https://orcid.org/0000-0002-8543-5833
Abstract
A key step in metabolic pathway evolution is the recruitment of promiscuous enzymes to perform new functions. Despite the recognition that promiscuity is widespread in biology, factors dictating the preferential recruitment of one promiscuous enzyme over other candidates are unknown. Escherichia coli contains four sugar kinases that are candidates for recruitment when the native glucokinase machinery is deleted—allokinase (AlsK), manno(fructo)kinase (Mak), N-acetylmannosamine kinase (NanK), and N-acetylglucosamine kinase (NagK). The catalytic efficiencies of these enzymes are 103- to 105-fold lower than native glucokinases, ranging from 2,400 M−1 s−1 for the most active candidate, NagK, to 15 M−1 s−1 for the least active candidate, AlsK. To investigate the relationship between catalytic activities of promiscuous enzymes and their recruitment, we performed adaptive evolution of a glucokinase-deficient E. coli strain to restore glycolytic metabolism. We observed preferential recruitment of NanK via a trajectory involving early mutations that facilitate glucose uptake and amplify nanK transcription, followed by nonsynonymous substitutions in NanK that enhance the enzyme's promiscuous glucokinase activity. These substitutions reduced the native activity of NanK and reduced organismal fitness during growth on an N-acetylated carbon source, indicating that enzyme recruitment comes at a cost for growth on other substrates. Notably, the two most active candidates, NagK and Mak, were not recruited, suggesting that catalytic activity alone does not dictate evolutionary outcomes. The results highlight our lack of knowledge regarding biological drivers of enzyme recruitment and emphasize the need for a systems-wide approach to identify factors facilitating or constraining this important adaptive process.
Introduction
Environmental change, including access to new resources, spurs the evolution of new metabolic pathways. A key step in metabolic innovation is enzyme recruitment, whereby an existing catalyst is enlisted to provide a new function (Jensen 1976). Enzyme recruitment facilitates real-world, health-relevant processes such as antibiotic resistance and pollutant bioremediation. A recent example is provided by the emergence of a degradative pathway for the toxic herbicide atrazine following recruitment of two hydrolases from melamine and cyanurate catabolism (Seffernick et al. 2001; Seffernick and Wackett 2001). Comparative phylogenetics of extant pathways also provides evidence of past recruitment events within various central and secondary metabolic processes (Rison and Thornton 2002). Together, these data support the patchwork model of metabolic evolution, in which new pathways are cobbled together from enzymes borrowed from other metabolic pathways (Lazcano and Miller 1996).
Enzyme recruitment is powered by the inherent functional promiscuity of modern enzymes, which allows them to transform multiple substrates and perform multiple types of chemical reactions (Schulenburg and Miller 2014). Experimentalists have cataloged a variety of promiscuous activities hidden within modern proteomes, and estimates suggest that a single bacterium may harbor thousands of promiscuous enzymes lying in wait for future recruitment (Copley 2012; Kim et al. 2019). For example, genome-scale models indicate that ∼40% of Escherichia coli enzymes are generalists, meaning they can perform multiple physiologically relevant transformations inside the cell (Nam et al. 2012). Such widespread promiscuity suggests a multitude of potential solutions to new metabolic challenges. Nevertheless, the solution selected by evolution remains unpredictable, partly because enzyme recruitment occurs amidst a complex and dynamic physiological setting.
To investigate factors that facilitate or constrain enzyme recruitment, we sought to develop an experimental system to evaluate the recruitment process during adaptive evolution. As a model, we chose the critical metabolic enzyme glucokinase. Glucokinase catalyzes the ATP-dependent phosphorylation of glucose in the first step of glycolysis. Nearly all organisms use the glycolytic pathway to extract energy from glucose, and glycolysis was likely among the first pathways to have emerged on our planet (Fothergill-Gilmore and Michels 1993). Furthermore, flux through glycolysis is high, suggesting that the selective pressure to maintain robust operation of the pathway is strong (Court et al. 2015; Park et al. 2016). Within the E. coli proteome, we identified four sugar kinases that are candidates for recruitment when the native glucokinase machinery is deleted (Miller and Raines 2004, 2005). The candidate enzymes include allokinase (AlsK), manno(fructo)kinase (Mak), N-acetylglucosamine kinase (NagK), and N-acetylmannosamine kinase (NanK). The physiological substrates of these enzymes are glucose analogs, indicating that their promiscuous glucokinase activities originate from substrate ambiguity (fig. 1A). The latent glucokinase activities of these four catalysts are 103- to 105-fold less active than native glucokinases; however, plasmid-mediated overexpression of each gene can restore glycolysis to a glucokinase-deficient bacterium (Miller and Raines 2004, 2005).
Biochemical and genomic features of recruitment candidates. (A) Structures and Km values for the native carbohydrate substrates of the four promiscuous glucokinases. Kinetic values are from references Curtis and Epstein (1975), McDonald et al. (1997), and Porco et al. (1997). (B) Transcriptional units for each candidate gene are highlighted in gray. Associated σ factor binding sites for the transcriptional units are labeled followed by their specific identifier. Each unit's approximate location and transcription orientation is indicated by the arrow attached to the associated σ factor. All repressor proteins that are proximal to the transcriptional units are underlined.
Here, we report the results of adaptive evolution experiments investigating the connection between promiscuous activity levels and enzyme recruitment outcomes. We challenged a glucokinase-deficient bacterium to recruit a new enzyme to fill the missing gap in metabolism and enable survival on glucose minimal medium. The results reveal a series of mutational events leading to the preferential recruitment of NanK within 100 generations. During adaptation, we observed the acquisition of two active site substitutions in NanK, L84P and L84R, that enhance the glucokinase activity of the enzyme. Both substitutions decreased the enzyme's native activity and reduced organismal fitness during growth on an N-acetylated carbon source, indicating that NanK recruitment comes at a cost for metabolizing other substrates. Notably, the estimated in vivo glucokinase activity of NanK is lower than two other promiscuous recruitment candidates, NagK and Mak. This observation suggests that promiscuous activity levels alone do not dictate the results of individual enzyme recruitment events. Our findings demonstrate an inability to predict evolutionary outcomes based solely on functional biochemical properties and emphasize the need for a systems-wide approach to identify factors that facilitate or constrain enzyme recruitment.
Results
Biochemical Characteristics of Promiscuous Candidates
Previous studies identified four sugar kinases capable of restoring growth of a glucokinase-deficient bacterium on glucose minimal medium via plasmid-mediated multicopy suppression: AlsK, Mak, NagK, and NanK (Miller and Raines 2004, 2005). To explore the relationship between the relative catalytic efficiencies of promiscuous candidates and their recruitment, the expected glucokinase activities provided by each enzyme were estimated from in vitro biochemical data and in vivo protein concentration measurements. Promiscuous enzymes generally operate under subsaturating substrate concentrations making the second-order rate constant, kcat/Km, the most relevant kinetic parameter to consider when comparing relative activity (Fendt et al. 2010; Nam et al. 2012; Gerosa et al. 2015). In our experimental system, this supposition is supported by the high glucose Km values displayed by each candidate enzyme, which exceed the estimated cellular glucose concentrations in E. coli (∼0.3 mM) (table 1) (Hu et al. 2018). Prior kinetic analyses demonstrate that NagK is the most efficient latent glucokinase, with a kcat/Km value of 2,400 M−1 s−1 (Miller and Raines 2004). In contrast, AlsK is the least efficient glucokinase, displaying a kcat/Km value of 15 M−1 s−1 (Miller and Raines 2005). The catalytic efficiencies of the other two candidates, NanK and Mak, are similar in value, with kcat/Km values of 510 M−1 s−1 and 200 M−1 s−1, respectively (Miller and Raines 2004). By comparison, the kcat/Km value of E. coli glucokinase is 5.4 × 106 M−1 s−1 (Miller and Raines 2004).
