Phenotypic and Transcriptional Plasticity Directed by a Yeast Mitogen-Activated Protein Kinase Network

Corresponding author: Mike Tyers, Room 1080, Mount Sinai Hospital, 600 University Ave., Toronto, ON M5G 1X5, Canada., tyers{at}mshri.on.ca (E-mail)

Communicating editor: M. JOHNSTON

	ABSTRACT

TOP ABSTRACT MATERIALS AND METHODS RESULTS DISCUSSION LITERATURE CITED

The yeast pheromone/filamentous growth MAPK pathway mediates both mating and invasive-growth responses. The interface between this MAPK module and the transcriptional machinery consists of a network of two MAPKs, Fus3 and Kss1; two regulators, Rst1 and Rst2 (a.k.a. Dig1 and Dig2); and two transcription factors, Ste12 and Tec1. Of 16 possible combinations of gene deletions in FUS3, KSS1, RST1, and RST2 in the 1278 background, 10 display constitutive invasive growth. Rst1 was the primary negative regulator of invasive growth, while other components either attenuated or enhanced invasive growth, depending on the genetic context. Despite activation of the invasive response by lesions at the same level in the MAPK pathway, transcriptional profiles of different invasive mutant combinations did not exhibit a unified program of gene expression. The distal MAPK regulatory network is thus capable of generating phenotypically similar invasive-growth states (an attractor) from different molecular architectures (trajectories) that can functionally compensate for one another. This systems-level robustness may also account for the observed diversity of signals that trigger invasive growth.

MITOGEN-activated protein kinase (MAPK) modules are ubiquitous signaling elements that typically link receptor-mediated events to regulation of gene expression. Signals are conveyed through sequential phosphorylation and activation of three kinases, from a proximal MAPK kinase kinase, to a MAPK kinase, and finally to a distal MAPK (reviewed in ROBINSON and COBB 1997 ; GUSTIN et al. 1998 ; CHANG and KARIN 2001 ). The three-tiered kinase module is often physically tethered together by scaffold proteins and may transmit signals in either a switch-like or a graded fashion (WHITMARSH and DAVIS 1998 ; BURACK and SHAW 2000 ; FERRELL 2000 ; HARRIS et al. 2001 ; PARK et al. 2003 ). Because the same MAPK module may be used to transmit different signals within the same cell, a critical issue is how MAPK activation yields specific biological responses (discussed in TAN and KIM 1999 ). This puzzle is epitomized by the mating pheromone/filamentous growth pathway of budding yeast, which appears to transmit signals for mating and invasive growth within the same haploid cell (Fig 1A; reviewed in HERSKOWITZ 1995 ; MADHANI and FINK 1998B ; BREITKREUTZ and TYERS 2002 ).

View larger version (87K): In this window In a new window Download PPT slide   Figure 1. Genetic analysis of the mating pheromone/filamentous growth MAPK cascade. (A) Two distinct developmental programs, mating and invasive growth, share multiple components of a single MAPK signaling cascade. (B) Differential interference-contrast images of combinatorial deletion strains in exponential phase in liquid rich medium at 30°.

The budding yeast Saccharomyces cerevisiae, as well as human and plant fungal pathogens, can adopt filamentous growth forms that correlate with virulence and may represent a foraging mechanism (reviewed in MADHANI and FINK 1998A ). S. cerevisiae filamentous growth usually entails prolonged polarized morphogenesis toward the tip of the bud, reorganization of budding pattern to a unipolar form, altered cell cycle control, and increased cell-cell adhesion, which together result in the generation of long branching filaments (reviewed in PAN et al. 2000 ; RUA et al. 2001 ; PALECEK et al. 2002B ). On solid media containing poor nitrogen sources, diploid cells undergo a dimorphic transition from oval cells to chains of elongated cells, a response termed pseudohyphal growth (BROWN and HOUGH 1965 ; GIMENO et al. 1992 ). On rich medium, haploid cells under mature colonies penetrate into subsurface agar in a response termed "invasive growth" (ROBERTS and FINK 1994 ). Haploid invasive growth and diploid pseudohyphal growth are collectively referred to as "filamentous growth."

Filamentous growth depends in part on two conserved signaling pathways, the mating/filamentous MAPK cascade and the Ras/cAMP pathway (GIMENO et al. 1992 ; MOSCH et al. 1996 ; ROBERTSON and FINK 1998 ; PAN and HEITMAN 1999 ). The filamentous response is also closely linked to the cell cycle machinery since inhibition of B-type cyclin-Cdc28 or hyperactivation of Cln1/2-Cdc28 causes a filamentous-like response (RUA et al. 2001 ). Ancillary factors including Ash1, Fkh1/2, Mep2, Sok2, Srb10, and Phd1 are implicated in filamentous growth but do not appear to fall within the MAPK or cAMP pathways (MOSCH and FINK 1997 ; PAN et al. 2000 ; PALECEK et al. 2002B ). All told, filamentous growth represents a complex biological output that is influenced to a greater or lesser degree by many parameters.

Components of the mating pheromone MAPK cascade, including the MAPKK Ste7, the MAPKKK Ste11, the PAK-like kinase Ste20, and the downstream transcription factor Ste12, are necessary for both mating and the filamentous growth response (LIU et al. 1993 ). Two related MAPKs, Fus3 and Kss1, lie at the bottom of the mating/filamentous growth pathway. Deletion of FUS3 causes a moderate decrease in mating efficiency and deletion of KSS1 has little or no effect, but elimination of both MAPKs results in sterility, suggesting an overlapping function for Kss1 and Fus3 in the mating response (ELION et al. 1991 ; BREITKREUTZ and TYERS 2002 ). Consistently, removal of either individual MAPK has little influence on the transcriptional response to pheromone, and both Fus3 and Kss1 are activated by pheromone in wild-type cells (BREITKREUTZ et al. 2001A ; CHERKASOVA and ELION 2001 ; SABBAGH et al. 2001 ). The dominant role of Fus3 in the mating response may be explained by its specificity for critical substrates such as Far1, which mediates G₁ arrest and polarization, while other substrates such as Ste12 are phosphorylated equivalently by either MAPK (PETER et al. 1993 ; BREITKREUTZ et al. 2001A ). In contrast, cells that lack Kss1 are defective in filamentous growth (BARDWELL et al. 1994 ; ROBERTS and FINK 1994 ). However, the role of Kss1 is complex as it also represses the filamentous response when in the inactive state (COOK et al. 1997 ; MADHANI et al. 1997 ; BARDWELL et al. 1998 ).

