Structural analyses of the Haemophilus influenzae peptidoglycan synthase activator LpoA suggest multiple conformations in solution

In many Gram-negative bacteria, the peptidoglycan synthase PBP1A requires the outer membrane lipoprotein LpoA for constructing a functional peptidoglycan required for bacterial viability. Previously, we have shown that the C-terminal domain of Haemophilus influenzae LpoA (HiLpoA) has a highly conserved, putative substrate-binding cleft between two α/β lobes. Here, we report a 2.0 Å resolution crystal structure of the HiLpoA N-terminal domain. Two subdomains contain tetratricopeptide-like motifs that form a concave groove, but their relative orientation differs by ∼45° from that observed in an NMR structure of the Escherichia coli LpoA N domain. We also determined three 2.0–2.8 Å resolution crystal structures containing four independent full-length HiLpoA molecules. In contrast to an elongated model previously suggested for E. coli LpoA, each HiLpoA formed a U-shaped structure with a different C-domain orientation. This resulted from both N-domain twisting and rotation of the C domain (up to 30°) at the end of the relatively immobile interdomain linker. Moreover, a previously predicted hinge between the lobes of the LpoA C domain exhibited variations of up to 12°. Small-angle X-ray scattering data revealed excellent agreement with a model calculated by normal mode analysis from one of the full-length HiLpoA molecules but even better agreement with an ensemble of this molecule and two of the partially extended normal mode analysis-predicted models. The different LpoA structures helped explain how an outer membrane-anchored LpoA can either withdraw from or extend toward the inner membrane-bound PBP1A through peptidoglycan gaps and hence regulate the synthesis of peptidoglycan necessary for bacterial viability.

LpoA is a two-domain outer membrane lipoprotein and is encoded in the genomes of many proteobacteria (8,10). In E. coli, EcLpoA was shown to tightly bind to EcPBP1A and activate its transpeptidase and to be essential in a mutant that lacked PBP1B (8,10,11). Moreover, the presence of the LpoA C domain was required for these activities. In Vibrio cholera, deletion of LpoA caused growth defects in minimal media. Such mutants also showed less virulence in vivo (12). In the Haemophilus influenzae bacterium, the orthologous gene lpoA (previously yraM or HI1655) was shown to be essential for cell growth in vitro (13). The likely reason is that the ponA-encoded PBP1A in H. influenzae is also essential for cellular growth, as indicated by high-density transposon mutagenesis experiments (14,15).
Experimentally determined structures have previously been reported for the individual N-and C-terminal domains of LpoA. We described the 1.35 Å resolution crystal structure of the C-terminal domain of H. influenzae LpoA (LpoA-C; residues 257-575) (16). This domain shares a homologous fold, but low sequence identity, with periplasmic substrate-binding proteins involved in the high-affinity import of small molecules by ABC-like transporters (17). The bilobate structure of LpoA-C adopts an "open" conformation, with a large cleft between the two lobes, typical of unliganded substrate-binding proteins (16). In periplasmic substrate-binding proteins, ligand molecules tightly bind in the cleft and stabilize a closed-cleft conformation (18). Because the residues exposed in the LpoA-C cleft are highly conserved in all LpoA sequences and form nonpolar as well as polar surfaces, this cleft probably binds another molecule, possibly PBP1A or peptidoglycan. Truncation experiments confirmed that the EcLpoA C domain was sufficient for the activation of the EcPBP1A transpeptidase in vitro (8).
The N-terminal domain of H. influenzae LpoA (LpoA-N), which includes residues 26 -256, was predicted, based on significant sequence similarity to other periplasmic proteins of known structure, to contain repeats of the tetratricopeptide (TPR)-like motif, a helix-turn-helix motif commonly involved in protein-protein interactions (16). This prediction was more or less borne out by the recent NMR structure of E. coli LpoA-N, which indeed showed adjacent helix-turn-helix motifs, but without the usual curvature associated with TPRcontaining domains (19).
Knowledge of these structures by themselves appears to be insufficient for fully understanding how LpoA promotes PG synthesis. As the cell grows, PG hydrolases and amidases form holes in the sacculus to expose acceptor peptides for crosslinking to new strands (reviewed in Refs. 7 and 21), and turgor may cause these gaps to widen (22). Whereas movement of PBP1A is known to be controlled by the cytoskeletal actin-like protein MreB and associated inner membrane proteins MreC, MreD, RodA, and RodZ, what directs PBP1A specifically to the regions requiring new PG strands is unknown. As an outer membrane lipoprotein, LpoA would be in a position to sense the integrity of PG (2,7). Only at gaps or holes could LpoA extend its C domain through the PG and activate PBP1A to insert and cross-link new PG strands. Consistent with this proposal, recent studies of E. coli LpoB (an activator of PBP1B but structurally unrelated to LpoA) predicted that its N-terminal ϳ50 residues are unstructured and probably long enough to position the LpoB globular domain so that it can interact with PBP1B (23,24). Moreover, from small-angle X-ray scattering and NMR data, Jean et al. (19) proposed that full-length E. coli LpoA is a relatively rigid rod that is narrow enough to fit through the gaps and pores in the PG and that it is about 150 Å long, a sufficient length to interact with PBP1A. But it remains unknown how such an elongated and rigid protein would be accommodated in the absence of a suitable PG hole and how this LpoA structure could adapt to different distances between the inner and outer membrane in different species.
To clarify what structural elements of LpoA will allow it to pass through gaps in the PG and interact with PBP1A, we set out to provide experimentally determined structural data for an intact full-length LpoA. Here we present structures from three crystal forms of full-length H. influenzae LpoA, determined to resolutions of 2.0, 2.8, and 2.6 Å, as well as the crystal structure of its N domain (LpoA-N) to a resolution of 2.0 Å. Unlike the E. coli model (19), the full-length LpoA was not elongated in the crystals but forms a variable jaw-like structure. Also, the N domain assumed the canonical superhelical twist associated with other TPR-containing proteins. These structures provide insight into the possible role of the N domain and how conformational flexibility allows LpoA to adapt to different PG architectures.

