Directed evolution of mammalian paraoxonases PON1 and PON3 for bacterial expression and catalytic specialization

Results and Discussion

Family DNA Shuffling and Screening. The wild-type PON1 genes, human (R-form; Hu), mouse (Mo), rat (Rat), and rabbit (Rab) (79–93% nucleic acid sequence identity), and wild-type PON3 genes Hu, Mo, and Rab (81–87% identity), were cloned into an expression vector encoding the PON genes fused to thioredoxin by a peptide linker. Fusion to thioredoxin slightly increased the solubility of PONs (24) and permitted detection of trace activities in lysates of Mo-, Rat-, and Rab-PON1. To achieve efficient expression in E. coli and increase the specific activity of PON toward different substrates, we used family DNA shuffling followed by screening for different activities. DNA shuffling of families of homologous genes has proven a powerful technique for directed evolution of enzymes with specific functions (21, 25). We created two separate libraries by DNA shuffling, a PON1 library derived from shuffling of four wild-type PON1 genes and a PON3 library derived from three wild-type PON3 genes. The libraries were transformed to E. coli. Sequencing of five randomly chosen clones from each library indicated >10 crossovers per gene with almost even distribution of the parental genes (Fig. 5, which is published as supporting information on the PNAS web site). The libraries were screened on agar plates for esterase activity by using 2NA and an azo dye (Fast Red) that reacts with the released 2-naphthol to generate an insoluble red product (23); 10³ to 10⁴ individual agar colonies were screened in each round. Typically, 200–500 clones exhibiting the highest rate of color formation were picked from a replica plate and grown in liquid medium; the crude cell lysate was assayed spectrophotometrically for the hydrolysis of various substrates, including paraoxon and 2NA. The 20 best clones were subcloned and used as templates for further rounds of DNA shuffling and screening.

Directed Evolution of PON1 for E. coli Expression. After three rounds of shuffling and screening, improved clones of PON1 were isolated. The crude lysates of these clones exhibited ≈20-fold higher rates with both 2-NA and paraoxon. Further analysis indicated increased levels of soluble, active enzyme rather than increased specific activity. The best first-generation variants were expressed fused to thioredoxin tag (trx-rePON1 variants G1A5 and G1C4) and exhibited a 3- to 6-fold increase in the amount of soluble protein relative to trx-RabPON1. The best clones from the second round of evolution (G2D6 and G2E6) exhibited a 14- to 18-fold increase (Fig. 1). We then recloned the second-generation library to an expression vector without the fusion protein (thioredoxin) and linker peptide, retaining the unmodified, 355-aa PON1 proteins (rePON1). Screening of this library generated variants G3C9 and G3H8, which yielded ≈12 mg/liter culture of pure, active, unmodified PON1 (Fig. 6, which is published as supporting information on the PNAS web site). These newly evolved variants stand in clear contrast to the unmodified wild-type PON1s, including RabPON1, that yield no soluble expression without the thioredoxin tag.

The sequences of the selected variants reveal that, as early as the first round of evolution, convergence to the RabPON1 gene has occurred, with relatively small contributions from the other three parental genes (Fig. 7A, which is published as supporting information on the PNAS web site). The convergence toward RabPON1 is not surprising, given that the wild-type RabPON1 was the only wild-type PON1 that, after fusion to thioredoxin, expressed measurable amounts of soluble enzyme in E. coli (Fig. 1). It may be related to the higher stability of serum RabPON1 (certainly relative to HuPON1) and its higher affinity for calcium (26). In all selected variants, position 192 is a lysine as in the rabbit wild-type sequence. This position is either an argininine or asparagine in the human R/Q PON1 isoenzymes, of which the R form exhibits ≈8-fold higher catalytic activity (27). Lysine is probably equivalent to arginine, hence the resemblance of the rate parameters of the new variants to the R form human PON1 (Table 1). Eight conserved mutations were identified in all the selected variants, located in only two regions, residues 126–142 and 301–343 (Fig. 8, which is published as supporting information on the PNAS web site). The mutations in the first region are I126T, M130L, K137S, and L142V, and in the second region the mutations are A301G, A320V, M341L, and V343I. All the mutations are quite subtle, and half are from one hydrophobic amino acid to another (M130L, L142V, M341L, and V343I). The two mutated regions in the newly evolved PON1s may be involved in correct folding of PON1 and in its oligomeric packing. It has been suggested that exposed hydrophobic surfaces of PON1 are involved in its oligomerization and aggregation (17). These putative hydrophobic surfaces may also lead to aggregation of wild-type PON1s when expressed in E. coli. Subtle mutations in these exposed hydrophobic surfaces may have prevented misfolding and formation of high-order aggregates and thereby facilitated soluble expression.

