Mutations in the phosphomannomutase 2 gene, PMM2, are the most common cause of CDG. Lao et al., 2019 constructed yeast models of PMM2-CDG by making the yeast-equivalent mutations for five human-disease-associated alleles in SEC53, the yeast ortholog of PMM2. These strains have growth rate defects that correlate with enzymatic activity and promoter strength. Increasing expression levels of the mutant sec53 alleles twofold, by replacing the endogenous promoter (pSEC53) with the ACT1 promoter (pACT1), modestly improves the growth rate of sec53 disease-associated alleles (Figure 1A).

We chose to focus on sec53-V238M and sec53-F126L because they are two of the most common disease alleles in humans and because the mutations have distinct and well-characterized effects on protein structure and function (Figure 1A; Andreotti et al., 2015; Briso-Montiano et al., 2022; Citro et al., 2018; Kjaergaard et al., 1999; Pirard et al., 1999; Silvaggi et al., 2006). We evolved 96 haploid populations of each mutant sec53 genotype (pSEC53-sec53-V238M, pACT1-sec53-V238M, and pACT1-sec53-F126L) and 48 haploid populations of each wild-type SEC53 genotype (pSEC53-SEC53-WT and pACT1-SEC53-WT) for 1000 generations in rich glucose medium (Figure 1B). Strains with the pSEC53-sec53-F126L allele grew too slowly to keep up with a 1:2¹⁰ dilution every 24 hr. Every 50 generations we measured culture density (single time-point OD₆₀₀) for each population as a proxy for growth rate (Figure 1C; Figure 1—figure supplement 1).

Populations with the pACT1-sec53-V238M allele show a range of dynamics, with some populations reaching maximum saturation (OD₆₀₀ ≈ 1.0) early in the experiment, while some do not even by Generation 1000 (Figure 1C). Three pACT1-sec53-V238M populations went extinct before Generation 200. While saturation levels of pACT1-sec53-F126L populations also increased over the course of the evolution experiment, none reached an OD₆₀₀ close to 1.0, indicating that these populations are likely less-fit than the evolved pACT1-sec53-V238M populations. Each pSEC53-sec53-V238M population started the evolution experiment with growth rates comparable to SEC53-WT populations, indicating compensatory mutation(s) likely arose in the starting inocula (Figure 1—figure supplement 1). We therefore exclude these populations from subsequent analyses. Together, our data show that the PMM2-CDG yeast models acquired compensatory mutations throughout the evolution experiment and that the extent of compensation depends on the specific disease-associated allele.

To identify the compensatory mutations in our evolved populations, we sequenced single clones from 188 populations isolated at Generation 1000 to an average sequencing depth of ~50×. These included 91 pACT1-sec53-V238M clones, 36 pACT1-sec53-F126L clones, 32 pSEC53-SEC53-WT clones, 11 pACT1-SEC53-WT clones, and 18 of the initially suppressed pSEC53-sec53-V238M clones (Supplementary file 1). Autodiploids are a common occurrence in our experimental system (Fisher et al., 2018; Johnson et al., 2021). We attempted to avoid sequencing autodiploids by screening our evolved populations for sensitivity to benomyl, an antifungal agent that inhibits growth of S. cerevisiae diploids more severely than haploids (Venkataram et al., 2016).

We find that the average number of de novo mutations (single nucleotide polymorphisms [SNPs] and small indels) varied between SEC53 genotypes, with the pACT1-sec53-F126L clones accruing more mutations per clone (8.14 ± 1.00, 95% confidence interval [CI]) compared to pACT1-sec53-V238M (5.76 ± 0.54) and SEC53-WT clones (combined pSEC53-SEC53-WT and pACT1-SEC53-WT) (6.49 ± 0.90) (Figure 2—figure supplement 1). Despite screening for benomyl sensitivity, most of the sequenced clones (131/188) appear to be autodiploids, with evidence of multiple heterozygous loci. In addition, we detect several copy number variants (CNVs) and aneuploidies shared between independent populations (Figure 2—source data 1, Figure 2—source data 2).

We identified putative adaptive targets of selection as genes with more mutations than expected by chance across replicate clones. Some of these common targets of selection are not specific to sec53. For example, mutations in negative regulators of Ras (IRA1 and IRA2) are found in clones from each experimental group and are also observed in other laboratory evolution experiments across a wide range of conditions (Figure 2A; Fisher et al., 2018; Gresham and Hong, 2014; Johnson et al., 2021; Kvitek and Sherlock, 2013; Lang et al., 2013; Venkataram et al., 2016).

