Outer-membrane translocation of bulky small molecules by passive diffusion

Significance

The outer membrane (OM) of gram-negative bacteria forms a protective layer on the outside of the cell that prevents unrestricted access of harmful compounds. For the acquisition of ions and nutrients, the OM contains two types of transport proteins: passive diffusion channels and active transporters. Due to the limited diameters of passive diffusion channels, bulky molecules such as iron–siderophores and complex oligosaccharides are assumed to be taken up exclusively by active transporters. Here we assert that this assumption is incorrect. Using a combination of biophysical and computational approaches, we show that the OM protein CymA (cyclodextrin metabolism A) from Klebsiella oxytoca represents a previously unidentified paradigm in OM transport by mediating the passive diffusion of cyclic oligosaccharides (cyclodextrins) with diameters of ∼15 Å.

Abstract

The outer membrane (OM) of gram-negative bacteria forms a protective layer around the cell that serves as a permeability barrier to prevent unrestricted access of noxious substances. The permeability barrier of the OM results partly from the limited pore diameters of OM diffusion channels. As a consequence, there is an “OM size-exclusion limit,” and the uptake of bulky molecules with molecular masses of more than ∼600 Da is thought to be mediated by TonB-dependent, active transporters. Intriguingly, the OM protein CymA from Klebsiella oxytoca does not depend on TonB but nevertheless mediates efficient OM passage of cyclodextrins with diameters of up to ∼15 Å. Here we show, by using X-ray crystallography, molecular dynamics simulations, and single-channel electrophysiology, that CymA forms a monomeric 14-stranded β-barrel with a large pore that is occluded on the periplasmic side by the N-terminal 15 residues of the protein. Representing a previously unidentified paradigm in OM transport, CymA mediates the passive diffusion of bulky molecules via an elegant transport mechanism in which a mobile element formed by the N terminus acts as a ligand-expelled gate to preserve the permeability barrier of the OM.

The outer membrane (OM) of gram-negative bacteria serves as an efficient permeability barrier for water-soluble and hydrophobic molecules. To obtain the necessary compounds for cell growth and function, the OM contains various β-barrel membrane proteins that serve as diffusion channels (1). Two classes of OM diffusion channels can be distinguished: nonspecific porins and substrate-specific channels. Due to the need to prevent unrestricted access of potentially toxic molecules, OM diffusion channels from both classes have pores that are relatively narrow, with diameters of at most 7–8 Å. Because of these limited pore sizes, cellular entry by diffusion is thought to be effectively prevented for molecules larger than ∼600 Da. However, many physiologically important molecules are larger than this “OM size-exclusion limit.” Such compounds (e.g., iron–siderophores, vitamin B₁₂, and complex oligosaccharides) are taken up by active OM transport proteins termed TonB-dependent transporters (TBDTs) (2). Although TBDTs have large barrels of 22 β-strands they do not have a permanently open pore, due to an N-terminal “plug” or “cork” domain that completely fills the lumen of the barrel. The plug has a short sequence termed the TonB box that, upon substrate binding by the transporter, interacts with a periplasmic domain of the TonB protein. TonB is part of the ExbBD–TonB inner-membrane protein complex, which functions as a proton pump to induce conformational changes in TonB as a result of the movement of protons. These conformational changes may then result in partial unfolding or ejection of the TBDT plug, resulting in a transient large channel through which the substrate passes into the periplasmic space. Thus, transport of bulky small molecules by TBDTs is an energy-dependent process that requires the proton motive force (PMF) across the inner membrane (2).

The OM protein CymA of Klebsiella oxytoca, a close relative of K. pneumoniae, is part of the cym (cyclodextrin metabolism) operon dedicated to the utilization of cyclodextrins (CDs), cylindrical oligosaccharides containing six (α-CD), seven (β-CD), or eight (γ-CD) glucose units. Orthologs of CymA are present in the Enterobacteriaceae and Vibrionaceae (including Vibrio cholerae), but only the protein from K. oxytoca has been studied. CDs are bulky molecules, with outer diameters of 13.7 Å for α-CD (molecular mass 973 Da) and 15.3 Å for β-CD (molecular mass 1,135 Da); that is, they are clearly too large to pass through known OM diffusion channels. CDs are formed extracellularly from starch by the action of secreted cyclodextrin-glucanotransferases (cgts). Besides cymA and cgt, the cym operon contains cymE, coding for a periplasmic binding protein; cymDGF, encoding an ABC transporter; and cymH, encoding a cytoplasmic cyclodextrinase that converts CDs into linear maltooligosaccharides that then enter the maltose degradation pathway (Fig. 1) (3). Escherichia coli transformed with the cym operon grows on α-CD and β-CD as sole sources of carbon (4). In the absence of the cymA gene, growth on CDs is lost (4). In addition, CymA confers to an E. coli ΔlamB strain the ability to use maltopentaose and maltohexaose for growth (5). These data indicate that CymA functions as an efficient OM uptake channel for α- and β-CD as well as for linear maltooligosaccharides. The relatively small size of CymA (∼39 kDa) and sequence analysis suggest that CymA is not a TBDT but a regular diffusion channel, raising the question of how such large substrates are transported across the OM without the requirement for an external energy source and without compromising the OM permeability barrier.

Fig. 1.

Schematic overview of cyclodextrin uptake and metabolism in K. oxytoca. IM, inner membrane.

To answer these questions, we report here the X-ray crystal structures of CymA in the absence and presence of the substrates α- and β-CD. CymA forms a 14-stranded β-barrel with a simple architecture for the extracellular loops. The N-terminal ∼15 residues fold into the lumen from the periplasmic side and constrict the channel. The CymA–CD cocrystal structures reveal two α-CD binding sites inside the channel, one at the extracellular mouth and the other near the periplasmic exit, delineating the passageway through the OM. The structures, together with molecular dynamics simulations and single-channel electrophysiology experiments, reveal a previously unidentified and elegant mechanism for the passive diffusion of bulky molecules while at the same time preserving the permeability barrier of the OM.

