Predicting Phenotype from Genotype

through Automatically Composed Petri Nets

Mary Ann Bltke1 , Monika Heiner2 , and Wolfgang Marwan1
1

Magdeburg Centre for Systems Biology and Lehrstuhl fr Regulationsbiologie,

Otto-von-Guericke-Universitt, Magdeburg, Germany
2
Chair of Data Structures and Software Dependability,
Brandenburg Technical University, Cottbus, Germany
[email protected]

Abstract. We describe a modular modelling approach permitting curation, updating, and distributed development of modules through joined
community eort overcoming the problem of keeping a combinatorially
exploding number of monolithic models up to date. For this purpose, the
eects of genes and their mutated alleles on downstream components are
modeled by composable, metadata-containing Petri net models organized
in a database with version control, accessible through a web interface
(www.biomodelkit.org). Gene modules can be coupled to protein modules
through mRNA modules by specific interfaces designed for the automatic,
database-assisted composition. Automatically assembled executable models may then consider cell type-specific gene expression patterns and the
resulting protein concentrations. Gene modules and allelic interference
modules may represent eects of gene mutation and predict their
pleiotropic consequences or uncover complex genotype/phenotype relationships. Forward and reverse engineered modules are fully compatible.
Keywords: Biomodel engineering, formal language, data integration,
high-throughput, quantitative trait loci.

Introduction

Systems biology witnesses the evolution of experimental high-throughput methods with steadily increasing power regarding the quantification of nucleic
acids, proteins, covalent modifications, and metabolites. Soon, these methods
will broadly allow omics scale analyses of molecules involved in cellular regulatory control that capture the time-resolved response to stimulation or (genetic)
network perturbation [22]. These advances challenge the development of integrative and modular modelling frameworks that support the combination of
findings obtained through dierent, qualitative and quantitative experimental
approaches. To be useful, such models will need to be multi-level in terms of
integrating multiple levels of abstraction in causally connected and experimentally well established processes at the molecular level with adjustable resolution
of details for higher level phenomena. These may include cell fate decisions for
D. Gilbert and M. Heiner (Eds.): CMSB 2012, LNCS 7605, pp. 87106, 2012.
c Springer-Verlag Berlin Heidelberg 2012

88

M.A. Bltke, M. Heiner, and W. Marwan

simulating the intrinsic heterogeneity of clonal populations of cells following individual trajectories during development. Petri nets are an ideal formalism for
the formal description of processes at multiple levels of abstraction for systems
biology purposes [13,17,18,28].
For the sake of creating realistic scenarios, it will presumably become indispensable to compare, on a regular basis, simulation results on one and the
same network topology as obtained by employing continuous, stochastic, and
hybrid paradigms. In the software tool Snoopy, a given Petri net graph can be
interpreted and simulated as continuous (ODE), stochastic, hybrid, or simply as
qualitative model with export option to SBML [29]. Interpretation as coloured
Petri net, again in Snoopy, provides advanced options for biomodel engineering [19] as coloured Petri nets combine the formalism of Petri nets with the
expressive power of a programming language. For this reason, and because of
the intuitively accessible graphical representation, we have chosen Petri nets as
framework for modelling and simulation.
Repeated iterations of experimental data acquisition, modelling, and simulation can evaluate the consistency of the interpretation of experimental results.
This is especially true when high-throughput data come into play. However, conventional monolithic models usually represent certain aspects of a phenomenon
at a certain resolution in detail and are restricted to a certain mathematical
modelling paradigm. Such models can neither be easily combined with other
models nor can they be easily updated by persons other than the author of the
particular model. One solution to this limitation is to create a collection of Petri
net modules that can be automatically linked in order to obtain and to update
coherent models covering all or selected aspects of a biological process. Conceptually, such modules may be contributed, curated, and updated by individuals of
the community with special expertise in certain aspects. Being organized within
a database with version control, modules obtained by reverse engineering of
experimental data can also be integrated and, as we will show below, help to
import complex data sets into models that have been automatically generated
by composition of pre-existing modules.
This paper makes three major contributions which fundamentally enhance
the versatility of modular Petri net modelling by (1) linking regulated gene
expression to protein concentration, (2) allowing the fully automated generation
of models for the application in genome-wide (omics) approaches, and (3) linking
gene mutation to complex phenotypic consequences in generating predictable
models. Let us briefly elaborate on these claims.
1. The introduction of gene modules and mRNA modules allows to model regulated gene expression and protein biosynthesis. As the gene expression
pattern of a cell is not constant and can drastically change dependent on
cell type, physiological state, or experimental conditions, cells are definitely
equipped with specific sets of proteins of variable relative abundance. By
introducing gene modules into the model, dierentially regulated gene activity and the resulting gene expression patterns translate into the marking of
places of the protein modules. As the rates of biochemical reactions always

