As I am about to embark on my brief lecturing stint in the department, thoughts about the subject matter come to mind. One that always concerns me is the value of epistasis (the exercise to create functional hierarchies of genes through the analysis of the phenotypes of double mutants) to learn about processes in developmental biology.
The problem is simple. Genetics is to Biology what mathematics is to Physics: a formal language that allows us to pose questions and find answers. A mutant screen is, in many ways, the formulation of a problem: what controls the decision of a bacterium to use glucose or galactose? What determines the polarity of an embryo? How does a cell orient iself in space? What we find is a collection of mutants which, through some use of our experience (and the rules of genetic analysis), we relate to particular genes. Today, thanks to advances in genomics and the accumulated knowledge of molecular biology, we learn quickly about the kind of proteins that the genes encode and we can weave one of those â€œJust so storiesâ€ that HIF (High Impact Factor) journals like so much: how the gene X affects the nose of the mouse, how the chicken got its wings, or how a neuron lost its way on the way to a muscleâ€¦â€¦in any event, an important element of the story is to weave the genes together into a â€œfunctional networkâ€.Â If the story is good it will be bought by a HIF journal and you will have a grant, a job, etcâ€¦.Now, an important element of the story is the formulation o the relationship between the genes that relies on the analysis of double (and some times triple) mutants: epistasis.
When considering a regulatory process, epistasis allows us to create a linear chain of relationships between the elements (mutants) of the screen for pairwise interactions. The mechanics of epistasis is simple. It requires that the mutants under study (say A and B) have different phenotypes and then asks what is the phenotype of the double mutant (A, B). If the phenotype is B, this means that B is downstream of A i.e. A->B. If the phenotype is A, it is the other way around and A is downstream of B (B->A). If we now can do this with B and a new mutant C, we can construct an A->B->C chain and continue forward.
NB. This is fine in the abstract genetic world of phenotypes since, as those who grew up in metabolism know, when dealing with metabolic pathways, the reverse is true. Suppose you have two steps in which enzyme A catalyzes the change of X into Y, and then enzyme B catalyzes the change of Y into W; if A is mutant the reactions will accumulate Y and the action of B will be redundant i.e. in a metabolic pathway if the double mutant AB is A it means that B is downstream of A!!!! This difference highlights something interesting and perhaps deep about what it is that we measure with mutation; why the difference in this formal language between biochemistry and signalling? There are other uses of epistasis (see Phillips, P. (2008) Epistasis: the essential role of gene interactions in the structure and evolution of genetic systems. Nature Rev. Genet. 9, 855-867).Â
Leaving aside terminological issues, there are several problems with the use of epistasis as a way to uncover the functional structure of a biological system, which should make us concerned. Here I would like to highlight three:
1. Machines do not have linear blueprints. The physiology of cells (and this broadly speaking encompasses the sum of the processes that allow them to survive, interact. process information and reproduce) is operated through molecular machines which are often multicomponent. What counts is the operation of these machines as wholes and this does not emerge from a functional linear arrangement of their component parts. I am sure that being clever one could make an epistasis of the ribosome; but this is nonsensical, as what matters is its function as one piece, not our interpretation of its function as the result of a nonexistent array of linear instructions.
2. Feedbacks create problems of interpretation. Feedbacks and time series, create loops in linear systems which can complicate the analysis of epistasis. For example, imagine a ligand X for a signalling pathway Y and a second ligand Z for a second pathway W. It could be that X activates Y which leads to the expression of Z and the activation of W. Epistasis would easily show that X/Y is upstream of Z/W but if now W activation leads to the expression of X (of course now X would act in a different context), we would have a problem for epistasis as X is both up and downstream of Z/W and the genetics could become very confusing. If to this scenario we add the possibility of a feedback of Z/W onto X/Y, the situation that we extract from the genetics would not be simple and certainly not informative.
3. Double mutants are misleading. The reason for this is simple, in a double mutant AB one is not removing two genes from one organism, but rather one is removing gene A from mutant B, or gene B from mutant A; and which one is not possible to gauge. The problem is that, independently of the system, what we are doing in a double mutant is to study the response of a new situation; the loss of B in an A mutant, or the loss of A in a B mutant i.e the reaction of an established and adapted genetic system (the single mutant), one we know little about, to the loss of a gene. This is an important point which deserves some consideration.
These are serious flaws in an analytical tool that is so widely used, and all of them need to be considered when interpreting genetic data. Single mutants are simple to read but even these, sometimes, can become obscure. What can be done?
There are no bad tools, there are bad uses of tools. You cannot cut steak with spoon, nor eat soup with a knife, at least not effectively. It is important to know the limitations of our tools and this is something we should do with epistasis. It can be useful and, indeed, sometimes double mutants are informative, particularly of one is close to a molecular understanding. For example, if a mutant for MAP Kinase is epistatic with a mutant for MEK (the MAP Kinase Kinase) or a Receptor Tyrosine Kinase Receptor (RTK) -in particular if these are activated- we can put our money that MAPK is downstream of both (and this is indeed the case), and built a pathway: RTK->MEK->MAPK which tells us something useful about the system. But, as in the case of the ribosome, there is little use in trying to work out a linear pathway in the dynamics of actin polymerization and we should think more carefully about double mutants.
Final comment: High throughput genetics has brought up a high throughput version of epistasis. Some studies have looked for synthetic lethal phenotypes which reveal that the % of lethal mutant phenotype is higher in double/synthetic mutants than in single mutants. About 16% of single mutants are lethal (Tong, A. H. Y. et al. 2004 Global mapping of the yeast genetic interaction network. Science 303, 808â€“813) whereas, surprisingly, example in a screen targeted to genes required for chromosome segregation, out of 230 interactions between otherwise viable mutants, 18% (42) were synthetic lethal (Measday et al. (2005) Systematic yeast synthetic lethal and synthetic dosage lethal screens identify genes required for chromosome segregation. Proc Nat Acad Sci 102, 13956-13961). Other studies have corroborated these figures and provide a basis to think about QTLs and human multigenic disease (see Phillips above for a discussion of these matters).
Epistasis is about interactions and, as what we want is to uncover function, we should be careful in the way we interpret them. Particularly in these screens, whether they tell us about the ways to break the system when it is weakened or whether they are telling us something about how the system works. My money would be on the first and on the message that this contains: biological systems are redundant, highly connected networks, with multiple backups (see for example Guet et al.Â (2002) Combinatorial synthesis of genetic networks. Science 296, 1466-1470). Epistasis is a tool, and it needs to be complemented by many others when trying to infer the workings of the system. Some people call these secondary screens.