A Calcium-gated Lid and a Large β-Roll Sandwich Are Revealed by the Crystal Structure of Extracellular Lipase from Serratia marcescens*

Lipase LipA from Serratia marcescens is a 613-amino acid enzyme belonging to family I.3 of lipolytic enzymes that has an important biotechnological application in the production of a chiral precursor for the coronary vasodilator diltiazem. Like other family I.3 lipases, LipA is secreted by Gram-negative bacteria via a type I secretion system and possesses 13 copies of a calcium binding tandem repeat motif, GGXGXDXUX (U, hydrophobic amino acids), in the C-terminal part of the polypeptide chain. The 1.8-Å crystal structure of LipA reveals a close relation to eukaryotic lipases, whereas family I.1 and I.2 enzymes appear to be more distantly related. Interestingly, the structure shows for the N-terminal lipase domain a variation on the canonical α/β hydrolase fold in an open conformation, where the putative lid helix is anchored by a Ca2+ ion essential for activity. Another novel feature observed in this lipase structure is the presence of a helical hairpin additional to the putative lid helix that exposes a hydrophobic surface to the aqueous medium and might function as an additional lid. The tandem repeats form two separated parallel β-roll domains that pack tightly against each other. Variations of the consensus sequence of the tandem repeats within the second β-roll result in an asymmetric Ca2+ binding on only one side of the roll. The analysis of the properties of the β-roll domains suggests an intramolecular chaperone function.

Lipase LipA from Serratia marcescens is a 613-amino acid enzyme belonging to family I.3 of lipolytic enzymes that has an important biotechnological application in the production of a chiral precursor for the coronary vasodilator diltiazem. Like other family I.3 lipases, LipA is secreted by Gram-negative bacteria via a type I secretion system and possesses 13 copies of a calcium binding tandem repeat motif, GGXGXDXUX (U, hydrophobic amino acids), in the C-terminal part of the polypeptide chain. The 1.8-Å crystal structure of LipA reveals a close relation to eukaryotic lipases, whereas family I.1 and I.2 enzymes appear to be more distantly related. Interestingly, the structure shows for the N-terminal lipase domain a variation on the canonical ␣/␤ hydrolase fold in an open conformation, where the putative lid helix is anchored by a Ca 2؉ ion essential for activity. Another novel feature observed in this lipase structure is the presence of a helical hairpin additional to the putative lid helix that exposes a hydrophobic surface to the aqueous medium and might function as an additional lid. The tandem repeats form two separated parallel ␤-roll domains that pack tightly against each other. Variations of the consensus sequence of the tandem repeats within the second ␤-roll result in an asymmetric Ca 2؉ binding on only one side of the roll. The analysis of the properties of the ␤-roll domains suggests an intramolecular chaperone function.
Lipases (EC 3.1.1.3) hydrolyze the ester bonds of long-chain acylglycerides (1). Biotechnologically, they have gained considerable importance as biocatalysts that are able to catalyze not only hydrolysis but also synthesis reactions, with the latter occurring in nearly water-free organic solvents (see Refs. 2 and 3 for detailed reviews) and often with high regio-and enantioselectivity (4). Virtually all of the technologically employed lipases are of microbial origin, because the respective genes are relatively easy to access and can efficiently be expressed. Consequently, it is a lipase produced by the Gram-negative bacterium Pseudomonas aeruginosa that currently constitutes the best studied example of creating an enantioselective enzyme by directed evolution (5)(6)(7). A characteristic feature to distinguish lipases from esterases is called interfacial activation and describes the observation that lipase activity sharply increases as soon as the monomolecular substrate starts to form a micellar emulsion (8). An obvious explanation was provided by several lipase crystal structures that revealed the presence of a surface-exposed ␣-helical polypeptide chain termed the "lid" which covers the active site and moves away upon contact with the micellar substrate interface (9,10). However, exceptions have also been observed of lipases that possess a lid and still do not show interfacial activation. Therefore, lipases are presently defined as carboxylesterases catalyzing the hydrolysis of long-chain (Ͼ10 carbon atoms) acylglycerols (11).
