Repeated Cis-Regulatory Tuning of a Metabolic Bottleneck Gene during Evolution

Abstract

Repeated evolutionary events imply underlying genetic constraints that can make evolutionary mechanisms predictable. Morphological traits are thought to evolve frequently through cis-regulatory changes because these mechanisms bypass constraints in pleiotropic genes that are reused during development. In contrast, the constraints acting on metabolic traits during evolution are less well studied. Here we show how a metabolic bottleneck gene has repeatedly adopted similar cis-regulatory solutions during evolution, likely due to its pleiotropic role integrating flux from multiple metabolic pathways. Specifically, the genes encoding phosphoglucomutase activity (PGM1/PGM2), which connect GALactose catabolism to glycolysis, have gained and lost direct regulation by the transcription factor Gal4 several times during yeast evolution. Through targeted mutations of predicted Gal4-binding sites in yeast genomes, we show this galactose-mediated regulation of PGM1/2 supports vigorous growth on galactose in multiple yeast species, including Saccharomyces uvarum and Lachancea kluyveri. Furthermore, the addition of galactose-inducible PGM1 alone is sufficient to improve the growth on galactose of multiple species that lack this regulation, including Saccharomyces cerevisiae. The strong association between regulation of PGM1/2 by Gal4 even enables remarkably accurate predictions of galactose growth phenotypes between closely related species. This repeated mode of evolution suggests that this specific cis-regulatory connection is a common way that diverse yeasts can govern flux through the pathway, likely due to the constraints imposed by this pleiotropic bottleneck gene. Since metabolic pathways are highly interconnected, we argue that cis-regulatory evolution might be widespread at pleiotropic genes that control metabolic bottlenecks and intersections.

Introduction

Repeated use of the same genes to achieve similar phenotypic outcomes is thought to reflect a combination of similar selective pressures and genetic constraints (Christin et al. 2010; Stern 2013). Both coding changes (Hoekstra et al. 2006; Christin et al. 2007) and cis-regulatory changes (Sucena et al. 2003; Prud'Homme et al. 2006; Rogers et al. 2013; Rebeiz and Williams 2017) have been shown to underlie repeated phenotypic alterations. Cis-regulatory changes have been hypothesized to be the key genetic causes of morphological evolution because strong pleiotropic constraints are imposed when key developmental genes are reused spatially and temporally (Prud'Homme et al. 2006; Carroll 2008; Stern and Orgogozo 2008; Rebeiz et al. 2009; Rebeiz and Williams 2017). In contrast, physiological and metabolic traits have frequently evolved through changes in both protein-coding and cis-regulatory regions (Ihmels et al. 2005; Lin et al. 2013; Roop et al. 2016; Sood and Brickner 2017). Mutations in coding regions have been frequently shown to lead to the acquisition of novel enzymatic activities and radical modifications in specificity (Thomson et al. 2005; Des Marais and Rausher 2008; Voordeckers et al. 2012). Cis-regulatory rewiring has also been associated with many physiological changes, including the transition from aerobic respiration to aerobic fermentation in yeasts, even as most central metabolic functions were conserved (Ihmels et al. 2005; Jiang et al. 2010; Lin et al. 2013; Roy et al. 2013). However, it is unknown whether there are any general principles that would lead specific cis-regulatory changes to occur frequently in metabolic evolution. Here we address this question by taking advantage of trait variation among yeast species in catabolism of the sugar galactose. As a paradigm of eukaryotic molecular biology, the metabolic and regulatory pathway for GALactose utilization in the budding yeast model Saccharomyces cerevisiae offers a suite of well-characterized molecular features. The availability of several high-quality genome assemblies from diverse budding yeast species further allows us to determine which pathway features are conserved, which are variable, and which are associated with trait variation.

Galactose concentrations are highly variable across environments (Marsilio et al. 2001; Nierop et al. 2001), providing the opportunity for diverse yeast species to adopt different strategies for consuming this resource. Qualitative differences in galactose utilization that evolved due to the parallel losses of entire GAL networks have received considerable prior attention (Hittinger et al. 2004; Riley et al. 2016). Recent functional comparisons of the GAL network in multiple yeast species have also demonstrated quantitative variation across diverse budding yeasts, generally focusing on pairwise comparisons (Martchenko et al. 2007; Peng et al. 2015; Dalal et al. 2016; Kuang et al. 2016; Roop et al. 2016; Sood and Brickner 2017). The GAL network of S. cerevisiae encodes three enzymes in the galactose-specific Leloir pathway (Gal1, Gal7, and Gal10), a transporter (Gal2), and three regulators (Gal3, Gal4, and Gal80) (fig. 1). In S. cerevisiae, the metabolic bottleneck for galactose metabolism is controlled by the enzyme phosphoglucomutase, which catalyzes the interconversion of glucose-1-phosphate into glucose-6-phosphate (Bro et al. 2005; Garcia Sanchez et al. 2010; Hong et al. 2011). Phosphoglucomutase controls the flux from the Leloir pathway into glycolysis and integrates flux from several other pathways. This enzyme is encoded by PGM1, or a pair of paralogs (PGM1 and PGM2) in yeasts that underwent whole genome duplication (WGD) (Wolfe and Shields 1997; Marcet-Houben and Gabaldon 2015). In S. cerevisiae, PGM1 encodes the minor isoform of phosphoglucomutase, whereas PGM2 encodes the major isoform (Bevan and Douglas 1969). Overexpressing PGM2 is among the best ways to increase flux through the pathway in S. cerevisiae (Bro et al. 2005; Garcia Sanchez et al. 2010). However, PGM1 and PGM2 are not directly regulated by Gal4 in S. cerevisiae (Ren et al. 2000). The expression of PGM1 in S. cerevisiae is not induced by galactose, whereas PGM2 is only mildly induced (∼4 fold) in a Gal4-independent manner (Oh and Hopper 1990; Rubio-Texeira 2005). In contrast, the Leloir enzymes are highly induced (up to 1,000 fold) in a Gal4-dependent manner (Lohr et al. 1995; Rubio-Texeira 2005). Unlike these galactose-specific enzymes, phosphoglucomutase is also involved in other metabolic pathways, including the pentose phosphate pathway (Cherry et al. 2012), glycogen biosynthesis (Hirata et al. 2003; Cherry et al. 2012), and trehalose biosynthesis (Mulet et al. 2004; Cherry et al. 2012).

