Figure 1.
Model assumption on protein complexes as hidden factors underlying strain's fitness.
Our model assumes that a protein complex (PC) is a functional unit to perform the biological processes in a cell, whose growth and survival is determined by the cooperative operation of a collection of PCs in a cell. (A) Suppose that a strain of Gene Y deletion has three PCs (PC 1, PC 2, and PC 3), and PC 2 and PC 3 are genetically and physically linked with Gene Y in a normal strain, respectively. (B) In general, genetic interactions have been involved in the genetic buffering in the redundant or parallel pathways, and the physical interactions tend to be involved in a sequential biological event through a serial pathway [49]. From this, the following scenario is plausible. When PC 2 and PC 3 are supposed to be the targets of a drug that blocks their functions, the inhibition of PC 2 by the drug will affect the growth of the strain of the gene Y deletion because some component of PC 2 cannot play a role of genetic buffer to some biological process involving Gene Y. In the extreme situation, the synthetic lethality or sickness will occur in the strain of gene Y deletion. In addition, the inhibition of PC 3 by the drug will also affect the growth of the strain because some component of PC 3 cannot interact with some component in the sequential biological process involving Gene Y anymore. Overall, the combined deleterious effects of the first neighbors PCs physically or genetically linked to gene Y will cause the unbalance of homeostasis of the strain. Consequently, the growth fitness of a strain treated with the drug (B) will be observed relatively lower than growth fitness of a strain treated with no drug (A).
Figure 2.
Procedures for inferring the hidden activities of a collection of protein complexes in a cell.
(A) A bipartite network illustrating the first-order relationships between protein complexes (PCs) and strains they are associated with. The definitions of “protein complex”, “strains”, and “association” in the study are as follows: the first yeast comprehensive protein complexes reported by Gavin et al. [8] are used as a collection of “protein complexes” in a cell. The “strains” are defined as a collection of pooled deletion mutants released from Saccharomyces Genome Deletion Project [1]. The “association” is defined as the existence of physical or genetic interactions between at least one of components in PCs and knockout gene product of a strain. In bipartite network, we assume that the relative growth fitness of strains under different chemicals (called drugs or bioactive compounds in the text) is mainly caused by the deleterious associations of PCs and strains (Figure 1). (B) The bipartite PC-strain network reconstructed by applying PC-based Bayesian factor analysis (PCBA). The bar charts within dotted circles in the top of panel show the relative activities of PCs depending on chemicals inferred from our analysis. The bar charts within each strain in the bottom represent the relative growth fitness under different chemicals, which are used as observed data for our analysis. The thicknesses of arrows in the middle denote the association strength between PCs and strains inferred from our analysis. The colors of red and blue indicate “positive” and “negative” association, respectively. (C) It shows two types of input data for PCBA, one of which is a prior knowledge data of genetic and physical interactions in the left. It is represented in the form of matrix containing binary associations of each strain (row) to PCs (column) (called Z matrix in the text). If there was the association between the knockout gene of a strain i and at least one of components in a protein complex j, we set zij = 1. Otherwise we set zij = 0. The other is the chemical-genetic profiles representing relative growth fitness of pooled deletion strains under various chemicals. As the observed data for PCBA, it is shown in the right (called E matrix in the text).
Figure 3.
Clustering analysis of protein complex activities.
(A) Bird eyes of two-dimensional hierarchical clustering analysis of protein complex activities (488 PCs by 82 drugs). (B) The PC-based hierarchical dendrogram of 82 drugs using relative activities of all of 488 protein complexes. The red star and red vertical bar indicates the group of drugs very closely clustered in the PC-based clustering but not in strain-based clustering using the same compendium [5]. The blue star and blue vertical bar represent the group of drugs very closely clustered in both clustering. The rectangular dashed line box shows the group of drugs clustered slightly differently in each clustering. The alphabets in parenthesis denote the groups of drugs with similar mode-of-action supported by literatures. (C) The strain-based hierarchical dendrogram of 82 drugs using the growth fitness of 3,418 strains with at least one measurement above Log2 ratio 0.5 as performed in the compendium paper [5]. (D) The examples of the clusters of PCs in similar biological functions. We selected three clusters which have at least two PCs with known function. The annotation in the right denotes “complex ID: complex name: complex localization”.
Figure 4.
Significantly sensitive protein complexes to drugs known for their target pathway.
The significance score of the effect of a compound on a protein complex was estimated using error function (see Method for details). When a significance score of a compound on a complex was less than a given threshold, and also that complex has relatively positive value of an activity, the complex was defined as the “sensitive complex” to that compound. For rapamycin in cluster I, there is only one sensitive complex, PC 321. There are four sensitive complexes, PC 379, PC 413, PC 148, and PC 366 to compounds in cluster II which are microtubule-poisons. Those DNA-damaging agents of cluster III have more than five sensitive complexes, PC 181, PC 170, PC 65, PC 290, and PC 424. The types of biological associations between each sensitive complex and their sensitive strains are available, and also GO analysis results of the set of those genes of such sensitive strains at http://pombe.kaist.ac.kr/CMA/ModeOfAction.pl.
Figure 5.
(A) The most sensitive protein complex to rapamycin, PC 321, is composed of ERG1 and SEC2, both of which are essential genes. Any of components in such complex may be physically or genetically associated with ELP3, TEF4, and TOR1 among gene products deleted in all the sensitive strains to rapamycin. According to model assumption, it can be interpreted as follows: Strains of complex-associated gene deletions have deleterious biological interactions with that complex. It may lead to decrease in the growth fitness of those strains in rapamycin. (B) Target of Rapamycin (TOR) pathway primarily regulated by Target of Rapamycin Complex 1(TORC1) with/without rapamycin. When the rapamycin is treated in a yeast cell, it binds FKBP12, forming toxic complex, which inhibits specifically TOR1, an essential component of TORC1. It gives rise to abnormal TORC1 signaling cascades related to broad biological functions, transcription, translation, mRNA stability and permeability.
Figure 6.
Target pathway of camptothecin.
(A) The most sensitive protein complex to camptothecin, PC 181, is composed of ULA1 and UBA3, both of which are non-essential genes. Any of components in such complex may be physically associated with RUB1, UBA3, ULA1, UBC12, and RPN4 among gene products deleted in all the sensitive strains to camptothecin. (B) Our proposed model of neddylation-enhanced and ubiquitin-dependent proteasomal degradation of Topoisomerase I-DNA complex stabilized by camptothecin in yeast (details in the text). In this model, we suggest that RUB1-attachment of CUL3 may enhance the degradation of TOP1-cleavable complexes and, therefore, the blocking of RUB1-conjugation pathway contributes to significant increase of the level of camptothecin toxicity in cell growth as shown in PC 181-associated strains.