The N-terminal domain of LpoA
The crystal structure of the H. influenzae LpoA N-terminal domain (LpoA-N), containing LpoA residues 33-253 and selenomethionine in place of Met, was refined to an R work /R free of 0.18/0.22 for data extending to 1.95 Å resolution (see Table 1 and "Experimental procedures"). The crystal structure includes two independent LpoA-N molecules with residues 33-248 and 33-249, respectively. The two molecules superpose with an RMSD of 0.49 Å for 202 of 215 residues. The molecules are arranged N-to-C along a pseudo-4 3 -screw axis (supplemental Fig. S1). The two molecules make intermolecular polar interactions between non-conserved residues on helices H2-H4 and H11-H12.
Thestructureispredominantlyhelical,consistingof14␣-helices (Fig. 1). Near the N terminus of the structure, there are two approximately antiparallel ␣-helices, H1 and H2, connected by a 5-residue ␤-like strand. These helices do not contact each other, but both straddle helix H3 lying beneath them. The H2-H3 loop (positions [53][54][55][56][57][58] in each molecule has poor electron density, but the path of the main chain could be discerned in molecule A. Helical pairs H3 and H4, H5 and H6, and H7 and H8 as well as H9 form 3 1 ⁄ 2 tetratricopeptide repeat-like (TPRlike) helix-turn-helix motifs. They pack against each other in a canonical superhelical fashion to provide the protein with concave and convex surfaces (Fig. 1B). The 6-residue helix H10 is approximately perpendicular to H9 and is not part of a TPR-like motif. H11 and H12 form a TPR-like motif but do not participate in superhelical packing, as do the other helix-turn-helix motifs. The short C-terminal helices H13 and H14 precede what in the full-length LpoA molecule is the linker strand (residues 251-257) between the N and C domains (see below). Helices of each TPR-like motif have a unique color, except for gray (H1-H2, H13-H14) and green (H9-H10), which are not TPR-like motifs. Views are perpendicular into the groove (A) and a side view (B). C, contact surface of the LpoA-N structure with coloring calculated by Consurf (62) from a sequence alignment of 74 putative LpoA amino acid sequences taken from the RP75 (March 2014) set of representative genomes (25) in Pfam PF04348 and aligned with Muscle in Jalview (61). Alignments are available in the supplemental materials. Sequences Ͼ96% identical or Ͻ200 residues in length were excluded. Front and back views of the domain are shown. Cyan surface, residues that are most variable; magenta, residues that are the most conserved. The black arrow points to a small highly conserved pocket (see "Results"). Note that surface accessibilities of certain residues may differ in LpoA-N structures from other species. D, same as C, but the input to Consurf was an alignment of only the seven sequences from the family Pasteurellaceae. E, same as C, but the input to Consurf was an alignment of 51 sequences from the family Enterobacteriaceae. Yellow surfaces, residues that the Consurf algorithm was not able to assign a statistically significant conservation score.

Multiple conformations of LpoA
The structure of HiLpoA-N can be considered as composed of two subdomains. Subdomain 1, composed of residues 33-148 (H1-H7), contains two three-helix bundles, H3-H5 and H5-H7, each having nonpolar residues at its core. In subdomain 2, residues 149 -249, nonpolar side chains from H11 interact with those from H8 -H10, and these helices interact with residues downstream to form a second nonpolar core that includes three Trp residues and other aromatic residues. Other than contacts between H7 and H8, there are few interactions between subdomain 1 and subdomain 2, which allows for flexibility within the N domain, as discussed below.
This division of LpoA-N into two distinct subdomains is also suggested by a comparison of the HiLpoA-N crystal structure with the NMR structure of the E. coli LpoA N domain (19). HiLpoA-N and EcLpoA-N have sequences that are 30% identical, and both form similar TPR-like motifs. However, whereas the structures of residues 33-149 and of 150 -248 of HiLpoA-N are similar, respectively, to the equivalent regions in EcLpoA-N (Fig. 2, A and B), the orientation of these two regions relative to one another in the HiLpoA-N structure differs markedly from that in the EcLpoA-N structure, by a ϳ45°rotation (Fig. 2C). This difference is consistent with the EcLpoA-N structure not showing the typical superhelical conformation of TPR-containing domains (19).

Most solvent-exposed residues of LpoA-N are not conserved
To identify the surfaces of LpoA-N that are conserved, we aligned the RP75 subset of LpoA sequences from the Pfam family PF04348 (accessed March 2014) (25,26) that were Ͼ200 residues long. This alignment resulted in 74 sequences from 40 genera of ␥-proteobacteria. This analysis, when including all of the sequences, revealed the LpoA N domain to have few conserved solvent-exposed residues (Fig. 1C). A small pocket on the convex side of LpoA-N surrounded by the highly conserved residues His-235, Ala-236, and Trp-204 was also found (arrow in Fig. 1C). The lack of conservation of the surface of the N-terminal domain stands in marked contrast to the C domain (see below). In the context of the full-length HiLpoA structure, the C domain has 4 times the percentage of solvent-exposed area considered highly conserved (magenta in Fig. 1C) compared with the N domain. The C domain has been shown in E. coli to be functionally indispensable for cell growth in an LpoA-dependent cell growth assay and for binding to EcPBP1A (8,10). Also, this lack of conservation anywhere on the LpoA N-domain surface contrasts with an assessment that only included sequences from the Pasteurellaceae or Enterobacteriaceae families (Fig. 1, D and E). This latter assessment did reveal highly conserved regions, on the concave side of the N domain, including the loop between H3 and H4 (residues 76 -78 and 81), the surface of H2 and H3, and the H9 -H14 interface, including residues Gln-172, Asp-176, and Lys-243.

