Horizontal transfer of a pathway for coumarate catabolism unexpectedly inhibits purine nucleotide biosynthesis

Summary:

A microbe’s ecological niche and biotechnological utility are determined by its specific set of co-evolved metabolic pathways. The acquisition of new pathways, through horizontal gene transfer or genetic engineering, can have unpredictable consequences. Here we show that two different pathways for coumarate catabolism failed to function when initially transferred into Escherichia coli. Using laboratory evolution, we elucidated the factors limiting activity of the newly acquired pathways and the modifications required to overcome these limitations. Both pathways required host mutations to enable effective growth with coumarate, but the necessary mutations differed. In one case, a pathway intermediate inhibited purine nucleotide biosynthesis, and this inhibition was relieved by single amino acid replacements in IMP dehydrogenase. A strain that natively contains this coumarate catabolism pathway, Acinetobacter baumannii, is resistant to inhibition by the relevant intermediate, suggesting that natural pathway transfers have faced and overcome similar challenges. Molecular dynamics simulation of the wild type and a representative single-residue mutant provide insight into the structural and dynamic changes that relieve inhibition. These results demonstrate how deleterious interactions can limit pathway transfer, that these interactions can be traced to specific molecular interactions between host and pathway, and how evolution or engineering can alleviate these limitations.

Graphical Abstract:

Abbreviated Summary:

A pathway for catabolism of coumarate, a biotechnologically relevant lignin-derived aromatic compound, was engineered into Escherichia coli. Accumulation of an intermediate, 4-hydroxybenzaldehyde, was shown to inhibit a key enzyme in purine nucleotide biosynthesis, IMP dehydrogenase (IMPDH). Inhibition could be relieved through point mutations to IMPDH that altered enzyme dynamics without disrupting the catalytic center.

Introduction:

Microbes can use a wide variety of compounds as carbon and energy sources. Expanding the breadth of compounds that a strain can catabolize can allow access to new environmental niches or enable engineered microbes to use new feedstocks. Correspondingly, catabolic pathways are frequently transferred between strains, either in nature through horizontal gene transfer (HGT) or in the laboratory through metabolic engineering (Pál et al., 2005; Nielsen and Keasling, 2016). However, newly-acquired pathways often fail to function effectively in their new host (Porse et al., 2018). In these cases, productive use of a new pathway may require post-transfer refinement to optimize expression and minimize deleterious interactions (Michener et al., 2014; Clark et al., 2015). The pathway activity immediately following transfer may be very different from the potential activity after optimization, complicating predictions about engineering or HGT.

We have explored this issue using pathways for catabolism of lignin-derived aromatic compounds, since these pathways are widespread in nature (Bugg et al., 2011), are often transferred by HGT (Chain et al., 2006), have biotechnological applications (Ragauskas et al., 2014), and involve challenging biochemistry (Masai et al., 2007). We previously constructed strains of E. coli that grow with the model lignin-derived compounds protocatechuate (PCA) and 4-hydroxybenzoate (4-HB) as sole sources of carbon and energy using the 3,4-cleavage pathway for protocatechuate catabolism from Pseudomonas putida and a 4-hydroxybenzoate 3-monooxygenase from P. putida or Paenibacillus sp. JJ-1b (Clarkson et al., 2017; Standaert et al., 2018). Introduction of the relevant catabolic pathways was not sufficient to enable rapid growth with either carbon source. We then used experimental evolution to select for strains with improved growth. By resequencing the evolved variants and reconstructing mutations in the parental strains, we identified causal mutations that improved function of the heterologous pathway.

In this work, we extended those pathways to allow growth with a model phenylpropanoid, coumarate. There are two known oxidative routes for coumarate catabolism, differing in their specific reaction chemistry and resulting intermediates (Figure 1). These pathways are exemplified by the hca pathway from Acinetobacter baylyi ADP1 (Parke and Ornston, 2003) and the cou pathway from Rhodococcus jostii (Otani et al., 2014). Both pathways begin by conjugating the phenylpropanoid substrate to coenzyme A. The hca pathway then uses a retro-aldol reaction to produce an intermediate benzaldehyde derivative, while the cou pathway uses a hydrolytic retro-Claisen reaction to produce the benzoate derivative directly. Since these two phenylpropanoid pathways use different biochemistry and intermediates, their interactions with the host may also differ substantially (Kim and Copley, 2012). Identifying the likeliest pairing of host and pathway, either for engineering or HGT, will depend on understanding the specific challenges imposed by each potential pathway and the mechanisms to overcome these challenges available to the host.

Figure 1:

Two routes, exemplified by the hcaABC pathway from A. baylyi ADP1 and the couLMNO pathway from R. jostii, convert the phenylpropanoid coumarate to 4-HB, which is then oxidized to PCA. For simplicity, cofactors and the resulting acetyl-CoA are not shown.

Using a combination of engineering and evolution, we constructed and optimized both representative pathways for phenylpropanoid catabolism in E. coli. We show that pathway activity is initially limited due to pathway-specific molecular interactions that can readily be alleviated through point mutations to the host. Similar compensatory mechanisms are present in a strain that natively contains the appropriate pathway. Molecular dynamics simulations of the wild-type and mutant enzymes demonstrate how subtle modifications to the enzyme distant from the active site can relieve inhibition while preserving catalysis. Identifying and alleviating the specific molecular interactions between an engineered metabolic pathway and its heterologous host will aid our efforts to rapidly engineer metabolic capabilities.