Fig. 1.

Diagram of the GAL network in the budding yeast model S. cerevisiae and the role of Gal4-binding sites in explaining quantitative variation in growth on galactose. In the absence of galactose, the transcription factor Gal4 is inhibited by the corepressor Gal80, which prevents the expression of the GALactose utilization genes. When galactose is present, Gal80 is sequestered by the co-inducer Gal3, allowing Gal4 to induce GAL gene expression. Other GAL genes encode the transporter Gal2 and three enzymes in the Leloir pathway that catabolize galactose. Glucose-1-phosphate, the end product of the Leloir pathway, is converted by the phosphoglucomutases Pgm1/2 into glucose-6-phosphate, which then enters glycolysis. For each protein, except for Gal2 (where paralogous hexose transporters complicated analyses), the correlation (R²) between the number of predicted Gal4-binding sites upstream of the gene encoding it and growth on galactose across diverse yeast species (see fig. 2) is color-coded according to the key.

Here, we show that expression of the bottleneck gene PGM1/2 has been repeatedly tuned across the budding yeast family Saccharomycetaceae, which spans about 100 My of evolution, quantitatively modulating galactose metabolism by the addition or subtraction of Gal4-binding sites in its promoter. We show that Gal4-mediated regulation is necessary for vigorous galactose metabolism in multiple yeast species and that the addition of Gal4-regulated copies of PGM1/2 are sufficient to confer vigorous growth to species that lack this regulation, such as S. cerevisiae. In contrast to increasing the basal expression level, modifying this cis-regulatory connection during evolution would have provided a mechanism for PGM1/2 expression to respond specifically to galactose, which we hypothesize resolved the constraints imposed by converging metabolic pathways. These genetic constraints and continually shifting ecological niches likely underlie the repeated evolutionary gain and loss of the Gal4-binding sites upstream of this metabolic bottleneck gene.

Results

Galactose-Mediated Regulation of Its Bottleneck Gene Is Associated with Quantitative Variation

Yeasts of the family Saccharomycetaceae display dramatic variation in their abilities to grow on galactose. Growth experiments with galactose as the sole carbon source revealed widespread quantitative variation in galactose metabolism (fig. 2A, supplementary fig. S1A, Supplementary Material online). To understand the genetic basis of this variation, we applied a comparative approach that leveraged mechanistic understandings of the GAL network in S. cerevisiae and recent functional studies on Saccharomyces uvarum (formerly S. bayanus var. uvarum). Prior studies showed that S. uvarum grows faster on galactose because its GAL network is more active than that of S. cerevisiae, in part due to cis-regulatory changes that affect the expression of multiple GAL genes (Caudy et al. 2013; Kuang et al. 2016; Roop et al. 2016; Sood and Brickner 2017). To determine which, if any, genetic and molecular features were associated with quantitative variation in galactose growth between species, we examined 19 DNA sequence features of the GAL networks of 17 species from the family Saccharomycetaceae. We chose at least two species that lacked obvious mutations in their GAL genes from six genera with published high-quality genome assemblies. The 19 features included the number and position of Gal4-binding sites upstream of every GAL gene, copy number variation, and peptide motifs in the encoded enzymes. We found that the Gal4-binding sites upstream of the bottleneck gene PGM1/2 had the strongest association with growth on galactose among all the characteristics examined (figs. 1 and 2; supplementary table S1 and fig. S1A, Supplementary Material online, regression coefficient R² = 0.81, P = 1.9e-7). This correlation was observed in several different media formulations containing galactose as the sole carbon source and when the temperature was varied (supplementary fig. S4, Supplementary Material online), but it was not seen on glucose (supplementary fig. S1B and C).

Fig. 2.

Predicted Gal4-binding sites upstream of PGM1/2 strongly correlate with growth on galactose across the yeast family Saccharomycetaceae. (A) The number of predicted Gal4-binding sites upstream of PGM1 and PGM2 strongly correlates with relative growth on galactose. The phylogeny and whole genome duplication (WGD) are shown at the left as published in prior genome-wide analyses (Shen et al. 2016). “Num. Sites” denotes the number of predicted Gal4-binding sites upstream of PGM1 (blue for WGD species, black for non-WGD species) or PGM2 (purple). The shades separate each monphyletic genus. Each data box is color-coded based on the number of binding sites, with the darkness of the blue color corresponding to the number of predicted binding sites, light blue indicating that the predicted binding sites had no detected function, and orange indicating the absence of any predicted binding sites. Relative growth (n = 6) denotes the number of cell divisions after 15 h in synthetic complete medium (SC) +2% galactose, which was calculated as log₂[(OD_strain − OD_media)/(OD_start − OD_media)]. This calculation was applied for all figures. The 15-h time point was chosen because most strains have started to initiate growth after 15 h in galactose (supplementary fig. S1A). Each strain is designated by a 4-letter species abbreviation (Vpol: Vanderwaltozyma polyspora, Tpha: Tetrapisispora phaffii, Tbla: Tetrapisispora blattae, Suva: Saccharomyces uvarum, Sarb: Saccharomyces arboricola, Skud-Port: Saccharomyces kudriavzevii Portuguese population, Skud-Jap: Saccharomyces kudriavzevii Japanese population (a negative control whose genome lacks a functional GAL network), Scer: Saccharomyces cerevisiae, Ncas: Naumovozyma castellii, Ndai: Naumovozyma dairenensis, Zrou: Zygosaccharomyces rouxii, Zkom: Zygosaccharomyces kombuchaensis, Lklu: Lachancea kluyveri, Lthe: Lachancea thermotolerans, Kaes: Kluyveromyces aestuarii, Klac: Kluyveromyces lactis, Kdob: Kluyveromyces dobzhanskii, Kmar: Kluyveromyces marxianus). (B) There was a strong correlation between the number of predicted binding sites and growth on galactose. The data were extracted from 1A, and the median was used to represent each species. The gray shaded area corresponds to the 95% confidence interval. (C) Absence/presence of predicted binding sites (converted from 1B) revealed the same pattern (P = 2.9e-5, n = 9, df = 13.9, t = −6.1, Welch's two-sample t-test). Ancestral state reconstruction shows that putative Gal4-PGM1/2 connections are evolutionarily dynamic with limited phylogenetic signal, supporting the treatment of taxa as independent (supplementary fig. S9, Supplementary Material online). Note that we were not able to obtain consistent growth with Kazachstania africana, so the genus Kazachstania was excluded. Otherwise, we included every characterized genus in this family where at least two species had published genome sequences.

