Multistep mutational transformation of a protein fold through structural intermediates

Abstract

New protein folds may evolve from existing folds through metamorphic evolution involving a dramatic switch in structure. To mimic pathways by which amino acid sequence changes could induce a change in fold, we designed two folded hybrids of Xfaso 1 and Pfl 6, a pair of homologous Cro protein sequences with ~40% identity but different folds (all‐α vs. α + β, respectively). Each hybrid, XPH1 or XPH2, is 85% identical in sequence to its parent, Xfaso 1 or Pfl 6, respectively; 55% identical to its noncognate parent; and ~70% identical to the other hybrid. XPH1 and XPH2 also feature a designed hybrid chameleon sequence corresponding to the C‐terminal region, which switched from α‐helical to β‐sheet structure during Cro evolution. We report solution nuclear magnetic resonance (NMR) structures of XPH1 and XPH2 at 0.3 Å and 0.5 Å backbone root mean square deviation (RMSD), respectively. XPH1 retains a global fold generally similar to Xfaso 1, and XPH2 retains a fold similar to Pfl 6, as measured by TM‐align scores (~0.7), DALI Z‐scores (7‐9), and backbone RMSD (2–3 Å RMSD for the most ordered regions). However, these scores also indicate significant deviations in structure. Most notably, XPH1 and XPH2 have different, and intermediate, secondary structure content relative to Xfaso 1 and Pfl 6. The multistep progression in sequence, from Xfaso 1 to XPH1 to XPH2 to Pfl 6, thus involves both abrupt and gradual changes in folding pattern. The plasticity of some protein folds may allow for “polymetamorphic” evolution through intermediate structures.

Introduction

Metamorphic proteins undergo major changes in the folding pattern of a domain as a function of solution conditions, ligand binding, posttranslational modification, or accumulation of sequence mutations.1, 2, 3 Metamorphism in proteins plays a role in functional regulation, disease, and the evolution of new protein folds. Metamorphic evolution,1, 2 in which a continuous process of sequence change gives rise to major structural transitions, is particularly intriguing given its fundamental implications for the landscape of correlated sequence and structure change in proteins.

Evolution of an α + β fold from an all‐α fold has occurred in Cro, a family of homodimeric helix‐turn‐helix (HTH) DNA‐binding domains (Fig. 1). The C‐terminal region of Cro proteins, which coincides with the homodimer interface, contains two to three α helices in the ancestral all‐α fold, but is remodeled to a β‐hairpin structure in the α + β descendant.4, 5, 6, 7 The N‐terminal region has three conserved α helices, including the HTH motif, along with a single β strand at the N terminus in the α + β fold. Overall, the domain can be divided into a conserved HTH subdomain and a divergent dimerization subdomain, localized mostly but not strictly to the N‐and C‐terminal halves of the sequence, respectively.

Figure 1.

Evolution of fold in the Cro family of transcription factors. (a) Xfaso 1 Cro, representing the ancestral all‐α fold. (b) Pfl 6 Cro, representing the descendant α + β fold. (c) Sequence alignment of Xfaso1 and Pfl 6 Cro, with regions of interest highlighted. Xfaso 1 and Pfl 6 are 38–40% identical in sequence, depending on exactly how the sequences are aligned. The helix‐turn‐helix (HTH) subdomain is structurally conserved and has an HTH motif spanning α2‐α3, which contains most of the DNA‐contacting residues. The dimerization subdomain is not structurally conserved. For each Cro protein in (a) and (b), the second subunit in the biological homodimer is shown in gray. Cro proteins function as homodimers but many are monomeric as free proteins in solution.

The remodeling of Cro structure is an example of metamorphic evolution.2, 3 The sequences of all‐α and α + β Cro proteins have as much as ~40% identity and show similarity across both N‐and C‐terminal halves [see Fig. 1(c)].5, 7 The fold switch thus resulted from the accumulation of amino‐acid substitutions or small insertion/deletion events rather than through heterologous recombination or frameshifting.5, 7 In addition, cro genes are orthologs4 and part of the conserved immunity region of lambdoid bacteriophages. Cro is essential for lytic growth of phage λ 8, 9, 10, 11 and appears to be under purifying selection.12 The importance of Cro proteins in phage biology and the lack of apparent gene duplication make it likely that the fold switch occurred with nearly continuous preservation of stable folding and function.

The likelihood that Cro evolution passed through a metamorphic transition raises questions of mechanism at the level of sequence, structure, and function. Here we focus on how sequence evolution could bring about a change in fold with continuous preservation of folding stability. A two‐state switch between all‐α and α + β structures (Fig. 1) implies that highly similar sequences, or even single sequences, existed that simultaneously encoded the two folds. Either a key mutation abruptly triggered a fold switch, or the mutational trajectory passed smoothly through evolutionary bridge sequences13, 14 that encoded both ancestral and descendant folds. A multistate switch involving structural intermediates is also possible, and might circumvent the difficulty of encoding radically different folding patterns in a single sequence. However, all known Cro structures to date hew closely to two folding patterns, offering no positive evidence for intermediate forms.

To explore the potential sequence and structure space available to Cro evolution, we have studied hybrids of Cro sequences with the all‐α and α + β folds. We have examined hybrids of P22 and λ Cro,15, 16 which have 25% sequence identity, and hybrids of Xfaso 1 and Pfl 6 Cro,17, 18 which have about 40% sequence identity (Fig. 1). Hybrid point mutagenesis and chimeragenesis showed that the sequence determinants of all‐α and α + β folds within the C‐terminal region itself are substantially different, and that none of these four natural Cro sequences has a C‐terminal region that can switch readily to the alternate fold (although the C terminus of Xfaso 1 may be within a few substitutions).16, 17 However, mutagenesis of P22 and λ Cro also suggested that the key C‐terminal determinants of the two Cro folds are essentially compatible and partly orthogonal, despite being different.16 This compatibility was explicitly demonstrated through heuristic design of C‐terminal “chameleon“ sequences19, 20, 21 that adopted the all‐α or α + β fold depending on attachment to the N‐terminal parts of P22 or λ Cro, respectively.15, 16 These studies suggested that evolutionary bridges between Cro folds are plausible at least as far as the key C‐terminal region is concerned, and also suggested that strong fold determinants exist outside of the C‐terminal region.15

Hybrid insertion–deletion studies of the Xfaso 1‐Pfl 6 pair later showed that the sequence length at or near the subdomain junctions (see Fig. 1) exert a strong topological influence on the overall fold.18 Pfl 6 (α + β fold) prefers a short linker to connect helix α3 to strand β2, and a longer N terminus to allow formation of strand β1, which interacts with the β2‐β3 hairpin. Xfaso 1 (all‐α fold) tolerates either a long or short sequence at the N terminus but requires a longer central linker sequence to make the α3‐α4 connection. These findings offer a rationale for why P22‐λ based C‐terminal chameleons switch structure as a function of determinants elsewhere in the sequence.

