made some progresses towards GBS of out-crossing species
lacking reference genomes and of many agriculturally important polyploids crops
such as wheat, cotton and potato [9, 14, 15]. Despite its benefits, GBS shows some
limitations such as the presence of large amount of missing data, largely due to the
use of low coverage sequencing and uneven genome coverage [16].

3 GBS Protocol and Data Analysis

The GBS protocol includes four major steps: (1) sample preparation, (2) NGS
library construction, (3) SNP discovery and (4) genetic analysis (Fig. 1).

1.00
0.80
0.60
0.40
0.20
0.00
1.00
0.80
0.60
0.40
0.20
0.00

Fig. 1 Schematic steps of a genotyping by sequencing experiment

104 F. Taranto et al.

3.1 Sample Preparation

Sample preparation includes DNA extraction, assessment of quality parameters

and library development. Total genomic DNA (gDNA) is extracted from plant
tissue (young leaves or fruits) preferably using columns (i.e. DNeasy kits). The
purity of DNA is confirmed by UV-Vis spectrophotometer and agarose gel elec-
trophoresis. Using any spectrophotometer, the absorbance measurements consider
the 260:280 nm to assess the purity of DNA. Usually, a ratio of 1.8 is generally
accepted as ‘pure’ for DNA. Agarose gel electrophoresis is used to check the
quantity and quality of DNA and evaluate its possible degradation or contaminations
from RNA, proteins, polysaccharides and other pigments that are difficult to
separate from the DNA. However, DNA free from RNA can be obtained by
treating the extracts with the RNAse A enzyme. Usually, thanks to the support of
DNeasy kits, the purified DNA is free of proteins, nucleases and other contaminants
and inhibitors, and therefore it is suitable for NGS. DNA requirements also
include concentration above 50 ng/l and trial digestions by methylation sensitive
restriction enzyme (RE) (i.e. HindIII/EcoRI).
High-quality DNA is then digested with appropriate REs to reduce genomic
complexity. The choice of the appropriate RE is a critical step in developing a GBS
protocol for an organism. Since during library preparation no size selection step
is performed, it is important to maximize the proportion of predicted restriction
fragments that fall within the desired size range (100–400 bp) for sequencing.
The original protocol provides ApeKI [14], which recognizes a degenerate 5-bp
sequence (GCWGC, where W is A or T), creates a 50 overhang (3-bp) and is partially
methylation sensitive [13]. Other enzymes such as the rare cutter, PstI (CTGCAG)
and a frequent cutter, MspI (CCGG) can be alternatively used. A larger pool of
restriction fragments and consequently more unique sequences are generated by
ApeKI. However PstI and MspI provide a greater degree of complexity reduction
and uniform library for sequencing. Both REs have been widely used in genetic
diversity studies on crops [14, 17]. Sonah et al. [18] described a modified library
preparation protocol, in which selective amplification is used to increase both the
number of SNPs called and their depth of coverage, resulting in a high efficiency
and a reduction in per sample cost. After digestion, fragments are directly ligated
to a pair of enzyme-specific adapters, which contain specific priming sites for the
Illumina sequencing. Following ligation, the fragments are PCR amplified.

3.2 NGS Library Construction

Up to four amplicons with similar concentrations are generally pooled in order

to assemble the library. A selection of Illumina-specific sequences is carried out
followed by appropriate quantification and adjustment to preferred concentration.
Afterwards, fragments are combined to form a sample library. As described in
An Overview of Genotyping by Sequencing in Crop Species and Its Application. . . 105

Peterson et al. [16], the GBS protocol uses the Illumina ‘Generate FASTQ’
workflow, the ‘FASTQ Only’ application and ‘TruSeq HT’ assay to generate a
de-multiplexed set of FASTQ files with the adapter sequences removed upon
completion of the sequencing run. After sequencing run, raw data are downloaded.
Each sample has two FASTQ files representing the forward and reverse sequenced
reads. FASTQ files are text-based files for storing biological sequences (FASTA)
with embedded quality scores.

3.3 SNP Discovery

Different computational pipelines have been specifically developed for SNP dis-
covery and genotyping from FASTQ files. The TASSEL-GBS Discovery Pipeline
is the most used in diploid plants with a reference genome [19]. It uses the first
64 nucleotides (nts) of the reads to minimize the effects of sequencing errors.
As mentioned above, the sequencing produces million reads, split across multiple
FASTQ files. All unique sequence tags from each sequence file are captured and
then collapsed to generate a master tag file. The alignment of the unique 64-nts
reads (tags) to reference genome is carried out using Bowtie2 [20] or BWA.
A ‘TagsOnPhysicalMap’ (TOPM) file is returned as output and it can be used for
SNP calling. SNP call is carried out for each set of tags originating from the same
restriction enzyme cut site. Every set of tags aligns to the exact starting genomic
position and strand, where the starting genomic position of a tag is identify by
the cut site residue at the beginning of the tag. Raw SNP data output produced by
the TASSEL-GBS pipeline are further filtered for studying purposes. Usually, the
parameters considered are: inbreeding coefficient (FIT ) and minimum minor allele
frequency (mnMAF). FIT is largely used to filter SNPs from NGS data in inbred lines
[21] and it is calculated based on the expectation–maximization (EM) algorithm
[22]. In GBS analysis, spurious SNPs will appear to be excessively heterozygous,
so it is necessary to calculate the FIT and apply the minimum FIT filter, generally 0.8
[19]. To detect and filter out error-prone SNPs, the TASSEL-GBS pipeline relies on
population-genetic parameters such as MAF. The minimal filter used is in general
set to MAF>0.01. Minimum minor allele count (mnMAC) and minimum locus
coverage (mnLCov) are two additional parameters used in GBS analysis to count
the number of minor alleles for each marker and to evaluate the proportion of taxa
with a genotype, respectively [19].
The TASSEL-GBS pipeline provides SNP calls in both HapMap and VCF
formats. The pipeline provides two sets of HapMap files: (1) a set without post
SNP calling filtering; (2) a set with additional filtering on missingness and allele
frequency. VCF format is an alternative format for holding SNP information that
retains information on depth of coverage for each allele, and the genotype likelihood
scores are calculated according to Etter et al. [23]. Specific software packages, such
as VCFtools and VCFlib, have been developed for working with and manipulating
VCFfiles [24]. For species with no reference genomes, a network-based algorithm
106 F. Taranto et al.

(UNEAK) and a computational pipeline, npGeno, were specifically developed for

SNP discovery and genotyping [9, 16]. These two last bioinformatic pipelines
have been developed particularly for polyploidy species, such as wheat, cotton and
potato.

3.4 Genetic Analysis

The output data files generated from the bioinformatic pipelines are widely used in
different genetic studies including conventional analysis to evaluate heterozygosity
and genetic relationships among individuals, genetic diversity and population
structure in large germplasm collections, high density linkage maps development,
phylogenetic and association mapping studies. Each of these aspects requires
complex analysis. In the next paragraph, a brief overview of the main methods used
for genetic diversity studies is given.

4 Methods for Studying Genetic Diversity in Crops

The study of genetic variation is of great interest for trait association analysis and
evolutionary researches. The first step is to investigate the population structure (the
presence of genetic differences among groups of individuals and their assignment
to different clusters based on allele frequency) given the large amount of SNP data.
So far, several algorithms have been proposed which can be divided into two major
computational paradigms: parametric and non-parametric. Parametric approaches
assume a model in which there are K populations, each of which characterized by
a set of allele frequencies at each locus. The assignment of individuals to a specific
cluster is based on statistical likelihood method, using assumption such as Hardy–
Weinberg equilibrium (HWE) for each marker and linkage equilibrium (LE) among
markers [25]. The structure paradigm consists in a model-based clustering approach
to infer the presence of distinct populations, assign each individual to a population
and estimate ancestral population allele frequencies based on a statistical method
known as the allele-frequency admixture model [26]. The most popular software
to investigate the genetic structure in plants is STRUCTURE [26], although, in the
last years, the ADMIXTURE [27, 28] program usage is growing. Both software
used the same statistical model and input files (i.e. HapMap by the TASSEL-GBS
pipeline) although ADMIXTURE performs much more rapidly since it employs
a fast numerical optimization. STRUCTURE uses a Markov Chain Monte Carlo
(MCMC) stochastic algorithm to produce sample-based estimates of a target
distribution of choice and Bayesian approach based on the posterior distribution of
defined population quantities. ADMIXTURE employs the same likelihood model
but focuses on maximizing the likelihood rather than the posterior distribution.
ADMIXTURE makes the further assumption of linkage equilibrium among the
markers where dense marker sets should be pruned to mitigate background linkage
disequilibrium (LD).
An Overview of Genotyping by Sequencing in Crop Species and Its Application. . . 107

Both software estimates the best value of K. STRUCTURE HARVESTER

(http://taylor0.biology.ucla.edu/structureHarvester/) is a web-based program devel-
oped to analyse the results generated by the program STRUCTURE. The algorithm
implemented in STRUCTURE HARVESTER allows to assess and visualize like-
lihood values across multiple values of K and requires at least three values of
sequential K with at least three replicates, and that the sample standard deviation
of the log likelihood values across all K values is non-zero.
ADMIXTURE uses a cross-validation procedure to identify which K has the best
predictive value, as determined by ‘holding out’ data points [27, 29].
The non-parametric approaches provide an alternative series of statistical meth-
ods that require few assumptions for data analysis [30]. There is a wide range
of methods that can be used for different purposes. A viable tool to understand
population diversity and structure is AWclust. AWclust has been firstly used to
investigate genetic diversity in a human population [31] and it calculates the
allele sharing distance (ASD) matrix, which represents the underlying genetic
distance between every pair of individuals. The non-parametric analysis generated
a multidimensional scaling (MDS) 2D/3D plots to recognize how the samples
grouped, and the dendrogram tree to get a general relationships among individuals
and to identify the number of population clusters. Furthermore, AWclust calculates
the Gap statistics for estimating the optimal number of group (K) based on sample
genetic relatedness.
Once the population structure is assessed, it is possible to select individuals in
order to define a core collection and reduce the number of genotypes for downstream
association mapping studies.

5 GBS Application in Crop Species

GBS technology is becoming pivotal as a cost-effective and unique tool for

genomics-assisted breeding in numerous crop species. GBS has been successfully
applied for a range of studies including genetic mapping [13, 17], assaying genetic
diversity and population structure [32] and genomic selection [14]. Both monocots
and dicots have been optimized by GBS for the efficient, low-cost and high-
throughput SNP marker discovery (Table 1). The results achieved with GBS depend
mainly on the type of population, the genome of the species and the protocol used.
Several examples of GBS studies were reported in diploid species and are
focused on recombinant inbred lines (RILs) and germplasm collections. RILs are
particularly feasible for GBS because of their high homozygosity, minimizing
heterozygote genotyping errors caused by low read depth. Considering RILs, 2,815
maize inbred accessions were genotyped and 681,257 SNP markers distributed
across the entire genome were detected, some of which linked to known candidate
genes for quality traits and flowering time [33]. In sorghum, GBS analysis was
conducted with an F6 RIL population derived from an intra-specific cross between
two Sorghum bicolour cultivars. The pilot study was performed by a single MiSeq
108 F. Taranto et al.

Table 1 Recent crop species analysed by GBS approach (non-exhaustive list)

Genome Sample
Crops Species size (Mb) size N. SNPs References
Maize Zea mays L. 2,600 33,000 2,200 K Romay et al. [33]
Rice Oryza sativa L. 400 850 60 K Spindel et al. [34]
Barley Hordeum vulgare L. 5,427 160 1,949 Poland et al. [14]
Brassica Brassica oleracea 628 89 683 Kim et al. [9]
Sorghum Sorghum bicolor L. 700 90 576 Kim et al. [9]
Soybean Glycine max L. 1,024 301 16,502 Jarquín et al. [35]
Potato Solanum tuberosum L. 840 636 129,156 Uitdewilligen et al. [15]
Cotton Gossypium hirsutum L. 197,632 39 956 Kim et al. [9]
Alfa-alfa Medicago sativa spp. 800 48 11,694 Rocher et al. [36]
Oat Avena sativa L. 11,300 2,664 45,117 Huang et al. [37]
Oil palm Elaeis guineensis 1,800 108 21,471 Pootakham et al. [38]
Cassava Manihot esculenta 530 917 56,489 Rabbi et al. [39]
Crantz
Watermelon Citrullus lanatus var. 425 183 11,483 Nimmakayala et al. [32]
lanatus
Guinea Dioscorea spp. 1,200 95 6,371 Girma et al. [40]
yams
Peach Prunus persica (L.) 227 57 9,998 Bielenberg et al. [41]
Batsch
Miscanthus Miscanthus sinensis 2,592 230 49,007 Ma et al. [42]
Pine Pinus spp. 25,000 99 7,000– Pan et al. [43]
14,751
Genome size is in megabases (Mb), number of individuals genotyped (sample size) and number of
SNPs used to assess genetic diversity

run ( 25 million reads in total) with 90 mapping individuals plus parents (three
redundant samples each), resulting in a total of 576 SNPs genetically mapped with
the aid of the reference genome [9]. In rice, 30,894 SNPs were identified on 176
RILs and used to map the recombined hot and cold spots and QTLs for leaf width
and aluminium tolerance [34].
Few studies have been performed on germplasm collections to characterize the
genetic structure and to provide a tool for association mapping analysis for complex
traits. As an example we report the work by Nimmakayala et al. [32], where the
genetic structure of 183 domesticated watermelon accessions is investigated using a
data set of 11,485 SNPs. Based on 5,254 filtered SNPs, linkage disequilibrium and
population structure were estimated in order to identify agronomically important
candidate genes. GBS has also been used for marker development in cassava
(Manihot esculenta Crantz). Using a set of 917 accessions, 56,489 SNP loci were
genotyped to assess population structure and perform varietal identification [39].
GBS was applied also in polyploid species such as potato, wheat and cotton. In
potato, 12.4 gigabases of high-quality sequence data and 129,156 sequence variants
have been identified [15]. In bread wheat, GBS was used to develop a high density
An Overview of Genotyping by Sequencing in Crop Species and Its Application. . . 109

map of 20,000 SNPs. To further evaluate GBS in wheat, a de novo genetic map was
also constructed using only SNP markers from GBS experiment. The GBS approach
presented here provides a powerful method of developing high density markers in
species without a sequenced genome while providing valuable tools for anchoring
and ordering physical maps and whole genome shotgun sequences [17].
Successful results in markers assisted breeding programs are reported. In cotton,
GBS was used to genotype two BC4 F1 populations and design strategies to obtain
near isogenic lines (NILs) [9]. Two reciprocal sets of NILs by introgression between
two tetraploid species were developed. In the first, 956 SNPs were used to genotype
39 individuals, which resulted in a total finding of 106 introgressions on average.
The second set consisted of 39 individuals genotyped with 914 SNPs for a total
of 114 introgressions. In pepper, GBS technology was used to develop a marker-
assisted backcrossing (MABC) program for the constitution of new pepper varieties
containing capsinoids, starting from BC1 F1 and BC2 F1 populations [44].
Despite the economical and nutritional importance of Solanaceae and the
huge variability within, analytical studies on the genetic variability in germplasm
collections using GBS are lacking. In the next paragraph we illustrate our research
activity aiming to investigate genetic diversity in a population of cultivated pepper
(Capsicum annuum) accessions.

