YidC and SecYEG form a heterotetrameric protein translocation channel

The heterotrimeric SecYEG complex cooperates with YidC to facilitate membrane protein insertion by an unknown mechanism. Here we show that YidC contacts the interior of the SecY channel resulting in a ligand-activated and voltage-dependent complex with distinct ion channel characteristics. The SecYEG pore diameter decreases from 8 Å to only 5 Å for the YidC-SecYEG pore, indicating a reduction in channel cross-section by YidC intercalation. In the presence of a substrate, YidC relocates to the rim of the pore as indicated by increased pore diameter and loss of YidC crosslinks to the channel interior. Changing the surface charge of the pore by incorporating YidC into the channel wall increases the anion selectivity, and the accompanying change in wall hydrophobicity is liable to alter the partition of helices from the pore into the membrane. This could explain how the exit of transmembrane domains from the SecY channel is facilitated by YidC.

ATP-independent steps during protein transport 22 . YajC is a small, single-spanning membrane protein encoded upstream of SecD in the yajC-secDF operon. Its expression level is tenfold higher than that of SecDF 23 , which might explain the presence of YajC complexes that do not contain SecDF 9, 24 . SecDFYajC was cross-linked to SecG  63 ) and the SecY channel interior and pore ring (right panel). ResiduesI91, L127, I191, I278 and Y332 where pBpa was inserted are indicated in red. The lateral gate is indicated in blue and the residues used for crosslinking are indicated. Residue F383 is at the back of the SecYEG complex and not visible in this front view. (b) In vivo crosslinking of Bl21 (wt) cells expressing SecY, either without pBpa insertion (SecY) or with pBpa inserted at the indicated positions. Crosslinking was induced by UV exposure of whole cells when indicated. SecY and SecY crosslinking products were purified after cell breakage and separated on SDS-PAGE. SecY-YidC crosslinking products were detected after western blotting using α-YidC antibodies. Indicated is the SecY-YidC cross-linking product (upper panel) and the SecY content in these cells, as revealed by α-SecY antibodies (lower panel). (*) indicates a weak UV-dependent SecY-YidC crosslinking product of SecY(I191pBpa). (c) In vivo crosslinking was performed and analyzed as in (b) with pBpa inserted at the indicated residues. (c) In vitro crosslinking with purified inner membrane vesicles (INV) (2.5 nM SecY) derived from BL21 cells expressing the indicated pBpacontaining SecY derivatives. Samples were processed as above. Crosslinking experiments were repeated at least three times and representative blots are shown. in a complex consisting of SecYEG, YidC and SecDFYajC 25 , supporting an earlier observation of a functional SecDFYajC-SecG interaction 26 .
SecDFYajC is possibly involved in tethering YidC to SecYEG. This was deduced from the observation that co-purification of YidC with SecYEG was detected upon SecDFYajC overexpression 10 and that SecDF forms a complex with YidC 27 . Nonetheless, the interaction between YidC and the lateral gate of SecY is not significantly influenced by the absence of SecDF 20 .
In the current study, we uncovered that YidC not only interacts with the surface of SecY's lateral gate, but that it also contacts the channel interior of SecY. Electrophysiology analyses revealed that YidC directly influences the channel properties of the SecYEG channel by altering the channel cross-section and anion selectivity. Our data demonstrate that upon complexation with SecYEG, YidC becomes part of the protein-conducting channel.

Results
YidC is in contact with the SecY channel interior. YidC's position in front of the lateral gate of SecY 20 likely explains the observation that during insertion, membrane proteins were found at the SecY-YidC interface 12,19 . Crosslinking data indicate that the SecY-YidC contact is influenced by ribosome-nascent chains (RNCs), demonstrating that the SecY-YidC interaction is different in resting and active translocons 20 . For analyzing these conformational changes, we employed in vivo site-directed photo-crosslinking using the phenylalanine derivative para-benzoyl-L-phenylalanine (pBpa) 28 . pBpa was incorporated into residues I91 (TM2b), L127 (TM3), I191 (TM5), I278 (TM7), Y332 (TM8) and F383 (TM9) (Fig. 1a). I91 and L127 are located on the surface of the lateral gate and have been shown to contact YidC 20 . Residues I191 and I278 are part of the internal pore ring and thus located inside of the protein conducting channel (Fig. 1a). Y332 is also part of the lateral gate, but located on the other site compared to I91 and L127. Finally, residue F383 is located at the back of SecY and thus not in vicinity to the lateral gate.