Direct Regulation of PGM1/2 by Gal4 Quantitatively Modulates Growth on Galactose in Multiple Species

Although PGM1/2 has not been shown to be directly regulated by Gal4 in any species, including S. cerevisiae, S. uvarum offered a particularly attractive model to test the hypothesis of direct regulation for several reasons: we previously showed that S. uvarum PGM1 was induced 18-fold by galactose, has two predicted Gal4-binding sites, and its expression was further increased in mutants lacking the corepressor pair Gal80/Gal80b (Kuang et al. 2016). To test whether S. uvarum PGM1 is directly regulated by Gal4, we mutated one base pair of a predicted Gal4-binding site (CGGN₁₁CCG) upstream of S. uvarum PGM1 (Godecke et al. 1991; Hittinger and Carroll 2007; Robasky and Bulyk 2011). The Gal4-binding motif (CGGN₁₁CCG) has been shown to be conserved in both Kluyveromyces lactis and S. cerevisiae, which span about 100 My of evolution (Godecke et al. 1991; Hittinger and Carroll 2007), and the same motif is also enriched upstream of S. uvarum GAL genes (Kuang et al. 2016). In S. cerevisiae, point mutations in Gal4-binding motifs are sufficient to greatly decrease Gal4-binding strength and disrupt regulation (Giniger et al. 1985). We therefore reasoned that, if mutating the Gal4-binding motif resulted in galactose-specific growth defects, the predicted binding motif would be highly likely to be functional. Indeed, a single point mutation was sufficient to slow down growth on galactose by 20% compared with wild-type S. uvarum (fig. 3A), a defect that was galactose-specific (supplementary fig. S2A, Supplementary Material online). To further examine its impact on metabolic flux, we tested whether removing Gal4-mediated induction from this bottleneck gene sufficiently reduced the flux to rescue the temporary growth arrest phenotype caused by galactose metabolic overload seen in S. uvarum mutants lacking Gal80 corepressors (Kuang et al. 2016). As expected, the deletion of both of the predicted Gal4-binding sites upstream of S. uvarum PGM1 rescued the temporary growth arrest phenotype, but also led to a slower maximum growth rate on galactose (supplementary fig. S3, Supplementary Material online). This evidence suggests that S. uvarum PGM1 is a bottleneck gene that controls flux through the GAL pathway. Thus, the high metabolic flux and vigorous growth of wild-type S. uvarum on galactose requires direct Gal4-mediated induction of PGM1, a novel regulatory connection that S. cerevisiae lacks.

Fig. 3.

Gal4-mediated regulation of phosphoglucomutase is necessary and sufficient to support vigorous growth on galactose for multiple yeast species. (A) Mutation of one of two predicted Gal4-binding sites upstream of S. uvarum PGM1 reduced growth on galactose compared with wild type (P = 6e-6, n_WT = 15, n_mutant = 11). (B) Mutation of the predicted Gal4-binding site upstream of L. kluyveri PGM1 reduced growth on galactose compared with wild type. (P = 7.6e-9, n_WT = 21, n_mutant = 14). Note that CCG was used as the PAM site for CRISPR/Cas9-engineering and was mutated to CAA to disrupt both the Gal4-binding site and the PAM site. (C) Hypothesis of GAL network activity tuning with a novel (dotted line) Gal4-Pgm1/2 feedforward loop. (D) S. uvarum PGM1 increased growth on galactose in S. cerevisiae (P = 5e-3, n = 6). (E) S. uvarum PGM1 increased growth on galactose in V. polyspora (P = 0.01, n_{Empty vector} = 6, n_SuvaPGM1 = 5). All tests are Wilcoxon rank sum tests.

