Introduction
The virtual physiological human (VPH) initiative is intended to support the development
of patient-specific computer models and their application in personalised and predictive
healthcare. The VPH, a core target of the European Commission's 7th Framework Programme,
will serve as a ‘methodological and technological framework that, once established,
will enable collaborative investigation of the human body as a single complex system'
(http://www.europhysiome.org/roadmap/). As such, the VPH initiative constitutes an
integral part of the international Physiome Project (http://www.physiome.org.nz/),
a worldwide public domain effort to develop a computational framework for the quantitative
description of biological processes in living systems across all relevant levels of
structural and functional integration, from molecule to organism, including the human
(Kohl et al, 2000; Bassingthwaighte et al, 2009).
So, what is the connection between this grand challenge and systems biology? To explore
this, we must first agree on what we take systems biology to mean.
Systems biology
Description versus definition
Descriptions of systems biology range from the view that it is merely ‘new wording,
more fashionable, for physiology' (http://is.gd/tQJL), to the all-inclusive ‘systems
biology involves the application of experimental, theoretical, and computational techniques
to the study of biological organisms at all levels, from the molecular, through the
cellular, to the organ, organism, and populations. Its aim is to understand biological
processes as integrated systems instead of as isolated parts' (http://is.gd/tQK0).
At the same time, attempts to concisely define systems biology have not yielded definitive
form of words that is acceptable to the majority of researchers engaged in what they
consider to be systems biology.
One of the reasons for this situation may be that many different scientific streams
have come together in the systems biology pool (see also Bassingthwaighte et al, 2009),
each with its own conceptual and terminological legacy.
But, another possible explanation for this apparent shortcoming is that systems biology
may constitute an approach (as detailed below), rather than a discipline (such as
biology), or a destination (such as the VPH). Such a scientific approach can be explained
descriptively, but cannot necessarily be defined prescriptively.
In either case, the lack of a generally acceptable definition of systems biology need
not be regarded as a surprise, or even as a disadvantage, as the artificial uniformity
that could be associated with a definition might exclude important current or future
work.
Terminological origins
It may be helpful, at this stage, to step back and consider the etymology of terms,
before discussing their possible interrelation.
Biology is contracted from bios (Greek for ‘life') and logos (Greek for ‘reasoned
account'). It is the science, or the logic, of life (Boyd and Noble, 1993).
A system is ‘the object' of the activity synthithemi (Greek for ‘I put together')
and has been defined as follows: ‘A system is an entity that maintains its existence
through the mutual interaction of its parts' (von Bertalanffy, 1968). In keeping with
this concept (Figure 1), research into systems therefore must combine:
the identification and
detailed characterisation of the parts, with the
investigation of their interaction with each other and
with their wider environment, to
elucidate the maintenance of the entity.
Subject matter
On the basis of the definition of a system, systems biology can be seen as a conceptual
approach to biological research that consciously combines ‘reductionist' (parts; points
i and ii) and ‘integrationist' (interactions; points iii and iv) research, to understand
the nature and maintenance of entities (point v). In biological systems, preservation
of entity includes a broad range of behaviours, including growth and development,
adaptation and maladaptation, and progeny, which explains why streams from so many
different research directions must be pooled.
In addition, the ‘parts' of a biological system (e.g. organs of a body, or tissues
within an organ, etc.) can usually be broken down into smaller biologically relevant
entities (such as cells, proteins, amino acids), which—when focussing at a lower level
of structural integration—form ‘systems' in their own right. This illustrates two
further points: first, systems biology as an approach can be applied to research targets
independent of their ‘scale', that is, their level of structural and functional complexity
and second, no particular scale has privileged relevance for systems biology (Noble
2008a, 2008c). From the multi-scale nature of biological systems, it follows further
that systems biology inherently involves a multi-scale approach (see below).
So, does this mean that there is nothing special about systems biology? Is it really
just another, more fashionable label for good old physiology?
Probably not. Systems biology forms a logical juxtaposition to the recently prevailing
‘reductionist' drive, serving as the ‘post-genomic' manifestation of the need to balance
dissection and synthesis. Certain aspects of systems biology do indeed mirror the
‘pre-genomic' approach of subjects such as physiology, but at a higher level. Thus,
Claude Bernard showed the way as early as the 19th century and specifically called
for the mathematical analysis of biological phenomena (see Noble, 2008a). However,
with a few notable exceptions, such as the Hodgkin–Huxley equations for the nerve
impulse (Hodgkin and Huxley 1952), their application to the heart (Noble, 1962), or
the early ideas of Guyton for a quantitative model of the circulation (Guyton et al,
1972), classic physiology largely lacked the ability to pursue the quantitative integration
of observed behaviour. This may be one reason why it failed to compete with the rise
of molecular biology, which was perceived to be more solidly quantitative. In fact,
many academic departments of physiology became molecular or cellular, in focus and
in name.