Automatically Composed Petri Nets

89

depend on both, the kinetic rate constants and the concentrations of the
reactants, changed gene expression will also change the rates of biochemical reactions, which in turn may drastically alter the dynamic behaviour of
a regulatory network. Moreover, changed concentrations in regulatory proteins (e.g. transcription factors) may feed back in a complex manner onto the
gene level by changing gene expression profiles. This circuitry of interwoven
regulatory control becomes systematically accessible through the model.
2. Gene and mRNA module prototypes permit the fully automated generation
of modules by simply uploading lists of gene names. This allows the automatic creation of models representing hundreds or even thousands of genes,
their mRNAs and the proteins they form. By importing transcriptomic or
proteomic data sets obtained in high throughput experiments [30], one can
infer rate constants and reverse engineer regulatory mechanisms with the
help of the model and predict changes in the proteome in response to dierential gene regulation. Such models will also support the interpretation of
phenomena observed in systematic RNAi screens where individual genes are
knocked down [25,10,9].
3. Introduction of allelic influence modules extends the modelling of gene activity to the modelling of the regulatory consequences, which gene mutations
have on cellular processes. This sets the formal framework to reverse engineer biomodels from complex phenotype data sets resulting from genotypic
variation e.g. by employing Petri net compatible algorithms [16,12,11]. It
is obvious that such type of models have a high potential for the application to various areas from basic research to synthetic biology or personalized
medicine.
We are not aware of any modelling framework providing a comparable versatility and integrative power in terms of combining forward and reverse biomodel
engineering.
This paper is organised as follows. In the next section we briefly summarise
relevant own previous work on modular Petri net modelling and the automatic
composition of modules with the help of a module database. Then we will provide
the rules for the generation of modules and modular models and introduce gene
and mRNA modules in Section 3, the application of which is demonstrated by the
first case study in Section 4. Section 5 explains the specific features of gene and
mRNA modules required for modelling gene expression and its dierential regulation in eukaryotes. We continue in Section 6 with a second case study on cell
dierentiation and eukaryotic gene regulation, define allelic influence modules,
and explain how these work together with gene, mRNA, and protein modules
in generating models that integrate forward and reverse biomodel engineering
approaches. In Section 7 we conclude with discussing the versatility of our approach and provide future perspectives regarding the application to synthetic
biology and omics approaches.

90

M.A. Bltke, M. Heiner, and W. Marwan

Previous Work

Initially, we developed our modular modelling approach to represent biochemical reaction networks made of protein-protein interactions. The core idea is to
take an object-oriented approach where the objects correspond to the natural
modular building blocks of life. We represent individual proteins as independent
and self-contained hierarchically structured Petri nets, called modules. Thus,
modules correspond to natural units, each of which comprises intramolecular
regulatory mechanisms of the respective protein and of all its interactions with
other molecules. Modules of interacting proteins can be coupled by logical nodes
of identical shared subnets describing their interaction with each other. The assembly of models from a set of modules needs no further modifications at the
module level. An essential advantage of this approach is the reusability of all constructed modules in arbitrary combination to obtain models representing specific
pathways [4,5].
A crucial point of our modular modelling approach is that each module is selfcontained and can be evaluated on its own. Composed networks which in terms of
their behaviour correspond to the conventional, monolithic networks need never
to be explicitly shown. Their validity is entirely assessable by understanding the
individual modules and the connection rules.
Our modular modelling approach has been deployed to construct modular
Petri nets for two non-trivial case studies: (1) the JAK/STAT pathway in IL-6
signalling [4] and (2) nociceptive network in pain signalling [5]. The JAK/STAT
pathway is one of the major signalling pathways in multicellular organisms controlling cell development, growth and homeostasis by regulating the gene expression. The modular network of the JAK/STAT pathway in IL-6 signalling
comprises 7 protein modules (IL6, IL6-R, gp130, JAK1, STAT3, SOCS3, and
SHP2). Overall, the model consists of 92 places, 102 transitions spread over
58 pages with a nesting depth of 4. The nociceptive network in pain signalling
consists of several crucial signalling pathways, which are hitherto not completely
revealed and understood. The latest version of the nociceptive network consists of
38 modules, among them are several membrane receptors, kinases, phosphatases
and ion-channels. So far, the model is made up by 713 places and 775 transitions
spread over 325 pages, again with a nesting depth of 4.
To support convenient module selection and network composition, we developed a database prototype, which is accessible by a web-interface [4,6], see
Figure 1. The database holds the qualitative Petri net structure of each module,
as well as the kinetic information assigned to each transition. In addition, the
database contains also meta-information about the corresponding proteins (extracted from UniProt) and information about the literature used to construct
the modules (extracted from PubMed), which can be associated with each module. The organization of the modules in such a database enables the user to
(1) search for individual modules, places, transitions, etc., (2) store modules in
collections, and (3) assemble a modular network from a chosen collection.

Automatically Composed Petri Nets

91

Fig. 1. Selected screenshots of the prototype database [6]. (a) The web-interface
enables the user to browse and/or search for modules of specific proteins. Modules can
be stored in collections by selecting them. (b) Detailed information about each module
can be shown on a separate page.

3
3.1

Petri Net Modules

How Modules Are Built and Composed

We use Petri nets structured in the form of modules that allow the automatic
composition of executable models [4,5] based on Snoopy [29]. The modules and
their associated metadata are organized in a database accessible through a web
interface designed to manage multiple versions of each module and supporting
the automatic composition of models from modules (interactively) selected from
a potentially rich collection [4]. Initially, the approach was designed to model
protein-protein interactions (see Section 2) in the context of signal transduction
networks [4,5]. In this paper we go one step further by defining gene modules,
mRNA modules and allelic influence modules which considerably enhance power
and versatility of our modelling approach.
A module in general is centred around one entity describing all its interactions
with other components to which the module can be linked. The entity can be

92

M.A. Bltke, M. Heiner, and W. Marwan

a protein, an mRNA, a gene, or the specific allele of a gene. Conceptually, it is

possible to extend this definition to admit other entities as well. Before going into
detail, let us first briefly explain how the modular modelling approach technically
works and how modules are connected to each other.
The basic principle is explained taking the reaction of an autophosphorylating
kinase with its substrate as example (Figure 2). The kinase X autophosphorylates to give XP. XP transfers its phosphate group to the substrate Y. The
phosphate group of YP then hydrolyses spontaneously. The Petri net describing
these reactions (Figure 2a) can be decomposed into two modules each representing the reactions of one of the two proteins involved, X and Y, respectively
(Figure 2b, 2c). Places and transitions that occur in more than one module are
defined as so-called logical nodes. Logical nodes appear as multiple graphical
copies of a given place or transition. In this paper, logical nodes are shaded in
grey. Braking a Petri net up into modules introduces redundancy in terms of
the graphical display of nodes which might appear unnecessarily complicated
at first. For complex modules or when many modules are involved, the benefits
are indeed tremendous as we have previously shown ([4,5], and see Discussion).
The biosynthesis and the degradation of a protein or a nucleic acid are modelled
by separate biosynthesis or degradation modules, respectively. Accordingly, the
user can choose for each protein or mRNA whether or not its biosynthesis and
degradation should be considered in the assembled model.

