Synthetic genetic interactions occur between two genes when the double mutant displays a phenotype much more severe than does either single mutant alone. Global networks of such interactions are now being systematically determined, spearheaded by the budding yeast genome. Genetic interactions reflect in vivo relationships between gene products. Extracting that functional information from such genetic networks is now possible by exploiting and modifying the key concept of congruence. Here, we focus on synthetic genetic interactions between pairs of null mutations in non-essential yeast genes. We summarize how to identify biological pathways from these emerging networks, using illustrative examples.
- chitin synthase 3 (Chs3)
- chitin synthesis
- functional pathway
- genetic interaction network
- synthetic lethal interaction
Full genome sequences should provide the complete inventory of genes that function together to specify the particular organism. Knowing gene sequences is not enough. Understanding gene function is a necessary prerequisite to bridge the gap between gene lists and the organism. Genetics should provide a very powerful tool in this quest. Although genome-scale genetics may be largely restricted to a select number of genetic model organisms, the information derived from these analyses should be relevant to many, if not most, homologous genes in other organisms.
Here, we will focus on genome-scale genetics and how it can be used to inform gene function by identifying functional pathways. We will concentrate on the genome of the budding yeast, Saccharomyces cerevisiae, because it is the vanguard of this approach. We will focus on interactions between genes and their utility to the individual researcher in probing the function of their favourite gene, using illustrative case studies.
Most of the recognized ∼6000 protein-encoding genes in the yeast genome  have been inactivated, one at a time . Here, we will focus on the approx. 5000 non-essential genes whose null (complete loss-of-function) alleles are viable.
Linking mutants to phenotypes should directly identify the in vivo roles of the 5000 gene products. Unfortunately, sufficiently diagnostic and discriminating phenotypes are not common: the link between gene and phenotype can be very opaque [2,3]. Even worse, many mutations do not result in any obvious phenotype . Why? First, appropriate phenotypes may not have been examined to date. Secondly, many genes may play a subtle role in the life of the cell. Thirdly, many genes may do a critical task, but their role is obscured by another gene that performs or can perform the same function in a redundant manner. Alternatively, another gene may indirectly compensate for, or buffer, the loss of the gene of interest.
Redundancy and compensation mechanisms are rife in the yeast genome. It is relatively robust to single mutations but is surprisingly sensitive to additional mutations [4,5] (see below). Pairwise combinations of mutations that together cause a dramatic phenotype represent a genetic interaction between the corresponding genes. Such genetic interactions are termed synthetic genetic interactions or synthetic enhancements .
In yeast, two systematic, high-throughput strategies have been developed to identify the synthetic genetic interactions that cause inviability or lethality: SGA (synthetic genetic array) analysis  and SLAM (synthetic lethality analysis on microarrays) . The methodologies are well described and discussed elsewhere . For our purpose, both methods take a null mutation in a ‘query’ gene of interest and screen the entire set of viable null mutations (∼5000) for those that cause lethality when present with the query mutation.
Some hundreds of yeast query mutations have been systematically screened for synthetic genetic interactions to date. The emerging global genetic interaction network is striking . First, most query mutations tested partake in one or more synthetic lethal interaction [5,7]. Most genes can thus become essential when in an altered genetic context. So much for genes only having subtle roles. Redundancy or robustness  is rife. Secondly, the average query gene partakes in >30 synthetic genetic interactions, almost all with a gene whose products act in distinct biological processes and sharing little or no sequence similarity to the query protein . Robustness is thus more common than simple redundancy.
What functional relationships between gene products underlie such synthetic genetic interactions between gene pairs? For null mutations in non-essential genes, the relationship must be parallel, since each null mutation alone obliterates the local pathway in which it acts [4,5,7]. The gene products must therefore act independently to support some biological process. Unfortunately, these parallel relationships can be very distant. The link between the genetic network around a gene of interest and the biological function of its product can be obscure. However, the pattern of synthetic lethal interactions holds the key.
Genes whose synthetic genetic networks are similar or congruent (‘the same shape’) should encode gene products of similar or related in vivo function (Figure 1) [5,9,10]. At one extreme, genes that act in the same linear functional pathway should have identical networks of genetic interactions, i.e. be fully congruent. In reality, full congruence is never seen not least because of experimental errors and because pathways are rarely strictly linear. In practice, an observed network overlap of >80% appears indistinguishable from full congruence (Figure 1).