Biochemical Features of Promiscuous Candidates.
| Enzyme | kcat (s−1)a | Km (M)a | kcat/Km (M−1 s−1)a | Copy #b | Normalized rate at [glucose]c | ||
|---|---|---|---|---|---|---|---|
| 0.5 mM | 5 mM | 50 mM | |||||
| AlsK | 1.5 | 0.10 | 15 | 7 | 0.0075 | 0.071 | 0.5 |
| Mak | 12 | 0.059 | 200 | 279 | 3.59 | 33.3 | 196 |
| NagK | 9.2 | 0.0038 | 2400 | 131 | 20.0 | 97.8 | 160 |
| NanK | 9.1 | 0.018 | 510 | 14 | 0.49 | 3.96 | 13.4 |
| Enzyme | kcat (s−1)a | Km (M)a | kcat/Km (M−1 s−1)a | Copy #b | Normalized rate at [glucose]c | ||
|---|---|---|---|---|---|---|---|
| 0.5 mM | 5 mM | 50 mM | |||||
| AlsK | 1.5 | 0.10 | 15 | 7 | 0.0075 | 0.071 | 0.5 |
| Mak | 12 | 0.059 | 200 | 279 | 3.59 | 33.3 | 196 |
| NagK | 9.2 | 0.0038 | 2400 | 131 | 20.0 | 97.8 | 160 |
| NanK | 9.1 | 0.018 | 510 | 14 | 0.49 | 3.96 | 13.4 |
Michaelis–Menten parameters (Miller and Raines 2004, 2005).
Based on Ribo-Seq data (Li et al. 2014).
Calculated using normalized protein copy number and Michaelis–Menten parameters.
Biochemical Features of Promiscuous Candidates.
| Enzyme | kcat (s−1)a | Km (M)a | kcat/Km (M−1 s−1)a | Copy #b | Normalized rate at [glucose]c | ||
|---|---|---|---|---|---|---|---|
| 0.5 mM | 5 mM | 50 mM | |||||
| AlsK | 1.5 | 0.10 | 15 | 7 | 0.0075 | 0.071 | 0.5 |
| Mak | 12 | 0.059 | 200 | 279 | 3.59 | 33.3 | 196 |
| NagK | 9.2 | 0.0038 | 2400 | 131 | 20.0 | 97.8 | 160 |
| NanK | 9.1 | 0.018 | 510 | 14 | 0.49 | 3.96 | 13.4 |
| Enzyme | kcat (s−1)a | Km (M)a | kcat/Km (M−1 s−1)a | Copy #b | Normalized rate at [glucose]c | ||
|---|---|---|---|---|---|---|---|
| 0.5 mM | 5 mM | 50 mM | |||||
| AlsK | 1.5 | 0.10 | 15 | 7 | 0.0075 | 0.071 | 0.5 |
| Mak | 12 | 0.059 | 200 | 279 | 3.59 | 33.3 | 196 |
| NagK | 9.2 | 0.0038 | 2400 | 131 | 20.0 | 97.8 | 160 |
| NanK | 9.1 | 0.018 | 510 | 14 | 0.49 | 3.96 | 13.4 |
Michaelis–Menten parameters (Miller and Raines 2004, 2005).
Based on Ribo-Seq data (Li et al. 2014).
Calculated using normalized protein copy number and Michaelis–Menten parameters.
The capacity of each candidate to fill the metabolic gap caused by glucokinase deletion depends not only on catalytic efficiency but also on the stability and quantity of each enzyme available within the cell to perform the promiscuous function (Park et al. 2016; Hanson et al. 2021). The ProtParam tool on the ExPASy server predicts half-lives > 10 h for each promiscuous glucokinase in E. coli, suggesting that candidate lifetime is unlikely to contribute to recruitability in this system (Gasteiger et al. 2005). The Km values for the native substrates of AlsK, NagK, and NanK span a relatively narrow range of 0.19–0.36 mM, whereas the Km value of Mak acting on fructose is 1.3 mM (fig. 1A). The ability of glucose to effectively compete with native substrates for access to candidate active sites can be gauged by comparing these Km values with the physiological concentrations of the native substrates. Unfortunately, the intracellular concentrations of the native substrates are unknown. However, metabolite concentrations meet or exceed Km values for many substrate–enzyme pairs in glucose-fed E. coli cells, suggesting that substrate competition could impact any candidate whose native substrate is present at appreciable levels (Park et al. 2016).
Ribosome profiling (Ribo-Seq) experiments performed on E. coli grown in glucose minimal medium provide protein copy numbers of all four enzymes (Li et al. 2014). These studies demonstrate that two candidates, Mak and NagK, appear to be constitutively expressed during growth on glucose minimal medium, with per cell copy numbers of 279 and 131, respectively. These data are consistent with the fact that nagK and mak are not located within an operon and do not possess specific transcriptional repressors (fig. 1B) (Sproul et al. 2001; Castaño-Cerezo et al. 2011). In contrast, the alsK and nanK genes are located within dedicated operons, with associated repressor proteins encoded by the alsR and nanR loci, respectively (fig. 1B) (Vimr and Troy 1985; Kim et al. 1997; Sproul et al. 2001; Kalivoda et al. 2003). Although alsK transcription is reportedly not under the control of the AlsR protein (Poulsen et al. 1999), AlsK is present at low levels during growth on glucose minimal medium. Incorporating the estimates of protein concentration provided by Ribo-Seq data with previously determined Michaelis–Menten parameters for each enzyme, we calculated the normalized rate of glucose phosphorylation expected from each promiscuous candidate at three different glucose concentrations (table 1) (Miller and Raines 2004, 2005). The data suggest that in the absence of any competing substrate, NagK provides the highest predicted level of glucokinase activity in cells, with Mak ranking second, NanK ranking third, and AlsK providing the lowest level of promiscuous activity. If relative activity was the primary factor determining recruitment outcome, these results suggest that NagK would be the preferred candidate to replace a missing native glucokinase.
An Experimental System to Investigate Recruitment
Wild-type E. coli strain BW25113 was cultured for 500 generations in 0.2% glucose minimal medium to identify adaptive mutations associated with laboratory propagation. The maximum growth rate was extracted from growth curves at specific generations to assess fitness changes during adaptation. A comparison of maximum growth rates revealed a plateau in growth rate, with no further changes after generation 400 (supplementary fig. S1, Supplementary Material online). Based on these results, adaptation to laboratory conditions was deemed complete after 500 generations. Whole-genome sequencing of the resulting strain (M9.2.50) revealed several mutations associated with acclimation to laboratory conditions (supplementary Dataset 1, Supplementary Material online). Mutations that reached fixation in the population include mutations within the intergenic region of the rph gene, which encodes a defective ribonuclease PH, and mutations in the gene coding for RNA polymerase B (rpoB). Prior experiments in other E. coli strains have observed similar mutations at these loci following laboratory adaption in glucose minimal medium (Knöppel et al. 2018).
