Comparative Laboratory Evolution of Ordered and Disordered Enzymes*♦

Background: Intrinsic protein disorder has been hypothesized to be advantageous for protein evolution. Results: Parallel evolution of structured and disordered dihydrofolate reductases afforded comparable increases in activity without substantially changing the dynamic properties of the starting scaffolds. Conclusion: Ordered and disordered enzymes can be similarly evolvable. Significance: Understanding how structural (dis)order influences evolutionary pathways may aid efforts to engineer existing proteins for new tasks. Intrinsically disordered proteins are ubiquitous in nature. To assess potential evolutionary advantages and disadvantages of structural disorder under controlled laboratory conditions, we directly compared the evolvability of weakly active ordered and disordered variants of dihydrofolate reductase by genetic selection. The circularly permuted Escherichia coli enzyme, which exists as a molten globule in the absence of ligands, and a well folded deletion mutant of the Bacillus stearothermophilus enzyme served as starting points. Both scaffolds evolved at similar rates and to similar extents, reaching near-native activity after three rounds of mutagenesis and selection. Surprisingly, however, the starting structural properties of the two scaffolds changed only marginally during optimization. Although the ordered and disordered proteins accumulated distinct sets of mutations, the changes introduced likely improved catalytic efficiency indirectly in both cases by bolstering the network of dynamic conformational fluctuations that productively couple into the reaction coordinate.

Intrinsically disordered proteins, found in all domains of life, are essential to key cellular processes such as signaling and regulation (1)(2)(3). Rather than adopt rigid tertiary structures, disordered proteins exist as dynamic ensembles of many roughly isoenergetic conformational states (4). Such plasticity confers fast binding kinetics and high specificity, despite modest affinity, and can be advantageous for molecular recognition and protein modification (5,6). Substantial structural disorder has even been shown to be compatible with enzyme catalysis (7).
Although intrinsic disorder contravenes the traditional view that proteins should possess a stable, well defined tertiary fold to be fully functional, intrinsically disordered proteins are evolutionarily persistent (8,9). Indeed, such biopolymers often possess properties linked to high evolvability, including low sequence complexity, few intramolecular contacts, and a large fraction of solvent-exposed amino acid residues (10,11). Statistical surveys of protein families have suggested that disordered sequences generally undergo more rapid evolution than their ordered counterparts (12). Disordered regions in transcription factors, for example, have been found to be particularly susceptible to positive Darwinian selection, facilitating functional adaptation (9). Because residues introduced into a flexible structure are free to sample multiple structural conformations, divergence and innovation are potentially easier than in a rigid scaffold.
Although disordered sequences may increase adaptive potential, other studies have emphasized the importance of structure and stability for accessing new or improved functions (13)(14)(15). Mutations often cause deleterious losses in thermodynamic and kinetic stability (16), so an ordered scaffold may compensate for destabilization more easily than a disordered one. In fact, high inter-residue contact density has been shown to correlate with increased rates of sequence evolution in yeast proteins (17); a large number of buried residues imparts robustness, enhancing the evolvability of solvent-exposed residues. Predictions that extra stability allows a protein to accept a wider range of mutations while maintaining its native structure have been borne out in selection experiments with enzymes like cytochrome P450 (18) and chorismate mutase (19). Because stable proteins from thermophilic organisms tolerate a greater number of destabilizing mutations than their mesophilic counterparts, they are often preferred as starting points for enzyme design projects.
Directed evolution (20,21) provides a means of directly comparing the evolvability of intrinsically disordered and homologous structured proteins under controlled laboratory conditions. The small (ϳ18-kDa), well studied enzyme dihydrofolate reductase (DHFR) 3 is an attractive model system for this pur-pose. DHFR catalyzes the reduction of 7,8-dihydrofolate (DHF) to 5,6,7,8-tetrahydrofolate by stereospecific hydride transfer from the cofactor NADPH (Fig. 1A), an activity essential for de novo purine biosynthesis and hence cell growth and proliferation (22).
Despite low sequence homology, DHFRs typically fold into two-domain structures, consisting of a central eight-stranded ␤-sheet and four flanking ␣-helices (Fig. 1B). Circular permutation of the well studied Escherichia coli enzyme (EcDHFR) yields intrinsically disordered yet weakly active catalysts (23,24). Biochemical and spectroscopic characterization showed that such permutants behave like molten globules in the absence of ligands but become more structured upon addition of NADP ϩ and the folate antagonist methotrexate (MTX). Although detailed kinetic analyses were never performed, the catalytic activity of circularly permuted DHFR variants in the presence of substrates was reported to be 10 -100 times lower than for EcDHFR (23,24).
