Media used in this study, along with their component concentrations, are listed in Table S1 in Supplementary Material. For both genetic manipulations and growth and tracer experiments, all strains were grown in non-baffled shake flasks at 30°C at 200 rpm in either minimal glucose medium (Min), minimal glucose medium with galactose (Min + Gal), YPD, or Sc-Ura. All strains were stored in 20% glycerol stocks at −80°C. Labeled media used 80% 1-¹³C glucose and 20% U-¹³C glucose at the same total concentration of 2% glucose. Exponential-phase cells were obtained by streaking from −80°C glycerol stocks on YPD plates, incubating 5-mL YPD cultures overnight, inoculating into 40 mL of unlabeled media of the final desired composition, and grown until exponential phase (usually 0.6–0.9 OD₆₀₀).

Prototrophic base (base) and mutant (sip1Δ) S. cerevisiae strains were constructed in a haploid CEN.PK113-7D (MATa URA3 HIS3, LEU2 TRP1 MAL2-8c SUC2) (Entian and Kötter, 2007) background containing a URA3 knockout. All strains used in this study are listed in Table 1 with their strain designations, parent strain, genotype descriptions, and Inventory of Composable Elements (ICE) reference numbers (Ham et al., 2012) (https://public-registry.jbei.org). All knockouts were constructed via a near-markerless loxP/Cre recombinase strategy (Güldener et al., 1996) and PCR verified. Briefly, each knockout cassette was amplified from a loxP-kanMX-loxP plasmid, pUG6 (Güldener et al., 1996), using the primers listed in Table 2, transformed into yeast using a heat shock method (Agatep et al., 1998), selected on YPD + G418 (geneticin) plates, and PCR verified using primers listed in Table 3. In order to loop out the kanMX marker, a Cre recombinase plasmid was transformed in the resulting kanMX cassette integrants and plated on selective medium. The selective plate varied depending on the knockout. For knockout of SIP1, the Cre recombinase promoter was Gal1p [pSH47 (Güldener et al., 1996)] and selection occurred on pSH47 plates. A different plasmid was necessary for knockout of GAL1, since the strain could not grow on galactose. We opted for expression of the Cre recombinase under a constitutive TEF1 promoter. This new plasmid, pCMS1, was constructed via yeast cloning using SacI and XbaI digested pSH47, to excise Gal1p, and Tef1p amplified with regions homologous to the cut ends and subsequent selection on Sc-Ura plates. All loop-outs were PCR verified using the same verification primers in Table 3 and pCMS1 was sequence verified. Cre recombinase plasmids were cured by streaking on YPD plates, growing overnight in liquid YPD medium, streaking to single colonies on YPD plates, simultaneously streaking on YPD and Sc-Ura plates, and glycerol storing YPD plate colonies whose corresponding Sc-Ura colonies did not grow. Prototrophic final base and mutant strains were completed via transformation of a URA3 plasmid, pRS416 (Sikorski and Hieter, 1989).

Cell and extracellular metabolite concentrations were monitored during exponential phase in strain characterization batch experiments. These data were necessary to calculate extracellular fluxes and specific growth rates used to mathematically constrain flux profile inference. Exponentially growing cells, obtained as described in the Media and culturing conditions section, were used to inoculate, in quadruplicate, the final 40-mL shake flask cultures to achieve exponential growth the following morning. Optical density was monitored at 600 nm via UV–VIS, and 200 µL samples were spin-filtered and kept at −20°C for subsequent HPLC analysis.

Exponentially growing cells (obtained as described above) were used to inoculate 40 mL labeled shake flask cultures in quadruplicate and monitored via UV–VIS. To prevent changes in intracellular metabolite labeling patterns, 1 mL mid-log (~0.75 OD₆₀₀) samples were taken, spun down (1 min, max speed, 4°C), immediately quenched with 300 µL ice-cold methanol, and kept at −80°C.

Labeling distributions were obtained from processed labeled biomass samples for intracellular 3-phospho-d-glycerate (3 pg), alanine (Ala), arginine (Arg), asparagine (Asp), glutamine (Gln), glutamate (Glu), isoleucine (Ile), leucine (Leu), lysine (Lys), phenylalanine (Phe), threonine (Thr), tyrosine (Tyr), valine (Val), citrate (cit_m), fructose 1,6-bisphosphate (fdp), and succinate (succ_m). Succinate and citrate were assumed to be mitochondrial, while the rest are assumed to be cytosolic as has been done in previous studies (Moxley et al., 2009). The closeness of fit of these data with corresponding simulated values provided a measure of the quality of inferred flux distributions. Labeled biomass samples were mixed with 300 µL ice-cold chloroform and 150 µL ice-cold water, spun down, bead-beated with 500 µL acid-washed beads (10 times, 10 s, 1 min on ice between sonication bursts) in 1.7 mL screw cap tubes, the bottom of the tube was punctured with a needle, and the beads were separated from the solution by spinning (1 min, 1,000 g, 4°C) into a 2-mL collection tube. The aqueous layer was filtered (3 k MW cut-off (Amicon), 1.5 h, 13,000 g, 4°C), mixed with 1 mL ice-cold H₂O, and snap frozen in liquid nitrogen. Three holes were punched in the tube cap, and the samples were lyophilized for 24 h. Lyophilized samples were resuspended in 40 µL 50/50 MeOH/H₂O, and stored at −80°C. Samples were analyzed to obtain intracellular amino acid and non-amino acid labeling data via LC-MS as previously described (Bokinsky et al., 2013; Weaver et al., 2015).

