Dynamic Changes in Yeast Phosphatase Families Allow for Specialization in Phosphate and Thiamine Starvation

Convergent evolution is often due to selective pressures generating a similar phenotype. We observe relatively recent duplications in a spectrum of Saccharomycetaceae yeast species resulting in multiple phosphatases that are regulated by different nutrient conditions – thiamine and phosphate starvation. This specialization is both transcriptional and at the level of phosphatase substrate specificity. In Candida glabrata, loss of the ancestral phosphatase family was compensated by the co-option of a different histidine phosphatase family with three paralogs. Using RNA-seq and functional assays, we identify one of these paralogs, CgPMU3, as a thiamine phosphatase. We further determine that the 81% identical paralog CgPMU2 does not encode thiamine phosphatase activity; however, both are capable of cleaving the phosphatase substrate, 1-napthyl-phosphate. We functionally demonstrate that members of this family evolved novel enzymatic functions for phosphate and thiamine starvation, and are regulated transcriptionally by either nutrient condition, and observe similar trends in other yeast species. This independent, parallel evolution involving two different families of histidine phosphatases suggests that there were likely similar selective pressures on multiple yeast species to recycle thiamine and phosphate. In this work, we focused on duplication and specialization, but there is also repeated loss of phosphatases, indicating that the expansion and contraction of the phosphatase family is dynamic in many Ascomycetes. The dynamic evolution of the phosphatase gene families is perhaps just one example of how gene duplication, co-option, and transcriptional and functional specialization together allow species to adapt to their environment with existing genetic resources.

, and cyclins (Wolfe and Shields 1997;Carroll and O'Shea 2002). In some cases, duplications and/or specializations exhibit convergence -i.e. the same characteristics in divergent species that were likely not in the ancestor, and thus, arose independently (Stern 2013). For example, S. cerevisiae and S. pombe each have three mating type gene cassettes that are understood to have arisen from independent gene duplication events, suggesting that this genetic architecture was adaptive (Klar et al. 2014;Hanson and Wolfe 2017;Dujon and Louis 2017).
During our studies with Candida glabrata, we identified a threegene family of phosphatases (CgPMU1-3) that encodes phosphatases of differing enzyme specificity (Orkwis et al. 2010;Orlando et al. 2015). We previously identified CgPMU2 as the phosphate-starvation inducible acid phosphatase gene that is analogous to the PHO5 gene in S. cerevisiae (Orkwis et al. 2010;Kerwin and Wykoff 2012). C. glabrata does not contain any homologs of ScPHO5 based on sequence similarity. CgPMU1 appears to encode a narrow-range phosphatase, which is likely similar to the ancestral gene conserved in most Ascomycetes. CgPmu3 appears to have some functions in common with CgPmu2, but it is unknown what selective pressures led to the preservation of CgPMU3. ScPHO5 also has a number of paralogs in S. cerevisiae, including ScPHO3, which has been suggested to be important for thiamine recycling through its activity as a thiamine phosphatase (Nosaka et al. 1989b(Nosaka et al. , 1989a. This led us to explore thiamine metabolism in C. glabrata. Thiamine and its metabolically active form, thiamine pyrophosphate (or TPP), are essential for a number of core intermediary carbon metabolism reactions (Gibson et al. 2016). TPP allows for decarboxylation of pyruvate, and was originally isolated as a cofactor for pyruvate dehydrogenase (Lipschitz et al. 1938). In S. cerevisiae, thiamine starvation activates the transcription factor ScThi3, which, together with its transcriptional co-activators, ScThi2 and ScPdc2, induces the expression of many genes related to the biosynthesis of thiamine. C. glabrata has subtle differences in how it synthesizes, recycles, or transports thiamine relative to S. cerevisiae (Iosue et al. 2016). C. glabrata lacks the Thi2 transcriptional coactivator in S. cerevisiae, lacks the ability to synthesize a pyrimidine precursor for thiamine synthesis, and in standard laboratory medium is a thiamine and pyridoxine (vitamin B 6 ) auxotroph (Iosue et al. 2016). However, both species have many of the same biosynthetic genes, and those genes are regulated by the transcription factors, Pdc2 and Thi3. These subtle alterations are likely influenced by the environments of the two species (Gabaldón Estevan et al. 2013;Gabaldón et al. 2016).
We performed RNA-seq on C. glabrata cells to identify the genes that are regulated by thiamine starvation and identified CgPMU3 as a highly induced gene during thiamine starvation. The identification of an analogous system to S. cerevisiae in C. glabrata, where there is a phosphatase that appears to be important for phosphate starvation responses and a paralog that is important for thiamine starvation responses, led us to explore whether there has been convergent evolution in multiple species to duplicate and specialize the phosphatase genes. Indeed, in S. pombe, there are two ScPHO5 related phosphatases encoded by SpPHO1 and SpPHO4, and transcriptional induction of these genes is dependent on different environmental conditions. SpPHO1 is upregulated during phosphate and adenine starvation (Henry et al. 2011;Estill et al. 2015), and SpPHO4 transcription is repressed by addition of thiamine to the medium (Yang and Schweingruber 1990;Yang et al. 1991). Given the similarities in behavior of paralogs, we characterized this behavior more fully to understand how flexible the evolutionary architecture for generating two phosphatases that are tailored to two different environmental conditions was.