In this study, we evolved a circularly permuted variant of the E. coli enzyme (cpDHFR) in parallel with a more structured homolog from Bacillus stearothermophilus (BsDHFR), exploiting in vivo selection for DHFR activity to increase catalytic efficiency. Analysis of the optimized catalysts and their respective evolutionary trajectories sheds light on how (dis)order shapes protein evolution.

EXPERIMENTAL PROCEDURES
Bacterial Strains-E. coli XL1-Blue (Stratagene) was used for general cloning and amplification purposes (25). The DHFRdeficient E. coli strain MH829, created by replacement of folA by a kanamycin resistance cassette, was used for selection experiments (26). The tetracycline-resistant DHFR-deficient E. coli strain MH829 tet was used for protein expression. It was constructed by horizontal gene transfer using XL1-Blue cells as the tet(M) donor. Tetracycline-resistant cells were subsequently selected for tetracycline and kanamycin resistance.
DNA Manipulation-Molecular cloning was performed according to standard procedures (27). Oligonucleotide synthesis and DNA sequencing were carried out by Microsynth (Balgach, Switzerland).
Plasmid Construction-Plasmids pKTCTET-0 and pKTNTET-0 were obtained from Dr. Kathrin Roderer (28). Plasmid His-pKTNTET-0 (4011 bp) was constructed by ligating the NdeI/ NheI cleavage fragment from pKTNTET-0 with a synthetic DNA fragment (5Ј-TA ATG CAT CAT CAT CAT CAT CAT TCG TCG GGC CAT ATG) encoding the His tag and linker sequence. In this way, the original NdeI restriction site of pKTNTET-0 was eliminated and a new NdeI restriction site introduced.
The gene encoding cpDHFR was derived from the gene encoding EcDHFR. It was constructed by overlap extension PCR of two fragments. The first fragment was obtained with primer A (5Ј-AC TAT CAT ATG CAG AAA CTG TAT CTG ACC CAT AT), which has an NdeI restriction site, and primer B (5Ј-CAG GCT AAT CAT GCC GCT GCC GCG GCG TTC CAG AAT TTC AAA GC), which includes a sequence that encodes a GSG linker and is partially complementary to the second fragment. The second fragment was amplified using primer C (5Ј-CTG GAA CGC GCG GGC AGC GGC ATG ATT AGC CTG ATT GCA GCA CT), which contains the GSG linker sequence, and primer D (5Ј-C CTA CAC TAG TTA TTA CGC TTT CGG CAG AAA CTG TTC AT), which contains two stop codons and an SpeI restriction site. The combined fragments were ligated into NdeI/SpeI-cleaved pKTCTET-0 to give pKTCTET-cp.
The gene for BsDHFR was a generous gift from Prof. Judith Klinman. It was amplified by primers bsDHFR_fwd (5Ј-AC TAT CAT ATG ATT TCG CAC ATT GTG GCA AT), which introduced an NdeI restriction site, and bsDHFR_rev (5Ј-CCT ACA CTA GTT ATT ATT TCG CCT TTT CCG CTC), which introduced two stop codons and an SpeI cleavage site. Ligation into NdeI/SpeI-cleaved pKTCTET-0 afforded pKTCTET-bs.
The gene encoding the ⌬121BsDHFR variant was constructed by a QuikChange reaction, using the primers ⌬121bs_fwd (5Ј-GAC AAA AAT TTT TGC TTC CCC CGG CGA TAC GT) and ⌬121bs_rev (5Ј-GGA AGC AAA FIGURE 1. Dihydrofolate reductase. A, DHFR catalyzes the reduction of DHF to 5,6,7,8-tetrahydrofolate (THF) by stereospecific hydride transfer from NADPH. B, the enzyme has a mixed ␣/␤ structure and binds the nicotinamide (blue) and folate (orange) in a cleft between two domains (Protein Data Bank 1RX2). The circularly permuted variant cpDHFR was constructed by linking the original termini with a GSG peptide linker and generating new termini by cleavage between residues 107 and 108 (dashed arrow). The deletion variant was generated by removing residue 121 from BsDHFR (indicated by ⌬).
