What happens if you pose the wrong questions?

Abstract

In their recent review article entitled ‘Ten questions about systems biology’ Drs Joyner and Pedersen provide a comprehensive rebuke of what they consider the failure of genomic sciences and the lack of novelty of the concepts of translational research and systems biology (Joyner & Pedersen, 2011). In doing so, they draw analogies from the physiological sciences that they claim precede these new [sic] concepts. In fact their points are largely moot and I would argue that such misinterpretations and misrepresentations could ultimately be damaging to the field of Physiology and its future development if perpetuated among our scientific community. First we should attempt to define Systems Biology and its origins. A System can be defined as the ‘whole compounded of several components’ and thus systems biology aims to provide a model or integrated ‘signal’ that represents a complex biological process. Importantly the concept of a ‘system property’ or systems science originates from the time of Socrates – at the same time Hippocrates was fathering Physiology and Medicine. Joyner and Pedersen imply that the ‘vast intellectual and technical progress’ in the area of genetics and gene sequencing has consumed vast sums of funding to the detriment of the Physiological Sciences. Further, lack of progress in gene-therapy and disease risk stratification for diabetes is provided as a key example of the failure of Genomics. However, these arguments represent a substantial misdirection. Failure of Genome Wide Association Study (GWAS) to impact on the prognosis of Type II diabetes, for example, is not a failure of Systems Biology or even Genomics. It confirms what many guessed: Type II diabetes is not a singular entity driven by a handful of genetic variants. Failure of Physiologists to engage, coupled with a propensity to unjustifiably criticise (Joyner & Pedersen, 2011), has probably limited Physiology's impact in the GWAS field. We should not repeat this, and furthermore GWAS per se is nothing to do with Systems Biology. Indeed a systems biology approach has yet to be taken with GWAS-type data sets and the ROC (Receiver Operator Characteristic) analysis presented by Joyner and Pedersen is not an example of validation of a set of diagnostics but the sensitivity of a one-off data set. Specificity and reproducibility would also need to be addressed to guide treatment strategies. It is noteworthy that aspects of their presented analysis, when subject to intervention, did not produce even the numerical change expected in cardiovascular mortality or morbidity (Uusitupa et al. 2009). In reality, the first GWAS studies are crude and represent an incomplete Genomics strategy for the study of complex common disease – an immature discipline in short. Investment in new OMICs technology will drive that maturation process. And before we condemn the approach as one of misleading rhetoric and false hope, we have to reflect on what Physiology had achieved 10 years after its emergence as a science? When GWAS data can be yielded in a manner that allows a Systems Biology approach combined with more robust physiological phenotyping, then the true value of GWAS will emerge. Either way, Systems Biology remains untainted as GWAS is only one strategy that can feed into Systems Biology approaches. With cost-effective OMIC techniques, together with detailed individual physiological characterization we will be able to apply Systems Biology approaches to estimate each person's OMIC and environmental burden and provide diagnostic advice on disease risk. Currently, there is too much talk of ‘predicting’ physiological outcome using only one data set with no independent predictive assessment (Krabbe et al. 2009). A genuine systems-based prediction of disease risk, based on genomics data (for example) also does not mean that we will then ‘understand’ the genes that ‘cause’ disease – because many genes will contribute and the relative contribution of each gene will vary greatly from person to person and have differing contributions from organ to organ. A reductionist level ‘understanding’ is organic fool's gold and it is not an essential feature of Systems Biology. Joyner and Pedersen state ‘It [systems biology] also seems like an indiscriminate effort (see also the comments about hypothesis neutral science below) to throw more and different combinations of technology at the idea that if we only understood “what is wrong with the genes” we would gain vast new insights into disease’. This misses the point. Systems Biology could be applied to population-based marine biology with no interest or involvement in gene measurement. Large scale, expensive studies, be they epidemiological, gene sequencing or transcriptomic, all generate data which could be analysed using a Systems Biology approach. Even Systems Biology efforts that first rely on processes aimed at generating large amounts of gene-sequence or transcript (RNA) data are neither ‘hypothesis neutral’ nor simply aiming to find the ‘wrong genes’. They are the first step in a sophisticated multi-step process. Systems Biology takes over when the raw data is generated. Nevertheless, it is a fair question to ask if Systems Biology has done something useful with OMIC data that would convince a Physiologist. First and foremost the gene-sequencing revolution has yielded a new set of tools for physiologists and systems biologists alike to apply to medical research. If we did not have the annotated sequences of the genomes, there would be no possibility of having cost-effective gene chip technology. Both RNA- and DNA-detection technologies have impacted dramatically in cancer diagnostics and chemotherapy prognosis (Willenbrock et al. 2004; Kang et al. 2011). Joyner and Pedersen also fail to mention a novel Systems Biology approach to cardiovascular