A classification of bacterial lipases was first put forward by Arpigny and Jaeger (12), who defined eight families (I-VIII) based on amino acid sequences and biological properties. The largest family I is further subdivided in seven subfamilies (I.1-I.7) with the first three subfamilies comprising true lipases from Gram-negative bacteria. All lipases are members of the ␣/␤ hydrolase superfamily, where the canonical fold comprises a mostly parallel, central eightstranded ␤-sheet. The active site carries the Ser-His-Asp/Glu catalytic triad, where the serine residues are usually located within a characteristic GXSXG motif, which forms a sharp ␥-like turn between a ␤-strand and the following ␣-helix (13).
Although lipases of families I.1 and I.2 are clearly homologous with amino acid sequence identities above 30%, the family I.3 enzymes exhibit only a very low sequence similarity to the former two families. Also, they are translocated to the extracellular medium by a different mechanism. Whereas family I.1 and I.2 lipases are secreted by a type II secretion system (T2SS, 3 also named general secretory pathway) (14 -16), the family I.3 lipases are transported by a type I secretion system (T1SS) (16 -18). This secretion pathway utilizes a C-terminal, non-cleavable secretion signal, and translocation occurs in one step from the cytoplasm to the extracellular medium without the occurrence of periplasmic intermediates. The Escherichia coli hemolysin secretion system Hly constitutes the paradigm of a T1SS (19), which like the other T1SSs consists of three membrane-associated proteins; that is, an inner membrane ABC transporter (HlyB), an outer membrane protein (TolC), and a periplasmic membrane fusion protein (HlyD). T1SS passenger proteins possess as a characteristic feature a variable number of tandem repeats of a glycine-rich nine-residue motif (GGXGX-DX(U)X) n (U, apolar residue) that precedes the C-terminal secretion signal. The number n of the repeats correlates positively with the molecular weight of the protein. This so-called RTX signature (repeats in toxins) (20) is responsible for Ca 2ϩ binding. X-ray crystallographic analyses of serralysin-type metalloproteases showed that these repeats fold into a parallel ␤-roll, where the first six amino acids build a turn that forms two half-sites for Ca 2ϩ binding, and the remaining three residues fold into a short ␤-strand (21)(22)(23). The calcium ions bridge the spatially adjacent aspartic acids (position 6 in the motif) from two sequence motifs 18 residues apart. The nonpolar amino acids (abbreviated U) at position 8 build the hydrophobic core of the ␤-roll. This peculiar structure is unstable in the absence of Ca 2ϩ but does fold spontaneously in the presence of calcium concentrations in the mM range (24,25).
Serratia marcescens SM6 produces an extracellular lipase LipA that has been a subject of research for more than 30 years. Its production was found to be stimulated up to 100-fold when the standard growth medium was supplemented with various polysaccharides like glycogen or hyaluronate (26). Later it was shown that this enzyme is a family I.3 lipase of 613 amino acids in length (27). The catalytic residue Ser 207 is located in a conserved GHSLGG motif. From inspection of the sequence, S. marcescens LipA possesses about 12-14 repeats of the glycine-rich sequence motif that come in two blocks spanning residues 369 -418 and 489 -564. Furthermore, S. marcescens LipA is a biotechnologically very important enzyme as it serves for the large-scale kinetic resolution of racemic 3-(4Ј-methoxyphenyl)glycidic acid methyl ester. LipA catalyzes the enantioselective hydrolysis of the (2S,3R) enantiomer, which is then easily separated from the desired (2S,3R) enantiomer (28). This biocatalytic reaction reduces by 4 steps the original 9-step chemical synthesis to yield diltiazem, a calcium-channel blocker and coronary vasodilator produced worldwide in an excess of 100 t a Ϫ1 (29).
Presently, crystal structures have not been reported for S. marcescens LipA nor for any other lipase of family I.3 despite ongoing work (30). Furthermore, homology modeling would be unreliable because the lipase domain of LipA shows only a very limited sequence identity (Ͻ20%) to other known lipase structures. Therefore, we have determined the high resolution crystal structure of LipA from S. marcescens.

EXPERIMENTAL PROCEDURES
Protein Expression and Purification-The LipA expression construct was made by cloning the PCR product derived from S. marcescens genomic library into the NdeI and HindIII cloning sites of the pET28a vector (Novagen) using the primers 5Ј-GGAATTC-CATATGGGCATCTTTAGCTATAAGG-3Ј and 5Ј-CCC-AAGCTTTTAGGCCAACACCACCTGATCGG-3Ј. The respective restriction sites within the primer sequences are underlined.