6 Genetic Diversity Analysis in Capsicum annuum

Using GBS

Pepper (Capsicum spp.) belongs to Solanaceae, which is an economically important

family of flowering plants consisting of 102 genera and 2500 species. Plants
belonging to the genus Capsicum had their origins in the South American regions
and now are widely cultivated in tropical and temperate areas. Estimates report the
existence of about 40 species (www.theplantlist.org) five of which (C. annuum,
C. baccatum, C. chinense, C. frutescens and C. pubescens) were domesticated
through distinct events at different primary diversification centres [45] and are
widely consumed as sweet and hot peppers. Most of the species are diploids
with 24 and 26 chromosomes and are distributed in three main gene pools based
on morphological characteristics, chromosome banding or hybridization studies.
Among the domesticated Capsicum spp., C. annuum (2n D 2x D 24) is the most
widely grown species in the world, consumed as food or processed product and it is
the most used in pepper breeding programs [46]. For most cultivated species the loss
of genetic variability started as soon as the domestication process and subsequent
steps of artificial selection. This led to the great variation in size, shape, colour and
pungency of contemporary C. annuum fruits, depending on consumers’ preference
and product destination (fresh or powder). By contrast, the large number of
landraces and ecotypes developed as a consequence of farmers’ selection represents
a wide source of diversity, particularly for alleles of agricultural interest and
110 F. Taranto et al.

related to local adaptation. Therefore, the availability of large germplasm collections

facilitates the evaluation of population diversity and genetic structure, providing
vital information for genome-wide association mapping and allele mining studies
to be exploited by plant breeders for the development of novel varieties and seed
conservation programs [47, 48]. More recently, several approaches were developed
in pepper to assess the genetic diversity and track allelic variants associated with
phenotypic variations. The recent whole genome sequencing of Capsicum [49, 50]
provides a more complete view to estimate chromosome wide molecular diversity
and precisely infer pepper population structure.
Aiming to contribute in this scenario, we determined population structure and
estimated genetic diversity in a collection of cultivated pepper (Capsicum annuum)
using GBS.
Our approach consisted to collect and phenotype two hundred accessions of
cultivated pepper for a wide range of agronomical and morphological traits.
Genetic materials were retrieved from farmers and producers from over 20 world
countries. Plants were previously stabilized through self-fertilization cycles and
DNA was extracted using DNeasy plant column (QIAGEN). Quality parameters
were measured by absorbance at 260 and 280 nm, respectively, using a UV-
Vis spectrophotometer (ND-1000; Nanodrop, Thermo Scientific, USA). GBS was
performed following the protocol described in Elshire et al. [13] using the ApeKI
enzyme. About 8 million master tags were aligned to reference CM334 genome
[49]. The TASSEL-GBS pipeline allowed to identify almost 100k filtered SNPs,
which have been used to determine the population structure. Hereafter we show
preliminary results on the population structure using a parametric approach of
study. According to the Evanno’s test [51] (Fig. 2), the population was divided
into 3 clusters (Fig. 3a). A main group was identified which includes most of the

Fig. 2 Evanno’s plot

generated by STRUCTURE 30000
HARVESTER for the
detection of the true number 25000
of clusters (the most likely
value of K). The best value
was at K D 3 20000
Delta K

15000

10000

5000

2 4 6 8 10 12
K
An Overview of Genotyping by Sequencing in Crop Species and Its Application. . . 111

Fig. 3 Estimate of genetic diversity of C. annuum accessions using GBS-SNP markers. Bar-
plot describing the population structure estimated by the Bayesian clustering. Each individual is
represented by a thin vertical line, which is partitioned into K coloured segments whose length
is proportional to the estimated membership coefficient (q). Population structure at (a) K D 3, (b)
K D 6, (c) K D 10 is reported. Three, six and ten groups are identified, respectively. The asterisk
shows the most informative K value (K D 3). Genotypes retrieved from the same geographical
areas are represented by yellow and brown lines at K D 6 and K D 10, respectively

accessions having a common geographical origin. Considering the large variability

within these sub-populations we performed the STRUCTURE analysis on C.
annuum collection using other different K, particularly K D 6 and K D 10. At
K D 6 (Fig. 3b) the clusters displayed some admixture, and the genetic structure
of collection was not informative. Considering K D 10 (Fig. 3c), there was a
better distinction considering other characteristics such as fruits morphology and
pungency. As observed, increasing the number of sub-populations (K) it was
112 F. Taranto et al.

possible to distinguish the accessions considering both geographical origin and fruit
characteristics. Detailed assessment of morphological fruit-related characteristics
was carried out using automated tools for the analysis (i.e. Chroma metre, 2D
scanner). In total over 300 thousand data points for 38 fruit size and shape
attributes were obtained. Main phenotypic variation was due to fruit size traits
(i.e. perimeter, area, fruit height and fruit width) which could be considered the
most relevant attributes for breeding new varieties. In order to identify genomic
regions responsible for the phenotypic variation, high-quality SNP (mmMAF 0.01,
coverage 90 %) were further selected. A first attempt to associate SNP alleles and
morphological traits was carried out on the basis of General Linear Model. Several
SNP highly correlated to the phenotypic variation were identified. For the main
traits responsible for fruit size variation as well as for shape traits of high interest
in breeding, highly correlated SNP were detected on chromosomes 2, 3, 6 and 9.
Next step will involve the integration of a parametric (STRUCTURE) with a non-
parametric approach (AWclust) in order to better refine the population structure with
the aim to select a core-set of accessions. Moreover, Mixed Linear Model will be
used for future association mapping analysis.

7 Conclusion

GBS is a high-throughput and low-cost technology used in several crop species

in order to genotype breeding population, assess genomic diversity, discover and
develop new molecular markers useful in plant breeding programs, and carry out
GWAS. GBS, has proven useful and reliable for the identification of high-quality
SNPs. It has several advantages, including the fact that no preliminary sequence
information is required and that all newly discovered markers originate from the
population under investigation.
Our study aims to unlock the genetic potentiality of cultivated pepper, which
represents a major vegetable crop given its nutritional properties. GBS has been
chosen to identify a large number of SNPs useful to precisely define the structure of
a C. annuum population. Moreover large-scale phenomics has been carried out for
fruit-related traits. Information concerning SNP markers and population structure
developed in this study are the first step towards future genome-wide association
mapping studies and marker-assisted selection programs in cultivated pepper.

Acknowledgements This project was supported by the ‘GenHort’ project funded by the Italian
Ministry of University and Research (MIUR, PON02_00395_3215002) and the ‘PEPIC’ project
funded by the Italian Ministry of Agricultural, Food and Forestry.
An Overview of Genotyping by Sequencing in Crop Species and Its Application. . . 113

References

1. Hammer, K., Arrowsmith, N., Gladis, T.: Agrobiodiversity with emphasis on plant genetic
resources. Naturwissenschaften 90, 241–250 (2003)
2. Pei, C., Wang, H., Zhang, J., Wang, Y., Francis, D.M., Yang, W.: Fine mapping and analysis of a
candidate gene in tomato accession PI128216 conferring hypersensitive resistance to bacterial
spot race T3. Theor. Appl. Genet. 124(3), 533–542 (2012)
3. Di Dato, F., Parisi, M., Cardi, T., Tripodi, P.: Genetic diversity and assessment of markers linked
to resistance and pungency genes in Capsicum germplasm. Euphytica 1, 103–119 (2015)
4. Xu, Y., Ma, R.C., Xie, H., Liu, J.T., Cao, M.Q.: Development of SSR markers for the
phylogenetic analysis of almond trees from China and the Mediterranean region. Genome
47(6), 1091–1104 (2004)
5. Alseekh, S., Ofner, I., Pleban, T., Tripodi, P., Di Dato, F., Cammareri, M., Mohammad, A.,
Grandillo, S., Fernie, A.R., Zamir, D.: Resolution by recombination: breaking up Solanum
pennellii introgressions. Trends Plant Sci. 18(10), 536–538 (2013)
6. Laidò, G., Mangini, G., Taranto, F., Gadaleta, A., Blanco, A., Cattivelli, L., Marone, D.,
Mastrangelo, A.M., Papa, R., De Vita, P.: Genetic diversity and population 387 structure of
tetraploid wheats (Triticum turgidum L.) estimated by SSR, DArT and Pedigree Data. PLoS
One 8(6), e67280 (2013)
7. Wu, F., Eannetta, N.T., Durrett, Y.X.R., Mazourek, M., Jahn, M.M., Tanksley, S.D.: A COSII
genetic map of the pepper genome provides a detailed picture of synteny with tomato and new
insights into recent chromosome evolution in the genus Capsicum. Theor. Appl. Genet. 118,
1279–1293 (2009)
8. Wu, F., Eannetta, N.T., Xu, Y., Plieske, J., Ganal, M., Pozzi, C., Bakaher, N., Tanksley, S.D.:
COSII genetic maps of two diploid Nicotiana species provide a detailed picture of synteny with
tomato and insights into chromosome evolution in tetraploid N. tabacum. Theor. Appl. Genet.
120(4), 809–827 (2010)
9. Kim, C., Guo, H., Kong, W., Chandnani, R., Shuang, L.S., Paterson, A.H.: Application of
genotyping-by-sequencing technology to a variety of crop breeding programs. Plant Sci. 242,
12–14 (2016)
10. Rafalski, A.: Applications of single nucleotide polymorphisms in crop genetics. Curr. Opin.
Plant Biol. 5, 94–100 (2002)
11. He, J., Zhao, X., Laroche, A., Lu, Z.X., Liu, H., Li, Z.: Genotyping-by-sequencing (GBS), an
ultimate marker-assisted selection (MAS) tool to accelerate plant breeding. Front. Plant Sci. 5,
484 (2014)
12. Deschamps, S., Llaca, V., May, G.D.: Genotyping-by-sequencing in plants. Biology 1, 460–483
(2012)
13. Elshire, R.J., Glaubitz, J.C., Sun, Q., Poland, J.A., Kawamoto, K., Buckler, E.S., Mitchell, S.E.:
A robust, simple genotyping-by-sequencing (GBS) approach for high diversity species. PLoS
One 6(5), e19379 (2011)
14. Poland, J.A., Brown, P.J., Sorrells, M.E., Jannink, J.L.: Development of high-density genetic
maps for barley and wheat using a novel two-enzyme genotyping-by-sequencing approach.
PLoS One 7, e32253 (2012)
15. Uitdewilligen, J.G., Wolters, A.A., D’hoop, B.B., Borm, T.J., Visser, R.G., van Eck, H.J.:
A next-generation sequencing method for genotyping by-sequencing of highly heterozygous
autotetraploid potato. PLoS One 8(5), e62355 (2013)
16. Peterson, G.W., Dong, Y., Horbach, C., Fu, Y.B.: Genotyping-by-sequencing for plant genetic
diversity analysis: a lab guide for SNP genotyping. Diversity 6, 665–680 (2014)
17. Poland, J., Endelman, J., Dawson, J., Rutkoski, J., Wu, S., et al.: Genomic selection in wheat
breeding using genotyping-by-sequencing. Plant Genome 5, 103–113 (2012)
114 F. Taranto et al.

18. Sonah, H., Bastien, M., Iquira, E., Tardivel, A., Legare, G., Boyle, B., Normandeau, E.,
Laroche, J., Larose, S., Jean, M., Belzile, F.: An improved genotyping-by-sequencing (GBS)
approach offering increased versatility and efficiency of SNP discovery and genotyping. PLoS
One 8(1), e54603 (2013)
19. Glaubitz, J.C., Casstevens, T.M., Lu, F., Harriman, J., Elshire, R.J., Sun, Q., Buckler, E.S.:
TASSEL-GBS: a high capacity genotyping-by-sequencing analysis pipeline. PLoS One 9(2),
e90346 (2014)
20. Langmead, B., Salzberg, S.L.: Fast gapped-read alignment with Bowtie 2. Nat. Methods 9,
357–359 (2012)
21. Vieira, F.G., Fumagalli, M., Albrechtsen, A., Nielsen, R.: Estimating inbreeding coefficients
from NGS data: impact on genotype calling and allele frequency estimation. Genome Res. 23,
1852–1861 (2013)
22. Smith, C.A.B., Thomson, R.: Estimation of inbreeding from population samples. J. Appl.
Probab. 25, 127–135 (1988)
23. Etter, P.D., Bassham, S., Hohenlohe, P.A., Johnson, E.A., Cresko, W.A.: SNP discovery and
genotyping for evolutionary genetics using RAD sequencing. Methods Mol. Biol. 772, 157–
178 (2011)
24. Danecek, P., Auton, A., Abecasis, G., Albers, C.A., Banks, E., Depristo, M.A., Handsaker, R.,
Lunter, G., Marth, G., Sherry, S.T., McVean, G., Durbin, R., 1000 Genomes Project Analysis
Group: The variant call format and VCFtools. Bioinformatics 27, 2156–2158 (2011)
25. Deejai, P., Assawamakin, A., Wangkumhang, P., Poomputsa, K., Tongsima, S.: On assigning
individuals from cryptic population structures to optimal predicted subpopulations: an empiri-
cal evaluation of non-parametric population structure analysis techniques. Comput. Syst. Biol.
Bioinform. 115, 58–70 (2010)
26. Pritchard, J.K., Stephens, M., Donnelly, P.: Inference of population structure using multilocus
genotype data. Genetics 155, 945–959 (2000)
27. Alexander, D.H., Novembre, J., Lange, K.: Fast model-based estimation of ancestry in
unrelated individuals. Genome Res. 19(9), 1655–1664 (2009)
28. Frichot, E., Mathieu, F., Trouillon, T., Bouchard, G., Francois, O.: Fast and efficient estimation
of individual ancestry coefficients. Genetics 196(4), 973–983 (2014)
29. Alexander, D.H., Lange, K.: Enhancements to the ADMIXTURE algorithm for individual
ancestry estimation. BMC Bioinf. 12, 246 (2011)
30. Whitley, E., Ball, J.: Statistics review 6: nonparametric methods. Crit. Care 6, 509–513 (2002)
31. Gao, X., Martin, E.R.: Using allele sharing distance for detecting human population stratifica-
tion. Hum. Hered. 68, 182–191 (2009)
32. Nimmakayala, P., Levi, A., Abburi, L., Abburi, V.L., Tomason, Y.R., Saminathan, T., Vajja,
V.G., Malkaram, G., Reddy, R., Wehner, T.C., Mitchell, S.E., Reddy, U.K.: Single nucleotide
polymorphisms generated by genotyping-by-sequencing to characterize genome-wide diver-
sity, linkage disequilibrium, and selective sweeps in cultivated watermelon. BMC Genomics
15, 767 (2014)
33. Romay, M.C., Millard, M.J., Glaubitz, J.C., Peiffer, J.A., Swarts, K.L., Casstevens, T.M.,
Elshire, R.J., Acharya, C.B., Mitchell, S.E., Flint-Garcia, S.A., McMullen, M.D., Holland, J.B.,
Buckler, E.S., Gardner, C.A.: Comprehensive genotyping of the USA national maize inbred
seed bank. Genome Biol. 14, R55 (2013)
34. Spindel, J., Wright, M., Chen, C., Cobb, J., Gage, J., Harrington, S.: Bridging the genotyping
gap: using genotyping-by-sequencing (GBS) to add high-density SNP markers and new value
to traditional bi-parental mapping and breeding populations. Theor. Appl. Genet. 126, 2699–
2716 (2013)
35. Jarquín, D., Kocak, K., Posadas, L., Hyma, K., Jedlicka, J., Graef, G., Lorenz, A.: Genotyping
by sequencing for genomic prediction in a soybean breeding population. BMC Genomics 15,
740 (2014)
36. Rocher, S., Jean, M., Castongyay, Y., Belzile, F.: Validation of genotyping-by-sequencing anal-
ysis in populations of tetraploid alfalfa by 454 sequencing. PLoS One (2015). doi:10.1371/jour-
nal.pone.0131918
An Overview of Genotyping by Sequencing in Crop Species and Its Application. . . 115