Fig. 2. The principle of modular modelling based on the use of logical nodes
(connectors). (a) A Petri net model of the phosphorylation of protein Y by the
autophosphorylating kinase X is split into modules for (b) protein X and (c) protein Y,
respectively. Places and transitions that are shared by the two modules are implemented
as logical nodes shaded in grey.

3.2

Definition of Modules Representing the Function of Genes

We now extend the initial approach constrained to protein modules by defining

gene modules. A gene module considers the mechanisms of a gene being regulated
through the reversible transition between its on and o state. Assigned metadata
information includes the Genbank database accession number providing the DNA
sequence information as a cross-reference. Logical nodes of a gene module are
used to link the gene to other modules.
In general, multiple forms of each gene, so-called alleles, do exist that dier in
one or more base pairs. These dierences are due to mutations. Mutations can
be silent in not altering the amino acid sequence of the encoded protein. Mutations can also be neutral in not changing the properties of the encoded protein

Automatically Composed Petri Nets

93

although its amino acid sequence is changed due to mutation. These sequence
polymorphisms are commonly found in wild-type populations. Alternatively, a
mutation can change the properties of the encoded protein due to its altered
amino acid sequence or it may even prevent the formation of the protein at
all, e.g. by introducing a stop codon. Following strictly the modularity principle
in designing Petri nets, a separate gene module is created for each allele of a
gene to represent mutations that change relevant properties of the gene products (RNAs and proteins) as compared to the wild-type. Entering a query for a
gene, the module database will list all modules corresponding to alleles of that
gene.
We will now use two case studies to demonstrate how gene modules work.
The first case study (see Section 4) concerns metabolic regulation in bacteria.
It is taken to explain the principle of gene modules with the help of a simple example. In Section 5, we explain the specific features of gene and mRNA
modules required for modelling gene expression and its dierential regulation in
eukaryotes, which is the topic of our second case study (see Section 6). Here, we
show a more complex scenario involving multiple layers of regulatory control.
In addition, the second case study will provide an example of how modules can
be obtained through reverse engineering of experimental data, it will introduce
allelic influence modules and it will reveal the scalability of the approach.

Case Study: The Phosphate Regulatory Network

In the first case study, we consider the response of enteric bacteria like Escherichia coli to the limitation in inorganic phosphate which is required for the
biosynthesis of nucleic acids and other cellular components (compare Figure 2).
When inorganic phosphate (Pi ) becomes low in the environment, it may turn
into a growth-limiting factor even if sucient nutrients are available [26]. When
present, inorganic phosphate is taken up from outside of the cell through an ABC
transporter system, the PstSCAB transmembrane complex (Figure 3 [21]). With
sucient Pi outside, the PstSCAB complex actively pumps Pi across the cell
membrane into the cytoplasm. Under this condition, the PhoU protein forms,
according to the proposed mechanistic model [21], a complex with the pstSCAB
transporter system and the PhoR histidine kinase. Complex formation prevents
the kinase to autophosphorylate caused by the interaction with PhoU. PhoU is a
chaperone-like PhoR/PhoB inhibitory protein. When external Pi is low and the
PstSCAB complex is inactive, PhoU dissociates and allows the autophosphorylation of PhoR. PhoR then phosphorylates and thereby activates the transcription
factor PhoB. The phosphorylated form of PhoB, namely PhoBP, then activates
the transcription of at least 31 genes organised into 9 transcriptional units (eda,
phnCDEFGHIJKLMNOP, phoA, phoBR, phoE, phoH, psiE, pstSCAB-phoU, and
ugpBAECQ ) [21]. One of the activated genes, phoA, encodes the PhoA protein
which is a bacterial alkaline phosphatase. PhoA is exported across the membrane into the periplasm where it degrades organic phosphorous compounds
to liberate Pi which is then taken up into the cell to overcome the limitation.

94

M.A. Bltke, M. Heiner, and W. Marwan

Inhibition
[Pi] > 4 M

Activation
[Pi] < 4 M

Deactivation
[Pi] > 4 M & 3KR83VW%"
Pi

Pi

Periplasm

PstS

PstS

PstS

PstA PstC

PstA PstC

PstA PstC

PstB PstB

PstB Pst
PstB
B

PstB Pst
PstB
B
A
PhoR
PhoR
RA

A
PhoU Ph
PhoR
PhoR
RA

D
PhoU Ph
PhoR
PhoR
RD

PhoU

Pi
PhoB

PhoB

Pi

PhoB

Cytoplasm

Gene activation or repression

Fig. 3. Biochemical model for sensing extracellular inorganic phosphate (Pi )

and transduction of the signal to control gene expression. The PstSCAB transmembrane complex serves as an ABC transporter for the uptake of environmental Pi .
At high extracellular Pi concentration, the binding protein PstS is fully saturated and
this signal is relayed to the cytoplasmic part of the receptor that forms an inhibitory
complex with a second transmembrane protein, the PhoR kinase via the cytoplasmic
protein PhoU. If Pi is low, the complex dissociates and the autophosphorylating kinase
PhoR phosphorylates PhoB which, in its phosphorylated form, binds DNA to induce
gene expression. When i subsequently increases, the compex with PhoU is formed again
and PhoBP is dephosphorylated. The figure was redrawn from Hsieh and Wanner published in Current Opinion in Microbiology [21].