LipA was expressed in E. coli BL21 (DE3) cells (Stratagene). Cells were grown at 37°C to an A 600 of 1.2 and then induced with 1 mM isopropylthiogalactopyranoside. 5 h after induction, cells were harvested and resuspended in 20 mM Tris-HCl, pH 8.0, 300 mM NaCl and disrupted using a French press. Inclusion bodies were isolated from the crude cell lysate by centrifugation and washed twice with 100 mM Tris-HCl, pH 7.5, 100 mM EDTA, 10% (w/v) Triton X-100. Purified inclusion bodies were stirred for 40 min in 100 mM NaH 2 PO 4 , 10 mM Tris, pH 8.0, 8 M urea. The solution of solubilized inclusion bodies was clarified by centrifugation at 10,000 ϫ g for 15 min at room temperature. 20 ml of solubilized LipA inclusion bodies was then diluted into 180 ml of 250 mM Tris, pH 8.0, 50 mM CaCl 2 , 800 mM L-arginine, 8% glycerol. Refolded protein solution was concentrated in an Amicon ultrafiltration cell (Millipore) to 80 ml. A gel filtration on a Superdex 75 column (Amersham Biosciences) in GPC buffer (20 mM Tris-HCl, pH 8.0, 100 mM NaCl, 20 mM CaCl 2 , 0.05% (w/v) Triton X-100) was performed with 5 ml of the concentrated refolded LipA. One major peak representing the LipA fraction was pooled and concentrated to 18 mg/ml and ϳ0.2% (w/v) Triton X-100 (about 3 mM).
Activity Assays-LipA was dialyzed overnight at 4°C against buffer 20 mM Tris, 100 mM NaCl, 0.05% Triton X-100, pH 8.0, to remove excess of calcium deriving from the gel filtration buffer. After dialysis, LipA was incubated with 1, 2, and 5 mM EDTA for 1 h at room temperature. The enzymatic activity of the hydrolysis of p-nitrophenyl palmitate was determined at a concentration of 0.1 M in 1000 l of 25 mM Tris-HCl, pH 8.0, 10% acetonitrile, and 50 M p-nitrophenyl palmitate at room temperature. Protein concentrations were determined by a bicinchoninic acid assay (Pierce). The absorption of the p-nitrophenolate produced by the reaction was determined at 412 nm using an Ultrospec 2100 pro (Amersham Biosciences) spectrophotometer.
Crystallization and Data Collection-Crystals belonging to space group P321 with one molecule per asymmetric unit were obtained at 18°C using a hanging drop vapor diffusion method. Drops were set up by mixing 3-6 l of protein solution (18 mg/ml) with 1 l of reservoir solution (0.1 M sodium acetate, pH 4.6, 22-28% (v/v) 2-methyl-pentane-2,4-diol) and equilibrated against 500 l of reservoir solution at 18°C. Hexagonal-shaped crystals were observed after 2 days and grew to an average size of 100 ϫ 100 ϫ 25 m. These crystals belong to space group P321 with cell parameters of a ϭ 115.1 Å and c ϭ 104.6 Å. Heavy atom derivative crystals were prepared by soaking crystals in the mother solution described above containing additional 5 mM Pb(II) acetate for 30 and 1 h. After soaking, crystals were back-soaked to mother liquor containing 30% (v/v) 2-methyl-pentane-2,4-diol.
A second crystal form appeared in the same drops. These crystals belong to the space group R3 with cell constants in the hexagonal setting of a ϭ 202.4 Å and c ϭ 317.7 Å. This crystal form contains six molecules in the asymmetric unit. "Interme-diates" between the two crystal forms, i.e. twinned crystals, were observed very frequently.
Before data collection, crystals were flash-cooled in a nitrogen stream at 110 K after raising the 2-methyl-pentane-2,4-diol concentration of the crystallization solution to 30% (v/v). Data were collected at the Swiss Light Source (Paul-Scherrer Institute, Villigen, Switzerland) using beamlines X10SA (PXII) and X06SA (PX) at 100 K employing a MAR225 CCD detector (MAR x-ray-research, Hamburg, Germany). The crystal to detector distance was 240 mm for native data and varied from 350 to 400 mm for derivatives. Oscillation angles were 0.5°per frame, and exposure times ranged from 0.5 to 1 s. Data were integrated and scaled with XDS (32) or MOSFLM (33) ( Table 1).