Library Construction-The genes encoding cpDHFR and ⌬121BsDHFR were mutagenized by error-prone PCR (Gene-Morph II Random Mutagenesis Kit, Stratagene) using the T7_fwd and T7_rev primers, which hybridize upstream and downstream of the gene, respectively. The PCR products were purified via a 2% agarose gel and the NucleoSpin gel extraction kit (Macherey-Nagel). The NdeI/SpeI-digested fragments were ligated into NdeI/SpeI-cleaved pKTCTET-0. About 100 ng of the ligation product was transformed into 50 l of electrocompetent XL1-Blue cells for library quality control. After 1 h of incubation in SOC medium at 37°C, the cells were plated on LB agar plates containing 50 g/ml ampicillin. Six colonies were picked and sequenced to estimate the average number of newly introduced mutations. About 1 g of ligation product was transformed into 50 l of electrocompetent MH829 cells. The cells were transferred to 1 ml of SOC medium containing 50 g/ml thymidine and incubated for 1 h at 30°C. Afterward, 100 l of the cell suspension was plated on LB agar plates containing 50 g/ml thymidine and 50 g/ml ampicillin to determine the transformation efficiency. The remaining cells were incubated overnight in 10 ml of LB liquid medium, supplemented with 50 g/ml thymidine and 50 g/ml ampicillin to ensure optimal growth conditions. The resulting cultures were used for library selection experiments.
Library Selection Experiments-The precultured cells were washed three times in M9 liquid medium (6 mg/ml Na 2 HPO 4 , 3 mg/ml KH 2 PO 4 , 1 mg/ml NH 4 Cl, 0.5 mg/ml NaCl, 0.2% (w/v) D(ϩ)-glucose, 5 g/ml thiamine-HCl, 1 mM MgSO 4 , 0.1 mM CaCl 2 , 50 g/ml ampicillin) and subsequently incubated in M9 liquid medium for 1 h at 30°C. The cultures were diluted to 10 3 -10 5 cells (assuming that an A 600 of 1 corresponds to 10 8 cells/ml), and 5 ϫ 10 2 to 10 3 cells were plated on nonselective and selective plates containing increasing concentrations of tetracycline. Selection experiments were carried out on M9 agar plates, pH 7.0, containing M9 liquid medium plus 15 g/liter agar and 0 -100 ng/ml tetracycline. The plates were incubated at 30°C for 1-5 days. Full complementation was achieved when the number of cells growing on the selective plates equaled the number of cells growing on nonselective plates.
Complementation Experiments-For liquid growth assays, single colonies were used to inoculate 5-ml precultures in LB-AT medium (25 g/liter LB medium, 50 g/ml ampicillin, 50 g/ml thymidine). After overnight incubation, 1 ml of precultured cells was grown in 5 ml of fresh LB-AT medium for 4 h at 30°C, 230 rpm. For growth tests, 1 ml of cells was washed three times in 500 l of M9 medium and subsequently incubated in the same medium for 1 h at 30°C. For growth curves, 4 ml of liquid LB-AT medium or 4 ml of liquid M9 medium was inoculated to a starting A 600 of 0.05. The cultures were incubated at 30°C and 230 rpm, and the A 600 was recorded for 1-5 days. The experiments were conducted in triplicate in the presence and absence of 5 ng/ml tetracycline.
For growth assays on M9 agar plates, single colonies were used to inoculate 5 ml of LB-AT precultures. The precultured cells were washed three times in liquid M9 medium and subsequently incubated in M9 medium for 1 h at 30°C. The cultures were used for three-dimensional streak outs on M9 agar plates (pH 7.0) containing 0 -100 ng/ml tetracycline. The plates were incubated at 30°C for 1 to 5 days.
EcDHFR Production and Purification-Electrocompetent MH829 tet cells were co-transformed with pKTCTET-ec, containing the gene for EcDHFR, and pT7POLTS, which encodes the T7 polymerase (28). 500 ml of LB medium, containing 50 g/ml ampicillin, 30 g/ml chloramphenicol, and 50 g/ml thymidine (LB-ACT), were inoculated with an overnight preculture to a starting A 600 of 0.05 and incubated at 37°C until an A 600 of 0.8 -1 was reached. Gene expression was induced with 2 g/ml tetracycline. Cells were harvested by centrifugation after 4 -16 h of growth at 37°C and washed with 50 ml of 0.9% NaCl. Cell pellets were stored at Ϫ20°C.
