Lipids assist the membrane insertion of a BAM-independent outer membrane protein

Like several other large, multimeric bacterial outer membrane proteins (OMPs), the assembly of the Klebsiella oxytoca OMP PulD does not rely on the universally conserved β-barrel assembly machinery (BAM) that catalyses outer membrane insertion. The only other factor known to interact with PulD prior to or during outer membrane targeting and assembly is the cognate chaperone PulS. Here, in vitro translation-transcription coupled PulD folding demonstrated that PulS does not act during the membrane insertion of PulD, and engineered in vivo site-specific cross-linking between PulD and PulS showed that PulS binding does not prevent membrane insertion. In vitro folding kinetics revealed that PulD is atypical compared to BAM-dependent OMPs by inserting more rapidly into membranes containing E. coli phospholipids than into membranes containing lecithin. PulD folding was fast in diC14:0-phosphatidylethanolamine liposomes but not diC14:0-phosphatidylglycerol liposomes, and in diC18:1-phosphatidylcholine liposomes but not in diC14:1-phosphatidylcholine liposomes. These results suggest that PulD efficiently exploits the membrane composition to complete final steps in insertion and explain how PulD can assemble independently of any protein-assembly machinery. Lipid-assisted assembly in this manner might apply to other large OMPs whose assembly is BAM-independent.


Results
sPulS facilitates rapid PulD 28-42/259-660 multimerisation in lecithin liposomes. We previously observed that PulD 28-42/259-660 multimerisation in vitro is inversely dependent on the concentration of lecithin in the coupled synthesis and insertion reaction 29 . To find conditions under which the effects of adding PulS to the in vitro coupled transcription-translation reaction could be measured, we added a non-lipidated form of PulS (sPulS) to the reaction mixture in the presence of increasing amounts of lecithin before commencing PulD 28-42/259-660 synthesis. Although the overall production of PulD 28-42/259-660 is lower at lecithin concentrations above 27 mM, we have shown previously that this does not impair the analysis of initial PulD multimerisation 29 . sPulS was used because the presence of a lipid anchor would (1) require the use of detergent that interferes with liposome integrity and (2) physically restrain PulS on the lipid surface rather than being free in solution to recruit PulD monomers. We established previously that PulS produced in this way interacts efficiently with PulD 19,30,31 .
The observed increase in initial PulD 28-42/259-660 multimerisation in the presence of sPulS could stem from a direct kinetic advantage by rapid association of the PulD 28-42/259-660 /sPulS complex with the lecithin surface, from a PulS-induced conformational advantage for PulD 28-42/259-660 oligomerisation, or from a change in the PulD 28-42/259-660 folding mechanism. If sPulS binding induces the formation of small oligomers prior to dodecamerisation, then the membrane dependent multimerisation reaction would reduce in order, lowering its inverse dependence on the lecithin concentration: dodecamerisation by monomeric addition results in a dependence of the initial multimerisation of up to twelve, hexamerisation of dimers in a dependence of up to six and so forth. Such oligomers can be observed by creating mixed multimers between full-length PulD (PulD fl ) and PulD 28-42/259-660 , which separate as a regular 13-step ladder by SDS-polyacrylamide electrophoresis 29 . If sPulS helps form oligomers that are obligatory intermediates in dodecameristion, then distinct steps should disappear when PulD fl and PulD 28-42/259-660 synthesised separately in the presence of sPulS are mixed post-synthesis 29 . However, PulD 28-42/259-660 formed regular ladders with PulD fl in the presence or absence of sPulS (Fig. 1d), making it unlikely that sPulS induces the formation of small obligate oligomers prior to dodecamerisation. sPulS