When enough Pi has been formed, this system is switched o again and PhoBP
is dephosphorylated.
In [24], we gave a monolithic Petri net of a simplified version of the phosphate regulatory circuitry. To cover the entire functionality we now construct a
modular Petri net model which is composed of three types of modules: (1) one
protein module representing the reactions of the PhoB protein, (2) gene modules representing the regulated genes, and (3) mRNA modules representing the
transcription of the gene, the translation of mRNA into the protein, and the
degradation of mRNA. The degradation of the encoded proteins is represented
by degradation modules which have been introduced previously [4].
4.1

The PhoB Module

The PhoB module (Figure 4) represents the reactions of the PhoB protein in its
phosphorylated (PhoBP) and dephosphorylated (PhoB) states. It also represents
the complex formation of PhoB with its regulatory proteins as well as the binding
of PhoB to regulatory sequences in the DNA.

Automatically Composed Petri Nets

95

Fig. 4. Petri net representation of the phosphate regulatory network. (a)

Top-leve presentation of the modules of the phosphate regulatory system in the form
of coarse places as they appear in Snoopy. (b) Protein module of the PhoB protein
displaying the direct interactions with binding partners.(BS, binding site).

The PhoB module models the regulatory mechanism schematically shown in

Figure 3. Binding and dissociation of PhoBP in its complex with PhoRP is represented separately for each transcriptional unit. Displaying the binding of PhoBP
to each regulatory site on the DNA separately, keeps the Petri net graph clearly
arranged and allows to reuse the structural motif of binding and unbinding reactions via copy/paste for the various regulated transcriptional units. Accordingly,
PhoRP_PhoBP is declared as logical place.
To save space in Figure 4, we only show binding of PhoBP to one of the nine
transcriptional units.
4.2

The Gene Modules

A gene module represents the regulation of the gene by other factors (e.g. transcription factors) through the transition between its on and o state (Figure 5).
In the quantitative interpretation of the Petri net (as stochastic, continuous, or
hybrid Petri net) the regulatory factors influence the equilibrium between the
on and the o state of the gene. The on state means that the gene is transcriptionally active and that mRNA molecules can be accordingly formed as modeled
in the mRNA module.
In prokaryotes, several functionally related genes can be organized into a
single regulatory and transcriptional unit, a so-called operon. The genes of an
operon are transcribed in the form of a single mRNA molecule, called a polycistronic message, which may encode several proteins at once. In the case of
the phnCDEFGHIJKLMNOP transcriptional unit (Figure 6), one polycistronic
message encoding 14 dierent proteins (PhoC to PhoP) is formed upon the

96

M.A. Bltke, M. Heiner, and W. Marwan

Fig. 5. Prototype of a gene module. The module may represent the activation of any
gene responsive to the phosphorylated PhoB protein. Binding of PhoRP_PhoBP to the
promotor renders the gene active. These regulatory interactions are modeled through a
test arc activating transition T2 and an inhibitory arc blocking T4. The basal activity
of the gene in the absence of PhoRP_PhoBP is maintained by T1 and T3.

Fig. 6. mRNA module modelling the formation of the polycistronic message

synthesized by transcription of the active phnC-P gene. The mRNA is translated into the proteins PhnC . . . PhnP. For simplicity, the places for only five of the 14
proteins that are formed are shown.

initiation of transcription. The probability per unit of time for the initiation
of transcription to occur depends on binding of PhoRP_PhoBP to the regulatory region of the phnCDEFGHIJKLMNOP operon on the DNA. From the
biological point of view, polycistronic messages provide a simple mechanism for
co-regulation of genes encoding proteins that work together in a cellular process. In order to obtain a model including the transcriptional regulation of all
31 genes organized into the 9 transcriptional units that are under control of the
PhoB protein, 9 gene modules (phoA, phoBR, phoE etc.) of analogous structure
(as shown in Figure 5) are required. For large scale modelling approaches, these
network structures could be generated automatically or semi-automatically.
4.3

The mRNA Modules

The mRNA modules models the reactions of the respective mRNA species,
namely its biosynthesis by transcription, its degradation, and its translation
into the encoded proteins (Figure 6). An mRNA module may in addition represent the binding of regulatory proteins to the RNA, the binding of antisense

Automatically Composed Petri Nets

97

RNA influencing the stability of the message, or the processing (e.g. splicing)
of the transcript, as it occurs in eukaryotes (shown in Figure 8 and discussed in
Section 7.2). The transcription of the bacterial phnCDEFGHIJKLMNOP transcriptional unit leads to the formation of a polycistronic message, which encodes
for 14 proteins (PhoC . . . PhoP). The reactions (e.g. the catalytic activity) of
the encoded proteins might then be considered in separate protein modules.

5
5.1

Modelling Eukaryotic Gene Regulation with Gene and

mRNA Modules
Eukaryotic Gene Modules

The gene modules designed to model the regulation of eukaryotic genes are very
similar to the models of prokaryotic genes as presented in the previous section for
the phosphate regulatory network. However, the regulation of eukaryotic genes
is typically more complex than in prokaryotes as more protein factors and more
binding sites for regulatory proteins on the DNA may be involved. All these
factors together may control the on state of a gene.
Gene regulation may involve protein binding sites on the DNA functioning
as enhancers or silencers that are located several thousand base pairs distant
from the genes they regulate. Proteins bound to these sites may influence the
probability for transcription to be initiated through physical interactions with
the protein complexes bound to the promotor of the regulated gene. These regulatory sites and the binding of regulatory proteins to these sites are represented
as part of the gene module.
A prototype of a module representing the regulatory control of a eukaryotic
gene is shown in Figure 7. Making regulatory sites part of the gene module comes
with the advantage that potentially cooperative eects in protein binding and
gene activation can be considered as part of the module.
5.2

