Abstract
Mutational robustness is the extent to which an organism has evolved to withstand the effects of deleterious mutations. We explored the extent of mutational robustness in the budding yeast by genome wide dosage suppressor analysis of 53 conditional lethal mutations in cell division cycle and RNA synthesis related genes, revealing 660 suppressor interactions of which 642 are novel. This collection has several distinctive features, including high co-occurrence of mutant-suppressor pairs within protein modules, highly correlated functions between the pairs, and higher diversity of functions among the co-suppressors than previously observed. Dosage suppression of essential genes encoding RNA polymerase subunits and chromosome cohesion complex suggest a surprising degree of functional plasticity of macromolecular complexes and the existence of degenerate pathways for circumventing potentially lethal mutations. The utility of dosage-suppressor networks is illustrated by the discovery of a novel connection between chromosome cohesion-condensation pathways involving homologous recombination, and Huntington’s disease.
Biological interaction networks are robust to perturbation1,2,3,4,5 because of several features, including power-law network topology, redundancy, modularity, and their dynamic properties2,6,7,8,9,10,11,12,13,14,15,16,17. Although modularity is a common feature of interaction networks12,18,19,20,21,22, the contribution of modularity to genetic robustness is difficult to determine. Recent studies have revealed dynamic interaction among apparently unrelated gene modules in response to genotoxic stress, suggesting the existence of highly reconfigurable networks of gene and protein modules as well as of unexpectedly plastic macromolecular complexes19. Rewiring of signaling and/or metabolic networks have been observed in cancer cells that evolved chemotherapy resistance23,24. It has been speculated that genetic and epigenetic changes could accomplish network rewiring25,26. Complex genomic changes reminiscent of functional rewiring are associated with rapid evolutionary adaptation in response to mutation in an essential gene27.
The suppression of essential gene mutations has been classically employed to investigate gene function, and suppressors provide clues to mechanisms of evolution28. More recently, genome wide dosage suppressor discovery has contributed novel insights into biological processes29,30. By seeking dosage suppressors of 53 mutant genes, mostly encoding cell cycle/DNA replication related genes or RNA polymerase/RNA modification related genes, we have uncovered 660 pairs of “dosage-suppressor” interactions, which supplement 1,626 dosage-suppressor pairs reported previously in the literature31 and 254 interactions that were discovered in a recent high throughput experiment29. The set of interaction pairs discovered in this study overlaps significantly with known protein-protein and genetic interaction pairs, suggesting related molecular mechanisms that connect the suppressed mutant gene product with the suppressor gene’s product. The mutant-suppressor pairs tend to belong to the same protein-protein interaction modules as determined by several independent criteria. We confirm all previously noted mechanisms of suppression29, including stabilization of heat sensitive macromolecular complexes by over-producing a physically interacting protein partner (Fig. 1a), stabilization of unfolded heat sensitive proteins by chaperones, and modulation of expression of other genes participating in the same biological process as affected by the mutant gene. Here we present evidence of at least one additional mechanism of network robustness, namely, the existence of degenerate pathways. The utility of dosage-suppression network is illustrated by the serendipitous discovery of a novel connection between chromosome cohesion-condensation pathways and the toxic effects of the Huntington’s disease protein.
Results
A genome-wide screen for dosage suppressors
We transformed 108 isogenic yeast strains each containing a temperature sensitive (ts) point mutation (85 mutants) or a ts deletion mutation (23 mutants)32 with pools of the entire MORF (Movable Open Reading Frame) library plasmids (Materials and Methods, Supplementary Table S1). The use of this library permits interrogating every conditional mutant with virtually every yeast ORF, under conditions where each ORF is expressed in the presence of galactose or glucose (Fig. 1b-d). We transformed each mutant strain with high copy 2μ based plasmids expressing open reading frames (ORFs) under pGAL promoter control33, selected transformants at the permissive temperature (25°C), then tested for the ability of the recombinant plasmids to rescue growth defect above the specific restrictive temperature for the corresponding mutant strain (see Methods). Each candidate suppressor ORF was individually confirmed by multiple retransformation experiments, and a subset by sequencing. In this manner we obtained 660 confirmed dosage suppressor interactions for 53 of the 108 ts mutants we tested (Fig. 1c, Supplementary Table S2), involving 517 suppressor genes, all of which were individually confirmed through retransformation and repeats of the assays, with all suppressors of a third of the mutant strains collection having been independently confirmed twice at two different sites (Rochester and KGI). 642 out of 660 interactions are novel; 18 were reported earlier. We did not detect 147 interactions previously reported in the literature corresponding to the 53 query mutants (Supplementary Table S3. We directly tested 57 of these previously reported suppressor interactions (corresponding to seven query mutant genes), chosen arbitrarily from among the 147 interactions that we did not discover, and were able to confirm by our methods only 15/57 interactions (Supplementary Table S4). These observations suggest that strain background and allele differences, copy number of the plasmids used for suppression, and/or the levels of galactose-induced gene expression obtainable by these various methods are sufficiently different such that a direct comparison is not possible. Therefore each case of validated suppression should be considered as a suppression event that is true under at least one set of conditions.
General properties of the dosage-suppressor network
The dosage suppressor network discovered here (dataset DS-A), containing 53 ts mutants and 517 suppressor genes, exhibits a large connected component (560 nodes, 656 edges) (Supplementary Fig. S2) that excludes 29 nodes and 20 edges. Of the 517 suppressor genes, 134 are of unknown function at the time of writing. Previous work29 had examined the entire collection of dosage suppressors of essential gene mutants known at the time (dataset DS-B, in BioGrid), including 214 suppressor genes they discovered for 29 mutants (dataset DS-C). All three network datasets together (DS-ABC) comprise 2,286 dosage-suppressor interactions (Fig. 2; Cytoscape network output file: http://tinyurl.com/kmsb6nm).