MATERIALS AND METHODS
RNA-seq C. glabrata wild-type and Cgthi3D strains (Table S2) were grown in thiamine replete conditions at 30°to logarithmic growth phase (OD 600 $0.2-0.5). Cells were harvested by centrifugation, washed three times with water, and transferred to SD medium (Sunrise Science, CA) with thiamine (thiamine replete: 0.4 mg/L) and without thiamine (thiamine starvation) and grown at 30°. C. glabrata wild-type was grown for 2 and 4 hr while the Cgthi3D strain was grown for 4 hr. This resulted in six samples: C. glabrata wild-type at 2 h, C. glabrata wild-type at 4 h, and Cgthi3D at 4 h, with thiamine replete and starvation conditions for all. RNA was harvested from these six samples for next generation sequencing. Eurofins Genomics generated an Illumina specific library for sequencing, performing 2x50 bp sequencing on a HiSeq2000, and yielding 92 million reads for the six samples. We utilized the Geospiza (Perkin Elmer) bioinformatics suite to perform analyses after reads per kilobase per million (rpkm) were grouped to NCBI annotated coding sequences of the C. glabrata CBS138 genome. Two percent of genes change expression twofold in response to thiamine starvation (in either 2h or 4h). An .xls file is included (Table S1) with the expression (rpm) per gene both logarithm transformed and as rpkm counts. The raw FASTQ files are submitted to NCBI under accession SRP131893.

Flow Cytometry
To assay induction of C. glabrata PMU2 and PMU3 using flow cytometry, we constructed plasmids where the full-length promoters of these genes (3 Kb CgPMU2p and 1 Kb CgPMU3p) were driving expression of YFP. The promoter sequences were amplified by PCR (Table S3) then repaired into a plasmid containing YFP by homologous recombination (Corrigan et al. 2013). These plasmids were transformed into C. glabrata wild-type, Cgthi3D (Iosue et al. 2016), and Cgpho4D (Kerwin and Wykoff 2009) strains (Table S2). Cells were grown at 30°in thiamine/phosphate replete medium to logarithmic growth phase (OD 600 $0.2-0.5). Cells were harvested by centrifugation, washed 3 times with water, inoculated into thiamine replete (0.4 mg/L) and starvation and phosphate replete (1 g/L) and starvation conditions, and grown at 30°overnight ($18 h). Fluorescence (in arbitrary units) of each strain was measured using a flow cytometer with a 533/30 FL1 filter set (Accuri C6, BD Biosciences).

Phylogenetic Analysis
We downloaded PHO5 homologs from 23 species (Wapinski et al. 2007). These species represent all major clades of the Ascomycota phylum, including the Saccharomycotina (budding yeast), Pezizomycotina (filamentous fungi), and Schizosaccharomycetes (fission yeast), although sampling is biased toward the first group, with the latter two being represented by just two and three species, respectively. To identify any homologs missing from the Fungal Orthogroup Repository, we performed BLAST searches on the following online databases: Saccharomyces Genome Database, Candida Genome Database, and Genome Resources for Yeast Chromosomes. In addition, we used HMMER webserver (https://www.ebi.ac.uk/Tools/hmmer/) to search against the Reference Proteomes to leverage the high sensitivity of the HMM-based tool for identifying distant homologs (Finn et al. 2015). For three species in the sensu stricto group (S. paradoxus, S. mikatae, S. bayanus), we downloaded the annotated FASTA file containing gene features from the Saccharomyces sensu stricto genome website (Scannell et al. 2011), and performed local BLAST searches. In total, we identified 53 putative PHO5 homologs in the 23 species. The number of homologs per species ranges between zero (C. glabrata, N. castellii, and E. gossypii) to five (K. lactis and S. cerevisiae), with a median of 2 and a mean of 2.3 per species.
The amino acid sequences were aligned using ProbCons with the following parameters: 2 consistency reps, 1000 iterative refinement reps, 0 pre-training reps (Do et al. 2005). Protein phylogeny was reconstructed using PhyML website service (v3.0), with the aligned protein sequences as input and the following parameters: AIC for model selection, using BIONJ to construct the initial tree, SPR for tree improvement, no random starting tree used, 1000 bootstraps for Figure 3 and only 100 bootstraps for Figure S3, due to the much larger sample size (Guindon et al. 2010).