Eukaryotic mRNA Modules

In addition to the biosynthesis of proteins there are several RNA-dependent

processes that may be of regulatory importance especially in eukaryotic cells.
Typically, the occurrence of these mechanisms depends on the considered RNA
species and may also depend on physiological conditions as well as on developmental states. Each of the mechanisms described in the following has been
implemented in a basic form using the mRNA module prototype shown in
Figure 8.
Alternative Splicing. Primary transcripts in eukaryotes are processed during
the maturation of the final protein-encoding mRNA. Processing includes the
splicing of the RNA where non-coding introns are excised from the primary
transcripts. Due to the occurrence of alternative splicing sites, it may be that
dierently spliced mRNAs are formed from one and the same primary transcript

98

M.A. Bltke, M. Heiner, and W. Marwan

Fig. 7. Prototype of a eurkaryotic gene module. The regulation of the eukaryotic gene (a) depends on more protein factors than the regulation of the prokaryotic
gene shown in Figure 5 does. By binding to the promotor region of the gene, these
factors form a multimeric protein complex, as shown for transcription factors TF1
and TF2, both of which directly bind to the promotor (b). The third transcription
factor, TF3, binds first to an enhancer sequence distant from the promotor (c) and
subsequently can bind to the promotor-TF1-TF2 complex to form the gene activating
complex GeneX_AC (b) which switches the gene into its transcriptionally active state
(a). As in Figure 5, gene activation by protein binding to the promotor is modeled by
control arcs and the basal level of gene activity occurs through firing of transitions T1
and T3 (a). According to our module notion the gene module displays all direct molecular interactions of GeneX with the proteins that bind to its regulatory sequences. Note
that binding of regulatory proteins may involve cooperative mechanisms which would
be represented accordingly in the context of the gene module.

giving rise to proteins of partially dierent amino acid sequence. This splicing
depends on protein factors that may be regulated depending on the physiological
or developmental state of the cell. When necessary, the biochemical reactions
of these slicing factors (like posttranslational modification or protein-protein
interaction) may be represented in the context of protein modules with the help
of logical places.
RNA-Binding Proteins. The half-life of mRNA species may vary between
minutes and months. This variation may have dierent reasons in addition to
the specific secondary structure of the RNA. One mechanism influencing the
half-life of a given mRNA species is the binding to specific RNA-binding proteins that may store or degrade the RNA. Being bound to an RNA-binding
protein, like Pumilio for example, the mRNAs can be stored in the cell while
being translationally inactive. Upon release from the RNA-binding proteins the
mRNA may suddenly become translationally active and hence become available
at relatively high concentration [14].

Automatically Composed Petri Nets

99

Fig. 8. Prototype of an eukaryotic mRNA module. Transcription of the active

Gene (GeneX) leads to the formation of a primary transcript which is processed. In
the example shown, the primary transcript is spliced into two alternative mRNAs,
GeneX_mRNA_1 and GeneX_mRNA_2, respectively that are subsequently translated into the corresponding proteins. The mature mRNAs may bind to RNA binding
proteins that regulate the stability of the mRNA and its availability for the translational maschinery. For simplicity, this reaction is shown for one of the two mRNAs
only.

RNA Interference. RNA interference is a natural mechanism for the specific

inactivation of the expression of eukaryotic genes, e.g. by small interfering RNAs
that bind to the target RNA. The degradation of the RNAs depends accordingly
on specific protein factors and on the availability of the interfering RNA [8,33].
These mechanisms may have to be considered in the context of mRNA modules and receive regulatory input from specific cellular proteins.
5.3

Why Is the Integration of Bottom-Up and Top-Down Models

Essential, Especially in Eukaryotes?

In some respects systems biology appears to be more dicult for eukaryotic

as compared to prokaryotic cells. This may in part be due to the occurrence
of fundamentally dierent regulatory processes in the two domains of life with
non-obvious consequences of certain eukaryotic regulatory phenomena.
There is indeed highly detailed knowledge on the canonic pathways of eukaryotic signal transduction which allows the formulation of well-structured bottomup models representing the biochemical interactions of regulatory components
like e.g. the MAP kinase cascade. Such models can be tremendously useful in
understanding mechanisms in health and disease [23,27,31] and in finding new
and powerful drugs.
However, it is also true that many experimental findings on canonic pathway
components seem to be contradictory. This may in part be due to the fact that
gene expression patterns in dierent experimental systems and under dierent
physiological conditions are dierent leading to a dierent composition of bio-

100

M.A. Bltke, M. Heiner, and W. Marwan

chemical reactants within the cell at a given time point or between replicas of a
particular experiment.
This certainly restricts the current practical value of bottom-up models without disclaiming their general usefulness. When a certain cellular process, for
example the dierentiation phenomenon in a eukaryotic cell or the progression
through the cell cycle, is to be rigorously analysed at the transcriptomic or the
proteomic level, the changes in many perhaps in most of the observed components and their consequences can currently not be explained on the basis of the
established bottom-up models of the canonic pathways. This suggests that there
are tremendous gaps in our current understanding.
On the other hand, it seems for the time being impossible that the thousands
of components accessible through omics approaches can be analysed with such
experimental eort as invested for the exploration of canonic pathway components. Therefore it seems self-evident that omics data are used to rigorously
reverse-engineer models. Then, a next and essential step is to integrate these
top-down models with relevant bottom-up models to obtain integrated models
with predictive power.