Selected PON1 variants were purified, and their kinetic parameters were determined (Table 1 and Fig. 9, which is published as supporting information on the PNAS web site). Little variation occurs in the catalytic parameters for paraoxon and phenylacetate hydrolysis between variants of the first, second, or third generations. Other substrates (2NA and dihydrocoumarin) show the same trend (Table 5, which is published as supporting information on the PNAS web site). This suggests that the increased catalytic activity in the isolated clones is mainly due to increased protein solubility and that the enzymatic properties of PON1 were not significantly altered. Detailed kinetic parameters for wild-type RabPON1 (to which the newly evolved clones are most homologous) are not available. However, serum Rab-PON1 and serum HuPON1 display very similar rates (26), and the newly evolved rePON1s do exhibit kinetic parameters similar to serum HuPON1 (Table 1). The substrate specificity of serum RabPON1 has also been determined (13), and our rePON1s appear quite similar (Table 6, which is published as supporting information on the PNAS web site). Other key properties of PON1 seem to be retained in the newly evolved variants. Like HuPON1 (17), rePON1s appear to be in monomer–dimer equilibrium in the presence of 0.6 mM C₁₂-maltoside, as analyzed by gel filtration (data not shown).

Homocysteine Thiolactonase Activity of rePON1 and Exclusion of Multiple Sites of Hydrolytic Activity. A recent and intriguing finding regarding PON1 is its putative ability to hydrolyze L-HcyT, which is a major risk factor in atherosclerosis (2, 11). We measured such activity in both trx-reRabPON1 and in the newly evolved variants (e.g., G1A5 and G3H8; Table 5 and Fig. 10, which are published as supporting information on the PNAS web site). The K _M measured for L-HcyT hydrolysis (≈19 mM) is in good agreement with the K _M for serum HuPON1 (23 mM) (2).

The substrate specificity of PON1 is unusually broad; its catalytic efficiencies span five orders of magnitude between the best substrate known (phenylacetate) and L-HcyT (k _cat/K _M of 10⁶ to 10 s^–1 M^–1). To establish that the various catalytic activities all occur at the same active site, we used the competitive inhibitor 2-hydroxyquinoline (11). Despite the vast differences in k _cat and K _M values, the inhibition constants for phenylacetate, paraoxon, and L-HcyT hydrolysis were found to be 3–11 μM (Table 7 and Fig. 11, which are published as supporting information on the PNAS web site). These values agree with the inhibition constant of ≈5 μM measured for hydrolysis of phenylacetate by serum HuPON1 (11). The close similarity of the inhibition constants above is strong evidence that all three substrates are hydrolyzed at the same active site; however, the complete proof can come only from deciphering the enzyme structure and catalytic mechanism.

Newly Evolved PON3 Variants (rePON3). Relative to PON1, little is known about PON3. Because of its low level in the serum (≈50-fold lower than PON1), PON3 was only recently purified and characterized (13, 14), and it was not expressed in heterologous expression systems. The specific kinetic parameters of serum PON3 have not been determined, but relative activities of various substrates of RabPON3 are available (13).

Unlike wild-type PON1, we found that wild-type genes of both rabbit and mouse PON3, when fused to thioredoxin, could be expressed in E. coli at reasonable levels (4–6 mg/liter) and appear to be stable and active. The serum-purified and E. coli-expressed PON3s show the same pattern of specificity with dihydrocoumarin (a lactone substrate) and the ester substrates phenyl and naphthyl acetate (Table 8, which is published as supporting information on the PNAS web site). However, we have observed very weak paraoxonase activity with both mouse and rabbit PON3. Activity was very low (<0.1% of that of PON1) and exhibited a prolonged lag (≈30 min) in onset (hysteresis). These two factors may account for the earlier reports of no paraoxonase activity in the serum-purified RabPON3 (13, 14). Thus, the enzymatic properties of RabPON3 are not significantly altered on fusion to thioredoxin and expression in E. coli, and the detailed kinetic parameters obtained here are, in general, relevant to wild-type PON3s.