Results

Combining engineering and evolution enabled coumarate catabolism

We designed and synthesized two constructs containing genes for phenylpropanoid import and degradation, each of which converts coumarate into 4-hydroxybenzoate (Figure 1 and Figure S1). Each pathway was introduced into E. coli strains, JME38 and JME50, that had previously been engineered to grow with 4-HB using pobA and praI, respectively (Standaert et al., 2018). None of the engineered strains acquired the immediate ability to grow with coumarate as the sole source of carbon and energy (Figure S2).

To understand the factors preventing pathway function, we used experimental evolution to select for strains with the ability to catabolize coumarate. Three replicate cultures of each engineered strain were propagated in minimal medium containing 1 g/L coumarate (~6.1 mM). After 300 generations, individual mutants were isolated from each population and characterized for growth with protocatechuate (PCA), 4-HB, coumarate, and caffeate. Representative isolates were chosen for each replicate population for further characterization. All isolates could grow with PCA and coumarate, though growth with caffeate and 4-HB varied between replicates (Figure S2).

Genome resequencing and reconstruction identified causal mutations

The genomes of the selected isolates were resequenced to identify new mutations (Table S1 and Table S4). Several of the mutations have previously been described for their effects on catabolism of 4-HB, such as synonymous mutations to the gene encoding the 4-hydroxybenzoate monooxygenase pobA (Standaert et al., 2018). Among the strains with the hca pathway, five of the six isolates had additional mutations to the native gene guaB, encoding inosine monophosphate (IMP) dehydrogenase (IMPDH), and to the intergenic region between hcaB and hcaC in the engineered pathway. The exception was JME96, which had a mutation to rpoS, encoding the RNA polymerase sigma factor σ³⁸, that is expected to be highly pleiotropic (Saxer et al., 2014).

In the strains with the cou pathway, the acquired mutations were less consistent across replicates, with several mutations to genes that are expected to be pleiotropic. However, parallel mutations were observed in JME106 and JME109, with mutations to both couL and nadR. The mutations to couL, which encodes the CoA ligase, were coding mutations, L192R and S134Y. NadR is involved in both regulation and catalysis for NAD salvage (Kurnasov et al., 2002). One of the mutations to nadR led to a frameshift that precisely removed the C-terminal ribosylnicotinamide kinase (RNK) domain, which converts N-ribosylnicotinamide into β-nicotinamide mononucleotide during NAD salvage (Kurnasov et al., 2002). Similarly, the second nadR mutation also occurred in the RNK domain. The physiological consequences of these mutations are unclear.

To test the causality of the identified mutations, we reconstructed representative mutations in the engineered parental strains. We assumed that parallel mutations to a given gene produced similar effects, and therefore only tested one representative mutation (e.g. D243G in guaB). Two mutations, to pobA and hcaABCK, were necessary for growth with coumarate in JME64, while a third mutation to guaB significantly increased growth (Figure 2A). Similarly, mutations to pobA, couLHTMNO, and nadR were all required for growth with coumarate using the cou pathway in JME65 (Figure 2B).

Figure 2:

Reconstruction identifies causal mutations. (A) For the hca pathway, a triple mutant, containing mutations to pobA (synonymous, T3T), guaB (D243G), and hcaC (intergenic mutation, G→A 47 bp upstream of start codon), and the three double mutants were grown in minimal medium with 1 g/L of the indicated substrate as the sole source of carbon and energy. (B) As in 3A, except using the cou pathway and mutations to pobA (T3T), nadR (Δ326–410), and couL (S134Y). Error bars show one standard deviation, calculated from three biological replicates. Specific mutations are described in Tables S1 and S4.

Synonymous mutations to pobA mutation have previously been shown to increase expression of PobA by destabilizing secondary structures in the mRNA (Standaert et al., 2018). To understand the effect of the intergenic mutations upstream of hcaC, we measured protein expression levels in the engineered strains. As expected, the mutation to pobA increased expression of PobA by approximately 9-fold, while the hcaC mutation increased expression of both HcaB and HcaC by roughly 2-fold (Figure S3). The intergenic mutation before hcaC is predicted to increase the translation rates by approximately 10-fold through modulation of the strength of the ribosome binding site (Espah Borujeni et al., 2014).

Parallelism of mutations within replicates of a pathway, but divergence between pathways, strongly suggests that the mutations are specific to a particular pathway. To test this hypothesis, we replaced the hca pathway in JME131 with either the wild-type or evolved cou pathways. Neither strain was able to grow with coumarate as the sole source of carbon and energy.

Inhibitory cross-talk between engineered and native pathways limits function

A mutation to guaB was necessary for growth with coumarate using the hca pathway. IMPDH, encoded by guaB, converts inosine monophosphate (IMP) to xanthosine monophosphate (XMP) with the reduction of NAD⁺ during guanine nucleotide biosynthesis (Hedstrom, 2009). Five independent amino acid replacements in IMPDH were identified: A48V, D243G, G330D, L364Q, and P482L. IMPDH uses different conformations to catalyze each step of the catalytic cycle: an open conformation for hydride transfer that produces a covalent intermediate with the catalytic C305 (E-XMP*) and a closed conformation for hydrolysis of the E-XMP* (Figure S4). We generated homology models of wild-type E. coli IMPDH in both the closed and open conformations. The mutations are distant from each other and from the active site, with no obvious effect on catalysis (Figure 3).