does not accelerate other PulD 28-42/259-660 folding steps in lecithin liposomes. Since sPulS caused more rapid initial multimerisation of PulD 28-42/259-660 at 53 mM lecithin, it was next investigated whether the effect of PulS extended to subsequent kinetic steps in PulD 28-42/259-660 folding. PulD 28-42/259-660 folding under these conditions was monitored by SDS-treatment and by subjecting PulD 28-42/259-660 to trypsinolysis at increasing time points after 10 min synthesis (Fig. 2a). Despite the more rapid PulD 28-42/259-660 multimerisation in the presence of sPulS immediately after 6 min synthesis, sPulS did not significantly accelerate other phases in PulD 28-42/259-660 folding (Fig. 2b). Previously, PulD 28-42/259-660 folding was characterised by two rate constants in a multistep sequential process with simultaneous acquisition of SDS-and urea-resistance, followed by trypsin-resistance of the native protein core upon membrane insertion 29 . Here, data scattering warranted data fitting to a single exponential equation only. The rate constant of 0.18 ± 0.01 min −1 agreed well with the fast rate constant of 0.14 ± 0.04 min −1 reported in the absence of sPulS 29 . As was the case in the absence of sPulS, trypsin-resistance in lecithin liposomes was acquired after a delay of approximately 20 min (Fig. 2b). Together, the data demonstrate that sPulS only improves the efficiency of initial steps in PulD 28-42/259-660 folding in lecithin liposomes in vitro. E. coli lipids do not enhance sPulS dependent PulD 28-42/259-660 folding. Lipids with phosphatidylcholine (PC)-headgroups (like lecithin) have proven successful for the in vitro folding of many OMPs from their chemically denatured state 24 , but bacterial membranes rarely contain PC-headgroups 33,34 . The inner leaflet of bacterial outer membranes consists mostly of lipids with phosphatidylethanolamine (PE)-headgroups (up to 90 %) and phosphatidylglycerol (PG)-headgroups 34 . The outer leaflet is exclusively composed of lipopolysaccharides 1 . Most chemically denatured bacterial OMPs fold only inefficiently in vitro in membranes derived from native sources like E. coli 24 . However, PulD 28-42/259-660 , multimerised efficiently after 10 min synthesis in the presence of E. coli liposomes prepared from E. coli polar extract lipids (Fig. 2c). The absence of a delay in the acquisition of trypsin-resistance indicated that PulD 28-42/259-660 acquired its final native state efficiently and faster in the presence of E. coli lipids than in the presence of lecithin liposomes (Fig. 2b,c,e).
To investigate whether sPulS could further accelerate PulD 28-42/259-660 membrane insertion into E. coli membranes, the effect of adding sPulS prior to PulD 28-42/259-660 synthesis on the acquisition of PulD 28-42/259-660 trypsin-resistance was measured. As in lecithin liposomes, the addition of sPulS to the synthesis reaction did not increase the rate of PulD 28-42/259-660 folding into its trypsin-resistant native state in the presence of E. coli lipids (Fig. 2d,e). Thus, whereas PulD 28-42/259-660 membrane insertion is rapid in the presence of liposomes prepared from E. coli lipids, the addition of sPulS did not reveal an additional kinetic advantage in the presence of these lipids. We cannot exclude that the PulS lipid-anchor plays an additional (e) Plot of the multimerisation kinetics measured by SDS-resistance () and by trypsin-resistance in the absence (☐) and presence of (◊) sPulS in (c,d) following band quantification by densitometry. Errors represent S.D. over 3 independent measurements. Mu tr and Mo tr indicate the migration position of multimeric and monomeric PulD 28-42/259-660 species, respectively. Only multimers are shown for the trypsin resistant state, as monomers were completely digested. role in the PulD folding process, for example by creating membrane defects or enforcing a particular orientation with respect to the membrane.

PulS attachment does not prevent PulD assembly in vivo.