LpoA-N forms a groove like TPR-containing peptide-binding proteins
The packing of TPR-like motifs in LpoA-N creates a groove on the concave surface directly over H7 with side chains from H3 and H5 and H8 and H9 forming the walls of the groove (Figs. 1B and 3 (A and B)). This groove is closed at one end by residues Asp-41, Arg-136, and Arg-170, which are located on the H1-H2 connector, H7, and H8, respectively (Fig. 3A). Arg-136 and Arg-71 are conserved in several bacterial families, and these two residues form a pocket directly under the H1-H2 connector at the closed end of the groove (Fig. 3C). A sulfate ion is bound in the pocket of the LpoA-N chain A molecule. Electron density at the same position in chain B could not be identified and was modeled as glycerol. Electrostatic calculations showed the LpoA-N cleft to have an overall positive charge (Fig. 3B), with the basic residues Arg-158, Arg-170, Arg-75, Arg-106, and Lys-177 facing this groove (Fig. 3C). Glu-146, Asn-147, and Arg-158, also point into the groove and are mostly conserved in Pasteurellaceae and Enterobacteriaceae.
A comparison of the crystal structure of LpoA-N with other proteins suggests that the groove in LpoA-N may be suitable for binding peptides or proteins. To find structural homologs of the LpoA-N structure, we compared it with other protein structures with DALI (27). All of the homologs superposed onto the TPR-like motifs (H3-H9) of HiLpoA-N but often had more

Multiple conformations of LpoA
repeats than did LpoA-N. No significant structural homologs were found for the non-TPR region (helices H10 -H14) of LpoA-N. The highest scoring homologs included members of a family of cytoplasmic quorum-sensing receptors from Grampositive bacteria (28) that bind small imported peptides. The RapI receptor (PDB entry 4ILA) without bound peptide displayed the highest sequence identity (21%) with LpoA-N (139 residues superposed with an RMSD of 3.1 Å). NprR (PDB entry 4GPK) from Bacillus cerius is another member of this transcription regulator family and had the highest Z-score (13.0; a measure of structural similarity) in the DALI search. This structure contains the signaling peptide from NprX bound in the concave cleft of its TPR domain (29). When LpoA-N was superposed onto NprR by aligning the first TPR motif (H3 and H4) of LpoA-N to the second TPR motif of NprR, the peptide-binding cleft of NprX coincided with the groove of LpoA-N (Fig. 3D). Similar to LpoA-N, the NprR-binding groove also contains polar side chains, which in NprR interact with the main chain of the bound NprX peptide.

Full-length LpoA is not extended, and the linker mostly interacts with the C domain
Structures of the full-length HiLpoA were determined from three different crystal forms. The orthorhombic crystal structure (LpoA Ort ), with one molecule in the asymmetric unit, included residues 34 -573 and was refined to 2.0 Å resolution with an R work /R free ϭ 0.17/0.22 ( Fig. 4A and Table 1). We also refined structures with data from two different monoclinic crystal forms of LpoA: LpoA Mon to 2.8 Å with R work /R free ϭ 0.21/0.25, and LpoA Mon2 with two molecules in the asymmetric unit to 2.6 Å and R work /R free ϭ 0.22/0.26. Whereas detailed differences between these HiLpoA structures relevant to the flexibility of the molecule were observed and are discussed later, here we first describe the structure of LpoA in the orthorhombic form (LpoA Ort ), being of highest resolution.
The N domain and C-terminal domain (C domain, residues 257-575) are arranged to form a U-shaped (jaw-like) structure, with the longest dimension being 91 Å between residues 91 and 505 in LpoA Ort , and a width ranging from 35 to 75 Å ( Fig. 4A; see also "LpoA flexibility suggested by structure comparisons, normal mode analysis, and small-angle X-ray scattering"). We estimated the maximum length of HiLpoA Ort to be ϳ105 Å, on the assumption that residues 26 -32 (not present in the crystallized protein) are unstructured as they are in the EcLpoA-N NMR structure (19).
The N domain is joined to the first ␤-strand of the C domain through a 7-residue linker polypeptide, residues 251-257 (Fig.  5B). The linker is in an extended conformation with residues 252-257 primarily in a crevice on the side of the C domain and Phe-251 making van der Waals contacts with both N and C domains (Fig. 5B). About 50% of the surface area of the linker is exposed to bulk solvent. There are only a few direct hydrogen bonds between the linker residues and the rest of the protein (see the legend to

Multiple conformations of LpoA
the N domain through van der Waals interactions (buried surface area ϭ 270 Å 2 ) and at least six water-mediated polar interactions (Fig. 5C).

The C-domain structures in LpoA Ort and LpoA-C are very similar
The C-terminal domain of the LpoA Ort structure is very similar to the LpoA-C (PDB entry 3CKM, C-terminal domain only) structure determined previously (supplemental Fig. S2) (16). The domain consists of two lobes (N lobe and C lobe) connected by three polypeptide strands and separated by a large cleft (Fig. 4B). Each lobe superposes on the corresponding lobe of the 3CKM structure with an RMSD of 0.3 Å. Based on the concentration of highly conserved residues between the lobes (Fig. 4C), we previously proposed it to be a ligand-binding cleft (16). Two regions of LpoA Ort differed significantly from the

Multiple conformations of LpoA
3CKM structure of LpoA-C. Residues 425-434 have a different conformation, which is probably due to different crystal packing. One loop near the cleft (residues 345-350) was reported to have ambiguous density in the previous 3CKM structure but is wellresolved in LpoA Ort and adopts a conformation different from that previously modeled in 3CKM (green in Fig. 4E and supplemental Fig. S3). Consequently, this opened a path from the top of the binding cleft into a partially conserved and electronegative region at the "top" of the C domain, which faces the N-domain groove described above (Fig. 4, D-F). Differences between the C domain in the four full-length structures are discussed below.