Results

The CymA N Terminus Forms a Mobile Element in the Channel Lumen.

After unsuccessful molecular replacement trials with crystals of the native protein, the structure of CymA was solved with phases obtained from a single anomalous dispersion experiment on a selenomethionine (SeMet)-substituted crystal. The CymA structure shows a monomeric 14-stranded β-barrel with relatively short extracellular loops (Fig. 2). A Dali search identified OmpG as the closest structural homolog in the database [Protein Data Bank (PDB) ID code 2IWV; Z score = 24, rmsd 2.9 Å (6, 7)]. There is clear helical density inside the barrel lumen on the periplasmic side, resembling a postcleavage autotransporter structure (8). Surprisingly, the helical density corresponds to the entire 15-residue-long cloning region including the heptahistidine sequence (Methods) and the first 2 residues of CymA. Electron density for residues 3–10 is absent, presumably due to disorder, but the intervening distance of ∼15 Å could easily be bridged by those missing 8 residues. To our knowledge, this is the first example of a histidine tag forming part of a well-defined α-helix, and is a striking example of how environmental context shapes protein structure.

Fig. 2.

Overview of CymA crystal structures. (A) Cartoon model of SeMet CymA viewed from the side (Left) and from the extracellular side (Right) in rainbow coloring. The numbering for the visible residues of the N terminus (blue) is shown. Negative numbers correspond to residues of the cloning region. (B) Comparison of the interior of the barrel lumen for His-tagged SeMet CymA (PDB ID code 4V3G), His-tagged CymA (PDB ID code 4V3H), and wild-type CymA (PDB ID code 4D51). The cloning region is colored blue and the first 17 residues of CymA are shown in magenta. Selected residues are numbered.

In addition to the SeMet structure, we also solved a structure for the unlabeled (“native”) His-tagged protein using data to 1.9 Å. Interestingly, this structure, obtained using different crystallization conditions relative to the SeMet protein, does not show any density for both the cloning region and the first 10 amino acids of CymA (Fig. 2), indicating that the N terminus has likely moved out of the barrel. For both structures, none of the extracellular loops fold inward to constrict the channel, contrasting with virtually all other OM diffusion channels (1). Consequently, the lumen of the barrel of the high-resolution structure is empty and forms a circular channel with a fairly uniform diameter of ∼11–14 Å that is filled with water and has a length of ∼45 Å. Compared with E. coli OmpF, it is clear that the CymA channel is much wider, with dimensions that appear a close match to that of α-CD (Fig. S1). Taken together, the structural data for the tagged protein suggest that the N terminus of CymA is mobile and might be able to move in and out of the large-diameter channel.

Fig. S1.

Relative dimensions of the channels of CymA (Left) and E. coli OmpF (Right) compared with the CymA substrates α-CD (Middle, Bottom) and β-CD (Middle, Top). Both substrates are clearly too large to pass through the OmpF channel. The channel-constricting loop L3 in OmpF is colored blue. Hydrogen atoms are included both for the channels and the substrates.

We next asked where the N terminus is located in the wild-type, tagless protein. To answer this question, we purified and determined the crystal structure of nontagged CymA (termed “wild type”). As shown in Fig. 2, the N terminus of wild-type CymA is present inside the barrel. The side chain of Arg5 interacts with three glutamic acid residues in the barrel wall (Glu30, Glu287, and Glu289), and is likely to be important for positioning of the N terminus given that few other interactions are evident. Interestingly, residue Arg(-11) of the cloning tag makes very similar interactions in the structure of the SeMet protein (Fig. S2). In the wild-type protein structure, density is missing for residues 10–21 but, similar to the His-tagged protein, the intervening distance (∼17 Å) can be bridged by those missing residues. Both His-tagged CymA and wild-type CymA migrate at their expected molecular weight in SDS/PAGE gels, arguing against proteolytic cleavage in the N terminus as a source of the disorder. To obtain a more complete picture of the wild-type protein, we modeled the missing residues (Glu10–Phe21) followed by 100 ns of equilibrium molecular dynamics (MD) simulation (Fig. 3). The simulation suggests that the structure is stable (Fig. S3) and confirms the likely importance of Arg5 for positioning of the N terminus inside the barrel. Moreover, high root-mean-square fluctuation (rmsf) values and the presence of relatively few interactions between Glu10–Phe21 and residues in the barrel wall provide an explanation for the fact that this segment is not observed in the crystal structure (Fig. 3). The wild-type CymA model shows that the presence of the N terminus inside the barrel restricts the diameter of the channel dramatically (Fig. 3), and this has important implications for substrate transport.

Fig. S2.

Stereo diagram of cartoon superpositions for His-tagged SeMet CymA (white) and wild-type CymA (green), highlighting the interactions of the N-terminal arginine residue (Arg5 in wild-type CymA) with glutamic acid residues in the barrel wall. Nitrogen atoms are colored blue and oxygen atoms are colored red.

Fig. 3.

Constriction of the CymA channel by the N terminus. (A) Cartoon of wild-type CymA with residues Glu10–Phe21 (magenta) modeled followed by a 100-ns MD simulation. Residues 1–9 visible in the crystal structure are shown in blue. (B) Root-mean-square fluctuations (Top) and hydrogen-bonding interactions (Bottom) of the N terminus (residues 1–25) with residues of the barrel wall. (C) Surface views from the periplasmic space for CymA with the N terminus expelled from the barrel (PDB ID code 4V3H) (Left) and for the wild-type CymA model (Right), illustrating the size difference of the channel. (D) Average pore radii derived from unbiased MD simulations with the corresponding SDs for wild-type CymA with (red) and without (black) the N terminus bound inside the barrel. For comparison, the pore radius from the high-resolution crystal structure of OmpF (PDB ID code 2OMF) is shown as a blue line. The pore radii were determined using the program HOLE (29). EC, extracellular; PP, periplasmic.