Having turned full circle on what the dialectic method depicts as a three-dimensional
spiral of development, we have come ‘back to the future', now that bio-science can
harness the power of mathematics and computation and apply it to a re-integration
of the pieces of the jigsaw—which have been amply provided by reductionist research
approaches. Systems biology therefore thrives on the revolutionary improvement of
experimental techniques to investigate system components and their interactions, and
on significant advances in computational power, tools, and techniques, which allow
quantitative modelling and reintegration at hitherto unimaginable detail and breadth.
Modern computational models thus address points (i) to (v) above, and project between
them, while observing elementary rules such as conservation of mass, energy, and matter
and taking into account natural restrictions imposed on parts and interactions by
the system's own properties (e.g. a water-based solute system will impose different
constraints compared to a hydro-carbon based one; dark-blue background in Figure 1).
So, perhaps this is where the essence of systems biology lies: by providing a framework
for the re-unification of biological studies with ‘the other' sciences, and their
joint application to iterative reduction and synthesis, it forms the approach on which
quantitative descriptions of parts (i and ii) and their interactions (iii and iv)
give rise to an understanding of the maintenance of biological entities (v) across
all relevant levels of structural and functional integration (Figure 2).
An important aspect of this summary is the plural of ‘quantitative description'. Like
their experimental counterparts, computational models are—by the very definition of
the term ‘model'—simplified representations of reality. Like tools in a toolbox, models
for biomedical research, whether ‘wet' or ‘dry', have a range of applications for
which they are suitable. This suitability is affected by the extent to which models
are representative of the aspect of reality that they mimic; relevant for the question
under investigation; reasonable in terms of their cost (including not merely financial
considerations, but also resources such as time, training requirements, or ethical
dimensions); and reproducible (a challenge also for computational models, not only
when they include descriptions of stochasticity, but also when they exhibit language-,
compiler-, or hardware-dependence) (Kohl et al, 2006). Thus, the multi-level nature
of biological systems must find suitable reflection in an integrated set of multiple
models, both experimental and computational. This will be discussed next in the context
of the VPH initiative.
Systems biology and the VPH
The VPH initiative
As its name suggests, the VPH initiative targets the whole human body as the system
of interest. But, it does not herald a return to classical top-down physiology from
entity to parts. The aim is to understand human physiology quantitatively, as a dynamic
system, and at all relevant levels between genes and the organism.
Equally, it is not a bottom-up analysis from parts to entities. This would be impossible,
both conceptually (as the ‘parts' of the whole organism form systemic ‘entities' of
their own), and practically (as the number of possible combinations of interactions
between the products of 25 000 genes is simply too vast (Feytmans et al, 2005)).
The approach is better characterised by a term introduced by Sydney Brenner, ‘middle-out'
(Brenner et al, 2001), which is based on conceptualising insight at whichever level
there is a good understanding of data and processes, and on then connecting this to
higher and lower levels of structural and functional integration. In a system of multi-level
interactions that involves both regulatory feedforward and feedback pathways, as well
as environmentally prescribed parameter constraints, there is really no alternative
to breaking in at one level (the ‘middle' part of the metaphor) and then reaching
‘out' to neighbouring levels using appropriate, experimentally founded and validated
mathematical methods (Bassingthwaighte et al, 2009).
Of course, one has to be aware of the possible (and in the present case counterproductive)
association of the expressions ‘higher' or ‘lower' level with ‘superior' or ‘inferior'
in terms of relevance for systems function. Regulatory interactions are, by definition,
two-way (‘regulatory loop'), and the metaphoric use of high and low is associated
here simply with the notion of spatial scale, not relevance. Furthermore, it is important
to realize that influences from ‘outer' levels to the ‘middle' are equally relevant.
One might call this an outside-in approach, illustrating the utility and limitations
of metaphors, simplified representations of a concept or idea (models!), which are
not necessarily of much help when used outside the applicable contextualisation for
which they were developed.