We examined the functional similarity of the genes discovered through this study with those in previous studies in the following way. We mapped the genes of each of the three networks on to a curated and integrated protein-protein interaction (PPI) network18,34,35,36,37 (see Methods), and calculated topological properties of the corresponding protein nodes in this PPI network (Supplementary Table S5). DS-A comprising 660 interactions overlapped significantly with the PPI network (P=2.8 × 10−15, see Tables 1, S5, and S6). This overlap, though significant, is small: 3.4% (23/660) of the dosage suppressor interactions in this work are edges in the PPI network compared with 19.5% (445/2,286, including 45 reciprocals) for DS-ABC (Fisher’s exact P < 0.0001). The previous dosage-suppressor collection (DS-A and DS-B) also showed a statistically significant overlap with the curated PPI. The degree and clustering coefficient of the dosage suppressor gene nodes in the PPI network are not significantly different among the three DS datasets. By contrast, the values of another topological property, betweenness centrality (BC; see Methods for definition) of the nodes in DS-A or DS-C are significantly higher than the nodes in DS-B (Supplementary Table S5). Moreover, the mutant-suppressor gene pairs are not statistically enriched for co-expression or for genes with similar evolutionary age (Table 1), again suggesting functionally different mechanisms than stabilization by PPI, for which the interacting protein pairs tend to possess correlated gene expression patterns38 and similar evolutionary ages39. A statistically significant overlap with PPI network implies that the mutant-suppressor pairs may belong to the same protein modules or complexes (below, we explicitly examine this question). However, the relatively small, though statistically significant, overlap of DS-A with the PPI network implies that DS-A is qualitatively different from DS-B or DS-C. Since a common mechanism of dosage suppression in the previously described suppressor networks was rescue of protein function through protein-protein contact29, the present results open the possibility that a significant proportion of dosage suppressors discovered in this work might suppress through other mechanisms.
Modular organization in dosage-suppressor network
Biological networks have underlying modular sub-structures that reflect functional organization and evolutionary origins of gene products38,40. Because there is no unambiguous definition of a module within protein networks, we here examine three concepts of modularity that are intrinsically different. First, we examined protein complexes that are manually curated clusters obtained from physical protein-protein interaction data41 (See Methods for the integrated PPI dataset), and determined the overlap of the 660 mutant-suppressor pairs with these complexes31,41. The products of 54 pairs of the 660 dosage-suppressor pairs were found within the same protein interaction complexes (binomial P < 10−15) (Table 1). When the entire collection of 2,286 dosage-suppressor pairs known so far (DS-ABC) was so examined, 558 pairs co-occur in the same protein complexes (binomial P < 10−15). Second, modules were obtained by computationally optimizing a modularity measure on the PPI network42, and similar associations were observed within the computationally predicted PPI clusters (Table 1). Third, we identified modules dynamically, by sequentially removing genes from a curated PPI network constructed from the dosage suppressor pairs (see Methods), starting with the highest BC gene, and re-computing the BC of nodes in each resulting network, leading to a measure of modularity43. Gene pairs that predominantly lie within modules should remain connected in the PPI network longer than the average pair, while gene pairs that straddle modules should separate earlier. Throughout this iterative process the mutant-suppressor pairs in DS-A were more likely to be found within module boundaries (Wilcoxon rank-sum P = 2.26 × 10−14; Supplementary Fig. S4) than were the randomly chosen protein pairs, even more so than the pairs in DS-C29 (P = 1.55 × 10−13) or those DS-B (P = 2.14× 10−10). That the mutant-suppressor pairs lie preferentially within module boundaries is consistent with the observed distribution of mutant-suppressor distances within the PPI network (number of edges along the shortest path in the PPI network connecting the mutant to the suppressor; see inset in Supplementary Fig. S4).
To determine the identity of the complexes within which the mutant-suppressor pairs co-occur, we culled from BioGRID database 4,632 direct PPIs, and used Netcarto module clustering algorithm42 to generate 41 modules (Supplementary Table S7). We queried each mutant-suppressor pair discovered in DS-A for their co-occurrence in these 41 modules, and found eight such modules containing five clusters and three singletons (Fig. 3a-e). In one module (Fig. 3a), mutations in cell cycle control genes cdc28 (a CDK), cdc20 and cdc16 (both anaphase promoting complex protein genes), and cdc37 (encodes an HSP90 co-chaperone), are suppressed by several genes including a ribosomal protein gene MRPL50, and MPD1 that encodes an endoplasmic reticulum chaperone interacting protein, underscoring the importance of molecular chaperones and ribosomal proteins in facilitating the suppression of point mutant alleles. In a second module (Fig. 3b), mutations in cdc9 (DNA ligase), tfb3 (transcription coupled DNA repair), and pob3 (encodes a member of the FACT complex for nucleosome reorganization) are in the same module with several suppressors including ADE2 (purine biosynthesis), PSY3 (DNA repair-recombination), and SSL2 (DNA repair helicase). While it is surprising that an adenine biosynthesis gene ADE2 is a suppressor of cdc9, Ade2p has predicted protein-protein interactions with Cdc9p, Pob3p, Tfb3p, and Ssl2p, suggesting that the suppression of cdc9 by ADE2 likely has a biological basis through the DNA repair-recombination pathway. When this module is expanded by querying the yeast integrated genetic and physical interaction network44 with members of this module, the network output provides further indirect evidence for linking the purine biosynthesis related genes to DNA repair-recombination (Supplementary Fig. S3). These results demonstrate the utility of using modules instructed by the dosage-suppressor network for novel gene-function discovery.
Functional similarity within mutant-suppressor pairs
Dosage suppressor genes and their corresponding suppressed mutant genes are functionally related. 120 out of 660 mutant-suppressor pairs in DS-A shared the same gold standard gene ontology (GO)45 terms (binomial P < 10−15), and 922 of 2286 known mutant-suppressor pairs (in DS-ABC) shared the same (binomial P < 10−15) (Supplementary Table S8). The mutant-suppressor pairs have similar GO molecular functions (P = 6.02 × 10−4 for the 660 pairs in DS-A, and P < 10−15 for DS-ABC) and similar GO biological process (Wilcoxon rank-sum P = 1.88 × 10−12 for DS-A, and P < 10−15 for DS-ABC).