Fig. S3.

Unbiased MD simulation of full-length CymA, showing that the structure is stable over the length of the simulation. CA, alpha carbons.

To provide supporting evidence for the structural data, we performed single-channel electrophysiology on CymA proteins. Wild-type CymA forms ion-permeable channels with an average conductance of 1.16 nS in solvent-free membranes (Fig. 4). The conductance states and the open probabilities are dependent on the magnitude and polarity of the applied voltage. At low voltages, the channel is predominantly in a nonconducting state (Fig. 4A). Because OM Donnan potentials are low to nonexistent under physiological conditions [<30 mV (9)], the CymA channel is likely to be mostly closed in vivo. With increasing positive voltages, CymA adopts a stable open state whereas at negative voltages the channel displays frequent channel closures even at high potential differences (Fig. 4). Taken together with the crystal structures, we propose that the open state results from a voltage-induced displacement of the N terminus from the lumen of the barrel. To confirm this notion, we constructed and purified an N-terminal truncation mutant that has the first 15 residues removed (ΔCymA) (Methods). Single-channel experiments demonstrate that the mutant channel is completely open at positive and negative voltages (Fig. 4E). Moreover, ΔCymA is silent, indicating that the extracellular loops are rigid and do not fold inward to transiently close the channel. These results confirm previous electrophysiology data, including the suggestion that the channel “contains a section, which is mobile and can block the channel at least partially” (9). Our data show that this assumption is correct and that the mobile element corresponds to the N terminus of CymA moving into and out of the barrel lumen. The final protein that we tested was the R5A site-directed mutant. Based on the crystal structure of WT CymA (Fig. 3B and Fig. S2), Arg5 is likely to play an important role in constraining the N terminus inside the lumen of the barrel by interacting with several side-chain carboxyl groups in the barrel wall. The substitution of Arg5 with alanine should abolish those interactions. The electrophysiology data for the R5A mutant protein indeed show that even at low voltages the channel is open almost permanently (Fig. 4F), confirming the important role of Arg5 in constraining the N terminus.

Fig. 4.

Single-channel recordings of CymA show dynamics of the N terminus. (A–D) Traces are shown for the wild-type channel at (A) +25 mV and −25 mV, (B) +50 mV and −50 mV, (C) +100 mV and −100 mV, and (D) +150 mV and −150 mV, Left and Right, respectively. (E and F) Conductance traces for ΔCymA at +100 mV (Left) and −100 mV (Right) in E, whereas F shows traces for the R5A mutant protein at +50 mV (Left) and −50 mV (Right). The all-point histograms are shown in Insets and represent quantitation of the different conductance states of the channel.

CymA Contains Two Binding Sites for Cyclodextrins.

We next sought to characterize the interaction of CD substrates with the CymA channel. For this, we soaked α- and β-CD into pregrown His-tagged CymA crystals that diffracted to high resolution. The structure for the complex of CymA and α-CD (1.7 Å resolution) does not show any density for the N terminus and the cloning region. Instead, well-defined density is present inside the lumen of the barrel, indicating two binding sites for the cyclic oligosaccharide (Fig. 5 and Fig. S4). One site is formed by the extracellular mouth of the channel and likely represents the initial CD binding site. We will refer to this site as the “entry site.” The second binding site is ∼30 Å away and at a position close to where the center of the OM bilayer would be (Fig. 5). Beyond this position the channel widens toward the periplasmic space, and we will therefore designate the second binding site the “exit site.” Strikingly, the orientation of the α-CD molecule is completely different in both sites. In the entry site, the plane of the CD ring is approximately parallel to the membrane surface. By contrast, the CD ring in the exit site has rotated almost 90° and is now almost perpendicular to the membrane plane (Fig. 5). Thus, during diffusion through the channel the substrate molecule reorients itself.

Fig. 5.

Cyclodextrins bind at two positions inside the CymA channel. (A) View of the interior of the CymA channel complexed with α-CD (PDB ID code 4D5B), shown as a space-filling model with carbons colored blue and oxygens colored red. (B) View from the extracellular side, highlighting the similar dimensions of α-CD and the channel. (C) Cocrystal structure of CymA with β-CD (PDB ID code 4D5D; carbons are shown in yellow). In this structure, an α-CD molecule is bound in the entry site.

Fig. S4.

CD binding sites in the complex of CymA and β-CD (PDB ID code 4D5D). (A and B) Stereo diagrams of 2F_o − F_c electron density contoured at 1.0 σ for the entry site (A) and the exit site (B), showing the α-CD and β-CD molecules, respectively. Aromatic residues making stacking interactions with sugars are shown, as well as the glutamic acid residues contacting the substrate within the entry site. (C) Stereo view of a superposition of the β-CD cocrystal structure with that of α-CD (PDB ID code 4D5B), showing the differences in binding between the exit-site α-CD and the β-CD molecule.

An additional interesting feature of the CymA–CD cocrystal structures is the presence of tubular density inside the CD cavities (Fig. S4). These densities most likely correspond to partially ordered C₈E₄ molecules used for purification and crystallization, and demonstrate the well-established property of CDs to form complexes with hydrophobic compounds (10). It may therefore be possible to use CymA as a means of delivery of useful hydrophobic molecules in synthetic biology or biotechnological applications involving gram-negative bacteria.

CD Binding to CymA Results in Only Minor Conformational Changes.