To determine whether the four promiscuous enzymes described above are viable candidates for recruitment during adaptive evolution, a glucokinase-deficient bacterium was challenged for survival with glucose as the sole carbon source. E. coli utilizes two protein-based systems to phosphorylate glucose. The phosphoenolpyruvate-dependent phosphotransferase (PTS) system, encoded by the ptsH, ptsI, and crr genes, phosphorylates glucose concurrent with transport into the cell via a phosphoryl-group transfer cascade (Postma et al. 1993). E. coli also contains a cytosolic glucokinase, encoded by glk, responsible for ATP-dependent phosphorylation of glucose liberated from other intracellular sources (Curtis and Epstein 1975). To generate a glucokinase-deficient host, the complete pts operon and the glk gene were deleted from our lab-adapted strain to produce BW25113(Δpts, Δglk). This strain failed to produce colonies on M9 agar supplemented with 0.2% glucose and did not produce a color change during growth on glucose-supplemented MacConkey agar, confirming its inability to efficiently metabolize glucose. Whole-genome sequencing validated the genotype of this strain and revealed an additional truncation mutation in the adenylate cyclase encoding cyaA gene that appeared during strain construction. The cyaA mutation removes the enzyme's regulatory C-terminal domain, resulting in amplified cyclic adenosine monophosphate (cAMP) production (Roy et al. 1983; Aiba et al. 1984; Inada et al. 1996). Similar cyaA mutations have been previously observed in PTS-deficient E. coli strains (Crasnier et al. 1994).
Adaptive Evolution Recruits NanK to Restore Glycolytic Metabolism
Adaptive evolution was conducted on BW25113(Δpts, Δglk) by culturing two biological replicates of this strain at 37 °C in M9 minimal medium supplemented with 0.2% glucose. The strain's ability to metabolize glucose was monitored using a phenotypic assay based upon growth in medium containing the pH-sensitive indicator neutral red. Glucose metabolism causes acidification of the growth medium, which can be quantified by monitoring the increase in absorbance at 522 nm (supplementary fig. S2A, Supplementary Material online). We performed adaptive evolution until the neutral red phenotype of BW25113(Δpts, Δglk) returned to the level of the progenitor strain, M9.2.50. A steady increase in glucose metabolism, as reflected by red medium color, was apparent along the evolutionary trajectory. By generation 100 glucose metabolism had been restored to wild-type levels in both biological replicates, at which point adaptation was deemed complete (fig. 2A; supplementary fig. S2B, Supplementary Material online).
Mutational trajectories and fitness parameters during adaptive evolution. Adaptive trajectory of strain BW25113(Δpts, Δglk) as measured by reconstitution of the (A) glucokinase phenotype with associated mutations and (B) fitness measured as growth rates. (C, D) Strains BW25113(Δpts, Δglk, Δmak) and (E, F) BW25113(Δpts::kanR, Δglk) display a similar adaptive trajectory. Mutations observed from whole-genome sequencing are annotated at the respective generation when they appear at 20% or greater frequency in the population. Points represent averages (n = 3) ± SD. Dashed lines indicate corresponding parameters for strain M9.2.50.
To associate the observed increase in glucose metabolism with specific mutational events, we performed whole-genome sequencing of evolved BW25113(Δpts , Δglk) at generations 10, 25, 50, 75, and 100. Our sequencing approach gave a robust, 84- to 254-fold genome coverage. Read coverage mapping of the coding sequences of the four candidate enzymes demonstrated that there were no gene duplication events that could lead to increased activity (supplementary fig. S3A&B, Supplementary Material online). A polymorphism cutoff filter of 0.2 (20%) was used to identify high-frequency mutations in each replicate population (Miura et al. 2021). The resulting adaptive mutational trajectory provided specific evidence of NanK recruitment (fig. 2A). The individual steps associated with recruitment were strikingly similar in each replicate. In early generations, we observed chromosomal mutations correlated with increased transcription of genes associated with glucose uptake and the N-acetylneuraminate (nan) operon, which contains nanK (Vimr and Troy 1985). At later generations, we observed nonsynonymous substitutions that enhanced the promiscuous glucokinase activity of NanK.
We also quantified growth rates extracted from growth curves at generations spanning the adaptive trajectory to correlate changes in glucose metabolism and mutational events with organismal fitness. The results revealed an increase in maximum growth rate by the end of the adaptive evolution (fig. 2B). The growth rate of adapted BW25113(Δpts, Δglk) did not reach the maximum growth rate of the progenitor lab–adapted strain M9.2.50 by generation 100, despite the restoration of glucose metabolism. This observation is not surprising since individual components of the PTS system, most notably crr, are involved in multiple regulatory processes that extend beyond glucose phosphorylation (Postma et al. 1993; Crasnier et al. 1994). The established and putative biochemical consequences of the specific mutational events observed during adaptive evolution of BW25113(Δpts , Δglk) are summarized below (a list of all mutations and their associated frequencies is provided in supplementary Dataset 1, Supplementary Material online).
Mutations Altering Promiscuous Candidate Transcription
Mutations associated with alterations in transcriptional levels were among the earliest changes observed during the adaptive metabolic evolution of BW25113(Δpts, Δglk). In both biological replicates, we observed a seven-base–pair deletion in the nanR gene at generation 10. NanR is the transcriptional repressor of the sialic acid catabolism operon, and the observed deletion is postulated to abolish NanR function, resulting in elevated transcription of downstream genes including nanK (Kalivoda et al. 2003, 2013). To test this hypothesis, we performed reverse transcription quantitative real-time polymerase chain reaction (RT-qPCR) analysis on the nanK locus using RNA isolated from the population samples. The results reveal a statistically significant increase in nanK transcription at generation 10 compared with the parental strain (fig. 3A). Interestingly, elevated nanK transcription returned to a level comparable to the progenitor strain by generation 100, even though the nanR deletion persisted in the population (fig. 3B).
Transcriptional changes during adaptive evolution. Changes in gene expression for promiscuous candidates at generations where mutations associated with nanK transcription manifest for two biological replicates of (A) BW25113(Δpts, Δglk), (C) BW25113(Δpts, Δglk, Δmak), and (E) BW25113(Δpts::kanR, Δglk). Changes in gene expression for promiscuous candidates at generation 100 for two biological replicates of (B) BW25113(Δpts, Δglk), (D) BW25113(Δpts, Δglk, Δmak), and (F) BW25113(Δpts::kanR, Δglk). Bars represent averages (n = 3) ± SD. Relative expression was normalized to cysG (Zhou et al. 2011). ****P < 0.0001, ***P < 0.001, **P < 0.01, and *P < 0.05, as determined by two-tailed Student's t-test.
Transcriptional changes in the three other candidate enzymes during adaptive evolution were also investigated via RT-qPCR using RNA isolated from the population samples. The results revealed smaller magnitude reductions in mak transcription and elevations in alsK transcription at early generations. At generation 100, transcription of mak returned to near wild-type levels whereas alsK transcription remained elevated. Given the very low normalized rate of glucose phosphorylation provided by AlsK (table 1), the small increase in alsK transcription is unlikely to substantially impact glucose processing. Unlike nanK, no mutations were observed near the alsK or mak loci. A predicted binding site for the global σ32 transcriptional factor is upstream of the alsK gene (fig. 1B). Past investigations have shown genes possessing σ32 binding sites are often overexpressed during exposure to stress conditions (VanBogelen et al. 1987; Jenkins et al. 1991). To test whether this characteristic might explain the observed changes in alsK transcription, we quantified transcriptional levels of yiaA, another gene under σ32 factor control. The results indicate that yiaA transcriptional changes mirror those of alsK during adaptation (supplementary fig. S4, Supplementary Material online). Together, these results are consistent with prior studies demonstrating that changes in protein abundance, which may be mediated by direct and/or global transcriptional derepression, are a common first step of adaptive evolution (Jensen 1976).