To test whether Gal4-mediated induction of PGM1/2 contributes to vigorous growth on galactose in multiple species, we examined the impact of mutating predicted Gal4-binding sites upstream of PGM1/2. We first developed a genome-editing approach potentially universal across yeasts by integrating a CRISPR/Cas9 system with an autonomously replicating sequence (ARS) that functions in diverse genera (Liachko and Dunham 2014). We applied this method to PGM1 and PGM2 genes with predicted upstream Gal4-binding sites in species with published transformation protocols. In addition to the genus Saccharomyces, this genome-editing system can induce targeted point mutations in at least the genera of Lachancea and Kluyveromyces, which diverged from S. cerevisiae about 100 Ma. Through either CRISPR/Cas9 or traditional approaches (Alexander et al. 2014), we mutated a predicted Gal4-binding site upstream of Lachancea kluyveri PGM1, Kluyveromyces lactis PGM1, and Saccharomyces kudriavzevii PGM2 (in a Portuguese strain capable of growth on galactose [Hittinger et al. 2010]). The L. kluyveri mutant grew 32% more slowly on galactose compared with wild type, but not on glucose, indicating that its Gal4-binding site is specifically required for vigorous growth on galactose (fig. 3B and supplementary fig. S2B, Supplementary Material online). However, the K. lactis and S. kudriavzevii mutants lacked observable growth defects, indicating that predicted Gal4-binding sites were not required for robust growth in the conditions we tested (supplementary fig. S5A and B, Supplementary Material online). However, full induction of S. kudriavzevii PGM2 by galactose did require the predicted Gal4-binding site (supplementary fig. S5E, Supplementary Material online). Even though this site was required to reach expression levels similar to S. uvarum PGM1, the higher basal expression of S. kudriavzevii PGM2 might render Gal4 induction dispensable to support its relatively modest growth (supplementary fig. S5C and D, Supplementary Material online). Thus, we conclude that direct regulation of PGM1/2 by Gal4 supports vigorous galactose growth in some species (e.g., S. uvarum and L. kluyveri), whereas other mechanisms are important for other species.

To examine whether up-regulating phosphoglucomutase expression alone was sufficient to improve growth on galactose across yeast species, we introduced a Gal4-regulated PGM1 gene (S. uvarum PGM1 with both of its predicted Gal4-binding sites) into multiple species that lack predicted Gal4-binding sites upstream of PGM1/2. S. uvarum PGM1 has a relatively low level of basal expression and a high level of galactose induction, so it is predicted to enhance flux through the GAL pathway while minimizing pleiotropic effects when cells are not grown on galactose (supplementary figs. S2 and S5C and D, Supplementary Material online) (Kuang et al. 2016). Introduction of S. uvarum PGM1 into S. cerevisiae enhanced its growth on galactose by 56% (fig. 3D), whereas it enhanced the growth of Vanderwaltozyma polyspora by 110% (fig. 3E). Therefore, up-regulation of PGM1 alone is sufficient to increase the growth on galactose of multiple species.

GAL Network Atrophy Did Not Affect the Protein-Coding Sequence of the Pleiotropic Bottleneck Gene

To test whether the strong association of galactose growth with predicted Gal4 regulation of PGM1 represents a general trend, we examined several additional closely related species in the genus Lachancea, which includes one of the species whose Gal4-mediated regulation we verified above (fig. 3B). This genus also provided a good opportunity to examine gene-trait association because all 12 known species in this family have high-quality genome assemblies available (Sarilar et al. 2015; Vakirlis et al. 2016), more than any other monophyletic budding yeast genus, including Saccharomyces. Similar to what we saw in our sparser sampling across the Saccharomycetaceae (fig. 2), we found that the presence of a predicted Gal4-binding site was highly correlated with galactose growth (fig. 4 and supplementary fig. S6, Supplementary Material online). Perhaps most strikingly, the topology of the species tree and the novel location of its predicted Gal4-binding site (318 bp upstream of PGM1 and overlapping with the neighboring predicted coding sequence) suggest that Lachancea dasiensis may have reacquired the ability to grow vigorously on galactose by acquiring a new Gal4-binding site (fig. 4). In fact, even though the genomes of several species of Lachancea were predicted to encode functional GAL genes, they grew so slowly on 2% galactose that taxonomists had previously scored them as nongrowing, weak, or variable (Kurtzman et al. 2011). In these cases, the presence of a predicted Gal4-binding site upstream of PGM1 was actually a better predictor of galactose growth in the conditions we tested than the presence of GAL genes. We hypothesized that these slow-growing species may have experienced changes in the strength of purifying selection acting on their GAL genes, and indeed, we found statistically significant relaxations of the selective pressure acting against nonsynonymous substitutions in all three GAL genes encoding enzymes (table 1, P = 7e-5, Fisher’s method). In contrast, we did not detect any signal of relaxed selection in the pleiotropic gene PGM1 (table 1). Some of these species lost GAL genes through pseudogenization or deletion, including GAL4 and GAL80 in some cases, as well as experiencing translocations and gene duplications (fig. 4 and supplementary fig. S7 and supplementary notes, Supplementary Material online). Thus, we propose that the loss of the Gal4-PGM1 regulatory connection and the relaxed selection on components dedicated to galactose metabolism may represent an early stage of GAL network atrophy that, in some cases, led to degeneration and complete loss.

Fig. 4.

Predicted Gal4-binding sites upstream of PGM1 predicted galactose growth differences among closely related species in the genus Lachancea. (A) A genome-wide consensus phylogeny (Vakirlis et al. 2016) shows branches under normal purifying selection as black and those under relaxed selection as red (table 1). In the boxplot (n = 6), each data box is colored coded by either blue (presence of predicted Gal4-binding sites) or orange (absence). The organization of the required GAL1/10/7 gene cluster is represented on the right. Each arrow denotes the direction of transcription. Distances are proportional, except for regions marked by two slashes. Genes on the same chromosome are connected with a black line. Each homolog is color-coded. Pseudogenes (Ψ) or likely pseudogenes are represented with their gene names in red. Each species is designated by a 4-letter abbreviation (Lklu: Lachancea kluyveri, Lcid: L. cidri, Lfer: L. fermentati, Lmir: L. mirantina, Lwal: L. waltii, Lthe: L. thermotolerans, Lque: L. quebecensis, Lnot: L. nothofagi, Ldas: L. dasiensis, Lmey: L. meyersii, Lfan: ′L. fantastica′ nom. nud., Llan: L. lanzarotensis). All strains were cultured in SC + 2% galactose. (B) The presence of predicted Gal4-binding sites associates with faster growth on galactose (median of each species from 3A is plotted as individual dots) (P = 5.8e-3, n_no= 4 n_yes= 8, t = −5.30, df =4.08, Welch's two-sample t-test).