Structure Solution and Refinement-The structure was solved by SAD at the LI edge using Pb 2ϩ derivatized P321 crystals. 6 heavy atom positions were found using the program SHELXD (34). Heavy occupancies were initially low, and because Pb 2ϩ soaking damaged the crystals after some time, two very different soak times were employed to get the optimal trade-off for occupancy and diffraction. Phases were computed with SHARP (35) employing both derivatives, making use of the different occupancies. This resulted in a figure-of-merit of 0.47 and 0.14 for acentric and centric reflections, respectively. Phases were extended by solvent-flipping using SOLOMON (36), and automated model building was done by ArpWarp (37). Refinement was carried out by using REFMAC (38), and the model was completed and corrected using COOT (39). The rhombohedral structure was solved by molecular replacement employing the program PHASER (40,41). Non-crystallographic symmetry restraints in the rhombohedral crystal form led to an increase in R free and were, therefore, not applied. Most figures were prepared using PyMol (Delano Scientific LLC, Palo Alto, CA).

Overall Structure
The structure of LipA was determined and refined in two crystal forms at 2.0 and 1.8 Å of resolution ( Table 1). The two crystal forms grew under the same conditions. Virtually all amino acids are defined in the electron density map in both crystal forms. The seven crystallographically independent molecules totally present in the two crystal forms showed no significant deviations. The root mean square pairwise difference between the structures is 0.366 and 0.451 Å for backbone atoms and all protein atoms, respectively, whereas the root mean square difference from the mean structure is 0.240 and 0.295 Å for backbone atoms and all protein atoms, respectively.
LipA forms an ellipsoid-shaped molecule with overall dimensions of ϳ80 ϫ 45 ϫ 30 Å (Fig. 1); 8 Ca 2ϩ ions were identified per molecule. The N-terminal domain comprising amino acids 1-320 (supplemental Fig. S1) shows a modified ␣/␤ hydrolase fold (supplemental Fig. S2). Compared with the canonical fold, strand ␤1 is split, ␤3 and ␣A are missing, and ␣D has been replaced by a loop (supplemental Fig. S2). Furthermore, the eight ␤ strands are now split into a six-stranded mixed and a two-stranded parallel sheet.
A DALI search (42) was performed, revealing significant similarity of S. marcescens LipA to the fungal Thermomyces lanuginosa lipase (Z-score 15), whereas family I.1 and I.2 lipases proved to be much more distantly related (Z-score 3.8 for P. aeruginosa lipase). Hence, S. marcescens LipA resembles more closely eukaryotic than prokaryotic lipases (see supplemental Table S1).
Comparison to the most similar lipase structure from T. lanuginosa (PDB entries 1EIN and 1DT3) (43) revealed a number of large insertions in LipA, most notably amino acids 33-74, which form an ␣-helix hairpin, i.e. two antiparallel helices connected by a short loop. This element is located at one side of the active site cleft (Figs. 2 and 3A). At residue 343 the polypeptide chain leaves the lipase domain and enters the first ␤-roll domain at amino acid 369. After 2.5 turns of the ␤-roll, at residue 417, the polypeptide chain winds through an antiparallel ␤-structure which extends the first ␤-roll domain. Eventually, the second ␤-roll domain starts at residue 489. This domain is displaced laterally from the major long axis of the molecule and packs to the first ␤-roll as well as to the antiparallel ␤ structure succeeding it. The interface is hydrophobic, consisting mainly of leucines and aromatic residues. After 4.5 turns the polypeptide chain extends into an antiparallel ␤ structure enlarging the second ␤-roll domain and placing the C terminus in the neighborhood of the N terminus. This last bit of the structure is very similar to the C-terminal segments of the structurally characterized ␤-roll domain proteins, namely the serralysins.

Active Site
Catalytic Triad-The active site of lipases is defined by a canonical catalytic triad, which in LipA consists of Ser 207 , Asp 256 , and His 314 . These residues superimpose quite well with the corresponding amino acids from other lipases, e.g. for T. lanuginosa or P. aeruginosa the average distance difference is 0.5 and 1.5 Å, respectively.