A lead example: systems biology of the virtual heart
We will illustrate the ideas discussed above by considering the modelling of cardiac
structure and function, partly because that is the area of our own research, but also
because, by common consent, it is the most highly developed example of a virtual organ,
with applications already within the pharmaceutical industry and in the development
of medical devices (Hunter et al, 2001; Noble 2008b). There are three reasons for
this situation.
First, cardiac cell models have now benefited from a track record of nearly 50 years
of iterative interaction between modelling and experimentation, with an accumulating
body of insights derived as much from the ‘failures' as from the ‘successes' of theoretical
prediction and experimental validation (Noble 2002). In fact, the contradiction of
predictions—whether based on hypotheses formed in thought experiments (conceptual
models) or quantitative simulation (computer models)—is usually more instructive than
their confirmation. Although confirmation increases the confidence associated with
a particular concept or model, contradiction highlights shortcomings in the quality
and/or quantity of data input, processing, or interpretation. This will prompt additional
observation, consideration, and conceptualisation, with the potential of advancing
models and insight (Kohl et al, 2000).
Second, despite its complexity, the heart shows pronounced spatial regularity in structural
properties (from the tissue level right through to the arrangement of subcellular
protein- and membrane-structures), and it is governed by a very high degree of spatio-temporal
coordination of key functional behaviour (such as the spreading wave of electrical
excitation that invokes every single cardiomyocyte during each heartbeat, or the highly
orchestrated sequence of ionic fluxes and protein interactions that give rise to remarkably
optimised pressure generation some 2.5 billion times in the healthy human heart during
a life time).
Third, systems of interaction in the heart show a considerable degree of modularity.
Basic models of cardiac electrophysiology, for example, do not need to take into account
interactions with cardiac mechanics, circulation, metabolism, and so on, to predict
important aspects of the interplay between ion distributions, currents, and voltage
changes. As they become increasingly detailed, however, wider interactions become
more and more relevant, as entities that were classically believed to be linked in
a one-directional manner are subject to cross-talk and interaction. Examples include
the interdependence of cardiac structure and function (Allessie et al, 2002), of ion
channels and cell or tissue behaviour (Hodgson et al, 2003), or of electrophysiology
and mechanics (Kohl et al, 2006).
Work on the virtual heart has advanced with progressively increasing complexity. The
earliest cell models had just three differential equations that represented the summary
kinetics of multiple ‘lumped' electrical mechanisms which, by and large, had not yet
been identified and were not, therefore, strictly related to individual protein channel
subtypes as we know them now. Cell models today may contain 50 or more equations (Ten
Tusscher et al, 2004), depending on the extent to which individual ion handling mechanisms
are represented (e.g. through Markov models of ion channels (Clancy and Rudy, 1999))
and the complexity with which intracellular structural features are simulated (Pásek
et al, 2008). The insertion of such models into tissue and organ models has also occurred
at different levels of tissue size and complexity. Although the goal of reconstructing
the whole organ with representative histo-anatomical detail is important for some
applications (Burton et al, 2006; Plank et al, 2009), much insight can be gleaned
from multi-cellular simulations using one-dimensional strands of cells, two-dimensional
sheets, and three-dimensional simplified tissue geometries (Garny et al, 2005). The
overall lesson from these simulations has been that theoretical models of biological
behaviour are most efficient when they are as complex as necessary, yet as simple
as possible.
Extension of principles from heart to other systems: opportunities and challenges
We do not have the space here to review the modelling of other organs and systems.
Readers can find out more by accessing the websites of the Physiome Project (http://www.physiome.org.nz/)
and the VPH (http://www.vph-noe.eu/). However, some of the approaches and principles
developed for, and applied to, cardiac modelling may be transferrable to other aspects
of the VPH initiative. Among the features that are already being tackled with some
success by the Physiome community are several general issues related to the various
types of modelling approaches and their role in the discovery process (Box 1). These
principles have emerged largely from grass-roots development of model systems in the
cardiac field. Although instructive, there is of course no reason to regard them as
prescriptive indicators of how other VPH-related projects should be pursued.
The reason for this is straightforward and bears relevance for systems biology in
general: we simply do not know what approach will eventually succeed. Researchers
pursuing a systems approach can be likened more to people finding their way through
unchartered territory, than to those walking a path that has already been mapped.