Figure 3:

Beneficial mutations to IMPDH are distant from the active site. (A) Structural model of E. coli IMPDH colored by chain. (B) Chain A of IMPDH in the closed conformation, highlighting the loop containing C305-XMP* and the active site flap containing the R401-Y402 catalytic dyad.

To understand the consequences of these mutations, we measured metabolite levels in the parent and engineered strains during growth with coumarate. Consistent with our genetic analysis above, we chose to focus on the D243G mutation. Compared to the D243G guaB mutant (JME131), the strain with wild-type guaB (JME129) showed higher levels of inosine nucleotides (Figure 4A). We hypothesized that growth with coumarate led to inhibition of IMPDH and accumulation of IMP, and that this inhibition was relieved in the guaB mutants. To determine whether inhibition of nucleotide biosynthesis limited growth with coumarate, we supplemented the growth medium with guanosine. Addition of guanosine increased growth with coumarate in a strain with the wild-type IMPDH, but not the mutant (Figure 4B).

Figure 4:

Accumulation of 4-hydroxybenzaldehyde inhibits IMPDH. (A) Metabolomics of purine nucleobases and derivatives. Different phosphorylation states could not be distinguished and are reported as the total nucleotide pool. Each of the four biological replicates is shown as a separate measurement, normalized to the mean for JME129. The solid bar indicates the median, as a guide for the eye. N.S.: Not significant; **: p < 0.01. (B) Strains with wild-type and mutant versions of IMPDH were grown in medium containing 1 g/L coumarate with and without the addition of 5 mg/L guanosine (~18 μM). No growth was observed with guanosine alone. Error bars show the standard deviation, calculated from three biological replicates. (C) Enzyme variants were purified and assayed in vitro for inhibition by 4-hydroxybenzaldehyde. Curves show a model fit, using the calculated inhibition constants. Error bars show the standard deviation, calculated from three biological replicates.

Mutations to guaB improved growth with the hca pathway but not with the cou pathway. The hca pathway produces an intermediate, 4-hydroxybenzaldehyde, that is not present in the cou pathway (Figure 1). To test whether this intermediate was responsible for the inhibition of IMPDH, we grew strains containing WT and mutant IMPDH in varying concentra tions of 4-hydroxybenzaldehyde (Figure S5). Both strains were inhibited by high concentrations of 4-hydroxybenzaldehyde, but the mutation to guaB decreased inhibition. In strain JME129, addition of guanosine provided no advantage during growth with glucose but relieved inhibition by 4-hydroxybenzaldehyde (Figure S6).

Next, we purified WT and mutant IMPDH and measured inhibition in vitro with 4-hydroxybenzaldehyde. This compound is a weak inhibitor of WT EcIMPDH, with a K_i,app of 320 ± 20 μM (Figure 4C). Introduction of the D243G replacement had little effect on catalytic activity (Table 1) but increased the K_i,app to 1250 ± 50 μM, indicating a substantial reduction of inhibition in the mutant. The hca pathway that we used came from A. baylyi ADP1, and we hypothesized that the native IMPDH of this strain would have faced similar selective pressures to minimize inhibition by 4-hydroxybenzaldehyde. As a surrogate, we tested the IMPDH of A. baumannii, since this strain contains a homologous hca pathway (83–94% amino acid identity) and IMPDH (91% amino acid identity). As predicted, the A. baumannii IMPDH has a K_i,app of 720 ± 30 μM, substantially higher than that of wild-type EcIMPDH. Further kinetic characterization of these IMPDH orthologs is summarized in Table 1.

Table 1:

Kinetic Parameters for EcIMPDH/WT, EcIMPDH/D243G and AbIMPDH. The values are the average and range of two independent experiments. a. UC inhibition versus IMP. b. NC inhibition versus NAD⁺. n.d., not determined.

	EcIMPDH/WT	EcIMPDH/D243G	AbIMPDH
kcat (s−1)	0.97 ± 0.05	0.50 ± 0.04	0.86 ± 0.02
Km(IMP) (μM)	21 ± 3	7.9 ± 1	16 ± 3
Km(NAD) (μM)	210 ± 20	210 ± 40	208 ± 17
Kii(NAD) (μM)	4300 ± 530	3500 ± 690	10600 ± 1100
Ki,app(4HB) (μM)	320 ± 30	1250 ± 50	720 ± 30
IMP Kii (μM) a	930 ± 70	570 ± 30	n.d.
NAD+ Kis (μM) b	490 ± 100	450 ± 90	n.d.
Kii (μM) b	1200 ± 100	450 ± 90	n.d.

Subtle changes in enzyme dynamics relieve inhibition while maintaining catalysis

To gain insight into possible mechanisms of inhibition by 4-hydroxybenzaldehyde, we measured enzyme activity at varying inhibitor concentrations. 4-Hydroxybenzaldehyde is an uncompetitive inhibitor with respect to IMP and a noncompetitive (mixed) inhibitor with respect to NAD⁺ for both wild type and D243G EcIMPDH (Figure S7). Similar patterns of inhibition have been observed for compounds that bind in the NAD⁺ site (Makowska-Grzyska et al., 2015). In addition, we computationally docked 4-hydroxybenzaldehyde to a model of wild-type IMPDH in the open conformation of the apoenzyme as well as to IMP-bound and IMP/NAD⁺-bound states. In both the apoenzyme and IMP-bound models, the majority of the top poses of 4-hydroxybenzaldehyde were found to occupy the NAD⁺ binding site, approximately 20 Å from D243 (Figure S8 and Table S6). Therefore, to relieve inhibition by 4-hydroxybenzaldehyde, the D243G substitution would need to perturb the structure or dynamics of the distant active site.