6
6.1

Case Study: The Sporulation Control Network in

Physarum polycephalum
Sporulation Is Controlled by a Gene Regulatory Network

Physarum polycephalum is a unicellular eukaryote belonging to the amoebozoa

group of organisms [1,2,15]. During its relatively complex and branched life cycle,
Physarum develops into various cell types. These cell types occur in temporal
order and dier in morphology (shape), molecular composition, and physiological
function ([7] and references therein).
One of these cell types is the plasmodium, a multinucleate macroscopic single
cell. Dierentiation of the plasmodium can be easily studied under lab conditions
as the response can be experimentally triggered by applying a brief pulse of farred light. The light pulse sets a defined starting point on the time axis on which
subsequent events are observed. During about 18 hours after the trigger, the
entire plasmodial cell is extensively remodeled and fruiting bodies are formed
that give rise to mononucleate haploid spores that are precursor cells of amoebal
gamets which will develop at later stages of the cycle [7]. This process is called
sporulation. Please note that sporulation in bacteria, although the name is the
same, is biochemically a completely dierent process than in eukaryotic cells.
Five to six hours after an inductive far-red pulse, the cell is irreversibly committed to sporulation. The associated molecular events are of scientific interest
as the plasmodium loses some stem cell-like capabilities during commitment.
The expression pattern of hundreds of genes changes [15,7]. These changes in
gene expression that normally occur can be compared to the changes that are
seen in mutant cells that have lost their ability to be committed to sporulation
[3,20].

Automatically Composed Petri Nets

101

A widely-used method in the biosciences is the genetic dissection of gene

regulatory networks by generating mutants which are altered in the regulatory
control and analysing the phenotype which a mutation produces. Mutants may
be produced through forward and/or reverse genetic approaches. In forward
genetics, randomly mutated cells or organisms are screened for phenotypes of
interest and the gene which causes the phenotype is identified subsequently. In
the reverse genetic approach, a gene of interest is mutated and the phenotypic
consequence of the generated mutation is analysed. Mutation of a gene, both
in forward and reverse genetics, may either cause the loss of a protein or a
change in its function. Mutation may change the activity of a protein (up or
down) or it may change the specificity of its catalytic activity. In many cases,
the molecular mechanisms of how a given mutation translates into the observed
phenotype remain unknown for a number of years. Despite this ambiguity, the
genetic approach is powerful as it rigorously establishes causal dependencies
within the living organism. In most cases biochemistry alone could not fulfill
this task.
A powerful way to employ genotype/phenotype relationships for modelling
and simulation is the reverse engineering of genetic data. Reverse engineering
of gene expression data provides a direct link to bottom-up models of proteinprotein interactions. By reverse engineering one can establish eects, which a
mutated gene (the allele of a gene) exerts on a cellular process. We define allelic
influence modules to represent these influences. We will now show how allelic
influence modules are built and how they are useful for the integration of topdown and bottom-up model parts into one coherent model.
6.2

Linking Genotype to Phenotype: Allelic Influence Modules

As gene modules, allelic influence modules are centred around the allele of a given
gene. However, allelic influence modules dier from gene modules in representing the regulatory influences exerted by the allele on cellular processes by controlling the firing activities of respective transitions through read or inhibitory
arcs. In reality, these influences can be rather indirect by involving numerous
other, potentially unknown components. Accordingly, the allelic influence module may represent the control of molecular events like the biosynthesis of RNA
by transcription or even more complex processes of potentially unknown molecular mechanism as inferred from functional studies. To make this more clear, let
us consider the case study.
In response to far-red light, Physarum plasmodia dierentially express a large
number of genes several hours before the cell is irreversible committed to sporulation [20]. Genes with both up- and down-regulated RNAs have been identified
at a genomic scale [15,3], and the precise expression kinetics of some of them
have been investigated in detail [20]. Currently, we do neither know the molecular mechanisms through which these genes are controlled nor do we know which
causal consequences the change in expression level in detail have. However, the
majority of the dierentially expressed genes encodes proteins with high sequence
similarity to proteins of important regulatory function.

102

M.A. Bltke, M. Heiner, and W. Marwan

Fig. 9. Allelic influence module. The module represents the dierential regulation
of four genes, pcnA, pldA (down regulated), hcpA, and rsgA (up regulated) by the
spoN48 allele of the spoN gene in Physarum polycephalum. The logic transitions shown
here are also part of the gene modules of the four dierentially regulated genes (not
shown). Genes and names of their orthologs in the UniProt database: pcnA, proliferating cell nuclear antigen; pldA, Phosphatidylinositol-glycan-specific phospholipase D;
hcpA, Histone chaperone ASF1A; rsgA, Regulator of G-protein signalling 2.

We have genetically dissected the underlying regulatory network with the help
of mutants that are altered in the photocontrol of sporulation, as isolated in phenotypic screens ([32]; Rtzel et al., unpublished results). Representatives of one
class of these mutants have lost their ability to be committed to sporulation and
remain forever in a proliferative state. Although these mutants do not respond
to far-red light by sporulation, they clearly respond at the transcriptional level.
However the pattern of genes that are dierentially expressed in response to
the stimulus significantly diers in the mutants as compared to the wild-type
and also diers between mutants. The altered gene expression patterns clearly
reveal the regulatory influence of the mutated genes and can be used to infer
the network of regulatory control in which the dierent regulators inactivated
in each of the mutants are interwoven. The changed pattern of dierentially
expressed genes can be used to reverse engineer the regulatory influence of the
gene mutations. In Figure 9 the allelic influence module obtained for the phoN48
allele of the phoN mutant gene is shown. The proteins encoded by some of the
dierentially expressed genes have well-known biochemical functions in the developmental control of eukaryotic cells. In a bottom-up approach, these genes
can be linked to corresponding protein modules in terms of changed protein
concentrations as predicted by the model.