We identified putative sec53-compensatory targets as genes mutated exclusively in pACT1-sec53-V238M and/or pACT1-sec53-F126L populations (Figure 2A). The most frequently mutated genes among pACT1-sec53-F126L clones are common targets in other evolution experiments such as IRA1, IRA2, and KRE6, suggesting that mutations capable of compensating for Sec53-F126L dimerization defects are rare or not easily accessible. In contrast, while recurrently mutated genes in pACT1-sec53-V238M populations include common targets, we also identified a number of unique targets of selection, most notably in homologs of other Type 1 CDG-associated genes. We find an enrichment of mutations in CDG homologs in pACT1-sec53-V238M clones (Figure 2B; Figure 2—source data 3).

Across all sequenced populations we identified ~620 nonsynonymous mutations. As there are nearly 6000 yeast genes, the probability of any given gene receiving even a single nonsynonymous mutation is low. We are therefore well powered to detect enrichment of mutations, but underpowered to detect depletion of mutations. To gain statistical power we aggregated our SEC53-WT data with three other large datasets using similar strains and propagation regimes (Fisher et al., 2018; Johnson et al., 2021; Lang et al., 2013). In this aggregate dataset, we observe a statistically significant depletion of mutations in Type 1 CDG homologs, suggesting that they are typically under purifying selection in experimental evolution (Figure 2B).

One of the CDG homologs, PGM1, is the most frequently mutated gene among pACT1-sec53-V238M clones. PGM1 encodes the minor isoform of phosphoglucomutase in yeast (Bevan and Douglas, 1969). We only find PGM1 mutations in pACT1-sec53-V238M populations which suggests their compensatory effects are unique to the sec53-V238M mutation and not glycosylation deficiency in general. Each of the five mutations in PGM1 are missense mutations, rather than frameshift or nonsense. This is consistent with selection acting on alteration-of-function rather than loss-of-function (LOF; posterior probability of non-LOF = 0.96, see Methods).

To identify the compensatory mutation(s) that arose prior to the start of the evolution experiment in the initially suppressed pSEC53-sec53-V238M populations, we analyzed the 18 sequenced clones. For each clone, we find both wild-type SEC53 and sec53-V238M alleles among sequencing reads. This could result from integration or maintenance of the covering plasmid. No sequencing reads align to the URA3 locus of our reference genome, suggesting that the plasmid marker was lost, as expected following counterselection.

In order to validate putative compensatory mutations we reconstructed evolved mutations in PGM1 and ALG9, two genes whose human homologs are implicated in Type 1 CDG (Frank et al., 2004; Timal et al., 2012). These heterozygous evolved mutations were constructed in a homozygous pACT1-sec53-V238M diploid background (note that although each pgm1 and alg9 mutation assayed here arose in haploid-founded populations, they arose as heterozygous mutations in autodiploids). We quantified the fitness effect of each mutation using a flow cytometry-based fitness assay, competing the reconstructed strains against a fluorescently labeled, diploid version of the pACT1-SEC53-WT ancestor.

The pACT1-sec53-V238M ancestor has a fitness deficit of −25.86 ± 2.3% (95% CI) relative to wild-type. We find that each of the five evolved pgm1 mutations are compensatory in the pACT1-sec53-V238M background. The fitness effects of the sec53/pgm1 double mutants range between −19.10 ± 1.07% and −9.84 ± 0.77% (i.e., the pgm1 mutations confer a fitness benefit between 6.76% and 16.02% in the pACT1-sec53-V238M background) (Figure 3A). We also find that the evolved alg9-S230R mutation is compensatory, as the sec53/alg9 double mutant has a fitness effect of −17.89 ± 0.59% (Figure 3—figure supplement 1). We compared the fitness of these reconstructed double mutants to the fitness of the evolved clones which carry three to nine additional SNPs. In each case, the evolved clones were more fit than the reconstructed strain, except for the evolved clone containing the pgm1-D295N mutation (Figure 3—figure supplement 1). Thus, while the alg9 and pgm1 mutations compensate for the sec53-V238M defect, other mutations contribute to the overall fitness of these evolved clones.

We constructed and assayed the fitness effects of the pgm1 mutations in diploid pACT1-sec53-F126L and pACT1-SEC53-WT backgrounds. Each pACT1-sec53-F126L strain was too quickly outcompeted by the reference strain to measure fitness, suggesting that the compensatory effects of the evolved pgm1 mutations are specific for the sec53-V238M mutation or any fitness improvements in the pACT1-sec53-F126L background are too minimal to detect. Each pgm1 mutation is nearly neutral in the pACT1-SEC53-WT background (Figure 3—figure supplement 2).