Table 1.

Compared with Reference (Ref) Branches, Slow-Growing Lachancea spp. that Grew Slowly and Lacked Predicted Gal4-PGM1 Regulation (test branches) Experienced Relaxed Selection at Dedicated GAL Genes, but not at the Pleiotropic Gene PGM1.

	GAL1 (P Value = 0.0457, K = 0.8281)		GAL7 (P Value = 0.0005, K = 0.5429)		GAL10 (P Value = 0.0258, K = 0.8336)		PGM1 (P Value = 0.8985, K = 0.9934)
Alternative Model	Test Branches	Ref Branches	Test Branches	Ref Branches	Test Branches	Ref Branches	Test Branches	Ref Branches
Site class I	0.0086 (79.85%)	0.0032 (79.85%)	0.0044 (77.79%)	0.0005 (77.79%)	0.0000 (68.51%)	0.0000 (68.51%)	0.0101 (94.25%)	0.0098 (94.25%)
Site class II	0.3413 (19.46%)	0.2730 (19.46%)	0.4451 (19.92%)	0.2251 (19.92%)	0.1633 (27.18%)	0.1137 (27.18%)	1 (4.26%)	1 (4.26%)
Site class III	56.9198 (0.69%)	131.6834 (0.69%)	2.4805 (2.29%)	5.3297 (2.29%)	1.9259 (4.31%)	2.1951 (4.31%)	1.7126 (1.49%)	1.7188 (1.49%)

Note.—Test branches refer to internal and terminal branches leading to species growing slowly on galactose (fig. 4A). Reference (Ref) branches refer to all the remaining branches on the tree. The analysis was conducted using the RELAX method implemented in the HYPHY package, and K is the relaxation-of-selection parameter (Wertheim et al. 2015). Site classes I–III refer to groups of sites evolving under a shared selection regimen, in a descending order of the number of sites they represent (i.e., class I is the most frequently observed, and class III is the least frequently observed). Values indicate d_N/d_S ratios (ω) for each class of sites, along with the proportion of sites that belong to that class in parentheses. Note that the Test GAL genes are evolving significantly nearer to neutrality than the Ref GAL genes (i.e., ω nearer to 1 for Test than Ref), whereas Test and Ref rates are similar for PGM1.

Conditional Benefits of Direct Regulation

To further model how Gal4-PGM1/2 regulatory connections evolved in the family Saccharomycetaceae, we performed ancestral state reconstruction using the R packages Geiger (Harmon et al. 2008) and phytools (Revell 2012). Although there was limited signal to resolve individual nodes, all likely evolutionary trajectories involve multiple gains and losses of predicted Gal4-PGM1/2 regulatory connections during evolution (supplementary fig. S9, Supplementary Material online). Among taxa that underwent the whole genome duplication, there did not appear to be any pattern of which paralog gained or lost sites, but all of the species examined were predicted to only have Gal4-binding sites upstream of a single PGM1/2 gene (fig. 1 and supplementary table S1, Supplementary Material online). The repeated gains and losses of Gal4-binding sites upstream of PGM1/2 during evolution made us wonder whether the effects of Gal4-binding sites might be associated with specific galactose conditions. We hypothesized that the galactose-inducibility of PGM1/2 could affect growth more strongly as galactose concentrations increased. Consistent with this hypothesis, when the galactose-inducible S. uvarum PGM1 was added to V. polyspora, it grew much better than the wild-type strain at higher concentrations (5% galactose), but not at lower concentrations (0.5% galactose) (fig. 5B). Additionally, mutation of the native Gal4-binding site of L. kluyveri PGM1 caused stronger defects as galactose concentrations increased (fig. 5C). Also consistent with this model, several species of Lachancea that had lost Gal4-PGM1 sites actually grew better at low concentrations than high concentrations of galactose, suggesting that they might be specialists at low concentrations of galactose (supplementary fig. S8, Supplementary Material online). This trait was not exclusive to this genus but was shared with the distantly related species V. polyspora, where it was particularly pronounced (fig. 5A, Supplementary Material online). These data suggest that direct induction of PGM1/2 by Gal4 has stronger phenotypic impacts at high concentrations of galactose, but its effects are limited at low concentrations.

Fig. 5.

The importance of Gal4-mediated regulation of PGM1 depends on galactose concentrations. (A) V. polyspora grew better at low concentrations of galactose (P = 1.7e-2, n = 5, Wilcoxon rank sum test). Box plots are shown at the left and growth curves at the right. (B) When added to V. polyspora, S. uvarum PGM1 only increased galactose growth at high galactose concentrations. Growth differences were calculated as: (Div_SuvaPGM1 − Div_{Empty vector})/Div_{Empty vector} × 100%, where “Div” denotes the number of cell divisions at 15 h on 0.5% galactose and 30 h on 2% or 5% galactose. Note that time points were chosen immediately prior to when the wild type saturated. The gain in growth conferred by SuvaPGM1 significantly increased as the galactose concentrations increased (P = 1.6e-2, n = 5, Jonckheere–Terpstra test). (C) Mutating the Gal4-binding site upstream of PGM1 caused greater growth defects in L. kluyveri as galactose concentrations increased. Growth differences were calculated as: (Div_WT − Div_Mutant)/Div_WT × 100%, where “Div” denotes the number of cell divisions after 15 h on 2% galactose, 10 h on 0.5% galactose, and 9 h on 0.2% galactose. Note that time points were chosen immediately prior to when the wild type saturated. The growth penalty caused by deleting the Gal4-binding site significantly decreased as the galactose concentrations decreased (P = 2.4e-3, n = 4, Jonckheere–Terpstra test). The error bars in B and C are standard deviations.