UV-exposure of E. coli cells expressing pBpa-containing SecY derivatives induces the formation of a covalent bond between SecY and any protein in close proximity. We observed a strong crosslink between residue 91 and YidC at approx. 90 kDa (Fig. 1b), which was recognized by α-YidC antibodies and not present when cells expressing SecY without pBpa were UV-exposed. This crosslinking product had been observed before 20 and was identified by mass spectrometry as a SecY-YidC crosslinking product 20 . This confirms YidC's position close to the lateral gate. We also observed a crosslink between residue 278 and YidC, suggesting that YidC is not only in contact with the surface of the lateral gate, but that it actually extends into the SecY channel (Fig. 1b). We did not observe a strong crosslink to YidC from residue 191 in vivo, although we did detect a very weak UV-dependent band that was recognized by YidC antibodies (Fig. 1b, *). This band migrated slightly higher than the SecY-YidC crosslink products from positions 91 and 278. Irregular migration of SecY crosslink products on SDS-PAGE is commonly observed and probably reflect different folding states of the crosslink product 20,[29][30][31] . For position 332 of the lateral gate, we observed only a very weak crosslink product and position 383 at the back of SecY did not show any crosslink to YidC (Fig. 1c). Thus, our data confirm the contact between the lateral gate of SecY and YidC, but surprisingly also show that YidC is in contact with interior of the SecY channel. No crosslink was observed between the back of SecY and YidC, indicating that YidC is primarily in contact with the front of the SecYEG complex.
In vivo crosslinking is advantageous for monitoring protein-protein interactions in living cells, but since it is unsynchronized, only rather strong interactions are detected. We therefore also analyzed the interaction between YidC and the channel interior by in vitro crosslinking. Inner membrane vesicles (INV) of cells expressing pBpa-containing SecY derivatives were purified and subsequently UV exposed for activating pBpa. This approach confirmed the 90 kDa SecY-YidC crosslink product for residue 91 but also showed an additional weak crosslink product at approx. 70 kDa (Fig. 1d). This band was previously identified by mass spectrometry as additional SecY-YidC crosslink product 20 . The occurrence of two crosslink products either indicates that this SecY residue is in contact with two different regions of YidC. Such a position-dependent mobility of pBpa crosslink products on SDS-PAGE is frequently observed for SecY and is assumed to reflect different three-dimensional structures of the crosslink products 29,31 . Alternatively, the lower band could reflect a proteolytic cleavage product. In particular SecY is subject to proteolysis under in vitro conditions 32,33 , which would explain why the 70 kDa band is primarily observed under in vitro conditions.
In vitro, the 90 kDa and 70 kDa crosslink products were also observed for residues 278 and 191, although the latter showed only very weak crosslinks to YidC in vivo. As in vivo, the in vitro 90 kDa YidC crosslink product from residue 191 displayed slightly reduced mobility on SDS-PAGE.