At the other extreme, genes of no related function should have completely independent networks, i.e. zero congruence. Again, reality is not so simple. Because a minority of genes buffer a large number of others [e.g. HSP82 (heat-shock protein 82) has 272 known synthetic lethal interactions, 5.5% of the genome], some frequency of shared interactions can be observed even for functionally unrelated gene pairs . Although statistical analyses can suggest what minimal cut-offs may be prudent , informed intuition and biological understanding are likely to serve the individual investigator better. In our experience, an overlap of <20% is at best marginally significant.
Congruence can be assessed in a number of ways including by clustering the large-scale network by pattern (near neighbours) , or mathematically (based on the number and fraction of overlapping interactions shared between each gene pair) [9,10]. A simple and intuitive measure of congruence in genetic networks is the percentage of shared interactions for each of the two genes being considered (Figure 1; see below).
From congruence to pathway
Is congruence sufficient to specify in vivo functional relationships from genetic interaction networks? Very high congruence is certainly strong evidence for a close functional relationship [9,10]. However, many highly congruent gene pairs in the yeast genetic interaction network do not simply act in the same functional pathways: the genes themselves share a synthetic lethal interaction [9,10]. The gene products may have related roles but they must act, in part at least, independently. Congruence alone can focus on certain functional categories but not always with sufficient resolution (Figure 1).
We can now predict how genes that act in the same pathway or biological process (if they exist in the genome) would behave in the large-scale genetic interaction network: they would be highly congruent but not synthetic lethal with each other (Figure 1). This qualified congruence should constitute a very strict but potentially robust criterion for predicting whether two genes act in a single in vivo pathway or process.
Hereafter, we analyse some published synthetic lethal networks to illustrate the utility of congruence and qualified congruence in exploring and analysing gene function and in vivo pathways.
CHS3 (chitin synthase 3): chitin synthesis at the bud neck
The three chitin synthases of yeast are plasma-membrane-localized enzymes that manufacture the polymer chitin for subsequent incorporation into the extracellular cell wall [11,12]. The Chs3 isoenzyme is the best understood (Figure 2). The synthetic genetic network around CHS3 has been determined [6,13] and the gene lies in a very well explored part of the known global genetic interaction network .
CHS3 participates in 63 synthetic lethal interactions. Direct analysis of that network is minimally informative of Chs3 action . How about congruence? The genes whose networks overlap that of CHS3 will be discussed in descending order according to the extent of overlap.
The three most highly congruent networks with that of CHS3 belong to the following three genes: CHS5 encodes Chs5, a protein required for transport of Chs3 to the cell surface [11,12]; SKT5 encodes Skt5, a protein that is a direct activator of Chs3 and that also localizes the enzyme to the bud neck (Figure 2) [11,12]; and CHS7 encodes Chs7, another protein required for transport of Chs3 to the cell surface [11,12]. Because all the genes whose networks appear congruent with that of CHS3 have themselves been used as queries (they all lie in a well-explored part of the genetic network) [6,13], their full genetic networks are known. We can thus directly compare their networks with that of CHS3. The percentage overlaps (% of shared interactions, calculated with respect to both CHS3 and test gene networks) underlie congruence. For CHS5, shared interactions constitute 49% of its network and 59% of the CHS3 network. For SKT5, shared interactions constitute 75% of its network and 57% of the CHS3 network. For CHS7, the values are 84 and 41% respectively. It seems that Chs3 activity is not totally dependent on any of these three proteins but that Chs7 and Skt5 outperform Chs5 in their dedication to Chs3 function. Not only are three key regulators/mediators of Chs3 activity easily identified by using congruence, but we can estimate their devotion to Chs3 and how dependent it is on them.
The next two most overlapping networks belong to HSP82 and CHS6. The latter encodes a protein Chs6 that is required for transport of Chs3 to the cell surface [11,12]. Although it is devoted to Chs3, 72% of its network overlaps that of Chs3, that overlap accounts for only 29% of the CHS3 network. Chs3 thus appears to be only partly dependent on Chs6.
Could Hsp82 affect Chs3 activity? Shared interactions account for 29% of the CHS3 network, but a mere 7% of its own network. Is 7% overlap significant? Unlikely. HSP82 partakes in a large number of synthetic genetic interactions, over 270 corresponding to 5.5% of the genome. A 7% overlap is thus unimpressive, although chaperones can affect many different targets . Congruence tends to fail for targets that have many distinct functions or a very general function.