Ste12-dependent transcriptional activation is critical for both the pheromone and filamentous growth responses. In unstimulated cells, Ste12 is held in check by two physically associated regulatory factors, Dig1/Rst1 and Dig2/Rst2 (COOK et al. 1996 ; TEDFORD et al. 1997 ). An rst1 rst2 double deletion strain displays constitutive filamentous growth and mating pheromone responses in a manner that depends on STE12 but not on other components of the MAPK cascade, thereby placing RST1/2 function between the MAPK cascade and STE12 (TEDFORD et al. 1997 ). Like Ste12, both Rst1 and Rst2 are phosphorylated in vitro and in vivo by Fus3 and Kss1 (COOK et al. 1996 ; TEDFORD et al. 1997 ). Phosphorylation of Ste12 by the RNA-polymerase-II-associated kinase Srb10 is also essential for filamentous growth (NELSON et al. 2003 ). The genes necessary for filamentous growth are not well defined but depend in part on the transcription factors Ste12, Tec1, and Flo8 (PAN et al. 2000 ). Genome-wide expression profiles have identified a number of genes regulated by TEC1 (MADHANI et al. 1999 ), but the expression of this gene set under filamentous growth conditions has not been examined. Ste12 and Tec1 appear to act in close proximity in cis on the same promoter, as in the case of FLO11, which encodes a cell-surface flocculin (MADHANI and FINK 1997 ; PAN et al. 2000 ). In addition, Ste12 and Tec1 form a transcriptional cascade since TEC1 expression is induced by mating pheromone in a Ste12-dependent manner (OEHLEN and CROSS 1998 ; ROBERTS et al. 2000 ; BREITKREUTZ et al. 2001A ; KOHLER et al. 2002 ).

The upstream signals that initiate filamentous growth are also not well understood. In diploid cells, nitrogen limitation triggers pseudohyphal growth (BROWN and HOUGH 1965 ; GIMENO et al. 1992 ), while in haploid cells depletion of fermentable carbon sources or the presence of various alcohols causes invasive growth (ROBERTS and FINK 1994 ; CULLEN and SPRAGUE 2000 ; LORENZ et al. 2000 ). Exposure of haploid cells to pheromone also stimulates vigorous invasion, particularly in the absence of cycle arrest, as in a far1 strain (ROBERTS et al. 2000 ; ERDMAN and SNYDER 2001 ). This effect is Kss1 independent, implying that Fus3 is fully able to activate an invasive response. Mating pheromone-induced invasion occurs under wild-type physiological conditions since a-cells exhibit highly polarized divisions toward -cell mating partners when placed in close proximity (LEVI 1956 ). The many signals that trigger filamentous growth and the nonlinearity of the pathway complicate interpretation of mutant phenotypes and genetic interactions (BREITKREUTZ and TYERS 2002 ).

To analyze the means by which MAPK activity regulates invasive growth and transcription, we undertook a systematic genetic and genome-wide DNA microarray analysis of strains lacking all possible combinations of factors that directly interact with Ste12, namely Fus3, Kss1, Rst1, and Rst2. Because commonly used laboratory strain backgrounds, such as S288C and W303, are defective for invasive growth (LIU et al. 1996 ), previous genome-wide analyses in these backgrounds may not reveal true correlations between the invasive phenotype and the transcriptional profile (ROBERTS et al. 2000 ; BREITKREUTZ et al. 2001A ). To overcome this problem, we undertook our analysis in the 1278 strain background, which is fully permissive for invasive growth (GIMENO et al. 1992 ). Contrary to expectations, Fus3, Kss1, Rst1, and Rst2 do not control the invasive response in a simple additive manner nor do genome-wide expression profiles indicate a unified program of gene expression associated with invasive growth. These findings suggest plasticity in the control of the invasive response by the MAPK network and illustrate that complex phenotypes such as filamentous growth may emanate from functionally coupled regulators along quite distinct pathways.

	MATERIALS AND METHODS

TOP ABSTRACT MATERIALS AND METHODS RESULTS DISCUSSION LITERATURE CITED

Yeast strains and culture:
Standard methods used for yeast strain construction, determination of invasive growth, and microscopy were as described previously (BREITKREUTZ et al. 2001A ). All gene deletions were confirmed by Southern analysis (see online supplementary Table 1 at http://www.genetics.org/supplemental/).

DNA microarrays and analysis:
Sample workup, hybridization, and analysis were performed as described previously (BREITKREUTZ et al. 2001A , BREITKREUTZ et al. 2001B ). Spreadsheets containing raw data for all microarrays are available at http://www.mshri.on.ca/tyers/. Any genes shown in black either were below background or did not change expression >1.5-fold. Gene ontology and functional annotations were obtained from the Saccharomyces Genome Database (http://www.yeastgenome.org/; DWIGHT et al. 2002 ). Microarray data were analyzed using hierarchical clustering (EISEN et al. 1998 ) and binary tree-structured vector quantization (BTSVQ) programs (SULTAN et al. 2002 ). Genes are considered pheromone-regulated if induced/repressed >1.5-fold (452 genes total) in wild-type pheromone-treated cells (BREITKREUTZ et al. 2001A ).