Given the large size of the CD molecules, it might be expected that substantial conformational changes in CymA would be required for sugar binding. However, the structures show that for both binding sites the structural changes in the channel upon substrate binding are very small. Within the entry site, just three amino acid side chains undergo conformational changes of more than 1 Å: Asn40 in loop L1, Glu80 in loop L2, and Trp303 in loop L7 (Fig. S5A). The latter side chain reorients itself to form stacking interactions of its aromatic ring with a sugar moiety, as commonly observed in sugar-binding and transport proteins (11, 12). Additional stacking interactions with the entry site-bound CD are mediated by Tyr84 and Trp169. In addition to those interactions, numerous hydrogen bonds are present between the entry-site CD and polar and charged side chains of CymA. Compared with the entry site, the exit-site α-CD molecule makes fewer interactions with residues in the barrel wall due to its orientation and off-center position inside the channel. The structural changes in the exit site upon substrate binding are also very small and restricted to the side chains of residues E113, E275, and Q319. Only one aromatic residue (Tyr154) makes stacking interactions with sugar moieties (Fig. S5B).

Fig. S5.

CD binding causes minor conformational changes in CymA. Views from the extracellular side (A) and from the periplasmic space (B and C) for the superposition of apo-CymA (PDB ID code 4V3H; gray) and the complex of CymA and α-CD (PDB ID code 4D5B; rainbow) and that of CymA and β-CD (PDB ID code 4D5D; rainbow). The aromatic residues Y154 and Y184 as well as the residues that change conformation (>1 Å) upon CD binding are shown. The CD molecules are colored red.

CD Binding by CymA Is Unidirectional.

The CymA structure complexed to α-CD suggests that the extracellular mouth of the channel has evolved to bind the substrate with high affinity. We next asked whether we could obtain evidence for directionality of substrate binding by single-channel electrophysiology. When substrate is added to wild-type CymA from the cis side, no significant current blockages are observed. By contrast, addition of substrate from the trans side blocks the channel very efficiently with a binding constant of 34,000 ± 2,700 M⁻¹ (Fig. 6C), confirming the high affinity of CymA for α-CD from previous multichannel titrations (13), and suggesting that the trans side corresponds to the extracellular milieu. Even though the channel is fully open at the applied voltage (+100 mV; Fig. 4), it could be argued that the presence of the mobile N terminus could hamper substrate binding from the periplasmic side. To test this notion, we also performed substrate addition experiments for ΔCymA. Remarkably, the dependence of substrate binding on the orientation of the channel is also observed for the N-terminal deletion mutant, with efficient channel blockage only observed from the trans side (Fig. 6; binding constant 35,000 ± 1,300 M⁻¹). These observations indicate that the substrate only very inefficiently enters the channel from the periplasmic space, which in turn suggests that the exit-site CD molecule observed in our cocrystal structure has diffused there from the entry site. Furthermore, the affinities of α-CD for the N-terminal deletion variant and the wild-type protein are very similar, indicating that the N terminus does not play a role in substrate binding.

Fig. 6.

Directionality of CD binding to CymA probed by electrophysiology. (A–C) Single-channel conductance traces at +100 mV applied voltage for wild-type CymA (Left) and ΔCymA (Right) in the absence of α-CD (A) and in the presence of 10 μM α-CD added to the cis side (B) or to the trans side of the channel (C). (D) Traces obtained in the presence of 30 μM β-CD added to the trans side. All-point histograms are shown in Insets and represent the different conductance states of the channel.

The closely matching dimensions of the CymA channel and α-CD (Fig. 5) raise the question of how the larger β-CD molecule traverses the channel, and we therefore also determined a cocrystal structure of CymA with β-CD. As for α-CD, electron density for a large cyclic oligosaccharide is present in two sites. Surprisingly, the density in the entry site corresponds to α-CD whereas the density in the exit site can be assigned unambiguously to β-CD (Fig. 5 and Fig. S4). The β-CD molecule has a similar orientation as the exit-site α-CD, with the plane of the ring being approximately perpendicular to the OM plane. Although the orientation is similar, the β-CD molecule is shifted ∼4–5 Å toward the periplasmic space compared with α-CD and also occupies a more central position inside the CymA channel (Fig. S4C). As is the case for α-CD, the binding of β-CD introduces only small conformational changes in a few residues lining the channel (Fig. S5C). The fact that α-CD is bound in the entry site must be due to a contamination in the β-CD preparation coupled to a much higher affinity of α-CD for the entry site relative to β-CD. Indeed, single-channel measurements clearly show the lower affinity of β-CD for CymA (Fig. 6D), with a binding constant of 1,600 ± 200 M⁻¹ compared with 34,000 ± 2,700 for α-CD. We hypothesize that the size of β-CD makes it unlikely to bind in the entry site in a similar way as α-CD, that is, with the plane of the ring parallel to the OM plane. In addition, the CymA channel is likely too narrow for transport of γ-CD (17 Å outer diameter), which could explain why K. oxytoca does not grow on this compound (3). It appears that CymA has evolved for transport of α-CD, and it would therefore be interesting to determine whether α-CD predominates as a carbon source in vivo.

A Dynamic View of CD Transport.

To complement the crystallographic snapshots, we performed unbiased MD simulations of α-CD bound to the entry and exit sites of CymA, starting from the crystal structure coordinates. For the entry site, we observe stable binding of the α-CD molecule only when the three acidic residues (Glu80, Glu85, and Glu205) that contact the sugar are protonated. When those residues are charged, the α-CD molecule is clearly unstable and assumes a wide range of orientations during the simulation (Fig. S6). The calculated pK_a values for the glutamic acid residues are elevated by ∼1 pH unit in the presence of the bound substrate, but are still well below the crystallization and soaking pH of 7.5 (Methods). Further studies are required to shed light on the possible role of binding-site protonation in substrate binding and release. By contrast, the α-CD molecule in the exit site is stably bound for the duration of the simulation regardless of the protonation states of the two acidic molecules (Glu113 and Glu275) that contact the substrate.

Fig. S6.

Unbiased MD simulations (50 ns each, performed twice) for the α-CD molecules bound in the entry site (A) and exit site (B) of CymA (PDB ID code 4D5B), showing the orientation (angle) of the substrate and its position within the channel. The simulations were performed with charged (Left) and protonated (Right) carboxylate groups contacting the substrate.

