Biophysics and systems biology

Is downward causation always discrete feedback or feedforward? The answer is no and the basis for that answer is profound, forming one of the reasons why I think that systems biology is revolutionary. A feedback loop can be closed. Feedback loops could exist between the levels of an organism, while the organism itself could still be modelled as a closed system. Yet, we know that organisms are not closed systems. Firstly they exchange energy and matter with the environment, including particularly other organisms whose existence forms a major part of the selection pressure. That is well recognized as a reason for regarding organisms as open systems. But there are other reasons also. I think that the best way to explain that is mathematical.

We model many biological processes as systems of differential equations. These equations describe the rates at which those processes occur. The number of such equations depends on the kind of question we are asking. At a cellular or subcellular (protein network) level, there may be a few dozen equations for the protein and other chemical entities involved. When we include structural details at the tissue or organ level, we may be dealing with millions of equations. Whatever the number, there is an inescapable requirement before we can begin to solve the equations. We must know or make plausible guesses for the initial and boundary conditions. They are not set by the differential equations themselves. These conditions restrain the solutions that are possible. In fact, beyond a certain level of complexity, the more interesting question becomes the explanation of that restraining set of conditions, not just the behaviour of the system, since the restraints may completely change the behaviour of the system. A restraint, therefore, is not necessarily a feedback. Restraints can be simply the background set of conditions within which the system operates, i.e. its environment. Through these interactions organisms can adapt to many different conditions. Their robustness in doing so distinguishes them from complex nonlinear systems that are highly sensitive to initial conditions or which end up unable to escape attractors.

This is a suitable point at which to return to the question of ‘genetic programs’. As we have seen, DNA sequences act as templates for proteins and as switches for turning genes on and off when they are in an organism, starting with the fertilized egg cell and maternal environment in the case of higher animals. A possible objection to my conclusion that the DNA sequences are better viewed as a database rather than as a program is that all programs require a computer to implement them. It was part of Monod and Jacob’s idea that, if DNA is the program, the organism is equivalent to the computer. Programs also do nothing outside the context of a computer. Could we somehow update this approach to save the ‘program’ metaphor? It is so ingrained into modern thought, among laypeople as well as most scientists, that it may now be difficult to convince people to abandon it. It is therefore worth spelling out, once again, what the difficulties are.

DNA sequences alone are not capable of being parsed as the complete logic of a program. Whenever we talk of a genetic program we must also include steps that involve the rest of the organism (e.g. my discussion of the ‘circadian rhythm’ program in Noble (2006, pp. 69–73), and this is certainly true for the analysis of cardiac rhythm (Noble 2006, pp. 56–65)). Much of the logic of living systems lies beyond DNA. To save the program metaphor therefore we would have to say that the ‘program’ is distributed between the tape and the machine. This would, incidentally, explain an important fact. Virtually all attempts at cross-species cloning fail to develop to the adult (Chung et al. 2009). A possible explanation is that the egg cell information is too specific (Chen et al. 2006). In fact, in the only case so far, that of a carp nucleus and goldfish egg, the egg cytoplasm clearly influences the phenotype (Sun et al. 2005). Strathmann (1993) also refers to the influence of the egg cytoplasm on gene expression during early development as one of the impediments to hybridization in an evolutionary context. There is no good reason why cells themselves should have ceased to evolve once genomes arose. But if we need a specific (special purpose) ‘computer’ for each ‘program’, the program concept loses much of its attraction.

The way to save the genetic program idea would therefore be to abandon the identification of genes with specific sequences of DNA alone and return to the original idea of genes as the causes of particular phenotypes (Kitcher 1982; Mayr 1982; Dupré 1993; Pichot 1999; Keller 2000b; Noble 2008c) by including other relevant processes in the organism. The problem with this approach is that the closer we get to characterizing the ‘program’ for a particular phenotype, the more it looks like the functionality itself. Thus, the process of cardiac rhythm can be represented as such a ‘program’ (indeed, modellers write computer programs to reproduce the process), but it is not a sequence of instructions separate from the functionality itself. This is another way to understand the quotation from Coen referred to earlier. The clear distinction between the replicator and the vehicle disappears and, with it, a fundamental aspect of the ‘selfish gene’ view.