To determine if LOF of PGM1 would phenocopy evolved mutations, we deleted one copy of PGM1 (pgm1Δ) in the diploid pACT1-sec53-V238M background and again measured fitness. We find that the heterozygous pgm1Δ mutation improves fitness (−15.28 ± 0.76%), indicating that LOF of PGM1 is compensatory in the conditions of the evolution experiment (Figure 3A). It is not immediately clear, then, why we only identify missense mutations in PGM1. Each of the mutated Pgm1 residues is located around the active site of the enzyme, based on predicted protein structure (Jumper et al., 2021; Stiers and Beamer, 2018). It could be that the evolved pgm1 mutations are LOF given this clustering but maintaining protein expression provides an ancillary benefit. To account for this possibility, we constructed a catalytically dead allele of PGM1 by mutating the enzyme’s catalytic serine (pgm1-S120A) (Stiers et al., 2017a). We find that pgm1-S120A improves pACT1-sec53-V238M fitness (−13.32 ± 0.88%) and this does not significantly differ from pgm1Δ (Figure 3A), indicating that PGM1 LOF is compensatory regardless of if that arises from loss of coding sequence or enzyme activity. We also measured fitness of homozygous pgm1 mutations (Figure 3—figure supplement 3). In all cases, the fitness effect of heterozygous and homozygous pgm1 mutants is not substantially different from each other, indicating that, by this assay, the evolved pgm1 mutations are dominant in the pACT1-sec53-V238M background, as are the pgm1Δ and pgm1-S120A mutations.

We have established that pgm1 mutations compensate for the growth rate defect of the sec53-V238M allele. To determine whether the evolved pgm1 mutants also compensate for the molecular defects in N-linked glycosylation we examined two representative yeast glycoproteins, invertase and carboxypeptidase Y (CPY), in our reconstructed pgm1 strains. Invertase forms both a non-glycosylated homodimer (120 kDa) and a secreted homodimer that is heavily glycosylated (approximate range of 140–270 kDa) (Gascón et al., 1968; Zeng and Biemann, 1999). Mature CPY only exists in its glycosylated form, with a molecular weight of 61 kDa (Hasilik and Tanner, 1978).

The ancestral pACT1-sec53-V238M strain shows underglycosylation of invertase and CPY. We find a reduced abundance of the higher-molecular-weight glycosylated form of invertase and mature CPY, accompanied by the appearance of underglycosylated forms of these proteins (Figure 3B). We find that evolved pgm1 mutations, pgm1Δ, and pgm1-S120A each restore glycosylation of these proteins to near-wild-type levels (Figure 3B, Figure 3—figure supplements 4 and 5). We do not observe rescue of protein glycosylation in pACT1-sec53-F126L strains carrying pgm1 mutations (Figure 3—figure supplement 6). Together, these suggest that pgm1-mediated rescue of pACT1-sec53-V238M fitness is due to restoration of protein glycosylation and the effect is specific for the pACT1-sec53-V238M background.

To further demonstrate that pgm1-mediated compensation is specific for sec53-V238M, we performed a genetic analysis by dissecting tetrads from strains that are heterozygous for both sec53 and pgm1. In the absence of a compensatory mutation, we expect 50% large colonies and all tetrads showing a 2:2 segregation of colony size due to the single segregating locus (SEC53/sec53). However, if an evolved mutation is compensatory then we expect 75% large colonies with three types of segregation patterns: 2:2, 3:1, and 4:0, large to small colonies, respectively. These three segregation patterns are expected to follow a 1:4:1 ratio assuming no genetic linkage (Figure 4A). For each genetic test, we dissected ten tetrads and performed a log-likelihood test to determine whether the segregation pattern is more consistent with genetic suppression than non-suppression (Figure 4—source data 1).

We find each of the evolved pgm1 mutations compensates for the V238M allele of SEC53 (Figure 4—figure supplement 1). For example, dissections of a strain heterozygous for pACT1-sec53-V238M and pgm1-T521K show 78% large colonies (one 4:0 and nine 3:1, with a log-likelihood ratio of 27.6, Figure 4B). These genetic tests corroborate the results from the fitness assays by showing that all five evolved pgm1 mutations, the pgm1Δ and pgm1-S120A mutations, as well as the evolved alg9 mutation all suppress sec53-V238M (Figure 4—figure supplement 1). We next assayed each of the pgm1 mutations in a sec53-F126L background. In contrast to the fitness assays, we find several pgm1 alleles (T521K, D295N, S120A, and pgm1Δ) that suppress sec53-F126L (Figure 4—figure supplement 1). It is worth noting, however, that the degree of compensation is less than in the sec53-V238M background based on colony sizes.