Mutational Optimization of Glucose Transport
Whole-genome sequencing of evolved populations following 25 and 50 generations of adaptation revealed mutations attributable to enhanced glucose uptake (fig. 2A). Frameshift/nonsense mutations in galR appeared at generation 25. GalR is the transcriptional repressor of the galactose:H+ symporter GalP, a protein previously shown to promiscuously transport glucose (Macpherson et al. 1983; Weickert and Adhya 1993; McDonald et al. 1997; Patching, Henderson, et al. 2008; Patching, Psakis, et al. 2008). Derepression of this gene product via inactivation of GalR is expected to increase glucose transport capabilities. Nonsynonymous mutations appear in galP itself at generation 50. AlphaFold modeling of the GalP structure, combined with prior biochemical studies of this protein, indicates that the two resultant amino acid substitutions (T265A and N268D) are located near two tryptophan residues previously implicated in sugar recognition (supplementary fig. S5, Supplementary Material online) (Patching, Henderson, et al. 2008; Tunyasuvunakool et al. 2021; Varadi et al. 2022). We postulate that these substitutions enhance the promiscuous transport of glucose by GalP. Prior studies have reported galP mutations in other PTS-deficient strains following growth on glucose minimal medium, which is consistent with the hypothesis that the mutations observed in this study facilitate glucose uptake (Hernández-Montalvo et al. 2003; Báez-Viveros et al. 2007; Sawisit et al. 2015; Jung et al. 2019; Minliang et al. 2021).
Active Site Mutations Enhancing NanK's Promiscuous Activity
We observed two different point mutations at the same nucleotide within the coding sequence of nanK at generation 75. In one replicate a thymine-to-guanine transversion occurred at this position, whereas a thymine-to-cytosine transition was observed in the second replicate. These mutations result in the nonsynonymous substitution of Leu84 with arginine or proline, respectively. The L84R transversion mutation reached fixation in the replicate one population by generation 100, whereas the L84P transition mutation reached a frequency of 30.5% within the replicate two population. To quantify changes in the promiscuous and native activities of NanK caused by these substitutions, we assessed their functional impact using enzyme assays (table 2; supplementary fig. S6–S11, Supplementary Material online). Compared with wild-type NanK, the glucokinase catalytic efficiencies of the L84R and L84P variants increase by 6.1-fold and 6.9-fold, respectively. In both variants, the increase in kcat/Km for glucose phosphorylation stems from small increases in kcat values and small decreases in glucose Km values. The L84R and L84P substitutions reduce the kcat/Km value for phosphorylation of the native substrate N-acetylmannosamine by 1.5-fold and 3.0-fold, respectively. Thus, the L84R and L84P substitutions enhance the promiscuous activity of NanK with a modest trade-off in native function. Notably, the L84P variant was previously identified from error-prone PCR libraries designed to identify individual NanK variants with the highest glucokinase activities (Larion et al. 2007). A multiple sequence alignment demonstrates that residue 84 is highly variable among NanK homologs (supplementary fig. S12A, Supplementary Material online) (Conejo et al. 2010). A comparison of the unliganded E. coli NanK structure with substrate-bound homologs from other bacterial species demonstrates that this position is located on a loop near the enzyme active site (supplementary fig. S12B, Supplementary Material online), suggesting it may participate in substrate discrimination (Coombes et al. 2020; Gangi Setty et al. 2020).
Kinetic Parameters for Substrate Phosphorylation by NanK Variants.
| Parameter | Wild-type NanK | L84P NanK | L84R NanK |
|---|---|---|---|
| N-Acetylmannosamine | |||
| kcat (s−1) | 29 ± 0.5 | 13 ± 0.2 | 17 ± 0.3 |
| Km, sugar (M) | (2.1 ± 0.3)×10−4 | (2.8 ± 0.4)×10−4 | (1.8 ± 0.3)×10−4 |
| Km, ATP (M) | (1.4 ± 0.2)×10−4 | (1.4 ± 0.3)×10−4 | (8.3 ± 1.4)×10−5 |
| kcat/Km, sugar (M−1 s−1)a | (1.4 ± 0.2)×105 | (4.6 ± 0.7)×104 | (9.4 ± 1.6)×104 |
| Glucose | |||
| kcat (s−1) | 13 ± 0.2 | 34 ± 0.8 | 37 ± 0.3 |
| Km, sugar (M) | (1.5 ± 0.3)×10−2 | (5.7 ± 1.2)×10−3 | (7.0 ± 0.4)×10−3 |
| Km, ATP (M) | (1.2 ± 0.3)×10−3 | (9.8 ± 1.3)×10−4 | (9.7 ± 0.2)×10−4 |
| kcat/Km, sugar (M−1 s−1)a | (8.7 ± 1.7)×102 | (6.0 ± 1.3)×103 | (5.3 ± 0.3)×103 |
| Parameter | Wild-type NanK | L84P NanK | L84R NanK |
|---|---|---|---|
| N-Acetylmannosamine | |||
| kcat (s−1) | 29 ± 0.5 | 13 ± 0.2 | 17 ± 0.3 |
| Km, sugar (M) | (2.1 ± 0.3)×10−4 | (2.8 ± 0.4)×10−4 | (1.8 ± 0.3)×10−4 |
| Km, ATP (M) | (1.4 ± 0.2)×10−4 | (1.4 ± 0.3)×10−4 | (8.3 ± 1.4)×10−5 |
| kcat/Km, sugar (M−1 s−1)a | (1.4 ± 0.2)×105 | (4.6 ± 0.7)×104 | (9.4 ± 1.6)×104 |
| Glucose | |||
| kcat (s−1) | 13 ± 0.2 | 34 ± 0.8 | 37 ± 0.3 |
| Km, sugar (M) | (1.5 ± 0.3)×10−2 | (5.7 ± 1.2)×10−3 | (7.0 ± 0.4)×10−3 |
| Km, ATP (M) | (1.2 ± 0.3)×10−3 | (9.8 ± 1.3)×10−4 | (9.7 ± 0.2)×10−4 |
| kcat/Km, sugar (M−1 s−1)a | (8.7 ± 1.7)×102 | (6.0 ± 1.3)×103 | (5.3 ± 0.3)×103 |
Values are means (n = 6) averaged from two independent preparations (n = 3 per preparation) ± SD.
SD for these values are propagated using the following equation: , with“” representing the SD of the associated value.
Kinetic Parameters for Substrate Phosphorylation by NanK Variants.