To test whether the dosage-suppressor pairs are enriched for other types of genetic interactions, we intersected these pairs with known genetic interaction pairs reported in the literature46, and found significant enrichment (P = 5.34 × 10−8) for 11 kinds of genetic interactions considered by Costanzo et al46 (Supplementary Table S9). For 30 of the 660 gene pairs in DS-A (∼4.5%), a previously reported genetic interaction was found to exist, including 18 pairs of dosage-suppressor interactions, one each of positive genetic interaction and phenotypic suppression, seven negative genetic interactions, 11 synthetic lethal/rescue or growth defect interactions, and three phenotypic enhancements. A large fraction (547/660) was not found to overlap with known physical or genetic interaction pairs, thus reinforcing the previous conclusion that dosage-suppressor data provide additional information on biological function of genes that are not provided by other types of genetic interaction data29.
Co-suppressors are functionally diverse
We analyzed whether co-suppressors (suppressors of the same ts mutant) are functionally related to each other. First, suppressors in DS-A are not significantly enriched (Fisher’s exact P = 0.65) for paralogs: only 93 out of 517 genes in the suppressor network have at least one known paralog, as determined by the curated list of “ohnologs” (paralogs descended from whole genome duplication events)47. Among the 93 paralogs, we found 20 paralogous partners within mutant-suppressor sets (Table 2). We determined functional similarity between proteins by comparing their MIPS (Munich Information Center for Protein Sequences, http://www.helmholtz-muenchen.de/en/mips/services/index.html) functional catalog annotations48. Functional similarity, defined by the functional congruence (see Methods) of a ts-mutant gene with its suppressors, is significantly lower than that which would be expected for proteins having a direct physical interaction (Wilconxon rank-sum P = 2.26 × 10−14) (Supplementary Fig. S5A). As a comparison, the functional congruence between a ts-mutant and its suppressors is also lower for DS-C network than that for DS-B (BioGrid), whereas that of DS-B is comparable to that in the curated PPI network (Supplementary Fig. S5A). These observations suggest that DS-A and DS-C reveal suppressors that are qualitatively different from those revealed by the focused methods used by earlier workers, which predominate in DS-B. Moreover, the functional congruence among co-suppressors of the same mutant in DS-A is comparable with the congruence between proteins that share a physical edge in the curated PPI network (Supplementary Fig. S5B). However, the co-suppressors of the same mutant for DS-A are considerably more diverse than those in DS-B or DS-C (Supplementary Fig. S5B), once again demonstrating a distinct collection of dosage-suppressors discovered here. We schematically illustrate this diversity in Supplementary Fig. S6.
Plasticity and robustness of RNA polymerase II complex
To test how often genes whose products are known to function within the same macromolecular complex can suppress mutations that affect products within the same or related complexes, we chose as a test bed a well studied protein machine—the RNA polymerase II (RNA Pol II) complex49. RNA Pol II core (12 subunits) is recruited to the promoter by the general transcription factors TBP, TFIIA and TFIIB (1 subunit each), and TFIIF (3 subunits), with the help of the SRB/mediator complex (25 subunits), to form the pre-initiation complex, following which TFIIE (2 subunits) and TFIIH (9 subunits) are recruited50. The SWI/SNF (11 subunits) and SAGA (22 subunits) complexes facilitate chromatin remodeling during transcription initiation49. Extensive genetic interaction studies among mediator complex proteins have been reported51. We scanned eight mutant genes, each of which encoded a defective (or had a complete loss of one) transcription initiation complex protein52,53, for gene dosage suppression by 75 sub-complex genes (Supplementary Table S10; Supplementary Fig. S7). Six mutants were temperature sensitive (ts) lethal due to missense mutation (med4, med11, tfb3, rad3, kin28, taf12); two were ts deletion mutants (rpb4-Δ, and taf14-Δ). The deletion mutants had complete deletion of the respective structural genes, such that there was no possibility of expression of any remnant protein fragment32,54.
31 out of 122 dosage suppressor-mutant interactions in DS-A reside within the respective complexes. For example, MED11, NUT1, GAL11, ROX3, SRB5, and SRB7 each suppressed med4 (Fig. 4a). By contrast, nearly three times as many mutant-suppressor interaction pairs (91/122 compared to 31/122; P = 0.0001 by Fisher’s exact test) overlapped two separate sub-complexes (e.g., suppression of med4 by TFB3 of TFIIH complex and RPC10 of Pol II core complex, respectively). Suppressor interactions were specific: e.g., TFA1 (TFIIE complex) suppressed med4 but not med11—both of the latter encode mediator proteins. Similarly, SNF11 and SNF12 of the SWI/SNF complex each suppressed tfb3 but not kin28 (TFIIH); RPO21 (Pol II core) suppressed both tfb3 and kin28.
We discovered 36 suppressors of the eight RNA Pol II gene mutations, which are not known to be members of the RNA Pol II complex genes (Supplementary Table S2). The most striking example involves taf14-Δ (Fig. 4b), which yielded 27 suppressors, most of them encoding proteins outside the RNA Pol II complex, 12 of which are genes of unknown function. While the co-suppressor network of taf14-Δ was not enriched in PPI, a number of negative and positive epistatic edges connect several co-suppressors, including several genes of unknown function, to RNA Pol II complex genes. These results demonstrate the remarkable ability of this essential protein machine to function despite drastic genetic perturbations.
Suppressors of a chromosome condensation defect illustrate molecular rewiring
The genome wide suppressor dataset allowed us to explore suppression mechanisms quite different from that of the RNA Polymerase II findings discussed above: we provide examples wherein increased expression of genes allowed the bypass of an essential gene function by engaging alternate genetic pathways. Results described below show that the suppression of smc2 mutation by at least two different suppressor genes appear to proceed through this general mechanism, including the engaging of proteins important for or control of meiosis to an otherwise mitotic cellular division cycle.
Smc2p is a DNA-binding subunit of the Smc2p/Smc4p condensin complex55,56,57 required for sister chromatid (SC) alignment, separation, and inhibiting SC recombination during mitosis55,58. We identified four strong dosage suppressors of smc2: UME1, MEK1, HTA2, and SNU66 (Fig. 5a). Strikingly, the first two are known to play mutually opposing functional roles59,60,61—UME1 is a mitotically expressed gene required for the repression of a subset of meiotic genes, including those important for meiotic homologous recombination, and MEK1 is a meiosis specific protein kinase that promotes meiotic homologous recombination by suppressing sister chromatid (SC) recombination. To provide more insights into the mechanisms of suppression, we analyzed the time course of mRNA expression by these four suppressed smc2-8 mutant strains at the restrictive temperature and compared their global gene expression patterns with that of the mutant complemented by pGAL:SMC2 (See Methods).