Versatility of the Approach and Future Perspectives

We have described a strictly modular approach to Petri net modelling based on

clearly defined types of modules corresponding to the dierent types of molecular
entities around which each module is centred: genes, RNAs, and proteins. The
small number of module types and the few easy-to-follow rules for creating a
module are expected to encourage community eorts in creating a collection of
modules in analogy to how Wikipedia collects pages. In Petri net modules, cross
references to other modules allow the automatic composition of large models
that are directly executable [4]. A web-accessible database was constructed to
manage dierent versions of each module which is available as a prototype [4]. It
allows that dierent explanations of molecular mechanisms directly translate into
alternative computational models that predict experimental findings. Moreover,

Automatically Composed Petri Nets

103

working with modules provides several options for the engineering of biomodels
and their scalable application to systems and synthetic biology.
7.1

Regulatory Interactions Appear in Clear Graphical Structure

Because each module summarizes all functional interactions of a given molecular

component including its influences on other components, even complex regulatory interactions can be always displayed in the form of an easy to perceive
graphical layout. Certainly, not all of the functional interactions that appear in
a module necessarily have to be part of a composed model. With the support of
a database, modules can be selected according to user-defined criteria and then
automatically linked to give a functional model. Those interactions that do not
find a counterpart in one of the selected modules remain inactive because the
respective places remain unmarked in the composed Petri net.
7.2

Modules May Be Added, Removed, or Exchanged: in silico

Mutation of Networks

The modular structure allows to remove or exchange modules when automatically composing a model without touching or even considering modules that
remain unchanged. This is a crucial advantage as compared to the manual reengineering of monolithic models which requires careful consideration of how the
components are wired up with each other since overlooking connections might
accidentally introduce modelling errors. With the modular approach and the
built-in version control of the database, dierent versions of a given module can
be easily exchanged. This can be very helpful to analyse how alternative kinetic
mechanisms of molecular interactions would influence the overall behaviour of
the system.
For example for the phosphate regulatory network, one might wish to analyse whether or not dierent activation mechanisms of the PhoB protein would
change the gene expression response and the performance of the feed-back loop.
When working with really complex models, the modeller investigating a local
mechanism has not to care about the inner life of all the numerous modules, in
analogy to programming where the procedures of an approved library of subroutines do not have to be reconsidered each time they are used for building a new
program.
At the moment where the database will contain a relatively high number of
modules, automatically generated models might reveal nonobvious regulatory
interactions of molecular components, bring them into a quantitative context
and predict nonobvious and eventually counterintuitive consequences of network
activation or perturbation. This will definitely be the case when regulatory interactions at a genome-wide scale change gene expression levels that translate
into an updated marking of the places of protein modules due to the change in
cellular concentration of perhaps many proteins. Even without modelling, this
is already obvious by just looking at the gene expression data mentioned in the
Physarum case study.

104

M.A. Bltke, M. Heiner, and W. Marwan

One might systematically probe components for their global role in the
biomolecular system by simply removing modules from a model. This is the
in silico complement to systematic mutant screens that are performed in genetic
model organisms. In silico mutational studies may turn out to be of great benefit for synthetic biology in all cases where systematic mutant screens cannot be
applied for what reasons soever.
7.3

Modules Can Be Automatically Generated at Large Scale

For genome scale models where the regulatory control of hundreds or thousands
of genes or proteins is to be considered, automatic generation of models becomes
an issue. By creating multiple copies of the module prototypes described here,
modules can be generated fully automatically simply by assigning names to
places and transitions. This is especially straightforward for the gene and mRNA
modules but also for protein degradation modules. We expect that automatically
generated large scale Petri nets will transmute into helpful tools for the reverse
engineering of models from transcriptomic and proteomic data sets.
7.4

Allelic Influence Modules Integrate Forward and Reverse

Approaches to Biomodel Engineering

Allelic influence modules were designed to represent regulatory influences of mutated genes (the alleles of a gene) onto the system. Unlike in the other module
types, these influences may be rather indirect and may involve a number of potentially unknown components. As we have shown, defining these modules allows
to reverse engineer networks from data collected on mutants. These reverse engineered networks are indeed fully compatible with the molecule-centred modules
through transitions that control the active states of a gene as shown in the case
study.
Acknowledgement. We thank Mostafa Herajy, Fei Liu, Christian Rohr and
Martin Schwarick for their continuous support in developing Snoopy.

References
1. Baldauf, S.L., Doolittle, W.F.: Origin and evolution of the slime molds (Mycetozoa). Proc. Natl. Acad. Sci. USA 94, 1200712012 (1997)
2. Baldauf, S.L., Roger, A.J., Wenk-Siefert, I., Doolittle, W.F.: A kingdom-level phylogeny of eukaryotes based on combined protein data. Science 290, 972976 (2000)
3. Barrantes, I., Glckner, G., Meyer, S., Marwan, W.: Transcriptomic changes arising
during light-induced sporulation in Physarum polycephalum. BMC Genomics 11,
115 (2010)
4. Bltke, M.A., Dittrich, A., Rohr, C., Heiner, M., Schaper, F., Marwan, W.:
JAK/STAT signalling - an executable model assembled from molecule-centred
modules demonstrating a module-oriented database concept for systems- and synthetic biology (submitted 2012)