Cells were solubilized in 50 mM Tris-HCl, 1 mM EDTA, pH 7.0, and lysed by sonication in the presence of 1 mg/ml lysozyme. After DNase (5 g/ml) and RNase (10 g/ml) treatment, the solution was centrifuged at 14,000 rpm for 20 min at 4°C. The supernatant was stirred with 45% (v/v) NH 4 (SO 4 ) 2 for 30 min at 4°C. After centrifugation at 14,000 rpm for 20 min at 4°C, the supernatant was dialyzed against 10 mM potassium phosphate buffer, pH 7.0, 50 mM KCl, 0.1 mM EDTA at 4°C. The protein was purified by ion exchange chromatography on a diethylaminoethyl (DEAE)-Sepharose column pre-equilibrated with 10 mM potassium phosphate buffer, pH 7.0, 50 mM KCl, 0.1 mM EDTA. Protein was eluted with a gradient of 0 -1 M KCl. Enzyme-containing fractions were concentrated with a 3-kDa Amicon Ultra-15 Centrifugal Filter Unit (Millipore) and injected onto a Superdex 75 size exclusion column pre-equilibrated with 50 mM HEPES, pH 6.8, 150 mM NaCl, and 3 mM 2-mercaptoethanol. The correct mass of the purified protein was verified by electrospray mass spectrometry (ESI-MS).
Production and Purification of cpDHFR and Variants-Electrocompetent MH829 tet cells were co-transformed with pT7POLTS and the pKTCTET-cp plasmid encoding the cpDHFR variant of interest. The proteins were produced as described for EcDHFR. After cell lysis and centrifugation, the pellet was resuspended by stirring for 40 min in 25 ml of 1.5 M NaCl, 60 mM EDTA, 2% Triton X-100 at 4°C. The solution was centrifuged at 14,000 rpm for 20 min at 4°C, and the pellet washed three times (100 mM Tris-HCl, pH 7.0, 20 mM EDTA). After solubilization in 100 mM potassium phosphate buffer, pH 7.0, containing 6 M urea and 3 mM 2-mercaptoethanol at room temperature, the solution was centrifuged at 14,000 rpm for 20 min at 4°C. The protein was dialyzed against refolding buffer (10 M protein, 25 mM Tris-HCl, pH 8.0, 0.5 mM EDTA, 3 mM 2-mercaptoethanol) at 4°C. Precipitated proteins were separated by centrifugation, and the supernatant was dialyzed against 10 mM potassium phosphate buffer, pH 7.0, 50 mM KCl, 0.1 mM EDTA at 4°C. The protein was subsequently purified as described for EcDHFR. The correct mass of all variants was verified by ESI-MS.
BsDHFR Production and Purification-Electrocompetent MH829 tet cells were co-transformed with pT7POLTS and pKTCTET-bs. The protein was produced as described for EcDHFR and purified in analogy to the procedure described previously (30). Briefly, after cell lysis, the supernatant was heat-precipitated at 55°C for 20 min. The re-centrifuged pro-tein solution was concentrated with a 3-kDa Amicon Ultra-15 Centrifugal Filter Unit (Millipore) and injected into a Superdex 75 size exclusion column in 50 mM HEPES, pH 6.8, 150 mM NaCl. The correct mass of the purified protein was verified by ESI-MS.
Production and Purification of ⌬121BsDHFR and Variants-Electrocompetent MH829 tet cells were co-transformed with pT7POLTS and the His-pKTNTET-bs plasmid encoding the ⌬121BsDHFR variant of interest. Gene expression was induced with 0.5 g/ml tetracycline, and the cells were harvested by centrifugation after 14 -16 h at 18°C. Cell pellets were stored at Ϫ20°C. After cell lysis, the soluble fraction was loaded onto a nickel-nitrilotriacetic acid-agarose column (Qiagen) in 50 mM potassium phosphate buffer, pH 8.0, 300 mM NaCl, 10 mM imidazole and washed with the same buffer containing 30 mM imidazole. The protein was eluted in the same buffer with 250 mM imidazole. The elution fraction was concentrated with a 3-kDa Amicon Ultra-15 Centrifugal Filter Unit (Millipore) and injected onto a Superdex 75 size exclusion column preequilibrated with 50 mM HEPES, pH 6.8, 150 mM NaCl. Proteins were eluted from the column with the same buffer. The mass of all variants was verified by ESI-MS.
Protein Concentration Determination-Protein concentrations were determined with the Coomassie Plus TM kit (Pierce) and by absorbance spectroscopy using the calculated extinction coefficient at 280 nm determined by the Bioinformatics Resource Portal ExPASy. Extinction coefficients for ⌬121BsDHFR and its variants were determined as described previously (31).