The above studies suggest one plausible mechanism for Cro evolution, namely that the mutational trajectory proceeded through sequences with C‐terminal chameleon potential, which were then directly induced to switch fold by small indels. Missing from this hypothetical picture, however, is a detailed view of the structures adopted by intermediate sequences in such a mechanism. The chameleon design studies based on the P22‐λ Cro pair did not include high‐resolution structure determinations, stopping instead at the level of secondary structure assignments.

Here we report NMR solution structures of two designed hybrid sequences of the more closely homologous Xfaso1‐Pfl6 pair. As with the P22‐λ chameleons, we designed these hybrids to contain C‐terminal chameleon sequences affixed to the Xfaso 1 or Pfl 6 N terminus. Strikingly, the hybrid sequences adopt folding patterns that are intermediate in secondary structure content between the two parent structures. The progression of sequence from Xfaso 1 to Pfl 6 through the hybrids is accompanied by a complex progression in structure. Metamorphic evolution can, at least in principle, involve multiple steps and intermediate structures.

Results

Hybrid chameleon design strategy

In most respects our design approach to Xfaso1‐Pfl6 hybrids is analogous to the previous P22‐λ chameleon studies. In both cases we have focused on subunit structure and stability rather than on preserving function. We made no attempt to maintain a functional dimer interface in the C‐terminal region, and in fact included design elements to disfavor solution dimerization, as it is a potential complicating factor in NMR studies. Wild‐type Pfl 6 dimerizes with K_d ~1 mM, while wild‐type Xfaso 1 does not dimerize detectably in solution even at millimolar concentrations.7 Pfl 6 Cro is rendered fully monomeric by introduction of a hybrid I58D substitution from Xfaso 1,22, 23 and we included this residue in our chameleon design.

The preference for working with monomers led to other preliminary considerations. Monomerized versions of λ Cro retain the α + β fold and all its secondary structure elements, but lose order in the last two to three residues of strand β3,22 which make dimer interface contacts but not intrasubunit contacts. The same is true of Pfl 6 I58D, based on temperature factors22 as well as analysis with TALOS, RCI, and CS‐ROSETTA (Supporting Information Fig. S1). In Xfaso 1, the short helix α6 aligns partly to the end of strand β3 of Pfl 6 (see Fig. 1), and is likewise integral to the dimer interface, but remains helical in the monomer. Specifically, in the crystal structure of Xfaso 1 helix α6 is present in a third subunit of the asymmetric unit that is not part of a biological dimer. This region thus may be important for the folding of Xfaso 1 subunits, although some other all‐α Cro proteins such as P22 Cro lack a sixth helix. Our chameleon design [Fig. 2(C)] incorporates the entire sequence of helix α6 from Xfaso 1 (PDVF) to account for its possible importance to the all‐α fold and because of the low order in this region in the α + β monomer.

Figure 2.

Xfaso1‐Pfl 6 hybrid chameleon design considerations. (a) Subunit of Xfaso 1 (PDB ID 3BD1, chain C). (b) Subunit of Pfl 6 (PDB ID 2PIJ, chain A). (c) Sequences of designed Xfaso 1/Pfl 6 hybrids used in this study, annotated with secondary structures observed in wild‐type Xfaso 1 and Pfl 6. In (a) and (b), putative stabilizing features of each protein in the C‐terminal chameleon region (e.g. hydrophobic core residues, hydrogen bonding side chains, and glycines), are illustrated and labeled. These were preserved in the design where possible. Stabilizing helix‐turn‐helix residues of Pfl 6 (relative to Xfaso 1) are also shown, which were used to generate a more stable version of the Xfaso 1 background sequence (XPH0 in C). The ribbon for the chameleon region is highlighted in color in panels a and b, with each structure showing residues retained from the parent sequence in the chameleon hybrid design (red for Xfaso 1, cyan for Pfl 6). These are highlighted in color in (c), with the final chameleon design boxed. Some residues are identical in Xfaso 1 and Pfl 6 and thus retained in the chameleon relative to both parent sequences. The last six residues of the chameleon constructs derive entirely from Xfaso 1, as this region is disordered in Pfl 6 monomers.

For the portion of the C‐terminal region that is ordered in both all‐α and α + β monomers (positions 39–59 of Xfaso 1 and 36–56 of Pfl 6), we chose between the wild‐type Pfl 6 or Xfaso 1 residues, depending on which one appeared more important (Fig. 2). In the P22‐λ studies we had made extensive use of hybrid mutagenesis information to make such choices. The Xfaso1‐Pfl 6 designs, however, predate any of the hybrid mutagenesis work on these proteins,17 and we relied instead on heuristic analysis of crystal structures.7 We identified all buried side chains, salt bridges, hydrogen bonds, as well as unusual backbone conformations in the Xfaso 1 and Pfl 6 crystal structures [Fig. 2(a,b)]. As with previously reported P22‐λ chameleon designs, one or two positions posed a dilemma in which preserving one apparently important interaction led to the destruction of another. In the P22‐λ studies, we resolved these dilemmas by incorporating a third residue type not present in either template sequence. In the current study, we confined ourselves to the “binary sequence space”24 of Xfaso 1 and Pfl 6, and resolved such dilemmas in part through testing of multiple trial designs. Our final chameleon design is shown in Figure 2(C).

Effect of disulfide bond formation on Xfaso 1

One notable feature of the chameleon design is the preservation of Cys 42 and Cys 55 of Xfaso 1 [Fig. 2(a)]. Cys 42 and Cys 55 are free cysteines in the Xfaso 1 crystal structure, but are closely proximal to each other and oxidize readily in air to form a disulfide bond.7 Cysteines at the aligned positions in Pfl 6, meanwhile, would be too far apart in the α + β structure to form a bond. Thus an interesting test of our designs is their behavior under reducing versus nonreducing conditions.