Distance matrix is a method to visualize multidimensional data. Each hexagon on the distance matrix represents a d-dimensional data vector (called code vectors) selected by vector quantization (VQ). VQ is a method in BTSVQ used for lossless dimensionality reduction of the problem space. An additional hexagon is placed between existing hexagons to render Euclidean distance between the corresponding data vectors. The distance is rendered using a color scheme: dark blue represents minimum distance and dark brown represents maximum distance. The scale is changed to maximize the differential effect of the projection. Placement of individual vectors on the matrix depends on their local similarity. During the training process the self-organizing map (SOM) algorithm assigns best-matching vectors to the set of code vectors, and they are randomly distributed to the whole topology; thus, similar vectors can be placed at different parts of the map. The self-organizing part of the algorithm (neighborhood function) brings similar groups together in a manner that depends on three parameters: the size of the map, the radius of the neighborhood function, and the number of iterations. We thus used a Gaussian neighborhood function with medium radius. In the U-matrix, high-intensity colors (red) represent cluster boundaries and locally low-intensity colors (blue) represent similar data vectors.

To monitor how invasive and noninvasive strains cluster on the basis of gene expression, we used statistical filters to select the highly differential regulated genes. Two sets of genes were used for clustering using partitive K-means clustering and SOM. In one set, genes high in invasive and low in noninvasive strains were filtered by calculating the mean expression profile and selected if mean expression was higher than some threshold T1 in invasive strains and lower than some threshold T2 in noninvasive strains. Thresholds were selected using distribution profiles to select only a highly differential subset of genes (online supplementary Table 6 at http://www.genetics.org/supplemental/). In the current study, T1 = 0.75 and T2 = 0.75, which corresponds to 98.5 and 0.5% probability cutoffs. In a second set, genes high in noninvasive and low in invasive strains were filtered and selected if mean expression was higher than some threshold T3 in noninvasive strains and lower than some threshold T4 in invasive strains. Thresholds were selected using distribution profiles to select only a highly differential subset of genes (online supplementary Table 6 at http://www.genetics.org/supplemental/). In the current study, T3 = 0.33 and T4 = 0.75, which corresponds to 98.5 and 0.5% probability cutoffs.

	RESULTS

TOP ABSTRACT MATERIALS AND METHODS RESULTS DISCUSSION LITERATURE CITED

Combinatorial deletion analysis of the distal MAPK network:
To analyze how the MAPKs Fus3 and Kss1 and the regulators Rst1 and Rst2 control transcription (Fig 1A), we generated mutants lacking all possible combinations of these factors (see online supplementary Table 1 at http://www.genetics.org/supplemental/). Each of the fus3 kss1 rst1, fus3 rst1 rst2, and kss1 rst1 rst2 triple mutants had a growth defect and the fus3 kss1 rst1 rst2 quadruple-mutant strain was nearly unviable with an unusual cavitated colony morphology (data not shown). Analogous colony morphologies have been observed for wild-type yeast strains under certain stress conditions (ENGELBERG et al. 1998 ; CAVALIERI et al. 2000 ; KUTHAN et al. 2003 ). The rst1 mutation caused highly polarized growth and a unipolar budding pattern that typify filamentous morphology (Fig 1B). This cellular phenotype was slightly exacerbated by removal of one or more of any of the other three components. Deletion of FUS3 resulted in modest hyperpolarized growth, as described (ROBERTS and FINK 1994 ). All other single or multiple mutant combinations, including the fus3 kss1 rst2 triple mutant, grew with yeast form morphology.

RST1 is the main negative regulator of haploid invasive growth:
Strains that lacked RST1 exhibited pronounced agar invasion compared to a wild-type control (Fig 2A). Deletion of FUS3 and/or KSS1 in an rst1 strain further increased invasion, consistent with the previously described inhibitory roles for both MAPKs (ROBERTS and FINK 1994 ; COOK et al. 1997 ; MADHANI et al. 1997 ). As expected, the hyperinvasive growth of an rst1 strain was independent of upstream components within the mating/filamentous MAPK cascade as ste5 rst1, ste7 rst1, and ste11 rst1 double-mutant strains all invaded agar to the same extent as an rst1 strain did (Fig 2B). Consistent with a requirement for TEC1 in filamentous growth (PAN et al. 2000 ), a tec1 rst1 strain did not exhibit the hyperinvasive phenotype (Fig 2C). Deletion of FLO8 or PHD1, which encode two other transcription factors implicated in filamentous growth (PAN et al. 2000 ), did not affect the invasiveness of an rst1 strain (data not shown).

View larger version (116K): In this window In a new window Download PPT slide   Figure 2. Invasive phenotypes of combinatorial deletion strains. Patches of the indicated haploid 1278 strains were grown for 60 hr on rich media and then photographed before and after vigorous washing to remove surface growth.

As KSS1 has been reported to be a negative regulator of invasive growth (ROBERTS and FINK 1994 ; COOK et al. 1997 ; MADHANI et al. 1997 ), we compared its regulatory activity to that of RST1 by examining the phenotype of ste7 rst1 vs. ste7 kss1 double-delete strains (Fig 2D). Although deletion of KSS1 restored some invasive growth to a ste7 strain (COOK et al. 1997 ), deletion of RST1 restored invasion to a greater extent, suggesting that it is the primary inhibitor of the invasive-growth response. Deletion of RST2 was unable to restore invasion to the ste7 mutant, consistent with its limited role as a negative regulator. As expected, given the requirement for STE12 in expression of various signaling components, invasive growth of rst1, rst1 rst2, kss1 rst1, and fus3 rst1 strains was STE12 dependent (Fig 2E).

In contrast to the invasive phenotype of rst1 1278 strains, an rst1 mutation in either the W303 or the S288C strain background causes only a slightly elongated cell phenotype, whereas the rst1 rst2 double mutant exhibits florid invasive growth in all strain backgrounds (COOK et al. 1996 ; TEDFORD et al. 1997 ). In a natural S. cerevisiae isolate from a vineyard (MORTIMER et al. 1994 ), an rst1 mutation, but not an rst2 mutation, resulted in hyperinvasive growth (Fig 2F). The differences between wild-type and laboratory strain backgrounds presumably reflect the loss of activators of invasive growth during decades of propagation in the laboratory (LIU et al. 1996 ).