For Figure S3, we additionally performed gene-tree and species-tree reconciliation using an algorithm-based approach implemented in Notung (v2.9) (Durand et al. 2006). We first loaded the gene tree from the last step and a species tree based on a 1233-gene data matrix from (Shen et al. 2016). Reconciliation was performed with the default settings. The resulting gene tree was rooted by selecting the branch that minimized the total event score, following the software's recommendation. Finally, the gene tree was "rearranged" using the Rearrangement function, which swaps weakly supported branches to identify the alternative topology that minimizes the total event score. We followed the software's default setting to define weakly supported branches as those with less than 90/100 bootstrap values. This reduced the total number of duplications and losses from 29/48 to 23/15. Figure  S3 shows the reconciled and rearranged gene tree with inferred duplication and loss events.
qPCR Assay Cells were grown and washed as described for flow cytometry but inoculated into thiamine/phosphate replete, thiamine starvation, and phosphate starvation conditions for 4 h at 30°. RNA was extracted with the Direct-zol RNA Miniprep Plus kit (Zymo Research) and reverse transcribed to cDNA using the iScript cDNA synthesis kit (Bio-Rad). cDNA was amplified in a 25 mL reaction using the Sso Advanced Universal SYBR Green Supermix (Bio-Rad) with a CFX quantitative PCR machine (Bio-Rad). Primers were designed to amplify phosphatase genes and ACT1 in various species (Table S3). Transcript amounts for each gene were normalized to ACT1, the expression of which does not change in different thiamine and phosphate conditions (Kerwin and Wykoff 2009;Iosue et al. 2016). Ten-fold genomic DNA dilutions were also amplified with each primer set as amplification controls.
Hydrolysis of TPP, PNPP, and 1-Napthyl-Phosphate To assay hydrolysis of TPP in various yeasts, C. glabrata PMU3 was deleted and precisely replaced with phosphatase genes using a CgURA3 selection/counter selection scheme with 5-FOA. Multiple transformants were confirmed by PCR and tested in assays to verify that the observed phenotype was not an outlier. Strains (Table S2) were grown overnight in thiamine/TPP replete conditions. Cells were harvested and washed as described for flow cytometry and inoculated at a low density (OD 600 = 0.001) into TPP replete (30 mg/L), thiamine replete, and thiamine/TPP starvation conditions and grown at 30°for 24 h. Cell density (OD 600 ) was measured to indirectly assay the ability to hydrolyze TPP.
Because the CgPMU3 promoter does not highly express genes under phosphate starvation, we cloned the phosphatase ORFs under the control of a phosphate-regulated promoter (ScPHO5) (Orkwis et al. 2010). The same genes from the TPP assay were recombined onto a ScPHO5 promoted plasmid through gap repair and transformed into Cgpmu2D (Corrigan et al. 2013). CgPMU2 was deleted to reduce background phosphatase activity. Strains (Table S2) were grown overnight in phosphate replete conditions, then cells were harvested and washed as for flow cytometry. Cells were inoculated into phosphate starvation conditions at a cell density of OD 600 $0.1 and grown for $18 h. Cells were assayed for PNPP hydrolysis as previously described in (Kerwin and Wykoff 2009;Orkwis et al. 2010). Cells growing on agar plates lacking phosphate were assayed for hydrolysis of 1-napthyl phosphate as described in (Kerwin and Wykoff 2009;Orkwis et al. 2010

RNA-seq of Thiamine Starved C. glabrata Identifies CgPMU3 as a Highly Induced Phosphatase
To understand the transcriptional response of C. glabrata to thiamine starvation, we performed RNA-seq analysis on cells that were grown in medium either containing or lacking thiamine and identified genes up-regulated by thiamine starvation that were dependent on the transcriptional regulator CgTHI3 ( Figure 1). There is large overlap with the orthologous genes in S. cerevisiae, when examining a microarray of thiamine starvation in S. cerevisiae (Nosaka et al. 2005), including THI4, PET18, THI20, THI6, and THI10 (Table S1), and we confirmed a statistically significant increase in these transcripts during thiamine starvation using qPCR on separately grown samples in triplicate ( Figure  S1). However, numerous genes induced in S. cerevisiae (SNO/SNZ/ THI5/THI2/PHO3) lack orthologs in the C. glabrata genome. To determine any previously unidentified genes that might serve an analogous function in C. glabrata as in S. cerevisiae, we examined genes that were induced during thiamine starvation but did not have a known role in thiamine metabolism in C. glabrata. We noted that there was CgTHI3-dependent induction of CgPMU2 and CgPMU3 during thiamine starvation. We had previously identified CgPMU2 as a ScPHO5 analog, which is highly induced during phosphate starvation (Orkwis et al. 2010). Compared to CgPMU2, CgPMU3 is induced more strongly and at an earlier time point (Figure 1). Given this observation, we hypothesize that CgPMU3 functionally replaced the PHO3 gene, which is missing from the C. glabrata genome.