As part of the groundwork for our design study, we thought it prudent to study the effect of Cys 42‐Cys 55 disulfide bonding on the structure and stability of wild‐type Xfaso 1. Formation of the disulfide bond increases the T _m of Xfaso 1 by about 17°C [Fig. 3(A)] but has previously been found to have little effect on its CD spectrum, suggesting little effect on secondary structure.7 Analysis of chemical shifts with TALOS [Supporting Information Fig. S2(c)] also suggests that the backbone conformation (ϕ,ψ) of oxidized Xfaso 1 differs from that of reduced Xfaso 1 only in the helix α5/α6 junction (residues 58–59), along with increased backbone dynamics in part of helix α6 (residues 62–64).

Figure 3.

Disulfide bond formation and helix‐turn‐helix substitutions from Pfl 6 increase the thermal stability of Xfaso 1 while preserving its α‐helical secondary structure. (a) Fitted thermal denaturation profiles monitored by circular dichroism at 222 nm of 25 μM protein in SB250 buffer (50 mM Tris [pH 7.5], 250 mM KCl, 0.2 mM EDTA), with or without DTT. (b) Far ultraviolet circular dichroism spectra at 20°C, at 25 μM protein in a 1 mm cuvette in SB250 buffer with or without reducing agent. Oxidized wild‐type Xfaso 1 was previously shown to have a nearly identical circular dichroism spectrum as reduced Xfaso 1.0037 Spectra in (b) were digitally smoothed to facilitate comparison. Variants in (b) are truncated after residue 65 to remove the disordered C‐terminal region of Xfaso 1. XPH0 represents Xfaso 1 with three helix‐turn‐helix substitutions from Pfl 6: S20A/R24N/N30Q (see also Fig. 1).

The disulfide bond does perturb the tertiary structure of Xfaso 1 somewhat. Overlays of HSQC NMR spectra and chemical shift perturbation analysis suggest some global structural effects [Supporting Information Fig. S2(a,b)]. CS‐Rosetta calculations (Supporting Information Fig. S3) of the oxidized Xfaso 1 structure failed to converge on a single conformation in the helix α5/α6 region, but yielded two conformational families that in our hands could not be distinguished by further experiments. Both of these conformations differ from the crystal structure of reduced Xfaso 1. Formation of the disulfide bond has been found to destroy the DNA binding activity of Xfaso 1 (Branwen Hall, PhD dissertation). The changes in helix α5/α6 suggest that the disulfide bond may disrupt dimer interface contacts made by this region.

Stabilization of the Xfaso 1 helix‐turn‐helix motif

A final consideration that we faced in laying the groundwork for our study was the marginal folding stability of Xfaso 1. Trial efforts to characterize the Xfaso 1‐Pfl 6 chameleon designs were impeded by the poor folding stability of designs introduced into the Xfaso 1 sequence background, even under oxidizing conditions. Reduced wild‐type Xfaso 1 is less thermally stable [T _m~51°C; Fig. 3(a)] than wild‐type Pfl 6 [T _m~64°C; Fig. 3(a)],18 and less tolerant of hybrid mutations in the C‐terminal region.17 Computational studies also suggest that fewer sequences can fold stably to the all‐α fold than to the α + β fold.25

We raised the baseline stability of Xfaso 1 using hybrid substitutions from Pfl 6 within the conserved helix‐turn‐helix motif (HTH). Based on α‐helical propensity and helix N‐capping preference,26, 27 we predicted that three residues in the HTH of Xfaso 1 (Ser 20, Arg 24, and Asn 30) were destabilizing relative to aligned residues in Pfl 6 (Ala 22, Asn 26, and Gln 32, respectively; see Fig. 2). Indeed, the thermal stability of a Ser20Ala/Arg24Asn/Asn30Gln variant of Xfaso 1 under reducing conditions (T _m~65°C) is approximately equal to that of wild‐type Pfl 6 [Fig. 3(a)]; under oxidizing conditions this variant is hyperstable (T _m~82°C). CD spectra [Fig. 3(b)] and chemical shift perturbation analysis (Supporting Information Fig. S4) confirmed that the mutations do not perturb the structure of Xfaso 1 significantly, either in the oxidized or reduced form. Thus, we used this more stable hybrid rather than wild‐type Xfaso 1 as all‐α parent sequence for introduction of C‐terminal chameleon designs (Fig. 2). We refer to it henceforth as XPH0, for Xfaso 1/Pfl 6 Hybrid 0.

Circular dichroism of chameleon hybrid designs

We grafted the chameleon design onto the N‐terminal halves of stabilized Xfaso 1 (XPH0) and wild‐type Pfl 6, generating the hybrid sequences XPH1 and XPH2, respectively [Fig. 2(c)]. Both chameleon hybrids folded successfully (Fig. 4). Far ultraviolet circular dichroism spectra at 20°C under reducing conditions suggest that both XPH1 and XPH2 have helical content at least as strong as that of Pfl 6 based on the ellipticity at 222 nm [Fig. 4(a)]. Such a result is expected given that a structurally conserved HTH subdomain is common to all designs. Under reducing conditions, both XPH1 and XPH2 show thermal denaturation curves with apparent midpoints (T _m) well above ambient temperature [Fig. 4(b)]. However, reduced XPH1 and XPH2 are both less stable than their parent sequences: the apparent T _m of reduced XPH1 is 12°C lower than that of reduced XPH0 (stabilized Xfaso 1 background), and the apparent T _m of XPH2 is 17°C lower than that of Pfl 6.

Figure 4.

Secondary structure and stability of designed hybrid sequences of Xfaso 1 and Pfl6. (a) Far ultraviolet circular dichroism spectra of XPH1 and XPH2 and parent sequences at 20°C. Samples contained 25 μM protein in a 1 mm cuvette in SB250 buffer with or without DTT. (b) Thermal denaturation curves for XPH1 and XPH2 monitored by circular dichroism at 222 nm.

XPH2 and Pfl 6 have similar far ultraviolet CD spectra [Fig. 4(a)], suggesting that XPH2 retains the basic secondary structure, and probably the α + β fold, of the Pfl 6 parent sequence. The cysteines of XPH2 also resist air oxidation to form a disulfide bond, as expected based on their anticipated locations in the α + β fold. At the same time, the more evident minimum dichroism at 222 nm of XPH2 than Pfl 6 suggests a slight increase in helical content.