To test if pgm1 mutations are dominant, we integrated a second wild-type copy of PGM1 to each of the strains such that each haploid spore will contain either two wild-type copies of PGM1 (PGM1::PGM1) or both wild-type PGM1 and mutant pgm1 (PGM1::pgm1). We find that the five evolved pgm1 mutations and pgm1-S120A, but not pgm1Δ, are dominant suppressors of sec53-V238M in the genetic assay (Figure 4C, D; Figure 4—figure supplement 2). For the sec53-F126L background, we find no dominant suppressors.

To determine how the evolved mutations alter Pgm1 enzymatic activity, we cloned mutant and wild-type PGM1 alleles into bacterial expression vectors, purified recombinant enzyme, and assayed phosphoglucomutase activity using a standard coupled enzymatic assay (Figure 5A; Figure 5—figure supplement 1). We find that mutant Pgm1 have variable activities, ranging from near-wild-type (Pgm1-T231A) to non-detectable (Pgm1-R64C and Pgm1-D295N) (Figure 5B, C). The near complete loss of activity of Pgm1-D295N is unsurprising since D295 coordinates a Mg²⁺ ion that plays an essential role during catalysis (Stiers et al., 2016) and we verify that all the active forms show very low activity in the presence of ethylenediaminetetraacetic acid (EDTA) (Figure 5—source data 1). Thermostability did not differ between enzymes with detectable activity (Figure 5—figure supplement 2).

Comparison of the kinetic parameters of active mutant enzymes shows a reduction of maximal velocity (V_max) between 27% and 72% relative to wild-type Pgm1. Based on the apparent Michaelis constant (K_m), substrate affinities of Pgm1-G514C and Pgm1-T521K are lower than wild-type Pgm1 (t = 4.30 and 3.14, df = 35.78 and 20.47, p = 0.0001 and 0.0051, respectively, Welch’s modified t-test). Pgm1-T231A shows an increased affinity compared to wild-type Pgm1, but this difference is not statistically significant (t = 1.94, df = 34.68, p = 0.0601, Welch’s). Together these indicate that the five evolved pgm1 mutations have varying effects on enzymatic activity. The relationship between pgm1 fitness and Pgm1 V_max is best fit by a quadratic, rather than a linear, function (Figure 5D). Therefore, optimal fitness is attained by tuning down, but not eliminating, Pgm1 activity.

Phosphoglucomutases can exhibit low levels of phosphomannomutase activity, and we reasoned that active-site mutations in Pgm1 could alter epimer specificity from glucose to mannose, directly compensating for the loss of Sec53 activity (Lowry and Passonneau, 1969). We tested this in two ways. First, we directly measured phosphomannomutase activity of wild-type Pgm1 and two mutant Pgm1, using a ³¹P-NMR spectroscopy-based assay (Figure 5—figure supplements 1 and 3A). We find that wild-type Pgm1 and Pgm1-G514C show comparably low phosphomannomutase activity, accounting for 3.5% and 3.0% of their phosphoglucomutase activity, respectively. We could not detect any phosphomannomutase activity for Pgm1-D295N. We also tested for phosphomannomutase activity indirectly by determining whether M1P acted as a competitive inhibitor in the phosphoglucomutase assay. If mutant Pgm1 had altered epimer specificity, we would expect M1P to compete with glucose-1-phosphate (G1P) for residency within the active site, leading to an apparent decrease in phosphoglucomutase activity. However, we find no effect of the addition of M1P on phosphoglucomutase activity for any of the Pgm1 enzymes tested (Figure 5—figure supplement 3B). Together, these results indicate that mutant Pgm1 does not have enhanced phosphomannomutase activity.

Glucose-1,6-bisphosphate (G16P) is required as an activator of both Pgm1 and Pmm2 (Sec53) and can stabilize pathogenic variants of Pmm2 (Monticelli et al., 2019). The only known source of G16P in yeast is low-level dissociation during the phosphoglucomutase reaction of Pgm1 and Pgm2. We reasoned that evolved pgm1 mutations may increase intracellular levels of G16P which, in turn, would stabilize Sec53-V238M. Each active form of Pgm1 exhibits barely detectable activity in the absence of G16P and we determined an apparent half maximal effective concentration (EC₅₀) of G16P for several Pgm1 enzymes (Figure 5—source data 1). Given that Pgm1 is the minor isoform of phosphoglucomutase in yeast, we also reasoned that it could act as a glucose-1,6-bisphosphatase, similar to human PMM1 (Veiga-da-Cunha et al., 2008). We performed a specific glucose-1,6-bisphosphatase assay on wild-type and mutant Pgm1 enzyme (Figure 5—figure supplement 1). However, none of the mutant Pgm1 enzymes have detectable glucose-1,6-bisphosphatase activity.