| Parameter | Wild-type NanK | L84P NanK | L84R NanK |
|---|---|---|---|
| N-Acetylmannosamine | |||
| kcat (s−1) | 29 ± 0.5 | 13 ± 0.2 | 17 ± 0.3 |
| Km, sugar (M) | (2.1 ± 0.3)×10−4 | (2.8 ± 0.4)×10−4 | (1.8 ± 0.3)×10−4 |
| Km, ATP (M) | (1.4 ± 0.2)×10−4 | (1.4 ± 0.3)×10−4 | (8.3 ± 1.4)×10−5 |
| kcat/Km, sugar (M−1 s−1)a | (1.4 ± 0.2)×105 | (4.6 ± 0.7)×104 | (9.4 ± 1.6)×104 |
| Glucose | |||
| kcat (s−1) | 13 ± 0.2 | 34 ± 0.8 | 37 ± 0.3 |
| Km, sugar (M) | (1.5 ± 0.3)×10−2 | (5.7 ± 1.2)×10−3 | (7.0 ± 0.4)×10−3 |
| Km, ATP (M) | (1.2 ± 0.3)×10−3 | (9.8 ± 1.3)×10−4 | (9.7 ± 0.2)×10−4 |
| kcat/Km, sugar (M−1 s−1)a | (8.7 ± 1.7)×102 | (6.0 ± 1.3)×103 | (5.3 ± 0.3)×103 |
| Parameter | Wild-type NanK | L84P NanK | L84R NanK |
|---|---|---|---|
| N-Acetylmannosamine | |||
| kcat (s−1) | 29 ± 0.5 | 13 ± 0.2 | 17 ± 0.3 |
| Km, sugar (M) | (2.1 ± 0.3)×10−4 | (2.8 ± 0.4)×10−4 | (1.8 ± 0.3)×10−4 |
| Km, ATP (M) | (1.4 ± 0.2)×10−4 | (1.4 ± 0.3)×10−4 | (8.3 ± 1.4)×10−5 |
| kcat/Km, sugar (M−1 s−1)a | (1.4 ± 0.2)×105 | (4.6 ± 0.7)×104 | (9.4 ± 1.6)×104 |
| Glucose | |||
| kcat (s−1) | 13 ± 0.2 | 34 ± 0.8 | 37 ± 0.3 |
| Km, sugar (M) | (1.5 ± 0.3)×10−2 | (5.7 ± 1.2)×10−3 | (7.0 ± 0.4)×10−3 |
| Km, ATP (M) | (1.2 ± 0.3)×10−3 | (9.8 ± 1.3)×10−4 | (9.7 ± 0.2)×10−4 |
| kcat/Km, sugar (M−1 s−1)a | (8.7 ± 1.7)×102 | (6.0 ± 1.3)×103 | (5.3 ± 0.3)×103 |
Values are means (n = 6) averaged from two independent preparations (n = 3 per preparation) ± SD.
SD for these values are propagated using the following equation: , with“” representing the SD of the associated value.
To investigate whether the decrease in native NanK function caused by the L84P and L84R substitutions impacts organismal fitness during growth on N-acetylated carbon sources, we performed growth curve analysis of BW25113(Δpts, Δglk) in N-acetylneuraminate minimal medium at generations 10, 25, 50, 75, and 100. Due to inefficient transport, E. coli cannot grow on the physiological substrate of NanK, N-acetylmannosamine (Plumbridge and Vimr 1999). However, E. coli can efficiently transport and survive on N-acetylneuraminate, which is converted into the NanK substrate via N-acetylneuraminate lyase (Plumbridge and Vimr 1999; Park and Uehara 2008). The maximum growth rate of BW25113(Δpts, Δglk) on N-acetylneuraminate remained unchanged from the parental strain at generations 10, 25, and 50 for both biological replicates (fig. 4A); however, the growth rate began to decrease at generation 75, the point in time when the L84P or L84R substitutions began to appear in both replicate populations. By generation 100, when the glucokinase-enhancing substitutions are fixed in replicate one and reach a frequency of 30.5% in replicate two, a further decrease in fitness on N-acetylneuraminate was observed. These results demonstrate that the acquisition of active site substitutions associated with the recruitment of NanK to fulfill the glucokinase deficiency of BW25113(Δpts, Δglk) appears to cause a decrease in organismal fitness on N-acetylneuraminate.
Fitness changes during growth on N-acetylneuraminate. Maximum growth rate of two biological replicates of (A) strain BW25113(Δpts, Δglk), (B) BW25113(Δpts, Δglk, Δmak), and (C) BW25113(Δpts::kanR, Δglk) during growth in medium supplemented with 3 mM N-acetylneuraminate, the metabolic precursor of NanK's native substrate. Mutations observed from whole-genome sequencing are annotated at the respective generation when they appear at 20% or greater frequency in the population. Points represent averages (n = 3) ± SD. Dashed lines indicate corresponding parameters for M9.2.50.
Impact of Mak Removal upon Enzyme Recruitment
Previous work uncovered a promoter-up mutation in the 5′ untranslated region of mak that restored growth on glucose minimal medium to an E. coli strain harboring pts operon and glk deletions (Miller and Raines 2005). Thus, it is notable that we failed to detect point mutations near the mak loci at any frequency in any generation in either biological replicate. To investigate the impact of mak upon adaptive evolution of BW25113(Δpts, Δglk), we deleted the gene encoding this promiscuous candidate enzyme. Whole-genome sequencing validated the genotype of this strain. We identified a single, likely silent, synonymous mutation in the transcriptional repressor yegW that occurred during the generation of this mak deletion strain (Pérez-Rueda et al. 2004). We repeated adaptive evolution experiments on the BW25113(Δpts, Δglk, Δmak) strain for two biological replicates using the same growth conditions described above. The sequencing results of these experiments, again filtered using a polymorphism cutoff value of 0.2 to identify high-frequency mutations with 101- to 198-fold genome coverage, revealed a mutational trajectory very similar to that observed during BW25113(Δpts, Δglk) adaption (fig. 2C and D, supplementary Dataset 1, Supplementary Material online). Additionally, no gene duplication events that could lead to increased activity were detected in any of the latent glucokinases (supplementary fig. S3CandD, Supplementary Material online).
A key difference in the adaptation trajectory of these strains is the rapid accumulation of numerous mutations in genes associated with translation within one of the BW25113(Δpts, Δglk, Δmak) replicates, specifically the genes coding for the 23S ribosomal RNA of the rrnA operon (rrlA) and the translation elongation factor Tu 2 (tufB). Further, the mutations associated with translation leave the population by generation 25, corresponding with the appearance of mutations in previously discussed genes associated with the transport of glucose and the regulation of nanK transcription. Prior work investigating the dynamics of E. coli adaptation has demonstrated that E. coli can tune their mutation rate to rapidly adapt to environmental stressors (Wielgoss et al. 2013; Swings et al. 2017). Furthermore, under carbon-limiting conditions, E. coli has been shown to rapidly alter its translational strategy to a paradigm that slows growth in nutrient-poor environments (Li et al. 2018; Biselli et al. 2020; Zhang et al. 2022). Taking literature evidence into context, we postulate that the observed rapid accumulation of mutations in genes associated with translation results from the cells adapting to the stress of growth in a carbon-limited environment. We further postulate that these mutations leave the population when mutations associated with the transport of glucose and the regulation of nanK transcription manifest and partially alleviate the stress of growing in a carbon-limited environment.
By generation 10, one replicate of BW25113(Δpts, Δglk, Δmak) accumulated the same seven-base–pair deletion in the nanR gene previously observed in BW25113(Δpts, Δglk). In addition, the population also accumulated a nonsynonymous R128G substitution in nanR. The functional consequence of this substitution is unknown; however, this residue resides within the predicted ligand binding domain of the repressor and may alter effector binding (Horne et al. 2021). BW25113(Δpts, Δglk, Δmak) replicate two accumulated a nine-base–pair deletion in the predicted NanR binding site within the nan operon at generation 10 (Kalivoda et al. 2013). RT-qPCR analysis using RNA isolated from the population samples demonstrated that nanK transcription is elevated at generation 10 in both replicates and returned to near wild-type levels by generation 100. Transcriptional analysis also revealed changes in alsK transcription during BW25113(Δpts, Δglk, Δmak) adaptation (fig. 3C and D), which mirrors the timing and magnitude of alsK transcriptional activation observed in BW25113(Δpts, Δglk). RT-qPCR analysis using RNA isolated from the population samples also revealed a decreased level of nagK transcription at generation 10 of both replicate populations, which is quantitatively different from the unchanged levels of nagK in BW25113(Δpts, Δglk). Reduced transcription of nagK remained apparent at BW25113(Δpts, Δglk, Δmak) generation 100.