Meanwhile, XPH1 has much less helical signal than its parent sequences Xfaso 1 and XPH0. Like Xfaso1 and XPH0, it does undergo air oxidation, but both the reduced and oxidized forms show less helicity than expected. The low apparent native helicity of reduced XPH1 could be an artifact of populating the unfolded state at 20°C, given the fairly low T _m, broad melt profile, and strong negative dichroism below 210 nm. But it is unlikely that this can be true of oxidized XPH1, which has a much higher T _m and a steeper denaturation profile. Although the ability of XPH1 to be stabilized by oxidation is consistent with the folding pattern of the Xfaso 1 parent, the low helicity is not. This conundrum led us to seek high‐resolution structures of both hybrids, especially XPH1.

Solution structures of hybrid designs

XPH1 and XPH2 exhibited poor to marginal quality 15N‐¹H correlation NMR spectra (Supporting Information Fig. S5) under reducing conditions, possibly because their low folding stability makes exchange effects between the folded and unfolded state significant [Fig. 4(b)]. Consistent with this interpretation, addition of 1 M trimethylamine N‐oxide (TMAO) as a stabilizing osmolyte improves (but does not otherwise significantly perturb) the spectrum of reduced XPH2 (Supporting Information Fig. S5). Unfortunately, this approach did not work for reduced XPH1, which precipitates extensively upon addition of TMAO. This prevented a high‐resolution solution comparison of XPH1 and XPH2 in reduced form. Instead, we focused our NMR studies of XPH1 on the oxidized form, as disulfide bond formation improves its stability and spectral quality (Supporting Information Fig. S5).

We obtained resonance assignments for oxidized XPH1, as well as for reduced XPH2 in 1 M TMAO, and calculated NMR solution structures using CYANA followed by CNS water refinement (Table 1, Fig. 5). Both solution ensembles are well defined throughout the folded region typical of α + β monomers (up to residue 55–56 in Pfl 6 numbering). Both XPH1 and XPH2 become less ordered beyond this point based on RCI‐S² analysis (Supporting Information Fig. S6) and greater coordinate variability in the ensemble. The ensemble of XPH2 does retain some order near the C terminus, however, in the form of a short helix for residues 57–59; secondary chemical shifts suggest that partial helical character may be present in this region of XPH1 as well.

Table 1.

Statistics for XPH1 and XPH2 solution structures

	XPH1	XPH2
Experimental restraints
Distance restraints
All NOE	699	458
Intraresidue	183	175
Sequential (|i‐j| = 1)	187	105
Medium‐range (2 ≤ |i‐j| ≤ 4)	139	83
Long‐range (|i‐j| ≥ 5)	190	95
Hydrogen bond distances	42	28
Backbone angle restraints	107	108
χ1 angle restraints	13	12
Total restraints	861	606
Average RMSD to mean, residues 1–58 (1–62) for XPH1, residues 1–55 (1–59) for XPH2
Backbone atoms (Å)	0.34 (0.61)	0.50 (0.57)
Heavy atoms (Å)	0.86 (1.13)	1.10 (1.15)
Measure of structural quality
Ramachandran distribution
Most favored, %	90.5	93.8
Additional allowed, %	8.8	6.2
Generously allowed, %	0.5	0.0
Disallowed, %	0.1	0.0
Overall dihedral G factor	−0.26	−0.14

Figure 5.

Progression of sequence and structure from Xfaso 1 to Pfl 6 through hybrid intermediates. (a) Secondary structures of Xfaso 1 and Pfl 6 (α‐helix: red; β‐strand: cyan) and designed chameleon hybrids mapped onto sequence alignment. The number of helical and strand residues is shown at right for each sequence, showing the gradual progression in secondary structure content as the sequence changes. Secondary structure elements at the C terminus tend to show dynamic character and are highlighted with dashed boxes and colored different. Helix α6 of Xfaso 1 (orange) is dynamic in the oxidized form; this region has only weak or transient helical character (pink) in XPH1, and is well formed but apparently mobile in XPH2; the last three residues of strand β3 of Pfl 6 (cyan) are disordered in the Pfl 6 monomer. (b) Ribbon diagrams of parent and hybrid structures, with the conserved N‐terminal region shown in grey, and the variable C‐terminal region (boxed in panel a) shown in color according to secondary structure as in (a). Structural elements in the chameleon hybrids that are distinct from the parent structures are annotated. Global percent identities between each pair of sequences in the progression are calculated as in Table 2. (c) Final solution ensembles of XPH1 and XPH2 from explicit solvent refinement in CNS. Note the less well‐defined nature of the C‐terminal 6–7 residues.

The designed C‐terminal chameleon sequence common to XPH1 and XPH2 is in fact a chameleon. The chameleon region, which includes residues 36–59 in Pfl 6 numbering, or 36–55 if the less ordered C terminus is omitted, has distinct secondary structures and chemical shifts in XPH1 and XPH2 (Supporting Information Fig. , S7). The backbone RMSD for a best‐fit superposition is ~5 Å, and only about 40% of the individual residues have similar backbone conformations (Table 2). The XPH1 and XPH2 structures show global TM‐align scores of ~0.7 with their respective parent sequences, Xfaso 1 and Pfl 6, indicative of retention of a similar global fold; by contrast, XPH1 and XPH2 have much lower scores (~0.5) with each other and with the noncognate parent sequences, similar to the score between wild‐type Xfaso 1 and Pfl 6 (Table 2). These comparisons suggest, at least at first glance, that our attempt to design a C‐terminal sequence capable of adopting or “bridging” both Cro folds was successful.

Table 2.

Pairwise structural comparisons of Xfaso 1, Pfl 6 and chameleon hybrids

	Global					Chameleon region only
Pair	%IDa	RMSD, Åb	TMc	Zd	Lalignd	RMSD, Åb	%
sim%(φ,ψ) e
Xfaso1‐Pfl6	41	5.9 (5.9)	0.53	3.2	42	5.5 (6.1)	25 (25)
Pfl6‐XPH1	55	6.5 (6.6)	0.48	3.1	46	4.8 (5.4)	45 (46)
Xfaso1‐XPH2	55	6.1 (7.2)	0.50	2.4	38	5.0 (4.9)	35 (46)
XPH1‐XPH2	69	6.5 (6.8)	0.47	2.0	38	4.6 (5.6)	35 (42)
Pfl6‐XPH2	85	1.8 (4.6)	0.74	8.7	54	2.3 (4.2)	90 (79)
Xfaso1‐XPH1	85	2.7 (3.8)	0.69	7.1	60	2.7 (3.8)	60 (58)
Pfl6‐Pfl6	100	0.4 (0.4)	0.93	12.4	59	0.3 (0.3)	100 (100)
Xfaso1‐Xfaso1	100	0.3 (0.3)	0.96	15.0	63–64	0.1 (0.1)	100 (100)

^aPercent identity is calculated based on the alignment shown in Figure 2(c) but does not include the last three residues shown in that alignment, which are not part of the ordered region of both Xfaso 1 and Pfl 6 in the crystal structures.

