We selected the ammonia metabolism of S. cerevisiae as a model system for engineering control over fitness tradeoffs. Ammonia is one of the preferred sources of nitrogen for yeast. Uptake of ammonia is performed by three transporters: two high affinity permeases, Mep1p and Mep2p, and one low affinity but high capacity permease, Mep3p 29. Ammonia is used to either perform reductive amination of 2-ketoglutarate to form glutamate, the source of 80% of cellular nitrogen, or to synthesize glutamine from glutamate, which accounts for the other 20% 29. These reactions are performed by two NADPH-dependent glutamate dehydrogenases, Gdh1p and Gdh3p, and a glutamate synthase, Glt1p. Gdh1p provides the primary route of ammonia assimilation under aerobic conditions and is regulated by at least four transcription factors, Gat1p, Gln3p, Gzf3p, and Dal80p (Fig. 1).

The toxic effects of ammonia have been well documented in animals and plants 33; however, recent work by Hess et al demonstrated that yeast are also susceptible to ammonia toxicity 30. Yeast possess mechanisms to excrete excess nitrogen in the form of amino acids, a rudimentary form of ammonia detoxification analogous to urea in mammals. The authors also demonstrated that ammonia toxicity is increased under low potassium ion levels, potentially due to the ability of ammonia ions to enter the cell via potassium channels. Therefore, while ammonia is an essential source of cellular nitrogen, environmental conditions, such as excess ammonia or low potassium, can be toxic to the cells. This paradoxical nature of ammonia (essential but toxic) provides a promising experimental system to examine the synthetic regulation of a fitness tradeoff between resource abundant and resource limited environments.

We implemented synthetic transcriptional control to examine the effects of changing the expression profile of a key enzyme in the ammonia assimilation pathway on the fitness tradeoff exhibited by the population under different ammonia concentrations. Previous studies have shown that mutations in promoter regions can contribute to both abundance and noise profiles in the expression of a downstream gene 34. Therefore, we generated a set of Gdh1p expression mutants by random mutagenesis of the GDH1 promoter in a strain harboring a GDH1:GFP fusion (Fig. 2a). We examined the distribution of noise and mean abundance in Gdh1p expression for a set of randomly selected clones from the mutant library through flow cytometry. The clones exhibited a range of abundance and noise levels in Gdh1p (Fig. 2b), such that sets of strains were identified with constant abundance but different noise levels and constant noise but different abundance levels. Although we noted a slight correlation between noise and abundance, which has been observed in other studies of gene expression variability 2135, we selected sets of mutants exhibiting different abundance levels for further characterization.

We used a previously reported direct competition assay 13 to examine the population fitness of the GDH1 promoter mutant clones under different ammonia concentrations. The relative contributions of noise and abundance in Gdh1p levels to the fitness tradeoff were determined by examining the fitness of sets of strains exhibiting constant abundance but different noise levels and constant noise but different abundance levels. To represent the relative contributions of fitness and stress resistance, we measured fitness via direct competition with a reference strain at both 17 and 556 mM ammonia. We defined a fitness term, W_env, as the ratio of fitness at the higher ammonia concentration to fitness at the lower ammonia concentration, with W_env for a wildtype strain containing the endogenous Gdh1p promoter equal to 1. Thus, clones with W_env values greater than 1 exhibit enhanced growth at high ammonia versus low ammonia concentrations, while clones with W_env values lower than 1 exhibit higher growth at low ammonia than high ammonia concentrations. Constant abundance-varying noise mutant sets with low, medium, and high abundance Gdh1p expression exhibited variations in W_env that correlated with noise values (Fig. 2c). In contrast, W_env did not show any correlation with abundance values under constant noise values (Fig. 2d). Our results demonstrate a correlation between Gdh1p noise and a fitness tradeoff across varying ammonia environments.