Understanding the mechanism of this perturbation required additional analysis. The hydrolysis of E-XMP* is the slow step in the IMPDH reaction, so E-XMP* is the predominant enzyme complex (Hedstrom, 2009). Therefore, a decrease in the affinity of 4-hydroxybenzaldehyde for E-XMP* can account for resistance to inhibition. Hydrolysis of E-XMP* requires a conformational change wherein a mobile protein flap folds into the cofactor binding site. We assessed the effect of the D243G mutation on the active site by performing molecular dynamics (MD) simulations of wild-type and D243G mutant IMPDH in the covalently bound E-XMP* state using the closed conformation model, since the flap is disordered in crystal structures of the open conformation. In the simulations of the wild-type E. coli IMPDH, D243 forms stable hydrogen bonds with the side chains of K87 and R219 and also with the backbone of V220 (Figure 5A). In the absence of this hydrogen bonding network in the mutant (Figure 5B), G243 adopts two different conformations, one that resembles the wild type in which G243 is close to but not interacting with K87, R219, and V220, and another in which G243 is positioned farther away from these residues when a new hydrogen bond is formed with Q272 (Figure 5C and Figure S9). However, it was not obvious how these local changes around the mutation site propagate to the active site, which is located on the opposite side of the β-barrel.

Figure 5:

A beneficial mutation to IMPDH affects enzyme structural dynamics. Hydrogen bond network around the D243G mutation site for (A) wild-type and (B) mutant IMPDH from MD simulations. (C) Heavy atom distance distributions from five independent simulations of each closed conformation model are shown.

To identify changes in protein dynamics resulting from the D243G mutation, we calculated root-mean-squared fluctuations (RMSFs) for chain A in both the wild-type and mutant. In both systems, high RMSFs were observed over the entire flap region (Figure S10). Upon inspection of specific interactions of flap residues, we found that the mutation leads to changes in hydrogen-bonding interactions with other residues on chain A or the adjacent chain D (Figure S11 and Figure S12). These interactions resulted in reorientation of a loop on the flap that could alter inhibitor binding. Details of specific hydrogen bonding interaction changes are described in the Supporting Information. Despite these changes in flap conformations and dynamics, the catalytic dyad remains in close proximity to the covalent intermediate, poised for catalysis (Figure S13).

RMSF analysis also revealed that helix α2 (residues 76–89) and helix α8 (residues 230–241) fluctuate more in the mutant than in the wild type (Figure S10). Helix α8 is downstream of the mutation site (Figure 6A). Therefore, the higher fluctuations of this helix and the adjacent helix α2 in the mutant are likely due to the loss of the hydrogen bonding network formed by D243 with K87, R219, and V220 as well as the formation of a new hydrogen bond between G243 and Q272 in the mutant. The N-terminal end of helix α2 and the C-terminal end of helix α8 are located near the NAD⁺ binding site, which is also the predicted binding site for 4-hydroxybenzaldehyde based on docking to the open conformation model (Figure S8). Therefore, these perturbations to helices α2 and α8 represent another mechanism for the D243G mutation to affect inhibitor binding.

Figure 6:

(A) Selected snapshots of the flap from MD simulations of wild-type and mutant IMPDH in the closed conformation. Cα atoms of key residues whose interactions differ between wild type and mutant simulations (D50, R386, S399, S405, D410, E418) are shown as spheres and labeled. A selected docking pose is shown for 4-hydroxybenzadehyde in the IMP-bound open conformation after superposition with the closed conformation snapshots. (B) Selected docking pose showing 4-hydroxybenzaldehyde interactions with residues D248 and S250. Flap residues were omitted as they were not resolved in the templates used to model the open conformation.

To understand the mechanism by which helix dynamics could affect inhibitor binding, we combined these MD results with our previous docking studies. In six of the top 10 docking poses, the phenolic hydrogen of the inhibitor forms hydrogen-bonding interactions with either the side chain or backbone of D248. The carbonyl oxygen of the inhibitor also interacts with the side chain of S250 (Figure 6B). D248 and S250 are located on the β-sheet (β11) downstream of the mutation site and are in close proximity to helices α2 and α8 as well as the loop on the flap that showed different interactions in the wild-type and mutant simulations. Thus, changes in the structure and dynamics of these regions around the NAD⁺ binding site likely disrupt inhibitor access and binding.

Discussion

In this work, we have recapitulated the process of HGT and demonstrated the necessity for host adaptations to accommodate the hca pathway in both E. coli and A. baumannii. We identified a novel interaction between the newly introduced pathway and the endogenous metabolism, as well as the physiological and biochemical consequences of this interaction. Finally, we demonstrated how single point mutations to an essential host protein alter its conformational dynamics to prevent binding of the novel inhibitor while still preserving catalysis.

Highly similar hca pathways are present in various beta- and gamma-proteobacteria. Further HGT of this pathway would require either a host with an IMPDH homolog that is resistant to inhibition by 4-hydroxybenzaldehyde, or post-transfer selection for mutations that relieve inhibition. Understanding these types of limitations on HGT, and the mechanisms by which organisms evolve to avoid them, will aid in our ability to predict and manipulate horizontal gene transfer (Michener et al., 2014; Clark et al., 2015).