CgPMU2 and CgPMU3 Are Regulated By Two Different Environmental Conditions
To confirm the transcriptional regulation that we observed in the RNAseq experiment with CgPMU2 and CgPMU3, we constructed CgPMU2and CgPMU3-promoted YFP plasmids and assessed their ability to recapitulate phosphate-(PMU2) and thiamine-(PMU3) repression of YFP expression in cells (Figure 2). We also tested expression of each promoter-YFP construct in the alternate repressing condition. CgPMU3 is tightly regulated by thiamine conditions (phosphate starvation has little effect), whereas CgPMU2 is de-repressed by both thiamine and phosphate starvation. Additionally, the induction of phosphatases in the two conditions requires the appropriate transcription factor -i.e. CgThi3 regulates thiamine starvation induction and CgPho4 regulates phosphate starvation induction. It is worth noting that we see some defect in the appropriate induction of a phosphatase when the other transcription factor is absente.g., CgPMU2-YFP is not fully induced during phosphate starvation in the Cgthi3D strain, suggesting some potential cross-talk between the transcription factors. Additionally, because the PMU1 gene family is in tandem in the C. glabrata genome, one might argue that this cross-talk is a consequence of relaxation of chromatin structure. However, we eliminated this possibility because the plasmids used did not contain any part of each other's promoters, and these plasmids are likely not exposed to the same chromatin structure as in the genome. We conclude that CgPMU3 transcription is tightly regulated by thiamine concentration, and CgPMU2 is primarily induced during phosphate starvation, but partially upregulated in response to thiamine starvation.

Repeated Expansion of Gene Families Encoding Phosphatases in the Saccharomycetaceae
Despite similar specialization of CgPMU2 and CgPMU3 vs. ScPHO5 and ScPHO3, the PHO5-related genes are not evolutionarily related to the PMU1-related genes (Orlando et al. 2015). Therefore, the gene family expansion and the subsequent specialization among the genes occurred independently in these two species. This led us to hypothesize that such expansion and specialization of phosphatase-encoding genes has occurred repeatedly during the evolution of this group of diverse yeasts.
To test this hypothesis, we reconstructed the evolutionary history of the PHO5 gene family in four Ascomycota yeasts that span a wide range of evolutionary distance, and which contain multiple members of the PHO5 family. Three of the four species belong to the class of Saccharomycetes, including S. cerevisiae, S. mikatae, and K. lactis; the fourth species is S. pombe, which comes from the distantly related class of Schizosaccharomycetes, i.e., the fission yeast. Under our hypothesis, there should be repeated, independent duplications that happened after speciation. To infer the relative timing of duplication to speciation, we reconstructed the gene tree for the PHO5 homologs in these four species and compared it to the species tree. In a cartoon example with two paralogs in two extant species-designating the two paralogs as A and B, and the two species as 1 and 2-if gene duplication preceded speciation, gene 1A and 2A would cluster before they are joined by the cluster of 1B and 2B ( Figure 3A). Conversely, if gene duplication occurred after speciation, 1A and 1B would coalesce first, and so would 2A and 2B, before the two clusters would coalesce ( Figure 3B).
The maximum likelihood tree for PHO5 family genes in these species strongly supports independent duplication in S. pombe, K. lactis, and the sensu stricto clade, which includes S. cerevisiae ( Figure 3C). Within the closely related sensu stricto clade, which is estimated to have speciated in the past $20 million years, the protein sequence divergence between homologs was often too close for confident reconstruction of the phylogeny, as shown by the low bootstrap values in some of the internal nodes. Nonetheless, there is evidence for recent duplications. Examining the syntenic relationships, e.g., for the genomic region that contains ScPHO5 and ScPHO3 arranged in tandem, we found that the syntenic region in S. mikatae contains three genes instead of two, one of which, S.mik|2.235, appears to have no orthologs in other species, suggesting a recent origin in that species ( Figure S2). Similarly, ScPHO11 and ScPHO12 appear to have no syntenic orthologs in the other sensu stricto species.
We next asked if this pattern of repeated duplications post speciation is general across the Ascomycota fungi. To do this, we extended our analysis to 53 homologs of PHO5 in 23 Ascomycota yeasts (Wapinski et al. 2007;Gabaldón et al. 2016;Shen et al. 2016). For this large dataset, we used an algorithm-based approach to reconcile the gene tree with the species tree, allowing for duplications and losses. The result revealed extensive duplication and loss events not just at the tip of the tree, but rather throughout the species phylogeny ( Figure S3). There were both ancient and recent duplications, accompanied by multiple loss events, revealing an incredibly dynamic history of the phosphatase gene family. In conclusion, phylogenetic analysis confirmed our hypothesis that the PHO5 family has experienced repeated duplication (and loss) in multiple yeast species in the Ascomycota phylum.