Contrary to the Genome Project, we do neither know the smallest part that we need
to identify (there is no elementary set of generic building blocks from which we can
assemble the jigsaw), nor the extent of the overall entity (in terms of the types
and number of interactions that need to be quantified). We have to determine the best
approach as we try out various ideas on how to modularise, simplify, connect multiple
levels, relate different aspects at the same level, and incorporate increasingly fine-grained
structural and functional data. At the same time, we are also seeking mathematical
approaches and computational resources that will enable models to be run in a reasonable
period of time (Fink and Noble, 2009), while using user interfaces that allow utilisation
by non-experts in computational modelling (Garny et al, 2003). These considerations
are associated with a number of additional challenges that have also been experienced
in the cardiac modelling field, but are far from being resolved (some examples are
listed in Box 2).
Of particular relevance, in our view, is the need to establish public access to data
and models derived from publicly funded work. This could be regarded as a make-or-break
issue, as crucial for systems biology as was the decision by a majority of Genome
Project investigators to publish and share information on annotated gene sequences,
obtained through publicly funded research (rather than patenting them, which would
have invoked a whole host of ethical, scientific, and socioeconomic dilemmas).
In this context, a range of ethical issues arise. We will refer briefly to just three
of them here. The first is one of scientific integrity and social responsibility (and
inherently underlies the drive towards public access to data and models): to serve
the usual criteria of scientific scrutiny and public accountability, and to avoid
‘re-inventing wheels', it is required to enable others to review, (re-)use, develop,
and efficiently apply prior work. From this, a second issue arises, related to professional
development and career progression: as long as the prevailing approach to assessing
‘academic merit' disproportionately rewards ‘peer-reviewed' publications as the output
of academic endeavour, compared with the (often very time consuming) development of
‘peer-used' tools, sharing data and models may end up disadvantaging those professionals
who generate them (by relieving them of control over and, conceivably, co-authorship
in their follow-on use). A third ethical aspect is the obvious need to protect the
privacy of individuals' data (a common challenge to using, re-using, and sharing human
data). An international solution to these challenges may be regarded as a second make-or-break
issue for systems biology and the VPH.
Conclusions
Systems biology may be interpreted as a scientific approach (rather than a subject
or destination) that consciously combines ‘reductionist' (identification and description
of parts) and ‘integrationist' (internal and external interactions) research, to foster
our understanding of the nature and maintenance of biological entities. During the
decade or so in which systems biology has become popular, it has often been interpreted
as an extension of molecular biology, here to foster the understanding of subcellular
regulation networks and interaction pathways, essentially equating ‘system' with ‘cell'.
While representing an important aspect of the systems approach, there is no a priori
reason why one level of structural or functional complexity should be more important
than any other (Noble, 2008a). Work involving more complex levels of structural and
functional integration is essential if systems biology is to deliver in relation to
human physiology and health care. In addition to this vertical integration across
multiple scales, we also need horizontal integration across boundaries such as between
organ systems, and between ‘wet' and ‘dry' modelling. Often, the best results are
obtained when theoretical work is pursued in close and continuous iteration with experimental
and/or clinical investigations. An essential task for systems biology is therefore
the quantitative integration of in-silico, in-vitro, and in-vivo research. Key make-or-break
issues are the extent to which we can harvest the synergies between the multiple international
efforts in the field by sharing data and models, and the question of how to address
the ethical dimensions of relevant research and development in this area.
Editorial Note
This Guest Editorial was commissioned on the occasion of the EMBL/EMBO Science & Society
Conference on ‘Systems and Synthetic Biology: Scientific and Social Implications',
Heidelberg, November 7–8, 2008. Additional contributions from several speakers are
available on the EMBO Reports website (http://www.nature.com/embor).
Box 1 General principles learned from the cardiac modelling field
Conceptual Duality: the combined application of reductionist and integrationist tools
and concepts lies at the very heart of successful development of a quantitative understanding
of systems behaviour. The analysis of heart rhythm resulting from individual protein
interactions (reductionist aspect) and their integration through feedback from the
overall cell electrical activity (integration) is a good example (Noble, 2006, chapter
5).
Iteration of Theory and Practice: ‘wet' experimental and ‘dry' theoretical models
need to be developed in continuous iteration, where new experimental (or clinical)
data feed model development and/or refinement, while computational predictions are
used to guide hypothesis formation and experimental design, the outcome of which is
the used to validate model predictions. A good example of this approach can be found
in the papers of Lei and Kohl (1998) and Cooper et al (2000), which used modelling
to interpret experiments showing an unexpected effect of cell swelling on pacemaker
frequency, leading to work using axial stretch to yield the expected result, also
explained by the modelling.