In combination, our results suggest that introduction of the hca pathway allowed only limited growth with coumarate because accumulation of 4-hydroxybenzaldehyde inhibited the native E. coli IMPDH. This inhibitory cross-talk results in nucleotide starvation and impairs growth and phenylpropanoid catabolism. Mutations to guaB prevent inhibition by 4-hydroxybenzaldehyde and allow growth with coumarate. There is no a priori reason to expect that a pathway for degradation of an aromatic compound would interact with a native pathway for nucleotide biosynthesis. Phenolic amides such as feruloyl amide have been shown to inhibit a different step in nucleotide biosynthesis (Pisithkul et al., 2015), but neither the substrate nor products of coumarate degradation are toxic at the relevant concentrations (Figure S5, and Clarkson et al., 2017; Standaert et al., 2018). These types of inhibitory cross-talk are likely to be common with heterologous engineered metabolic pathways, though they are rarely identified and alleviated (Kizer et al., 2008; Kim and Copley, 2012; Michener et al., 2012).

In particular, inhibition of microbial growth by aldehydes is commonly observed, though the mechanisms of toxicity can rarely be traced to a specific interaction (Mills et al., 2009; Clarkson et al., 2014; Yi et al., 2015). Enzymatic pathways have frequently evolved to limit the release of free aldehydes, for example through enzymatic channeling (Huang et al., 2001). It is unclear whether channeling between HcaA and HcaB limits the release of free aldehydes in either the native or heterologous hosts. Mutations that increase tolerance to free aldehydes generally do so either by increasing export of the toxic compound or by performing redox chemistry to remove the aldehyde functionality (Mukhopadhyay, 2015). In this work, we have shown an example of aldehyde toxicity that acts through a single protein and can be relieved by point mutations to the associated gene. For the D243G mutant, biochemical assays revealed mixed inhibition that was relieved through mutation. Other examples of nonspecific toxicity may prove to be similarly specific when characterized fully.

MD simulations provided insight into how a single amino acid substitution distant from the active site could relieve inhibition while maintaining catalysis. The mutation is located at the N-terminal end of β11, which is near the NAD⁺ binding site where the inhibitor was predicted to bind based on docking calculations. The simulations showed local changes in the hydrogen bonding networks at the mutation site, which led to changes in the dynamics of the catalytic flap and helices α2 and α8 near the inhibitor binding site. In addition, the catalytic dyad showed only minor perturbations and remained poised for catalysis.

Across the replicate populations, many mutations were highly pleiotropic, including large insertions and deletions flanked by insertion sequences as well as mutations to core transcriptional machinery such as rho and rpoB. Duplications frequently spanned the insertion sites for engineered operons, suggesting that expression of the heterologous genes was limiting. By comparing across replicates, we were able to identify a set of point mutations that allowed growth with coumarate as the sole source of carbon and energy. However, a reconstructed strain containing these mutations does not grow as quickly with coumarate as the evolved isolates, suggesting that some of the remaining mutations provided additional fitness benefits (Figure 2 and Figure S2).

The two 4-HB monooxygenases, praI and pobA, are 60% identical at the nucleotide level, and the associated enzymes have 54% amino acid identity. We previously demonstrated that these enzymes required different optimization solutions to enable growth with 4-HB (Standaert et al., 2018). In contrast, in this experiment, the evolutionary solutions were very similar. Even in the optimized strain, JME131, the growth rate with coumarate was lower than the growth rate with PCA. We hypothesize that the conversion of coumarate into 4-HB is the rate-limiting step, and that the conversion of 4-HB into PCA by either enzyme was sufficient under these circumstances.

Multiple mutations were identified in the heterologous cou and hca pathways. In the hca pathway, these mutations served to increase expression of the pathway, either through pathway duplication or by intergenic mutations that affected translation, specifically increasing expression of the CoA ligase HcaC. The cou pathway mutations were coding mutations to a single gene, the couL that encodes a CoA ligase, and decrease expression of that enzyme (SI Appendix, Figure S3). These differential evolutionary responses could arise from different initial expression levels of the two CoA ligases, for example due to the placement of couL at the beginning of an operon and hcaC at the end. Further biochemical analysis will be required to precisely identify the consequences of these mutations.

We have described the use of experimental evolution to identify and alleviate deleterious interactions between engineered metabolic pathways for coumarate catabolism and native pathways for nucleotide biosynthesis and cofactor salvage. Many engineered pathways place a substantial burden on the production host, yet understanding and accommodating these interactions remains challenging. Evolution can simplify this optimization process by directly selecting for mutations that eliminate the inhibition. As we did with guaB, researchers can then work backwards from the evolutionary solutions to understand the factors that were initially limiting productivity and the biochemical solutions to overcome those problems. By solving more problems of this sort, we will develop design rules for future forward engineering of metabolic pathways and better predictions of the likelihood of pathway transfer by HGT.

Experimental Procedures

Strains and chemicals

Unless otherwise noted, all chemicals were purchased from Sigma-Aldrich (St. Louis, MO) or Fisher Scientific (Fairlawn, NJ) and were molecular grade. All oligonucleotides were ordered from IDT (Coralville, IA). E. coli strains were routinely cultivated at 37 °C in LB containing the necessary antibiotics (50 mg/L kanamycin or 50 mg/L spectinomycin). Growth assays with aromatic substrates were performed in M9 salts medium containing 300 mg/L thiamine and 1 mM isopropyl β-D-1-thiogalactopyranoside (IPTG). PCA and 4-HB were dissolved in water at 5 g/L, filter sterilized, and added at a final concentration of 1 g/L. Coumarate and caffeate were dissolved in DMSO at 100 g/L and added at a final concentration of 1 g/L. The addition of 1% DMSO did not affect growth. The pH of the substrates was not controlled, as PCA oxidation occurred more rapidly at neutral pH.