Multiple Yeast Species Specialize the Transcription of Phosphatase Genes
Our data in combination with the phylogenetic analysis raised the question of whether multiple yeast species have this same specialization behavior -i.e. a phosphatase repressed by thiamine and a phosphatase repressed by phosphate. Others have observed similar partitioning of phosphatases in S. cerevisiae and S. pombe, but none have directly compared the two nutrient conditions side-by-side (Yang and Schweingruber 1990;Henry et al. 2011).
To understand the transcriptional regulation of duplicate phosphatase-encoding genes in response to thiamine and phosphate starvation, we used qPCR of reverse-transcribed RNA from cells grown in nutrient Figure 1 RNA-seq of thiamine-starved C. glabrata. (A) Normalized abundance of transcripts plotted for wild-type C. glabrata and a Cgthi3D strain that were grown for 4 h in thiamine starvation relative to cells grown in thiamine replete medium (strains grown in singlicate for this experiment). Transcripts that increased in abundance during starvation are visualized as circles that are above the majority of the genes. The seven most highly induced genes are CgTHI3-dependent. (B) A list of the most highly induced genes during thiamine starvation and their fold induction relative to the thiamine replete sample. Wildtype cells were grown for 2 h and 4 h to investigate the timing of induction; however, only the 4 h time point is presented in part (A).
replete, thiamine-starved, and phosphate-starved conditions, and assessed the abundance of phosphatase transcripts in S. cerevisiae, C. glabrata, and S. pombe (Figure 4). We find there is tight regulation of the phosphate-repressible phosphatases. Only ScPHO5, CgPMU2, and SpPHO1 are highly induced by phosphate starvation. The thiamine-regulated paralogs show no induction during phosphate starvation (Figure 4, Right). Gene induction during thiamine starvation is not as tightly regulated. The thiamine-repressible phosphatases, ScPHO3, CgPMU3 and SpPHO4 are induced (note the log scale in the x-axis) during thiamine starvation. However, CgPMU2 and SpPHO1 show some induction as well (Figure 4, Left), and only S. cerevisiae and C. glabrata have statistically significant differential paralog regulation, although in S. pombe the trend is present (but not statistically significant at a p value of 0.09). In general, we observe some repression of one paralog by thiamine, and another paralog strongly by phosphate, supporting our hypothesis that multiple yeast species have partitioned the transcription of one phosphatase for each condition.

Substrate Specialization in Both the PHO5 and PMU1 Phosphatase Protein Families
Since we observed specialization of the phosphatases in terms of initiation of transcription, we explored whether the phosphatases had specialized their activity toward substrates -i.e. are the phosphatases that are induced during thiamine starvation well-tailored to TPP as a substrate? When cells die, they often release the active cofactor TPP, and in order for the thiamine to be accessed it must be hydrolyzed to thiamine, as thiamine transporters appear to be unable to transport TPP (Nosaka et al. 1989a). We expected that if CgPMU3 encodes a phosphatase required for the hydrolysis of TPP to thiamine, a Cgpmu3D strain might not utilize TPP as a thiamine substrate. Because C. glabrata behaves as a thiamine auxotroph in standard growth medium, we expected that a lack of TPP hydrolysis to release thiamine could limit growth, as measured by optical density. Indeed, deletion of CgPMU3 led to a defect in growth when TPP was supplied as the sole thiamine source, whereas deletion of CgPMU2 did not ( Figure 5). We introduced plasmids containing the phosphatases under the control of a low-level, non-thiamine regulated promoter (Orkwis et al. 2010). Expressing CgPMU3 in a Cgpmu3D suppressed this defect, but CgPMU2 did not ( Figure 5).
The previous experiment did not eliminate the possibility that CgPMU2, when highly expressed, was capable of hydrolyzing TPP at a low rate which could allow for growth. To test for this possibility, as well as to test the other known phosphatases for the ability to cleave TPP at a physiological level, we precisely deleted the CgPMU3 open reading frame (ORF) with the CgURA3 gene and then subsequently replaced the CgURA3 gene with an ORF of our choosing, through 5-FOA selection. This allowed for the ORF to be regulated by the endogenous CgPMU3 promoter and determined whether high level induction of CgPMU2 would allow for physiological hydrolysis of TPP (Figure 6, Left). CgPMU2 does not encode a phosphatase that recognizes TPP at physiological conditions; thus, CgPmu2 and CgPmu3 have different functional specificities in vivo. As expected, in the PHO5 phosphatase gene family, we observe high activity for TPP with the phosphatase (ScPHO3) that is induced during thiamine starvation.
We were also curious whether the phosphatases might have altered specificity toward other substrates. To test this hypothesis, we measured the activity of the phosphatases toward commonly used laboratory organic phosphates such as p-nitrophenyl phosphate (PNPP) and 1-napthyl phosphate (1-NP) (Orlando et al. 2015) in addition to TPP hydrolysis. We first cloned each phosphatase ORF under the control of the ScPHO5 promoter in a plasmid so that the phosphatase would be highly expressed during phosphate starvation (Orkwis et al. 2010).