The lack of any effect on late steps in PulD 28-42/259-660 folding kinetics might reflect that the PulD 28-42/259-660 /sPulS complex dissociates once it adsorbs onto the lipid surface or forms a dodecamer. Complex dissociation could be a prerequisite for the formation of a native, secretion-competent PulD complex in the outer membrane. To examine further whether this is the case, a series of cysteine variants in PulD fl and PulS was created that would cross-link spontaneously upon interaction in vivo. Their capacity to form an efficient cross-linked PulD fl -PulS product, to secrete PulA and to induce a PspA-response was examined. The choice of residues for substitution by cysteines was based upon the crystal structures of PulS 31 and that of the PulS homologue OutS containing the binding peptide of the S-domain of the PulD homologue OutD 35 (Fig. 3a). OutS has the Of the residues lining the PulS binding cleft, Q38 on α -helix 1 and Q95 on α -helix 3 appeared to be good candidates (in distance and orientation relative to the PulD peptide) for substitution by cysteines: Q38C might cross-link with two residues of the PulD S-domain, A649C and F654C, while Q95C might cross-link with A643C of the PulD S-domain (Fig. 3a). All of the pairs tested produced PulD fl multimers, albeit less efficiently than observed with wild-type PulD fl and wild-type PulS (Fig. 3b). An upper shift in the electrophoretic migration of PulD fl A643C monomers recognized by anti-PulD antibodies indicated that it cross-linked with high efficiency to PulS Q95C , while PulD fl A649C and PulD fl F654C showed no or limited cross-linking with PulS Q38C , respectively (Fig. 3b). PulD fl A643C monomers and multimers were also recognised efficiently by anti-PulS antibodies (Fig. 3b). Cross-linking was specific between the engineered cysteines A643C and Q95C, as efficient cross-linking was not achieved when either of the variants was produced in the presence of the wild-type binding partner (Fig. 3c). The slower migrating cross-linked PulD fl A643C -PulS Q95C hetero-dimer disappeared upon treatment with dithiothreitol (DTT), which resulted in a concomitant increase in the amount of non-cross-linked PulD fl A643C monomers. In addition, broadening of the multimer band upon DTT treatment also suggested that PulD fl A643C multimers were cross-linked to PulS Q95C (Fig. 3c). Phenol treatment dissociated the multimers and predominantly resulted mainly in an increase in the amount of cross-linked PulD fl A643C -PulS Q95C hetero-dimers (Fig. 3c). Neither phenomenon occurred with any of the other combinations or between wild-type proteins and the variants (Fig. 3c).
Having established that at least the majority of the PulD fl A643C multimers formed were cross-linked to PulS Q95C , we next investigated whether these multimers behaved like the wild-type upon production in E. coli; i.e., do they induce the Psp response and permit PulA secretion. A Psp response is induced when PulD multimers associate with or insert into the inner membrane. For example, PspA production is high when even a small amount of PulD fl multimers is assembled in the absence of PulS (Fig. 4a). In this case, PulD fl is no longer transported to the outer membrane, which results both in rapid PulD fl degradation and in insertion of PulD that escapes degradation into the inner membrane, leading to PspA induction. PspA induction is lower, but still above background, when PulD fl is rendered resistant to degradation and is targeted to the outer membrane by PulS (e.g., Fig. 4a). PspA induction remained at similar levels when either PulD fl A643C or PulS Q95C were produced in the presence of their wild-type binding partner or when PulD fl A643C and PulS Q95C were produced together and cross-linked (Fig. 4a). Thus, cross-linked hetero-dimers were likely transported to the outer membrane. Furthermore, since PspA levels were above background, a small proportion of the PulD fl A643C multimers apparently inserted into the inner membrane and formed small pores, as is the case with wild-type PulD and PulS 21,22 .
PulS production and lipidation are critical to achieve PulA secretion through PulD 30,38 . Regardless of the presence of the substituted cysteines and regardless of the formation of a disulfide bridge, all multimeric secretins secreted PulA efficiently (Fig. 4b).
Hence, it appears that PulD fl A643C -PulS Q95 cross-linked secretins are fully functional and that PulS dissociation is not required to allow PulD assembly and function.  Fig. 5. PulD 28-42/259-660 appears as two stacked rings when examined by EM 15 . As previously, particle averaging revealed two orientations: a stack when viewed in the plane of the membrane (side view) and a disc perpendicular to the axis of symmetry (top view) (Fig. 5).
The results suggest that the hydrophobic thickness of the membrane affects the rate of PulD 28-42/259-660 folding; however, hydrophobic thickness alone is not a critical determinant in PulD 28-42/259-660 folding.