LpoA flexibility suggested by structure comparisons, normal mode analysis, and small-angle X-ray scattering
By comparing the four full-length crystal structures of LpoA, we observed at least three modes of flexibility: twisting of the TPR-like motifs in the N domain, hinge bending between the two lobes of the C domain, and C-domain rotation at the end of the interdomain linker. Surprisingly, most of the linker region as defined above assumes a very similar conformation in all LpoA crystal structures (supplemental Fig. S4).
To assess the conformational variability of the LpoA N domain, we superposed subdomain 1 (i.e. residues 33-148 (H1-H7)) of each of the two N domains in the LpoA-N crystal and the N domains from LpoA Ort and LpoA Mon onto each other and calculated the rotations necessary to superpose each of the subdomain 2 regions (residues 149 -246) (Fig. 6A). They all showed 4 -5°rotations, but along different axes, implying a general flexibility between these two subdomains. Such rotations within the N domain, although relatively small in magnitude, can lead to considerable effects on the position of the C domain. For example, the twisting of the N domain in the LpoA Mon structure is responsible for rotating the C domain (N lobe) by 4.4°, effectively shifting it an average of 5 Å in the direction of the linker in comparison with the LpoA Ort structure (supplemental Fig. S5). No change in the linker conformation was observed.
Previously, we reported that normal mode calculations of LpoA-C (PDB entry 3CKM) indicated that a hinge region between the two lobes of the C-terminal domain would allow the cleft between the two lobes to open and close, similar to differences actually observed in multiple crystal forms of the Leu/Ile/Val-binding protein and other periplasmic binding proteins (16,18). Comparing the two lobes of the C domain in all full-length crystal structures shows that the maximum difference is between the C domains of LpoA Mon and LpoA Mon2 chain B, where the cleft opening increases by 12.1°around an axis passing near the hinge region between the two lobes ( Fig.  6B; also see supplemental Fig. S2B). Notably, residues Arg-393 and Lys-322, which block access to the electronegative region on the top side of the C domain in the LpoA Ort and LpoA Mon structures (Fig. 4E), are farther apart in LpoA Mon2 chain B, extending the putative binding cleft region (Fig. 6C).
Results of normal mode analyses (NMA) often reflect actual protein dynamics observed in solution (30 -33). To examine the potential for flexibility in full-length LpoA Ort , we calculated the normal mode-derived motions of LpoA Ort with an arbitrary amplitude of 500 (34) and compared the starting structure with the final structure. In the final NMA structure (NMA(LpoA Ort ); Fig. 7A, left), the N-domain superhelix was slightly unwound, and the C domain was rotated ϳ25°with respect to the starting conformation.
To assess whether the conformation predicted by the normal mode analysis actually occurs experimentally, we compared the NMA(LpoA Ort ) prediction with the two molecules (chains A and B) in the asymmetric unit of the LpoA Mon2 crystal form. Each of these two molecules (green and red in Fig. 7B and supplemental Fig. S6A) was observed to adopt a conformation on the path to a more extended structure similar to that predicted

Multiple conformations of LpoA
by the above normal mode calculation. After superposing subdomain 1 (residues 33-148) of the N domains, there were two components of the conformational change between LpoA Ort and the LpoA Mon2 structures. In the N-terminal domain, subdomain 2 (residues 149 -250) of both molecules from LpoA Mon2 was observed to be rotated 5°with respect to subdomain 1, resulting in a 2-3 Å movement of the C domain (not shown); second, the C-terminal domains of the two molecules (A and B) of the LpoA Mon2 asymmetric unit were rotated by 20 and 30°, respectively, relative to the LpoA Ort structure (Fig. 7B).
In the latter case, the net rotation is around an axis almost parallel to the linker strand (Fig. 7C). As in NMA(LpoA Ort ), the conformational change was observed to be mediated primarily by slight changes in the main-chain dihedral angles of residues 256 -259 at the end of the linker (Fig. 7C). The resulting largest difference between LpoA Ort and LpoA Mon2 molecule B was at residue 453 of the C domain, which moves 33 Å (Fig. 7D). The largest change in an intramolecular distance was 25 Å between residues 54 and 454 of the two structures (supplemental Fig.   S6B). The longest dimension in the LpoA Mon2 molecule B structure, at 99 Å, is 8 Å greater than that measured in the LpoA Ort structure (supplemental Fig. S6C). The consequence of this conformational change is that the center of the putative binding groove (around residue 360) in the C domain tilts and moves 13 Å away from the N terminus, which in the periplasm would be toward the inner membrane. To further quantify the similarity, we calculated the RMSDs of all of the full-length structures superposed onto the extended structure predicted by normal mode analysis (Fig. 7B). Molecule B of the LpoA Mon2 structure was most similar to the extended structure predicted by the NMA of LpoA Ort with an RMSD of 2.5 Å. Repeating the NMA calculation (with amplitude of 500) starting from molecule B of the LpoA Mon2 structure resulted in NMA(LpoA Mon2 ) with another ϳ25°rotation of the C-domain movement relative to LpoA Mon2 and the longest dimension of the protein now extending 116 Å (at maximum length excluding residues 26 -32) and a predicted radius of gyration (R g ) of 34.9 Å (Fig. 7,  B and E). shows two other C-domain comparisons. C, the consequence of the two lobes of the C domain spreading apart in B is that the LpoA Mon2 chain B (bottom; green solvent contact surface) has a more direct pathway (black arrow) to the electronegative region on top of the C domain (red arrow; also see Fig. 4E). In contrast, in the LpoA Mon (top, orange) C-domain structure, access to the electronegative region is blocked. This may affect the site accessibility to a bound polypeptide or PG strand.