None of the above genes themselves share a synthetic lethal interaction with CHS3. This, coupled with high congruence, supports a ‘same pathway’ relationship for all except HSP82. The next four genes, SLT2, SMI1, HOC1 and RVS167, share 23–38% of their networks with that of CHS3, accounting for 27–29% of the latter. However, each of these genes shares a direct synthetic lethal interaction with CHS3. The corresponding proteins must therefore act, in part or in whole, independently of Chs3.
Of the remaining genes, only BNI4 stands out as sharing much of its network with that of CHS3, with overlaps of 74 and 27% respectively. BNI4 and CHS3 are not linked by a synthetic genetic interaction, consistent with a ‘same pathway’ relationship. Indeed, Bni4 binds to Skt5 and thereby localizes Chs3 to the bud neck [11,12]. The percentage overlap suggests correctly that Bni4's main role is to affect Chs3, whereas Chs3's absolute activity is only marginally dependent on it. The genetic network finds no compelling evidence for the involvement of other genes in the Chs3 pathway but Kre11 and Bem2 may have subtle roles.
FKS1 and CHS1
FKS1 encodes a major glucan synthase responsible for the synthesis of another key cell wall polymer . FKS1 partakes in 75 synthetic genetic interactions. Analysis of qualified congruence points to the SMI1 gene as encoding a component of the same pathway. Shared interactions account for 43% of the FKS1 network and 40% of the SMI1 network. Smi1 indeed regulates Fks1 activity, albeit indirectly . The main known regulator of glucan synthase is the Rho1 GTPase, an essential protein whose gene is not included in the analysis. Importantly, no components of the Chs3 pathway are predicted to act in the glucan synthase pathway (Figure 3). Analysis of genetic networks can clearly distinguish between closely related biological functions.
Can analysis of genetic networks distinguish between the differentially regulated chitin synthase isoenzyme Chs1 from Chs3 [11,13]? CHS1 partakes in 61 synthetic lethal interactions. In contrast with CHS3, analysis of qualified congruence fails to identify BNI4, SKT5, CHS6 or CHS7. Indeed, no clear ‘same pathway’ component emerges, consistent with the lack of any known, dedicated, non-essential control of Chs1 activity [11,12]. There is some modest overlap between the networks for CHS5 and CHS1, 20 and 25% respectively . The genes do not share a synthetic lethal interaction with each other, consistent with a ‘same pathway’ relationship if supported by sufficient levels of congruence. Chs5 may thus have a minor role in trafficking Chs1 in addition to its major role in delivering Chs3 to the cell surface. The prediction is not strong. In any event, analysis of genetic networks can even distinguish between the in vivo roles of isoenzymes (Figure 3).
Analysis of genetic networks can systematically cluster genes into functional pathways and with remarkable specificity and sensitivity. Analysis is also relatively straightforward. The data underlying the ever-expanding genetic interaction network of yeast can be readily accessed and analysed, using for example the Osprey™ network visualization tool .
There are, however, some limitations to using genetic networks. First, the more synthetic interactions that exist around a query gene of interest, the more powerful the analysis. Genes with few or no known interactions cannot be studied productively in this way (but see below for a qualification). Secondly, most genes products are unlikely to act in strict, linear pathways. For those that do, or whose function is largely overlapping, the method should be useful. Highly branched pathways or wide-ranging biological functions are less amenable to analysis. Thirdly, non-essential genes are the easiest to analyse. Networks around essential genes are poorly explored and their nature poorly understood. Fourthly, the screening methodology is prone to false negatives: it is not saturating. Fifthly, only a small subset of genes has been used as queries to date. As a consequence, the synthetic genetic network around most of the genes is at best only partly explored. For these genes, only interactions with query genes are known. All is not lost, but caution is recommended. Observing some overlap between the network around a query of interest and the partly known network around a non-query gene can be useful if extensive overlap is already evident. However, little can be concluded about non-query genes that have few or no known interactions: such genes are either refractive to interactions or, more likely, interact with an as yet unexplored part of the genome-wide genetic network.
In spite of such limitations, analysis of genetic interaction networks is proving very useful and will increasingly do so as network coverage improves. Determining the synthetic genetic network around a favourite gene of interest is an investment worth making.
British Yeast Group Meeting 2007: Independent Meeting held at the Paramount Palace Hotel, Buxton, U.K., 26–28 March 2007. Organized and Edited by A. Goldman (Sheffield, U.K.).
Abbreviations: CHS3, chitin synthase 3; HSP82, heat-shock protein 82
- © The Authors Journal compilation © 2007 Biochemical Society