Enzyme Kinetics-Kinetic measurements were performed at 25°C using final enzyme concentrations of 2-200 nM in PBS, pH 7.4, containing 5 mM DTT. The concentrations of substrate stock solutions were determined for NADPH at 340 nm (⑀ 6,220 M Ϫ1 cm Ϫ1 ). For DHF, an extinction coefficient of 28,000 M Ϫ1 cm Ϫ1 at 282 nm was used. Initial rates were determined by monitoring the decrease in signal at 340 or 365 nm at low and high substrate concentrations, respectively. The extinction coefficient of 11,800 M Ϫ1 s Ϫ1 at 340 nm was reported previously (32), and the extinction coefficient of 5,740 M Ϫ1 s Ϫ1 at 365 nm was determined from the difference in absorbance at 365 nm before and after the reaction in a two-chamber cuvette. Initial velocities were plotted as a function of NADPH or DHF concentration, and the steady-state parameters k cat and K m were determined using the Michaelis-Menten equation.
Circular Dichroism (CD) Spectroscopy-Far-UV CD experiments were performed on a CD spectropolarimeter (model 202) from AVIV Instruments Inc. Spectra were recorded at 25°C in 10 mM potassium phosphate buffer, pH 7.4, in the presence or absence of 25 M MTX with 5-7 M protein. Spectra were obtained by averaging five wavelength scans taken in 0.5nm steps, with a signal averaging time of 2 s and a bandwidth of 1 nm.
FluorescenceSpectroscopy-ForANSbindingstudies,thefluorescence spectra were recorded at 25°C in 100 mM potassium phosphate buffer, 50 mM KCl, 0.1 mM EDTA, pH 7.4, with a fluorescence spectrometer (Photon Technology International). Spectra were obtained in 1-cm quartz cuvettes by averaging three wavelength scans with a signal averaging time of 1 s, in 1-nm steps. Excitation and emission slits were 10 nm. For ANS binding studies, 10 M ANS solution was excited at 370 nm, and fluorescence was monitored at 400 -600 nm in the presence and absence of 10 M protein.
Chemical Denaturation-GdnHCl-induced transition curves were obtained by fluorescence spectroscopy on a 96-well plate reader (Thermo Scientific) at 25°C using 0.15 mg/ml protein in 100 mM potassium phosphate buffer, 50 mM KCl, 0.1 mM EDTA, pH 7.4, containing 0 to 7.4 M GdnHCl. Samples were FIGURE 2. Sequence alignment of EcDHFR, cpDHFR, and ⌬121BsDHFR. EcDHFR is represented in duplicate for better comparison with cpDHFR. The first sequence is highlighted in green, the second sequence in red. The linker sequence inserted for cpDHFR construction is highlighted in yellow. ⌬121BsDHFR is numbered according to BsDHFR; the deleted residue is indicated by -. APRIL 10, 2015 • VOLUME 290 • NUMBER 15 prepared starting from native and unfolded protein to check for reversibility. Fluorescence spectra were recorded from 300 to 450 nm. The bandwidth was 1 nm for excitation at 280 nm and 10 nm for emission. The shift of the wavelength of the emission maximum was determined as a function of the GdnHCl concentration and evaluated by nonlinear regression according to Ref. 33  Tryptic Digests-DHFR variants (0.5 mg/ml) were incubated with 0.05 mg/ml bovine pancreatic trypsin (Fluka) at 15 and

Directed Evolution of a Disordered Enzyme
25°C. At different time points, 30-l aliquots were removed from the reaction mixture and quenched with 10 l phenylmethanesulfonyl fluoride (50 mM in 2-propanol). The samples were separated via SDS-PAGE, and band intensities were quantified with the program ImageJ (National Institutes of Health, Bethesda). Intensities were normalized to the band intensity of an undigested sample.