Wild-type PON3s exhibit 100- to 1,000-fold lower rates of ester and phosphotriester hydrolysis than PON1, although lactonase activities of the two enzymes are similar (13, 14). Their sequences are ≈65% identical at the amino acid level. We were interested to see how far the two enzymes had diverged and whether, under directed selection pressure, they might reconverge to exhibit similar, if not identical, phenotypes. Indeed, three rounds of shuffling and screening of wild-type PON3 genes yielded clones with greatly increased rates of both ester and phosphotriester hydrolysis. Comparison of selected variants from the three rounds of evolution (Table 2) indicated that the evolutionary process was directed primarily toward an increase in the catalytic efficiency of PON3 (see also Table 9, which is published as supporting information on the PNAS web site). After the first round, a mild improvement in catalytic efficiency is observed (Table 2, variants G1A7 and G1B11). After the second round of evolution, one clone was isolated (G2C2) that exhibits (relative to wild-type RabPON3) 62- and 23-fold higher activity for paraoxon and 2NA hydrolysis, respectively. After the third round of evolution, three clones were isolated (G3A5, G3G3, and G3H9) that exhibit an overall improvement of up to 240-fold in paraoxonase activity. Thus, although wild-type PON3s exhibit almost no paraoxonase activity, shuffled variants of the same genes exhibit paraoxonase activity only 60-fold lower than PON1. However, no further improvement in 2NA activity was observed with these clones (Fig. 2 and Table 2). A 4- to 7-fold increase in the expression level of PON3s was also observed after three rounds of evolution (Table 9). Like PON1, the newly evolved PON3 could be expressed and purified (by adding a C-terminal 6×His tag; rePON3 variant G3H9) without the thioredoxin fusion protein, and its catalytic parameters were similar to those of trx-rePON3.

Kinetic parameters for other ester and lactone substrates (Table 9) revealed that the improvements in the catalytic efficiency can be divided into two groups. For the first group, which includes 2NA, phenylacetate, dihydrocoumarin, and L-HcyT, the increase in k _cat/K _M values for the third-generation variants is 2- to 20-fold (relative to the wild-type trx-reRabPON3). The second group includes paraoxon and p-nitrophenylacetate, for which the third-generation variants exhibit k _cat/K _M values up to 240-fold higher than wild-type PON3 (Fig. 2). Thus, the activity toward p-nitrophenylacetate co-evolved with that of paraoxon, whereas all the other substrates show a much milder improvement. In our view, this does not indicate the existence of two catalytic sites or subsites. Rather, the common feature of p-nitrophenylacetate and paraoxon is the p-nitrophenol-leaving group. The coevolution of the two substrates suggests that hydrolysis of both carboxy- and phosphoesters occurs by means of the same mechanism and at the same site.

Sequence analysis of the selected rePON3 variants indicates that the parental genes are evenly represented in variants of the first round of evolution (Fig. 7B). Alignment of variants from the second and third rounds of evolution reveals that, although the sequence similarity between the selected clones is very low, the RabPON3 gene is mildly overrepresented relative to the HuPON3 and MoPON3 genes. These subtle patterns of convergence in the sequences of the selected PON3 are in contrast to the clear convergence of the PON1 newly evolved variants. They show that many routes exist by which the enzymatic activity of PON3 can be increased.