physiology. Using a combination of global transcriptomics and targeted genetics a molecular predictor of the magnitude of cardiovascular physiological adaptation in vivo in humans was produced (Timmons et al. 2010). Indeed the RNA abundance of 30 gene products had substantial prognostic value when predicting the outcome of 10–20 weeks of supervised fully compliant aerobic exercise training. The fact that the same genes also contained DNA variants that tracked with cardiovascular trainability confirmed the robustness of new research strategy. While this omission is rather surprising as Pedersen was a co-author on this new Systems Biology work and Joyner is a cardiovascular physiologist, the study presents new opportunities to tailor cardiovascular interventions in humans. It has been argued that one could easily use ‘physiological insight’ to pick these key genes that regulate cardiovascular biology or disease risk (Joyner & Pedersen, 2011). However, anyone examining this new article (Timmons et al. 2010) would see that most genes are poorly ‘understood’ in the classic reductionist physiological sense of the word and none of them are regulated by exercise. It is critical to point out that no pre-intervention physiological measure held predictive power in the same experiment. Furthermore, the Systems Biology and Physiological end-points in this study were fundamentally different and published separately. Thus Joyner and Pedersen confuse Genetics, and more particularly the study of DNA sequence, with broader concepts of genomics and other sequence-based strategies. They also ignore epigenetic modifications which can easily contribute rapid changes in human population phenotype when confronted with large environmental changes and will be increasingly determined using large scale OMICS technologies. Indeed, we would probably not have survived as a species if all we had was mendelian genetics as a strategy. Thus, measurement of the Transcriptome or the Metabolome can provide a fingerprint that integrates many factors, including physiological status, diet, gene sequence variation and so forth. Such large scale strategies are already proving useful in Type II diabetes and cardiovascular research (Shah et al. 2009; Wang et al. 2011) Consider Lists A and B in Table 1. Let us assume that both sets of ‘genes’, when measured in tissue, can predict vascular remodelling in response to hypoxia. Which list of gene products provides you immediate understanding of the biological process being studied? I would hazard a guess that to most people, at least those not hard-wired to PubMed, neither list is informative. In fact, even after literature analysis, list A would leave you with little clue. To a Systems Biologist, either list is entirely acceptable, as the names are of passing interest. Rather, it's the contribution the data measurements make to the development of a model. i.e. what is of value is the predictive capacity of the variance captured by their accurate measurement, and how this contributes to an overall model of how tissue remodels to become more vascular. The focus is on measurement and a clinically relevant endpoint per se. So, if determination of the seven hypothetically important genes above provides a model that can then be used a priori to assign a subject to a different experiment (testing the idea, for example, that they would remodel better at a greater or lesser degree of hypoxia) then Systems Biology can be seen to empower Physiological experimentation. Perhaps the signature explains tumour aggression and new ideas can be tested by assigning ‘high’ and ‘low’ responders to treatment alternatives? These can be post hoc analyses of clinical trials, to yield personalised medicine strategies, even without having ‘reductionist’ level ‘knowledge’ of the components of the predictive genomic signature. Thus such data can be used by the clinician to implement a more effective chemotherapy strategy. Perhaps the pharmaceutical company no longer requires classical physiological or biochemical understanding of the target, but rather screens for small-molecule compounds that bind to protein pockets, where potent hits are screened for efficacy and safety without trying to second guess mechanism of action in the classic sense. In fact, that is essentially how many effective prescribed drugs work, by binding many protein targets, some known, some unknown (Hopkins, 2009). Understanding the drug–genome interactions is critical, as lack of efficacy and unexpected toxicology have driven the failure rate in new drug development (Hopkins, 2009). Now armed with the seven hypothetical targets above, found by network analysis, we could tune the structure–activity relationship, in a network sense, using the protein target structures (even if we have no concept in the traditional sense of what the ‘gene does’) and in silico or ‘real’ ligand binding assays. This would be an example of taking a Systems Biology discovery, and using network biology and chemoinformatics, then applying it to an integrated physiological or disease endpoint (Hopkins, 2008). In conclusion, it will require great patience to implement the multi-stage process of Systems Biology. It also takes great vision to see past the early rhetoric of the OMICS technology pioneers and apply such methods in a robust manner to important medical questions. Physiologists have as big a role to play in integrating the ancient science of Systems analysis into medical research, a task that does not differ in principle from how physiology has integrated biochemical, genomic and pharmacological methods over the past decades. It may, however, hold much more long-term promise as long as we stop wishing to focus on what a single ‘gene’ does (Pedersen et al. 2003). I thank Professor Jonathan Elliot for insightful comments during the preparation of this commentary.