We next investigated whether the substrate reorientation during transport, inferred from the crystal structures, could be recapitulated in a steered MD simulation. To simplify our modeling, we used the coordinates of the cocrystal structure of CymA with α-CD (PDB ID code 4D5B); that is, the N terminus is absent during the simulation. The α-CD molecule was placed ∼40 Å from the extracellular mouth of the pore and was pulled with constant speed toward the periplasmic space (Methods). The results of the simulations indeed match the cocrystal structures, although we emphasize that the steered MD simulations should only be viewed as a qualitative indication of the translocation pathway. The substrate binds to the entry site with the plane of the CD ring almost parallel to the membrane plane (∼160° angle). Upon release from the entry site the substrate tilts immediately (∼90° angle) and remains in this orientation until it reaches the periplasmic space (Fig. 7 and Movie S1). We speculate that the rotation of the cyclic sugar during diffusion serves to lower its affinity for the channel, which would prevent substrate from stalling before exiting into the periplasmic space.

Fig. 7.

Orientation of α-CD and β-CD during translocation. Plots obtained from representative steered MD simulations showing the tilt angles of the CD molecules during transport through the CymA channel. The red and green crosses represent the orientations and positions of the α-CD and β-CD molecules in the respective crystal structures (PDB ID codes 4D5B and 4D5D). The error bars represent the standard deviation of the orientation calculated over a bin size of 2 Å along the channel axis (n = 3 simulations).

We also carried out steered MD simulations for β-CD to investigate whether this compound can translocate through the CymA channel, as implied by growth data (4). The force profiles suggest that whereas the average forces required for pulling β-CD through the channel are somewhat higher than those for α-CD, the maximum forces are similar (8–10 kcal⋅mol⁻¹⋅Å⁻¹) (Fig. S7). This suggests that β-CD can traverse the channel but with lower efficiency compared with α-CD. Compared with α-CD, β-CD appears to pass through CymA in a more restricted manner and close to its orientation seen in the crystal, with the plane of the ring approximately perpendicular to the OM plane (Fig. 7). This result makes sense given the larger size of β-CD, and reinforces our notion that β-CD likely does not bind to the entry site.

Fig. S7.

Force profiles from constant-velocity steered MD simulations of α-CD and β-CD molecules along the CymA pore axis. Each CD molecule was placed on the extracellular (EC) side of the pore and pulled along the permeation pathway toward the periplasmic (PP) side of the pore. The profiles shown are averages of three simulations, two in the forward direction (EC to PP) and one in the reverse direction (PP to EC).

Discussion

The most interesting part of the CymA structure is the N terminus, which forms a mobile element that can move in and out of the channel lumen. What is the role of the N terminus, given that our data indicate it is dispensable for substrate binding (Fig. 6)? By constricting the channel, the N-terminal ∼15 residues likely preserve the permeability of the OM, which might otherwise be compromised. Constriction elements composed of extracellular loops that fold inward to decrease the effective diameter of the barrel are found in OM channels composed of 16 or more β-strands such as OmpF (Fig. S1) (1). Those loops tend to be very stable because they make many interactions with the interior surfaces of the barrel or with other loops. In the case of CymA, a relatively short segment (the N terminus) inserts into the channel from the periplasmic side and interacts with the barrel wall. To gain insight into the energetics of this interaction, we performed a one-dimensional potential of mean force calculation of the N terminus in wild-type CymA using metadynamics (14). The result shows that the position of the N terminus in the crystal corresponds to a free energy minimum (Fig. S8). Moreover, movement of the N terminus into the periplasmic space requires passing several energy barriers. Thus, to generate an open channel, the interactions of the N terminus with the barrel need to be disturbed. Considering the fact that only a limited number of N-terminal residues interact with the barrel wall (Fig. 3B), we propose that the CD substrate moving down from the entry site could provide the driving force for channel opening, possibly by engaging the acidic residues of the channel wall that interact with the important residue Arg5 (Fig. 8). CymA can therefore be considered a ligand-gated OM diffusion channel. Future experiments and more detailed molecular dynamics simulations will be required to obtain insights into the displacement mechanism of the N terminus by CDs and whether other molecules can open the channel.

Fig. S8.

One-dimensional free energy profiles (PMF) of the N terminus along the channel axis determined using well-tempered metadynamics (14) implemented in PLUMED 2 (30). The reaction coordinate distance is defined as the COM (center of mass) difference between the N terminus (C_α atoms of residues 1–13) and the barrel (C_α atoms of residues 26–324) along the channel axis. Three representative snapshots of the protein are shown, including the crystallized state of the N terminus corresponding to the lowest energy state.

Fig. 8.

Schematic model for cyclodextrin transport by CymA. (i) The N terminus of CymA (magenta) is located inside the barrel lumen. Arg5 in the N terminus (+) interacts with carboxylate groups of glutamic acid residues (−) in the channel wall. (ii) After substrate capture by the entry site, the CD molecule rotates to lower its affinity for the channel and diffuses further. The N terminus is expelled from the barrel by perturbation of the interactions of the N terminus with the barrel wall. (iii) Upon expulsion of the N terminus, the CD molecule diffuses into the periplasmic space.

OM diffusion channels that allow permeation of molecules larger than the OM size-exclusion limit have been identified previously. In all cases, however, the substrates have structures allowing them to pass through the channel in a linear fashion, for example, maltooligosaccharide passage through E. coli LamB (11). Our study now shows that bulky molecules such as CDs can also enter cells efficiently through OM diffusion channels. This then poses the question as to why cells spend PMF-derived energy to take up bulky substrates via TonB-dependent transporters. As CymA illustrates, substrate size is not the decisive factor that necessitates active transport. Rather, it is likely the need to bind substrate very tightly because it is scarce and/or valuable. The external energy input is required to generate the large conformational changes necessary for substrate release. In the case of TBDTs, substrate release is coupled to formation of a transient channel into the periplasmic space. By contrast, the mobile N-terminal constriction domain of CymA is likely displaced by the incoming substrate, providing an elegant way for cells to take up bulky substrates without compromising the permeability barrier of the OM.