37. Huang, Y.F., Poland, J.A., Wight, C.P., Jackson, E.W., Tinker, N.A.: Using genotyping-by-
sequencing (GBS) for genomic discovery in cultivated oat. PLoS One 9(7), e102448 (2014)
38. Pootakham, W., Jomchai, N., Ruang-areerate, P., Shearman, J.R., Sonthirod, C., Sangsrakru,
D., Tragoonrung, S., Tangphatsornruang, S.: Genome-wide SNP discovery and identification
of QTL associated with agronomic traits in oil palm using genotyping-by-sequencing (GBS).
Genomics 105, 288–295 (2015)
39. Rabbi, Y.I., Kulakow, P.A., Manu-Aduening, J.A., Dankyi, A.A., Asibuo, J.Y., Parkes, E.Y.,
Abdoulaye, T., Girma, G., Gedil, M.A., Ramu, P., Reyes, B., Maredia, M.K.: Tracking crop
varieties using genotyping by-sequencing markers: a case study using cassava (Manihot
esculenta Crantz). BMC Genet. 16, 115 (2015)
40. Girma, G., Hyma, K.E., Asiedu, R., Mitchell, S.E., Gedil, M., Spillane, C.: Next generation
sequencing based genotyping, cytometry and phenotyping for understanding diversity and
evolution of guinea yams. Theor. Appl. Genet. 127, 1783–1794 (2014)
41. Bielenberg, D.G., Rauh, B., Fan, S., Gasic, K., Abbott, A.G., Reighard, G.L., Okie, W.R.,
Wells, C.E.: Genotyping by sequencing for SNP-based linkage map construction and QTL
analysis of chilling requirement and bloom date in peach [Prunus persica (L.) Batsch]. PLoS
One (2015). doi:10.1371/journal.pone.0139406
42. Ma, X.F., Jensen, E., Alexandrov, N., Troukhan, M., Zhang, L., Thomas-Jones, S., Farra, K.,
Clifton-Brown, J., Donnison, I., Swaller, T., Flavell, R.: High resolution genetic mapping by
genome sequencing reveals genome duplication and tetraploid genetic structure of the diploid
Miscanthus sinensis. PLoS One 7(3), e33821 (2012)
43. Pan, J., Wang, B., Pei, Z.Y., Zhao, W., Gao, J., Mao, J.F., Wang, X.R.: Optimization of the
genotyping-by-sequencing strategy for population genomic analysis in conifers. Mol. Ecol.
Resour. 15, 711–722 (2015)
44. Jeong, H.S., Jang, S., Han, K., Kwon, J.K., Kang, B.C.: Marker-assisted backcross breeding
for development of pepper varieties (Capsicum annuum) containing capsinoids. Mol. Breed.
35, 226 (2015)
45. Moscone, E.A., Scaldaferro, M.A., Grabiele, M., Cecchini, N.M., Sanchez Garcıa, Y., Jarret,
R., Davina, J.R., Ducasse, D.A., Barboza, G.E., Ehrendorfer, F.: The evolution of chili peppers
(Capsicum-Solanaceae): a cytogenetic perspective. Acta Hortic. 745, 137–170 (2007)
46. Hernández-Verdugo, S., Luna-Reyes, R., Oyama, K.: Genetic structure and differentiation of
wild and domesticated populations of Capsicum annuum (Solanaceae) from Mexico. Plant
Syst. Evol. 226(3–4), 129–142 (2001)
47. Cericola, F., Portis, E., Toppino, L., Barchi, L., Acciarri, N., Ciriaci, T., Sala, T., Rotino, G.L.,
Lanteri, S.: The population structure and diversity of eggplant from Asia and the Mediterranean
Basin. PLoS One 8(9), e73702 (2013)
48. Rodriguez, M., Rau, D., Bitocchi, E., Bellucci, E., Biagetti, E., Carboni, A., Biagetti, E.,
Carboni, A., Gepts, P., Nanni, L., Papa, R., Attene, G.: Landscape genetics, adaptive diversity
and population structure in Phaseolus vulgaris. New Phytol. 209(4), 1781–1794 (2015)
49. Kim, S., Park, M., Yeom, S.I., Kim, Y.M., Lee, J.M., Lee, H.A., Seo, E., Choi, J., Cheong, K.,
Kim, K.T., Jung, K., Lee, G.W., Oh, S.K., Bae, C., Kim, S.B., Lee, H.Y., Kim, S.Y., Kim, M.S.,
Kang, B.C., Jo, Y.D., Yang, H.B., Jeong, H.J., Kang, W.H., Kwon, J.K., Shin, C., Lim, J.Y.,
Park, J.H., Huh, J.H., Kim, J.S., Kim, B.D., Cohen, O., Paran, I., Suh, M.C., Lee, S.B., Kim,
Y.K., Shin, Y., Noh, S.J., Park, J., Seo, Y.S., Kwon, S.Y., Kim, H.A., Park, J.M., Kim, H.J.,
Choi, S.B., Bosland, P.W., Reeves, G., Jo, S.H., Lee, B.W., Cho, H.T., Choi, H.S., Lee, M.S.,
Yu, Y., Do Choi, Y., Park, B.S., van Deynze, A., Ashrafi, H., Hill, T., Kim, W.T., Pai, H.S., Ahn,
H.K., Yeam, I., Giovannoni, J.J., Rose, J.K., Sørensen, I., Lee, S.J., Kim, R.W., Choi, I.Y., Choi,
B.S., Lim, J.S., Lee, Y.H., Choi, D.: Genome sequence of the hot pepper provides insights into
the evolution of pungency in Capsicum species. Nat. Genet. 46(3), 270–278 (2014)
50. Qin, C., Yu, C., Shen, Y., Fang, X., Chen, L., Min, J., Cheng, J., Zhao, S., Xu, M., Luo, Y.,
Yang, Y., Wu, Z., Mao, L., Wu, H., Ling-Hu, C., Zhou, H., Lin, H., González-Morales, S.,
Trejo-Saavedra, D.L., Tian, H., Tang, X., Zhao, M., Huang, Z., Zhou, A., Yao, X., Cui, J., Li,
W., Chen, Z., Feng, Y., Niu, Y., Bi, S., Yang, X., Li, W., Cai, H., Luo, X., Montes-Hernández,
S., Leyva-González, M.A., Xiong, Z., He, X., Bai, L., Tan, S., Tang, X., Liu, D., Liu, J., Zhang,
116 F. Taranto et al.

S., Chen, M., Zhang, L., Zhang, L., Zhang, Y., Liao, W., Zhang, Y., Wang, M., Lv, X., Wen, B.,
Liu, H., Luan, H., Zhang, Y., Yang, S., Wang, X., Xu, J., Li, X., Li, S., Wang, J., Palloix, A.,
Bosland, P.W., Li, Y., Krogh, A., Rivera-Bustamante, R.F., Herrera-Estrella, L., Yin, Y., Yu, J.,
Hu, K., Zhang, Z.: Whole-genome sequencing of cultivated and wild peppers provides insights
into Capsicum domestication and specialization. Proc. Natl. Acad. Sci. U. S. A. 111(14), 5135–
5140 (2014)
51. Evanno, G., Regnaut, S., Goudet, J.: Detecting the number of clusters of individuals using the
software STRUCTURE: a simulation study. Mol. Ecol. 14, 2611–2620 (2005)
Hybridization-Based Enrichment and Next
Generation Sequencing to Explore Genetic
Diversity in Plants

Irma Terracciano, Concita Cantarella, and Nunzio D’Agostino

Abstract In plant research, targeted re-sequencing of enriched genomic DNA

regions has become a scalable and cost-effective method for the discovery of
genome-wide sequence variations to be exploited to address different biological
questions.
In this manuscript, we describe the main strategies to reduce genome complexity
in plants with a special focus on hybridization-based enrichment methods. Then,
we provide an overview of applications of target enrichment-based next generation
sequencing (NGS) protocols in plant genetics and illustrate major bioinformatic
approaches and tools for the analysis of NGS data, the identification of sequence
polymorphisms, and their annotation to predict possible biological effects. Finally,
we introduce our research activity on the use of hybridization-based target enrich-
ment system for the identification of interesting sequence variations at candidate
genes controlling carotenoid biosynthesis in tomato.

Keywords Sequence capture • Target enrichment • Sequence polymorphisms •

Variant calling • Allele mining

1 Introduction

Next generation sequencing (NGS) technologies have experienced an extraordinary

increase in capacity and a significant advancement in data generation. In plant
research, large datasets are now being generated across various model and non-
model species by sequencing whole genomes. Recently, whole exome sequencing
(WES) provided a cost-effective alternative aiming at the identification of nucleotide
variability across the exome [1, 2], which is defined as the sequences encompassing
all the exons of protein-coding genes in a genome. When there is interest on specific
candidate loci, target enrichment (in which genomic regions are selectively captured
from a DNA sample before sequencing) offers substantial reduction in sequencing

I. Terracciano • C. Cantarella • N. D’Agostino ()

Consiglio per la ricerca in agricoltura e l’analisi dell’economia agraria (CREA) - Centro di ricerca
per l’orticoltura, via dei Cavalleggeri 25, 84089 Pontecagnano Faiano (SA), Italy
e-mail: [email protected]

© Springer International Publishing Switzerland 2016 117

A. Rogato et al. (eds.), Dynamics of Mathematical Models in Biology,
DOI 10.1007/978-3-319-45723-9_10
118 I. Terracciano et al.

space and cost [3–6]. The hybridization-based method is one of the most efficient
and widely adopted among the available target enrichment techniques [7, 8]. It
has been demonstrated powerful, independently of the DNA capture protocol and
the sequencing platform used and it is often replacing PCR as the main target
enrichment method in plant sciences [3, 5].
Plant genomes can be extremely complex, repetitive, and are often polyploids;
as a consequence, some species are not well suited for whole genome sequencing
(WGS) approaches. By contrast, sequence capture and targeted re-sequencing have
the advantage of providing higher read depth for individual locus and support
the accurate identification of nucleotide polymorphisms also in plants with large
genomes and higher ploidy levels [9, 10].
In this manuscript, we provide a brief overview of the available strategies to
reduce genome complexity in plants with a special focus on hybridization-based
enrichment methods currently used for the characterization of natural/induced
genetic variation in plant species. Then, we highlight possible applications of these
technologies to plant research and describe a typical bioinformatic workflow for the
analysis of NGS data and the identification of sequence polymorphisms. Finally, we
discuss our experience in a project aimed at the identification of naturally occurring
sequence variation at candidate genes controlling carotenoid biosynthesis in tomato.

2 Strategies to Reduce Genome Complexity in Plants: Target

Enrichment

For plants that possess large size or polyploid genomes, for which whole genomes
cannot be readily assembled and the analysis of a large number of individuals
results still very expensive, an alternative strategy to WGS is to generate a
reduced representation of the genome. Genome reduction can be obtained using
target enrichment strategies. Target enrichment consists in the isolation of specific
genomic loci (e.g., genes, molecular markers, larger genomic regions, and organelle
genomes) coupled with NGS. Compared to WGS, the reduction in sequencing
space entails three main advantages: (1) sample multiplexing that implicates an
overall reduction of the sequencing cost per sample; (2) significant reduction in the
complexity of the analysis; and (3) the possibility of identifying the precise region
of interest given the depth of sequencing provided by NGS.
At present, transcriptome-based, restriction enzyme-based, PCR-based, and
hybridization-based methods, all compatible with the most popular NGS platforms,
have been developed to enrich specific targets [3].
Transcriptome-Based Enrichment is one of the most widely used strategies to
reduce genome complexity, since it focuses only on the transcribed portion of the
genome. The key aim of transcriptome sequencing, also known as RNA-seq, is to
determine gene expression profiles of each transcript during development and under
different conditions [11]. SNP discovery and molecular marker development via
Hybridization-Based Enrichment and Next Generation Sequencing to Explore. . . 119

RNA-seq are often performed, especially in organisms with large genomes [12].
Noteworthy, since RNA-seq is independent from any a priori knowledge on the
genome sequence of the species under investigation, it allows the analysis of poorly
characterized species.
Restriction Enzyme-Based Enrichment makes use of the discriminatory power
of the restriction endonucleases to produce restriction fragments among individuals
in a population. Three main techniques have been developed so far: RAD-seq
(restriction-site associated DNA sequencing) [13, 14], GR-RSC (genomic reduction
based on restriction site conservation) [15], and GBS (genotyping-by-sequencing)
[16]. All these methods, reviewed by Cronn et al. [3], are flexible and quite
inexpensive and have been used to identify and score, in a group of individuals,
thousands of genetic markers randomly distributed along the genome enabling SNP
discovery, genotyping as well as quantitative genetic and phylo-geographic studies.
PCR-Based Target Enrichment includes the direct sequencing of small and long
PCR products. NGS of PCR fragments has been preferentially applied to chloroplast
genomes in systematic studies [17] and in some cases also to nuclear genomic
regions despite their complexity [18]. The main disadvantages associated with
this method are the high level of failed target amplifications and/or non-specific
amplifications as well as the difficulty in obtaining an accurate pooling of samples
for NGS multiplexing [5]. Anyway, PCR-based enrichment remains feasible for
targeting small to medium-sized regions of the genome, but for high-throughput
sequencing of tens of thousands of PCR amplicons its efficiency falls off, given the
initial cost per sample and challenges in sample multiplexing. Microfluidic-based
multiplexing PCR can reduce costs but continues to be more expensive than other
enrichment methods [3].
Hybridization-Based Enrichment or sequence capture methods exploit the high
specificity of DNA or RNA probes (also called baits) which are designed to be
complementary to target genomic regions. RNA baits have significant advantages
over DNA probes because RNA–DNA hybrids have a higher affinity and melting
temperature than DNA–DNA hybrids. Two main technologies have been developed
for hybrid-capture applications: (1) on-array- or solid-based hybridization which
implies sample hybridization on a solid support (i.e., glass slide, microarray)
[8] and (2) in-solution- or liquid-based hybridization where pooled baits are
used in reaction tubes [7]. Due to their moderate costs and high specificity, low
amounts of required DNA per sample and power to simultaneously target large
numbers of markers, several protocols and commercial kits have been developed.
The most widespread ones and reliable in studies on plant species were provided by
Agilent Technologies (SureSelect), Roche NimbleGen (SeqCap EZ), MYcroarray
(MYbaits),and Ion Torrent (TargetSeq). Distinguishing features of these sequence
capture platforms are reported in Table 1.
120 I. Terracciano et al.

Table 1 List of the most important features of the commercially available target enrichment kits
On-array hybridization-based
capture In-solution hybridization-based capture
NimbleGen Agilent NimbleGen Agilent MYcroarray IonTorrent
Sequence Microarray SeqCap SureSelect MYbait TargetSeq
Capture Array EZ
Bait type DNA DNA DNA RNA RNA DNA
Bait length 60 bp 60 bp 55–105 bp 114–126 bp 80–120 bp 50–120 bp
Target size Up to 30 Mb N.D. Up to From 1 kb Up to From
200 Mb to 24 Mb 200,000 100 kb up
baits to 10 Mb
N.D. not determined

All these providers offer the opportunity to design custom kits for the species of
interest and make available services and tools to support probe design. Evidently,
it is necessary to have a reference sequence (complete or draft genome sequence,
transcripts, Expressed Sequence Tag database, etc.) to accomplish this task.