Plasmid construction

Plasmids pJM219 and pJM223, containing the cou and hca expression constructs, were synthesized by the Joint Genome Institute. As described previously, the pathway design used synthetic promoters, terminators, and custom ribosome binding sites (Salis et al., 2009; Chen et al., 2013; Kosuri et al., 2013; Espah Borujeni et al., 2014). Plasmids expressing sgRNA for chromosomal modifications were constructed as described previously, using an inverse PCR to linearize the expression vector followed by assembly with synthesized oligonucleotides (Clarkson et al., 2017). Plasmid pJM303, expressing the D243G mutant of the E. coli IMPDH, was constructed by amplifying the mutant guaB allele from JME89 and cloning it into pMCSG7 under the control of a T7 promoter. Plasmids and primers used in this work are described in Tables S2 and S3.

Strain construction

Genome modifications were performed as described previously, using the lambda-red recombineering system in combination with Cas9-mediated selection (Jiang et al., 2015; Clarkson et al., 2017). Integration cassettes were amplified from synthesized plasmids or the chromosomal DNA of mutant strains, as needed.

Experimental evolution

Parental strains were streaked to single colonies. Three colonies from each strain were grown to saturation in LB + 1 mM IPTG, then diluted 128-fold into M9 + 1 mM IPTG + 1 g/L coumarate + 50 mg/L PCA and grown at 37 °C. When the cultures reached saturation, typically after two days during the initial stages, they were diluted 128-fold into fresh medium. As the growth became more robust, the PCA concentration was decreased. Cultures derived from JME65 and JME67 required the addition of PCA for 150 generations, and in some cases took two days to reach saturation even after 300 generations. One replicate culture, JME65-C, became contaminated with a different coumarate-degrading strain. This contamination was not discovered until resequencing, and consequently the culture was not restarted.

After 300 generations, the evolved cultures were streaked to single colonies. Six isolates from each replicate culture were tested for growth with PCA, 4-HB, coumarate, and caffeate. Representative isolates were selected and validated, followed by genome resequencing.

Genome resequencing

Genomic DNA was isolated using a Blood and Tissue kit (Qiagen, Valencia, CA), according to the manufacturer’s directions. The DNA was then sequenced by the Joint Genome Institute on a MiSeq (Illumina, San Diego, CA) to approximately 75x coverage.

Growth rate measurements

Growth rates were measured as described previously (Clarkson et al., 2017). Briefly, cultures were grown overnight to saturation in M9 + 1 mM IPTG + 2 g/L glucose. They were then diluted 100-fold into fresh M9 + IPTG containing the appropriate carbon source and grown as triplicate 100 μL cultures in a Bioscreen C plate reader (Oy Growth Curves Ab Ltd, Helsinki, Finland). Growth rates were calculated using CurveFitter software based on readings of optical density at 600 nm (Delaney et al., 2013).

Proteomic measurements

Engineered E. coli strains were grown to saturation in 5 mL cultures of M9 + 2 g/L glucose + 1 mM IPTG. They were then diluted 100-fold into triplicate 5 mL of the same medium and grown to mid-log phase. The cells were separated by centrifugation, washed twice with water, and frozen in LN₂ for later analysis.

Processing for LC-MS/MS analysis was performed as previously described (Clarkson et al., 2017). Briefly, crude protein lysates were obtained by bead beating cells in sodium deoxycholate (SDC) lysis buffer (4% SDC, 100 mM ammonium bicarbonate, pH 8.0). Cleared protein lysates were then adjusted to 10 mM dithiothreitol and incubated at 95 °C for 10 min to denature and reduce proteins. Cysteines were alkylated/blocked with 30 mM iodoacetamide and 250 μg transferred to a 10-kDa MWCO spin filter (Vivaspin 500, Sartorius) for in situ clean-up and digestion with sequencing-grade trypsin (G-Biosciences). The tryptic peptide solution was then spin-filtered through the MWCO membrane, adjusted to 1% formic acid to precipitate residual SDC, and SDC precipitate removed from the peptide solution with water-saturated ethyl acetate extraction. Peptide samples were then concentrated via SpeedVac (Thermo Fisher) and quantified by BCA assay (Pierce) prior to LC-MS/MS analysis.

Peptide samples were analyzed by automated 2D LC-MS/MS analysis using a Vanquish UHPLC plumbed directly in-line with a Q Exactive Plus mass spectrometer (Thermo Scientific) outfitted with a triphasic MudPIT back column (RP-SCX-RP) coupled to an in-house pulled nanospray emitter packed with 30 cm of 5 μm Kinetex C18 RP resin (Phenomenex). For each sample, 5 μg of peptides were loaded, desalted, separated and analyzed across two successive salt cuts of ammonium acetate (50 mM and 500 mM), each followed by 105 min organic gradient, as previously detailed (Clarkson et al., 2017). Eluting peptides were measured and sequenced by data-dependent acquisition on the Q Exactive MS.