By generation 10, one replicate of BW25113(Δpts, Δglk, Δmak) accumulated two nonsynonymous mutations in galR, V36L and L43P, both of which colocalize within the helix–turn–helix DNA–binding domain of the gene product (Geanacopoulos and Adhya 1997). Similarly, a nonsynonymous mutation (V21F) and a 33-base–pair deletion in galR were also observed in biological replicate 2 at generation 10. At this generation, we also observe a mutation in the repressor binding site upstream of gntT, which encodes the high-affinity gluconate transporter (Porco et al. 1997). Overexpression of GntT has been previously observed in E. coli strains suffering from impaired glucose transport (Jung et al. 2019). Collectively, these mutations are postulated to increase glucose uptake, akin to the galR and galP mutations that appeared in evolved BW25113(Δpts, Δglk) populations.
At generation 25, we observed a 12-base–pair deletion in the fpb gene, which encodes the key gluconeogenic enzyme fructose-1,6-bisphosphatase 1 (FBP) (Fraenkel and Horecker 1965). This deletion removes four amino acids from a loop previously implicated in the allosteric regulation of FBP activity by AMP (Nelson, Iancu, et al. 2000; Nelson, Choe, et al. 2000; Hines et al. 2007). Modulation of FBP activity has been reported in other engineered E. coli strains following modification of GalP activity. Thus, we postulate that the fbp deletion observed in BW21153(Δpts, Δglk, Δmak) is associated with enhanced GalP-based glucose transport (Báez-Viveros et al. 2007; Jung et al. 2019). One BW25113(Δpts, Δglk, Δmak) replicate also accumulated a frameshift mutation in the nagC gene at generation 75, which is expected to abolish the function of the encoded N-acetylglucosamine transcriptional regulator (NagC). NagC inhibits transcription of galP; thus, this mutation may also be associated with enhanced glucose uptake via derepression of galP (El Qaidi and Plumbridge 2008). Finally, both BW21153(Δpts, Δglk, Δmak) replicates accumulated the same L84R glucokinase-enhancing substitution in NanK by generation 50. This mutation reaches fixation in both populations by generation 100. In summary, despite differences in the identity of specific mutations observed during BW25113(Δpts, Δglk, Δmak) adaptation, the overall mutational trajectory is similar to BW25113(Δpts, Δglk). Namely, early adaptive mutations amplify nanK transcription and facilitate glucose uptake, whereas later mutations cause a glucokinase-enhancing substitution in NanK. Similarly, the appearance of the glucokinase-enhancing NanK substitution in BW21153(Δpts, Δglk, Δmak) correlates with a decrease in organismal fitness during growth on N-acetylneuraminate (fig. 4B).
Impact of Antibiotic Presence upon Enzyme Recruitment
A critical difference that separates this work from the previous work that uncovered a promoter-up mutation in the 5′ untranslated region of mak is the absence of antibiotics in our adaptation medium (Miller and Raines 2005). Antibiotics are known to enhance mutation rates and alter the spectra of mutations sampled in a growing bacterial population (Long et al. 2016; Hickman et al. 2017; Blázquez et al. 2018). Thus, to investigate the impact of antibiotics upon the adaptive evolution of BW25113(Δpts, Δglk), we retained the kanamycin resistance-encoding aminoglycoside PTS gene in place of the pts operon following lambda red recombineering. By retaining this gene, kanamycin could be included in the medium throughout the adaptation experiment. Whole-genome sequencing validated the genotype of this strain. While picking individual clonal isolates to use as replicate starting strains for the adaptive evolution experiments, we identified a genetic variation within one replicate that was not observed in any other strain used in this study. The observed additional mutation for this replicate was an 11-base–pair deletion mutation in the gene ptsP that results in the early termination and removal of the C-terminal domain of Enzyme I of the nitrogen PTS system (PTSNtr) (Rabus et al. 1999). This mutation is expected to influence the ability of E. coli to efficiently process glucose, as prior mutagenesis studies have demonstrated the role of this enzyme in glucose processing (Deutscher et al. 2006).
We repeated adaptive evolution experiments on the BW25113(Δpts::kanR, Δglk) strain for two biological replicates using the same growth conditions described above, except kanamycin which was also supplied in the medium. The sequencing results of these experiments, again filtered using a polymorphism cutoff value of 0.2 to identify high-frequency mutations with 93- to 201-fold genome coverage, revealed a mutational trajectory similar to what was observed during BW25113(Δpts, Δglk) adaption (fig. 2E and F; supplementary Dataset 1, Supplementary Material online). Additionally, no gene duplication events that could lead to increased activity were detected in any of the latent glucokinases (supplementary fig. S3EandF, Supplementary Material online). By generation 10, both replicates of BW25113(Δpts::kanR, Δglk) also accumulated mutations in the same genes associated with translation observed in one replicate of BW25113(Δpts, Δglk, Δmak). Further, the mutations associated with translation followed the same trend of disappearing when mutations in genes associated with the transport of glucose and the regulation of nanK transcription appeared.
By generation 50, both replicates of BW25113(Δpts::kanR, Δglk) had accumulated various mutations in nanR that reached fixation. Replicate one accumulated the same seven-base–pair deletion in the nanR gene previously observed in BW25113(Δpts, Δglk), while replicate two accumulated a nonsynonymous K30T mutation in nanR. The functional consequence of this substitution is unknown; however, this residue resides within the HTH gntR-type DNA-binding domain and is predicted to disrupt the binding of nanR repressor to the regulatory site of the nan operon and increase expression of nanK (Kalivoda et al. 2013). By generation 100, the same seven-base–pair deletion in the nanR gene previously observed in BW25113(Δpts, Δglk) began manifesting. RT-qPCR analysis using RNA isolated from the population samples demonstrated that nanK transcription is elevated at generation 50 in both replicates (fig. 3E). RT-qPCR analysis using RNA isolated from the population samples also revealed an increased level of transcription at generation 50 for all the latent glucokinases within both replicate populations, which is quantitatively different from the unchanged or repressed levels of nagK and mak observed in BW25113(Δpts, Δglk). By generation 75, replicate one of BW25113(Δpts::kanR, Δglk) accumulated the same L84R mutation observed in BW25113(Δpts, Δglk), which reached fixation by generation 100. Conversely, replicate two of BW25113(Δpts::kanR, Δglk) did not accumulate any mutations within nanK by the end of 100 generations of adaptation. Additional RT-qPCR analysis using RNA isolated from the population samples at generation 100 for both replicates of BW25113(Δpts::kanR, Δglk) revealed a decreased transcription level of nanK for replicate one compared with generation 50 (fig. 3F). In contrast, the transcription level of nanK in replicate two remained elevated (fig. 3F). These findings suggest that the observed nanK mutation could be inducing a change in the transcription level of the gene, implying that a post-transcriptional mechanism of gene control, a mechanism observed in bacteria in response to stress, could be tuning the fitness of the organism in the aftermath of recruitment (Vargas-Blanco and Shell 2020).