3 Technical Considerations on Hybridization-Based

Enrichment Methods

As previously mentioned, affordable costs combined with ease of use, multiplexing

capacity, and scalability (from few genes or genomic regions to entire exomes) make
sequence capture an attractive alternative to large-scale PCR and a method of choice
for its wide potential use ranged from intra-specific population studies, typically for
polymorphism identification, to deeper-level phylo-genomics.
Several authors compared the most popular sequence capture technologies (both
liquid- and solid-phase) demonstrating that, although there are slight differences,
results in terms of coverage efficiency, accuracy in genotype assignment, and variant
discovery are basically very similar [19, 20].
However, liquid-based sequence capture systems are gradually replacing on-
array-based hybridization methods because all the reaction steps of the protocol
take place in a single tube making the process scalable to large numbers of samples
and suitable for robotic automation. Furthermore, it requires less input DNA and
simple laboratory equipment [7].
Regardless of the method used, several technical aspects and potential drawbacks
must be taken into account in order to plan a successful target enrichment project
and achieve predetermined objectives.
Ploidy level, genome size, and DNA compositional properties (e.g., high GC
content) of the species under investigation together with several features of the baits
(e.g., probe length, hybridization temperature) can affect enrichment efficiency. It
has been demonstrated that enrichment efficiency level can be considerably reduced
Hybridization-Based Enrichment and Next Generation Sequencing to Explore. . . 121

in promoter regions, 50 UTR regions, and in the first exon of genes because of high
GC content of these regions [21]. High or low GC content reduces the efficiency of
PCR amplifications [22], bait synthesis, and hybridization. Since this latter aspect
is related to nucleotide compositional properties of the probes, it can somehow
be corrected by probe design. The GC bias effect on sequencing coverage has
been studied by different authors, who plot GC content distribution against the
normalized mean read depth [19, 23]. Enrichment efficiency depends also on the
sequence capture protocol of choice as well as on the sequencing technology used.
The percentage of sequences that map to the selected targets (probe specificity)
can be influenced by the presence of closely related sequences (orthologs/paralogs)
of duplicated regions and/or interspersed repetitive elements in the genome [3].
Minimizing the number of off-target reads is desirable and it can be achieved by
selecting probes with high specificity.
A crucial parameter of a sequence capture experiment is the sensitivity, which
is the percentage of the target bases that are represented by one or more sequenced
reads. In other words, the higher the sequencing depth, the higher the confidence
that the base called at that position is correct, the better the estimation of SNP/InDel
frequency for any particular SNP/InDel. Also the experimental design has a great
impact on enrichment efficiency. Effectively, the right balance between the numbers
of targets to be sequenced and the expected sequencing depth must be found.

4 Overview of Hybridization-Based Enrichment

Applications in Plant Species

In the last few years, sequence capture and target enrichment followed by NGS
have been used to identify a high number of mutations in whole exomes, selected
gene families, and target genes or genomic regions of many plant species allowing
(1) generation of useful polymorphism resources in a quick and rather inexpensive
way; (2) biodiversity exploration and mining; (3) SNP marker development and
generation of genetic maps; (4) population structure definition or evolutionary
history in phylogenetics and phylogeography studies to be tracked; (5) QTL
mapping and candidate gene identification; and (6) genomic selection.
All these applications are intended to accelerate plant breeder activity for crop
improvement.
In this section we review recent literature on targeted re-sequencing of enriched
genomic DNA regions in crops and other economically important plant species and
briefly describe objectives and applications of each study (Table 2).
The first application of hybridization-based sequence capture in plant was
published by Fu et al. [9] who demonstrated the effectiveness of the enrichment
protocol in the identification of plant polymorphisms in divergent maize (Zea mays)
lines.
Table 2 List of several examples from the recent literature on the applications of hybridization-based enrichment protocols to plant genetics
122

Hybridization-
Target based
Species Nı Plant material Target size Assay type technology NGS platform References
Maize 1 2 inbred lines Non-repetitive portion 4.3 Mb Roche NimbleGen SB Roche 454 Fu et al. [9]
of 2.2 Mb and 43 array
genes
2 21 inbred lines Genomic regions 29 Mb Roche NimbleGen SB Roche 454 Muraya et al.
including genes for array [24]
biomass production
Rapeseed canola 3 10 genotypes 47 QTL-associated 51.2 Mb Roche NimbleGen SB and LB Roche 454 Clarke et al. [25]
genomic regions array and undefined and Illumina
liquid platform
4 4 accessions 29 regulatory 614 kb Agilent SureSelect LB Illumina Schiessl et al.
flowering-time genes [26]
Cotton 5 2 accessions 1000 geni (500 pairs of 550 kb Roche NimbleGen SB Roche 454 Salmon et al.
homeologs) array [27]
Sugarcane 6 2 genotypes (one Gene-rich regions 5.8 Mb Agilent SureSelect LB Illumina Bundock et al.
Saccharum from the close relative [28]
officinarum and one Sorghum bicolor
Saccharum hybrid)
Swichgrass 7 4 tetraploid lowland Whole exome 50 Mb Roche NimbleGen LB Illumina Evans et al. [29]
and 4 octoploid upland Seq-EZ
cultivars
Cassava 8 100 F1 progeny and 2 27,469 biallelic SNPs 2.49 Mb Ion TargetSeq Life LB Ion Torrent Pootakham et al.
parental strains from 10,105 regions Technologies Proton [30]
System
Loblolly pine 9 24 haploid samples Whole exome 6.57 Mb Agilent SureSelect LB Illumina Neves et al. [31]
10 72 haploid samples Whole exome 6.57 Mb Agilent SureSelect LB Illumina Neves et al. [32]
from a mapping
population
I. Terracciano et al.
Black 11 48 genotypes Predicted exons and 20.76 Mb Agilent SureSelect LB Illumina Zhou and
cottonwood upstream regulatory Holliday [23]
regions and random
genomic intervals
Eucalyptus 12 3 genotypes 94 genes involved in N.D. Agilent custom SB Illumina Dasgupta et al.
xylogenesis array [33]
Wheat 13 The wild emmer Exon regions of 3497 3.5 Mb Agilent SureSelect LB Illumina Saintenac et al.
accession (Triticum genes [10]
dicoccoides) and
durum wheat cultivar
Langdon (Triticum
turgidum var. durum)
14 8 UK alloexaploid Significant proportion 56.5 Mb Roche NimbleGen SB Illumina Allen et al. [34]
varieties of the wheat exome array
15 8 UK alloexaploid Significant proportion 56.5 Mb Roche NimbleGen SB Illumina Winfield et al.
varieties of the wheat exome array [35]
16 2 RILs chosen as Gene-rich regions of 110 Mb Roche NimbleGen LB Illumina Gardiner et al.
parent for a F2 the genome Seq-EZ [36]
mapping population
Rice and wheat 17 EMS-mutagenized rice Whole exome 42 Mb Roche NimbleGen LB Illumina Henry et al. [37]
(72) and wheat (6) (rice) Seq-EZ
individuals 107 Mb
(wheat)
Soybean 18 4 fast-neutron mutants Whole exome 52.3 Mb Roche NimbleGen SB Illumina Bolon et al. [38]
array
Hybridization-Based Enrichment and Next Generation Sequencing to Explore. . .

19 2 individuals of the Whole exome 52.3 Mb Roche NimbleGen SB Illumina Haun et al. [39]
cultivar Williams 82 array
Barley 20 36 samples from 13 Whole exome 61.6 Mb Roche NimbleGen LB Illumina Mascher et al.
barley cultivars and 7 Seq-EZ [40]
samples from 3 wild
relatives
123

(continued)
 124

Table 2 (continued)
Hybridization-
Target based
Species Nı Plant material Target size Assay type technology NGS platform References
21 Parental lines and Whole exome 61.6 Mb Roche NimbleGen LB Illumina Pankin et al. [41]
BC1 F2 lines enriched Seq-EZ
for early flowering
genotypes
22 18 mutant plants and Whole exome 61.6 Mb Roche NimbleGen LB Illumina Mascher et al.
30 randomly selected Seq-EZ [40]
wild type plants
23 3 Hv/Hb ILs and the Whole exome 61.6 Mb Roche NimbleGen LB Illumina Wendler et al.
respective donor lines Seq-EZ [42]
Tribe trifoliae 24 6 individuals 62 low-copy nuclear 185 kb MYbaits LB Illumina de Sousa et al.
representing major genes and 257 short MYcroarray [43]
lineages within loci (exon sequences)
Medicago and distributed across all
Melilotus Medicago
chromosomes
Compositae 25 15 species 763 conserved N.D. MYbaits LB Illumina Mandel et al.
ortholog set (COS) loci MYcroarray [44]
Rosidae, 26 24 species (22 eudicots 300 plastidial genomes N.D. Agilent SureSelect LB Illumina Stull et al. [45]
asteridae, and 2 monocots)
caryophyllales,
asparagaceae,
and poaceae
I. Terracciano et al.
Strawberry 27 48 F1 individuals from 200 bp surrounding 149 Mb MYbaits LB Illumina Tennessen et al.
MRD30xMDR60 and each of 6575 MYcroarray [46]
their parental lines previously identified
polymorphisms
Potato and 28 Rpi-ber2 and Rpi-rzc1 580 NB-LRR coding N.D. Agilent SureSelect LB Illumina Jupe et al. [47]
tomato F1 populations and sequence from
Solanum tuberosum Solanaceae
group Phureja clone
DM1-3 516 R44
Tomato 29 Solanum 743 NB-LRR-like N.D. Agilent SureSelect LB Illumina Andolfo et al.
pimpinellifolium sequences [48]
LA1589 and Solanum
lycopersicum Heinz
1706
Potato 30 83 tetraploid cultivars 807 genes 1.44 Mb Agilent SureSelect LB Illumina Uitdewilligen et
and 1 monoploid clone al. [49]
(DM 1-3 511)
LB liquid-based, SB solid-based, N.D. not determined
Hybridization-Based Enrichment and Next Generation Sequencing to Explore. . .
125
126 I. Terracciano et al.

In the course of recent years, different authors proved the efficacy of exome
capture for the investigation of nucleotide diversity in polyploid species with a large,
repetitive, and heterozygous genomes [10, 29, 34, 35] and of intra-cultivar genomic
heterogeneity in diploid species [39, 40].
Sequence capture assays have also been designed to target genomic regions
associated with agronomically important traits and capture DNA sequence diversity
in maize [24], rapeseed (Brassica napus) [25, 26], cotton (Gossypium hirsutum)
[27], and cassava (Manihot esculenta) [30] to generate novel data for both research
and breeding activities.
In addition, sequence capture and re-sequencing have been applied to several
tree species, namely loblolly pine (Pinus taeda), black cottonwood (Populus
trichocarpa), and eucalyptus (Eucalyptus globulus), in order to identify sequence
polymorphisms to be used for the generation of a dense reference gene-based
genetic map [31, 32], perform genotyping [23], and develop xylogenesis associated-
trait markers [33].
In order to reconstruct phylogenetic relationships across the Trifolie tribe [43]
and the Compositae family [44] hybridization-based enrichment has been used to
capture sequence variability within low-copy nuclear (LCN) and conserved ortholog
set (COS) markers, respectively.
Mapping-by-sequencing, which combines genetic mapping with targeted-re-
sequencing, has been exploited (1) to identify useful polymorphism to map candi-
date genes in barley (Hordeum vulgare) [41], wild strawberry (Fragaria vesca ssp.
bracteata) [46], and einkorn wheat (Triticum monococcum) [36] and (2) to detect the
precise allocation of Hordeum bulbosum introgression regions in the cultivated H.
vulgare genetic background [42].
WES has been used to re-sequence ethyl methanesulfonate (EMS)- and fast
neutron (FN)-mutagenized plant populations to discover induced mutations in
rice (Oryza sativa), bread wheat (Triticum aestivum) [37], and soybean (Glycine
max) [38].
The use of a closely related reference genome (i.e., Sorghum bicolor) for probe
design has been applied to capture genomic regions of two sugarcane (Saccharum
officinarum) genotypes [28] proving useful for polymorphism discovery in poorly
described species.
Recently, even chloroplast genomes were subjected to target enrichment and
massively parallel sequencing [45]. The strategy the authors adopted is based on
the design of a custom RNA probe set based on the complete sequences of 22
previously sequenced eudicot chloroplast DNAs. Using this probe set an enrichment
experiment was performed on 24 angiosperms (22 eudicots, 2 monocots), which
were subsequently sequenced leading to the generation of complete plastid genomes
with exceptionally high coverage (717 on average).
At present, very few studies are referred to solanaceous crops. In 2013, [49]
described liquid-phase capture method to identify sequence variants within and
across 84 potato (Solanum tuberosum) cultivars they afterwards used to genotype
the same plant material by genotyping-by-sequencing.
Hybridization-Based Enrichment and Next Generation Sequencing to Explore. . . 127

In both tomato (Solanum lycopersicum) and potato, resistance gene enrichment

and sequencing (RenSeq) has been used to discover and annotate novel NB-LRR
genes [47, 48] that will undoubtedly provide breeders with a valuable tool to identify
novel disease resistance traits.

5 Major Bioinformatic Strategies and Tools for

Genome-Wide Sequence Variant Discovery

The demand of targeted re-sequencing is paralleled by the development of

bioinformatic tools to analyze sequence data, with more than 500 tools published
within a span of only 2 years [50]. Some of them were specifically developed to
handle sequence data from targeted re-sequencing experiments and are constantly
being improved and updated. A multi-step analysis performed with a combination
of various tools is a general prerequisite to extract meaningful results from
sequence capture and targeted re-sequencing experiments (Fig. 1). The first
step of the analysis includes the evaluation of read quality. The command-
line tools FastQC (http://www.bioinformatics.babraham.ac.uk/projects/fastqc/),
FASTX-Toolkit (http://hannonlab.cshl.edu/fastx_toolkit/), and Trimmomatic [51],
are often combined during the primary analysis to assessing the overall quality
of a sequencing run, to trim off poor quality bases, and to filter on high quality
scores. The next step is critical and involves the alignments of high quality
reads to a reference genome or transcriptome when present. Read-to-reference
alignments are released in a compressed, indexed, binary form called Binary
sequence Alignment/Map format (BAM) [52].
The tool chosen for the alignment of the reads to the reference as well as the
compositional properties of the reference sequence itself (e.g., presence of low com-
plexity sequences, repetitive regions) affect the number of reads properly aligned
and often influence final coverage and depth values [53]. Different algorithms
dedicated to the mapping of short reads to a reference sequence have been developed
so far [54], being the most popular BWA [55], Bowtie2 [56], and SOAP2 [57].
Alignments of the reads to the reference genome/transcriptome are generally
used to estimate the coverage along target regions and to assess the level of on-
target enrichment efficiency. In this regard, the coverage utility of the bedtools
package [58] is normally used. Given a BAM alignment file and a BED file (a
tab-delimited text file that defines the coordinates of a feature along the reference
sequence) containing target regions, it computes the coverage over defined intervals.
This allows evaluating if the coverage depth is uniform among the re-sequenced
genotypes and at what extent the variation in coverage affects target regions. Indeed,
some off-target reads are expected and this depends on the nature of the target
sequences (e.g., genic, intergenic, etc.) as well as on the enrichment technology
and sequencing platform used.
128 I. Terracciano et al.