RST2 and KSS1 have both positive and negative roles in invasion:
Deletion of RST2 causes no obvious morphological phenotype, but does enhance the invasiveness of an rst1 mutant strain in both laboratory (COOK et al. 1996 ; TEDFORD et al. 1997 ; HUGHES et al. 2000 ) and 1278 backgrounds (Fig 2A). However, systematic analysis revealed that RST2 also plays a positive role in invasiveness in some circumstances as its deletion reduces invasion of fus3 and fus3 kss1 strains (Fig 2A). This positive function of Rst2 may be to help counteract inhibition by Rst1 because all triple- and quadruple-mutant combinations that bear rst1 and rst2 mutations vigorously invaded agar.

KSS1 also has properties of both an activator and an inhibitor of invasive growth (ROBERTS and FINK 1994 ; COOK et al. 1997 ; MADHANI et al. 1997 ). As shown previously (ROBERTS and FINK 1994 ), a fus3 strain displays hyperinvasive growth, whereas a kss1 strain displays hypoinvasive growth (Fig 2A). Elimination of either MAPK in an rst1 strain further stimulated invasive growth, supporting a negative regulatory role for both MAPKs (Fig 2A). Thus, Rst2 and Kss1 exhibit positive and negative effects on invasive growth depending on genetic context, while Fus3 always antagonizes invasive growth.

Strain background effects on transcriptional profiles:
Because strain background had a marked effect on the invasive outputs of the MAPK regulatory network, we examined the global transcriptional variations between wild-type and laboratory strains. DNA microarrays with >97% genome coverage were probed with differentially labeled cDNA pools and reported as the average of two independent experiments (see online supplementary Table 2 at http://www.genetics.org/supplemental/), in accord with Minimal Information About a Microarray Experiment guidelines (BALL et al. 2002 ). Strikingly, a haploid W303 strain had 496 genes induced greater than twofold and 112 genes repressed greater than twofold compared to a haploid 1278 strain (Fig 3A). Similar large-scale transcriptional variations have been observed between vineyard isolates (CAVALIERI et al. 2000 ; BREM et al. 2002 ). Hierarchical clustering of the 608 differentially regulated genes by gene ontology (GO) functional annotation (DWIGHT et al. 2002 ) revealed that the bulk of the differentially expressed genes (76%) fall under metabolism (136 genes), carbohydrate metabolism (21), transport (85 genes), or unknown (223 genes) processes (Fig 3A). The majority of transcriptional differences between W303 and 1278 genetic backgrounds were independent of ploidy (Fig 3B). Notably, the W303 strain had lower transcript levels for CLN1, SHO1, and TEC1, all of which are activators of filamentous growth (O'ROURKE and HERSKOWITZ 1998 ; MADHANI et al. 1999 ; PAN et al. 2000 ; Fig 3C). A number of genes encoding nutrient transporters were expressed at higher levels in the W303 background, including ZRT1 (high-affinity zinc uptake), FET4 (iron transport), PTR2 (small peptide uptake), HXT4 (hexose transport), PHO84 (inorganic phosphate transport), and MEP2 (ammonium transport). The most prominent coregulated gene cluster that showed elevated expression in the W303 backgrounds was the PHO regulon (OGAWA et al. 2000 ), including PHO84, PHO12, PHO11, PHO5, PHO3, SPL2, VTC1, VTC2, VTC3, and VTC4 (Fig 3C). Each of these gene promoters contains consensus-binding sites for Pho4, the bHLH transcription factor required for phosphate metabolism. PHO4 itself was expressed at similar levels in the W303 and 1278 backgrounds, consistent with the known post-transcriptional mechanisms that regulate Pho4 activity (CARROLL and O'SHEA 2002 ). Other misregulated genes in the W303 background are implicated in carbon and nitrogen metabolism, oxidative metabolism, and cell wall biosynthesis (Fig 3A). Higher transport and metabolic capabilities in the W303 strain background may buffer the effects of nutrient deprivation and thereby indirectly attenuate filamentous growth. These changes may have arisen by selection for altered nutritional responses during culture in the laboratory (LIU et al. 1996 ).

View larger version (46K): In this window In a new window Download PPT slide   Figure 3. Dependence of genome-wide transcriptional profile on strain background. (A) A total of 496 genes induced or 112 genes repressed greater than twofold in W303 haploid vs. 1278 haploid strains were compared to expression profiles of other indicated strains and clustered by the GO biological process (see online supplementary Table 9 at http://www.genetics.org/supplemental/ for gene list and expression ratios). The vertical line of induced genes in all strains corresponds to prototrophic marker genes used in strain construction. (B) Correlation plot of genome-wide transcriptional profiles of W303 haploid and W303 diploid strains compared to 1278 haploid and diploid strains, respectively. Correlation coefficients are given for 584 genes induced or repressed greater than twofold (sig) in W303 haploids and for all genes (all). (C) Top 30 genes induced (excluding URA3 and HIS3, which were used in construction of gene deletions) or repressed in W303 haploid vs. 1278 haploid strains are listed according to induction level and compared to expression profiles of other indicated strains.

Gene regulation by the MAPKs Fus3 and Kss1:
To correlate transcriptional changes with phenotypes caused by deletion of components in the mating/filamentous MAPK cascade, genome-wide vegetative expression profiles were generated for the 12 deletion mutant combinations shown in Fig 2A, as compared to a 1278 wild-type strain. Although genetic data indicate opposing functions for Fus3 and Kss1 for invasive growth (ROBERTS and FINK 1994 ), the genome-wide expression profiles of fus3 and kss1 strains in the 1278 background were similar ( = 0.60 for 82 genes that are induced/repressed greater than twofold in fus3 cells, or = 0.52 for 147 genes induced/repressed greater than twofold in kss1 cells). Four genes (YNL013C, SPC19, and the two pheromone-induced a-factor precursors MFA1 and MFA2) were induced and 30 genes were repressed specifically in kss1 cells and not in fus3 cells, whereas kss1 cells differed from fus3 kss1 cells by induction of the same 4 genes and repression of 47 genes (Fig 4A). Intriguingly, deletion of either MAPK results in the repression of YBR012W-B, YJR026W, YML045W, YJR028W, YAR010C, YBR012W-A, and YML040W, all of which are involved in Ty element transposition (SCHOLES et al. 2001 ). Three genes (FLO10 and the pheromone-induced genes YLR042C and PGU1) were induced and 14 genes repressed specifically in fus3 cells and not in kss1 cells, whereas fus3 cells differed from fus3 kss1 cells by the induction of 1 gene, YLR042C (ROBERTS et al. 2000 ; BREITKREUTZ et al. 2001A ), and the repression of 19 genes (Fig 4B). A TEC1-regulated set of 11 genes has been previously defined in haploid cells (MADHANI et al. 1999 ) but none of these genes were reduced in a kss1 strain and only 2, PGU1 and YLR042C, were induced in a fus3 strain (Fig 4B). While unexplained, these differences in part may be due to growth on rich (this work) vs. synthetic medium (MADHANI et al. 1999 ). In summary, while deletion of either MAPK influenced expression of a small gene set, there was little evidence of an invasive-specific gene program.