We selected two mutants from the GDH1 promoter library exhibiting similar abundances and different noise levels in Gdh1p expression to further examine the effects of Gdh1p noise values on the tradeoff between population growth in normal nitrogen environments and stress resistance in high ammonia environments. The 'high noise' mutant displayed 20% higher noise than the wildtype strain, while the 'low noise' mutant displayed 10% lower noise than the wildtype strain in minimal media with 40 mM ammonia (Fig. 3a). Noise remained unchanged when the mutants were grown in a range of ammonia concentrations (data not shown). We examined resistance to ammonia toxicity by transferring cultures growing in exponential phase to minimal media with high concentrations of ammonia (600 mM) and assaying the change in viable cell counts (measured as colony forming units, CFUs) by plating the cultures on rich media after a short amount of time (30 or 60 minutes; < 1 cell cycle) to observed initial fitness responses rather than long-term metabolic adaptation to ammonia stress. Wildtype cells displayed little change at 30 or 60 minutes after ammonia challenge (Fig. 3b). The high noise strain displayed a slight increase in CFUs after 30 minutes (~10%), whereas the low noise strain showed a marked decrease in CFUs after 30 minutes (~25%). Trends were similar in the low and high noise strains at 60 minutes after ammonia challenge. Results from this assay at later time points showed large variance between replicates, which necessitated a more sensitive assay of fitness as described below. Our results demonstrate that the high noise strain exhibits the greatest ability to adapt to stressful ammonia conditions, while the low noise strain exhibits the lowest resistance to toxic ammonia. These results suggest a correlation between noise in Gdh1p expression and stress resistance.

As previously indicated, ammonia toxicity is enhanced in environments with low amounts of potassium. To further examine the correlation between Gdh1p noise and stress resistance, we challenged the high and low noise strains in media with 600 mM ammonia and low potassium (17 mM), conditions at which ammonia is toxic to the cells. The high noise strain exhibited increases in CFUs at 30 minutes after ammonia challenge, but a decrease in viable cells at 60 minutes (Fig. 3c). In contrast, the low noise strain exhibited a slight loss of viable cells at 30 minutes after ammonia challenge that becomes more pronounced at 60 minutes. While it appeared that both strains were susceptible to the toxic effects of ammonia in low potassium media, the high noise strain displayed a significant lag period before the number of viable cells decreased. Such temporal differences in resistance may be critical in uncertain or fluctuating environments, where a high noise strain may be better able to buffer deleterious fluctuations in ammonia concentration.

We also examined the effects of noise in Gdh1p expression on fitness under 'physiological' ammonia environments. Yeast are commonly grown in 40 mM ammonia in the laboratory, while ammonia concentrations in the natural yeast habitat may be lower 30. Therefore, we measured population growth of the low and high noise strains across a range of low ammonia concentrations. The growth assays that examine increases in CFUs over time exhibited too much experimental error at these ammonia concentrations to reliably measure growth differences between the high and low noise strains and the wildtype strain. Therefore, we used the direct competition growth assay to measure fitness at multiple ammonia concentrations. The low noise strain exhibited similar fitness to wildtype at each assayed ammonia concentration (Fig. 3d). In contrast, the high noise strain exhibited significantly lower fitness than the wildtype and low noise strains at 10 mM and 20 mM ammonia, and similar fitness at 40 mM ammonia. Our results indicate that while the high noise strain exhibited increased resistance to ammonia stress, it also exhibited decreased fitness under lower ammonia concentrations, supporting the role of Gdh1p noise in affecting a tradeoff between stress resistance at high ammonia concentrations and fitness at low ammonia concentrations.

Synthetic networks that enable regulation over fitness tradeoffs in a given environment in response to a small molecule that can be exogenously added to the system, will provide mechanisms through which to tune ecological strategies. Our results demonstrate that Gdh1p noise levels are correlated with the tradeoff in fitness across varying ammonia concentrations. We attempted to build titratable control over the observed fitness tradeoff by manipulating Gdh1p noise. We based our design on a synthetic control system that allows for the manipulation of Gdh1p noise levels in response to an exogenous small molecule. Alterations in the levels of transcriptional regulators have been shown to affect noise levels in regulated genes 1634363738. Gdh1p is regulated by at least four transcription factors: two activators, Gat1p and Gln3p, and two repressors, Gzf3p and Dal80p 31 (Fig. 1). We based our initial system design on the inducible control of the Gdh1p negative regulator, Dal80p. We replaced the endogenous DAL80 promoter with the GAL1-10 promoter through chromosomal integration in a strain background that harbored the GDH1:GFP fusion and a deletion in the galactose transporter Gal2p to achieve galactose-tunable control 32 of Dal80p (Fig. 4a). In our engineering strain, Dal80p expression levels changed linearly with galactose concentration as measured by quantitative real-time PCR, with most galactose concentrations inducing DAL80 levels greater than the wildtype DAL80 levels (Fig. 4b).