RESULTS
Characterization of the Intrinsically Disordered Starting Point-EcDHFR was circularly permuted as described by Gudkov and co-workers (23) by linking the N and C termini with a Gly-Ser-Gly tripeptide and cleaving the chain between residues 107 and 108 (Fig. 1B). Our construct, cpDHFR, differs from the literature variant in retaining the starting methionine at the linkage site and the original glutamine at position 108 (Fig. 2). It nevertheless exhibits all the hallmarks of a molten globule (Fig. 3). Far-UV CD spectra showed that cpDHFR is pre-dominantly ␣-helical, whereas EcDHFR has higher ␤-strand content (Fig. 3A). Temperature denaturation (Fig. 3B) and GdnHCl-induced unfolding (Fig. 3C) are less cooperative for cpDHFR than for EcDHFR, indicating multiple unfolding conformations. As is typical for disordered proteins, the circularly permuted protein also binds the hydrophobic dye ANS, giving rise to enhanced and blue-shifted fluorescence (Fig. 3D). Finally, broad peaks and poor dispersion in the 1 H NMR spectrum of cpDHFR (Fig. 3E) attest to an ensemble of rapidly interconverting species.
As seen previously for other DHFR permutants (23,24), the competitive inhibitor MTX induces conformational changes in cpDHFR (Fig. 3). In contrast to EcDHFR, which showed only a small increase in the CD signal at 218 nm, MTX binding to cpDHFR causes a dramatic shift in the CD spectrum, indicating a transition from a predominantly ␣-helical to a mixed ␣/␤structure similar to apoEcDHFR (Fig. 3A). Substantial ordering of the cpDHFR structure by MTX is further indicated by increased cooperativity of temperature-induced denaturation (Fig. 3B), weaker ANS binding (Fig. 3D), and greater peak dispersion in the 1 H NMR spectrum (Fig. 3E). Regardless, cpDHFR still appears to be substantially less structured than EcDHFR even when complexed with ligand.
Although not natively folded, cpDHFR is catalytically active (Fig. 3F) but roughly 20,000-fold less efficient than EcDHFR ( Table 1). The permuted protein has an approximately 100-fold lower k cat value than EcDHFR and K m values for DHF and NADPH that are 15-and 8-fold higher, respectively.
Creation and Characterization of a Well Structured Counterpart-For the purpose of comparison, a well structured enzyme with activity similar to that of cpDHFR was engineered starting from a thermostable DHFR from B. stearothermophilus (BsDHFR) (30). To reduce the efficiency of this highly active catalyst (Table 1), two circularly permuted and three deletion variants were constructed. The permuted proteins were designed as described for cpDHFR, using the same linker to connect the original termini and introducing new termini between either Val-109/Asp-110 and Asp-110/Arg-111 (BsDHFR numbering). The deletion variants were obtained by deleting Arg-52 and Phe-121, which are located in mutationally

TABLE 3 ⌬121BsDHFR variants obtained after three rounds of directed evolution
Residue numbering corresponds to the ⌬121BsDHFR sequence shown in Figure 2.
The variant depicted in boldface type was biochemically characterized.  ⌬121BsDHFR (B). A, arrow indicates the original N terminus of EcDHFR. The amino acid residues are numbered as depicted in Fig. 2. The residues identified as DHFR sector residues are highlighted in green (A and B). The variants that were characterized in detail are highlighted in pink (A) and cyan (B), respectively. sensitive loop regions of the protein (35), alone and in combination. The resulting variants were analyzed by their ability to complement an E. coli DHFR knock-out strain that requires an extra-chromosomal DHFR for growth (26). Most of the transformants grew at wild-type levels; only cells producing ⌬121BsDHFR showed reduced growth.

Directed Evolution of a Disordered Enzyme
In vitro studies confirmed that ⌬121BsDHFR is less efficient than WT BsDHFR (Table 1 and Fig. 4). Deletion of Phe-121 reduced k cat values 48-fold and increased the K m value for DHF 17-fold; the K m value for NADPH remained unchanged. As a result, the catalytic efficiency of ⌬121BsDHFR is nearly 1000fold lower than that of the WT enzyme and only an order of magnitude higher than that of cpDHFR (Table 1).
Although WT BsDHFR is thermostable, removal of Phe-121 also decreased the T m of the enzyme by 20°C (Fig. 4C) and the ⌬G 0 for chemical denaturation by 5 kcal/mol (Fig. 4D and Table  1). Nonetheless, the unfolding behavior of ⌬121BsDHFR is still highly cooperative (Fig. 4, C and D). Moreover, its far-UV CD spectrum is indistinguishable from that of BsDHFR (Fig. 4E), and its 1 H NMR spectrum shows good peak dispersion (Fig. 4F), consistent with a well structured protein.
Laboratory Evolution under Selection Pressure-The disordered and structured scaffolds were subjected to three rounds of random mutagenesis and selection. Gene libraries, constructed by error-prone PCR, were placed under the control of the tunable tetracycline-dependent P tet promoter. Active catalysts were identified by their ability to complement the DHFR deficiency of the E. coli knock-out strain under auxotrophic conditions (26). In this system, selection pressure can be steadily increased simply by reducing the tetracycline concentration and thus lowering the intracellular concentration of the DHFR variants (Fig. 5) (36).