By generation 10, both replicates of BW25113(Δpts::kanR, ΔglK) accumulated the same mutation in the repressor binding site upstream of gntT previously observed in BW25113(Δpts, Δglk, Δmak) that is expected to influence glucose transport (Jung et al. 2019). Additionally, replicate two also accumulated a nonsynonymous L135M substitution in crp, which codes for the cAMP-activated global transcription factor (Buchet et al. 1999). The functional consequence of this substitution is unknown; however, this residue is proximal to a residue that enables crp to utilize cytidine and uridine in the absence of cAMP (Lauritsen et al. 2021). By generation 50, both strains accumulated mutations within the galS and galR genes. Like the previously discussed GalR, GalS is also a transcriptional repressor of GalP (Geanacopoulos and Adhya 1997). Thus, as with the aforementioned mutations in galR, these observed mutations are expected to derepress galP expression via inactivation of GalR and GalS, leading to an increase in glucose transport capabilities. In summary, despite the presence of kanamycin during adaptation and differences in the identity of specific mutations observed during BW25113(Δpts::kanR, Δglk) adaptation, the overall mutational trajectory is similar to BW25113(Δpts, Δglk). Namely, early adaptive mutations amplify nanK transcription and facilitate glucose uptake, whereas later mutations in one replicate causes a glucokinase-enhancing substitution in NanK. Similarly, the appearance of the glucokinase-enhancing NanK substitution in BW21153(Δpts::kanR, Δglk) replicate one correlates with a decrease in organismal fitness during growth on N-acetylneuraminate, while fitness in the same conditions for BW21153(Δpts::kanR, Δglk) replicate two remains primarily unaffected (fig. 4C).
Discussion
Promiscuous enzymes provide the foundations for metabolic innovation. Comparative phylogenetics supports the patchwork model of metabolic evolution, in which promiscuous enzymes are recruited from existing metabolic pathways to enable new metabolism (Rison and Thornton 2002). Despite the recognition that enzyme recruitment is an important adaptive event, the evolution of new metabolic pathways remains unpredictable, partly because recruitment occurs amidst a complex physiological backdrop. Little is known about the specific biological factors that facilitate or constrain enzyme recruitment en route to specific evolutionary endpoints. In particular, the extent to which the level of an enzyme's latent activity impacts its potential to be recruited is largely unknown. Investigating this question is difficult, however, because it requires prior knowledge of all potential recruitment candidates and their relative activity levels. Here, we used an experimental system comprised of a glucokinase-deficient bacterium harboring four previously identified candidates with varying degrees of latent glucokinase activity to investigate the connection between promiscuous activity levels and enzyme recruitability. The results demonstrate that the most active enzyme is not preferentially recruited under the experimental conditions employed here. The enzyme that was recruited, NanK, is the only candidate subject to transcriptional repression by a specialized repressor protein, NanR (fig. 1B). The NanK operon structure offers multiple mutational avenues capable of impairing transcriptional repression, thereby elevating enzyme concentration and facilitating NanK recruitment. This possibility is consistent with similar observations in other systems in which derepression of latent activities is often an early step in metabolic evolution (Jensen 1976).
What other factors might contribute to candidate recruitability? Mutations do not arise at uniform frequencies across genomes. Direct measurements of locus-specific mutation rates demonstrate variations of at least 10-fold across the E. coli genome (Jee et al. 2016). This observation suggests promiscuous enzymes in certain mutational hot spots on the chromosome might be preferentially predisposed for recruitment (Kassen 2019). Bacterial genes are more frequently encoded on the leading strand of DNA, where transcription occurs in the same direction as replication fork progression (Srivatsan et al. 2010). This preferential orientation is thought to result from selection against head-on replication–transcription conflict, which causes higher mutation rates in the lagging DNA strand compared with the leading strand (Merrikh et al. 2012). Interestingly, two of our recruitment candidates, AlsK and NanK, are encoded on the lagging strand of chromosomal DNA (fig. 1B). The strand orientation of nanK offers a possible explanation for its preferential recruitment in this study, considering that the glucokinase catalytic efficiency of NanK is ∼30-fold higher than AlsK. The extent to which strand orientation and/or operon structure contributes to the preferential recruitment of NanK remains to be investigated, but further studies are clearly warranted.
In this study, we failed to observe recruitment of NagK, the candidate with the highest putative in vivo level of latent glucokinase activity. NagK natively phosphorylates N-acetylglucosamine, a key step in murein cell wall recycling (Uehara and Park 2004). Murein recycling is a constant process with E. coli recycling up to 60% of its cell wall each generation (Park and Uehara 2008). Thus, NagK is expected to encounter a continuous supply of N-acetylglucosamine during cell growth. Kinetic assays show that NagK prefers N-acetylglucosamine over glucose by a factor of ∼50-fold (Larion et al. 2007). These observations are consistent with the possibility that NagK is functionally constrained from performing the glucokinase reaction, because access to the NagK active site is blocked by the ubiquitous presence of the native substrate, a consequence of the important metabolic role of its native function. Interestingly, the other candidate enzymes may not suffer from this same constraint. No known E. coli biosynthetic pathways exist for allose or N-acetylmannosamine, the native substrates of AlsK and NanK, respectively. Thus, neither compound is expected to be present in cells unless the bacterium is grown in the presence of these compounds or their metabolic precursors. Fructose is expected to be readily processed through the remaining glycolytic enzymes, and its accumulation seems unlikely. Moreover, the fructose Km value displayed by Mak is notably higher than the Km values of the other three candidate enzymes for their native substrates, suggesting that Mak might be more immune to substrate competition than the other enzymes. Other experimental evolution studies provide precedence for the involvement of such antagonistic pleiotropy during enzyme recruitment. Copley and coworkers identified compensatory derepression mutations that provide sufficient quantities of both the native and promiscuous activities of glutamyl phosphate reductase to enable survival of an argC-deficient strain on glucose minimal medium (Morgenthaler et al. 2019). Future studies aimed at alleviating antagonistic pleiotropy in NagK, perhaps via gene amplification, could be used to test whether NagK's native activity prevents its recruitment to perform the glucokinase reaction.
Our failure to uncover evidence of Mak recruitment is interesting considering prior studies in which a mak promoter-up single nucleotide polymorphism restored glucose processing of a genetically similar glucokinase-deficient bacterium (Miller and Raines 2005). The experimental conditions of the prior study differ from our present study in several important ways. First, the mak + revertant colonies were isolated on solid medium containing ampicillin, kanamycin, and chloramphenicol. Second, the glucokinase-deficient strain used in the earlier study harbored an extrachromosomal plasmid encoding the glucose facilitator protein (Glf) from Zymomonas mobilis, which restores efficient glucose uptake to the glucokinase-deficient strain (Parker et al. 1995). The presence of antibiotics is known to enhance mutation rates and alter the spectra of mutations sampled in a growing bacterial population (Long et al. 2016; Hickman et al. 2017; Blázquez et al. 2018). The presence of the Glf protein, which displays a K0.5 value for glucose of 4.1 mM, could significantly alter the concentration of glucose available within the cell (Weisser et al. 1995). The extent to which these experimental differences contribute to the differences in enzyme recruitment outcomes is unknown. However, this work begins to probe the impact of some of these differences. While the inclusion of the antibiotic kanamycin for the adaptation of the BW25113(Δpts::kanR, Δglk) replicates did not change the overall outcome of nanK recruitment via transcriptional amplification of nanK and accumulation of mutations that alter NanK's activity, our study did identify differences in mutational identity and time of appearance in the adaptation experiments containing kanamycin compared with the adaptation experiments lacking antibiotics. For instance, the average number of mutations detected at greater than 20% frequency throughout the adaptive process in strains adapted in medium lacking kanamycin was almost an order of magnitude smaller than the average number of mutations detected at greater than 20% frequency in strains adapted in medium containing kanamycin (supplementary Dataset 1, Supplementary Material online). Some of the new mutations accumulated in the galR and nanR genes, producing new mechanisms for repressor disruption in these proteins, as well as in the prior unobserved galS gene, indicating an alternative target to galR for recruitment (fig. 2E). Further, we were unable to reliably collect fitness measurements for the replicates from the kanamycin-containing adaptative evolution experiments before generation 25, whereas fitness could be reliably obtained as early as generation 10 from replicate strains in adaptive evolution experiments lacking kanamycin (fig. 2). This observation is consistent with literature observations that epistatic constraints between mutations in adaptive evolution experiments containing antibiotics could limit or delay evolutionary outcomes (Palmer et al. 2015; Lukačišinová et al. 2020). In addition to these differences in experimental conditions, the earlier study also did not perform laboratory conditioning before isolating the revertant colonies, and whole-genome sequencing was not performed before or after adaptation. Only chromosomal sites near the four candidate genes were sequenced after adaptation. Thus, the extent to which additional genetic differences might be responsible for Mak recruitment is unknown. Nevertheless, the differences in recruitment outcomes in the prior study, compared with the present results, further highlight the need to identify determinants influencing the recruitment process. The similarity in mutational trajectories for BW25113(Δpts, Δglk) and BW25113(Δpts, Δglk, Δmak) adaptation suggests that the presence of candidate Mak does not substantially impact the evolutionary outcome of NanK recruitment. This observation is interesting since mak expression is constitutive under the growth conditions used in our adaption experiments, and it could function as a metabolic placeholder until NanK is recruited (Li et al. 2014).