Automatically Composed Petri Nets

105

5. Bltke, M.A., Meyer, S., Marwan, W.: Pain signaling - A case study of the modular
Petri net modeling moncept with prospect to a protein-oriented modeling platform.
In: Proceedings of the 2nd International Workshop on Biological Processes & Petri
Nets (BioPPN 2011), Newcastle upon Tyne, United Kingdom, pp. 119 (2011)
6. Bltke, M.A.: BioModelKit (2012), http://www.biomodelkit.org
7. Burland, T.G., Solnica-Krezel, L., Bailey, J., Cunningham, D.B., Dove, W.F.: Patterns of inheritance, development and the mitotic cycle in the protist Physarum
polycephalum. Adv. Microb. Physiol. 35, 169 (1993)
8. Cerutti, H., Casas-Mollano, J.A.: On the origin and functions of RNA-mediated
silencing: from protists to man. Current Genetics 50(2), 8199 (2006)
9. Chia, N.Y., Chan, Y.S., Feng, B., Lu, X., Orlov, Y.L., Moreau, D., Kumar, P.,
Yang, L., Jiang, J., Lau, M.S., Huss, M., Soh, B.S., Kraus, P., Li, P., Lufkin, T.,
Lim, B., Clarke, N.D., Bard, F., Ng, H.H.: A genome-wide RNAi screen reveals
determinants of human embryonic stem cell identity. Nature 468(7321), 316320
(2010)
10. Cuomo, A., Bonaldi, T.: Systems biology "on-the-fly": SILAC-based quantitative
proteomics and RNAi approach in Drosophila melanogaster. Methods in Molecular
Biology 662, 5978 (2010)
11. Durzinsky, M., Marwan, W., Ostrowski, M., Schaub, T., Wagler, A.: Automatic
network reconstruction using ASP. Theory and Practice of Logic Programming 11,
749766 (2011)
12. Durzinsky, M., Wagler, A., Marwan, W.: Reconstruction of extended Petri nets
from time series data and its application to signal transduction and to gene regulatory networks. BMC Systems Biology 5(1), 113 (2011)
13. Fisher, J., Henzinger, T.A.: Executable cell biology. Nature Biotechnology 25(11),
12391249 (2007)
14. Gerber, A., Luschnig, S., Krasnow, M., Brown, P.: Genome-wide identification
of mRNAs associated with the translational regulator PUMILIO in Drosophila
melanogaster. Proceedings of the National Academy of Sciences 103(12), 4487
4492 (2006)
15. Glckner, G., Golderer, G., Werner-Felmayer, G., Meyer, S., Marwan, W.: A first
glimpse at the transcriptome of Physarum polycephalum. BMC Genomics 9, 6
(2008)
16. Hecker, M., Lambeck, S., Toepfer, S., Van Someren, E., Guthke, R.: Gene regulatory network inference: data integration in dynamic models-a review. Bio Systems 96(1), 86103 (2009)
17. Heiner, M., Gilbert, D., Donaldson, R.: Petri Nets for Systems and Synthetic Biology. In: Bernardo, M., Degano, P., Zavattaro, G. (eds.) SFM 2008. LNCS, vol. 5016,
pp. 215264. Springer, Heidelberg (2008)
18. Heiner, M., Lehrack, S., Gilbert, D., Marwan, W.: Extended stochastic Petri nets
for model-based design of wetlab experiments. Transactions on Computational Systems Biology XI, 138163 (2009)
19. Heiner, M., Herajy, M., Liu, F., Rohr, C., Schwarick, M.: Snoopy A Unifying Petri
Net Tool. In: Haddad, S., Pomello, L. (eds.) PETRI NETS 2012. LNCS, vol. 7347,
pp. 398407. Springer, Heidelberg (2012)
20. Homann, X.K., Tesmer, J., Souquet, M., Marwan, W.: Futile attempts to dierentiate provide molecular evidence for individual dierences within a population
of cells during cellular reprogramming. FEMS Microbiology Letters 329(1), 7886
(2012)
21. Hsieh, Y.J., Wanner, B.L.: Global regulation by the seven-component Pi signaling
system. Current Opinion in Microbiology 13(2), 198203 (2010)

106

M.A. Bltke, M. Heiner, and W. Marwan

22. Ideker, T., Krogan, N.J.: Dierential network biology. Molecular Systems Biology 8,
565 (2012)
23. Kolch, W., Calder, M., Gilbert, D.: When kinases meet mathematics: the systems
biology of MAPK signalling. FEBS Letters 579(8), 18911895 (2005)
24. Marwan, W., Rohr, C., Heiner, M.: Petri nets in Snoopy: a unifying framework
for the graphical display, computational modelling, and simulation of bacterial
regulatory networks. Methods in Molecular Biology 804, 409437 (2012)
25. Moat, J., Sabatini, D.M.: Building mammalian signalling pathways with RNAi
screens. Nature Reviews Molecular Cell Biology 7(3), 177187 (2006)
26. Neidhardt, F., Ingraham, J., Schaechter, M.: Physiology of the Bacterial Cell. A
Molecular Approach. Sinauer Associates, Sunderland (1990)
27. Orton, R., Adriaens, M., Gormand, A., Sturm, O., Kolch, W., Gilbert, D.: Computational modelling of cancerous mutations in the EGFR/ERK signalling pathway.
BMC Systems Biology 3(1), 100 (2009)
28. Pinney, J.W., Westhead, R.D., McConkey, G.A.: Petri net representations in systems biology. Biochem. Soc. Trans. 31, 15131515 (2003)
29. Rohr, C., Marwan, W., Heiner, M.: Snoopya unifying Petri net framework to
investigate biomolecular networks. Bioinformatics 26(7), 974975 (2010)
30. Schwanhusser, B., Busse, D., Li, N., Dittmar, G., Schuchhardt, J., Wolf, J., Chen,
W., Selbach, M.: Global quantification of mammalian gene expression control. Nature 473(7347), 337342 (2011)
31. Sturm, O.E., Orton, R., Grindlay, J., Birtwistle, M., Vyshemirsky, V., Gilbert, D.,
Calder, M., Pitt, A., Kholodenko, B., Kolch, W.: The mammalian MAPK/ERK
pathway exhibits properties of a negative feedback amplifier. Science Signaling 3(153), 90 (2010)
32. Sujatha, A., Balaji, S., Devi, R., Marwan, W.: Isolation of Physarum polycephalum
plasmodial mutants altered in sporulation by chemical mutagenesis of flagellates.
Eur. J. Protistol. 41, 1927 (2005)
33. Uhlmann, S., Mannsperger, H., Zhang, J.D., Horvat, E.A., Schmidt, C., Kblbeck,
M., Henjes, F., Ward, A., Tschulena, U., Zweig, K., Korf, U., Wiemann, S., Sahin,
O.: Global microRNA level regulation of EGFR-driven cell-cycle protein network
in breast cancer. Molecular Systems Biology 8, 570 (2012)