Methods

Cloning, Overexpression, and Purification of His-Tagged CymA.

The mature part of the cymA gene from K. oxytoca was synthesized (MWG Genomics) and cloned into the pB22 expression vector under control of the arabinose promoter. After cleavage by signal peptidase, the N-terminal sequence of the protein is as follows: (-14)ANVRLQHHHHHHHLE(0)-CymA. Protein expression was performed in C43 ΔcyoABCD cells by growing in LB medium until OD₆₀₀ ∼0.6 at 37 °C, followed by induction with 0.1% arabinose for 16 h at 20 °C (final OD ∼1.2–1.5). Cells were harvested by centrifugation (10,000 × g for 15 min), resuspended in TSB buffer (20 mM Tris, 300 mM NaCl, pH 8), and broken via one pass through a cell disrupter operated at 23 kpsi (Constant Systems). Total membranes and cell debris were collected by ultracentrifugation at 42,000 rpm for 45 min (45Ti rotor; Beckman). Membranes were homogenized with 3% (wt/vol) Elugent (Calbiochem) in TSB (100 mL for 6 L culture) followed by stirring at 4 °C overnight. After ultracentrifugation (30 min, 42 krpm; 45Ti rotor), the supernatant was loaded onto a 10-mL nickel column equilibrated in TSB + 0.2% lauryldimethylamine N-oxide (LDAO) (chelating Sepharose; GE Healthcare). The column was washed with 15 column volumes (CVs) of buffer with 25 mM imidazole and the protein was eluted with 3 CVs of buffer + 250 mM imidazole. After concentration (50 kDa cutoff; Millipore), the protein was applied to a Superdex 200 16/60 gel filtration column (GE Healthcare) equilibrated in 10 mM Hepes, 100 mM NaCl, 0.05% LDAO (pH 7.5). For polishing and buffer exchange before crystallization, a second gel filtration column was run in 10 mM Hepes, 100 mM LiCl, 0.35–0.4% C₈E₄ (pH 7.5). The protein was concentrated to 10–15 mg/mL, flash-frozen in liquid nitrogen, and stored at −80 °C. The yield of purified CymA was ∼4 mg/6 L rich medium. For production of selenomethionine-substituted protein, C43 ΔcyoABCD cells were grown at 37 °C in LeMaster–Richards (LR) minimal medium with 0.25% glycerol as the carbon source, using the methionine biosynthesis inhibition method (15). Approximately 20 min before induction, amino acids were added (K/T/F, 100 mg/L; L/I/V, 50 mg/L) as well as SeMet (60 mg/L). Induction was carried out by adding 0.5% arabinose, followed by 16 h of growth at 30 °C. The final OD₆₀₀ was ∼1.0. The SeMet-substituted protein was purified as described above for the native protein; in this case, the yield was ∼2.5 mg/10 L culture. For production of non–His-tagged (wild-type) CymA, the porin-deficient E. coli BL21 omp8 strain was transformed with the pUC18-derived pCYMA plasmid, which expresses CymA constitutively (5). Four liters of cells was grown in LB medium at 37 °C throughout; curiously, growth at lower temperatures did not result in any expression. For CymA purification, the total membrane pellet was extracted twice with 100 mL 0.5% sarkosyl in 20 mM Hepes (pH 7.5) to selectively remove inner-membrane proteins. The OM-enriched pellet was extracted with 1% LDAO in 10 mM Hepes, 50 mM NaCl (pH 7.5) by homogenization and stirring for 2 h at 4 °C. After ultracentrifugation (30 min, 42 krpm; 45Ti rotor), the supernatant was loaded onto 10 mL Q Sepharose Fast Flow columns equilibrated in 0.2% LDAO, washed with LDAO buffer containing 100 mM NaCl, and eluted with 500 mM NaCl. After concentration the protein was purified on Superdex 200 26/60 in 10 mM Tris, 100 mM NaCl, 0.05% LDAO (pH 8), followed by a linear salt gradient on Resource Q (6 mL) at pH 8. Appropriate fractions were pooled and run on a final Superdex S-200 16/60 gel filtration column in 10 mM Hepes, 100 mM LiCl, 0.35–0.4% C₈E₄ (pH 7.5). The protein was concentrated to 8 mg/mL, flash-frozen in liquid nitrogen, and stored at −80 °C. The yield of purified native CymA was ∼2 mg/4 L medium.

Construction, Expression, and Purification of ΔCymA.