Figure 2
CgPMU3 is tightly regulated by thiamine whereas CgPMU2 is regulated by both thiamine and phosphate. Plasmids containing the fulllength promoter of either CgPMU2 or CgPMU3 driving expression of YFP were transformed into C. glabrata wild-type, thi3D, and pho4D strains and assayed for fluorescence using flow cytometry. Error bars represent the standard deviation of three biological replicates. We performed a Student t-test to confirm that CgPMU2 expression declined in the Cgthi3D strain; however, there is still a significant increase in expression relative to the high phosphate conditions. We additionally confirmed no statistical decline in the Cgpho4D strain during thiamine starvation in the CgPMU3p-YFP. The ÃÃ indicates a p value less than 0.001 and NS indicates a p value higher than 0.05. (Left) Strains were grown in thiamine replete (High) and starvation (No) conditions. Both CgPMU2 and CgPMU3 promoters are induced by thiamine starvation and this induction is regulated by CgThi3. (Right) Strains were grown in phosphate replete (High) and starvation (No) conditions. Only the CgPMU2 promoter is induced by phosphate starvation and this induction is regulated by CgPho4.
These plasmids were then transformed into a Cgpmu2D strain to remove native PNPPase activity, as this strain has very low phosphatestarvation phosphatase activity (Orkwis et al. 2010). Because we used the same plasmid, promoter, and growth conditions, we expected that roughly the same amount of protein would be present regardless of ORF. We then measured the ability to cleave phosphate from PNPP in whole cells expressing the various phosphatases (Figure 6, Right). We observed differences between the paralogs of both the PHO5 and PMU1 families. The PMU1 family of phosphatases appears to have tuned phosphatase activity such that the phosphate starvationinduced phosphatase (CgPmu2) is extremely efficient at hydrolyzing PNPP, and the thiamine starvation-induced phosphatase (CgPmu3) is extremely efficient at hydrolyzing TPP. However, in S. cerevisiae, both members of the PHO5 family had significant PNPPase activity, while ScPho3 appears to be a more efficient TPPase.
We also assessed the ability of the cloned phosphatase genes to hydrolyze 1-NP. We have already observed differences in the PMU1 family with regards to these two phosphatase substratesboth CgPmu2 and CgPmu3 are efficient at hydrolysis of 1-NP, but only CgPmu2 can hydrolyze PNPP (Orlando et al. 2015). We do not observe these same differences between paralogs in S. cerevisiae and S. pombe. PNPP and 1-NP have different physical structures, and our results hint that identifying the in vivo biological substrate of a given phosphatase may be difficult, as there may be multiple relevant substrates or the promiscuity of the enzyme for many substrates may prevent strong conclusions.
In S. cerevisiae and S. pombe there does not appear to be a loss of activity against PNPP or 1-NP in the thiamine starvation induced phosphatases (ScPho3 and SpPho4), as we observe in the PMU1 family ( Figure 6, Right). This might be explained by the fact that these two Figure 4 Phosphate-repressible phosphatases are more tightly regulated than thiamine-repressible phosphatases. S. cerevisiae, C. glabrata, and S. pombe wild-type strains were grown in replete, thiamine starvation, and phosphate starvation conditions. qPCR on reverse-transcribed RNA from these strains determined the amount of transcript for various phosphatase genes. Transcript levels were normalized to transcript levels for ACT1, which does not change its expression in response to nutrient conditions, and fold induction was calculated relative to replete for both thiamine (Left) and phosphate (Right) starvation conditions. Error bars represent the standard deviation of the average of fold induction for three independently grown biological replicates. We compared with Student t-tests the two paralogs in each condition, and the p value below 0.05 is indicated with one Ã and 0.001 with ÃÃ . If the p value was above 0.05 it was considered not significant (NS). (Left) Thiamine-repressible phosphatases are not tightly regulated, as the other paralogs show some induction. (Right) Phosphate repressible phosphatases are tightly regulated, as there is no induction of the other paralogs. Figure 5 CgPMU3 encodes a functional thiamine pyrophosphatase (TPPase). C. glabrata wild-type, pmu2D, and pmu3D strains, containing an empty vector, were grown in thiamine/TPP starvation, thiamine replete, and TPP replete conditions. Plasmids containing CgPMU2 or CgPMU3 under the control of the ScPHO5 promoter were introduced into the deletion strains and grown in the same conditions. The ScPHO5 promoter expresses at a low level regardless of thiamine concentration in C. glabrata (Orkwis et al. 2010). Optical density at 600 nm was measured to determine growth. All strains grow to a high density when thiamine is supplied. Cgpmu3D shows a defect when TPP is the sole source of thiamine and only addition of CgPMU3 rescues this defect. Error bars represent the standard deviation of the average of three independently grown biological replicates. The ÃÃ indicate a p value below 0.001 when those samples are compared to wild-type with TPP, and all of the other TPP samples are not statistically significantly different from one another. protein families have different amino acids present in the active site (Orlando et al. 2015), and the PHO5 family is unable to discriminate between the organic compounds attached to the phosphate. Regardless, in S. cerevisiae the activity of the phosphatases against TPP are tailored for the phosphatase that is induced in thiamine starvation (ScPho3) (Figure 6, Left). Interestingly, we observe