Discussion

Cis-Regulatory Tuning of Gene Expression and Metabolic Flux during Evolution

In summary, we have shown that, unlike the model yeast S. cerevisiae, several yeast species contain direct regulatory connections between Gal4 and the metabolic bottleneck gene PGM1/2. Galactose-mediated induction of PGM1/2 is required for vigorous growth in at least two yeast species separated by about 100 My of evolution. Moreover, up-regulation of PGM1/2 alone is sufficient to increase galactose growth in multiple species. The addition of Gal4 regulation to PGM1/2 provides a way to specifically increase galactose metabolism during evolution. Across the family Saccharomycetaceae, the number of Gal4-binding sites upstream of PGM1/2 is one of the best predictors of how vigorously a species grows on galactose. In addition to the well-established link between qualitative differences in galactose metabolism and presence/absence polymorphisms in the GAL network (Hittinger et al. 2004; Riley et al. 2016), we propose that variation in the number of Gal4-binding sites upstream of PGM1/2 quantitatively tunes flux through the metabolic pathway in a condition-dependent manner (fig. 6). The repeated gains and losses of the Gal4-PGM1/2 regulatory connection may have been driven by variation in galactose availability across yeast ecological niches, leading to evolutionary changes through genetic drift, adaptation, or both.

Fig. 6.

The solutions to tune GAL network activities are limited by pleiotropic constraints on a metabolic bottleneck that integrates flux from multiple pathways. Metabolic flux from the Leloir/GAL pathway to glycolysis is controlled by the activity of the bottleneck gene(s) PGM1/2, which encode enzyme(s) that catalyze the interconversion of glucose-1-phosphate and glucose-6-phosphate. Unlike the rest of the GAL pathway, the activity encoded by PGM1/2 is evolutionarily constrained by at least four converging metabolic pathways, including the GAL pathway, glycogen biosynthesis, trehalose biosynthesis, and pentose phosphate pathway (PPP). The left panel shows the scenario without the Gal4-binding sites, such as in S. cerevisiae and V. polyspora. The right panel shows how a “cis dial” regulating PGM1/2 (with a varying number of Gal4-binding sites) is able to bypass the pleiotropic constraints to specifically increase flux from the GAL pathway to glycolysis in response to galactose, such as in S. uvarum and L. kluyveri.

Mutations leading to transcription factor binding site gains and losses are common (Stone and Wray 2001), raising the possibility that the Gal4-PGM1/2 regulatory connection is evolutionarily labile primarily because regulatory state changes are so easy to achieve. Nonetheless, except for PGM1/2, the Gal4-binding sites upstream of GAL genes are well conserved within the genus Saccharomyces (Cliften et al. 2003; Kellis et al. 2003), and our broader analyses extend this trend across the family Saccharomycetaceae (supplementary table S1, Supplementary Material online). Thus, PGM1/2 expression has been coupled or uncoupled from Gal4-regulation multiple times to a different degree than the canonical GAL genes. Other authors have proposed that enzymes controlling flux, including those at beginnings and intersections of metabolic pathways, are more likely to be under strong selective pressures (Flowers et al. 2007; Wright and Rausher 2010; Olson-Manning et al. 2013). Since Pgm1/2 controls the entry point from the Leloir pathway into glycolysis and enzymes acting on glucose-6-phosphate have previously been suggested to be targets of positive selection in Drosophila (Flowers et al. 2007), alterations to the expression or activity of this specific metabolic bottleneck may be particularly likely to affect phenotype.

Despite the importance of direct regulation of PGM1/2 by Gal4 in many species, other mechanisms to modify GAL network activity also exist. For example, galactose-mediated induction of PGM2 occurs in S. cerevisiae through a mechanism that is still undetermined (Oh and Hopper 1990). Our data suggest that a similar mechanism may exist in S. kudriavzevii. We reasoned that the combination of PGM1/2 basal expression and Gal4-independent induction may be sufficient to support low-to-moderate growth rates, perhaps because the metabolic activities upstream of phosphoglucomutase are lower such that phosphoglucomutase activity is not limiting (Hittinger et al. 2010). Thus, the benefit of direct induction of PGM1/2 may be strongest in cases, such as S. uvarum, where upstream network activities are already quite high (Kuang et al. 2016; Roop et al. 2016). Intriguingly, high-flux GAL networks, with the novel Gal4-PGM1/2 feedforward loop characterized here, also tend to have retained both copies of the duplicate genes encoding homologs of the Gal80 corepressors (Tetrapisispora blattae, Naumovozyma castellii, Naumovozyma dairenensis, and S. uvarum). These dual corepressors may lead to a more robust negative feedback loop that prevents the previously characterized phenomenon of metabolic overload (Kuang et al. 2016) (table 2 and supplementary table S1, Supplementary Material online). Thus, multiple genetic changes likely coordinate with the cis-regulatory changes in PGM1/2 to quantitatively tune GAL network activity.

Table 2.

The Three Species Growing Fastest on Galactose have the Highest Number of Predicted Gal4-Binding Sites and Retain the Corepressor Pair Encoded by GAL80/80B.