Extracellular concentrations for glucose, galactose, ethanol, glycerol, succinate, lactate, acetate, and formate were measured via HPLC. These concentrations, along with corresponding culture specific growth rates, were necessary to calculate extracellular fluxes. The 4 mM H₂SO₄ eluent flowed through a 1200 Series HPLC (Agilent Technologies, CA) outfitted with UV and refraction index detectors and an Organic Acid Analysis Column (Aminex HPX-87H Ion Exclusion Column, 300 mm 7.8 mm, 50°C, Cat# 125-0140 Bio-Rad, CA, USA) at a rate of 0.6 mL/min. Standards were used to identify metabolite retention times and sample concentrations.

Extracellular fluxes and specific growth rates were derived from extracellular concentration and optical density time curves. Their means and standard deviations were used to constrain exchange fluxes for consumed and excreted metabolites and biomass fluxes during flux profile inference. Flask-specific maximum specific growth rates were determined from the slope of lnOD versus time data in manually determined linear ranges. The same time points were used with corresponding concentration data to calculate extracellular fluxes. The extracellular flux of metabolite p, ν_p, is give by equation (1). M_p is the corresponding molecular weight of metabolite p, α is the conversion factor between OD₆₀₀ and cell mass concentration in grams of dry cell weight per liter (gDcW/L), and dC¯p/dOD is the slope of the concentration of metabolite p versus OD₆₀₀. The value of α was taken to be 0.7742 based on multiple in-house experiments (data not shown). The average plus and minus the corresponding standard deviation was used to constrain all extracellular fluxes and specific growth rates.

(1)νp=1,000 μMpα dC¯pdOD, Units are in mmol/gDcW/hr (hence the 1,000 factor).

Flux profiles were inferred from growth and tracer experiment data using 2-scale-¹³C metabolic flux analysis (2S-¹³C MFA) (García Martín et al., 2015). These fluxes, along with specific growth rates and extracellular fluxes were used to characterize the relative effects of the presence of galactose and/or knockout of SIP1. 2S-¹³C MFA was chosen over ¹³C MFA for its ability to describe metabolism more comprehensively through genome-scale models, iMM904 (Mo et al., 2009) in this case. The means and standard deviations for strain/condition-specific extracellular fluxes, intracellular metabolite LC-MS fractional labeling distributions, specific growth rates, and feed glucose labeling were used as inputs for the code included as Supplementary Material. The carbon transitions differed for each strain/condition pair as demanded by the ELVA requirements [see Figure 4 and supp. Fig 22 in García Martín et al. (2015)]. Starting from a base core reaction network, carbon transition information was added to reactions (i.e., the reaction was added to the core network) with the largest flux from non-core metabolism to core metabolism. This was performed iteratively until computational errors in the ELVA plot were minimal. This test guaranteed that labeling from outside the core model [e.g., labeling from CO₂ and formate (Gopalakrishnan and Maranas, 2015)] had a minimal impact in the measured labeling patterns. A final ¹³C Flux Variability Analysis (¹³C FVA) was used to find the maximum and minimum values of each flux compatibles with the experimental data [see (García Martín et al., 2015) for more details]. Our code uses the CONOPT Solver in a GAMS framework to perform the ¹³C MFA step using 30 initial flux starting points.

Confidence intervals and goodness-of-fit were calculated as in García Martín et al. (2015). Briefly, the usual ¹³C MFA goodness-of-fit estimates based on the chi square distribution, such as those proposed by Antoniewicz et al. (2006), are not applicable to 2S-¹³C MFA using genome-scale models [see page 24 in García Martín et al. (2015)]. This problem was surmounted by incorporating the goodness-of-fit considerations in the confidence intervals: good fits produce narrow flux confidence intervals (good flux resolution) and bad fits produce large confidence intervals (bad flux resolution). These confidence intervals are calculated by finding the maximum and minimum values of each flux compatible with experimental error [page 26 in García Martín et al. (2015)]. The experimental error for each m in the Mass Distribution Vector (MDV) was the maximum of the instrument error and the difference of the best fit computational labeling with the experimental labeling [equation (23) in García Martín et al. (2015)]. Hence a bad fit provides a large experimental error and begets large flux confidence intervals and less flux resolution, and a good fit provides narrower confidence intervals and better flux resolution.