Fig. 1 Typical variant calling workflow. Different analysis steps (each object in the figure)
are concatenated to identify reliable sequence polymorphisms and derive meaningful biological
interpretation of the results. The “RR” step is facultative in a variant calling NGS analysis

A BAM file can include reads with the same start and end coordinates. These
might represent PCR duplicates, which should be removed/flagged in the BAM file
since they are not informative and should not be counted as evidence of a putative
variant. The Picard MarkDuplicates (http://picard.sourceforge.net) is the preferred
tool for this task, although it only considers the starting position of the read as a
way to indicate a putative duplicated read. As an alternative, the SAMtools “rmdup”
command can be used [52].
Reads mapping to the edges of InDels often lead to mis-alignments and
produce artifactual mis-matches. Therefore, the local re-alignment of the reads
around InDels is necessary because it helps improve the accuracy of downstream
processing steps. The strategy developed to accomplish this task combines short-
read mapping with an assembly inspired approach to identify a local consensus
Hybridization-Based Enrichment and Next Generation Sequencing to Explore. . . 129

sequence. Programs that implement this approach include SRMA [59] and Indel-
Realigner from the Genome Analysis Toolkit (GATK) [60].
A further improvement may be achieved running the base quality score recalibra-
tion (BQSR) on re-aligned BAM files. One of the most commonly used programs
for BQSR is BaseRecalibrator from the GATK suite [60].
Variant calling, at first glance, may be pretty simple, as it involves the identifi-
cation of sites where one or more samples display possible genomic variations. All
the available tools allow the minimum coverage and minimum variant frequency
threshold to be fixed in order to extract significant variants. Of course additional
parameters can be configured to compute more stringent analyses. The GATK
HaplotypeCaller or UnifiedGenotyper [61], SAMtools mpileup [52], and Freebayes
(https://github.com/ekg/freebayes) are the most widely used programs for sequence
variant calling. The Variant Call File (VCF) format allows the most prevalent types
of sequence variations to be stored. In order to provide easily accessible methods for
working with VCF files the VCFtools program package has been developed [62].
A binary representation of the variant call format (BCF), which is more compact
and much faster to be processed than VCF, has also been implemented [63]. A
limitation of VCF tools is not supporting filtering of polyploidy data, but this can be
accomplished by VCFlib (https://github.com/vcflib/vcflib).
Identifying functionally relevant polymorphisms in a mare magnum of genetic
variations is the major challenge. Annotation of sequence variants includes the
classification of the effects of single nucleotide polymorphisms and insertion–
deletions (e.g., synonymous or non-synonymous SNPs, start-codon gain/loss, stop-
codon gain/loss, frame-shift, etc.) on annotated genes. Annotations can also be
based on the coordinate system used to describe the genomic position of each
polymorphism (e.g., intronic, 50 or 30 un-translated region, upstream, downstream,
inter-genic regions, etc.). In this regard, it is crucial to have an accurate, preferably
gold standard, structural annotation of the reference genome. ANNOVAR [64] and
SnpEff [65] are two of the most used tools in the variant annotation process.
Sequence polymorphisms in the coding regions are frequently associated with
aberrant protein modifications. The interpretation of novel missense mutations
(a type of non-synonymous substitutions) is challenging. Nevertheless, several
computational tools have been developed in order to predict possible impact of an
amino acid substitution on the structure and function of proteins [66, 67]. More
recently, the six best performing tools were combined into a consensus classifier,
called PredictSNP [68], which predictions on protein-related mutations represent a
robust alternative to the predictions delivered by individual tools. More complicated
is the study of splice-site polymorphisms as well as of sequence variations within
intronic regions. It is known that nucleotide variants very close to splice junctions
might alter the splicing pattern of a gene and/or affect splicing efficiency as well as
that introns can harbor functional polymorphisms that can influence the expression
of the genes that host them [69]. However, to the best of our knowledge, no tools
are available for the automatic classification of the effects of such polymorphisms
in plants. By contrast, strategies and tools to support investigations on promoters
are well-defined. Indeed, regulatory regions are generally scanned to identify
transcription factor binding sites (TFBSs). SNPs and InDels within these regions
might modify the TFBS pattern and alter gene expression. A variety of databases
130 I. Terracciano et al.

have been established during the time to collect cis-acting regulatory DNA elements
found in plant promoters. Most of them are now integrated into the most recent
PlantPAN resource [70]. Of course, in silico predictions must be always interpreted
with caution and additional experimental evidences are needed to confirm sequence
variations within the identified alleles.
The last step of the workflow consists in the visual representation of NGS data.
This can be amazingly useful when interpreting the obtained results. The integrative
genomic viewer (IGV) supports users by displaying, along a reference genome,
aligned reads (BAM files) and predicted genetic variants (VCF files) combined with
annotations from the reference [71]. Aggregation of data on a single platform has
significant consequences in the meaningful interpretation of sequencing data and it
is essential to facilitate knowledge discovery.

6 Use of Liquid-Phase Sequence Capture and Target

Enrichment System for Allele Mining in Tomato

Allele mining is a promising strategy to dissect allelic variation at candidate genes

controlling key agronomic traits for potential crop breeding applications. Several
factors determine a successful and efficient allele mining activity; mainly the
availability of (1) information on genome and gene sequences for the species under
investigation; (2) efficient and reliable phenotyping techniques; (3) high-throughput
methods for easy generation of allelic data points; (4) cost-effective sequencing
platforms, and (5) efficient bioinformatic tools for the identification of nucleotide
variations and molecular marker development [72].
Allele mining approaches for traits associated to yield, quality, and important
disease resistance have been successfully applied to many crop species, including
tomato (S. lycopersicum) [73]. Tomato is the most widely consumed vegetable
in the world and its fruits are an important source of bioactive compounds with
known beneficial effects on human health [74]. Carotenoids are the major class
of antioxidant compounds in ripe tomato fruits. They regulate pigmentation of
many fruits and flowers and are involved in photo-reception and photo-protection
mechanisms [75]. In plants, the carotenoid biosynthetic pathway is located into
plastids. In tomato this pathway is highly active during fruit ripening leading to the
accumulation of several metabolites, mainly ’-/“-carotene and lycopene. Because
a wide natural genetic variability associated with the accumulation of carotenoid
pigments in the fruit exists across tomato species [76, 77], its exploration can be
very useful to undertake breeding programs for tomato fruits bio-fortification [78].
The availability of the tomato genome sequence [79], combined with the existing
genetic resources and genomic tools, has undoubtedly expedited the investigation
on the genetic variability in large populations of individuals to identify sequence
variations across candidate genes.
Aiming to contribute in this scenario, we are performing a research activity
in order to capture interesting genetic variation affecting genes responsible for
carotenoid accumulation in tomato fruits.
Hybridization-Based Enrichment and Next Generation Sequencing to Explore. . . 131

We decided to apply a liquid-phase sequence capture followed by Illumina re-

sequencing to target a panel of genes responsible for “-carotene and lycopene
accumulation in ripe tomato berries.
The version 2.50 of the S. lycopersicum reference genome, the 2.40 iTAG
(international Tomato Annotation Group) annotation, and the Agilent’s Sure design
software have been used to generate 120mer RNA baits to cover all the target genes
and their the regulatory regions. Then, the SureSelect target enrichment system
(Agilent Technologies) for Illumina paired-end sequencing has been used to capture
230 kb target region in a panel of 48 genotypes differing for carotenoid content
in the ripe tomato fruits as determined through RP-HPLC analysis. Illumina data
were processed according to the workflow described in the previous paragraph.
About 10,000 polymorphisms, including both SNPs and InDels, were identified
and annotated to predict their biological effects (Terracciano et al.; manuscript in
preparation). The association between the identified genotypic variation and the
observed phenotypic variability is ongoing.
We are going to provide experimental validations of interesting mutations that
we feel could be employed to generate improved tomato varieties for fruit quality.

7 Conclusions

Liquid- or solid-phase sequence capture and target enrichment coupled with NGS
have been proven reliable in the identification of sequence polymorphisms in whole
exomes, target genes, or genomic regions of many plant species. The demand of
targeted re-sequencing is constantly growing and requires significant effort in data
analysis and management. Bioinformatic strategies are essential to extract meaning-
ful results from raw sequence data. Although all the steps of the complex workflow
are well defined, the tools developed to accomplish basic tasks are constantly being
improved and updated. Nevertheless, challenges associated with data analysis can be
taken on with confidence. Indeed, several applications intended to accelerate plant-
breeding activities for crop improvement can benefit from using this technology:
these include genotyping, SNP marker development and biodiversity exploration,
mapping-by-sequencing, etc. An additional application of sequence capture is the
identification and characterization of novel alleles from non-reference genomes. By
describing our research activity on the identification of sequence variation across a
panel of tomato genotypes at candidate genes controlling carotenoid biosynthesis,
we demonstrated that in solution-based hybridization method could be successfully
applied to detect and study the effect of novel alleles in economically important
crops.

Acknowledgements This work was carried out in the frame of the “GenoPom-pro -
Integrating post-genomic platforms to enhance the tomato production chain” project
(PON02_00395_3082360) and is supported by the PON R&C 2007–2013 grant funded by the
Italian Ministry of Education, University and Research in cooperation with the European Funds
for the Regional Development (FESR).
132 I. Terracciano et al.

References

1. Hashmi, U., Shafqat, S., Khan, F., Majid, M., Hussain, H., Kazi, A.G., John, R., Ahmad, P.:
Plant exomics: concepts, applications and methodologies in crop improvement. Plant Signal.
Behav. 10(1), e976152 (2015). doi:10.4161/15592324.2014.976152
2. Warr, A., Robert, C., Hume, D., Archibald, A., Deeb, N., Watson, M.: Exome sequencing:
current and future perspectives. G3 5(8), 1543–1550 (2015). doi:10.1534/g3.115.018564
3. Cronn, R., Knaus, B.J., Liston, A., Maughan, P.J., Parks, M., Syring, J.V., Udall, J.: Targeted
enrichment strategies for next-generation plant biology. Am. J. Bot. 99(2), 291–311 (2012).
doi:10.3732/ajb.1100356
4. Hodges, E., Xuan, Z., Balija, V., Kramer, M., Molla, M.N., Smith, S.W., Middle, C.M.,
Rodesch, M.J., Albert, T.J., Hannon, G.J., McCombie, W.R.: Genome-wide in situ exon capture
for selective resequencing. Nat. Genet. 39(12), 1522–1527 (2007). doi:10.1038/ng.2007.42
5. Mamanova, L., Coffey, A.J., Scott, C.E., Kozarewa, I., Turner, E.H., Kumar, A., Howard, E.,
Shendure, J., Turner, D.J.: Target-enrichment strategies for next-generation sequencing. Nat.
Methods 7(2), 111–118 (2010). doi:10.1038/nmeth.1419
6. Mertes, F., Elsharawy, A., Sauer, S., van Helvoort, J.M., van der Zaag, P.J., Franke,
A., Nilsson, M., Lehrach, H., Brookes, A.J.: Targeted enrichment of genomic DNA
regions for next-generation sequencing. Brief. Funct. Genomics 10(6), 374–386 (2011).
doi:10.1093/bfgp/elr033
7. Gnirke, A., Melnikov, A., Maguire, J., Rogov, P., LeProust, E.M., Brockman, W., Fennell,
T., Giannoukos, G., Fisher, S., Russ, C., Gabriel, S., Jaffe, D.B., Lander, E.S., Nusbaum,
C.: Solution hybrid selection with ultra-long oligonucleotides for massively parallel targeted
sequencing. Nat. Biotechnol. 27(2), 182–189 (2009). doi:10.1038/nbt.1523
8. Okou, D.T., Steinberg, K.M., Middle, C., Cutler, D.J., Albert, T.J., Zwick, M.E.: Microarray-
based genomic selection for high-throughput resequencing. Nat. Methods 4(11), 907–909
(2007). doi:10.1038/nmeth1109
9. Fu, Y., Springer, N.M., Gerhardt, D.J., Ying, K., Yeh, C.T., Wu, W., Swanson-Wagner,
R., D’Ascenzo, M., Millard, T., Freeberg, L., Aoyama, N., Kitzman, J., Burgess, D.,
Richmond, T., Albert, T.J., Barbazuk, W.B., Jeddeloh, J.A., Schnable, P.S.: Repeat subtraction-
mediated sequence capture from a complex genome. Plant J. 62(5), 898–909 (2010).
doi:10.1111/j.1365-313X.2010.04196.x
10. Saintenac, C., Jiang, D., Akhunov, E.D.: Targeted analysis of nucleotide and copy number
variation by exon capture in allotetraploid wheat genome. Genome Biol. 12(9), R88 (2011).
doi:10.1186/gb-2011-12-9-r88
11. Martin, L.B., Fei, Z., Giovannoni, J.J., Rose, J.K.: Catalyzing plant science research with RNA-
seq. Front. Plant Sci. 4, 66 (2013). doi:10.3389/fpls.2013.00066
12. Egan, A.N., Schlueter, J., Spooner, D.M.: Applications of next-generation sequencing in plant
biology. Am. J. Bot. 99(2), 175–185 (2012). doi:10.3732/ajb.1200020
13. Baird, N.A., Etter, P.D., Atwood, T.S., Currey, M.C., Shiver, A.L., Lewis, Z.A., Selker, E.U.,
Cresko, W.A., Johnson, E.A.: Rapid SNP discovery and genetic mapping using sequenced
RAD markers. PLoS One 3(10), e3376 (2008). doi:10.1371/journal.pone.0003376
14. Rowe, H.C., Renaut, S., Guggisberg, A.: RAD in the realm of next-generation sequencing
technologies. Mol. Ecol. 20(17), 3499–3502 (2011)
15. Maughan, P.J., Yourstone, S.M., Jellen, E.N., Udall, J.A.: SNP discovery via genomic
reduction, barcoding, and 454-pyrosequencing in Amaranth. Plant Genome J. 2(3), 260 (2009).
doi:10.3835/plantgenome2009.08.0022
16. He, J., Zhao, X., Laroche, A., Lu, Z.X., Liu, H., Li, Z.: Genotyping-by-sequencing (GBS), an
ultimate marker-assisted selection (MAS) tool to accelerate plant breeding. Front. Plant Sci. 5,
484 (2014). doi:10.3389/fpls.2014.00484
Hybridization-Based Enrichment and Next Generation Sequencing to Explore. . . 133