In summary, our results demonstrate that latent activity alone is not the sole determinant for preferential enzyme recruitment to solve a metabolic pathway impairment. Just as metabolic evolution involves the restructuring of multiple levels of biological organization, preferential enzyme recruitment likely depends on numerous characteristics across these levels. One potential example of such restructuring is the observed return of nanK transcription to initial levels at generation 100, which occurs in the absence of any specific repressor mutations (fig. 3). Interestingly these transcriptional alterations appear to coincide with the fixation of mutations that improve nanK's promiscuous activity. These observations suggest that higher order metabolic adjustments, such as posttranscriptional modifications that impact mRNA stability, may occur in response to changing conditions and could impact mutational trajectories (Vargas-Blanco and Shell 2020). Future work is needed to investigate how factors such as genomic context, strand orientation, and pleiotropic constraints influence the enzyme recruitment process, thereby shaping metabolic evolution. Understanding the contribution of these factors is expected to advance our fundamental understanding of the rules that determine enzyme recruitment success and ultimately dictate evolutionary outcomes. This information may also impact synthetic biology efforts to assemble new metabolic pathways from existing promiscuous activities. The present results suggest that leveraging the most active isolated enzymes to generate a new pathway might not produce a pathway capable of operating most efficiently in vivo.
Materials and Methods
A comprehensive list of materials and methods is provided in supplementary materials, Supplementary Material online.
Strain Construction and Laboratory Adaptation
A list of strains, plasmids, and primers used in this study is provided in the supplementary materials, Supplementary Material online (supplementary table S1–S3, Supplementary Material online). Adaptation to laboratory conditions of E. coli BW25113 was performed via growth on 0.2% (w/v) glucose M9 minimal medium following established procedures (Knöppel et al. 2018). P1 transduction was used to transduce targeted portions of genomic DNA from the Keio collection containing the desired glk deletion, and lambda red recombineering was employed for deletion of all other genes (Baba et al. 2006; Thomason et al. 2007; Jensen et al. 2015).
Adaptive Evolution
Adaptive evolution was performed by serial passaging in 500 mL of M9-medium supplemented with 0.2% (w/v) glucose. Replicate lineages were initiated from independent colonies derived from the sequence-confirmed starting strain. Serial passaging was performed by transferring the evolving populations during logarithmic growth after the populations reached OD600 ≈ 1.0 (∼4.4 × 1011 cells). A transmission bottleneck size of ∼1.2 × 1010 cells was utilized per transfer (OD600 ≈ 0.03), resulting in the serial passaging producing approximately five generations per passage. Glucose phosphorylation ability was assayed via growth on MacConkey agar or in liquid neutral red medium, a phenotypic measure of glucose metabolism. Growth rates, a measure of fitness, were extracted from growth-curve analysis performed in triplicate for each biological replicate using AMiGA (Midani et al. 2021).
Phenotypic and Fitness Measurements
Glucose metabolism was assayed at generations 10, 25, 50, 75, and 100 by quantifying the absorbance at 522 nm following overnight growth at 37 °C in neutral red indicator medium (supplementary fig. S2, Supplementary Material online). Growth curves were generated following growth of each strain for 6–8 h at 37 °C until OD600 ≈ 1.0, at which point 1 mL of culture was removed from each culture and washed three times by centrifugation at 4,000 × g and 4 °C for five minutes and resuspension in 1 mL of M9 with either glucose or N-acetylneuraminate. The cultures were used to inoculate M9 with either glucose or N-acetylneuraminate in a sterile 96-well clear microplate (Corning) to OD600 ≈ 0.1. Sample placement in the 96-well plate was randomized to avoid potential growth bias due to variations in temperature and evaporation in the plate reader. Each generation population was assayed in triplicate. The plate was sealed with a breathable membrane (Diversified Biotech). Plates were incubated at 37°C inside a SpectraMax iD5 Multi-Mode microplate reader (MolecularDevices), and OD600 measurements were recorded every 15 min, shaking between readings until all samples reached saturation. Growth rates were extracted from OD600 measurements using the AMiGA program (Midani et al. 2021) to measure fitness.
Whole-Genome Sequencing and RT-qPCR Analysis
Genomic DNA was extracted using the EZNA Bacterial DNA Kit (Omega Bio-Tek), and sample libraries were constructed using the NEBnext Ultra II DNA Kit (New England Biolabs [NEB]). Libraries were pooled and multiplexed for sequencing on a NOVASEQ 6000 sequencing system (Illumina). A comprehensive list of mutations observed in each lineage is provided in supplementary Dataset 1, Supplementary Material online. For RT-qPCR, RNA was extracted from frozen populations using the EZNA Bacterial RNA Kit (Omega Bio-Tek) and treated twice with DNase I RNAse free (NEB). cDNAs at a concentration of 100 ng/μL were used as a template for qPCR on a 7500 fast real-time PCR system (Life Technologies) using PerfeCTa SYBR Green SuperMix, Low Rox (Quanta Bioscience). Fold expression was determined by normalization to the endogenous housekeeping gene cysG (Livak and Schmittgen 2001; Zhou et al. 2011).
Enzyme Expression and Characterization
The nanK gene was amplified from BW25113 genomic DNA and inserted into pET-22b(+) for protein production and purification in E. coli strain BM5340(DE3). QuikChange II site–directed mutagenesis (Agilent) was used to generate the L84R and L84P variants of NanK. Activity with glucose or N-acetyl-D-mannosamine was measured using a coupled pyruvate kinase/lactate dehydrogenase enzyme–linked spectrophotometric assay (Larion et al. 2007).
Supplementary Material
Supplementary data are available at Molecular Biology and Evolution online.
Acknowledgments
This work was supported, in part, by National Institutes of Health grants GM133843 and GM115388.
Author Contributions
R.P.C., A.C.W., D.A.R.Z., and B.G.M. conceived and designed the research. R.P.C. and B.G.M. prepared the original manuscript. All authors edited and revised the manuscript. R.P.C. constructed bacterial strains, generated plasmids, conducted adaptive evolution experiments, performed growth curve analyses, analyzed whole-genome sequencing data, and performed enzyme kinetic assays. R.P.C and D.A.R.Z performed RT-qPCR experiments. A.C.W. and B.G.M. provided supervision and project administration.
Data Availability
All high-throughput sequencing files are archived in the NCBI Sequence Read Archive (SRA) database under accession number PRJNA952611. Full and annotated genome difference files for all evolving populations sequenced in this article are available in supplementary Dataset 1, Supplementary Material online. Materials described herein are available upon request to the corresponding author.