^bGlobal RMS deviation is calculated from superposition of backbone atoms for residues 1–33 and 39–58 of Xfaso 1/XPH1 and residues 3–55 of Pfl 6/XPH2, while chameleon region RMSD is calculated from superposition of backbone atoms for residues 39–58 of Xfaso 1 and XPH1 and 36–55 of Pfl 6 and XPH2. Values in parentheses also include residues 59–62 of Xfaso 1/XPH1 and residues 56–59 of Pfl 6/XPH2. This extended region is ordered in the crystal structures of wild‐type Xfaso 1 and Pfl 6 but is variably dynamic in oxidized Xfaso 1, monomeric Pfl 6, XPH1 and XPH2.

^cTM‐score from the program TM‐Align.

^dZ and Lalign refer to Z‐score and length of aligned region from the program Dali.

^e% sim (φ,ψ) refers to the percentage of residues in the chameleon region for which the mean φ,ψ values of the two proteins are within 100 degrees linear distance on a Ramachandran plot. Mean φ,ψ values were based on the solution ensemble average for XPH1 and XPH2, or on the average of the unique chains in the asymmetric unit of the crystal structure for wild‐type Xfaso 1 and Pfl 6.

However, closer inspection reveals that the folding patterns of the chameleon region in both XPH1 and XPH2 deviate significantly from the equivalent regions in the parent Cro structures. XPH1 differs from Xfaso 1, and XPH2 from Pfl 6, in backbone RMSD for the chameleon region (Table 2; ~2.5 Å, or 4 Å if the somewhat dynamic C‐terminal region is included). Each comparison shows differences in secondary structure assignment (Fig. 5) as well as qualitative changes in backbone conformation of individual residues, particularly between Xfaso 1 and XPH1 (Table 2). Each hybrid has a global TM‐score or DALI Z‐score against its cognate parent that is distinctly lower than those observed in comparisons between different chains of Xfaso 1 or Pfl 6 (Table 2). Thus, XPH1 and XPH2 have folding patterns that differ not only from each other, but also from those of Xfaso 1 and Pfl 6, respectively.

XPH1 and XPH2 have secondary structure contents that are intermediate between Xfaso 1 and Pfl 6, suggestive of a multistep progression in both sequence and structure. In going from Xfaso 1 to XPH1 to XPH2 to Pfl 6, 15–30% of the global sequence is changed at each step, with each pair in the progression showing at least ~70% global identity. Meanwhile, at the level of secondary structure, the number of helical residues declines gradually, and the number of strand residues increases gradually [Fig. 5(a)]. This progressive loss of helical structure seen in the NMR structures is qualitatively consistent with the observed decline of dichroism signal at 222 nm [see Fig. 4(a)]. A progressive structural change is also seen in analysis of backbone conformational similarity between aligned residues (Table 2). Xfaso 1 and Pfl 6 have similar backbone conformations for only 25% of the residues in the C‐terminal region. As the overall sequence progresses from Xfaso 1 to Pfl 6, XPH1 rises to 35% conformational similarity with Pfl 6 and retains only 60% backbone conformational similarity to Xfaso 1. XPH2 then has 90% conformational similarity to Pfl 6 and only 35% conformational similarity to Xfaso 1.

However, other measures of structural similarity show no evidence for a gradual sequence‐structure progression, and instead suggest an abrupt fold change in the step between XPH1 and XPH2. In particular, RMSD, TM‐align28 score, and DALI29 Z‐score for all four pairwise comparisons between the two Xfaso 1‐based sequences and the two Pfl 6‐based sequences (Pfl6‐Xfaso1; XPH1‐XPH2; XPH1‐Pfl 6; XPH2‐Xfaso 1) show similar (and poor) numbers (Table 2). By these measures, XPH1 does not represent a structural step in the direction of Pfl 6 from Xfaso 1, nor does XPH2 represent a structural step toward Xfaso 1 from Pfl 6. The structural intermediacy of XPH1 and XPH2 is less evident at the level of tertiary structure and may not be detected by many traditional structural comparison methods.

Detailed analysis of structural changes

The XPH1 structure contains several specific, notable changes relative to wild‐type Xfaso 1 (Fig. 5). XPH1 retains most of helix α4, but the N‐terminal end changes to a structure reminiscent of a type II β‐turn. Helix α5 converts to an extended structure that participates in a parallel β‐bridge structure with the loop between helices α3 and α4 [Figs. 5, 6(a)]. Helix α6 loses long‐range contacts with the rest of the fold and has only partial or transient helical character (dynamic behavior in this region was observed for the oxidized form of wild‐type Xfaso 1, however; see Supporting Information Fig. S2).

Figure 6.

Novel features of XPH2 including a new hydrophobic interface. (A) Superposition of wild‐type Pfl 6 (PDB ID 2PIJ, chain A; gray) and XPH2 (medoid structure of ensemble; tan). Leu 55 and Val 59 of XPH2 contact nonpolar residues Val 25, Ala 29, and Met 33 in or near helix α3. The contacts made by Val 59 are less well‐defined than those made by Leu 55 due to greater apparent disorder in the ensemble near the C‐terminus (see Fig. 5). In wild‐type Pfl 6 Val 25, Ala 29, and Met 33 are part of the dimer interface and contact Ile 58′ from a second subunit (PDB ID 2PIJ, chain B; green). (b) Chemical shift perturbation analysis annotated with secondary structures for Pfl 6 I58D (gray) and XPH2 (tan). Substituted residues are shown as cyan bars and unassigned or proline residues as red bars.

The XPH2 structure is also distinct from that of its parent, Pfl 6. XPH2 has a shorter version of the β2‐β3 hairpin. Strand β3 extends to residue 55–56 in monomeric Pfl 6 (I58D) and to residue 58 in the wild‐type dimer. Strand β3 of XPH2, however, terminates at residue 52, and residues 53–54 of XPH2 convert to a type I reverse β‐turn structure. Residues 57–59, which form the terminus of strand β3 in the Pfl 6 dimer and are disordered in the monomer, form a short helix in 18 of 20 XPH2 ensemble members, although RCI‐S² values and coordinate variability in the ensemble suggest some disorder (Supporting Information Fig. 5, S6).