We measured the effects of varying Dal80p expression levels on the Gdh1p expression profile through flow cytometry analysis. Mean Gdh1p abundance did not change with varying Dal80p levels (Fig. 4c), indicating that combinatorial interactions at the GDH1 promoter or a complex regulatory network may be associated with this nitrogen assimilation pathway. However, the noise in Gdh1p expression decreased with increasing Dal80p levels (Fig. 4d). Specifically, in our engineered strain, low levels of Dal80p (under the absence of galactose) resulted in ~15% higher noise in Gdh1p expression than the wildtype strain, whereas high levels of Dal80p (in the presence of ~1% galactose) reduced noise to ~80% of the wildtype strain.

We used the galactose-tunable control over Dal80p levels afforded by our engineered strain to measure population fitness at various Dal80p levels across a range of ammonia concentrations. We used the direct competition assay to measure fitness at ammonia concentrations spanning near growth limiting (17 mM) to near toxic conditions (600 mM) 30. At low expression of Dal80p (lower than wildtype expression) the engineered strain displayed lower fitness than the wildtype strain at low ammonia concentrations, and higher fitness with increasing ammonia concentrations (Fig. 5a). In contrast, high Dal80p expression levels led to high relative fitness of the engineered strain at low ammonia concentrations and progressively lower fitness as ammonia concentrations increased. This fitness tradeoff is specific to ammonia as a nitrogen source (Fig. 5b), indicating that our synthetic control system tunes this fitness tradeoff for a specified environmental condition. Our engineered system is based on the modification of noise levels in a pathway enzyme (Gdh1p) through titration of the levels of a transcriptional regulator for that pathway enzyme (Dal80p). Our results demonstrate that our engineered strain can be altered to perform as a superior competitor at either high or low ammonia concentrations through the exogenous addition of a small molecule (Fig. 5c), demonstrating the synthetic tuning of a tradeoff between stress resistance and fitness in a stable environment.

All manipulations were performed with the S288c background. Yeast were grown in synthetic complete media with the nitrogen source as specified. Primer sequences are provided in Additional file 1.

To construct mutant libraries of the GDH1 promoter, primers flanking 500 nucleotides upstream of the GDH1 coding region (1043500 – 1043050, chromosome XV) were used to amplify the fragment from yeast genomic DNA using KOD polymerase (Novagen) according to the manufacturer's instructions. The fragment was then diluted into mutagenic PCR buffer 41 (7 mM MgCl₂, 0.5 mM MnCl₂, 50 mM KCl, 10 mM Tris pH 8.3 with 1 mM dGTP, 0.2 mM dCTP, 0.2 mM dTTP, 0.2 mM dATP) and further amplified using Taq polymerase (Roche) according to the manufacturer's instructions. Separately, the leucine biosynthesis gene (LEU2) was amplified from pRS315 42 using KOD polymerase. The LEU2 gene fragment and the promoter library were then ethanol precipitated, resuspended in water, and PCR assembled together through overlapping primer sequences. The resulting large fragment was then transformed into yeast strains using a standard lithium acetate procedure 43. Transformants were selected in liquid synthetic complete leucine dropout media. The resulting library was grown to stationary phase and frozen in 15% glycerol at -80°C.