Multiple conformations of LpoA
To further validate the LpoA structures, we measured smallangle scattering curves at four concentrations of LpoA(29 -575) (supplemental Fig. S7A). The small-angle X-ray scattering (SAXS) data show that LpoA is monomeric in solution. We estimated the molecular weight from SAXS data in relative scale using two methods. The molecular weights are 57,489 and 63,859 from methods of correlation volume (35) and apparent volume (36), respectively. The estimations were made with scattering data extending to q ϭ 0.25 Å Ϫ1 (q ϭ 4sin/). The theoretical M r was 60,518. The Guinier plots (Fig. 8A) show that the intensity at zero angle is proportional to the protein concentration, and the experimental R g decreases with increasing concentration. Such an effect can be eliminated by extrapolation to infinite dilution (zero concentration). The R g at infinite dilution is 32.6 Ϯ 0.6 Å in reciprocal space from the Guinier plot (Fig. 8B) and 33.1 Ϯ 0.3 Å in real space calculated by the GNOM software (37).
Predicted R g values for each of the four LpoA models in the crystal structures and for the models predicted by normal mode analysis of LpoA Ort and LpoA Mon2B were calculated with Crysol (38) (Fig. 7B). The R g(calc) for the NMA of LpoA Ort (R g(calc) ϭ 32.4 Å) was closest to the R g from the SAXS experimental data. Predicted scattering profiles for the models were fit to the observed data with FoXS (39) that included parameters for a hydration layer, excluded volume adjustment, and implicit hydrogens and are shown in Table 2 (left). Again, NMA (LpoA Ort ) fit best with ϭ 1.25 (Fig. 8E).
The HiLpoA SAXS results are in contrast to the SAXS analysis of EcLpoA (19). After comparing the distance distribution function P(r) derived from EcLpoA SAXS data with those calculated from various models, Jean et al. (19) showed that, in comparison with other models, a fully extended EcLpoA was consistent with an observed R g ϭ 42.2 Å (Fig. 7B, far right). Although NMA(LpoA Ort ) best fit the HiLpoA SAXS data, two observations suggested that the molecule samples other conformations, perhaps more elongated than that observed in the crystal structures. The Porod-Debye plot (Fig. 8C) calculated from the SAXS data did not show the Porod-Plateau, which suggested that the LpoA protein in solution is more flexible (40). Second, an R g -based, dimensionless Kratky plot (Fig. 8D) (41) shows an asymmetric bell shape and the peak position (1.97, 1.21) shifted to higher values with respect to a globular protein (1.73, 1.11). This indicated that the protein has structured domains linked by a flexible segment and can adopt a more elongated shape than a globular protein.
Consistent with the above, we obtained a better fit ( ϭ 0.80) to the HiLpoA SAXS experimental data when an ensemble of three models (42) were considered together, each with a specific volume fraction: LpoA Ort 0.27, NMA(LpoA Ort ) 0.28, and NMA(LpoA Mon2 ) 0.45 (Table 2 (right) and Fig. 8 (F and G)). Note that the structure with the highest volume fraction, NMA(LpoA Mon2 ), is the most elongated, with a C-domain rotation of about 40 -50°compared with the LpoA Ort conformation, yet this structure by itself does not fit the experimental data as well as NMA(LpoA Ort ) ( ϭ 2.18 versus 1.25; Table 2, left). This means that the ensemble may be a better interpretation of the SAXS data with the closed and more extended structures existing simultaneously in solution. The presence of an extended structure like NMA(LpoA Mon2 ), with a maximum interatomic distance of 116.7 Å, is also consistent with the D max value of 118 Å determined from the pairwise distancedistribution plot, P(r) (Fig. 8H). This also would support the ability to adopt an extended conformation in vivo for interaction with PBP1A.

Discussion
This paper reports crystal structures for the N domain and full-length HiLpoA as well as alternative conformations suggested by normal mode calculations and consistent with the SAXS analysis. The U-shaped (or L-shaped) arrangement of the two domains in the full-length HiLpoA structures presented here is very different from the elongated model proposed for EcLpoA in solution (see below; Fig. 7B) (19). In HiLpoA Ort , the nonspecific interactions between the linker and C domain may restrain the position of the linker (along the side of the C domain) and result in the observed orientation of the two domains in the HiLpoA full-length structures. Despite the differences in domain orientation observed in the different fulllength crystal structures, the linkers have very similar structures. Whereas the linker is the same length in all LpoA homologs, its sequence is not highly conserved. Phe-251, which makes nonpolar contacts with both domains of HiLpoA, is only present in Pasteurellaceae and some Enterobacteriaceae sequences. Several sequence families (including the Enterobacteriaceae and Shewanellaceae) have proline at position 253 (HiLpoA numbering), which, when modeled in HiLpoA, does not affect the linker conformation but might affect its flexibility. Similarly, species from Shewanella, Pseudomonas, and Vibrio have a Pro at 256 in place of Val. In summary, the linker sequence may regulate the conformational flexibility of LpoA, Figure 7. LpoA crystal structures progressively extend until they reach the NMA prediction. A, overlay of LpoA Ort (gray surface representation) on the extended model (magenta) predicted by normal mode calculations by ElNèmo (34) for mode 7 and amplitude ϭ 500. The overlay was based on fitting residues 33-148 (subdomain 1) of the N domain of LpoA Ort to that of the normal mode model and revealed the largest intramolecular distance change (23 Å) between LpoA Ort and the predicted structure (right; helices (cylinders) and ␤-strands (arrows)). B, comparison of the four crystal structures and two models calculated by NMA (mode 7, amplitude ϭ 500). Subdomains 1 of the N domain of each structure are identically oriented. The RMSDs of the C␣ coordinates for each crystal structure compared with the NMA(LpoA Ort ) model were 6.3 Å for LpoA Mon , 5.5 Å for LpoA Ort , 3.6 Å for LpoA Mon2 molecule A, and 2.5 Å for LpoA Mon2 molecule B, showing that the latter structure is most similar to the NMA results. The NMA(LpoA Ort ) had a calculated radius of gyration (32.4 Å) very close to that determined directly from the SAXS data (32.6 Å). A similar figure with the molecules rotated ϳ90°is shown in supplemental Fig. S6A. The image at the far right is one of several hypothetical EcLpoA models, constructed by putting the two domains end-to-end, that was consistent with the R g from SAXS data (19)

Multiple conformations of LpoA
perhaps reflecting the peptidoglycan architecture of a particular species.

Need for LpoA flexibility
During PBP1A-mediated PG synthesis (at least in E. coli), LpoA must extend its C domain a sufficient distance toward the inner membrane to interact with the ODD domain (also called the OB domain due to its oligosaccharide-binding-like domain fold (43)) and TP domains of PBP1A (8). The LpoA architecture must provide conformational flexibility for at least two reasons. First, the width of the periplasm, the thickness of the peptidoglycan, and the distance of PG from the outer membrane probably differ between families of bacteria (e.g. see Ref. 44). Noting that the width of the periplasm in E. coli was shown to be 200 Å (44), Jean et al. (19) proposed, based on small-angle X-ray scattering data, that the EcLpoA full-length structure is extended with an overall length of 145 Å. They reasoned that this length would allow the C domain to bind to the PBP1A ODD domain. However, this model may not hold for other species because periplasm width varies among species. Analyses by atomic force and electron microscopy of the envelope of Aggregatibacter actinomycetem-comitans, a species in the same Pasteurellaceae family as H. influenzae, revealed that the outer membrane is highly rugose (ruffled) (45). The distance between the relatively smooth inner membrane and the outer membrane fluctuates by as much as 150 Å. According to the authors, this feature is also present in H. influenzae (45).
The second reason for flexibility is to differentiate between intact PG and the presence of gaps, as suggested previously (22). Intact sacculus has holes/pores of about 40 Å in diameter. Due to the elasticity of PG, these holes would enlarge during cell growth, resulting in gaps requiring repair by a PG synthase (22,46). It was proposed that the larger gaps would allow LpoA to extend through the sacculus and bind to PBP1A, resulting in a localized activation of PBP1A and insertion of new PG strands to repair the hole (7,22) (Fig. 9). Because LpoA appears randomly distributed throughout the outer membrane (OM) (8), it is conceivable that in regions of intact PG, LpoA would need to "curl up" or assume a more compact conformation to fit in the space between the OM and PG (22) (Fig. 9). This would argue for a non-rigid LpoA molecule (i.e. one with conformational flexibility).