In the first selection round, cells grew only at high tetracycline concentrations because of the low activity of the starting variants. A slight advantage was observed for the ⌬121BsDHFR library, most likely due to the higher starting activity of the deletion variant. After the second round of mutagenesis, 7% of the colonies producing cpDHFR variants and 21% of the colonies producing ⌬121BsDHFR variants grew in the absence of inducer. After three rounds of mutagenesis, the number of complementing clones growing under the most stringent selection conditions increased to roughly 50% for both libraries (Fig. 5).
After three rounds of evolution, representative members from each library were isolated and sequenced (Fig. 6). The successful variants harbored four to six substitutions (Tables 2  and 3). Eight different sites in cpDHFR, distributed across the entire scaffold, were mutated. These include four consensus mutations (H18Y, P24S, M57R, and I117V), plus four additional changes that appeared in various combinations. In contrast, 14 completely different sites were targeted in ⌬121BsDHFR, all of which are surface-exposed. Three consensus mutations (A49S, F118S, and G143W) emerged, but only A49S is present in every sequenced variant; eight of the remaining mutations are unique to a single clone.
Characterization of Improved Variants-For a better understanding of the evolutionary changes at the molecular level, the variant from each library containing the most frequently occurring mutations (cpDHFR.4 and ⌬121BsDHFR.3; see Tables 2   and 3) was characterized in detail. Following recloning, stringent growth tests with the DHFR-deficient strain transformed with the evolved enzymes confirmed full complementation.
Steady-state kinetic studies pinpointed increased catalytic activity as the reason for improved growth in both cases (Table  1 and Fig. 7). For the evolved cpDHFR variant, which has six mutations, catalytic efficiency rose 330-fold. The evolved ⌬121BsDHFR variant has four mutations and showed a 290fold increase. These improvements reflect 20 -30-fold increases in the k cat parameter and smaller decreases in the K m values for the two substrates. Although evolution of ⌬121BsDHFR recovered nearly full wild-type activity, the evolved cpDHFR variant is 17-fold less efficient, reflecting the lower activity of the disordered starting point.
The second-site mutations had relatively minor effects on protein structure and stability (Table 1). Evolved cpDHFR.4 showed a T m increase relative to cpDHFR (Fig. 7B) but no differences in GdnHCl-induced unfolding (Fig. 7C) or proteolytic degradation (Fig. 7D). Moreover, the broad thermal transition (Fig. 7B), considerable albeit diminished ANS binding (Fig. 7E), and poor NMR signal dispersion (Fig. 7F) indicated that the variant retains a significant degree of intrinsic disorder. In contrast, temperature-induced unfolding, chemical denaturation studies, proteolytic degradation, ANS binding, and NMR spectra were indistinguishable for ⌬121BsDHFR and its evolved variant (Fig. 7, B-F).

DISCUSSION
The active site of DHFR is located in a cleft between two protein subdomains (Fig. 1B). Structural and dynamic changes in these domains, induced by binding and dissociation of substrates, cofactors, and products (37), are critical to the function of the enzyme (22). NMR studies and molecular dynamics simulations of EcDHFR, supported by the results of site-directed mutagenesis, have produced maps of correlated conformational motions that productively channel bound intermediates along the reaction coordinate and facilitate hydride transfer (38 -42). Disrupting these interaction networks can cause significant losses in stability and catalytic efficiency as seen for the deletion variant ⌬121BsDHFR. In more extreme cases, such as the circularly permuted cpDHFR, structural integrity is lost as well, giving rise to intrinsic disorder.
Laboratory evolution provides a practical means to recover lost activity through second-site mutations. Three rounds of mutagenesis and genetic selection increased the efficiency of both ⌬121BsDHFR and cpDHFR about 300-fold, affording variants that fully complement the DHFR deficiency of an E. coli auxotroph under stringent selection conditions. At least in this specific case, structural plasticity appears to have had a relatively modest impact on evolvability. Contrary to suggestions that disordered proteins evolve significantly faster than ordered proteins in nature (12), the compact and molten globular DHFR variants evolved at similar rates in this system. An even less structured protein than the circularly permuted DHFR variant might be needed to recapitulate such effects.