We next assessed G16P levels for two pgm1 mutant alleles in both the pACT1-sec53-V238M and the pACT1-SEC53-WT backgrounds (Figure 5—figure supplement 4). We find that strains heterozygous for pgm1-T521K or pgm1-D295N show statistically significant increases in the amount of G16P present compared to wild-type PGM1, in the sec53-V238M background (W = 24 and 8, p = 0.003 and 0.0002, respectively, Mann–Whitney U-test). However, this difference is not significant in the SEC53-WT background (W = 41 and 15, p = 0.35 and 0.078, respectively, Mann–Whitney).

Finally, we also reasoned that pgm1 mutations could differently affect the forward and reverse directions of the phosphoglucomutase reaction, with possible effects on the flux of metabolites within the cell. We analyzed the forward and reverse reactions of Pgm1, Pgm1-T231A, and Pgm1-G514C using G1P or glucose-6-phopshate (G6P) as the substrate (Figure 5—figure supplement 1). We again used ³¹P-NMR spectroscopy, as G1P and G6P have clearly distinguishable signals (Figure 5—figure supplement 5). The ratio of forward to reverse reactions is 3.0 ± 1.2 and 4.3 ± 1.1 (± standard deviation) for Pgm1-T231A and Pgm1-G514C, respectively. Although a slightly higher ratio was obtained for wild-type Pgm1 (7.4 ± 3.1), the differences between wild-type and Pgm1-T231A or Pgm1-G514C are not statistically significant (W = 1 and 2, p = 0.057 and 0.53, respectively, Mann–Whitney).

The haploid strains used in the evolution experiment are S288c derivatives and were described previously (Lao et al., 2019). Each strain maintains a URA3-marked plasmid encoding wild-type SEC53 (pTC416-SEC53 described in Lao et al., 2019). Strains with the pSEC53-sec53-F126L allele grew too slowly to keep up with a 1:2¹⁰ dilution every 24 hr.

To initiate the evolution experiment, strains were struck to synthetic defined media (complete supplement mixture [CSM] minus arginine, 0.67% yeast nitrogen base, 2% dextrose) containing 5-Fluoroorotic Acid (5FOA) (Zymo Research, Irvine, CA, USA) to isolate cells that spontaneously lost pTC416-SEC53. A single colony was chosen for each strain and used to inoculate 10 ml of YPD (1% yeast extract, 2% peptone, 2% dextrose) plus ampicillin and tetracycline. These cultures grew to saturation then were distributed into 96-well plates. Every 24-hr populations were diluted 1:2¹⁰ using a BiomekFX liquid handler into 125 μl of YPD plus 100 μg/ml ampicillin and 25 μg/ml tetracycline to prevent bacterial contamination. Plates were then left at 30°C without shaking. The dilution scheme equates to 10 generations of growth per day at an effective population size of ~10⁵. Every 100 generations, 50 μl of 60% glycerol was added to each population and archived at −80°C.

OD₆₀₀ was measured every 50 generations over the course of the evolution experiment. After daily dilutions were completed, populations were resuspended by orbital vortexing at 1050 rpm for 15 s then transferred to a Tecan Infinite M200 Pro plate reader for sampling.

Benomyl assays were performed by spotting 5 μl of populations onto YPD plus 20 μg/ml benomyl (dissolved in dimethyl sulfoxide [DMSO]) and incubating at 25°C. Ploidy was then inferred by visually comparing growth of evolved populations to control haploid and diploid strains.

Single clones were isolated for sequencing by streaking 5 μl of cryo-archived populations to YPD agar. One clone from each population was randomly chosen and grown to saturation in 5 ml YPD, pelleted, and frozen at −20°C. Genomic DNA was harvested from frozen cell pellets using phenol–chloroform extraction and precipitated in ethanol. Total genomic DNA was used in a Nextera library preparation as described previously (Buskirk et al., 2017). Pooled clones were sequenced using an Illumina NovaSeq 6000 sequencer by the Sequencing Core Facility at the Lewis-Sigler Institute for Integrative Genomics at Princeton University.