N-domain flexibility
From comparing the crystal forms of HiLpoA, we observed three examples of flexibility: twisting and flexing of the N domain, rotation of the entire C domain with respect to the N domain, and hinge motion between the lobes of the C domain (Figs. 6 and 7). All of the structures showed small but significant flexing of the N domain (Fig. 6A). In the normal mode analysis of LpoA Ort , relative twisting of the TPR-like motifs changes the length of the N domain, causing changes in the position of the C domain with only minor changes in the main-chain torsion angles of the interdomain linker. The low sequence conservation of the N domain may reflect its ability to adopt a conformation or length specific to the cell wall structure of a particular species. The HiLpoA-N structure suggests that the N domain Figure 8. Small-angle X-ray scattering data. A, Guinier plots of X-ray small-angle scattering intensities measured at four concentrations of LpoA. Shown are the R g values derived from the linear portion of the Guinier plot. The reciprocal lattice vector length q is 4sin/ (wave number). The plot was in the q range, where R g ϫ q max is Յ1.3. B, this Guinier plot obtained after data extrapolation to zero protein concentration gives a radius of gyration of 32.65 Å. C, the Porod-Debye plot does not have the typical Porod plateau, indicating a flexible protein structure (40). D, SAXS data are plotted as a dimensionless Kratky plot based on R g (41). See "Results" for significance. E, predicted scattering from the NMA(LpoA Ort ) structure (red) fit best to the experimental scattering curve (black) with ϭ 1.25. The lower section of the graph shows the residual, calculated as the ratio between the two curves. F, same as E, but the scattering curve (red) for a 3-model ensemble (LpoA Ort , NMA(LpoA Ort ), NMA(LpoA Mon2 )) is fit to the observed data (black) with ϭ 0.80. G, the three structures constituting the ensemble. The most extended structure in the ensemble, NMA(LpoA Mon2B ), has a 45% volume fraction, suggesting that it is a significant conformer in solution. H, pairwise distance-distribution plot, P(r). The maximal distance (D max ) of 118 Å derived from this is consistent with the most extended structure considered here, NMA(LpoA Mon2B ), with maximum interatomic distance of 116.7 Å.   3UDF)). Shown from right to left, LpoA Mon or a more compact structure would rotate to fit above the PG, here assumed to be at 60 Å below the OM. A structure similar to LpoA Ort could detect the presence of a hole in the PG, and a more extended conformation like that observed for LpoA Mon2 chain B or NMA(LpoA Mon2B ) would be long enough to interact with PBP1A. IM, inner membrane; OB, oligosaccharide binding-like domain (also referred to as ODD); GT, glycosyltransferase domain; S, catalytic serine.

Multiple conformations of LpoA
may have the important role in the positioning of the C domain at the appropriate location to interact with PBP1A. This role in C-domain positioning is also supported by the observation that the N-domain sequences of LpoA from several genera are shorter (e.g. Proteus, Providencia, Morganella, and Acinetobacter), which may correlate with different cell wall morphologies.
The LpoA N-domain structure, which adopts a crescentshaped form typical of other TPR-containing domains, is strikingly different from the N-domain structure of EcLpoA determined by NMR (Fig. 2C) (19). This may be due to differences in sequence between HiLpoA-N and EcLpoA-N or reflect a difference in the crystal versus solution environment. Perhaps the conformational difference in EcLpoA-N reflects particular aspects of the periplasm (width?) or PG structure that are different from other Gram-negative bacteria, as mentioned above. For example, in E. coli, the Lpp outer membrane lipoprotein interacts both covalently and noncovalently with the PG, effectively fixing the PG at a relatively fixed distance below the outer membrane. However, H. influenzae lacks this protein, so the distance from the outer membrane to PG may vary.
The N-domain groove on the concave face of the domain resembles peptide-or protein-binding sites found on other TPR-like proteins. Recent reports suggest that LpoA may bind other polypeptides. CpoB is a periplasmic protein that localizes to the septum, where it interacts with E. coli PBP1B, LpoB, and the Tol proteins, and is necessary for PBP1B function. Gray et al. (47) showed that the N domain of LpoA could rescue the function of CpoB when PBP1B was the only functioning synthase. Because both CpoB and LpoA-N contain TPR-like motifs, the authors suggested that LpoA-N may interact with other PG-associated proteins or perhaps link to the Tol protein (47). Additionally, Dörr et al. (48) discovered in V. cholera that the periplasmic protein CsiV bound to both LpoA (specific domain not known) and PG and regulated the ability of LpoA to activate the PBP1A transpeptidase. CsiV is encoded in a subset of species that encode LpoA, but not in H. influenzae or E. coli. Perhaps there is another protein in E. coli or H. influenzae that serves this role.
Both molecules in the LpoA Mon2 structure when compared with LpoA Ort show a 5°rotation within the N domain, which has the effect of translating the C domain (similar to movement observed in LpoA Mon ; supplemental Fig. S5B). In addition, the C domain rotates 20 -30°around an axis parallel to the linker region, which in the cell positions the C domain farther away from the outer membrane and potentially closer to PBP1A. Interestingly, in the LpoA Mon2 chain B structure, the observed C-domain rotation does not affect linker conformation significantly (supplemental Fig. S4) but is a result of main-chain torsion angle differences at residues 257 and 259 following the linker (Fig. 7C). This is consistent with the NMR measurements of full-length EcLpoA, which suggested a relatively rigid linker (19). The presence of four conformations in three crystal forms suggests that LpoA is sampling many different conformations in solution.
The SAXS results confirm that an ensemble of structures varying from the LpoA Ort to the NMA(LpoA Mon2B ) conformation may occur in solution, although in vivo, we would expect the distance between the outer membrane and peptidoglycan layer, as well as the thickness of the peptidoglycan, to limit the number of conformations actually sampled. The protein dynamics may also be limited by molecular crowding due to the high concentration of proteins in the periplasm. The movements are probably necessary for LpoA binding to PBP1A. In the absence of LpoA, the movement of PBP1A in the periplasm is fast and diffusive (49). When LpoA activates PBP1A catalytic activity or binds to PBP1A, the net mobility of PBP1A decreases (49). This is probably necessary for productive catalytic function and may be a major regulatory function of LpoA (49).
At least in E. coli, the LpoA C domain is crucial for the binding and activating of PBP1A (8,10,16). Superpositions of C-domain pairs LpoA Mon on LpoA Ort and LpoA Mon on LpoA Mon2 showed a movement of the two lobes of the C domain and a narrowing or opening of the cleft between them (Fig. 6B and supplemental Fig. S2B). This observation suggests that LpoA-C (as well as the full-length structure) can sample multiple conformations in solution similar to that observed for LivJ, a structural homolog of LpoA-C (18). This may reflect an induced fit model for PBP1A (or possibly PG) binding. Although ABC transporter substrate-binding proteins often completely close the cleft on substrate binding, we suspect that the C-domain cleft of LpoA may not completely close because there is a highly conserved salt bridge between Arg-515 in the N lobe and Asp-488 (Glu in some species) in the C lobe that may restrain the open position of the two lobes (supplemental Fig. S8).
We estimate that the groove on the concave side of the N domain would lie ϳ50 Å below the outer membrane (Fig. 9).
Here it might make interactions with peptidoglycan. Consistent with this proposal is the finding that EcLpoA bound to peptidoglycan sacculi purified from E. coli (8). For the domain configuration observed in the LpoA full-length crystal structures, we expect LpoA (in the LpoA Mon2 chain B or NMA(LpoA Mon2 )) to extend about 115 Å below the outer membrane, positioning the center of the C domain just below the presumed location of the peptidoglycan layer. The model depicted in Fig. 9 assumes a periplasm width of 170 Å and a HiPBP1A model with dimensions similar to the known structure of Acinetobacter baumannii PBP1A (43). Future structural studies of a PBP1A-LpoA complex will help clarify how the two proteins interact.