Interestingly, the side chains of Leu55 and Val59 in XPH2 make novel long‐range hydrophobic contacts (Fig. 6). These new tertiary packing interactions overlap spatially with interactions observed in the wild‐type dimer interface. In particular, the position occupied by Leu 55 is close to that occupied by Ile 58′ of a second Pfl 6 chain in the wild‐type Pfl 6 dimer. Thus, a hydrophobic surface that mediates dimerization in wild‐type Pfl 6 acts as a template for novel tertiary structure in the XPH2 monomer.

Determinants of the XPH1 fold pattern

We were intrigued by the loss of helix α5 and gain of β‐bridge in XPH1, given that residues 55–58 are identical (CREL) in Xfaso 1 (and XPH0) and XPH1 (see Fig. 5). This led us to question what sequence changes cause the Xfaso 1 structure to switch to the XPH1 fold. Using extensive mutagenesis, we determined that only two of the six C‐terminal substitutions in XPH1, E40G, and V48Y, were necessary to trigger the switch (along with formation of the C42‐C55 disulfide). These substitutions are sufficient to induce the chemical shift changes seen in XPH1 (Fig. 7) as well as loss of helical signal (Supporting Information Fig. S8). The full loss of helicity associated with XPH1 does not occur in the reduced form of XPH0‐E40G/V48Y, suggesting that the disulfide is necessary for unwinding helix α5 (Supporting Information Figs. S8, S9). At the same time, disulfide formation alone is not sufficient, since oxidized wild‐type Xfaso 1 retains helix α5 (see Supporting Information Figs. S2, S3). The E40G and V48Y substitutions are both within helix α4, suggesting that perturbations of this helix trigger the broader changes in C‐terminal subdomain structure. E40G promotes conversion of the N‐terminal end of α4 to an approximate type II β‐turn. The V48Y substitution alters a hydrophobic packing interface and may shift the angle of helix α4 (Fig. 7). These changes may propagate to helix α5‐α6 via the C42‐C55 disulfide bond.

Figure 7.

Novel features and sequence determinants of the XPH1 folding pattern. (a) Superposition of Xfaso 1 (PDB ID 3BD1, chain C; gray) and XPH1 (medoid structure of solution ensemble; tan). An E40G substitution promotes conversion of the N‐terminal end of helix α4 to an approximate type‐II β‐turn, while a V48Y substitution pushes helix α4 away from helix α2. These changes, combined with disulfide bond formation, lead to unwinding of helix α5 and formation of a β‐bridge. (b) Chemical shift perturbation analysis annotated with secondary structures for Xfaso1/XPH0 (gray) and XPH1 (tan). XPH1 has six hybrid substitutions (cyan bars), but only two (E40G/V48Y) are necessary to give the same qualitative perturbations. Unassigned or proline resonances are shown as short red bars.

Discussion

We designed a three‐step progression of sequence and structure between all‐α (Xfaso 1) and an α + β (Pfl 6) Cro protein. The progression illustrates a developing mechanism for Cro fold evolution in which the C‐terminal region becomes structurally ambivalent through substitution mutations, but fully switches from α‐helix to β‐sheet only when triggered by other sequence changes such as indels.15, 16, 17, 18 More importantly, the current work introduces the possibility that sequence evolution could induce gradual and not just abrupt changes in structure. Specifically, we observe a stepwise transition in secondary structure content that expands our sense of the plasticity of protein structures toward sequence change and posttranslational modification.30 Metamorphic evolution of Cro or other proteins could, to borrow yet another geological term, be polymetamorphic 31 in mechanism, moving through multiple structural phases or intermediates.

Protein structure space has a geometrically semicontinuous nature32 that may facilitate stepwise transitions between folds. Almost any two protein folds can be interrelated by a series of limited geometric changes involving other known structures as intermediates.33 Undiscovered but geometrically plausible structures34 could provide further connections, especially since many may be similar to existing folds.35 Xfaso1, XPH1, XPH2, and Pfl 6 are an experimental illustration of multiple small structural steps that morph one fold into another (Supporting Information Movie 1). This progression is gradual at the level of secondary structure, but an abrupt change in tertiary structure still occurs between XPH1 and XPH2. The whole transformation changes about half the structure of the Cro domain, yielding a distinct fold classification. P22 Cro and λ Cro, two distantly related Cro proteins with all‐α versus α + β folds, respectively, belong to different classes (C), architectures (A) and topologies (T) within the hierarchy of the CATH classification system.36 Hypothetically, one could envision further steps that morph the helix‐turn‐helix subdomain and erase all traces of structural similarity to the original all‐α fold.

Folds are also connected in sequence space to an extent that may facilitate evolutionary flow between them, whether by stepwise or concerted mechanisms. There are now many demonstrations of identical or highly similar sequences with two different stable folds, ranging from changes in single secondary structure elements to global fold switches.13, 37, 38, 39, 40, 41, 42, 43 Computational studies have found sequence supernetworks that bridge large numbers of folds directly or indirectly.44, 45 A recent design study suggests that proteins with a 3‐α‐helix bundle fold can be bridged to two other folds.46 In at least two cases a single sequence has been shown to adopt three topologies, acting as a three‐way bridge.31, 47 This considerable flow between folds in sequence space can only be increased by the accessibility of intermediate structural forms.

Of course, evolutionary constraints of function can restrict this flow by making some sequences and structures inaccessible. Both XPH1 and XPH2 have features, like a disulfide bond in XPH1 and an occluded dimer interface in XPH2, which may hinder Cro function. Thus, a caveat to our study is that it only illustrates the plausibility of structural intermediates from the perspective of maintaining minimal stability. It will be interesting to see if intermediate forms are adopted by any natural, functional Cro sequences, and if designed Cro intermediate forms can carry DNA‐binding function.