Raw sequencing data were concatenated and then demultiplexed using a dual-index barcode splitter (https://bitbucket.org/princeton_genomics/barcode_splitter/src/master/). Adapter sequences were trimmed using Trimmomatic (Bolger et al., 2014). Modified S288c reference genomes were constructed using reform (https://github.com/gencorefacility/reform; Khalfan, 2021) to correct for strain-construction-related differences between our strains and canonical S288c. Trimmed reads were aligned to these modified S288c reference genomes using BWA and mutations were called using FreeBayes (Garrison and Marth, 2012; Li and Durbin, 2009). VCF files were annotated with SnpEff (Cingolani et al., 2012). All calls were confirmed manually by viewing BAM files in IGV (Thorvaldsdóttir et al., 2013). Clones were predicted to be diploids if two or more mutations were called at an allele frequency of 0.5. CNVs and aneuploidies were called via changes in sequencing coverage using Control-FREEC (Boeva et al., 2012). CNVs and aneuploidies were confirmed by visually inspecting coverage plots.

Evolved mutations were reconstructed via CRISPR/Cas9 allele swaps. For mutations that fall within a Cas9 gRNA site (alg9-S230R, pgm1-T231A, and pgm1-G514C), repair templates were produced by PCR-amplifying 500–2000 bp fragments centered around the mutation of interest from evolved clones. Commercially synthesized dsDNA fragments (Integrated DNA Technologies, Coralville, IA, USA) were used as repair templates for all other mutations (pgm1-R64C, pgm1-S120A, pgm1-D295N, pgm1-T521K, and pgm1Δ). These 500 bp fragments contained our mutation of interest as well as one to two synonymous mutations in the Cas9 gRNA site to halt cutting. The pgm1Δ repair template was a dsDNA fragment consisting of 250 bp upstream then 250 bp downstream of the ORF. Repair templates were cotransformed into haploid ancestral strains with a plasmid encoding Cas9 and gRNAs targeting near the mutation site (Addgene #83476) (Laughery et al., 2015).

Heterozygous mutant strains were constructed by mating haploid mutants to MATα versions of their respective ancestral strains. Homozygous mutant strains were constructed by reconstructing mutations in MATα versions of the ancestral strains and mating them to the respective MATa mutant strain. Successful genetic reconstructions were confirmed via Sanger sequencing (Psomagen, Rockville, MD, USA).

Linked PGM1 strains used for tetrad dissections were constructed by transforming strains with a linearized integrating plasmid (Addgene #35121) containing wild-type PGM1 (Goldstein and McCusker, 1999). Successful integration of the plasmid next to the endogenous PGM1 locus was confirmed via PCR. Strain information in Supplementary file 2.

We measured the fitness effect of evolved mutations using competitive fitness assays described previously with some modifications (Buskirk et al., 2017). Query strains were mixed 1:1 with a fluorescently labeled version of the pACT1-SEC53-WT ancestral diploid strain. Each pACT1-sec53-F126L strain was outcompeted immediately by the reference at this ratio, so we also tried mixing at a ratio of 50:1. Cocultures were propagated in 96-well plates in the same conditions in which they evolved for up to 50 generations. Saturated cultures were sampled for flow cytometry every 10 generations. Each genotype assayed was done so with at least 24 technical replicates (competitions) of 4 biological replicates (2 clones isolated each from 2 isogenic but separately constructed strains). Analyses were performed in Flowjo and R.

From reconstruction data from previously published evolution experiments (Fisher et al., 2018; Lang et al., 2013; Marad et al., 2018), we identified ten genes where selection is acting on LOF (ACE2, CTS1, ROT2, YUR1, STE11, STE12, STE4_haploids, STE5, IRA1, and IRA2) and six genes where selection is for non-LOF (CNE1, GAS1, KEG1, KRE5, KRE6, and STE4_autodiploids). These data established prior probabilities that selection is acting on LOF (0.625) or non-LOF (0.375). We then determined the conditional probabilities of missense and frameshift/nonsense mutations given selection for LOF and non-LOF using 240 mutations across these 16 genes as well as 414 5FOA-resistant mutations at URA3 and 454 canavanine-resistant mutations at CAN1 (Lang and Murray, 2008). From these data, we can estimate the log likelihood and posterior probabilities that selection is acting on LOF or non-LOF given the observed mutational spectrum for any given gene (Supplementary file 3).