MS/MS spectra were searched against the E. coli K-12 proteome concatenated with exogenous Pca, Hca, and Cou pathway proteins, common protein contaminants, and decoy sequences using MyriMatch v.2.2 (Tabb et al., 2007). Peptide spectrum matches (PSM) were required to be fully tryptic with any number of missed cleavages; a static modification of 57.0214 Da on cysteine (carbamidomethylated) and a dynamic modification of 15.9949 Da on methionine (oxidized) residues. PSMs were filtered using IDPicker v.3.0 (Ma et al., 2009) with an experiment-wide false-discovery rate controlled at < 1% at the peptide-level. Peptide intensities were assessed by chromatographic area-under-the-curve and unique peptide intensities summed to estimate protein-level abundance. Protein abundance distributions were then normalized across samples and missing values imputed to simulate the MS instrument’s limit of detection. Significant differences in protein abundance were assessed by pairwise T-test.

Metabolite measurements

Strains JME129 and JME131, picked from single colonies on freshly-streaked plates, were grown in M9 + 2 g/L glucose + 1 mM IPTG overnight at 37 °C. Cultures were diluted 100-fold into fresh M9 + 1 g/L coumarate + 1 mM IPTG and regrown to mid-log phase, then quickly concentrated to 100 μL by centrifugation and plated on sterile 50 mm mixed cellulose ester 0.2 μm Whatman filters placed on agar plates containing M9 + 1 g/L coumarate + 1 mM IPTG (Rabinowitz and Kimball, 2007). The plates were incubated for a further 1 h at 37 °C before analysis.

Metabolites were quickly extracted by scraping cells off the permeable membrane and immediately placing them in 100 μL of pre-chilled extraction solvent (40% acetonitrile, 40% methanol, 20% H₂O, 0.1 M formic acid; stored at −20°C) (Rabinowitz and Kimball, 2007). The extraction was then placed at −20°C for 15 min, cells/debris pelleted at 21,000 x g for 5 min at 4°C, and supernatants transferred to cold autosampler vials. Five microliters of each extract was then analyzed by high-resolution LC-MS/MS using a Vanquish UHPLC plumbed directly in-line with a Q Exactive Plus mass spectrometer (Thermo Scientific) outfitted with an in-house pulled nanospray emitter, as previously described (Cecil et al., 2018). Samples (n=4 per strain; biological replicates) were analyzed across 3 separate LC-MS injects, with technical replicates (n=3) and blank runs spread across the entire sampling campaign time of 20 h. Precursor abundances were derived for specific nucleotide phosphates (AxP, GxP, IxP) via Skyline (MacLean et al., 2010) at 5 ppm accuracy and normalized to IPTG, which is taken up by cells but not metabolized. Student’s T test was performed to compare the normalized nucleotide phosphate pools (merged signals from tri-, di-, and monophosphate versions) across samples to identify differences between the two E. coli strains.

IMPDH expression and purification

NAD⁺ was purchased from Roche, IMP and EDTA were purchased from Fisher, MOPS was purchased from Sigma, DTT and IPTG were purchased from GoldBio. EcIMPDH/WT and AbIMPDH were purified as previously described (Makowska-Grzyska et al., 2015). pJM303, expressing EcIMPDH/D243G was transformed into BL21(ΔguaB) cells that lack endogenous EcIMPDH (MacPherson et al., 2010). An overnight culture (5 mL) was diluted into 1 L of fresh LB broth containing 100 μg/mL ampicillin and grown at 37 °C. Once the culture reached an OD600 of 0.6–0.8, IPTG was added to a final concentration of 0.25 mM to induce expression of IMPDH. After 13 h at 30 °C, the cells were collected by centrifugation. All the operations below were performed at 4 °C. The pellet was resuspended in 50 mL phosphate buffer (pH = 8.0) containing 1 mM dithiothreitol (DTT) and sonicated. The debris was removed by centrifugation at 10,000 g at 4 °C for 1 h.

The enzyme in the supernatant was purified by nickel affinity chromatography. The Ni-NTA resin equilibrated with water and phosphate buffer (pH = 8.0). Lysate was loaded onto the 10 mL column with 5 mL resin and washed with 50 mL phosphate buffer (pH = 8.0) then 50 mL phosphate buffer containing 25 mM imidazole. Enzyme was eluted in 25 ml phosphate buffer with 250 mM imidazole. The fractions with IMPDH activity were identified by enzyme activity assays, combined and dialyzed in 25 mM HEPES (pH = 8.0), 1mM DTT and 1 mM EDTA. Protein concentration was determined by Bio-Rad Bradford assay using IgG as a standard. The assay over-estimates the concentration of IMPDH by a factor of 2.6, and protein concentration was adjusted accordingly (Wang et al., 1996).

IMPDH enzyme assays

The IMPDH reaction was monitored by measuring the rate of NADH production on a Shimadzu UV-1800 Spectrometer at λ = 340 nm. MOPS buffer (pH = 7.0) was used to reduce the background absorbance of 4HB. The assay buffer was composed by 20 mM MOPS (pH =7.0), 100 mM KCl, 1 mM EDTA and 1 mM DTT. The final volume of each cuvette was 1 mL.

Kinetic parameters with respect to NAD⁺ were determined by measuring the initial velocity for varying concentrations of NAD⁺ at a fixed saturating concentration of IMP (1.2 mM) and 50 nM of enzyme. Kinetic parameters with respect to IMP were determined by measuring the initial velocity for varying concentrations of IMP at a fixed saturating concentration of NAD⁺ (2.5 mM) and 50 nM of enzyme. Initial velocities were plotted against substrate concentrations and the data were fit using SigmaPlot. The values of K_i,app were determined by measuring the initial velocities for the reaction of IMPDH (20 nM) in the presence of varied concentrations (0–300 μM) of 4-hydroxybenzaldehyde at 12 μM of IMP and 500 μM of NAD⁺. The inhibition by 4-hydroxybenzaldehyde under physiological concentrations of IMP (270 μM) and NAD⁺ (2500 μM) were determined as well (Bennett et al., 2009; Park et al., 2016).