The presence of galactose in the medium increased the maximum specific growth rate for both base and sip1Δ mutant strains. Additionally, the deletion of SIP1 only resulted in a growth rate increase when galactose was present in the medium. Average maximum specific growth rates for the four strain/condition combinations are presented in Table 4 and a corresponding box-and-whisker plot is presented in Figure 3. An ~8.5% increase was observed when the base strain was grown in medium with supplemented galactose instead of glucose-only medium (base in 2% glucose versus base in 2% glucose + 0.2% galactose). Similarly, the sip1Δ mutant grew ~18% faster in galactose-supplemented medium relative to that without (sip1Δ in 2% glucose versus sip1Δ in 2% glucose + 0.2% galactose). The sip1Δ mutant grew ~8% faster than the base strain in mixed glucose/galactose medium (base in 2% glucose + 0.2% galactose versus sip1Δ 2% glucose + 0.2% galactose). Consistent with the literature (Breslow et al., 2008; Zhang et al., 2010), no change in maximum specific growth rate was observed between the base and sip1Δ mutant strains in glucose-only medium (base in 2% glucose versus sip1Δ in 2% glucose).

Neither the presence of galactose nor knockout of SIP1 resulted in any clear pattern in the extracellular fluxes. The means and standard deviations for all monitored extracellular metabolites for all strain/condition pairs can be found in Table 4. The extracellular flux input ranges, as the mean plus or minus one standard deviation, for strain condition pairs base in 2% glucose, sip1Δ in 2% glucose, base in 2% glucose + 0.2% galactose, and sip1Δ in 2% glucose + 0.2% galactose are presented in Table 4. The sip1Δ mutant consumed glucose and excreted ethanol at ~44% and ~72% lower rates, respectively, than the base strain in glucose-only medium (base in 2% glucose versus sip1Δ in 2% glucose). The sip1Δ mutant in glucose/galactose medium relative to the parental strain (base in 2% glucose + 0.2% galactose vs sip1Δ in 2% glucose + 0.2% galactose) exhibited a ~98% increase in absolute glucose flux, a ~70% increase in absolute ethanol flux, and a ~54% increase in acetate. Addition of 0.2% galactose to the medium of the base strain (base in 2% glucose vs base in 2% glucose + 0.2% galactose) resulted in a decrease in absolute glucose flux of ~51% and an increase of ~55% for acetate flux. Addition of galactose to the medium of the mutant strain (sip1Δ in 2% glucose vs sip1Δ in 2% glucose + 0.2% galactose) resulted in a ~75% increase in absolute glucose flux, a ~250% increase in ethanol flux, a ~130% increase in glycerol flux, and a ~175% increase in acetate flux. Lactate, formate, and succinate were not detected in any strain/condition pair, hence their extracellular fluxes were considered zero. All other absolute fluxes were the same for all strain/condition pairs within error. The presence of galactose in the medium of the sip1Δ mutant appeared to restore the ethanol flux to its value before the gene knockout (sip1Δ in 2% glucose versus sip1Δ in 2% glucose + 0.2% galactose), within error.

In spite of this lack of clear patterns, we will see below that the addition of ¹³C labeling information in the context of the genome-scale model results in noticeable patterns for the intracellular fluxes.

Detailed fits between simulated and measured LC-MS data for metabolites 3-phospho-d-glycerate (3 pg), alanine (Ala), arginine (Arg), asparagine (Asp), glutamine (Gln), glutamate (Glu), isoleucine (Ile), leucine (Leu), lysine (Lys), phenylalanine (Phe), threonine (Thr), tyrosine (Tyr), valine (Val), citrate (cit_m), fructose 1,6-bisphosphate (fdp), and succinate (succ_m) are displayed in Figure 4 for strain/condition pair base + 2% glucose and in Figures S1–S3 in Supplementary Material for sip1Δ in 2% glucose, base in 2% glucose + 0.2% galactose, and sip1Δ in 2% glucose + 0.2% galactose, respectively. We decided to exclude the labeling data for citrate from the fitting to test how well the fluxes fit by the other metabolites could predict its labeling. Predicted citrate labeling closely matches that measured. The ELVA plots (García Martín et al., 2015), used to confirm that reactions external to the core set do not significantly contribute to the core labeling, for all strain/condition pairs are presented in Figure S4 in Supplementary Material. Strain/condition pair sip1Δ in 2% glucose + 0.2% galactose exhibited more variability in both its measured and simulated data errors. Also, sip1Δ in 2% glucose + 0.2% galactose had a somewhat worse fit. The whole flux profiles corresponding to these strain/condition pair ELVA plots are displayed in Figures S5–S8 in Supplementary Material for base in 2% glucose, sip1Δ in 2% glucose, base in 2% glucose + 0.2% galactose, and sip1Δ in 2% glucose + 0.2% galactose, respectively. All values are normalized to the absolute glucose uptake rate. As indicated in the legend in the lower-right of the figure, differently colored small arrows indicate the use of particular cofactors. Cofactors displayed are NADPH, NADH, ATP, GLN-L, AKG-L, NADP, NAD, ADP, GLU-L, ACCOA (acetyl-CoA), FOR (formate), CO₂, AMP, and CoASH. Arrows pointing toward the main black reaction arrow indicate the cofactor is a reactant and vice versa.