Newly Evolved OP-Selective rePON1 Variants. Gene libraries of the highly soluble rePON1 variant G3C9 were prepared by random mutagenesis by using the wobble-base PCR method with either dPTP or 8-oxo-dGTP nucleoside analogs (22). The number of mutations per gene was adjusted by varying the concentration of dNTP analogs and the number of PCR cycles. Five libraries were generated: two dPTP libraries with 98% transitions, an average of 5–14 mutations per gene, and 20% to 6% residual activity in the pools of genes, respectively; and three 8-oxo-dGTP libraries with an average of 85% transversions, 3–10 mutations per gene, and 13% to 7% residual activity. In addition to the plate screen for 2NA hydrolysis described above, the libraries were directly screened on agar plates (with no need for replication) for the hydrolysis of a newly synthesized fluorogenic OP substrate, DEPCyC (Fig. 4). This substrate, which is a close homologue of the insecticide coumaphos previously used for the directed evolution of a bacterial phosphotriesterase (28, 29), exhibits higher fluorescence at a more convenient wavelength. Screening of ≈10³ to 10⁴ colonies from each library with 2NA yielded variants with mildly improved activity against OP substrates. The best variants were reshuffled and screened directly with DEPCyC, yielding a highly improved variant, reG3C9.49, that carried four mutations: G19R, S193P, N287D, and V346A. The V346A mutation appeared in other improved variants, suggesting its primary role in the enhancement of DEPCyC-hydrolyzing activity. The first-generation 8-oxo-dGTP libraries were also screened directly with DEPCyC to yield another highly improved variant, reG3C9.10, which contained two mutations, L69V and E218D. The two clones were purified and analyzed in detail (Fig. 3 and Table 3). The mutations affected the specific activity toward DEPCyC and phenylacetate, whereas the expression level of the two selected variants was not altered. Both variants exhibited a ≈40-fold increase in catalytic proficiency (k _cat/K _M) toward DEPCyC compared with wild-type PON1 (and its recombinant variant G3C9), and a simultaneous 50-fold reduction in catalytic proficiency toward phenylacetate. The change in the catalytic efficiencies resulted primarily from changes in k _cat values for both substrates. Both variants showed a similarly reduced activity with other esters (e.g., 2NA) and mildly improved activity with another OP substrate (paraoxon; data not shown).

Conclusions and Implications

The outcome of the in vitro evolutionary process described above is a range of recombinant PON variants with a broad range of enzymatic properties. Expression of these variants in E. coli clearly demonstrates that, in contrast to previous findings (16), glycosylation is not an absolute requirement for the stability or enzymatic functions of PON1 and PON3. Expression in E. coli provides an ample source of PON1 and PON3 free from other serum proteins and will facilitate mechanistic and structural characterization of these important enzymes. We were already able to establish unequivocally that homocysteine lactonase activity, which is probably the weakest activity ascribed to the serum-purified enzyme (2, 11) (≈10⁵-fold weaker than its activity toward phenylacetate), takes place at the same catalytic site at which paraoxon and phenylacetate hydrolysis occur.

PON1 and PON3 (as well as PON2) are members of the same gene family, but what caused an ancestral PON gene to duplicate and diverge in the course of natural evolution is not known (30). It is also unclear whether PON3 specializes in an activity that PON1 lacks or vice versa. Here we show that their responses to in vitro selection pressure toward higher paraoxonase and esterase activity are quite different. PON1 evolved to higher solubility, whereas its enzymatic properties seemed to have remained the same. In contrast, PON3 evolved primarily to higher catalytic efficiency; wild-type PON3s have almost no paraoxonase activity, but shuffled variants of the same genes exhibit paraoxonase activity only 60-fold lower than that of PON1. The sequence variations observed for PON1 and PON3 during the directed evolutionary process are also markedly different: whereas the newly evolved PON3 genes show considerable divergence, PON1 rapidly converged toward the rabbit sequence that seems to encode the most soluble and stable PON1. This example shows how closely related genes can behave very differently under the same selection pressure. The outcome of this directed selection pressure also indicates that, despite their divergence in natural evolution, PON1 and PON3 remain very close, not only in sequence but also in active-site structure and catalytic mechanism. This closeness is best exemplified by the rapid regain of paraoxonase activity in PON3 after the selection pressure.

Random mutagenesis and direct selection for OP hydrolysis led to >40-fold improvement in the catalytic proficiency of PON1 toward the phosphotriester DEPCyC (Table 3). This increase was accompanied by a >2,000-fold shift in specificity that was not selected for. Wild-type PON1 is essentially an esterase with very weak phosphotriesterase activity (k _cat/K _M for phenylacetate is 220- to 480-fold higher than for paraoxon or DEPCyC). The directly evolved variants were converted into a phosphotriesterase with weak esterase activity (k _cat/K _M for DEPCyC is 10-fold higher than for phenylacetate). Also notable is the exceptionally subtle nature of mutations that lead to the shift in specificity and the fact that mutations at two different locations (V346A in reG3C9.49 vs. L69V and E218D in reG3C9.10) produce the same shift. Finally, the demonstration that PONs are amenable to the powerful tool of directed evolution opens new prospects for improving their phosphotriesterase activity toward other hazardous OPs and toward catalytic activities related to the prevention of atherosclerosis.