View larger version (58K): In this window In a new window Download PPT slide   Figure 4. Differential gene regulation by Fus3 and Kss1. (A) Vegetative expression profiles of kss1 cells result in >2.5-fold induction of 9 genes and >2.5-fold repression of 56 genes. Of the KSS1-regulated genes, fus3 cells are defective in the induction of 4 genes and the repression of 30 genes (top), whereas fus3 kss1 cells are defective in the induction of the same 4 genes (in blue) and the repression of 46 genes (bottom). A total of 24 repressed ORFs are shared (in blue). (B) Vegetative expression profiles of fus3 cells result in >2.5-fold induction of 7 genes and >2.5-fold repression of 23 genes. Of the FUS3-regulated genes, kss1 cells are defective in the induction of 3 genes and the repression of 14 genes (left), whereas fus3 kss1 cells are defective in the induction of 1 gene and the repression of 19 genes (right). One induced and 11 repressed ORFs are shared (in blue).

Redundant function of RST1 and RST2 in transcriptional repression:
Despite the fact that RST1 acted as the primary negative regulator of haploid invasive growth, we found that genome-wide expression profiles of rst1 and rst2 cells were highly correlated ( = 0.78 for 90 genes that are induced/repressed >2-fold in rst1 cells, or = 0.87 for 36 genes induced/repressed >2-fold in rst2 cells). A total of 17 genes were induced and 1 gene was repressed >2.5-fold in an rst1 strain and not in an rst2 strain; of these, most were uncharacterized open reading frames (ORFs; Fig 5A). In an rst2 strain, the only induced gene was SST2, which encodes a GTPase activator for Gpa1 that desensitizes the pheromone response, whereas just 1 gene of unknown function, YAL064W, was differentially repressed (Fig 5A). Thus, despite the fact that rst1 strains are invasive and rst2 strains are not, only 21 genes differ in expression between the two strains.

View larger version (54K): In this window In a new window Download PPT slide   Figure 5. Partially redundant roles of Rst1 and Rst2. (A) Vegetative genome-wide expression profile of an rst1 strain revealed 25 genes induced and 12 genes repressed >2.5-fold (right), while that of an rst2 strain revealed 6 genes induced and 12 genes repressed >2.5-fold (left). Of the 37 genes induced/repressed in rst1 cells, 19 genes were shared with the rst2 profile and 11 with the rst1 rst2 profile. Seven genes differentially regulated in the rst1 strain were shared between both rst2 and rst1 rst2 strains (in blue). Of the 18 genes induced/repressed in the rst2 strain, 2 genes were shared with the rst1 profile and 8 genes with the rst1 rst2 profile. One gene, YAL064W, was repressed in neither rst1 nor rst1 rst2 strains (in blue). ORFs that are marked ** correspond to Ty1 elements. (B) Vegetative genome-wide expression profile of a rst1 rst2 strain revealed 166 genes induced and 18 genes repressed >2.5-fold, as compared to the union of rst1 and rst2 strain profiles (see online supplementary Table 3 at http://www.genetics.org/supplemental/). (C) A total of 382 genes induced or 135 genes repressed >2-fold in an rst1 rst2 strain were compared to the other indicated genome-wide expression profiles and clustered by the GO biological process (see online supplementary Table 10 at http://www.genetics.org/supplemental/ for gene list and expression ratios). The number of genes in each category is indicated. Of 30 of the most-induced genes classified as mating specific, 25 are listed at the top right, and 27 of 215 of the most-induced genes of unknown biological processes are listed at the bottom right. Profiles for competitive hybridization of W303 haploid and diploid strains vs. 1278 haploid and diploid strains and for haploid vs. diploid 1278 strains are shown for comparison.

In marked contrast to either single deletion strain, an rst1 rst2 double-mutant strain has a complex transcriptional profile, with 184 genes induced or repressed >2.5-fold that are not altered in rst1 or rst2 strains (Fig 5B; online supplementary Table 3 at http://www.genetics.org/supplemental/). This transcriptional profile of the rst1 rst2 strain inflated to 517 genes when the threshold was lowered to 2-fold induction or repression (Fig 5C). Clustering of this gene set by functional annotation revealed a preponderance of coregulated genes that have no known function or are not annotated by GO process (Fig 5C). Substantial gene sets of potential relevance for invasive growth included those implicated in metabolism, transport, and mating. On the basis of the massive transcriptional program of an rst1 rst2 strain, but not of either single deletion strain, RST1 and RST2 are largely redundant for transcriptional regulation despite their quite different individual contributions to the invasive response (Fig 2A; Fig 5B).