a different case in S. pombe. Both phosphatases appear to be capable of cleaving TPP, and the phosphate starvation-regulated phosphatase, SpPHO1, appears to be slightly better at cleaving TPP. In fact, the thiamine starvation-regulated SpPHO4 appears to be a better phosphatase against 1-NP and PNPP relative to SpPHO1, which is counter to published data that indicates that SpPHO1 is the phosphate starvation inducible phosphatase (Henry et al. 2011;Estill et al. 2015). However, the phenotype of lack of phosphatase activity in the Sppho1D strain is only observable in a medium mixture that contains thiamine, likely repressing SpPHO4 expression (standard S. pombe medium does not contain thiamine, and the phosphatase assays were performed in a SD/EMM medium mixture) (Henry et al. 2011;Carter-O'Connell et al. 2012). Additionally, the phosphatase genes in S. pombe appear to be expressed at higher levels under repressing conditions (Henry et al. 2011), so S. pombe may have a different approach from the other two species studied. The contrast in approaches taken by the different species to deal with phosphate and thiamine starvation highlights two important points. First, some species' phosphatases have a high degree of specialization at the level of the enzyme, while others do not, and second, because we are only looking at a few laboratory substrates, our results give guidance, but not strong conclusions, with regards to the in vivo functions of individual phosphatases.

The Ancestor of CgPMU2 and CgPMU3 Is Most Similar to CgPMU3
Because CgPMU2 and CgPMU3 are more closely related to one another than to CgPMU1, we wanted to determine the enzyme specificity of the likely ancestor of CgPMU2 and CgPMU3. To test the functional specificity of the ancestor, we generated a CgPmu1 protein where every amino acid that was common to CgPmu2 and CgPmu3 was introduced into the CgPmu1 background ( Figure S4). We posited that common sequences in both CgPmu2 and CgPmu3 were likely to have been common in the ancestor of both. The ancestor appears to resemble CgPmu3 specificity for TPP ( Figure 6). This result suggests that the primary selective condition for maintenance of the expansion of the PMU1 gene family may have been to recycle thiamine as opposed to Figure 6 PHO5 and PMU1 protein family members have different specificities against substrates -some are better TPPases and others are more broad-range organic phosphatases. (Left) The CgPMU3 ORF was deleted and replaced with ORFs of phosphatases from various yeasts so that they are under the control of the CgPMU3 promoter and inducible in growth medium containing TPP as the sole thiamine source. The ability to hydrolyze TPP to thiamine was assayed by measuring growth using optical density at 600 nm. (Right) The phosphatase ORFs were cloned into plasmids under the control of the ScPHO5 promoter, which is highly expressed during phosphate starvation, and these plasmids were introduced into a Cgpmu2D, which has minimal PNPPase activity. Cells were assayed for PNPP hydrolysis, indicated by an increase in OD 400 /OD 600 . (Right, images) The strains containing the plasmids were grown on solid medium lacking phosphate and assayed for 1-napthyl-phosphate hydrolysis, which is indicated by a red color. Activity was assayed after 5 min and 45 min. For the bar graphs, error bars represent the standard deviation of the average of three to six replicates. We compared with Student t-tests the two paralogs in each condition, and a p value below 0.05 is indicated with one Ã and 0.001 with ÃÃ . If the p value was above 0.05 it was considered not significant (NS). scavenge phosphate from organic phosphates; however, because we observe significant 1-NP hydrolase activity for the reconstructed ancestor protein, we cannot exclude recycling of phosphate as a selective force. These data are consistent with the CgPmu2 protein having neofunctionalized relative to its ancestor, and now being able to readily hydrolyze PNPP, losing the ability to hydrolyze TPP.

DISCUSSION
This work demonstrates that two different phosphatase gene families (the PHO5 and PMU1 families) experienced independent duplication events in multiple lineages, and that in both cases, the duplicates specialized at the transcriptional level to be responsive to either thiamine starvation or inorganic phosphate starvation. Additionally, this duplication in some species allowed for phosphatase specificity to be tailored for their respective substrates. An interesting evolutionary question is, under similar environmental pressures (or opportunities), does evolution repeatedly use the same raw genetic materials for adaptation? How predictable are evolution and adaptation? In our study, we found that C. glabrata lost the ancestral PHO5 family and recreated the same functional specialization phenotype by co-opting an expanded PMU1 family. This demonstrates that while the evolution of the phenotype is convergent (duplication of phosphatase genes and specialization), the genetic basis for the phenotype can be different (Stern 2013). In other cases, however, evolved phenotypes can appear to be highly constrained, and thus, predictable (Zhen et al. 2012). In extreme cases, limited evolutionary "solution space" can lead to parallel evolution involving the same amino acid changes in orthologs (Zhen et al. 2012). Our observation of multiple Ascomycota yeasts (e.g., S. cerevisiae, K. lactis, and S. pombe) independently duplicating and specializing the PHO5 family suggests that this family may be the preferred substrate for evolution, but when the PHO5 family is not available (for example, in C. glabrata), other options can be explored. Such dynamism suggests multiple "solutions" to the problem.