Plasmid construction
Oligonucleotides were purchased from Sigma Genosys Ltd. and Integrated DNA Technologies. Plasmids and PCR products were purified with kits from Qiagen or Promega. Restriction enzymes and polymerases were used as recommended by the manufacturers (Invitrogen, Promega, and New England Biolabs). Plasmid constructs were sequenced at the University of Michigan DNA Sequencing Core.
PCR primers and details of the plasmid constructs are shown in Tables 3 and 4. Genomic DNA isolated from the H. influenzae Rd strain (ATCC number 9008) (50) served as the PCR template. For plasmid pLpoAN-His 6 , the gene fragment corre-
Orthorhombic crystals of YraM-N and YraM-N(SeMet) were grown in hanging drops consisting of 2 l of 10 mg/ml protein solution and 2 l of precipitant (30% polyethylene glycol monomethyl ether 5000, 0.2 M ammonium sulfate, 0.1 M MES, pH 6.5) equilibrated against 1 ml of the precipitant by vapor diffusion. Crystals grew to a maximum size of 0.6 ϫ 0.4 ϫ 0.1 mm within a week. Crystals were harvested into precipitant solution containing 10% glycerol, mounted on loops, and immediately frozen in liquid nitrogen.
Native and heavy atom derivative intensity data from LpoA-N crystals were collected on a Rigaku R-Axis II detector and phased by multiple isomorphous replacement with SOLVE and RESOLVE (53). Electron density maps revealed two molecules per asymmetric unit, and an initial model was built with O (54). Diffraction images from LpoA-N(SeMet) crystals were collected at the absorption peak wavelength of 0.979 Å on a MAR-CCD detector at station 5ID, DND-CAT, Advanced Photon Source, Argonne National Laboratory. Reflection intensities to 1.95 Å resolution were integrated and scaled with d*TREK (55). Unit cell dimensions and statistics for the SeMet crystals are shown in Table 1. Structure refinement in PHENIX (56) with TLS parameters produced a model with R work /R free ϭ 0.19/0.22 and excellent geometry. One sulfate was modeled in the electropositive pocket of molecule A in the LpoA-N structure (Fig. 3C), whereas density tentatively modeled as glycerol was observed in the corresponding pocket of molecule B.

Multiple conformations of LpoA
A sulfate was also observed to interact with Arg-148 of each molecule.