pCYMA was amplified using Phusion DNA polymerase (Thermo Scientific) and the oligonucleotide forward primer (5′-GAAAGTTTTTTTTCGTTTGGTGGCCAT-3′) and reverse primer (5′-TGCAAATGAATGTACGGGCGCTGTGA-3′). The PCR product was digested with DpnI, phosphorylated using T4 polynucleotide kinase (Thermo Scientific), and ligated using T4 DNA ligase (Thermo Scientific). The ligation product was transformed into DH5α cells by electroporation. ΔCymA was expressed and purified as published previously for wild-type CymA with slight modifications (5). Cells were grown at 37 °C overnight to an OD₆₀₀ of 0.8–1.1 in LB medium, harvested by centrifugation (10,000 × g for 15 min), and resuspended in 50 mM potassium phosphate with protease inhibitor (1 mM PMSF) at pH 7.5 (buffer A). Three to five passages by French press were used to lyse the cells at 16,000 psi. Intact cells were removed by centrifugation at 6,000 × g for 1 h. Ultracentrifugation was done at 100,000 × g for 90 min to pellet cell envelopes. Buffer A with 0.5% sarkosyl was used to solubilize inner-membrane components, and stirred for 1 h at room temperature followed by centrifugation at 100,000 × g for 90 min. This preextraction procedure was repeated once. The pellet was washed in buffer A and resuspended in buffer A with 100 mg/mL lysozyme, 3 mM NaN₃ and the mixture was left for overnight stirring at 37 °C. The solution was centrifuged at 100,000 × g for 90 min and the OM was solubilized by resuspending in buffer A with 5 mM EDTA and 5% (wt/vol) octylpolyoxyethylene (OPOE; Bachem Biochemica) and stirred at room temperature for 1 h. After centrifugation (100,000 × g for 90 min), the supernatant was dialyzed against 10 mM potassium phosphate (pH 7.5) containing 0.6% OPOE (buffer B) and loaded onto a Mono Q-HR 5/5 column equilibrated in buffer B. ΔCymA was eluted with buffer B containing 1 M NaCl using a linear gradient to 1 M NaCl, concentrated, and flash-frozen for storage at −80 °C.

Single-Channel Electrophysiology.

Solvent-free membranes were made according to Montal and Mueller (16). Briefly, a Teflon cuvette with two symmetric chambers partitioned by a 25-μm-thick Teflon film with an aperture size of ∼50–100 μm was used for reconstitution of protein. The aperture was impregnated with 1% hexadecane in n-hexane to make it more hydrophobic. The aqueous phase was buffered with 10 mM MES at pH 6. The membrane was formed using a solution of 5 mg/mL diphytanoyl phosphatidylcholine (DPhPC) in n-pentane. Ag/AgCl electrodes (World Precision Instruments) were used to measure the electric current. One of the electrodes was connected to the ground (cis side of the membrane). The other electrode was connected to the head stage of an Axopatch 200B amplifier (Axon Instruments). Purified porin in detergent solution was added to the cis side of the membrane. The conductance measurements were performed by an Axopatch 200B amplifier in voltage-clamp mode and digitized by an Axon Digidata 1440A digitizer and controlled by Clampex software (Axon Instruments). The current traces were filtered by low-pass Bessel filter at 10 kHz and recorded with a sampling frequency of 50 kHz. Data were analyzed by the Clampfit program (Molecular Devices).

For evaluation of the binding constants, substrate was added to the extracellular side/periplasmic side in the specified concentration. The applied potential (100 mV) was measured between the periplasmic side (cis, corresponding to the electrical ground and to the side of protein addition) and the extracellular side of the channel. Single-channel analysis was performed to determine the association (k_on) and dissociation (k_off) rates (17). The ratio k_on/k_off = K relates to the binding constant K (1/M). The time for which the substrate is blocking the channel is called the average residence time (τ). At low substrate concentration, where [c] << K, τ is given by τ = 1/k_off and k_on = (no. of binding events)/[c].

Crystallization and Structure Determination.

Initial vapor diffusion crystallization trials were set up using a Mosquito crystallization robot (TTP Labtech) with sitting drops at 20 °C with in-house screens as well as the commercial MemGold 1, MemGold 2, and Morpheus screens (Molecular Dimensions). If necessary, crystal optimization was performed using larger-scale sitting drop or hanging drops. The SeMet crystal of His-tagged CymA used for data collection grew in space group P2₁ from Morpheus 1–35 [0.09 M nitrate, phosphate, sulphate (NPS), 0.1 M buffer system 3, pH 8.5, 20% (vol/vol) glycerol, 10% (vol/vol) PEG 4K]. The high-resolution His-tagged CymA dataset grew from MemGold 1, 2–13 [0.1 M Tris, pH 8.0, 0.3 M magnesium nitrate hexahydrate, 23% (vol/vol) PEG 2K] (space group P2₁), whereas the tagless wild-type protein crystals were obtained from ∼35% (vol/vol) PEG 400, 0.4 M sodium bromide, 50 mM MES (pH 6) or 50 mM glycine (pH 9) (space group C222₁). The His-tagged protein crystals used for CD soaking grew in Morpheus 1–30 [0.09 M NPS, 0.1 M buffer system 2, pH 7.5, 20% (vol/vol) ethylene glycol, 10% (vol/vol) PEG 8K] (space group P2₁2₁2). From substrate-induced channel blockages in multichannel electrophysiology experiments, saturation constants were previously estimated to be ∼30 μM for α-CD and 0.5 mM for β-CD (13). Based on these data, solutions of α- and β-CD (Sigma) were made in water at 100 mM and 50 mM, respectively, and diluted 10-fold into the crystallization hit solution containing 0.5% C₈E₄. This was then added in a 1:1 ratio to the crystal drop, giving final concentrations of 5 mM and 2.5 mM (β-CD), and incubated for 48 h at 20 °C. Cocrystallization experiments with α- and β-CD produced only small crystals.

Initially, we attempted molecular replacement for the native high-resolution CymA dataset using the distantly related OmpG OM channel (∼15% sequence identity to CymA) as a search model, but we were not successful. We then generated crystals of SeMet-substituted protein and collected three single anomalous dispersion datasets at the peak wavelength on a single, bar-shaped crystal. The datasets were processed using XDS (18) or HKL2000 (19) and merged to obtain a single, high-redundancy dataset. All four selenium sites within the asymmetric unit (AU) were found using SOLVE (20), and RESOLVE AUTOBUILD (20) was used to generate a partial model. The best-defined monomer of the AU was extended manually within Coot (21) and used as a search model for molecular replacement (MR) of the high-resolution native data using Phaser (22). A single, clear solution was found that was used for RESOLVE autobuilding, resulting in a model that was ∼80% complete. Several rounds of Phenix refinement (23) and manual model extension were subsequently carried out to arrive at a final model with good statistics (R/R_free = 17.4/20.7%; data to 1.8 Å resolution). This model was used to solve the other CymA datasets via MR. Data collection and refinement statistics are summarized in Table S1.