Rank of Galactose Growth in This Family	Species	Number of Predicted Gal4-Binding Sites Upstream of PGM1/2	GAL80 and GAL80B
1	Tetrapisispora blattae	3	Both
2	Naumovozyma dairenensis	2	Both
3	Saccharomyces uvarum	2	Both
4	Naumovozyma castellii	1	GAL80 only
5	Lachancea kluyveri	1	GAL80 only

Pleiotropic Constraints, Network Architectures, and the Predictability of Evolution

The dynamic evolution of the Gal4-PGM1/2 regulatory connection implies that the possible ways that this bottleneck gene responds to selection are constrained by pleiotropy. The growth rate on galactose is constrained by the activities of both the Leloir pathway and the phosphoglucomutase bottleneck. The enzymes of galactose catabolism are conserved from bacteria to yeasts to humans, including the PGM1 homologs (Lu and Kleckner 1994). Unlike the rest of the GAL pathway, PGM1 homologs are also involved in the pentose phosphate pathway (Cherry et al. 2012), glycogen biosynthesis (Hirata et al. 2003; Cherry et al. 2012), and trehalose biosynthesis (Mulet et al. 2004; Cherry et al. 2012) (fig. 6). Consistent with the hypothesis that the protein-coding sequence of PGM1 is pleiotropic, yeast species that have lost Gal4-PGM1 regulation have experienced relaxed selection on the protein-coding sequences of their dedicated GAL genes but not on their PGM1 homologs (table 1). Even in yeast species that have lost dedicated GAL pathway genes and cannot utilize galactose, PGM1 homologs are retained (Hittinger et al. 2004; Hittinger et al. 2010; Riley et al. 2016). The intersection of these metabolic pathways at the step controlled by PGM1/2 likely constrains the flux of the entire GAL pathway. Although there are many potential ways to modify the bottleneck activity, such as by modifying basal expression or coding sequence changes in PGM1/2, recruiting a Gal4-binding site to specifically induce PGM1/2 expression in response to galactose would increase flux through the GAL pathway, yet minimize the pleiotropic effects on other pathways in different environmental contexts.

If one envisions the rate of galactose growth as a continuous spectrum, there are many ways to marginally increase or decrease galactose growth during evolution. However, we propose that there are relatively few alternatives and many constraints to evolving a highly active GAL network. Up-regulating the bottleneck activity controlled by phosphoglucomutase through direct regulation of PGM1/2 by Gal4 provides a conditional way to increase expression on galactose without pleiotropic effects on other carbon sources (fig. 6). It is likely that changes in other GAL genes and interacting pathways are also involved, but our analyses (supplementary table S1, Supplementary Material online) argue these changes are less repeatable, and therefore, less predictable than the novel Gal4-PGM1/2 regulatory connection that we have characterized here. Thus, we propose that the architecture of the GAL network and the pleiotropy of the metabolic bottleneck gene PGM1/2 constrain the possible outcomes and lead to the repeated evolutionary mechanisms observed (fig. 6). We further hypothesize that greater constraints may lead to higher predictability more generally. Nonetheless, the high likelihood of cis-regulatory mutations means that genetic drift probably also contributes to the repeated variation observed at this locus.

In other conserved gene networks regulating metabolism, important genes with pleiotropic roles may also display predictable evolutionary patterns for reasons that are analogous to the spatial and temporal constraints imposed by developmental regulatory networks (Carroll 2005; Carroll 2008; Stern and Orgogozo 2008; Rebeiz et al. 2009; Stern 2013). Under this model, metabolic genes that handle flux from multiple pathways would be particularly likely to resolve conflicts between selective forces through cis-regulatory changes that enable environmentally specific responses. Indeed, decision points that integrate signals from multiple developmental pathways have been referred to as “bottleneck genes” and argued to be frequent targets of cis-regulatory changes (Stern and Orgogozo 2008). Similarly, we argue that metabolic genes that encode enzymes that are highly interconnected and reused by multiple pathways fall under constraints that favor cis-regulatory tuning of gene expression.

Materials and Methods

Strain Construction

All strains used this study are listed in supplementary table S2, Supplementary Material online. Two genetic engineering approaches were used to introduce point mutations to mutate the predicted Gal4-binding sites: traditional methods and a CRISPR/Cas9-based approach potentially generalizable to the family Saccharomycetaceae and beyond:

Traditional methods (S. uvarum and S. kudriavzevii)

We first deleted part of the promoter containing the predicted binding sites using a selectable and counterselectable marker. After removing the wild-type sequence with the marker gene, we then replaced the marker with sequences containing the desired point mutation that were introduced using PCR primers. Transformations of S. uvarum and S. kudriavzevii were conducted as previously described (Hittinger et al. 2010; Alexander et al. 2014; Kuang et al. 2016).

The SuvaPGM1 expression plasmids were constructed as follows: the S. uvarum PGM1 coding sequence, together with 800 bps upstream and downstream, was inserted into the panARS vector pIL75 (Liachko and Dunham 2014) at the multiple cloning site digested by SmaI. This sequence was assembled using Gibson assembly (Gibson et al. 2009). GFP reporters were constructed as previously described (Kuang et al. 2016). The modified loci of all transformants and constructs were verified by Sanger sequencing.

CRISPR/Cas9 Approach (Lachancea kluyveri and Kluyveromyces lactis)

The backbone of two vectors, a sgRNA expression cassette (GenBank MG680559) and pKOPIS (GenBank MG680557), were synthesized by the DOE Joint Genome Institute DNA Synthesis Science Program. The yeast sgRNA expression cassette contains the SNR52 promoter, HDV ribozyme linked to a cloning site for the sgRNA construct, and the SNR52-1 terminator. pKOPIS contains a KanMX selectable marker and encodes a Cas9 protein driven by the constitutive RNR2 promoter and codon-optimized for expression in S. cerevisiae. A pXIPHOS-panARS vector (GenBank MG835323) was subsequently constructed from pKOPIS through multiple modifications: adding of an Escherichia coli ampicillin resistance marker, swapping the KanMX marker with the NatMX marker, and swapping the 2-µ origin with an autonomously replicating sequence (ARS) that is stable in several yeast genera and was cloned from the panARS vector pIL75 (Liachko and Dunham 2014). Through Gibson assembly, a single final vector was assembled from pXIPHOS-panARS by inserting both a target-specific sgRNA that was amplified from the sgRNA cassette and a repair template generated by PCR next to NatMX. This final vector thus encoded Cas9, a custom sgRNA, and the repair template needed for the gene-editing event. Vectors were electroporated into Lachancea kluyveri and Kluyveromyces lactis following previously described methods (Gojkovic et al. 2000; Kooistra et al. 2004).