The presence of galactose appeared to greatly reduce pentose phosphate pathway (PPP) activity. The split between glycolysis and the PPP for strain/condition pairs base in 2% glucose, sip1Δ in 2% glucose, base in 2% glucose + 0.2% galactose, and base in 2% glucose + 0.2% galactose are displayed in Figure 5 and individual flux values and their absolute ranges from the ELVA are presented in Table 5. When switching from Min to Min + Gal the base strain’s PPP activity reduced by about 94% (base in 2% glucose versus base in 2% glucose + 0.2% galactose). Analogously, the PPP flux for the SIP1 null mutant decreased a similar ~93% when galactose was present in the medium (sip1Δ in 2% glucose versus sip1Δ in 2% glucose + 0.2% galactose).

Both the TCA cycle and the glyoxylate shunt, as expected from glucose repression, appeared to be almost completely repressed across all strain/condition pairs (~1%). This is mostly consistent with the ¹³C MFA literature, which indicates a small amount of activity (usually about 1–2% of total glucose consumption flux) in glucose-repressing conditions for CEN.PK113-7D (Gombert et al., 2001; Maaheimo et al., 2001; Blank and Sauer, 2004; Blank et al., 2005). Fluxes surrounding mitochondrial import and export for strain/condition pairs base in 2% glucose, sip1Δ in 2% glucose, base in 2% glucose + 0.2% galactose, and sip1Δ in 2% glucose + 0.2% galactose are displayed in Figures S9–S12 in Supplementary Material, respectively.

The presence of galactose in the medium for either the base or mutant strains, both with galactose metabolism deactivated, appears to greatly increase mitochondrial activity (Figures S9–S12 in Supplementary Material). Neither strain exhibits mitochondrial import of malate or pyruvate in glucose-repressing conditions. Addition of 0.2% galactose to the medium of both strains (base in 2% glucose vs base in 2% glucose + 0.2% galactose and sip1Δ in 2% glucose vs sip1Δ in 2% glucose + 0.2% galactose) resulted in a dramatic import of malate. This malate is fed through the NADP-dependent malic enzyme to generate mitochondrial pyruvate. Similarly, pyruvate import is activated. Finally, this pyruvate generation flux is directed toward branched-chain amino acids, particularly valine. It should be noted, however, that our ability to compare mitochondrial fluxes of the mutant strain in mixed-carbon medium (sip1Δ in 2% glucose + 0.2% galactose) in particular is limited due to rather wide flux confidence intervals resulting from the ¹³C FVA (García Martín et al., 2015). The best fit values, nonetheless, are consistent with these trends.

In the absence of galactose, both strains appear to direct pyruvate flux mainly toward ethanol and aspartate/threonine biosynthesis. The split of pyruvate carboxylase flux toward aspartate/threonine biosynthesis and production of cytosolic malate is displayed in Figure 6 for all strain/condition pairs and individual flux values and their absolute ranges from the ELVA are presented in Table 5. Deletion of SIP1 in glucose-repressing conditions (base in 2% glucose vs sip1Δ in 2% glucose) resulted in a ~35% decrease in pyruvate carboxylase and a ~41% decrease in flux toward aspartate/threonine biosynthesis. The presence of 0.2% galactose in the medium of the base strain (base in 2% glucose vs base in 2% glucose + 0.2% galactose) resulted in a ~10% increase in pyruvate carboxylase activity and a ~80% decrease in flux toward aspartate/threonine biosynthesis. Similarly, adding galactose to the medium of the sip1Δ mutant (sip1Δ in 2% glucose vs sip1Δ in 2% glucose + 0.2% galactose) resulted in a ~32% increase in pyruvate carboxylase flux and a ~66% decrease in flow toward aspartate/threonine biosynthesis. As before, the flux confidence intervals for SG are quite wide, limiting the quality of the inferences for this particular strain/condition.