Results (Fig. 5b-c) show that all four suppressors partially induce the expression of some meiosis-related genes. This led us to propose and test a simple model of smc2 dosage suppression mechanism, in which meiosis specific genes rescue smc2 defects in mitosis (Fig. 5d-e): smc2 mutation, which causes a failure of chromosome condensation in mitosis, allows the initiation but not the resolution of SC recombination, thus blocking mitosis55. Ectopic expression of meiosis-specific genes in the suppressed strains either prevents precocious SC recombination or resolves the SC recombination intermediates allowing mitotic division to progress. This model is a particular instance of the general mechanism shown in Fig. 1a by genes f and g. Specifically, a non-essential meiotic gene module controlling recombination replaces the essential mitotic gene module controlling chromosome condensation. We tested an implication of this hypothesis by introducing 37 MORF clones (Supplementary Table S12) corresponding to two categories of genes into smc2-8 and assaying their ability to suppress the ts growth defect: (a) other SMC complex genes, and (b) several meiotic recombination-promoting genes. 29/37 genes so tested suppressed smc2-8 (examples shown in Fig. 5d). Among those that suppressed smc2-8, were RAD51, DMC1, and MND1—all three are recombination-promoting genes, although RAD51 was a weak suppressor, and all suppressions were galactose independent. Several intron-containing meiosis-specific recombination genes, including DMC1 and MND1, are expressed and spliced to the mature form at low levels in mitosis but are expressed highly and spliced efficiently in meiosis62. The function of DMC1 is to promote recombination between homologous chromosomes and also to inhibit SC exchange in meiosis63. Dmc1p participates in a cascade of reactions activated through phosphorylation by Mek1p60, which we have found also to be a suppressor of smc2. These results support the idea that at least one mechanism of suppression of smc2 is through mitotic expression of meiotic recombination genes, which are expected to prevent the formation of or to promote the resolution of sister-chromatid junctions that occur at high frequency in smc2 mitosis.
A link between chromosome condensation-cohesion pathways and Huntington’s disease
A query of combined PPI and genetic interaction databases with six genes—smc2 and its five suppressors UME1, BNA5, SCC4, SMC1, and SMC3—generated an interaction network of 26 genes (including SCC2) containing three functional modules (enrichment significance, all Benjamini-Hochberg (B-H) corrected: chromatin remodeling complex, P = 2.37 × 10−5; sister chromatid cohesion complex, P = 1.38 × 10−7; and chromosome condensation, P = 6.4 × 10−13) (Fig. 6a). The members of each module exhibit extensive PPI and genetic interactions with members of the other two modules. We noticed that SCC2 encodes a member of the HEAT repeat proteins, which include Huntingtin (Htt)—a protein with expanded poly-glutamine residues in the N-terminal region that polymerizes to an insoluble aggregate (plaque) in certain brain cells of Huntington’s disease patients64. Moreover, BNA5 is a suppressor that linked two functional modules (Fig. 6a). A previous study65 had identified 28 deletion mutations that suppressed Htt toxicity in yeast, which included ume1Δ and bna4Δ, both of which have mammalian homologs. The transcriptional repressor Ume1p requires for its activity Rpd3p59, a part of the histone deacetylase (HDAC) complex. The inhibition of Rpd3 HDAC complex by Ume1p increases the toxicity of mutant human Htt in yeast, where Htt toxicity was related also to the function of BNA4 and BNA566. The latter two genes encode two successive enzymes in the NAD biosynthetic pathway from kynurenine, and the BNA4 product is a target for therapy against Huntington’s disease67,68,69. These results are consistent with the hypothesis that Htt toxicity is related to the chromosome condensation-cohesion processes, and might be influenced by homologous recombination pathways. We tested one prediction of this hypothesis by attempting to suppress Htt toxicity by SMC2, suppressors of smc2-8, and cohesin/condensin genes (Fig. 6b): 9 of 14 genes so tested suppressed Htt toxicity (SMC1, SMC3, SMC4, SCC4, MEK1, UME1, DMC1, HTA2, and MND1 suppressed; SMC2, SCC1, SCC3, CWC24 and RAD51 did not).
Discussion
A genome wide network of dosage suppressors allowed us to explore functional relatedness among the co-suppressors. Mapping of the mutant-suppressor pairs on a PPI network revealed boundaries of topologically defined protein modules. By examining specific protein modules through suppressors of selected RNA Pol II gene mutants, we found that high expression of specific component proteins within large protein assemblies can functionally replace the absence or the reduction of specific essential components. These latter results underscore the importance of including systematic dosage suppression data in deriving systems-level models of large protein complexes and their pathways of self-assembly.
In principle, dosage-suppressor interactions may involve high affinity and high probability protein-protein interactions that enable the system to return to the original state by positive and/or negative feedback effects (buffering interactions). In one scenario, over-produced suppressor products stabilize the corresponding thermo-sensitive missense mutant protein by direct interaction, through recruiting functionally competent folding intermediates from a distribution of folded states. By contrast, suppressors of deletion mutations, such as those of rpb4-Δ and taf14-Δ, in which the entire coding frames were deleted, cannot possibly exert their effects by stabilization through direct protein-protein interaction. Surprisingly, deletion ts alleles had a similar average frequency of suppressors (121 interactions with 10 mutants) as did missense ts alleles (539 interactions with 43 mutants). This suggests that heat sensitive biological processes (for ts deletion mutations) are as suppressible as processes catalyzed by individual heat-sensitive proteins (for ts missense mutations). Thus, the suppression mechanisms of rpb4-Δ by RPB3 (RNA Pol II core), and by ROX3 (mediator) likely involve direct or indirect functional replacement of Rpb4p. One mechanism might be the stabilization of the RNA Pol II preinitiation complex by the over-expression of one of the component proteins, such that the preinitiation complex rendered heat labile by rpb4-Δ is made more robust by an overabundance of Rpb3p or Rox3p. A similar argument holds for the suppression of taf14-Δ by TAF2, TAF13, and TAF10 (all TFIID), TFB3 (TFIIH), SRB7, PGD1, CSE2 (all mediator/SRB), SPT3 (SAGA), and by SNF5 and SNF6 (both SWI/SNF). Such alternate functional replacements within and between protein complexes reflect a high degree of compositional plasticity, and might also imply alternate pathways of assembly of multi-protein complexes. These observations are generally consistent with the recent observation that RNA Pol II open complexes can be reactivated by the TFIIE complex through stabilizing effects on relatively unstructured domains on mediator proteins70. Such “Intrinsically Disordered Regions” serve to functionally assemble RNA pol II complex subunit proteins54,71 and thus provide a high degree of modular functionality.