Integration of the GAL promoter was performed by amplifying the GAL1-10 promoter sequence from pRS314-Gal 42 using KOD polymerase. This fragment was PCR assembled with the leucine biosynthesis gene (LEU2) from pRS315 42 along with flanking homologous regions to the DAL80 upstream region (506000 – 504030 and 506500 – 506530 on chromosome XI) using KOD polymerase. The construct was transformed into yeast using a standard lithium acetate procedure and colonies were selected on synthetic complete leucine dropout agar plates. Integration was confirmed by colony PCR with primers flanking and internal to the integrated construct. Yeast DNA extraction was performed as previously described using the 'bust n' grab' method 44.

Cells were pelleted and frozen in liquid nitrogen. Pellets were resuspended in a 50 mM NaOAc (pH 5.2), 10 mM EDTA buffer. Cells were lysed by the addition of SDS to a final concentration of 1.6% and an equal volume of acid phenol. Solutions were kept at 65°C with intermittent vortexing for 10 min. After cooling on ice, the aqueous phase was extracted and further extraction was carried out with an equal volume of chloroform. RNA was further isolated and concentrated by use of RNeasy columns (Qiagen) according to manufacturer's instructions. Total RNA was quantified by OD₂₆₀ readings. RNA samples were treated with DNase (Invitrogen) according to manufacturer's instructions. cDNA was synthesized using gene-specific primers and Superscript III reverse transcriptase (Invitrogen) according to the manufacturer's instructions. qRT-PCR was carried out on this cDNA using an iCycler iQ system (BioRAD). Samples were prepared using the iQ SYBR green supermix and primer pairs specific for different templates. Data were analyzed using the iCycler iQ software.

Yeast cells growing in exponential phase (OD₆₀₀~0.5) were spun down, washed with sterile water, and resuspended in synthetic complete media with 600 mM ammonia. An aliquot of this culture was serially diluted (10-fold dilutions) and 50 μL of the dilutions were plated on YPD agar media. Cells were grown on the solid media for 2 days at 30°C and colonies were counted to measure the change in colony forming units (CFUs) over time.

Fitness was assayed by direct competition versus a common reference strain 13. The competitor and reference strain constitutively express different fluorescent proteins (GFP and CFP, respectively) from the ADH1 promoter integrated into the chromosome. The frequency of competitor and reference strains were quantified before and after the growth period by counting the numbers of GFP expressing cells to non-GFP expressing cells. Fitness (w) of the competitor strain is reported as the natural log of the change in frequency of the strain during the competitive growth period versus the change in frequency of the reference strain over the same growth period:

w = ln (Δ_frequency of competitor strain/Δ_frequency of reference strain)

All competitor strains were derivatives of the S288C background, while the reference strain was derived from the W303 background. These strains showed different electronic volume versus side-scatter distributions that can also be used to quantify population numbers, in good agreement with the values obtained from fluorescent measurements. Equal amounts of competitor and reference strain were mixed and grown in indicated liquid media for 3 generations (approximately 6 hours). The frequency of competitor and reference strain were quantified before and after the growth period by counting the numbers of GFP expressing cells to non-GFP expressing cells by flow cytometry using a Quanta SC flow cytometer (Beckman Coulter) equipped with the MPL system. Samples were excited with a 488 nm laser and GFP fluorescence was detected with a 525 nm bandpass filter. A gate was set above the non-GFP expressing cells in the Quanta analysis software to partition fluorescent from non-fluorescent cells. Samples of only reference or competitor strains and serial dilutions of ratios of competitor to reference strains were run in parallel as controls. 5,000 events were collected per sample.

Two gates were used to standardize each cell population for analysis using 'magnetic gating' in FlowJo flow cytometry analysis software (Tree Star, Inc.). The first gate isolated cells displaying regular morphology based on electronic volume and side-scatter, while the second gate removed non-fluorescent cells from the distribution. This gating method was compared against other methods previously described and the abundance and noise trends observed were consistent between methods 1934. Noise was calculated as the square of the coefficient of variation (σ²/p²) of the distribution 19. Abundance was calculated as the mean of the distribution. 50,000 events were analyzed to calculate noise for each sample. Noise trends were similar when calculated as the coefficient of variation (σ/p) and the variance (σ²).