Inhibitor Characterization

The inhibition mechanism was characterized by varying the concentration of 4HB from 0 to 200 μM at fixed concentration of substrates (see following) and enzyme (50 nM). The values of K_i,app with respect to IMP were determined using a fixed NAD+ concentration (1 mM; K_m(NAD⁺) = 210 μM for EcIMPDH/WT; K_m(NAD⁺) = 210 μM for EcIMPDH/D243G) and the values of K_i,app with respect to NAD⁺ were determined by using a saturating IMP concentration (300 μM; K_m(IMP) = 21 μM for EcIMPDH/WT; K_m(IMP) = 7.9 μM for EcIMPDH/D243G). The inhibition mechanism was determined by fitting to the equations for uncompetitive (Eq.1) and noncompetitive (Eq.2) inhibition using SigmaPlot.

Equation for uncompetitive inhibition:

$$\text{v}={\text{V}}_{\text{max}}\left[\text{S}\right]/\left\{K_m+\left[\text{S}\right]\left(1+\left[\text{I}\right]/K_{ii}\right)\right\}$$ (Eq. 1)

Equation for mixed noncompetitive inhibition:

$$\text{v}={\text{V}}_{\text{max}}\left[\text{S}\right]/\left\{(K_m\left(1+\left[\text{I}\right]K_{\text{is}}\right)+\left[\text{S}\right]\left(1+\left[\text{I}\right]/K_{ii}\right)\right\}$$ (Eq. 2)

Homology modeling

To generate models of IMPDH from E. coli (accession number P0ADG7) with the corresponding cofactors and substrates, HHpred (Zimmermann et al., 2017) was used to search the Protein Data Bank for suitable structural templates. Five templates were chosen on the basis of their similarity to the query sequence, inclusion of cofactors and substrates, or both (Table S5). The sequences were aligned with MAFFT (L-INS-i) (Katoh and Standley, 2013) (Figure S15) and homology models were generated using RosettaCM (Song et al., 2013) with fragment files obtained from the Robetta web server (Kim et al., 2004; Gront et al., 2011). The top-scoring model was used for docking of IMP, NAD⁺, and 4-hydroxybenzaldehyde.

The flap containing the catalytic dyad (R401 and Y402) is not resolved in most X-ray crystal structures of IMPDH but is present in the structure of the phosphate-bound “apoenzyme” from Bacillus anthracis (PDB entry 3TSB) (Makowska-Grzyska et al., 2015). Thus, we used this structure as a template to generate homology models of E. coli IMPDH in the closed conformation and the top scoring model was selected to generate a model of the C305-XMP* covalent intermediate for molecular dynamics simulations. Sequences were aligned with MAFFT (L-INS-i) (Katoh and Standley, 2013) (Figure S16).

Ligand docking

Structure files in mol2 format for IMP (ZINC04228242), NAD⁺ (ZINC08214766), and 4-hydroxybenzaldehyde (ZINC00156709) were obtained from http://zinc.docking.org (Irwin et al., 2012). RosettaLigand (Meiler and Baker, 2006) was used to dock 4-hydroxybenzaldehyde into the active site of the IMPDH model following a previously described protocol (Combs et al., 2013). Top binding poses were ranked on the basis of their ‘interface_delta’ score in Rosetta energy units. Additional details of the homology modeling and ligand docking are provided in the Supporting Information.

Molecular dynamics simulations

Initial coordinates for XMP were extracted from the crystal structure of B. anthracis IMPDH complexed with XMP (PDB ID 3TSD), and the covalently bound C305-XMP* complex was generated for chain A using the Molefacture plugin in VMD (Humphrey et al., 1996). The force field toolkit (ffTK) plugin in VMD was used to generate CHARMM-compatible force field parameters for C305-XMP*. Gaussian09 was used to perform geometry optimizations, compute Hessian matrices, and calculate water interaction energies of the C305-XMP* fragment. NAMD 2.11 was used for charge, bond, angle, and dihedral optimization (Phillips et al., 2005).

The CHARMM36 force field (Best et al., 2012) and TIP3P water model (Jorgensen et al., 1983) were used to describe the protein and solvent, respectively. Each system (wild-type and mutant) was solvated in a periodic box of 168 Å × 168 Å × 104 Å and 0.15 M KCl ions were added using CHARMM-GUI (Jo et al., 2008), resulting in a system of ~274,000 atoms. All-atom molecular dynamics (MD) simulations were performed using the OpenMM 7.0 package (Eastman et al., 2017) with GPU acceleration using CUDA 7.2. Ten thousand steps of energy minimization were performed to eliminate clashes, followed by equilibration in the NPT ensemble at 310 K. Temperature was maintained using the Langevin thermostat with a damping coefficient of 1 ps⁻¹. To enable a 5-fs time step, which was used in all simulations, all bond lengths were constrained to their equilibrium distances and the masses of hydrogens were repartitioned to the parent heavy atoms (Eastman et al., 2017). Five separate runs of 100 ns each were performed for both the wild type and mutant IMPDH and the last 50 ns of each run was used for analysis.