17. Uribe-Convers, S., Settles, M.L., Tank, D.C.: A phylogenomic approach based on PCR target
enrichment and high throughput sequencing: resolving the diversity within the South American
species of Bartsia L. (Orobanchaceae). PLoS one 11(2), e0148203 (2016). doi:10.1371/jour-
nal.pone.0148203
18. Durstewitz, G., Polley, A., Plieske, J., Luerssen, H., Graner, E.M., Wieseke, R., Ganal, M.W.:
SNP discovery by amplicon sequencing and multiplex SNP genotyping in the allopolyploid
species Brassica napus. Genome 3(11), 948–956 (2010). doi:10.1139/G10-079
19. Chilamakuri, C.S., Lorenz, S., Madoui, M.A., Vodak, D., Sun, J., Hovig, E., Myklebost,
O., Meza-Zepeda, L.A.: Performance comparison of four exome capture systems for deep
sequencing. BMC Genomics 15, 449 (2014). doi:10.1186/1471-2164-15-449
20. Parla, J.S., Iossifov, I., Grabill, I., Spector, M.S., Kramer, M., McCombie, W.R.: A comparative
analysis of exome capture. Genome Biol. 12(9), R97 (2011). doi:10.1186/gb-2011-12-9-r97
21. Dohm, J.C., Lottaz, C., Borodina, T., Himmelbauer, H.: Substantial biases in ultra-short read
data sets from high-throughput DNA sequencing. Nucleic Acids Res. 36(16), e105 (2008).
doi:10.1093/nar/gkn425
22. Aird, D., Ross, M.G., Chen, W.S., Danielsson, M., Fennell, T., Russ, C., Jaffe, D.B., Nusbaum,
C., Gnirke, A.: Analyzing and minimizing PCR amplification bias in Illumina sequencing
libraries. Genome Biol. 12(2), R18 (2011). doi:10.1186/gb-2011-12-2-r18
23. Zhou, L., Holliday, J.A.: Targeted enrichment of the black cottonwood (Populus
trichocarpa) gene space using sequence capture. BMC Genomics 13, 703 (2012).
doi:10.1186/1471-2164-13-703
24. Muraya, M.M., Schmutzer, T., Ulpinnis, C., Scholz, U., Altmann, T.: Targeted sequencing
reveals large-scale sequence polymorphism in Maize candidate genes for biomass production
and composition. PLoS One 10(7), e0132120 (2015). doi:10.1371/journal.pone.0132120
25. Clarke, W.E., Parkin, I.A., Gajardo, H.A., Gerhardt, D.J., Higgins, E., Sidebottom, C.,
Sharpe, A.G., Snowdon, R.J., Federico, M.L., Iniguez-Luy, F.L.: Genomic DNA enrichment
using sequence capture microarrays: a novel approach to discover sequence nucleotide
polymorphisms (SNP) in Brassica napus L. PLoS One 8(12), e81992 (2013). doi:10.1371/jour-
nal.pone.0081992
26. Schiessl, S., Samans, B., Huttel, B., Reinhard, R., Snowdon, R.J.: Capturing sequence variation
among flowering-time regulatory gene homologs in the allopolyploid crop species Brassica
napus. Front. Plant Sci. 5, 404 (2014). doi:10.3389/fpls.2014.00404
27. Salmon, A., Udall, J.A., Jeddeloh, J.A., Wendel, J.: Targeted capture of homoeolo-
gous coding and noncoding sequence in polyploid cotton. G3 2(8), 921–930 (2012).
doi:10.1534/g3.112.003392
28. Bundock, P.C., Casu, R.E., Henry, R.J.: Enrichment of genomic DNA for polymorphism
detection in a non-model highly polyploid crop plant. Plant Biotechnol. J. 10(6), 657–667
(2012). doi:10.1111/j.1467-7652.2012.00707.x
29. Evans, J., Kim, J., Childs, K.L., Vaillancourt, B., Crisovan, E., Nandety, A., Gerhardt,
D.J., Richmond, T.A., Jeddeloh, J.A., Kaeppler, S.M., Casler, M.D., Buell, C.R.: Nucleotide
polymorphism and copy number variant detection using exome capture and next-generation
sequencing in the polyploid grass Panicum virgatum. Plant J. 79(6), 993–1008 (2014).
doi:10.1111/tpj.12601
30. Pootakham, W., Shearman, J.R., Ruang-Areerate, P., Sonthirod, C., Sangsrakru, D., Jomchai,
N., Yoocha, T., Triwitayakorn, K., Tragoonrung, S., Tangphatsornruang, S.: Large-scale SNP
discovery through RNA sequencing and SNP genotyping by targeted enrichment sequencing
in cassava (Manihot esculenta Crantz). PLoS One 9(12), e116028 (2014). doi:10.1371/jour-
nal.pone.0116028
31. Neves, L.G., Davis, J.M., Barbazuk, W.B., Kirst, M.: Whole-exome targeted sequencing of the
uncharacterized pine genome. Plant J. 75(1), 146–156 (2013). doi:10.1111/tpj.12193
32. Neves, L.G., Davis, J.M., Barbazuk, W.B., Kirst, M.: A high-density gene map of loblolly
pine (Pinus taeda L.) based on exome sequence capture genotyping. G3 4(1), 29–37 (2014).
doi:10.1534/g3.113.008714
134 I. Terracciano et al.

33. Dasgupta, M.G., Dharanishanthi, V., Agarwal, I., Krutovsky, K.V.: Development of genetic
markers in Eucalyptus species by target enrichment and exome sequencing. PLoS One 10(1),
e0116528 (2015). doi:10.1371/journal.pone.0116528
34. Allen, A.M., Barker, G.L., Wilkinson, P., Burridge, A., Winfield, M., Coghill, J., Uauy,
C., Griffiths, S., Jack, P., Berry, S., Werner, P., Melichar, J.P., McDougall, J., Gwilliam,
R., Robinson, P., Edwards, K.J.: Discovery and development of exome-based, co-dominant
single nucleotide polymorphism markers in hexaploid wheat (Triticum aestivum L.). Plant
Biotechnol. J. 11(3), 279–295 (2013). doi:10.1111/pbi.12009
35. Winfield, M.O., Wilkinson, P.A., Allen, A.M., Barker, G.L., Coghill, J.A., Burridge, A., Hall,
A., Brenchley, R.C., D’Amore, R., Hall, N., Bevan, M.W., Richmond, T., Gerhardt, D.J.,
Jeddeloh, J.A., Edwards, K.J.: Targeted re-sequencing of the allohexaploid wheat exome. Plant
Biotechnol. J. 10(6), 733–742 (2012). doi:10.1111/j.1467-7652.2012.00713.x
36. Gardiner, L.J., Gawronski, P., Olohan, L., Schnurbusch, T., Hall, N., Hall, A.: Using genic
sequence capture in combination with a syntenic pseudo genome to map a deletion mutant in
a wheat species. Plant J. 80(5), 895–904 (2014). doi:10.1111/tpj.12660
37. Henry, I.M., Nagalakshmi, U., Lieberman, M.C., Ngo, K.J., Krasileva, K.V., Vasquez-Gross,
H., Akhunova, A., Akhunov, E., Dubcovsky, J., Tai, T.H., Comai, L.: Efficient genome-wide
detection and cataloging of EMS-induced mutations using exome capture and next-generation
sequencing. Plant Cell 26(4), 1382–1397 (2014). doi:10.1105/tpc.113.121590
38. Bolon, Y.T., Haun, W.J., Xu, W.W., Grant, D., Stacey, M.G., Nelson, R.T., Gerhardt, D.J.,
Jeddeloh, J.A., Stacey, G., Muehlbauer, G.J., Orf, J.H., Naeve, S.L., Stupar, R.M., Vance, C.P.:
Phenotypic and genomic analyses of a fast neutron mutant population resource in soybean.
Plant Physiol. 156(1), 240–253 (2011). doi:10.1104/pp.110.170811
39. Haun, W.J., Hyten, D.L., Xu, W.W., Gerhardt, D.J., Albert, T.J., Richmond, T., Jeddeloh, J.A.,
Jia, G., Springer, N.M., Vance, C.P., Stupar, R.M.: The composition and origins of genomic
variation among individuals of the soybean reference cultivar Williams 82. Plant Physiol.
155(2), 645–655 (2011). doi:10.1104/pp.110.166736
40. Mascher, M., Richmond, T.A., Gerhardt, D.J., Himmelbach, A., Clissold, L., Sampath,
D., Ayling, S., Steuernagel, B., Pfeifer, M., D’Ascenzo, M., Akhunov, E.D., Hedley, P.E.,
Gonzales, A.M., Morrell, P.L., Kilian, B., Blattner, F.R., Scholz, U., Mayer, K.F., Flavell,
A.J., Muehlbauer, G.J., Waugh, R., Jeddeloh, J.A., Stein, N.: Barley whole exome capture:
a tool for genomic research in the genus Hordeum and beyond. Plant J. 76(3), 494–505 (2013).
doi:10.1111/tpj.12294
41. Pankin, A., Campoli, C., Dong, X., Kilian, B., Sharma, R., Himmelbach, A., Saini, R.,
Davis, S.J., Stein, N., Schneeberger, K., von Korff, M.: Mapping-by-sequencing identifies
HvPHYTOCHROME C as a candidate gene for the early maturity 5 locus modulating the
circadian clock and photoperiodic flowering in barley. Genetics 198(1), 383–396 (2014).
doi:10.1534/genetics.114.165613
42. Wendler, N., Mascher, M., Noh, C., Himmelbach, A., Scholz, U., Ruge-Wehling, B., Stein,
N.: Unlocking the secondary gene-pool of barley with next-generation sequencing. Plant
Biotechnol. J. 12(8), 1122–1131 (2014). doi:10.1111/pbi.12219
43. de Sousa, F., Bertrand, Y.J., Nylinder, S., Oxelman, B., Eriksson, J.S., Pfeil, B.E.: Phylo-
genetic properties of 50 nuclear loci in Medicago (Leguminosae) generated using multi-
plexed sequence capture and next-generation sequencing. PLoS One 9(10), e109704 (2014).
doi:10.1371/journal.pone.0109704
44. Mandel, J.R., Dikow, R.B., Funk, V.A., Masalia, R.R., Staton, S.E., Kozik, A., Michelmore,
R.W., Rieseberg, L.H., Burke, J.M.: A target enrichment method for gathering phylogenetic
information from hundreds of loci: an example from the Compositae. Appl. Plant Sci. 2(2)
(2014). doi:10.3732/apps.1300085
45. Stull, G.W., Moore, M.J., Mandala, V.S., Douglas, N.A., Kates, H.R., Qi, X., Brockington,
S.F., Soltis, P.S., Soltis, D.E., Gitzendanner, M.A.: A targeted enrichment strategy for
massively parallel sequencing of angiosperm plastid genomes. Appl. Plant Sci. 1(2) (2013).
doi:10.3732/apps.1200497
Hybridization-Based Enrichment and Next Generation Sequencing to Explore. . . 135

46. Tennessen, J.A., Govindarajulu, R., Liston, A., Ashman, T.L.: Targeted sequence capture
provides insight into genome structure and genetics of male sterility in a gynodioecious
diploid strawberry, Fragaria vesca ssp. bracteata (Rosaceae). G3 3(8), 1341–1351 (2013).
doi:10.1534/g3.113.006288
47. Jupe, F., Witek, K., Verweij, W., Sliwka, J., Pritchard, L., Etherington, G.J., Maclean, D., Cock,
P.J., Leggett, R.M., Bryan, G.J., Cardle, L., Hein, I., Jones, J.D.: Resistance gene enrichment
sequencing (RenSeq) enables reannotation of the NB-LRR gene family from sequenced plant
genomes and rapid mapping of resistance loci in segregating populations. Plant J. 76(3), 530–
544 (2013). doi:10.1111/tpj.12307
48. Andolfo, G., Jupe, F., Witek, K., Etherington, G.J., Ercolano, M.R., Jones, J.D.: Defining the
full tomato NB-LRR resistance gene repertoire using genomic and cDNA RenSeq. BMC Plant
Biol. 14, 120 (2014). doi:10.1186/1471-2229-14-120
49. Uitdewilligen, J.G., Wolters, A.M., D’Hoop B.B., Borm, T.J., Visser, R.G., van Eck, H.J.:
A next-generation sequencing method for genotyping-by-sequencing of highly heterozygous
autotetraploid potato. PLoS One 8(5), e62355 (2013). doi:10.1371/journal.pone.0062355
50. Li, J.W., Robison, K., Martin, M., Sjodin, A., Usadel, B., Young, M., Olivares, E.C., Bolser,
D.M.: The SEQanswers wiki: a wiki database of tools for high-throughput sequencing analysis.
Nucleic Acids Res. 40(Database issue), D1313–D1317 (2012). doi:10.1093/nar/gkr1058
51. Bolger, A.M., Lohse, M., Usadel, B.: Trimmomatic: a flexible trimmer for Illumina sequence
data. Bioinformatics 30(15), 2114–2120 (2014). doi:10.1093/bioinformatics/btu170
52. Li, H., Handsaker, B., Wysoker, A., Fennell, T., Ruan, J., Homer, N., Marth, G., Abecasis,
G., Durbin, R.: Genome project data processing, S.: the sequence alignment/map format and
SAMtools. Bioinformatics 25(16), 2078–2079 (2009). doi:10.1093/bioinformatics/btp352
53. Sims, D., Sudbery, I., Ilott, N.E., Heger, A., Ponting, C.P.: Sequencing depth and cov-
erage: key considerations in genomic analyses. Nat. Rev. Genet. 15(2), 121–132 (2014).
doi:10.1038/nrg3642
54. Hatem, A., Bozdag, D., Toland, A.E., Catalyurek, U.V.: Benchmarking short sequence
mapping tools. BMC Bioinf. 14, 184 (2013). doi:10.1186/1471-2105-14-184
55. Li, H., Durbin, R.: Fast and accurate short read alignment with Burrows-Wheeler transform.
Bioinformatics 25(14), 1754–1760 (2009). doi:10.1093/bioinformatics/btp324
56. Langmead, B., Salzberg, S.L.: Fast gapped-read alignment with Bowtie 2. Nat. Methods 9(4),
357–359 (2012). doi:10.1038/nmeth.1923
57. Li, R., Yu, C., Li, Y., Lam, T.W., Yiu, S.M., Kristiansen, K., Wang, J.: SOAP2: an
improved ultrafast tool for short read alignment. Bioinformatics 25(15), 1966–1967 (2009).
doi:10.1093/bioinformatics/btp336
58. Quinlan, A.R., Hall, I.M.: BEDTools: a flexible suite of utilities for comparing genomic
features. Bioinformatics 26(6), 841–842 (2010). doi:10.1093/bioinformatics/btq033
59. Homer, N., Nelson, S.F.: Improved variant discovery through local re-alignment of short-
read next-generation sequencing data using SRMA. Genome Biol. 11(10), R99 (2010).
doi:10.1186/gb-2010-11-10-r99
60. McKenna, A., Hanna, M., Banks, E., Sivachenko, A., Cibulskis, K., Kernytsky, A., Garimella,
K., Altshuler, D., Gabriel, S., Daly, M., DePristo, M.A.: The Genome Analysis Toolkit: a
MapReduce framework for analyzing next-generation DNA sequencing data. Genome Res.
20(9), 1297–1303 (2010). doi:10.1101/gr.107524.110
61. Van der Auwera, G.A., Carneiro, M.O., Hartl, C., Poplin, R., Del Angel, G., Levy-Moonshine,
A., Jordan, T., Shakir, K., Roazen, D., Thibault, J., Banks, E., Garimella, K.V., Altshuler, D.,
Gabriel, S., DePristo, M.A.: From FastQ data to high confidence variant calls: the Genome
Analysis Toolkit best practices pipeline. Curr. Protoc. Bioinformatics 11(1110), 11.10.11–
11.10.33 (2013). doi:10.1002/0471250953.bi1110s43
62. Danecek, P., Auton, A., Abecasis, G., Albers, C.A., Banks, E., DePristo, M.A., Handsaker,
R.E., Lunter, G., Marth, G.T., Sherry, S.T., McVean, G., Durbin, R., Genomes Project
Analysis, G.: The variant call format and VCFtools. Bioinformatics 27(15), 2156–2158 (2011).
doi:10.1093/bioinformatics/btr330
136 I. Terracciano et al.