Although 45 of 53 suppressed mutant alleles are missense mutations, and several suppressors encode protein folding or processing enzymes such as chaperones or heat shock proteins (HSC82, HSP32, and CCT6) or chaperone interacting protein (MPD1), there is no significant statistical enrichment for these classes of proteins in the dataset produced in this work. By contrast, there is a significant enrichment for ribosomal proteins in our dataset (B-H corrected P = 2.64 × 10−4) as in the full set of known dosage suppressors (B-H corrected P = 4.5 × 10−6). The suppression of cdc37 (a co-chaperone mutant) by RPS18A, RPL25, and RPS24B is consistent with the possibility that some ribosomal proteins may have weak chaperone-like activity72.
Some suppressors of smc2-8 appear to act through direct protein-protein interaction with the mutant protein. For example, SMC1 and SMC3, required for SC cohesion, but not the condensin gene SMC4, can suppress smc2-8 (Fig. 5d). Protein-protein interaction between the mutant Smc2-8p and Smc1p/Smc3p cohesin complex might be able to stabilize misfolded Smc2-8p, whereas direct interaction between Smc2-8p and Smc4p, both members of the condensing complex, cannot do so. A recent report indicates that Smc2p homolog from S. pombe interacts with the S. pombe histone H2A and H2A.Z proteins in recruiting the condensin proteins to mitotic chromosomes73. Since S. cerevisiae HTA2 encodes a homolog of S. pombe H2A gene family, it is possible that Hta2p also recruits Smc2p to the nucleosome in an analogous manner. If true, at least a part of the suppression mechanism by HTA2 might be explained by the stabilization of mutant Smc2p through direct protein-protein interaction with Hta2p. Note that snu66 and hta2 mutations genetically interact46, suggesting that the mechanisms of suppression of smc2-8 by the two corresponding wild type genes might indeed be mutually related. Snu66p interacts with Snu40p46,74, and Snu40p in its turn interacts with proteins involved in chromosome condensation and cohesion, including Scc3p75. Scc3p is part of the Smc1p-Smc3p-Scc1p-Scc3p cohesin complex, which interacts with the Smc2p-Smc4p condensin complex, and the latter is stabilized by Scc2p-Scc4p complex. Therefore, one mechanism might be the stabilization and recruitment of thermosensitive Smc2-8p subunits into the condensin complex by direct protein-protein interaction with a cohesin complex protein that includes Scc4p. The suppression of smc2-8 by SCC4 is consistent with this mechanism.
Dosage suppression by rewiring, in contrast to that by direct PPI, may involve low affinity and/or low probability interactions that illustrate alternative—redundant or degenerate—pathways. These pathways of suppression appear to decouple physical interaction modules from the modules of functional activities, and the flexible interaction edges rearrange the functional modules to buffer the effects of genetic and environmental perturbations. We presented evidence that a mitotic chromosome condensation defect can be bypassed by ectopic expression of a series of meiotic genes involved in controlling sister chromatid recombination and resolution of recombination intermediates, suggesting that a defective biological process can be bypassed by alternate biological processes that are not engaged in cells during normal function. We proposed and tested a model in which the mitotic blockage by the smc2 mutation could be bypassed by UME1, MEK1, HTA2, and SNU66, by either preventing SC recombination initiation or augmenting SC recombination resolution; the predicted model was tested by further findings that several meiotic recombination genes DMC1, MND1, and RAD51 also suppressed smc2. Since SNU66, a strong suppressor of smc2, encodes a splicing component, it is possible that the over-expression of Snu66p augments the splicing of DMC1 and MND1 pre-mRNAs in mitotic cells.
Suppressor interactions that connect BNA4/5 to UME1, SCC2, and SMC2, point to a possible mode of action of the mutant Huntingtin protein involving the chromosome condensation/cohesion process. As a test of this hypothesis, we examined the ability of a set of suppressors of smc2-8 mutation (including SMC2) to also suppress 103Q toxicity, and discovered that several cohesin/condensin genes, and a few homologous recombination genes including MND1, are able to do so. Interference by Htt of the normal sister-chromatid cohesion process could potentially increase somatic recombination at CAG repeats, and thus might underlie the observed variable penetrance of high poly-glutamine expansion alleles on disease onset age in humans76,77. Consistent with this speculation, a recent report has linked a network of ploidy control genes in yeast to Htt toxicity78.
An unresolved question remains as to the extent of genetic robustness that is hardwired in the genome, which was first brought to the fore by the investigation of mutational effects on the lysis-lysogny decision circuit of bacteriophage lambda79. In this work, mechanisms of suppression of a defect in chromosome condensation revealed insights on the potential of unrelated genes that could be brought to bear on solving problems associated with defective cellular processes. It is conceivable that yeasts in nature, and organisms in general, depend on the rewiring of gene regulatory circuits to find new solutions to essential cellular processes during evolution under selective pressure, as observed in this work that meiotic genes relieve mitosis blockage. Such a possibility was investigated earlier in evolved yeast strains with a deletion in an essential gene (myo1), where aneuploidy and large-scale variation in the transcriptome were associated with survival27. While aneuploidy was a recurrent theme observed in that work, it was also estimated that the number of available genetic solutions to a lethal perturbation might be limited. Gene redundancy80, and promiscuity of gene function30,81 both contribute to genetic robustness. While the core set of essential genes might impede evolvability82, modular rewiring may in principle overcome this barrier83. Our finding that nearly six times as many genes can suppress 53 deleterious mutations indicates a high degree of robustness built into the genome, and illustrates potential pathways for rewiring of the genome. It is conceivable that a deleterious mutation in an essential gene, leading to growth arrest, is followed by genomic changes that are often observed in stationary phase cells27,84,85,86,87; these changes could in principle activate suppressor pathways to restore viability and provide adaptation.