Table S1.

Data collection and refinement statistics for CymA

	NHis-SeMet	NHis-native	Wild type	α-CD soak	β-CD soak
PDB ID code	4V3G	4V3H	4D51	4D5B	4D5D
Data collection
Beamline	DLS i02	DLS i02	DLS i04	DLS i02	DLS i02
Wavelength, Å	0.9796	0.9796	0.9795	0.9796	0.9796
Space group	P21	P21	C2221	P21212	P21212
Cell dimensions
a b c	42.6 105.2 111.8	61.2 61.4 108.3	59.6 228.1 60.6	140.5 77.4 110.7	139.6 77.8 109.5
α β γ	90 97.9 90	90 91.4 90	90 90 90	90 90 90	90 90 90
Molecules/AU	2	2	1	2	2
Solvent content, %	62	52	53	68	68
Resolution, Å	49.1–2.5	53.4–1.8	60.6–2.1	55.4–1.7	47.1–1.9
Completeness	99.4 (99.7)	96.5 (93.1)	97.9 (86.9)	100 (100)	100 (100)
Redundancy	20.4 (21.8)	3.4 (3.2)	6.1 (3.1)	7.4 (7.4)	9.3 (9.4)
Rmerge, %	8.1 (38)	6.5 (72)	5.8 (38)	6.7 (82)	7.3 (68)
Rpim, %*	1.8 (8.2)	4.2 (47)	2.4 (23)	2.7 (32)	2.5 (23)
CC (1/2)†	0.99 (0.99)	0.99 (0.67)	0.99 (0.91)	0.99 (0.78)	0.99 (0.91)
Phasing
Sites found (expected)	4 (4)	—	—	—	—
SOLVE FOM, Å	0.29 (20–3)
Refinement
Resolution, Å	49.1–2.5	43.4–1.8	53.5–2.3	47.1–1.7	47.1–1.9
Reflections, n	30,921	67,632	18,792	251,353	171,382
Rwork/Rfree, %‡	19.2/25.3	17.4/20.7	21.4/27.9	16.7/19.0	17.9/19.8
Atoms, n
Protein/solvent	5,465/52	5,241/556	2,605/44	5,242/785	5,205/576
Ligand/detergent	—/122	4/191	—/9	264/366	286/247
B factors, Å2
Protein/solvent	64/58	28/40	56/55	26/44	34/49
Ligand/detergent	—/73	33/52	—/60	39/54	54/57
Rmsd
Bond lengths, Å	0.009	0.006	0.008	0.007	0.0073
Bond angles, °	1.16	1.08	1.07	1.13	1.16
Ramachandran plot, %
Most favored/disallowed	97.1/0.5	97.9/0.3	96.8/0.0	98.7/0.2	98.9/0.0
MolProbity clashscore§	6.1	5.0	4.6	5.1	4.3

Values in parentheses refer to the highest-resolution shell. DLS, Diamond Light Source; FOM, figure of merit. NHis, N-terminally His-tagged CymA.

^*As defined by Weiss (31).

^†CC (1/2) is the correlation coefficient between two random half-datasets (32).

^‡R_free was computed as for R_work using a test set (2–6%) of randomly selected reflections that were omitted from the refinement.

^§As defined by Molprobity (33).

Molecular Dynamics Simulations.

To study the structure and stability of the CymA channel and especially of the N terminus inside the pore, CymA (PDB ID code 4D51) was simulated in a POPC (1-palmitoyl,2-oleoyl-sn-glycero-3-phosphocholine) bilayer consisting of 230 lipids per unit cell. The missing residues (Glu10–Phe21) of the CymA N terminus were predicted using Modeler 9.12 (24). The system was solvated using the TIP3P water model and neutralized with six potassium ions, leading to a total atom number of 91,342. Subsequently, an energy minimization was performed followed by a 10-ns equilibration in a canonical ensemble at 300 K with position restraints on the protein and lipids. In a further step, the position restraints were removed and the system was equilibrated for 100 ns at 1 bar. All MD simulations were performed using the GROMACS 4.6.5 package (25) using the standard CHARMM36 force field (26). The cutoff for short-range electrostatics and van der Waals interactions was set to 1.2 nm, and long-range electrostatics was treated using the particle-mesh Ewald (PME) method with grid size 0.1 nm. Moreover, all bonds were constrained using the LINCS method (27) to enable a time step of 2 fs.

In a second set of simulations, four different systems including α-CDs were built from the corresponding crystal structure (PDB ID code 4D5B). To this end, the N-terminal residues (Ala1–Ala13) were assumed to be outside the channel, unlike the system described above, which has an inwardly oriented N terminus. Two separate systems were built, containing an α-CD either at the entry or exit binding site. Initially all of the ionizable residues of the protein were kept in the charged state. However, the α-CD molecules were not very stably bound, especially at the entry site. Therefore, we performed pK_a calculations using PROPKA (28) in the absence and presence of α-CD. The output suggested that there is a change in the pK_a values by about 1 pH unit in the presence of α-CD mainly for the three glutamate residues (Glu80, Glu85, and Glu205) at the entry site and two glutamate residues (Glu113 and Glu275) at the exit site. Based on these findings, two more systems were built with the above sets of the glutamate residues in their protonated state and the remaining ionizable residues unchanged. The equilibration procedure described above was followed by unbiased simulations of two times 50 ns for each of the four system variants.

In addition, constant-velocity steered molecular dynamics was performed to obtain an initial, qualitative idea of the translocation of α-CD through the pore. For this simulation, the abovementioned five glutamate residues were protonated. Initially, the α-CD was placed outside the pore on the extracellular side and steered with a pulling velocity of 1 Å/ns and a spring constant of 100 kJ⋅mol⁻¹⋅nm⁻².