Discussion
This report examines the factors required for the folding and assembly of the OMP PulD, whose biogenesis is independent of the general OMP-specific assembly machinery (BAM) 23 . Previous in vivo and in vitro studies 20,21,46 demonstrated that the lipoprotein PulS plays an essential role in delivering PulD to the outer membrane. Here we show that PulS can improve the efficiency of early PulD multimerisation steps in vitro, but we failed to observe any major influence on later steps in PulD folding corresponding to the transition of the prepore into the native structure 29 . Nonetheless, PulS does not have to dissociate from its substrate for PulD to complete this transition. These observations clearly show that the PulD assembly pathway is quite different from the general pathway used by most OMPs, in which the broad specificity chaperones SurA and Skp must release their substrates near the membrane surface, passing them on to BAM for assisted membrane insertion 2 . Other members of the PulS family likely play equivalent roles in secretin assembly 30,37 . Dedicated chaperones AspS and PilF to the secretins GspD from Vibrio and E. coli EPEC and PilQ from P. aeruginosa, respectively, catalyse Bam-independent secretin assembly 9,10,47,48 , reflecting even closer involvement of this chaperone in secretin assembly than is the case for PulS. Other dedicated chaperones have somewhat different roles. In Neisseria meningitidis, for example, the chaperone PilW is critical for secretin PilQ stability 49 , while PilQ assembly is reported to rely on BAM 8 . The Pseudomonas aeruginosa secretin HxcQ is a lipoprotein that uses the Lol-pathway to reach the outer membrane 50 and might be stabilised by a second, smaller protein 47 . It is unknown whether HxcQ assembly is BAM-dependent. A systematic study of secretin assembly and folding pathways by a combination of in vivo and in vitro approaches would further define these differences and reveal whether secretins can be divided into two biogenesis classes: one relying totally on classical BAM-mediated membrane insertion and for which dedicated chaperones have roles not directly related to assembly, and a second class independent of BAM for membrane insertion, but for which dedicated chaperones are required for the correct localisation and for the catalysis of assembly steps of at least some.
Since PulS only catalyses initial PulD assembly steps, the question remains how PulD inserts into the membrane. Membrane insertion could depend on an as yet-unidentified assembly machinery, although this seems unlikely in view of its ability to insert both into artificial, protein-free liposomes in vitro and into the E. coli inner membrane when PulS is absent in vivo 21,22 . However, we observe that the lipid composition of the membrane influences PulD folding in vitro in an unusual fashion. Neither the lipid headgroup composition nor the hydrophobic thickness themselves appeared to be critical for folding. If the headgroup composition were critical, then PulD 28-42/259-660 folding would have been slow in diC 18:1 PC-liposomes, whereas the contrary was observed. If membrane thickness were the critical factor, then PulD 28-42/259-660 folding would have been more efficient in lecithin compared to E. coli lipid, whereas the contrary was observed. Instead, we propose that general physical membrane properties, like membrane curvature and membrane-stored energy, drive efficient PulD insertion, both in vitro and in vivo. Stored energy is high in PE-containing bilayers (as in E. coli extract liposomes) because of the non-bilayer packing conformations of the PE-lipids that lead to an increase in curvature stress 51,52 , and also in thick membranes composed of lipids with long saturated or mono-unsaturated acyl chains (as in diC 18:1 PC) 53 . In contrast, thick lecithin liposomes, which contain a high number of poly-unsaturated acyl chains, and thin PC-/PG-liposomes form highly elastic membranes with little stored energy.
As PulD insertion appears to be tuned to suit the in vivo membrane composition, the observations reported here might rationalise why PulD assembly is Bam-independent. BAM comprises five proteins: four peripheral lipoproteins (BamB-E) and one membrane embedded protein, BamA, which forms the central component that catalyses OMP insertion 3 . High-resolution structures and simulations reveal how the 16-stranded β -barrel of BamA distorts the membrane around strands 1 and 16, providing an access route for OMP insertion into the membrane 6 by lowering the energy barrier for OMP membrane insertion in the presence of PE-containing phospholipids 5,7 . If PulD exploits membrane-stored energy for its membrane insertion, it would not need a system such as BAM that lowers the energy. Insertion in this manner likely requires a high level of organisation to measure the amount of energy stored in the membrane, for example by sensing lateral pressure. Formation of the PulD prepore 39 to organise the C-domains could provide a means to achieve this. Therefore, the stability of the prepore structure could be a critical parameter in determining the fate of assembling PulD secretins.