If we do wish to retain the idea of a program, for example in talking about embryonic development where the concept of a ‘developmental program’ has its best applications (Keller 2000a), it might be better to think in the same terms in which we talk of neural nets being programmed. They are programmed by the initial setting up of their connections and then by the learning process, the set of restraints that allows them to ‘home in’ to a particular functionality. Those open-ended restraints are as much a part of the ‘program’ as the initial setting up of the system. The analogy with organisms as interaction machines is obvious. I am not proposing that organisms function as neural nets; only that the example of neural nets expands our concept of the word ‘program’ in a relevant way. The program is a distributed one (Siegelmann 1998) involving much more than DNA sequences, and is therefore far removed from Monod and Jacob’s original concept of a genetic program.

Where do the restraints come from in biological systems? Clearly, the immediate environment of the system is one source of restraint. Proteins are restrained by the cellular architecture (where they are found in or between the membrane and filament systems), cells are restrained by the tissues and organs they find themselves in (by the structure of the tissues and organs and by the intercellular signalling) and all levels are restrained by the external environment. Even these restraints though would not exhaust the list. Organisms are also a product of their evolutionary history, i.e. the interactions with past environments. These restraints are stored in two forms of inheritance—DNA and cellular. The DNA sequences restrict which amino acid sequences can be present in proteins, while the inherited cellular architecture restricts their locations, movements and reactions.

This is one of the reasons why systems biology cannot be restricted to the analysis of protein and gene circuits. The structural information is also crucial. Much of its evolution may have been independent of the cell’s own DNA since the early evolution of the eukaryotic cell involved many forms of symbiosis. The best known example is the mitochondria, which are now accepted to have originally been invading (or should we say ‘captured’?) bacteria, as were chloroplasts (Cavalier-Smith 2000, 2004). They even retain some of the original DNA, though some also migrated to the nucleus. There are other examples of symbiosis (Margulis 1981; Margulis & Sagan 2002; Williamson 2003, 2006; Williamson & Vickers 2007). Cooperativity may have been quite as important as competition in evolution (see also Goldenfeld & Woese 2007).

Cavalier-Smith has described some of these inherited features of animal and plant cells as the ‘membranome’, an important concept since lipids are not formed from DNA templates. An organism needs to inherit the membranome, which it does of course—it comes complete with the fertilized egg cell—yet another reason why it does not make sense to describe the organism as merely a vehicle for DNA. As I have argued elsewhere (Noble 2008c), the relative contributions of DNA and non-DNA inheritance are difficult to estimate (one is largely digital and so easy to calculate, whereas the other is analogue and hard to calculate), but the non-DNA inheritance is very substantial. It also contains many historical restraints of evolution.

This is the point at which I should attempt to explain the neo-Darwinian model and the modern synthesis and what is wrong with them from a systems viewpoint.

Neo-Darwinism brings together natural selection and nineteenth century genetics, while the modern synthesis (Huxley 1942) fuses Darwinism with twentieth century genetics. ‘Neo-Darwinism’ is the term often used for both of these syntheses. Darwin knew nothing of Mendel’s work on genetics. Moreover, he also accepted the idea of the inheritance of acquired characteristics, as did Lamarck (Lamarck 1809; Corsi 2001), who is incorrectly represented in many texts as inventing the idea. Darwin’s disagreements with Lamarck were not over the mechanisms of inheritance. Both were ignorant of those mechanisms. Their disagreement was more over the question of whether evolution had a direction or whether variation was random. Historically, we would do better to recognize Lamarck as the inventor of the term ‘biology’ as a separate science, and as championing the idea that species change (transformationism). Darwin can then be seen as discovering one of the mechanisms in his theory of natural selection, involved not only in transformations but also in the origin of species.