63. Li, H.: A statistical framework for SNP calling, mutation discovery, association mapping and
population genetical parameter estimation from sequencing data. Bioinformatics 27(21), 2987–
2993 (2011). doi:10.1093/bioinformatics/btr509
64. Yang, H., Wang, K.: Genomic variant annotation and prioritization with ANNOVAR and
wANNOVAR. Nat. Protoc. 10(10), 1556–1566 (2015). doi:10.1038/nprot.2015.105
65. Cingolani, P., Platts, A., le Wang, L., Coon, M., Nguyen, T., Wang, L., Land, S.J., Lu,
X., Ruden, D.M.: A program for annotating and predicting the effects of single nucleotide
polymorphisms, SnpEff: SNPs in the genome of Drosophila melanogaster strain w1118; iso-2;
iso-3. Fly 6(2), 80–92 (2012). doi:10.4161/fly.19695
66. Adzhubei, I.A., Schmidt, S., Peshkin, L., Ramensky, V.E., Gerasimova, A., Bork, P.,
Kondrashov, A.S., Sunyaev, S.R.: A method and server for predicting damaging missense
mutations. Nat. Methods 7(4), 248–249 (2010). doi:10.1038/nmeth0410-248
67. Ng, P.C., Henikoff, S.: Predicting the effects of amino acid substitutions
on protein function. Annu. Rev. Genomics Hum. Genet. 7, 61–80 (2006).
doi:10.1146/annurev.genom.7.080505.115630
68. Bendl, J., Stourac, J., Salanda, O., Pavelka, A., Wieben, E.D., Zendulka, J., Brezovsky,
J., Damborsky, J.: PredictSNP: robust and accurate consensus classifier for prediction of
disease-related mutations. PLoS Comput. Biol. 10(1), e1003440 (2014). doi:10.1371/jour-
nal.pcbi.1003440
69. Cooper, D.N.: Functional intronic polymorphisms: Buried treasure awaiting discovery within
our genes. Hum. Genomics 4(5), 284–288 (2010)
70. Chow, C.N., Zheng, H.Q., Wu, N.Y., Chien, C.H., Huang, H.D., Lee, T.Y., Chiang-Hsieh,
Y.F., Hou, P.F., Yang, T.Y., Chang, W.C.: PlantPAN 2.0: an update of plant promoter analysis
navigator for reconstructing transcriptional regulatory networks in plants. Nucleic Acids Res.
44(D1), D1154–D1160 (2016). doi:10.1093/nar/gkv1035
71. Thorvaldsdottir, H., Robinson, J.T., Mesirov, J.P.: Integrative genomics viewer (IGV): high-
performance genomics data visualization and exploration. Brief. Bioinform. 14(2), 178–192
(2013). doi:10.1093/bib/bbs017
72. Kumar, G.R., Sakthivel, K., Sundaram, R.M., Neeraja, C.N., Balachandran, S.M., Rani, N.S.,
Viraktamath, B.C., Madhav, M.S.: Allele mining in crops: prospects and potentials. Biotechnol.
Adv. 28(4), 451–461 (2010). doi:10.1016/j.biotechadv.2010.02.007
73. Rose, L.E., Langley, C.H., Bernal, A.J., Michelmore, R.W.: Natural variation in the Pto
pathogen resistance gene within species of wild tomato (Lycopersicon)I. Functional analysis
of Pto alleles. Genetics 171(1), 345–357 (2005). doi:10.1534/genetics.104.039339
74. Raiola, A., Rigano, M.M., Calafiore, R., Frusciante, L., Barone, A.: Enhancing the health-
promoting effects of tomato fruit for biofortified food. Mediators Inflamm. 2014, 139873
(2014). doi:10.1155/2014/139873
75. Giuliano, G.: Plant carotenoids: genomics meets multi-gene engineering. Curr. Opin. Plant
Biol. 19, 111–117 (2014). doi:10.1016/j.pbi.2014.05.006
76. Kavitha, P., Shivashankara, K.S., Rao, V.K., Sadashiva, A.T., Ravishankar, K.V., Sathish,
G.J.: Genotypic variability for antioxidant and quality parameters among tomato cultivars,
hybrids, cherry tomatoes and wild species. J. Sci. Food Agric. 94(5), 993–999 (2014).
doi:10.1002/jsfa.6359
77. Ruggieri, V., Francese, G., Sacco, A., D’Alessandro, A., Rigano, M.M., Parisi, M., Milone,
M., Cardi, T., Mennella, G., Barone, A.: An association mapping approach to identify
favourable alleles for tomato fruit quality breeding. BMC Plant Biol. 14, 337 (2014).
doi:10.1186/s12870-014-0337-9
78. Liu, L., Shao, Z., Zhang, M., Wang, Q.: Regulation of carotenoid metabolism in tomato. Mol.
Plant 8(1), 28–39 (2015). doi:10.1016/j.molp.2014.11.006
79. Tomato Genome Consortium: The tomato genome sequence provides insights into fleshy fruit
evolution. Nature 485(7400), 635–641 (2012). doi:10.1038/nature11119
DecontaMiner: A Pipeline for the Detection
and Analysis of Contaminating Sequences
in Human NGS Sequencing Data

Ilaria Granata , Mara Sangiovanni , and Mario Guarracino

Abstract Reads alignment is an essential step of next generation sequencing)

data analyses. One challenging issue is represented by unmapped reads that are
usually discarded and considered as not informative. Instead, it is important to fully
understand the source of those reads, to assess the quality of the whole experiment.
Moreover, it is of interest to get some insights on possible “contamination”
from non-human sequences (e.g., viruses, bacteria, and fungi). Contamination
may take place during the experimental procedures leading to sequencing, or be
due to the presence of microorganisms infecting the sampled tissues. Here we
propose a pipeline for the detection of viral, bacterial, and fungi contamination
in human sequenced data. Similarities between input reads (query) and putative
contaminating organism sequences (subject) are detected using a local alignment
strategy (MegaBLAST). For each organism database DecontaMiner provides two
main output files: one containing all the reads matching only a single organism;
the second one containing the “ambiguous” matching reads. In both files, data
is sorted by organism and classified by taxonomic group. Low quality, unaligned
sequences, and those discarded by user criteria are also provided as output. Other
information and summary statistics on the number of matched/filtered/discarded
reads and organisms are generated. This pipeline has successfully detected foreign
sequences in human Cancer RNA-seq data.

Keywords Contaminating sequences • Unmapped reads • NGS data

I. Granata • M. Sangiovanni ()

ICAR-CNR, Via Pietro Castellino 111, 80131 Napoli, Italy
e-mail: [email protected]; [email protected]
M. Guarracino
Laboratory for Genomics, Transcriptomics and Proteomics (Lab-GTP), High Performance
Computing and Networking Institute (ICAR), National Research Council (CNR),
Via Pietro Castellino 111, Naples, Italy
e-mail: [email protected]

© Springer International Publishing Switzerland 2016 137

A. Rogato et al. (eds.), Dynamics of Mathematical Models in Biology,
DOI 10.1007/978-3-319-45723-9_11
138 I. Granata et al.

1 Introduction

The study of the human genome and its relationship with the environment is a
crucial task in the context of modern biology.
The application of next generation sequencing technologies allows to charac-
terize the genome-wide map of organisms. Genome investigation has been made
possible by the construction of the reference genomes. Sequencing experiments
produce a large amount of small sequences that have to be mapped to the reference.
The alignment is probably the most challenging step of next generation sequencing
(NGS) data analyses. It allows to obtain several information—such as read density,
gene lists, and variant lists—crucial to the definition of the biological meaning
underlying the data.
Typically the amount of reads that correctly map onto the human reference
genome ranges between 70 and 90 % [1] leaving in some cases a consistent fraction
of unmapped reads. Underestimating this portion may determine loss of precious
information. Unmapped reads can be explained by errors during sequencing pro-
tocols, by the presence of repeat elements difficult to map, by novel transcripts
that can be investigated by de novo assembly, and lastly, they can derive from non-
human sequences. Indeed, microorganisms contamination can occur during samples
processing or can be part of the normal or pathological tissues microbiome [2].
The interest in detecting microorganisms-derived sequences has grown up
together with the spread of high-throughput approaches, allowing the extraction of
information both about the quality of the experimental procedures and about the
link between diseases and infections. The main appeal of these investigations is
represented by the possibility to find new pathogen-disease associations. In literature
there are many evidences which underline the importance of detecting contaminat-
ing organisms. Worth to note are the detection of polyomavirus in human Merkel
cell carcinoma [3] and a novel Old World arenavirus in a cluster of patients with
fatal transplant-associated disease [4]. Assembly of a novel bacterial draft genome
starting from tissue specimens sequencing of cord colitis patients suggested an
opportunistic pathogenic role for Bradyrhizobium enterica in humans [5].
Besides, environmental contaminations are routinely found in NGS datasets.
Downstream contaminations or cross-contaminations can compromise the reliabil-
ity of the whole experimental procedure. Strong et al. detected bacterial sequences,
belonging to different taxa, in cell line data coming from different sequencing
experiments and suggested the idea that a good portion of these bacterial reads
did not derive from the specimens themselves but from downstream contamination.
This suggestion has been supported by the detection of bacterial sequences in
polyA RNA-seq [6]. Indeed, the polyA selection step should remove upstream
contamination since bacteria are poorly polyadenylated. Moreover, to strengthen
the hypothesis of downstream contamination occurrence, the authors analyzed


Authors are contributed equally.
DecontaMiner: A Pipeline for the Detection of Contaminating Sequences 139

RNA-sequencing data of five Epstein-Barr virus (EBV)-positive lymphoblastoid

cell lines obtained in six different Illumina laboratories. Across these labs the level
of bacterial reads per million human mapped reads (RPMHs) differed by as much
as 30-fold, while the transcript levels of the EB virus were similar.
Furthermore, another study also confirmed this laboratory-peculiar contamina-
tion, showing that different sequencing centers had specific signatures of contami-
nating genomes as “time stamps” [7]. Unmapped ChIP-Seq reads from A. thaliana,
Z. mays, H. sapiens, and D. melanogaster datasets were investigated and found
contaminated by foreign sequences. Taxonomic classification of these reads allowed
authors to define the contaminants and to calculate the relative abundance for each
dataset [8].
Several tools, based on different computational approaches, have been developed
and used for the detection of pathogens in high-throughput sequencing data,
especially in cancer samples. In particular, PathSeq [9] and CaPSID [10] are worth
mentioning. Both are available as integrated open source softwares.
PathSeq applies a subtraction approach in which the reads are aligned on six
different human genomes. After, it uses local aligners such as Mega BLAST
and BLASTN [11] to re-align reads to microbial reference sequences and to
two additional human sequence databases. PathSeq is implemented in a cloud-
computing environment. However, the PathSeq pipeline can be computationally
intensive, mostly due to the numerous subtraction steps. CaPSID overcame this limit
using a single human reference genome with splice junctions. Although CaPSID
might face the risk to fail the correct alignment, it provides a large reduction in
elaboration time. Furthermore, PathSeq discards the ambiguous reads that map both
to human and pathogen genomes, while CaPSID stores them in a database.
It should be noted that PathSeq also requires a commercial computing platform
(i.e., Amazon Elastic Compute Cloud, EC2) to be used. CaPSID does not have
this kind of restriction but it requires two files in bam format as input, obtained by
the user with a separate alignment software. The user should take care of aligning
the sequences both to human and to each pathogen (bacteria, viruses, and fungi)
reference genome of interest, thus performing the most computationally intensive
steps before CaPSID. Hence, the CaPSID pipeline is lighter and faster, and it can
provide even gene annotations and a user-friendly web application that integrates a
genome browser.
Another cloud-compatible bioinformatics pipeline aimed to pathogen discovery
is SURPI (“Sequence-based Ultrarapid Pathogen Identification”) [12], which pro-
vides a very useful and complete tool for the analysis of complex metagenomic NGS
data. However, its purpose is the detection of microorganisms from complex clinical
metagenomic samples open to the environment, using the entire NCBI nt and/or
NCBI nr protein databases in comprehensive mode. The algorithm is particularly
sensitive but, as consequence, the pipeline is likely not appropriate for a rapid
analysis of the unmapped reads.
As far as we know, all the pipelines mentioned above are designed to analyze
data primarily aimed to the detection of pathogens in human samples. Due to this,
some of them, such as PathSeq and SURPI, provide intensive pipeline including
140 I. Granata et al.

alignment to host genome, while CaPSID, in order to reduce the required time and
computational efforts, works on BAM files provided by the user, containing the
resulted alignments to the human and to all the pathogen reference sequences.
Here we propose DecontaMiner, a pipeline designed and developed to detect
contaminating sequences in NGS data. Our main purpose is to understand the nature
of those reads that fail to map to the reference genome, as well as to provide
an automatic pipeline that allows the quality filtering and the processing of these
sequences.
From the detected output it is straightforward to extract information about the
eventual samples contamination and/or tissue infection. As in the above-mentioned
papers [6–8] the experimental setup and the study of the detected microorganism
species might suggest the possible contamination sources. In general, it is not possi-
ble to automatically discriminate between upstream and downstream contamination.
Concluding, it can be said that DecontaMiner lies in the middle between
the complex, intensive pipelines of PathSeq and SURPI, and the post-alignment
approach of CaPSID.

2 DecontaMiner Pipeline

The DecontaMiner pipeline is a suite composed of several command-line tools

wrapped together to identify, through digital subtraction, non-human nucleotide
sequences generated by high-throughput sequencing of RNA or DNA samples. It
is mainly written using Bash scripting and the Perl language. It requires in input the
BAM files or the raw fastQ files containing the unmapped reads (i.e., all the reads
discarded during the alignment on the human reference genome) if any. A schematic
view of the pipeline is shown in Fig. 1.
All the files that have to be submitted to DecontaMiner can be collected in the
same directory, and its path given as input. The entire pipeline can be subdivided
into three main phases.
The first phase involves the filtering and file format conversion steps, needed
to remove low quality reads and to obtain reads in fasta-format files, ready to be
aligned to the genome databases. More in detail, DecontaMiner wraps in its pipeline
two of the most used toolkits, Samtools [13] and Bedtools [14] used for the format
conversions, and FastX [15] for the quality filtering. The filtering is mainly based on
two parameters set by the user, namely the Phred quality threshold and the minimum
percentage of bases within that threshold.
DecontaMiner works both on paired- and single-end experiments, a parameter
that must be specified by the user. The conversion steps allow to sort the reads and
switch from bam to fastq and then to fasta formats.
Once terminated the conversion phase, the mapping module can start. In the case
of RNA-seq data, it is crucial to remove the ribosomal RNA (rRNA). Indeed, rRNA
represents up to 90 % of the total RNA. Although the wet lab procedures provide
an rRNA removal step, often this procedure is not totally satisfactory, due to high
DecontaMiner: A Pipeline for the Detection of Contaminating Sequences 141

Fig. 1 The pipeline

A scheme of the DecontaMiner pipeline. On the right, in blue are the input files, and in red the
tools used to process the data. In the central part, as a flux, the processing steps are described. On
the left, the parameters that can be set for each step are indicated in green. Several tab-delimited
files and one matrix are the pipeline outputs. All the discarded reads are also provided, as well as
all the different file formats generated (fastQ, FASTA, etc.). The matrix, containing all the samples,
can be easily used to create a bar plot

number of rRNA copies. We downloaded the fasta sequences of human ribosomal

RNA (28S, 18S, 5S, 5.8S and mitochondrial 12S, 16S) from NCBI website. The
rRNA alignment is performed using the SortmeRNA tool [16], which is a software
designed to this aim. All the reads that do not map to the human rRNA will undergo
mapping to bacteria, viruses, and fungi genome databases (NCBI nt) using the
MegaBLAST [17] algorithm.
The rRNA alignment reliability is evaluated using the E-value score. This
threshold can be either set by the user or left at the default SortMeRna value. The
user can specify also the alignment length and number of allowed mismatches/gaps
when aligning to contaminating genomes.
142 I. Granata et al.