The network of dosage suppressors of essential gene mutants is analogous to the network of genes that could potentially bypass, if aberrantly expressed, a drug target gene function (e.g., of a cancer-essential gene) for tumor cell survival. Such a network for a cancer cell is the equivalent of potential pathways for developing resistance to cancer chemotherapy, or, analogously, for evolving independence from the checkpoints that ensure non-proliferative growth, which evidently occurs frequently during the development of cancers.
Materials and methods
MORF plasmids
The movable open reading frame (MORF) library33 containing 5,871 2μ plasmids with galactose inducible promoter and a URA3 selectable marker were divided into 16 pools. Each pool, representing approximately 384 plasmids, was grown in 96 deep-well plates, pooled, and plasmid DNA samples were isolated for transformation.
Yeast strains, media and transformation
Temperature sensitive lethal Saccharomyces cerevisiae strains had specific mutations in BY4741 background (MATa his3-Δ1 leu2-Δ0 met15-Δ0 ura3-Δ0); point mutants were provided by Dr. Charlie Boone (University of Toronto) and ts deletion mutants were screened and selected from the deletion mutant library (OpenBiosystem). For each mutant, the range of growth and the threshold of non-permissive temperature on both inducing (+ galactose) and non-inducing (-galactose) conditions were determined. Yeast strains were grown in yeast complete media containing 1% raffinose, transformed with 1μg of each MORF plasmid pool, and plated at permissive temperature on synthetic defined medium lacking uracil with 1% raffinose. The transformants from 16 plates were pooled and selected at the restrictive temperature for that particular ts allele on complete and synthetic media containing either 2% glucose (repression) or 2% galactose (induction) (see Supplementary Table S1) for a list of restrictive temperatures corresponding to the ts alleles). The threshold restrictive temperature that cuts off the growth of each individual allele with or without (vector control) the candidate suppressor plasmid was determined for each suppressor by incubating identical multiple-replicate plates at a range of temperatures spanning at least ±2°C around the restrictive temperature for that allele. Transformants in each mutant strain were selected for growth above the respective restrictive temperature characteristic for the corresponding mutant strain containing the empty vector plasmid. Strains for testing suppression of Htt toxicity were in W303 background88 (can1-100, his3-11,15::FLAGhtt103Q-CFP,leu2-3,112trp1-1,ura3-1, ade2-1; and can1-100, his3-11,15::FLAGhtt25Q-CFP, leu2-3,112trp1-1,ura3-1,ade2-1).
Suppressor identification and confirmation
Suppressor candidate hits were identified by microarray hybridization of isolated plasmid DNA from colonies growing above the restrictive temperature for the respective ts allele as follows: Approximately 300 colonies were picked from selection plates at restrictive temperature, grown in 96 deep-well plates. The cells were pooled and plasmid DNA isolated using Cycle Pure Kit from Omega Bio-Tek, and labeled with Cy3 dye, whereas the pooled DNA of the MORF library was labeled with Cy5 by PCR amplification using two flanking primers (5’GGACCTTGAAAAAGAACTTC3’, 5’CCTCTATACTTTAACGTCAAGG3’). Labeled probes were hybridized to spotted microarrays (UHN Microarray; containing > 95% of all ORFs) at 65 °C for 16 h. Microarrays were scanned in Bio-Rad VersArray Chip Reader and the data were analyzed using ScanArray express software 3.0 (Perkin Elmer). For a limited number of suppressors, we purified the DNA from colonies growing on selection plates with raffinose and galactose, transformed into E. coli, re-isolated the corresponding plasmid DNA samples, PCR amplified the ORF off the MORF plasmids, and sequenced the DNA. In all cases, the suppressor MORFs identified by sequencing were identical to the MORFs identified by microarray hybridization. The putative suppressor genes identified by either microarray or by direct sequencing were retransformed individually to the respective mutant strains, their ability to suppress the mutants confirmed individually on at least three separate transformed colonies by isolating single colonies and testing on inducing or non-inducing plates at a range of temperatures above the growth cutoff temperature of the corresponding unsuppressed mutant, dependence of suppression on the introduced MORF plasmid was confirmed for each transformant on plates containing 5-fluroorotic acid, and for those transformants passing all above tests the titration-spotting test was carried out for final confirmation/quantification of suppressor strength. The background strain was always compared on the same plate with the candidate-suppressed strains at a range of temperatures spanning at least 1°C over the corresponding threshold temperature for growth of the given mutant. The extent of suppression was subsequently quantified through spotting of serial dilutions of each mutant/suppressor culture under both inducing and noninducing conditions. A subset of the final list of suppressors was again confirmed by sequencing. The negative control for each suppressor was the corresponding mutant strain carrying the empty MORF vector (BG1776). Positive control plasmids were the complementing genes under pGAL control, except for six mutants (cdc13, cdc4, cdc15, cdc35, cdc48, and abd1) that did not have the appropriate positive controls because either the over-expression of the corresponding MORF plasmids was lethal (CDC13, and CDC48) or they were absent in the MORF library (CDC4, CDC15, CDC35, and ABD1). The strength of suppression for each suppressor was normalized to the growth of the diluted spots against that of the corresponding vector control (BG1776) strain on the same plate on adjacent rows (Supplementary Table S2).
Protein detection
Cultures were grown in repressing and inducing media, and whole cell extracts were prepared by the bead beating method in yeast lysis buffer (25 mM Hepes-NaOH pH 7.5, 10 mM NaCl, 1 mM EDTA, 0.1% Triton X-100) containing EDTA–free complete protease inhibitor tablet (Roche). Proteins were detected by Western Blotting, probed by anti-HA antibodies (Covance) using standard methods.