Full-length LpoA crystal structures
His-tagged full-length LpoA expressed from Tuner TM (DE3)/ pLacI cells transformed with pLpoA-His 6 was susceptible to proteolysis and failed to produce crystals at 4 or 22°C. To prepare LpoA protein without a His 6 tag, 1 liter of Terrific Broth supplemented with 100 g/ml ampicillin and 33 g/ml chloramphenicol was inoculated overnight with a 10-ml starter culture of Origami TM (DE3)/pLacI previously transformed with pLpoA (Table 4). Cultures were grown at 37°C with shaking at 250 rpm to an A 600 ϭ 0.6 -0.8, transferred to room temperature, and induced with 0.2 mM IPTG. Cells were harvested 16 h later and lysed by applying sonication. LpoA was purified by carrying out ammonium sulfate fractionation at 30 and 50% and then dialyzed against 50 mM Tris-HCl, pH 8.0, overnight to remove salt. The protein was loaded successively onto Source Q and then Mono Q anion-exchange columns and eluted with a 0 -1 M NaCl gradient in 50 mM Tris-HCl, pH 8. The LpoA peak eluted at ϳ150 mM NaCl. Peak fractions were concentrated and loaded on a HiPrep TM Superdex 75 16/60 (GE Biosciences) gel filtration column, where LpoA eluted as a single peak with an apparent molecular mass of 60 kDa. LpoA in 150 mM NaCl, 50 mM Tris-HCl, pH 8.0, was concentrated to 33 mg/ml as determined by absorbance at 280 nm with an extinction coefficient calculated by ProtParam (http://web.expasy.org/protparam) and used immediately for crystallization.
This full-length LpoA formed clusters of branched needles under many conditions using vapor diffusion at both 22 and 4°C. To minimize the growth of these clusters, we added 0.2 l of 30% xylitol (Hampton Research) to hanging drops containing 1 l of concentrated protein and 1 l of precipitant (25% (w/v) PEG 1500, 0.1 M MMT buffer (20 mM DL-malic acid, 40 mM MES, 40 mM Tris-HCl), pH 4.0). The trays containing these hanging drops were incubated at 4°C for 2 days and then transferred in an insulated box to 22°C. Here, the drops produced rod-shaped orthorhombic crystals, which diffracted X-rays to a d min of 1.97 Å. Intensity data from these crystals, after being frozen, were collected at beamline 21ID-G at LS-CAT (Advanced Photon Source, Argonne, IL) and processed with HKL-2000 (57). These data were phased by the molecular replacement method with Phaser (58) using the previously determined LpoA-C (PDB entry 3CKM) and the LpoA-N structure as search models. Refinement with PHENIX and manual refitting with Coot resulted in final R work and R free values of 0.17 and 0.21, respectively (56,59). The final model included residues 34 -573, with poor electron density for residues 55-58 and 430 -433. Residues 473-478, which were well-ordered in the LpoA-C structure (PDB entry 3CKM), were observed to be disordered in LpoA Ort , probably due to the different arrangements of the molecules in their respective crystals. The two models also differed in the 345-350 loop above the cleft and between the two lobes. In the LpoA-C structure, density was ambiguous, and residues 346 -349 were omitted from the refinement calculations (16). In LpoA Ort , this loop was well-resolved but with a different conformation than in the LpoA-C structure (supplemental Fig. S4).
Monoclinic crystals of LpoA Mon (SeMet) were grown using similar procedures as above, but with 8% PEG 4000, 0.1 M sodium acetate trihydrate, pH 4.6, as the precipitant with added xylitol. These crystals were transferred to a precipitant solution containing 15% glycerol and then frozen in liquid nitrogen. These frozen crystals did not diffract as well as the orthorhombic crystals but provided a 90% complete data set to 2.8 Å resolution ( Table 1). The structure was solved by molecular replacement as above and refined with PHENIX. Restraining the structure to the orthorhombic structure in the final cycles resulted in an R work /R free ϭ 0.21/0.23 with excellent geometry.
Crystals of another monoclinic form, LpoA Mon2 , were grown from conditions similar to the above monoclinic crystal, and they showed different unit cell dimensions and two molecules per asymmetric unit. These crystals yielded an incomplete data set to a resolution of 2.6 Å that was phased as above with Phaser in PHENIX. The structure was refined with PHENIX to an R work /R free ϭ 0.23/0.29 (Table 1).
Software used for analyzing the structures included PyMOL for preparing the figures (60), Jalview for sequence alignments (61), Consurf for the calculation of sequence homology (62), and ElNémo for normal mode calculations (34). Coordinates were deposited in the Protein Data Bank (63).

Small-angle X-ray scattering
For the SAXS experiment, full-length HiLpoA protein with an N-terminal His 6 tag was expressed from Origami (DE3) cells transformed with pMCSG7-LpoA(29 -575). Protein was purified on a TALON (Clontech) cobalt affinity column, digested with TeV to remove the His 6 tag, rerun on the cobalt column, and, after dialysis and concentration, loaded on a Superdex-75 column. Fractions were evaluated by SDS-PAGE (supplemental Fig. S7A), pooled, and frozen at Ϫ80°C. Protein was bufferexchanged into 50 mM Tris-HCl, 150 mM NaCl, 10% glycerol, pH 7.3, and dilutions were prepared in the same buffer for SAXS. The final purified protein contained residues S-N-A-HiLpoA (29 -575) and migrated at ϳ60,000 on an SDS-polyacrylamide gel (supplemental Fig. S7A), consistent with the predicted molecular weight of 60,518.
Preliminary scattering profiles were measured on the inhouse SAXS instrument at the SAXS core (NCI, National Institutes of Health). Subsequently, small-angle scattering from buffer (50 mM Tris-HCl, 150 mM NaCl, 10% glycerol, pH 7.3) and four LpoA protein dilutions (8.8, 4.4, 2.2, and 1.1 mg/ml) were measured at the 12-ID-B beamline of the Advanced Photon Source (APS), Argonne National Laboratory, with photon energy of 14 keV and an off-center Pilatus 2 M detector. Simultaneous wide-angle X-ray scattering also was recorded, and the total q range covered was ϳ0.005 Ͻ q Ͻ 2.8 Å Ϫ1 . A total of 30 sequential data frames were recorded for each buffer and sample solution with an exposure time of 0.75-2 s to minimize radiation damage and obtain a good signal/noise ratio. The 2D images were corrected and reduced to 1D scattering profiles using the Matlab software package at the beamlines. The 1D SAXS profiles were grouped by sample and averaged, and background scattering due to the buffer was subtracted (supplemental Fig. S7B). The scattering intensity profiles were extrapolated to infinite dilution, to remove the scattering contribution due Multiple conformations of LpoA to interparticle interactions, and to low scattering angle following the results of the Guinier plot (Fig. 8, A and B). Both intensity at zero scattering angle (I 0 ) and R g calculated from each Guinier plot (Fig. 8A) were dependent on protein concentration, as expected. The R g was calculated for each of the six models with the Crysol software (38). Theoretical scattering profiles were generated from the six models and compared with the experimental SAXS data at q Ͻ 0.5 Å Ϫ1 using the FoXS software (64). The minimal ensemble search of the six structures to improve the fit to the observed data was calculated by the FoXS server (http://modbase.compbio.ucsf.edu/foxs) 5 (64).
Author contributions-Protein expression, crystallization, and structure determination were performed by K. S., J. V., and B. T.