Lid  Fig. S1) as lid helix. This lid, an ␣-helical surface-exposed segment, which presents a very hydrophobic face toward the active site cleft, is characteristic for many lipases and explains interfacial activation; it is assumed that this hydrophobic patch interacts with a micellar substrate surface (Fig. 3A) and opens the active site. Interestingly, in LipA the putative lid helix is anchored in its position by a Ca 2ϩ ion (termed Ca1), which is coordinated by aspartic acids 153 (monodentate) and 157 (bidentate), the carbonyl oxygens of Thr 118 and Ser 144 , and the side chain of Gln 120 (Fig. 3B). The coordination sphere is completed by a water molecule. This binding site is strongly conserved in lipases of family I.3 but absent in the structurally similar T. lanuginosa lipase. The calcium ion is deeply buried and at the opposite site of the active site cleft compared with the Ca 2ϩ ions found in lipases from P. aeruginosa (PDB entry 1EX9) (44) and Pseudomonas cepacia (PDB entry 1OIL) (45). For the homologous Pseudomonas sp. MIS38 lipase it was shown that calcium is essential for activity, since dialysis against calcium-free buffer reduced the activity by about 75% and dialysis against EDTA abolishes activity completely (46). Therefore, we have determined the influence of EDTA on the activity of S. marcescens LipA (supplemental Fig. S3), showing that calcium is strictly required. Mutation of Asp 157 to alanine abolished activity completely even in the presence of 20 mM Ca 2ϩ (data not shown). Another striking feature of S. marcescens lipase is the helix hairpin mentioned before, comprising amino acids 33-74 (Figs. 2 and 3). This element presents a conspicuous hydrophobic surface to the aqueous medium and may well act as a second lid.

The Parallel ␤-Roll Domains
The first ␤-roll domain spans residues 368 -418 and contains five GGXGXDXUX motifs. These conserved sequences  are somewhat degenerated, in particular at the beginning and at the end of the motif, like 374 GSDGNDLIK and 410 GGKGH-NIFD, where the second glycine and the aspartic acid of the consensus sequence are mutated, respectively. Consequently, there are only three calcium ions bound to this segment of the polypeptide chain because the sixth nine-residue motif is missing, and hence, the calcium binding site is incomplete.
The second ␤-roll domain spans residues 489 -564 and possesses eight nine-residue motifs. Interestingly, every other of these motifs is mutated in such a way that the aspartic acid at position 6 of the nine-residue motif is missing. As a consequence, only one side of this ␤-roll binds Ca 2ϩ ions ( Fig. 1 and  Fig. 4). In the lower part, two water molecules are found. Insertions occur here inside and between the nine-residue motifs. For example, the insertion of an alanine 517 GHAGGNLTFV occurs in the fourth motif, which places the leucine at the position 6, where normally the canonical aspartate resides. Two residues are inserted between the sixth and seventh motif 536 GVGNGNTFLFSGDFGRDQLY. These observations indicate that repetitive amino acid sequence motifs may appear interrupted in the tertiary structure of the respective protein making it difficult to determine by counting the exact number of repeats.

DISCUSSION
We have determined for the first time the atomic structure of a lipase belonging to family I.3. As expected, this enzyme consists of an N-terminal lipase domain and a C-terminal RTX domain involved in secretion of this enzyme. Additionally, our high resolution structure revealed several unusual and interesting features that deserve a more detailed discussion.
A New Type of Ca 2ϩ -dependent Lipase-S. marcescens LipA possess an N-terminal ␣/␤ hydrolase domain with some peculiar features. It is a calcium-dependent lipase for which interfacial activation has not been demonstrated yet. In the structure presented here, this lipase exhibits the open conformation that may have resulted from the presence of detergent in the crystallization medium, although the electron density map does not indicate the presence of detergent molecules. As a novel feature within microbial lipases, we have identified a Ca 2ϩ ion, which binds the putative lid helix and is, therefore, crucial for enzymatic activity. Consequently, treatment with EDTA resulted in inactivation of LipA. Besides the Ca 2ϩ ion discussed above, seven additional calcium ions were identified; six bind to the GGXGXDXUX motifs and a further one (designated Ca2 in Figs. 2 and 3) is coordinated by the side chains of Glu 254 (monodentate), Asp 276 (bidentate), Asn 285 , the carbonyl O of Asn 284 , and two water molecules. This site is also strongly conserved; however, it is located remote from the active site, and its function remains elusive. Calcium ions bound within ␤-roll domains are usually too tightly bound to be removed by dialysis or even by chelators (24,47). Hence, Ca1 (Fig. 2) must be the essential Ca 2ϩ ion, which is bound to the lid helix, and its removal will result in a permanent closure or even the complete distortion of the active site cleft. The inactivity of the D157A mutant strongly corroborates this hypothesis.