How general is this phenomenon of BAM-independent OMP assembly? Although BAM-independent assembly was initially reported for PulD 23 and then for other secretins in the same family 10 , the OMPs CsgG, GfcC and Wza were also shown recently to exhibit Bam-independent assembly. Like PulD, they also appear to form prepore structures 9,54,55 . Like PulD 29,39 , all of these complexes require the coalescence of multiple subunits to form a single transmembrane pore or channel. We hypothesise that a common assembly mechanism based on achieving a critical stability in the prepore and membrane-assisted insertion represents a new paradigm for complex OMP assembly. The characteristics that determine whether OMP assembly is BAM-dependent or not might be encoded in the three-dimensional structure of the OMP, which remains to be determined at high-resolution for secretins. In vitro analysis of the folding of OMPs with diverse structural features in the presence and the absence of BAM would greatly advance our understanding of the mechanisms involved.
Plasmids encoding for PulD variants were obtained by site-directed mutagenesis on the plasmids pCHAP3635 56 and pCHAP362 57 . The first is a pSU18 derivative that allows high levels of PulD production for cross-linking and PspA response assays, whilst the second is a pHSG575 derivative for a low production level used in secretion assays. PulS variants were generated from a pUC19 derived vector containing the pulS gene (pCHAP585) 56 . Primers used for mutagenesis are listed in Table 1.
To produce PulS Q95C in the presence of all other Pul proteins, the pulS gene was mutagenised through a cloning sequence rather than by site-directed mutagenesis. This was required because of the large size of the plasmids carrying the entire pul operon. First, the DNA fragment encoding for all the Pul proteins (except PulD fl ) was amplified from pCHAP1226 58 and ligated into pCHAP231 38 using restriction sites PsiI and HindIII. This created the plasmid pCHAP3402 that encodes for all the Pul proteins except PulD and has unique AscI and AsiSI restriction sites flanking a 2929 bp fragment carrying the pulS gene. Two separate, partially overlapping amplicons were generated to cover the entire 2929 bp: one fragment from the AsiSI-site up to the codon for Q102 on the pulS gene and a second from the codon for S89 on the pulS gene to the AscI-site. Primers annealing to the pulS gene carried the required codon change to substitute amino acid Q95 into C in PulS (Table 1, primers ING339 and 340). Primers annealing near the AsiSI and AscI-sites (in italics) are 5′ -AAACGACGGCCAGTGAATTCAGGCGATCGCCGTTGAAGGTC-3′ and 5′ -GACCATGATTACGCCAAGCTTTAACGGCGCGCCTGGCGG-3′ , respectively. Both primers also contained an EcoRI and a HindIII-site (underlined), respectively, to enable an intermediate cloning step into the pUC19-vector amplified with the primers 5′ -GAATTCACTGGCCGTCGTTTTAC-3′ and 5′ -AAGCTTGGCGTAATCATGGTC-3′ . The three fragments were assembled using the Gibson assembly master mix (NEB) to give plasmid pCHAP3404. The fragment carrying the mutagenised pulS gene was excised from pCHAP3404 using AsiSI and AscI and was ligated into the plasmid pCHAP3402 digested with the same enzymes to give pCHAP3405. All constructs were verified by DNA sequencing. analysed by immunoblotting with antibodies against PspA, PulS, PulD and OmpF, as indicated. Bands were analysed by densitometry.
PspA induction and PulA secretion assay. PspA induction was measured from the same cells used in the cross-linking assays. An empty vector and cells producing wild-type PulD in the absence of PulS were used as negative and positive controls, respectively.
PulA activity following its secretion to the outer surface was measured upon induction of the entire pul operon with 0.4% maltose in cells transformed with one of two plasmids encoding for the entire set of Pul proteins except PulD fl , pCHAP3402 (for wild-type PulS) or pCHAP3405 (for PulS Q95C ), and either pHSG575 (empty vector), pCHAP362 (wild-type PulD fl ) or pCHAP3406 (PulD fl A643C ). Pullulanase secretion was measured as a fraction of the pullulanase enzymatic activity on the bacterial surface in whole cells compared to that of octyl-polyoxyethylene lysed bacteria and relative to the activity upon PulA secretion in the presence of wild-type PulS and PulD fl 56,59 . sPulS production and purification. Production and purification of sPulS is described elsewhere 19 .