Another question is, what might be driving the gene family evolutionary dynamics? Our observation of two different phosphatase families being involved in generating a convergent phenotype suggests the possibility that there is a common selective pressure favoring this specialization, although the multiple losses seem to suggest that the selection for maintaining this specialization is weak or variable across evolutionary timescales (Stern 2013). We cannot exclude the possibility that neutral processes generate this seemingly complex genomic architecture repeatedly (Force et al. 1999), and potentially in some cases, genetic drift may have mimicked this apparent convergence. However, with the PMU1 family, the data are consistent with neofunctionalization, as the activities of CgPmu2 and CgPmu3 have diverged. The genetic architecture (especially with regards to transcriptional regulation) is reminiscent of the adaptive GAL switch in S. cerevisiae (Hittinger and Carroll 2007).
The convergent behavior has noticeable noise -i.e. not every species has tight transcriptional regulation, and the PHO5 phosphatase family appears to have significant TPPase activity even in the phosphate starvation-regulated phosphatases. Because we observe this convergent behavior in sensu stricto species and the distantly related Archaeascomycete S. pombe, it seems possible that the unicellular fungal lifestyle common to these species has imparted a selective advantage to the partitioning of phosphatase expression and activity to a single environmental condition. Because thiamine is critical for energy metabolism, it is appealing to consider that phosphatase duplication and evolution may have been shaped by a need to conserve energy by engaging the most efficient response to recycle thiamine and/or to acquire phosphate from organic phosphate sources. It is worth noting that there are no obvious, clear links between phosphate and thiamine metabolism, as the signaling pathways are different, and there is no clear single point where the two pathways would intersect. There are likely indirect interactions between the two metabolic pathways in that ATP/energy charge may impact the phosphorylation of thiamine, and that energy metabolism is influenced by phosphate concentrations. If duplication and loss happen at a relatively high rate over evolutionary timescales, as Figure S3 suggests, one possibility is that having multiple, specialized copies may only provide a small advantage, and thus, the selective pressure to make the specialization complete at both the transcriptional and enzymatic level may be low.
The two phosphatase families (that share no sequence homology based on BLASTp alignments) appear to have different ancestral functions. Little is known about PMU1, other than most Ascomycetes have one copy of the gene, the gene is involved in the dephosphorylation of trehalose, and it contains a phosphomutase-like domain (Elliott et al. 1996;Orkwis et al. 2010). In C. glabrata, there were two duplication events, and after the first duplication event which generated the ancestor of CgPMU2 and CgPMU3, the phosphatase acquired the ability to cleave TPP. After the second duplication event, CgPMU2 specialized into a broad-range acid phosphatase, and CgPMU3 maintained the TPPase activity. It is tempting to speculate that loss of the PHO5 family and gain of the PMU family is related to the pathogenicity of C. glabrata, but the same expansion/contraction did not occur in the pathogen C. albicans, so there is little evidence to support this coincidence. We believe that the PHO5 family ancestrally had both TPPase and some broad-range phosphatase activity. The dynamic duplication and gene loss events that occurred over evolutionary time allowed for likely rare cases of specializationwhat we observe in S. cerevisiae. In S. cerevisiae, changes in the amino acid sequence of the phosphatase genes allowed for specialization toward TPP, and specialization toward other organic phosphatases. However, this specialization is only one possibility, as S. pombe PHO5 homologs suggest that you can have both activities present at the same time (TPPase and PNPPase activity). It is worth emphasizing that we do not have access to the true ancestors, and so this evolutionary history is speculative in nature.
Often, phosphatases are thought of as releasing phosphate from an organic phosphate substrate, but we demonstrate that phosphatase expression might be more important for the recycling of the organic substrate (thiamine). In support of this argument are the results with a gene that mimics the ancestor of the CgPMU2 and CgPMU3 genes. This protein, which contains all of the amino acid changes common to CgPmu2 and CgPmu3 in the context of CgPmu1, has phosphatase substrate specificity that mirrors the thiamine-regulated CgPmu3 protein. Selection regimes that place energetic constraints on phosphatase expression and the recycling of thiamine, which is critical for sugar utilization, may have generated the behaviors that we observe in multiple Ascomycetes. Given the dynamic nature of the PHO5 and PMU1 gene families, it is not possible to ascertain what selective pressures drove the behaviors we observe, but this work highlights that recycling the organic compound connected to phosphate is also important.