The Parallel ␤-Roll as Intramolecular Chaperone-The structure of S. marcescens lipase revealed a novel and unique feature that we have termed "␤-roll sandwich" consisting of tandem repeats split into two groups and separated by a 70-residue spacer. The first ␤-roll domain is located at a position equivalent to the serralysin-type metalloproteases, whereas the The catalytic triad is shown as sticks with carbons in gray, nitrogens in blue, and oxygens in red. The orientation is similar to the one in Fig. 2. B, calcium coordination by amino acids from the lid helix. The lid helix is shown in cyan. second one, which is unique for S. marcescens lipase, packs from the side tightly against the first domain and other parts of the molecule. The tandem repeats, which separate the C-terminal secretion signal from the catalytic part of the molecule, consist mainly of highly hydrophilic amino acids, thus making them soluble even in the absence of calcium, where they adopt a random coil structure (24).
This has implications for the translocation mechanism and folding of the passenger protein; contrary to T2SS, secretion by T1SS is independent of chaperones, with the exception of HasA, the Serratia marcescens heme binding protein (48,49), which is also the only one not to possess a tandem repeat domain. HasA secretion is SecB-dependent. This cytoplasmic chaperone possibly prevents spontaneous folding of HasA, and folded HasA inhibits its own secretion (50).
The absence of chaperones poses a 2-fold challenge for the passenger protein; first, it must not fold into its native structure inside the cell because this would block at least its passage through the outer membrane protein tunnel (e.g. TolC) since crystallographic studies on TolC revealed a channel of only about 30 Å in width (51). Second, because the passenger protein is translocated into the chaperone-free external medium, it must be able to fold efficiently into its native conformation. Consequently, a characteristic feature of T1SS passenger proteins is the absence of disulfide bonds, presumably because there are no disulfidebond isomerases or similar helper proteins present that could rescue "scrambled" proteins. It is tempting to speculate that the parallel ␤-roll structure may act as an intramolecular chaperone that keeps the polypeptide chain unfolded in the cytoplasm where Ca 2ϩ concentrations are about 0.1 M, although in the external medium with Ca 2ϩ concentrations in the mM range the glycine-rich repeats would fold spontaneously (24,25,47) and provide a nucleus for the folding of the entire polypeptide chain.
Such a chaperone-like function would imply that the number of ␤-roll domains positively correlates with the size of the respective passenger proteins. Furthermore, this assumption is nicely supported by our recent finding that S. marcescens LipA can be autodisplayed on the surface of E. coli cells by fusing it to an autotransporter protein from P. aeruginosa. The autotransporter domain efficiently inserts into the bacterial outer membrane with the lipase domain facing the extracellular medium, where it folds into its enzymatically active conformation (52).
Studies on the homologous Pseudomonas sp. MIS38 lipase have shown that partial or complete knock-out of the tandem repeats leads to strongly reduced protein and secretion levels. Furthermore, knock-out of all repeats resulted in a virtually inactive enzyme (53). The intracellular stability is also affected being about 600 times lower than that of the wild type. This can be explained by rapid degradation of non-secreted protein rather than by a lower stability because a mutant lacking just the 19 C-terminal amino acids accumulates to an amount 100 times less than the wild type.
Another new and unexpected feature revealed by the structure of S. marcescens LipA is related to the second set of tandem repeats. These repeats do not simply extend the first ␤-roll domain along the helical axis but pack laterally against it. This appears to be a general feature of all RTX proteins possessing more than eight nine-residue motifs, e.g. E. coli hemolysin (28 repeats) or Bordetella pertussis cyclolysin, where the ϳ40 repeats come in blocks separated by linker sequences. In this way the ␤-roll domains become integral parts of the whole structure and can act as folding nuclei, stabilizing different parts of the catalytic N-terminal domains. The larger those N-terminal domains, the more ␤-rolls are needed. The Ca 2ϩinduced folding of the N-terminal passenger domains by the ␤-roll has been demonstrated directly in some cases (47,54,55).