Strains were grown in 5 ml YPD plus 100 μg/ml ampicillin and 25 μg/ml tetracycline until saturated, then pelleted, and frozen at −20°C. Cell pellets were lysed in Ripa lysis buffer (150 mM NaCl, 1% Nonidet P-40, 0.5% sodium deoxycholate, 0.1% sodium dodecyl sulfate, 25 mM Tris [pH 7.4]) with protease inhibitors (Thermo Fisher Scientific, Waltham, MA, USA) and 10 mM dithiothreitol (DTT), incubated for 30 min on ice, sonicated five times briefly, incubated for a further 10 min on ice, and centrifuged at 20,000 × g for 10 min at 4°C. Supernatant was quantified by Micro BCA Protein Assay (Thermo Fisher). 30 μg of each lysate were prepared in Lamelli buffer and were incubated at 95°C for 5 min and chilled at 4°C. Lysates were separated on a 6% or 8% sodium dodecyl sulfate–polyacrylamide gel electrophoresis (SDS–PAGE) gel. Protein was transferred to 0.2 µm pore nitrocellulose at 100 V for 100 min at 4°C in Towbin buffer. The membrane was rinsed and stained with Ponceau S for normalization. The membranes were blocked with 5% milk/TBST for 1 hr at room temperature.

CPY blots were incubated with an anti-carboxypeptidase antibody at 1:1000 overnight at 4°C and washed three times with TBST. The blots were then incubated with anti-rabbit IgG HRP diluted 1:2000 in milk/TBST for 1 hr, washed three times and developed with ECL reagent (Bio-Rad Laboratories, Hercules, CA, USA). Invertase blots were prepared similarly with an anti-invertase antibody and anti-goat IgG HRP. All images were captured on the Bio-Rad ChemiDoc MP Imaging System (Bio-Rad). Analysis was done with Image Lab Software (Bio-Rad). Raw images of blots and Ponceau stain controls are available in Figure 3—source data 1.

Heterozygous yeast strains (i.e., SEC53/sec53 PGM1/pgm1) were sporulated by resuspending 1 ml of overnight YPD culture in 2 ml of SPO++ (1.5% potassium acetate, 0.25% yeast extract, 0.25% dextrose, 0.002% histidine, 0.002% tryptophan) and incubated on a roller-drum for ~7 days at room temperature. Ten tetrads per strain were dissected on YPD agar and then incubated for 48 hr at 30°C. We performed a log-likelihood test to determine if patterns of segregation are indicative of suppression (Figure 4—source data 1). Note that in order to calculate conditional probabilities we had to include an error term (0.01) to account for biological noise in the real data. Our inference of suppression, however, is robust to our choice of error value.

Wild-type and mutant PGM1 alleles were cloned into the expression vector pGEX-2T. Cloned vectors were transformed into E. coli BL21(DE3). Bacteria were grown in lysogeny broth (LB) medium plus 0.1 mg/ml ampicillin. Each PGM1 allele was expressed by induction at 0.1 OD with 0.1 mg/ml IPTG for 16 hr at 15°C. Bacteria were harvested by centrifugation, washed with phosphate-buffered saline (PBS), and stored at −80°C. Bacterial lysis was accomplished in 25 mM Tris buffer (pH 8) containing 1 mM EDTA, 2 mM DTT, 5% glycerol, and 0.1 mM phenylmethylsulfonyl fluoride (PMSF), by adding 1 mg/ml lysozyme and 2.5 µg/ml DNase. Clear extract was then obtained by centrifugation.

Protein purification was accomplished using Glutathione Sepharose High Performance resin (GSTrap by Cytiva, Marlborough, MA, USA) and Benzamidine Sepharose 4 Fast Flow resin (HiTrap Benzamidine FF by Cytiva). Cleavage of the GST-tag was conducted on-column by adding thrombin. All the procedures were performed according to manufacturer’s instructions. Briefly, clear extract obtained from 50 ml of bacterial culture was loaded onto the GSTrap column (5 ml), previously equilibrated with 25 mM Tris (pH 8) containing 2 mM DTT and 5% glycerol. Unbound protein was eluted, then the column was washed with 20 mM Tris (pH 8) containing 1 mM DTT, 150 mM NaCl, 1 mM MgCl₂, and 5% glycerol (buffer A). Thrombin (30 units in 5 ml of buffer A) was loaded and the column was incubated for 14–16 hr at 10°C. GSTrap and HiTrap Benzamidine columns were connected in series, and elution of the untagged protein was accomplished with buffer A. Fractions were analyzed by SDS–PAGE and activity assays. Active fractions were finally collected and concentrated by ultrafiltration. A yield spanning 1.0–4.6 mg of protein per liter of bacterial culture was obtained.

Each protein was obtained in a pure, monomeric form as judged by SDS–PAGE and gel filtration analyses, with a single band observed at the expected molecular weight (63 kDa). Gel filtration analysis was performed on BioSep-SEC-S 3000 column (Phenomenex, Torrance, CA, USA) at 1 ml/min. The eluent was 20 mM Tris (pH 8), 1 mM MgCl₂, and 150 mM NaCl.