Materials and Methods

Mutagenesis, protein expression, and purification

Genes for the wild‐type proteins were previously cloned into pET21b vectors.7 Plasmids encoding Xfaso1 and Pfl6 variants were generated using the QuikChange mutagenesis method (Stratagene) and/or by cassette mutagenesis. Some Xfaso 1 variants studied here, including XPH1, lacked residues 66–78, which are not ordered in the crystal structure7 and can be removed without affecting stability or structure, as judged by comparisons of CD spectra, thermal denaturation curves, and 15N‐¹H HSQC spectra (both oxidized and reduced forms). All variants used had a LEHHHHHH C‐terminal purification tag and were overexpressed in Escherichia coli BL21(λDE3). LB was used as the standard growth medium for unlabeled proteins, while M9T minimal media were used for 15N and ¹³C uniform labeling using 0.8 g/L [U‐ ¹⁵ N]‐NH₄Cl and 2.5 g/L [U‐ ¹³ C ₆]‐D‐glucose as sole nitrogen and carbon sources, respectively.

All variants were purified by Ni‐NTA chromatography,⁴⁸ followed by dialysis of the eluate into SB250 refolding buffer [50 mM Tris (pH 7.5), 250 mM KCl, 0.2 mM EDTA] with or without 1 mM DTT. For variants to be characterized in reduced form, dialysates were centrifuged at 12,000g for 30 min to remove precipitates and loaded onto Sephacryl S‐100 16/60 size‐exclusion column for additional purification. Purified samples were concentrated using Amicon Ultracel‐3 K centrifugal filters (Merck Millipore Ltd, Tullagreen, Carrigtwohill, County Cork, Ireland) (3000 Da molecular weight cutoff, 5000g). Protein concentrations were determined from A₂₈₀ measurements using the following estimates for ε₂₈₀: 5559 cm⁻¹ for variants with one Trp and no Tyr residues (Xfaso1 and XPH0) and 6756 cm⁻¹ for variants with one Trp and one Tyr residue (XPH1, Xfaso1‐E40G/V48Y, XPH0‐E40G/V48Y).7 For Pfl6 variants containing no Trp and two Tyr residues (Pfl6‐I58D and XPH2), an ε₂₈₀ value of 3119 cm⁻¹ was used, measured for Pfl6 using the method of Edelhoch.49

For oxidation of Xfaso 1 variants containing the helix‐turn‐helix mutations, affinity‐purified, refolded proteins were first dialyzed into oxidation buffer [50 mM ammonium bicarbonate (pH 7.8), 150 mM KCl] and diluted to a protein concentration of 25 μM. Sodium azide (0.01%) was added to prevent microbial growth, followed by incubation of the protein solutions in small beakers for 3 days under laboratory atmosphere at ambient temperature. Oxidized proteins were then dialyzed back into SB250, centrifuged at 12,000 g for 30 min to remove precipitates, and loaded onto Sephacryl S‐100 16/60 size‐exclusion column for additional purification. Other Xfaso 1 variant samples readily oxidized in the NMR tube containing phosphate buffer (50 mM potassium phosphate [pH 6.5], 150 mM KCl) over a few days. Successful air oxidation of proteins was verified using ¹H NMR spectra, by monitoring shifts in the Hε1 resonance of the single Trp residue in every Xfaso 1 variant. Oxidized NMR samples could be rapidly reduced with the addition of 3 mM tris(2‐carboxyethyl)phosphine (TCEP), which achieved total reduction in minutes. Other reduced protein samples were prepared using 5 mM BME (β‐mercaptoethanol) during lysis and affinity purification, then dialyzed into SB250 with 3 mM DTT (dithiothreitol), maintaining the reducing agent in all subsequent purification steps.

Circular dichroism spectroscopy

Far‐ultraviolet CD spectra and thermal denaturation curves were obtained on an Olis DSM‐20 CD spectrometer (Olis, Inc, Bogart, GA, USA). Wavelength scans were performed using 25 μM protein samples in SB250 with or without 1 mM DTT in 1 mm pathlength cells at 20°C. Points were averaged from triplicate scans from 260 nm to 205 nm at 1 nm intervals with an integration time of 20–40 s, after baseline subtraction. Thermal denaturation curves were obtained using 25 μM protein samples in 1–2 mm pathlength cells at 222 nm wavelength. Samples were typically heated from 20°C to 90°C (with slighly different ranges used, depending on the variant) in 1–2°C intervals with a 2 min equilibration period per point and 25–55 s integration time. Baseline buffer (SB250) curves were subtracted from each denaturation curve. T _m values were obtained by fitting the baseline‐adjusted denaturation curves to the following relationship50:

$$\mathrm{\Delta }{G}_{\mathrm{u}}=\mathrm{\Delta }{H}_{\mathrm{u}}\left(1-T/T_{\mathrm{m}}\right)+\mathrm{\Delta }{C}_p\left[T-T_{\mathrm{m}}-Tln\left(T/T_{\mathrm{m}}\right)\right]$$

where ΔG _u is related to the fraction of the molecules in the unfolded state based on measured ellipticity between folded and unfolded baselines. The heat capacity of unfolding (ΔC _p) was set at 840 cal mol⁻¹ K⁻¹ based on an estimate for a 60‐residue folded region at 14 cal mol⁻¹ K⁻¹ per residue.51 All other parameters were allowed to vary during fitting.

NMR spectroscopy

Purified protein samples (0.4–0.8 mM) were dialyzed into phosphate buffer [50 mM potassium phosphate (pH 6.5), 150 mM KCl], and NMR samples were formulated by addition of 10% D₂O, 0.01% sodium azide, and 0.1 mM DSS. Trimethylamine N‐oxide (TMAO‐d ₉), a stabilizing osmolyte,52 was also added to XPH2 samples at a concentration of 1 M to improve spectral quality. Following addition of TMAO, the sample pH was readjusted to pH 6.5. All NMR spectra were acquired at 20°C on a Varian Inova‐600 spectrometer equipped with a triple‐resonance cryogenic probe.

NMR spectra were processed using NMRPipe53 and analyzed with Sparky (http://www.cgl.ucsf.edu/home/sparky). Chemical shifts were assigned using a combination of HNCO, HNCACB, CBCA(CO)NH, ¹⁵N‐TOCSY‐HSQC, HCCH‐TOCSY, ¹³C‐HSQC, and ¹⁵N‐HSQC spectra. Methyl groups of Val and Leu residues were stereospecifically assigned using [10%‐¹³C]‐labeled proteins.54 Backbone chemical shifts (H^N, Hα, N, Cα, Cβ, C′) were used in TALOS‐N55 to predict backbone dihedral angles (ϕ,ψ), some sidechain angles (χ ₁), and backbone order parameters (S ²) derived from the random coil index (RCI). NOEs were assigned manually for XPH1 and XPH2 using ¹⁵N‐NOESY‐HSQC and ¹³C‐NOESY‐HMQC spectra. Due to limited solubility and chemical shift dispersion, XPH2 had inherently poorer NOESY data than XPH1.