Structure–Function Relationship: biological function cannot be dissociated from underlying
structure. This finds a reflection in modelling, whether using ‘lumped parameters'
to describe general compartmentalisation (Orchard et al, 2009) or detailed representations
of three-dimensional morphology of proteins (Young et al, 2001), cells (Iribe et al,
2009), or organs (Zhao et al, 2009). Increasingly, this effort benefits from standards,
tools, and markup languages, such as SBML (http://sbml.org/Main_Page), CellML (http://www.cellml.org/)
and FieldML (http://www.fieldml.org/).
Multi-Scale Modelling: models at different scales of structural integration are required
to explore behaviour from molecule to organ or organism. This applies equally to ‘wet'
and ‘dry' research, and involves bridging spatial scales of (at least) nine orders
of magnitude (from nm to m) and temporal scales spanning 17 orders of magnitude or
more (from nanoseconds for description of molecular motion, to years or decades, for
longitudinal assessment of human development in norm and disease (Hunter and Borg,
2003). This requires application of ‘new maths' to systems modelling, for example,
scale relativity theory (Auffray and Nottale, 2008; Nottale and Auffray, 2008).
Multiplicity of Models (at each individual level): the availability of models of differing
levels of complexity, even at the same level of structural integration, allows the
treatment of the same biological aspect in different ways, depending on the nature
of the question being addressed (for examples see Noble and Rudy, 2001). Although
this is common practice in ‘wet' studies, it is often questioned in ‘dry' research.
Multi-dimensional Modelling: models from 0D to 3D+Time are needed to analyse parts
of the system that may, in some situations, be regarded as point-sources (e.g. cell
electrophysiology when looking at gross electrical behaviour such as reflected in
the electrocardiogram), and in others as complex spatio-temporally structured reaction
environments (such as the same cell when considering signal transduction cascades).
For an Open Source environment designed to address this see Bernabeu et al (2009).
Multi-physics Modelling: addressing questions of varying character, from the stochastic
behaviour of ion-channel-interactions to deterministic links between events, or from
multiple ODE systems to soft tissue mechanics and fluid dynamics, require different
implementations (e.g. finite differences, finite elements, or boundary element methods,
Hodgkin–Huxley versus Markov formalisms (see e.g. Fink and Noble, 2009), diffusion
reaction versus Monte Carlo approaches, etc).
Modularity of Models: a desirable but thus far ill-implemented need is the definition
of model interfaces. These may range from true modularity of components, to translation
tools or black-box style parameter inheritance. Likewise, model mapping is an area
where much more research into theoretical understanding and practical tools is called
for (Terkildsen et al, 2008).
High-Speed Simulation: application to real-world scenarios, in particular for time-critical
emergency settings, calls for faster-than-real-time simulation. The new generation
of supercomputers (e.g. the 10 petaflop machine being constructed for RIKEN in Kobe,
Japan) combined with improved algorithms is expected to make this possible (Bordas
et al, 2009).
Interactivity: interactive assessment of model behaviour is relevant for efficient
implementation of ‘dry' experiments, as well as for training, education, and interaction
between experts from different professional backgrounds (Garny et al, 2009).
Box 2 Issues and Challenges
Model Curation and Preservation: the long-term preservation of data and models and
the maintained ability to access digital data formats are recognised challenges of
modern IT infrastructures. They also present key concerns for the VPH initiative.
Tools, Standards, Ontologies and Access: concerted efforts have been launched to facilitate
the identification of suitable tools, standards, and ontologies to support model development,
interaction, and access (Hucka et al, 2003). This is one of the declared aims of the
VPH initiative and requires a willingness to
contribute to the development of standards;
adhere to ‘good practice', once standards are agreed; and
share and publish data, metadata, and models in a suitably annotated, re-usable format.
Patient-specific Analysis and Treatment: as non-invasive data-rich imaging methods
are becoming increasingly productive in the clinical setting, the goal of incorporating
patient-specific data into models for use in diagnosis, treatment planning, and prevention
is beginning to become a reality. This goal is desirable for a variety of reasons,
ranging from economic (it makes sense to choose treatments that are tailor-made for
the patient, rather than block-buster medicines that often miss the target) to ethical
(we should look forward to the day when we no longer tolerate disastrous side-effects
that could be eliminated by stratifying the patient population) and scientific considerations
(prevent, and if that is not possible, treat the patient—not the disease).