Phosphoglucomutase activity was assayed spectrophotometrically at 340 nm and 29°C by following the reduction of NADP⁺ to NADPH in a 0.3 ml reaction mixture containing 25 mM Tris (pH 8), 5 mM MgCl₂, 1 mM DTT, 0.25 mM NADP⁺, and 2.7 U/ml glucose-6-phosphate dehydrogenase in the presence of 0.3 mM G1P and 0.02 mM G16P. Additional experiments were performed with the addition of M1P (10, 50, and 200 μM). K_m were evaluated by measuring activity in the presence of G1P ranging from 0 mM to 0.6 mM. Analyses were performed in R using the drc package (Ritz et al., 2015). Bisphosphatase activity was measured spectrophotometrically at 340 nm and 29°C in the presence of 0.1 mM G16P, 0.25 mM NADP⁺, and 2.7 U/ml glucose-6-phosphate dehydrogenase. Activity expressed as the number of micromoles of substrate transformed per minute per mg of protein under the standard conditions.

Thermostabilities of Pgm1 were determined by incubating purified protein for 5 or 10 min at different temperatures (30, 35, 40, and 45°C) in 20 mM Tris (pH 8), 1 mM MgCl₂, 150 mM NaCl, 5% glycerol, and 0.1 mg/ml BSA. After cooling on ice, residual activity was measured under standard conditions.

Phosphoglucomutase and phosphomannomutase activity were measured by ³¹P-NMR spectroscopy (Citro et al., 2017). This assay allows us to exclude the effects on the change in absorbance due to impurities that could be substrates for the ancillary enzymes required for the assay. This is of utmost importance when the activity is particularly low and measuring the rate of change in absorbance requires long incubation times, as it is the case of phosphomannomutase activity for Pgm1. A discontinuous assay was performed. Appropriate amounts of proteins were incubated at 30°C with 1 mM G1P or M1P in 20 mM Tris (pH 8), 1 mM MgCl₂, 0.1 mg/ml BSA, in the presence of 20 μM G16P for up to 60 min. The reaction was stopped with 50 mM EDTA on ice and heat inactivation. The content of the residual substrate (M1P or G1P) and/or the product (M6P or G6P) were measured recording ³¹P-NMR spectra. Creatine phosphate was added as an internal standard for the quantitative analysis. The ¹H-decoupled, one-dimensional ³¹P spectra were recorded at 161.976 MHz on a Bruker Avance III HD spectrometer 400 MHz, equipped with a BBO BB-H&F-D CryoProbe Prodigy fitted with a gradient along the z-axis, at a probe temperature of 27°C. Spectral width 120 ppm, delay time 1.2 s, and pulse width of 12.0 μs were applied. All the samples contained 10% ²H₂O for internal lock.

Similarly, forward (G1P to G6P) and reverse (G6P to G1P) phosphoglucomutase activities were measured following the same procedure described above, with 1 mM G1P or G6P as the substrate.

Yeast metabolite extraction was performed as described previously (Gonzalez et al., 1997). Briefly, five replicate cultures per genotype were grown in 5 ml YPD plus 100 μg/ml ampicillin and 25 μg/ml tetracycline at 30°C. 10⁷ cells of mid-log culture were collected on nylon filters, washed with 10 ml cold PBS, frozen in liquid nitrogen, and lyophilized. Cells were then treated with an ice-cold solution containing 14% HClO₄ and 91 mM imidazole, frozen and thawed five times (dry ice/acetone followed by 40°C bath) with vigorous shaking between each cycle, then centrifuged at 13,500 × g at 4°C. Supernatant was neutralized with 3 M K₂CO₃ and centrifuged again at 13,500 × g at 4°C to eliminate salt precipitate.

G16P in the metabolite extracts was measured through stimulation of rabbit muscle phosphoglucomutase as described previously (Veiga-da-Cunha et al., 2008). Phosphoglucomutase activity was assayed spectrophotometrically at 340 nm, in a mixture containing 87 mM Tris (pH 7.6), 5.6 mM MgCl₂, 0.11 mM ethylene glycol tetraacetic acid (EGTA), 0.1 mg/ml BSA, 0.048 mM NADP⁺, 0.48 mM G1P, 0.26 U/ml glucose-6-phosphate dehydrogenase, at 36°C. G1P had previously been purified from G16P through anionic exchange chromatography on AG1x8 Hydroxide Form column in step gradient of triethylammonium bicarbonate from 0.01 M to 1 M (Monticelli et al., 2019). Three technical replicates for each of the five biological replicates were assayed. 0.0–0.15 µM commercial G16P (Merck & Co, Rahway, NJ, USA) was used as the standard.