Chemical shift Rosetta56, 57 structure calculations were performed on the CS‐Rosetta server at the BioMagResBank (http://condor.bmrb.wisc.edu), including disulfide restraints where appropriate, to determine the global fold and aid in NOE assignment. In each case 10,000 structures were calculated and the 10 lowest energy models selected. For Pfl 6 I58D and oxidized wild‐type Xfaso 1, CS‐Rosetta calculations were used primarily to establish that monomerization and disulfide bond formation, respectively, did not greatly alter the basic fold seen in crystal structures for the wild‐type proteins. No further structure calculations were performed for these proteins.

CYANA was used to calculate solution NMR structures of XPH1 and XPH2 using the manual method.58 TALOS‐N backbone angle predictions were converted to ϕ,ψ angle restraints. Side chain angle restraints were used when TALOS‐N had a strong prediction, with ± 40° allowed for each χ₁ angle restraint. NOE peak heights were converted to upper distance restraints manually using the ‘peaks calibrate’ routine in CYANA with “sum of r^‐6” averaging. Average restraint distance ‘dref’ was loosened and iteratively tightened with each round of CYANA calculation until RMSD decreased without an increase in target function due to violations. The parameter ‘dref’ was optimized separately for each NOESY dataset, allowing us to detect incorrectly assigned NOEs or to resolve ambiguous ones. Trivial NOE‐derived distance restraints, including duplicates or those that cannot be violated, were removed by CYANA prior to structure calculation. Helical regions predicted by TALOS‐N included i,i+4 hydrogen bond distance restraints of 2.5 and 3.5 Å for H—O and N—O pairs, respectively. Final CYANA structure calculations were performed using the restraints described in Table 1, starting with 100 random conformers, utilizing 20,000 minimization steps followed by 80,000 torsion angle dynamics steps, with 20 lowest energy conformers saved for each ensemble. Ensembles generated by CYANA were then refined in explicit solvent in CNS59 using the WaterRefCNS script to improve Ramachandran statistics.60 Structures were compared using DALI29 and TM‐align.28 RMS deviations were calculated using MOLMOL.61 Structures of XPH1 and XPH2 have been deposited in the Protein Databank under entries 5W8Y and 5W8Z.

Supporting information

Supplementary Figure S1. NMR analysis of Pfl 6 I58D monomer with (A) TALOS, including RCI‐S² and annotated with wild‐type Pfl 6 secondary structures, and (B) CS‐Rosetta (10 lowest energy structures), with the ensemble superimposed on chain A of the Pfl6 dimer crystal structure (gold), with chain B shown in green. Secondary structure elements are labeled. Residues 56–58 form the end of strand β3 in the wild‐type dimer but appear somewhat more flexible in the TALOS/RCI analysis and the CS‐Rosetta ensemble.

Supplementary Figure S2. NMR analysis of oxidized Xfaso 1. (A) Comparison of 15 N‐1H HSQC spectra of reduced and oxidized Xfaso 1. (B) Chemical shift perturbation analysis, with secondary structure element locations in reduced Xfaso 1 shown in red. (C) Difference in backbone conformations predicted by TALOS for oxidized Xfaso 1 compared to those observed in the crystal structure of reduced Xfaso 1. Deviations represent two‐dimensional distance in degrees on a Ramachandran plot.

Supplementary Figure S3. CS‐Rosetta analysis of oxidized Xfaso 1 and comparison to reduced Xfaso 1 crystal structure. (A) CS‐Rosetta ensemble (lowest energy 10 structures) with (B.C) two conformer families shown for helices α5 and α6. (D) Comparison of secondary Hα chemical shifts for the α5/α6 region of reduced (black) and oxidized (orange) Xfaso 1, suggesting retention of generally similar backbone conformation. (E,F) Comparisons of lowest energy oxidized conformation from the conformer family in panel B (E; light grey with α5/α6 region in cyan) or panel C (F; light grey with green) against the reduced Xfaso 1 crystal structure (dark grey with blue; PDB ID 3BD1, chain C).

Supplementary Figure S4. Comparison of XPH0 hybrid and wild‐type Xfaso 1 using NMR. (A) Overlay of HSQC spectra for oxidized forms. (B, C) Chemical shift perturbation analysis of reduced and oxidized forms.

Supplementary Figure S5. Quality of HSQC spectra for XPH1 (A) and XPH2 (B) are improved by oxidation (XPH1) or addition of TMAO (XPH2).

Supplementary Figure S6. TALOS and RCI‐S² analysis of (A) oxidized XPH1, annotated with secondary structures for the Xfaso 1/XPH0 parent sequence (grey) and for XPH1 itself (tan), as derived from the full structures, (B) reduced XPH2, similarly annotated with secondary structures for the Pfl 6 I58D parent (grey) and for XPH2 itself (tan).

Supplementary Figure S7. Chemical shift perturbation analysis of the chameleon sequence common to XPH1 and XPH2. Two comparisons are missing from the analysis (red) due to lack of resonance assignment in XPH1 or XPH2 or because the residue is proline. The plot is annotated with helical (red) or strand (blue) secondary structures of XPH1 and XPH2. The pink bar near the C terminus of XPH1 is a transient helix.

Supplementary Figure S8. Far ultraviolet circular dichroism analysis of E40G/V48Y variants in oxidized or reduced form. (A) Wavelength scans at 20°C. (B) Thermal denaturation curves with ellipticity monitored at 222 nm, fitted and converted to fraction unfolded.

Supplementary Figure S9. Chemical shift perturbation analysis of reduced E40G/V48Y variants. (A) In the reduced form, the E40G/V48Y mutations are insufficient to cause a major structural shift. (B) Disulfide bond formation causes numerous large changes in chemical shift. Unassigned or proline resonances are shown as short red bars. In panel A, substituted residues are shown as cyan bars.

Supplementary Movie S1. Conformational morph of Xfaso 1 to XPH1 to XPH2 to Pfl 6, and back again, generated with Chimera. The backbone is shown as plain ribbon with rainbow coloring from N (blue) to C (red) terminus. There is a gap at the junction of the two subdomains due to a deletion in Pfl 6 relative to Xfaso 1.