Liquid growth assay
Growth curves in liquid media along with maximal growth rates were determined using a Bioscreen C Automated Growth Curves Analysis System (Growth Curves USA). The suppressed strains were grown in 200 μl of S-URA with 1% raffinose and 2% galactose at various temperatures in 96-well plates. The optical density (OD) was measured at 600 nm every 30 minutes for 48 hours of growth.
RNA methods
Strains were grown in 5 mL S-URA media containing 2% raffinose for 24 hours at 28°C. Samples were diluted to OD 0.02 in 150 mL S-URA media containing 2% raffinose and 2% galactose, and then grown overnight at 25°C with agitation. At OD 0.1, cultures were shifted to 34.5°C with continuous agitation, and samples were harvested at 0, 45, 90 and 180 min by centrifugation. Total RNA was isolated using a hot phenol method89, followed by 2 chloroform extractions, RNA precipitated by addition of 1/10th volume of NaOAc pH 5.2, and 2.5 volumes of 100% ethanol and the pellet dissolved in 50μL ddH2O. Total RNA was deproteinized again after DNase treatment and resuspended in 10 μL of ddH2O, and quantified by a nanodrop spectrophotometer.
Gene expression measurements
1 μg total RNA samples from two individual biological replicates, four time points each (0, 45, 90 and 180 min), of smc2-8 mutant strains harboring pGAL:SMC2, pGAL:UME1, pGAL:MEK1, pGAL:HTA2, pGAL:SNU66, and the negative control MORF plasmid (pGAL:negative, BG1766), were reverse transcribed, hybridized to Affymetrix Yeast Genome S98 arrays and scanned with Affymetrix GeneChip Scanner 3000. The microarray data were analyzed with GeneSpring 6.2 software, and were deposited in Gene Expression Omnibus (accession number GSE24266). Microarray expression levels were verified for 7 reference genes by quantitative RT-PCR using samples of a third biological replicate (Supplementary Fig. S8).
Microarray Data Analysis
The Affymetrix array data were processed using the Robust Multi-array Analysis (RMA) as described previously90. A log scale, linear additive model represented the perfect match and mismatch data. For each experiment (time point or condition), the RMA analysis produced one numerical estimate of expression for every probe on the chip (two replicates for each treatment and time point). We combined the replicates using a median based normalization (given microarray replicates 1 and 2, determined the median intensities m1 and m21 of microarray 1 and microarray 2 respectively; adjusted the values of microarray 2 by adding m1 – m2 to the intensities of microarray 2) to produce an average of the adjusted intensities. For each time point and treatment we produced one intensity measurement for each probe. These numbers were used to find ratios of fold change from one time point to the next. For each chip, a background noise intensity measure was formed using the average intensity of the SPACER probes which act as a set of negative controls for the chip – if for a particular probe, the intensity level at time points t1 and t2 was below the background noise level at t1 and t2, then we assumed the probe was expressing at background noise level and the fold change was set to 1. We clustered genes that were significantly differentially expressed in at least one time point.
Quantitative RT-PCR
DNase treated RNA was reverse transcribed in 25 µL RT reaction mixtures (1x First Strand Buffer, 0.02 μg random hexamers, 0.01 M DTT, 0.5 mM dNTP mix, 0.6 U RNaseOUT, containing 3 U Superscript II Reverse Transcriptase) for 2 h at 37 °C, followed by heat inactivation at 100°C for 5 min and quick chilling on ice. A standard curve was generated for each gene target starting with 0.5 μg RNA, and four successive 2-fold serial dilutions. The cDNA templates generated by reverse transcription was used for quantitative RT-PCR in Applied Biosystems 7500 FAST Real Time PCR system. The PCR mix constitutes 20 µL buffer containing 1X FAST SYBR Green Master Mix and 0.2 μM forward and reverse primers (IDT), with the PCR conditions: [95°C → 20 sec] HOLD, [95°C 3 sec → 60°C 30 sec] 40 times). The following primers were used for real time PCR of seven genes:
Datasets
The full Dosage Suppressor (DS) dataset (DS-ABC, see text), consisting of data reported here combined with other available data, was culled from the following sources: Magtanong et al.29, the latest BioGRID version 3.1.7831, and this work (660 dosage suppressor interactions between 53 mutant ORFs and 517 suppressor ORFs). The full DS dataset contains 2286 interactions between 454 mutant ORFS and 1284 suppressor ORFS, and includes 60 reciprocal interactions, resulting in 2226 unique interactions. The PPI network was constructed from BioGRID version 3.1.7831 that contains 6,614 ORFs of which 5,955 nodes are connected by physical interaction edges. The 99,866 physical interactions produced 60,143 unique edges among 5,728 nodes. This network was curated to filter out indirect physical interactions and retain only the direct physical interaction data from eight types of experiments including Biochemical activity, co-crystal structures and reconstituted complexes, PCA, protein-peptide interactions, two-hybrid, far western and FRET. The resulting direct PPI network, containing a total of 20,034 unique interactions between 4683 nodes, was used for studying the enrichment of PPI interactions in the dosage-suppressor (DS) network (Supplementary Table S6). Genetic interaction (GI) data was downloaded using the latest BioGRID version 3.1.7831, which contains eleven kinds of genetic interactions, namely, Dosage growth defect, Lethality, Dosage Rescue, Negative Genetic, Phenotypic Enhancement, Phenotypic Suppression, Positive Genetic, Synthetic Growth Defect, Synthetic Haplo-insufficiency, Synthetic Lethality, and Synthetic Rescue. In all, the GI dataset includes 16898 interactions between 5411 nodes. The Stanford Microarray Database was used for inferring co-expression between yeast ORFs at a correlation coefficient cut-off of ± 0.5. The resultant co-expression network includes 623,224 unique edges between 5155 nodes from total 643 experiments reported by two groups91,92. Genome-wide protein complex data was inferred by combining the Curated Yeast Complexes (CYC2008), a comprehensive catalogue of manually curated 408 heteromeric protein complexes in S. cerevisiae with 400 complexes in the Annotated Yeast High Throughput (YHTP) complexes derived from high-throughput Tandem Affinity Purification/Mass Spectrometry (TAP/MS) studies41, and 72,016 pairs of indirect physical interactions from Affinity Capture, Co-Fractionation, Co-purification and Co-localization experiments (BioGRID version 3.1.78). The Pfam domains for many yeast proteins have been assigned one of four age groups ABE, AE/BE, E and F depending on their taxonomic distribution among archaea (A), bacteria (B), eukaryote (E) and fungi (F)93. This dataset was used to analyze whether DS pairs were likely to belong to the same age group. Both the DS datasets were tested for significant overlap with computationally predicted modules. For this, 41 Yeast Louvain modules were identified in the Yeast direct PPI network by using the NetCarto algorithm42. In addition, Markovian clusters were identified using the MCL-MLR clustering method 94 at an inflation value of 2.4.