Although protein folding and stability might have been expected to co-evolve with function, optimization of ⌬121BsDHFR and cpD-HFR resulted in only subtle structural changes to the starting scaffolds. Thus, the ⌬121BsDHFR.3 variant did not recover the thermostability of the parent scaffold but remained well packed after three rounds of evolution. The evolved circular permutant cpDHFR.4, however, retained significant disorder in the absence of ligands. In the latter case, reduced ANS binding and slightly more cooperative thermal denaturation provide evidence for some decrease in intrinsic disorder, but poor NMR signal dispersion indicates that the optimized cpDHFR variant is not native-like. Nevertheless, efficient complementation of the DHFR deficiency by cpDHFR shows that the dramatic disorder-to-order transition that accompanies ligand binding and subsequent catalytic events occurs on a time scale that is fully compatible with biologically relevant turnover times for folate metabolism (43).
Intriguingly, optimization of the compact and intrinsically disordered proteins followed distinct evolutionary trajectories. Indeed, the selected variants have no mutations in common. These distinct outcomes may reflect the low sequence identity (39%) between the starting scaffolds and/or the idiosyncratic perturbations to the parent enzymes, i.e. deletion as opposed to circular permutation. However, consistent with the greater tol-erance of more ordered scaffolds to substitution (13)(14)(15), the structurally compact ⌬121BsDHFR harbors a larger number of unique changes and fewer consensus substitutions than the disordered cpDHFR protein. Strikingly, all the ⌬121BsDHFR mutations are at surface sites, which generally evolve more rapidly than buried residues (11,44). Surface substitutions also predominate in the cpDHFR clones, although two buried residues, Ile-9 and Ile-117, were also conservatively mutated to valine.
Because the mutations in both proteins are generally distant from the active site, their influence on catalytic efficiency must necessarily be indirect. More detailed spectroscopic studies will be needed to elucidate the precise mechanistic consequences of these changes, but it is likely that mutation favorably alters the population distribution of ligand-bound conformations in both proteins, facilitating progression of the bound intermediates along the reaction coordinate (37). Both theoretical (45,46) and experimental (47)(48)(49) studies have shown that a dynamic network of coupled motions influences DHFR catalysis, and evidence suggests that this network has been maintained over billions of years of evolution (50,51).
Notable in this context, all the targeted sites in cpDHFR.4 are in contact with the sparse collection of physically contiguous and co-evolving amino acids identified by statistical coupling analysis of the DHFR protein family (Fig. 8) (52,53). This network, called a sector (54), functionally links the active site to distantly positioned surface sites and is strongly correlated with the network of residues in DHFR undergoing millisecond conformational fluctuations. Although these sites need not be directly relevant to catalysis (48), the mutations fixed during directed evolution conceivably increased protein fitness by bolstering or extending this network, restoring critical dynamic motions that were disrupted by circular permutation. In contrast, in ⌬121BsDHFR.3, which contains a less structurally disruptive deletion, the sector residue Ala-49 was mutated to serine. None of the other targeted residues is physically contiguous with the sector, suggesting that mutation at these sites either established new connectivities or enhanced activity by an alternative mechanism.
These comparative selection experiments show that, with respect to a weak pre-existing enzymatic activity, ordered and FIGURE 8. Location of mutations in the optimized DHFR variants relative to the sector of co-evolving residues in the DHFR family. A, in cpDHFR.4, all six mutations (red), mapped onto the crystal structure of EcDHFR (Protein Data Bank 1RX2), are physically contiguous with sector residues (yellow surface). B, mutations in ⌬121BsDHFR.3 are mapped onto the crystal structure of BsDHFR (Protein Data Bank 1ZDR); they include one sector residue (A49S, magenta) and three other mutations (blue) that are not in contact with sector sites. Note that different perspectives were chosen for A and B to make all of the mutated sites visible. NADP ϩ and folate are shown with black and gray carbons, respectively. disordered scaffolds can be similarly evolvable. In both scaffolds, relatively few, albeit distinct, mutations sufficed to counteract the deleterious effects of amino acid deletion or circular permutation and thereby improve binding and catalysis of DHF reduction. It is notable that optimization of the circularly permuted protein did not require sacrificing dynamic character to achieve high catalytic efficiency. Although further evolution of the disordered enzyme may afford more structured and stable catalysts, as seen during the optimization of an engineered molten globular chorismate mutase (55,56), additional studies will be needed to assess whether such plasticity ultimately enhances the adaptive potential of this scaffold for completely new tasks.