The transcriptional profile of an rst1 rst2 strain in the S288C genetic background correlates strongly with the mating pheromone-induced profile (ROBERTS et al. 2000 ). However, the S288C strain background bears a mutation in FLO8 and is incapable of penetrating an agar surface (LIU et al. 1996 ). The rst1 rst2 profile also correlated strongly with the pheromone response in the 1278 background as all pheromone-induced genes were elevated in the rst1 rst2 strain (Fig 6A). However, in contrast to results in S288C, we found that the transcriptional profile of the rst1 rst2 strain in the 1278 genetic background extended the profile of pheromone-treated wild-type cells by 25 highly induced genes (Fig 6A). The strongest differentially induced genes were FLO10, which promotes flocculation and agar invasion (GUO et al. 2000 ), and PGU1, a pectinase possibly required for filamentous growth (MADHANI et al. 1999 ; ROBERTS et al. 2000 ). Other genes induced in rst1 rst2 cells but not in the mating program include HXT family members (HXT4, HXT6, and HXT7, as well as HXK1) and seripauperin family members believed to encode cell wall mannoproteins (PAU5, PAU7, YDR542W, YOL161C, and YHL046C).

View larger version (63K): In this window In a new window Download PPT slide   Figure 6. An extended gene set regulated by RST1/2 does not correlate with invasive growth. (A) Correlation plot of genome-wide transcriptional profiles of an rst1 rst2 strain vs. a wild-type strain treated for 30 min with 5 µM -factor. Correlation coefficients are given for 145 genes induced/repressed greater than twofold (sig) and for all genes (all). A total of 25 genes induced greater than sixfold in the rst1 rst2 strain and not in the pheromone-treated wild-type strain are listed according to induction level and compared to other genome-wide expression profiles as indicated. (B) Northern blot analysis of representative mRNAs for mating (FUS1), haploid cell type (STE2), invasive growth (PGU1 and FLO11), and loading control (ACT1). (C) Levels of each mRNA from B were quantitated by PhosphorImager and normalized to actin. Genes are indicated in the inset. (D) The transcriptional profile of an rst1 rst2 strain depends on STE12. Correlation plot of an rst1 rst2 strain compared to an rst1 rst2 ste12 strain. Some highly induced mating genes and prototrophic marker genes used in strain construction are labeled.

The cell-surface flocculin encoded by FLO11/MUC1 has been implicated as a downstream effector of filamentous growth (PAN et al. 2000 ), although in at least one instance invasive growth does not correlate with FLO11 induction (PALECEK et al. 2000 ). We observed that FLO11 expression also did not correlate with invasion in rst1, rst2, and rst1 rst2 strains, for which FLO11 log₂ expression ratios were 0.26, -0.30, and 0.17, respectively. Direct Northern blot analysis of FLO11 mRNA levels showed that there was no obvious correlation between FLO11 expression levels and the invasive properties of any of the multiple deletion strains tested (Fig 6B and Fig C). With the exception of the fus3 kss1 double-mutant strain, expression of the pheromone-regulated gene PGU1 correlates with the haploid invasive phenotype (Fig 6B and Fig C), consistent with the idea that at least some forms of filamentous growth represent a feature of the pheromone response (BREITKREUTZ and TYERS 2002 ). As expected, all gene induction in the rst1 rst2 strain was STE12 dependent (Fig 6D).

Hierarchical clustering of RST1/2-dependent gene expression profiles:
To assign possible functions to RST1/2-regulated genes, hierarchical clustering analysis of gene sets induced/repressed specifically in the rst1, rst2, and rst1 rst2 vegetative profiles was performed on a data set totaling 635 expression profiles from previous studies (EISEN et al. 1998 ; SPELLMAN et al. 1998 ; GASCH et al. 2000 ; HUGHES et al. 2000 ; ROBERTS et al. 2000 ; BREITKREUTZ et al. 2001A ). As might be expected, the rst1 gene set (17 genes induced, 6 repressed in rst1 but not in either rst1 rst2 or rst2 strains; Fig 5A) clustered in a node containing fus3 rst1, fus3 kss1 rst1, kss1 rst1, and rst1 rst2 strains (Fig 7A). The expression profile resulting from the overexpression of STE12 also fell within the rst1 cluster (ROBERTS et al. 2000 ), consistent with the primary negative regulatory role of RST1 in STE12-dependent transcription.

View larger version (71K): In this window In a new window Download PPT slide   Figure 7. Hierarchical clustering of 635 expression profiles according to RST1- and/or RST2-regulated gene sets. (A) Clustering directed by the set of genes specifically altered in an rst1 strain (23 genes; see Fig 5A). (B) Clustering directed by the set of genes specifically altered in an rst2 strain (9 genes; see Fig 5A). (C) Clustering directed by the set of genes specifically altered in an rst1 rst2 gene set (184 genes; see Fig 5C). Enlarged images reveal the experimentally clustered node for each gene set. (D) Hierarchical relationship for genome-wide profiles of 12 deletion sets from this study.

The minor rst2 gene set (one gene induced, eight genes repressed in rst2 but not in either rst1 rst2 or rst1 strains) clustered adjacent to both rst1 and fus3 rst2. However, the rst2 profile also clustered with a number of nitrogen-depletion experiments, largely because of a shared set of seven corepressed genes (Fig 7B). Although loss of RST2 function was not sufficient to trigger invasive growth, the nitrogen-depletion signature response is consistent with the known role of nitrogen limitation as the physiological trigger for diploid filamentation (BROWN and HOUGH 1965 ; GIMENO et al. 1992 ). The rst2 profile also clustered with early time points in a 37°–25° temperature-shift experiment and with deletion of the genes encoding dihydrofolate reductase (DFR1) and a protein involved in membrane trafficking (ERP4). These results suggest that deletion of RST2 may mimic multiple nutrient and/or stress-associated responses.