Paralog identification
The list of paralogs (554 gene pairs) was described before47, and includes 457 pairs previously found95. Of the 1,108 paralogous genes, 1,001 were represented in the MORF library. To test for significant enrichment of paralogs, we performed a Fisher’s exact test using the 5,829 testable ORFs as the baseline.
Network properties
Betweenness centrality (BC) (fraction of shortest paths through a node) was , where bi is the ratio of the number of shortest paths between a pair of nodes in the network that pass through node i. BC was scaled as , where n is the number of nodes in the network96. Clustering coefficient (ratio of the actual number of degrees of a node to the possible degrees given a node’s neighbors), and shortest path between pairs were computed using the MATLAB Boost Graph Library toolset.
Functional congruence assessment
Functional gene annotations were derived from the MIPS FunCat database48. 449 of the 642 unique suppressors were annotated genes. Un-annotated genes (class ‘99’) were excluded. Here, functional congruence between two genes is defined as the extent of overlap between their respective MIPS annotations. Because MIPS functional annotations are hierarchical categories, we studied the congruence of MIPS annotations based on the 1st category alone, 1st and 2nd, 1st, 2nd, and 3rd, etc. categories, down to congruence in all 5 categories. Because genes can have several MIPS annotations, we compiled the possible pairs of categories for each pair of genes and calculated the fraction of pairs of annotations that agreed, divided by the number of possible pairs of annotations. Because many annotations do not include details up to the fifth functional categorization, we devised rules to match annotation strings of different length. For example, when matching the functional MIPS annotation “01.01.05.01.02” (degradation of polyamines) with “01.01.06” (metabolism of the aspartate family), we pad annotations such that the previous example would result in a match at level 1, a match at level 2, but not at level 3 and beyond. In other words, for incomplete annotation the omitted part is assumed to be different from that of any other annotations. The functional congruence of two genes at level n is the fraction of annotations of these genes that are identical up to level n. By design, the functional congruence of a pair at level n is larger or equal to the functional congruence of that pair at level n + 1.
Acknowledgements
We thank Dr. C. Boone (U. Toronto) for discussion, for sharing strains and their suppressor data before publication, Dr. Susan Lindquist (Whitehead Institute, Cambridge, MA) for providing strains with the human Htt genes, B. Kraynack and K. Svay for initial work in suppressor discovery, and C. Cordova, R. Vasudevan, V. Padmakumar, K. Bheda, K. Datta, A. Kamra, D. Lev, and M. Pollock for technical assistance, and Dr. C. Murre (UCSD) for comments on an earlier version of the manuscript. This work was supported by grants from the National Science Foundation (Frontiers in Integrative Biological Research #0527023 to D. J. G, E. J. G., E. M. P., C. A., A. Rav. and A. Ray; #0523643 to A. Rav. and A. Ray; #0941078 to A. Ray), the National Institutes of Health (# 1R01GM084881-01 to A. Ray) and CHDI Foundation (to A. Ray). The authors have no conflict of interest.
Contributions
B.P., Y.K., A.W.S., G.Y. J.F., R.V., A.H., B.Ø., A.B., B.M., and J.T. conducted experiments and analyzed data, A.K. conducted experiments; B.P., Y.K., A.W.S., J.F., R.V., A.H., B.Ø., A.B., B.M., N.S.B., E.J.G., D.J.G., G.Y., A.Rav., C.A., E.M.P, and A.Ray designed experiments and analyzed data; B.P., A.W.S., G.Y., J.F., A.H., B.Ø., A.B., D.J.G., A.Rav., C.A., and A.Ray wrote the manuscript; E.J.G., D.J.G., A.Rav., C.A., E.M.P., and A.Ray procured funding.
References
- 1.↵
- 2.↵
- 3.↵
- 4.↵
- 5.↵
- 6.↵
- 7.↵
- 8.↵
- 9.↵
- 10.↵
- 11.↵
- 12.↵
- 13.↵
- 14.↵
- 15.↵
- 16.↵
- 17.↵
- 18.↵
- 19.↵
- 20.↵
- 21.↵
- 22.↵
- 23.↵
- 24.↵
- 25.↵
- 26.↵
- 27.↵
- 28.↵
- 29.↵
- 30.↵
- 31.↵
- 32.↵
- 33.↵
- 34.↵
- 35.↵
- 36.↵
- 37.↵
- 38.↵
- 39.↵
- 40.↵
- 41.↵
- 42.↵
- 43.↵
- 44.↵
- 45.↵
- 46.↵
- 47.↵
- 48.↵
- 49.↵
- 50.↵
- 51.↵
- 52.↵
- 53.↵
- 54.↵
- 55.↵
- 56.↵
- 57.↵
- 58.↵
- 59.↵
- 60.↵
- 61.↵
- 62.↵
- 63.↵
- 64.↵
- 65.↵
- 66.↵
- 67.↵
- 68.↵
- 69.↵
- 70.↵
- 71.↵
- 72.↵
- 73.↵
- 74.↵
- 75.↵
- 76.↵
- 77.↵
- 78.↵
- 79.↵
- 80.↵
- 81.↵
- 82.↵
- 83.↵
- 84.↵
- 85.↵
- 86.↵
- 87.↵
- 88.↵
- 89.↵
- 90.↵
- 91.↵
- 92.↵
- 93.↵
- 94.↵
- 95.↵
- 96.↵
- 97.↵