The BLAST outputs, in table format, are then submitted to the third and last
phase, that involves the collection and extraction of information from the local
alignments.
This module, mainly composed of Perl scripts, is executed accordingly to
some user-specified parameters specifying the filtering and collecting options. In
particular, the filtering is based on the threshold number of total reads successfully
mapped and on the minimum threshold of reads mapped to a single organism.
Instead, the collecting options involve the choice of organizing the results according
either to genus or to species names.
DecontaMiner stores the output reads into three main files: unaligned, ambigu-
ous, and aligned. The “unaligned” file contains the reads that do not satisfy
the filtering parameters (i.e., length of alignment, number of allowed gaps, and
mismatches). The ambiguous reads are those that map to different Genera or, in case
of paired-end reads, those having mates mapping to different genera. Ambiguous
reads mapping to more than one Genus might derive from ortholog sequences. Since
Reads matching all the filtering criteria are stored into the “aligned” file.
The results are available in a tabular format, one for each sample, containing
the names of the detected organisms and the relative reads count. Furthermore,
DecontaMiner generates a matrix that can be easily used to create a barplot or other
types of diagrams in which all the data are collected together.
Lastly, the summary statistics about the number of matched/filtered/discarded
reads and organisms are generated and stored into tabular textual files.

3 Case Studies

3.1 Cancer Datasets

In order to assess the usefulness of the DecontaMiner pipeline and its efficiency
in detecting non-human sequences in NGS data, we used two publicly available
datasets downloaded from the GEO portal (GSE68086 and GSE69240).
The first study, from which the dataset GSE68086 was generated, concerns
the total RNA-sequencing experiments of blood platelet samples from patients
with six different malignant tumors (non-small cell lung cancer, colorectal cancer,
pancreatic cancer, glioblastoma, breast cancer, and hepato-biliary carcinomas) and
from healthy donors [18]. The experiment was performed with single-end 100 bp
reads.
The second one, GSE69240, derives from the expression profiling by high-
throughput sequencing of High-Grade Ductal Carcinoma In Situ (DCIS) [19]. The
dataset contains 25 pure HG-DCIS and 10 normal breast organoids samples. The
reads are paired-end 76 nucleotides long. This second dataset was used for testing
our pipeline on polyA RNA-seq data.
DecontaMiner: A Pipeline for the Detection of Contaminating Sequences 143

Table 1 Decontaminer parameter settings

Parameter name Value
Phred quality threshold 20
Minimum % of bases with the Phred set quality 100
E-value rRNA alignment 10–20
Match length D Query length
Mismatch number 1
Gap number 0
Minimum threshold of reads mapped to a single organism 100

3.2 Pre-processing

The Sequence Read Archive (SRA) file of each sample was downloaded and
converted to fastq format using the SRAToolkit [20]. The sequencing reads were
cleaned by eventual poor quality ends by Trimmomatic [21]. The quality assessment
of the trimmed reads was performed with FastQC [22]. The fast splice junction
mapper TopHat [23] was chosen to align the fastq files to the reference genome
(assembly hg19) guided by UCSC gene annotation. The sequence features in
mapped data were checked by SamStat [24]. The unmapped bam files provided
by TopHat were the input to our pipeline.
The parameter setting used for analyzing the two datasets is listed in Table 1.

3.3 Results

The analysis of the overall read mapping rate showed a high variability among the
samples of the GSE68086 dataset, with a range of 5–40 % of unmapped reads.
In the case of the GSE69240 dataset, instead, we observed a good mapping rate
in all the samples, with a percentage of unmapped reads below 10 %. The mapping
statistics of the two datasets immediately suggested a different probability to detect
non-human sequences.
In order to test the reliability of our pipeline we submitted to the analysis also the
samples with a small amount of unmapped sequences.
As we expected, we did not find any significant match to contaminating genomes
for the samples of the GSE69240 dataset. We also re-analyzed the data, lowering the
stringency of the parameters in terms of allowed mismatches and gaps (2 for each),
with the same negative outcome.
This result completely agrees with the type of experimental procedure used. As
mentioned before, an efficient polyA RNA-seq process and a set of samples not
contaminated by the environment should guarantee reads free of contamination.
Hence, this result supports the reliability of the pipeline in terms of false positives
detection.
144 I. Granata et al.

Table 2 Number of reads in the Decontaminer pipeline for two tumor

sample
Number of obtained reads (% of raw reads)
Pipeline step Sample A Sample B
Human unmapped (input) 4,698,672 (31.0 %) 4,961,067 (36.6 %)
Quality filtering 1,355,915 (22.5 %) 2,020,118 (14.9 %)
Ribosomal alignment 1,043,952 (22.2 %) 1,795,032 (13.3 %)
BLAST alignment 1,670,204 (11.0 %) 4478 (0.03 %)
Bacteria alignment filtering 1,434,098 (9.5 %) 49 (0.0004 %)

Instead, in the GSE68086 dataset DecontaMiner detected several matches to

bacterial reference sequences. In particular, we focused on those samples having
more than 10 % of human-unaligned reads. A modest amount of reads matched
to fungal genomes, whereas many reads aligned to bacteriophages specific for the
identified bacteria (namely Enterobacteria phage and Propionibacterium phage).
This last finding further confirmed the accuracy of the bacteria identification. As
an example, the number of reads in two samples before and after the filtering and
rRNA alignment processes are shown in Table 2. Sample A and Sample B had low
mapping rates on the reference genome 69 and 63.4 %, respectively. However, the
reason for such a high number of unmapped reads is completely different. Most of
the alignment failure of the Sample B is due to the presence of low quality reads,
that are approximately 22 % of the total raw reads, and only 0.0004 % reads matched
correctly to bacteria, according to our setting. Instead, only 7.5 % of the sample A
are low quality reads and almost 10 % significantly matched to bacteria. As shown
by the barplots generated by a Matlab in-house script, Figs. 2 and 3, both healthy
and tumor samples contain non-human sequences. For each sample, we plotted only
the organisms having number of matched reads greater than 20 % of the total. All
the species that do not fit this criterion are reported as “Others.”
Propionibacterium acnes and Escherichia coli species were detected in almost
all tumor samples and healthy donors, suggesting the possibility of a downstream
contamination of the samples or some kind of machine artifacts. P. acnes is a gram-
positive bacterium that forms part of the normal flora of the skin [25] and it is
usually considered a contaminant of blood cultures [26]. E. coli is a gram-negative
bacterium, host of the normal intestinal flora, but also one of the most common
responsible of a wide variety of hospital and community-onset infections, affecting
patients with normal immune systems as well as those immunodepressed [27].
One of the healthy samples, as well as one of the hepato-biliary carcinoma
group, did not have a significant number of reads matching to any bacterial species,
according to our thresholds.
DecontaMiner: A Pipeline for the Detection of Contaminating Sequences 145

Fig. 2 Healthy controls barplot. For each healthy sample a bar reports the detected contaminating
organism (colors) and percentage of unmapped reads assigned to each of them

Fig. 3 Tumor samples barplot

146 I. Granata et al.

The tumor samples barplot shows the presence of some bacterial species that
are absent in control samples, or present with a very low reads number. Among
them is worth to note the bacterium Acinetobacter baumannii. The percentage of
reads aligned to A. baumannii is particularly evident in hepato-biliary carcinoma,
although its presence seems to be independent of cancer type.
The genus Acinetobacter, as currently defined, comprises gram-negative, strictly
aerobic, nonfermenting, nonfastidious, nonmotile, catalase-positive, and oxidase-
negative bacteria [28]. A. baumannii normally inhabits human skin, mucous
membranes, and soil [29]. Acinetobacter baumannii, in particular, has become
one of the major causes of nosocomial infections during the past two decades
[28, 30–32] and its correlation with outcomes of cancer patients is a clinical issue
under study [33, 34].

4 Conclusions

The DecontaMiner pipeline was designed and developed to investigate the presence
of contaminating sequences in NGS data. It has a dual utility, both as a filtering tool
to remove foreign reads from the raw sequencing file, usually in fastq format, and
as a detection tool to identify contaminating sequences among the unmapped reads,
provided as a bam file. In order to test our pipeline we used two different RNA-
seq datasets. The lack of matches to microorganisms in case of the polyA-RNA
(GSE69240) demonstrates that the risk of incurring into false positive results is very
low. The reliability of our pipeline is further proved on the total RNA (GSE68086)
dataset analysis. Indeed, we found some kind of background contamination in
almost all the samples. The most present organisms are P. acnes and E. coli and,
in addition, some tumor samples significatively matched to A. baumannii, that it
is a well-known nosocomial pathogen, even probably associated with outcomes of
cancer diseases. It is important to underline that DecontaMiner can suggest the
presence of contaminating sequences, but this results must be confirmed by an
experimental validation. As an added value, the output fasta files and BLAST tables
can be easily uploaded to MEGAN5 [35], a metagenome analyzer, which allows
to obtain more detailed information about the taxonomy profile of the samples in
several graphical modes. We are currently working to provide DecontaMiner as a
Bash shell command-line tool, usable on a common laptop as well as in a distributed
computing environment. We are also planning to put together the pipeline here
developed and the Transcriptator tool [36] developed in our lab to provide an
integrated environment for the analysis of omics data.

Acknowledgements Ilaria Granata was supported by MIUR PON02-00619. Mara Sangiovanni

was supported by Interomics Italian Flagship Project. Mario R. Guarracino work has been
conducted at National Research University Higher School of Economics (HSE) and has been
supported by the RSF grant n. 14-41-00039.
DecontaMiner: A Pipeline for the Detection of Contaminating Sequences 147

References

1. Conesa, A., et al.: A survey of best practices for RNA-seq data analysis. Genome Biol. 17(1),
1–19 (2016)
2. Laurence, M., Hatzis, C., Brash, D.E.: Common contaminants in next-generation sequencing
that hinder discovery of low-abundance microbes. PLoS One 9(5), e97876 (2014)
3. Feng, H., et al.: Clonal integration of a polyomavirus in human Merkel cell carcinoma. Science
319(5866), 1096–1100 (2008)
4. Palacios, G., et al.: A new arenavirus in a cluster of fatal transplant-associated diseases. N.
Engl. J. Med. 358(10), 991–998 (2008)
5. Bhatt, A.S., et al.: Sequence-based discovery of Bradyrhizobium enterica in cord colitis
syndrome. N. Engl. J. Med. 369(6), 517–528 (2013)
6. Strong, M.J., et al.: Microbial contamination in next generation sequencing: implications for
sequence-based analysis of clinical samples. PLoS Pathog. 10(11), e1004437 (2014)
7. Tae, H., et al.: Large scale comparison of non-human sequences in human sequencing data.
Genomics 104(6), 453–458 (2014)
8. Ouma, W.Z., et al.: Important biological information uncovered in previously unaligned reads
from chromatin immunoprecipitation experiments (ChIP-Seq). Sci. Rep. 5, 8635–8635 (2015)
9. Kostic, A.D., et al.: PathSeq: software to identify or discover microbes by deep sequencing of
human tissue. Nat. Biotechnol. 29(5), 393–396 (2011)
10. Borozan, I., et al.: CaPSID: a bioinformatics platform for computational pathogen sequence
identification in human genomes and transcriptomes. BMC Bioinf. 13(1), 206 (2012)
11. Altschul, S.F., et al.: Basic local alignment search tool. J. Mol. Biol. 215(3), 403–410 (1990)
12. Naccache, S.N., et al.: A cloud-compatible bioinformatics pipeline for ultrarapid pathogen
identification from next-generation sequencing of clinical samples. Genome Res. 24(7),
1180–1192 (2014)
13. Li, H., et al.: The sequence alignment/map format and SAMtools. Bioinformatics 25(16),
2078–2079 (2009)
14. Quinlan, A.R., Hall, I.M.: BEDTools: a flexible suite of utilities for comparing genomic
features. Bioinformatics 26(6), 841–842 (2010)
15. Gordon, A., Hannon, G.J.: FastX Toolkit (2010) http://hannonlab.cshl.edu/fastx_toolkit/index
16. Kopylova, E., Noé, L., Touzet, H.: SortMeRNA: fast and accurate filtering of ribosomal RNAs
in metatranscriptomic data. Bioinformatics 28(24), 3211–3217 (2012)
17. Zhang, Z., Schwartz, S., Wagner, L., Miller, W.: A greedy algorithm for aligning DNA
sequences. J. Comput. Biol. 7, 203–214 (2000)
18. Best, M.G., et al.: RNA-Seq of tumor-educated platelets enables blood-based pan-cancer,
multiclass, and molecular pathway cancer diagnostics. Cancer Cell 28(5), 666–676 (2015)
19. Abba, M.C., et al.: A molecular portrait of high-grade ductal carcinoma in situ. Cancer Res.
75(18), 3980–3990 (2015)
20. Leinonen, R., Sugawara, H., Shumway, M.: The sequence read archive. Nucleic Acids Res. 39,
D19–D21 (2010).
21. Bolger, A.M., Lohse, M., Usadel, B.: Trimmomatic: a flexible trimmer for Illumina sequence
data. Bioinformatics 30, 2114–2120 (2014).
22. Andrews, S.: FastQC: a quality control tool for high throughput sequence data. Reference
Source (2010)
23. Trapnell, C., Pachter, L., Salzberg, S.L.: TopHat: discovering splice junctions with RNA-Seq.
Bioinformatics 25(9), 1105–1111 (2009)
24. Lassmann, T., Hayashizaki, Y., Daub, C.O.: SAMStat: monitoring biases in next generation
sequencing data. Bioinformatics 27(1), 130–131 (2011)
25. Perry, A., Lambert, P.: Propionibacterium acnes: infection beyond the skin. Expert Rev. Anti-
Infect. Ther. 9(12), 1149–1156 (2011)
26. Park, H.J., et al.: Clinical significance of Propionibacterium acnes recovered from blood
cultures: analysis of 524 episodes. J. Clin. Microbiol. 49(4), 1598–1601 (2011)
148 I. Granata et al.

27. Pitout, J.D.D.: Extraintestinal pathogenic Escherichia coli: an update on antimicrobial resis-
tance, laboratory diagnosis and treatment. Expert Rev. Anti-Infect. Ther. 10(10), 1165–1176
(2012)
28. Peleg, A.Y., Seifert, H., Paterson, D.L.: Acinetobacter baumannii: emergence of a successful
pathogen. Clin. Microbiol. Rev. 21(3), 538–582 (2008)
29. Manchanda, V., Sanchaita, S., Singh, N.P.: Multidrug resistant acinetobacter. J. Global Infect.
Dis. 2(3), 291 (2010)
30. Fukuta, Y., et al.: Risk factors for acquisition of multidrug-resistant Acinetobacter baumannii
among cancer patients. Am. J. Infect. Control 41(12), 1249–1252 (2013)
31. Al-Hassan, L., El Mehallawy, H., Amyes, S.G.B.: Diversity in Acinetobacter baumannii
isolates from paediatric cancer patients in Egypt. Clin. Microbiol. Infect. 19(11), 1082–1088
(2013)
32. Dijkshoorn, L., Nemec, A., Seifert, H.: An increasing threat in hospitals: multidrug-resistant
Acinetobacter baumannii. Nat. Rev. Microbiol. 5(12), 939–951 (2007)
33. Ñamendys-Silva, S.A., et al.: Outcomes of critically ill cancer patients with Acinetobacter
baumannii infection. World J. Crit. Care Med. 4(3), 258 (2015)
34. Nazer, L.H., et al.: Characteristics and Outcomes of Acinetobacter baumannii Infections in
Critically Ill Patients with cancer: a matched case-control study. Microb. Drug Resist. 21(5),
556–561 (2015)
35. Huson, D.H., Weber, N.: Microbial community analysis using MEGAN. Methods Enzymol.
531, 465–485 (2012)
36. Tripathi, K.P., Evangelista, D., Zuccaro, A., Guarracino, M.R.: Transcriptator: an automated
computational pipeline to annotate assembled reads and identify non coding RNA. PLoS One
10(11), e0140268 (2015)