The expansive rst1 rst2 gene set clustered with fus3 kss1 rst1 and kss1 rst1 expression profiles (Fig 7C). However, despite the extensive overlap of the rst1 rst2 profile with the mating pheromone program (Fig 6A), the additional set of genes controlled by RST1/2 in the 1278 background caused the rst1 rst2 profile to cluster away from the pheromone profile. Transcriptional subclusters within the mutant profile suggest at least two additional roles for RST1/2. First, several conditions that induce DNA damage or replication stress, including methyl methanesulfonate and hydroxyurea treatment and deletion of RRM3, SGS1, DIA2, RAD57, and RNR1, clustered with the rst1 rst2 gene set (HUGHES et al. 2000 ; SMITH and JOHNSON 2000 ). It is therefore possible that, in wild-type cells, RST1 and RST2 play a role in the maintenance of genome stability. A potential connection between Kss1-regulated genes and rates of Ty transcription and transposition has been noted previously (MORILLON et al. 2000 ) and mobilization of Ty elements may be a possible mechanism of adaptive mutagenesis under stress conditions (DUNHAM et al. 2002 ). Induction of DNA damage response genes, such as RNR3, GAD1, YDR516C, YNL200C, TFS1, and GLK1, could in turn reflect an elevated rate of DNA strand breaks through retrotransposition events. It is also notable that UV irradiation of wild-type yeast strains induces stalk-like structures similar to the cavitated colony morphology of the fus3 kss1 rst1 rst2 quadruple mutant (ENGELBERG et al. 1998 ). A second substantial gene set controlled by RST1/2 is also subject to TUP1-SSN6-mediated repression. Strains that lack RST1 and either RST2 or KSS1 are derepressed for 74 of 166 genes that are normally repressed by TUP1-SSN6 (HUGHES et al. 2000 ; SMITH and JOHNSON 2000 ). A link between galactose-induced invasive growth and SNF1-mediated derepression, which antagonizes TUP1-SSN6, has recently been established (PALECEK et al. 2002A ). In summary, RST1 and RST2 appear to be functionally redundant for transcriptional repression of a large suite of genes, including the well-defined mating program and also additional subsets of genes specific to the 1278 background. The functions of these additional gene sets in invasive growth remain to be determined.

Absence of a unified invasion-specific transcriptional profile:
By restricting the clustering analysis to genome-wide profiles of 12 deletion strains, three main branches were evident (Fig 7D). The branch composed of rst1 rst2, kss1 rst1, and fus3 kss1 rst1 shares the common element rst1 (branch a in Fig 7D). However, two other strains that bear rst1, namely the rst1 single mutant and the fus3 rst1 double mutant, fall into a different cluster that also contains the rst2 single mutant and the fus3 rst2 double mutant (branch b in Fig 7D). Each strain in this branch has only minimal differences in transcriptional profile, despite invasive-growth responses that range from nonexistent to hyperinvasive. Finally, a third branch composed of fus3, kss1, kss1 rst2, and fus3 kss1 rst2 strains also has a minimal transcriptional profile even though the fus3 strain is invasive (branch c in Fig 7D). Thus, by this clustering method, transcriptional profiles do not correlate with the invasive phenotype.

To extend the search for gene expression patterns associated with the invasive phenotype, we used the independent clustering method BTSVQ (SULTAN et al. 2002 ). Analysis of the 12 deletion strains by BTSVQ revealed a trend similar to hierarchical clustering. Thus, strains in branch a of Fig 7D were separated from strains in branches b and c. Visualization by a pseudocolor matrix, which illustrates the partitioning of individual strains on the basis of quantized gene expression profiles, did not segregate invasive from noninvasive strains (Fig 8A). BSTVQ analysis was also extended to include data from two other noninvasive strains, kss1 rst1 ste12 and rst1 rst2 ste12, but again overall transcriptional profiles and invasive growth did not cosegregate in the matrix (Fig 8B).

View larger version (46K): In this window In a new window Download PPT slide   Figure 8. Comparison of transcriptional states of invasive and noninvasive strains by pseudocolor distance matrix maps. The distance matrix maps illustrate the partitioning of transcriptional profiles of 12 (A) or 14 (B) individual strains using a SOM algorithm to cluster samples on the basis of all genes. (C) A distance matrix map of the 14 individual strains shown in B except that the high-variance (SD > 0.05) set of 475 genes (online supplementary Table 4 at http://www.genetics.org/supplemental/) was used in the SOM. The bar scale indicates arbitrary distance units with scale from blue to red. The location of the individual clusters is indicated by the strain genotype adjacent to the node. Unlabeled hexagons are interpolated to make homogeneous color. The asterisk indicates invasive strains.

To parse for a possible weak transcriptional signature within the complete data sets, we segregated the data across all 14 transcriptional profiles into genes with high (SD > 0.05) and low (SD < 0.05) variance subsets (for gene lists, see online supplementary Tables 4 and 5, respectively, at http://www.genetics.org/supplemental/). A set of 475 high-variance genes was enriched for pheromone-regulated genes (116 genes), as might be expected given the overlap between the rst1 rst2 and pheromone profiles (ROBERTS et al. 2000 ; this work). Even within this highly regulated set, visualization by pseudocolor distance matrix maps failed to reveal transcriptional homogeneity across invasive strains (Fig 8C). Conversely, the set of 1444 genes with low variability actually contained many genes previously implicated in filamentous or invasive growth, including FLO8, AXL1, YDJ1, GRR1, CDC53, ZUO1, MSN5, BPL1, GTR1, CDC55, GPR1, MSS11, RAS2, RGA2, SOK2, SPH1, STE7, TPK1, TPK3, MSM1, CHD1, MSN1, MGA1, MEP1, URE2, and DAL80 (LIU et al. 1996 ; LORENZ and HEITMAN 1998 ; PALECEK et al. 2000 ).

Finally, to ensure that we identified all differentially regulated genes between invasive and noninvasive strains, we performed supervised gene filtering by selecting genes that have high expression in invasive or noninvasive strains (see MATERIALS AND METHODS) and used SOMs to select a subset from the 14 transcriptional profiles. Intriguingly, of the 123 genes identified in the noninvasive group and the 139 genes identified in the invasive group, a large proportion (106 in total with 39 being pheromone regulated) was shared between the two sets (see online supplementary Table 6 at http://www.genetics.org/supplemental/). This substantial overlap between the two groups in part explains the failure to detect a common transcriptional signature for invasive growth. Even the genes that, on average, were expressed at high levels in invasive strains (15 genes) or in noninvasive strains (16 genes) failed to precisely correlate with one state or the other. Pheromone-regulated genes were also found within each of the gene sets that partially segregated with invasive or noninvasive growth (see online supplementary Table 6 at http://www.genetics.org/supplemental/).