Briefly, cells containing the plasmid for the expression of MalE-PulS with an N-terminal hexahistidine-tag were grown to a D 600 = 0.5 and induced with 0.5 mM IPTG for 4 h. Cells were harvested, lysed and debris was removed by centrifugation. The supernatant was applied to a nickel charged HiTrap column for affinity purification. After elution, MalE-PulS containing fractions were dialysed and digested overnight with Factor Xa. sPulS was further purified by cation exchange (HiTrap SP-Sepharose column) and gel filtration (HiLoad 16/60 Superdex 200 column).
PulD synthesis. PulD was synthesised by in vitro translation using an RTS100 E coli kit (5 Prime) as described 25,29 in the presence of 10 ng DNA (pCHAP3731 (PulD fl ), pCHAP3716 (PulD 28-42/259-660 ), pCHAP3803 (PulDΔ S 28-42/259-598 ), 10 to 60 μ g liposomes and 0.2 μ g sPulS (as indicated) per μ l RTS100 at 30 °C. Although the RTS100 kits were centrifuged at 100000 g for 30 min before use to remove most of the E. coli membranes, the trace amounts that remain are sufficient to allow limited PulD assembly. Synthesis was arrested with 3 ng streptomycin per μ l of reaction after 6 min for initial multimerisation experiments, after 10 min in all other kinetic experiments and for at least 6 h for structural characterisation. Synthesis reactions were further incubated for at least 6 h at 30 °C for complete folding to occur. Mixed multimers were produced by priming the reaction with the relevant DNAs in a 1:1 ratio. PulDΔ S 28-42/259-598 was used as a control for the effects of the addition of sPulS to the reaction mixture. PulDΔ S 28-42/259-598 no longer has the S-domain that binds PulS and behaves in all experiments performed as PulD 28-42/259-660 in the absence of sPulS. Monomeric and multimeric PulD were separated in SDS on a 10 % polyacrylamide (37.5:1 acrylamide/bis-acrylamide) gel without heating to 100 °C, transferred to nitrocellulose and analysed by immunoblotting with an antibody raised against native PulD-multimers. Bands corresponding to multimeric and monomeric PulD were analysed by densitometry. Resulting transients were fitted to a single exponential equation using Kaleidagraph 4.0. Fitting parameters are reported in the text.
Folding kinetics followed by SDS treatment. Folding transients of PulD 28-42/259-660 were obtained by mixing aliquots of the synthesis reaction at the time points indicated with SDS sample buffer in a 1:1 ratio to arrest folding and incubated on ice for 1 h before analysis by SDS-PAGE.

Folding kinetics followed by limited proteolysis by trypsin digestion. Trypsin was added to
PulD 28-42/259-660 aliquots at the times indicated to a final concentration of 4 μ g/μ l and incubated on ice for 5 min. Reactions were blocked using 150 ng/ml Pefabloc (Interchim) before mixing with SDS sample buffer for analysis. The fraction of trypsin resistant multimers was determined relative to the amount of SDS-resistant multimer at the endpoint of the reaction.
Protein solubilisation and transmission electron microscopy (TEM). Liposomes, purified as above, were resuspended in 100 mM Tris, pH 7.5, and 500 mM NaCl and diluted twice in 2% DDM. The lipid to detergent ratio was typically 1:5 (w/w) for solubilisation. For negative staining, 4 μ l of sample was adsorbed onto carbon film-coated copper EM grids, washed with three droplets of pure water and subsequently negative stained with 2% (w/v) uranyl-acetate. The prepared grids were imaged using a Philips CM10 TEM (FEI, Eindhoven, The Netherlands) operating at 80 keV. Images were recorded on a side-mounted Veleta 2 K × 2 K CCD camera (Olympus, Germany) at a magnification of 130000. The pixel size at the sample level is 3.7 Å. Image processing was performed in the EMAN2 software package 60 . The images were contrast transfer function corrected and the particles were semi-automatically