Media and Growth Assays

Strains were inoculated from frozen glycerol stocks into either synthetic complete (SC) medium plus 0.2% glucose (1.72 g/l yeast nitrogen base without amino acids, 5 g/l ammonium sulfate, 2 g/l complete dropout mix, 2 g/l glucose) or YPD (10 g/l yeast extract, 20 g/l peptone, 20 g/l glucose; fig. 2 and3E only) and precultured for 2–3 days. The growth assays were conducted in the stated media as previously described (Kuang et al. 2016). Briefly, the absorbance of each well was read by an unshaken BMG FLUOstar Omega plate reader every 10–20 min at 595 nm. “Relative growth” in each figure denotes the number of cell divisions after 15 h or at the indicated time point, which was calculated as log₂[(OD_strain − OD_media)/(OD_start − OD_media)]. This equation normalizes the optical density at each time point (OD_strain) by the starting optical density (OD_start) of that culture and the optical density of the medium (OD_media). In some cases (stated in the Y-axis label), relative growth was normalized to a no carbon source control: the division number was first calculated separately for the same strains cultured in media with or without carbon source in the same 96-well plate, and the division number was then calculated as Division_{carbon source} − Division_{no carbon source}. Strains from each species were tested at either room temperature or 30˚C to determine their preferred growth temperatures, unless their temperature preferences were already well known. In each case, each species was cultured at its expected growth temperature (room temperature or 30˚C), except when species with different optimal temperatures were cultured in the same 96-well plate. In these cases (figs. 2 and 4; supplementary figs. S1, S4B and C, S5, and S6, Supplementary Material online), strains were grown at room temperature (22–24°C). Replicates were defined as either biological replicates that were independent transformants or the same strains started from independent precultures, whereas technical replicates were where the same preculture was inoculated into independent growth assays.

Fluorescence Measurement Assays

Strains were cultured as described above. Both fluorescence levels and absorbance of each well were measured by an unshaken BMG FLUOstar Omega plate reader every 10–20 min, with the excitation filter at 485 nm, emission filter at 520 nm, and absorbance at 595 nm. The nonfluorescent wild-type strain was included as a control to correct for auto-fluorescence. The auto-fluorescence levels were first subtracted from the measured fluorescence levels, which were then normalized to absorbance to control for cell density variation. Replicates were defined the same as in the above growth assays.

Relaxed Selection Analysis

Nucleotide sequences of the GAL1, GAL7, GAL10, and PGM1 genes were extracted from every characterized Lachancea species (except Lachancea waltii, which lost its entire GAL network [Hittinger et al. 2004]) and three outgroup Kluyveromyces species (K. lactis, Kluyveromyces marxianus, and Kluyveromyces dobzhanskii). In cases where duplicate genes were found, all gene copies were analyzed. All sequences were used to obtain phylogeny-aware alignments with PRANK v150803 (Loytynoja 2014) run in the codon mode. Codon alignments were then used to reconstruct maximum-likelihood (ML) phylogenies with RAxML v8.2.10 (Stamatakis 2014), using the GTR model with evolutionary rate heterogeneity modeled by the gamma distribution, ML estimates of base frequencies, and 100 bootstrap pseudoreplicates. Tips and branches shared by species that showed poor growth on 2% galactose were marked as test branches. Finally, both the marked phylogenies and codon alignments were used together to run the RELAX module (Wertheim et al. 2015) implemented in the HYPHY package v2.220170606beta (Pond et al. 2005) to fit descriptive models and run the test for relaxed selection.

Ancestral State Reconstruction

Reconstruction of ancestral states was performed by first scoring each taxon shown in figure 2 for the absence (scored as 0) or presence (scored as 1) of predicted Gal4-binding sites upstream of PGM1/2. We then compared the Equal Rates and the All Rates Different models of discrete character evolution using the R packages Geiger v2.0.6 (Harmon et al. 2008) and phytools v0.4.56 (Revell 2012) to determine the best-fitting model based on the Corrected Akaike Information Criterion (Akaike 1974). Finally, we simulated 1,000 stochastic character maps on the phylogeny under the best-fitting model using stochastic mutational mapping (Bollback 2006) and obtained the posterior probability (PP) of each character state at each internode of the phylogeny.

Statistical Analysis

All P values were two-sided and calculated using Welch’s two-sample t-test (fig. 2C and supplementary figs. S1C and 4B and C, Supplementary Material online) or a conservative nonparametric test. Specifically, we used a Wilcoxon rank sum test that allows the rank data from multiple independent experiments to be pooled to account for day-to-day variation without making assumptions about the variance. To take into account the effects of decreasing galactose concentrations (fig. 5B and C, Supplementary Material online), we used the ordered Jonckheere–Terpstra test (two-sided), and P values from independent experiments were subsequently combined using Fisher’s method. These tests were performed using Mstat software version 6.1.4 (http://mcardle.oncology.wisc.edu/mstat/; last accessed May 21, 2018). For all boxplots, the elements are defined as follows: the bottom and top of the box are the lower and upper quartiles, respectively; the band within the box is the median; and the lower and upper whiskers represent 1.5 interquartile ranges.

Supplementary Material

Supplementary data are available at Molecular Biology and Evolution online.

Author Contributions

M.C.K. and C.T.H. designed the experiments and wrote the paper. M.C.K. conducted the experiments and performed most analyses. J.K. conducted the relaxed selection and ancestral state reconstruction analyses. W.G.A., J.-F.C., R.L.W., M.C.K., and C.T.H. designed and constructed the panARS-CRISPR/Cas9 system.