The problem with both revisions of Darwinism is that they involve a version of genetics that we need to revise. This version was one in which the central dogma of biology was taken to mean that the genetic material is never modified by the rest of the organism and the environment. Francis Crick’s original statements of the ‘central dogma of molecular biology’ (Crick 1958, 1970) do not in fact make such a strong claim. He stated a more limited chemical fact: that DNA sequences are used as templates to make proteins, but proteins are not used as reverse templates to make DNA. So, even if its proteins were to become modified during the lifetime of an individual, that modification cannot be inherited. The ‘dogma’ was then interpreted by many biologists to mean that information flows only one way. As we have seen, it does not. The quantities of proteins synthesized count as relevant information just as much as their amino acid sequences. But those quantities are most certainly dependent on signals from the rest of the system through the levels of transcription factors (including proteins and RNA) and the epigenetic marking of DNA itself and of the histone tails. All of this is open to the rest of the organism and to the environment to degrees we have yet to fully determine.

I will give just one example here to illustrate the potential significance of this openness. More examples can be found elsewhere (Jablonka & Lamb 1995, 2005). Neuroscientists have recently studied the epigenetic factors involved in maternal grooming behaviour in colonies of rats. Grooming depends on the environment. Colonies that are safe groom their young a lot. Colonies that are fighting off predators do not. This behaviour is inherited. The mechanisms are a fascinating example of epigenetic effects. The genome in the hippocampal region of the brain is epigenetically marked by the grooming behaviour and this predisposes the young to show that behaviour (Weaver et al. 2004, 2007). This is an important development, but as Weaver himself points out (Weaver 2009) it is currently restricted to one gene and one region of the brain. That underlines the importance of further research in this area. The implications of this form of epigenetic influence, however, are profound since it can transmit patterns of epigenetic marking through the generations even though they are not transmitted via the germline. This constitutes another form of inheritance of acquired characteristics to add to those reviewed by Jablonka and Lamb.

There is a tendency to dismiss such challenges to extensions of the central dogma as merely examples of cultural evolution. They seem to show rather that the boundaries between the different evolutionary processes are fuzzy. Once such interactions between behaviour and epigenetics are established and transmitted through the generations they can favour genetic combinations that lock them into the genome (Jablonka & Lamb 2005, pp. 260–270). This mechanism was originally described by Waddington (1942, 1957, 1959; Bard 2008), who demonstrated that, in fruitflies, just 14 generations of induced phenotype change could be assimilated into the genome. Mutations and genetic recombinations themselves are not random (Shapiro 2005). Moreover, they do not occur in a random context. They occur in the context of all the restraints exerted on the organism, including those of the environment. In such a process, it is the phenotype, not individual genes, that are the targets of selection (Keller 1999). Central building blocks of the neo-Darwinian synthesis are now known to be incompatible with the most recent discoveries in molecular biology.

I referred earlier to the ‘genetic differential effect problem’. In a previous article in this journal I have proposed that computational systems biology could provide a solution (Noble 2008c). The idea is basically simple. If our understanding and simulations are good enough they should include the robustness of biological systems, including their resistance to damage from mutations and knockouts. Moreover, if the models include representations of specific gene products (i.e. they extend down to the protein level) then it should be possible to reverse engineer to arrive at quantitative estimates of the contribution of each gene product to the functionality represented. That may be possible even if the system completely buffers the mutation or knockout so that no effect is observed in the phenotype. I give an example of this in the previous article from work on the heart (Noble 2008c). However, I would readily agree that, in its present state of development, computational systems biology is a long way from being able to do this in general. But it is worth bearing this in mind as an important long-term goal.

Work in the author’s laboratory is funded by the EU (Framework 6 and Framework 7), The British Heart Foundation, EPSRC and BBSRC. I acknowledge valuable criticisms from Charles Auffray, Jonathan Bard, Evelyn Fox Keller, Peter Kohl, Jean-Jacques Kupiec, Lynn Margulis, Laurent Nottale, James Shapiro, Hava Siegelmann, Eric Werner and Michael Yudkin.