Only the elevated expression of FLO10 strictly correlated with the invasive phenotype across the complete set of combinatorial deletion strains tested. FLO10 is a member of the flocculin gene family that includes FLO1, FLO5, and FLO9; Flo10 promotes cell-cell adhesion but is not strictly required for invasive growth (CARO et al. 1997 ; GUO et al. 2000 ). Increased PRM1 and PGU1 expression also closely paralleled the invasive phenotype, except that PRM1 was induced in a fus3 kss1 rst2 strain and PGU1 was induced in a fus3 rst2 strain, neither of which invade. In addition, both PRM1 and PGU1 are strongly regulated by mating pheromone (ROBERTS et al. 2000 ; BREITKREUTZ et al. 2001A ). Deletion of KSS1 or RST2 in any context resulted in the induction of AGA1, a receptor for cell-cell adhesion, whereas the removal of RST1 consistently resulted in induction of FLO10 and the pheromone-regulated genes PGU1 and Fig 2. No genes were uniformly altered by deletion of FUS3 in different genetic contexts.

The above conclusions are underscored by the marked transcriptional difference between kss1 rst1 and fus3 rst1, which have a virtually indistinguishable hyperinvasive phenotype (Fig 1B). Of the 55 genes induced/repressed >2.75-fold in both rst1 rst2 and kss1 rst1 strains, only 9 genes (PGU1, FLO10, AGA1, PRM1, YLL064C, YLR194C, DIA1, CMK2, and YOR385W) were similarly regulated in the fus3 rst1 strain (Fig 9A). As might be expected, most of this overlap was due to RST1-regulated genes (7 of 9 genes). Of the 99 genes induced/repressed >2-fold in both rst1 rst2 and kss1 rst1 strains, 48 were of unknown biological function (Fig 9B). This profile clustered adjacent to that of fus3 kss1 rst1 and GAL-STE12 strains (Fig 9C; online supplementary Table 7 at http://www.genetics.org/supplemental/). Yet the expression profiles of rst1 and fus3 rst1 strains did not cluster near this gene set, despite the invasive phenotypes of each strain. Only a few genes from the set were altered in either fus3 or fus3 kss1 strains, both of which are invasive (Fig 9B). Genes induced/repressed >2.5-fold in fus3 kss1 rst1 profile and not in the fus3 rst1 profile revealed differences only in the induction of 35 genes, including several PAU and HXT gene family members (online supplementary Table 8 at http://www.genetics.org/supplemental/), which are also partially upregulated in the kss1 rst1 and rst1 rst2 profiles as well. While this candidate Kss1-repressed gene set was not altered in other noninvasive kss1 strains, it was also not expressed in the fus3 rst1 strain and so is apparently dispensable for invasive growth. Taken together, the above comparisons indicate that while complex transcriptional changes can be wrought by genetic manipulation of the distal MAPK network, these alterations do not yield a universal program of gene expression that correlates with the invasive-growth response.

View larger version (79K): In this window In a new window Download PPT slide   Figure 9. Comparison of genome-wide expression profiles derived from highly invasive strains. (A) Genes induced >2.75-fold in both kss1 rst1 and rst1 rst2 strains (excluding URA3 and HIS3, 65 genes in total) were organized according to induction/repression level and compared to the indicated expression profiles for invasive and noninvasive strains. Of the 65 genes induced/repressed >2.75-fold that are shared between rst1 rst2 and kss1 rst1 strain profiles, 10 are shared with the fus3 rst1 strain profile (in blue). (B) Genes induced/repressed >2-fold in both rst1 rst2 and kss1 rst1 strains (99 genes in total) were compared to the indicated expression profiles within the combinatorial deletion series and clustered by the GO biological process (see supplementary Table 7 at http://www.genetics.org/supplemental/ for gene list and expression ratios). (C) The same gene set as in B was clustered with 635 expression profiles. Pink bars, low nitrogen profiles; blue bar, fus3 rst1 and rst1 profiles.

	DISCUSSION

TOP ABSTRACT MATERIALS AND METHODS RESULTS DISCUSSION LITERATURE CITED

Beginning with their initial discovery as components of the yeast mating pheromone pathway, MAPK cascades have been implicated in the transmission of environmental signals to the transcriptional machinery (ROBINSON and COBB 1997 ; GUSTIN et al. 1998 ; TAN and KIM 1999 ; CHANG and KARIN 2001 ). The genome-wide transcriptional response to mating pheromone conforms to this regulatory paradigm (ROBERTS et al. 2000 ). In contrast to the single signal that governs the pheromone response, multiple signals appear to trigger invasive growth (MADHANI and FINK 1998A ; PAN et al. 2000 ; RUA et al. 2001 ; PALECEK et al. 2002B ). Given the complexity of the interface between distal components of the MAPK pathway and the transcriptional machinery, we sought to systematically investigate the role of each component in invasive growth. Combinatorial deletion analysis revealed that RST1 is a primary inhibitor of haploid invasive growth, that FUS3 plays a secondary inhibitory role, and that RST2 and KSS1 can have either positive or negative roles in the invasive response, depending on the genetic context. The spectrum of phenotypes caused by disruption of different combinations of elements in the Rst1-Rst2-Fus3-Kss1 network is not easily reconciled with a linear pathway that impinges solely on a single transcriptional program. Rather, perturbation of the network in different ways leads to distinct transcriptional outputs that do not correlate with morphological or invasive phenotypes. These findings suggest that invasive growth is not a single unified state at the molecular level.

Transcriptional outputs and invasive growth:
Only one gene, FLO10, consistently correlated with invasive growth and filamentous morphology in our studies and yet flo10 strains are still competent for invasion (GUO et al. 2000 ). Indeed, the invasive phenotypes of rst1, fus3, and fus3 rst1 strains are accompanied by only minimal transcriptional profiles. In the few other cases where a genome-wide approach has been taken, an invasive-specific transcriptional program has not emerged. For example, a genome-wide analysis of TEC1-regulated genes in hapl