The different crosslink pattern of SecY(I191pBpa) in vivo and in vitro could reflect the fact that in sucrose-gradient purified INV most of the SecYEG channels are in a closed or inactive state, while the in vivo crosslinking monitors crosslinking of both active and inactive SecYEG channels. Thus, the observation that SecY(I191pBpa) forms strong crosslinks to YidC only in vitro could indicate that access of YidC to the channel interior is restricted in the presence of a nascent chain. This was further analyzed by performing the in vitro crosslink experiment in the presence of purified FtsQ-RNCs. FtsQ is a single spanning membrane protein that is inserted into the E. coli membrane via the SecYEG-YidC pathway 10,34 . Upon UV exposure of INV containing either SecY(I191pBpa) or SecY(I278pBpa), the two crosslink products at 90 kDa and 70 kDa were detected by α-YidC antibodies (Fig. 2a) and also by α-SecY antibodies (Fig. 2b), although the specificity of the α-SecY antibodies was lower. When these INV were pre-incubated with FtsQ-RNCs before UV-exposure, the crosslink products became significantly weaker for I278 and were almost absent for I191 (Fig. 2). In this experimental setting, we observed weak UV-independent crosslink formation, which is probably the result of light exposure during sample preparation and incubation. The presence of FtsQ-RNCs in the assay was monitored for SecY(I191pBpa)-INV by western blot using antibodies against the N-terminal HA-tag of the FtsQ-RNCs. This confirmed the presence of both FtsQ and the FtsQ-tRNA species in the RNC-containing sample (Fig. 2c).
Although all samples contained comparable SecY amounts as verified by western blotting (see Figs 1 and 2), we noticed that in all UV-exposed samples slightly more YidC co-purified with SecY, compared to the non-exposed sample. Co-purification of YidC with SecY has been observed before 10,20 and supports the conclusion that both proteins interact in the E. coli membrane. The increased co-purification of YidC with SecY after UV exposure could reflect the observation that YidC exists as both monomer and dimer 15,27,35 in the E. coli membrane. Therefore a crosslink of SecY to one copy of the YidC dimer will likely increase the amount of YidC that is found co-purifying with SecY. When samples were pre-treated with 50 mM DTT, the amount of YidC co-purifying with SecY was comparable between the UV-treated and non-treated sample (Supplementary   S1). The reasons for this obviously redox-sensitive interaction between SecY and YidC upon UV-exposure were not further analyzed in the current study.
The observed interaction of YidC with the channel interior of SecY could reflect a YidC molecule in the process of being co-translationally inserted via SecY. The size of the crosslink product, however, suggests that SecY is crosslinked to full length YidC, which would not be expected if YidC-RNCs were in contact with SecY. These highly purified INV are furthermore almost free of RNCs 36 . In summary, these in vivo and in vitro data indicate that YidC not only contacts SecY at the surface of the lateral gate, but that it is also in contact with the channel interior until it is reoriented by the presence of a nascent membrane protein.

Channel properties of the YidC-SecYEG channel. Since a protrusion of YidC into the SecY channel is
likely to decrease the channel lumen, we measured the single channel conductance g of SecYEG-or SecYEG-YidC complexes. To perform the electrophysiology measurements we first co-reconstituted SecYEG and YidC into proteoliposomes and then formed protein-containing monolayers on top of the proteoliposome suspensions. Two of these monolayers were then folded into a planar bilayer within the aperture of a Teflon septum 37 . This method of protein reconstitution into planar bilayers does not require the SecYEG channels in the proteoliposomes to be in their active state -in contrast to the widely used assays of proteoliposome fusion to planar lipid bilayers 38 . It thus allowed us to show that the YidC-SecYEG complexes, reconstituted in a 1:1 ratio, were closed in their resting state (Fig. 3a, 2 nd black trace). We recently discovered that ribosomes are able to open the SecYEG channel at low membrane potentials 39 . Thus, the SecYEG channel is both ligand dependent and voltage activated. At physiological membrane voltages, i.e. at about −130 mV, the channel is closed and ligands are unable to open it 40 . Most of the ribosome-bound SecYEG channels open upon membrane depolarization, i.e. at half the physiological value of the membrane potential. Channel opening is reversible, i.e. the channels close again upon restoration of more negative potentials. Opening of the SecYEG channel by ribosomes was not observed with a SecYEG mutant complex, in which two conserved arginine residues within the cytosolic loop C4 were replaced by glutamate ( Supplementary Figs 2 and 3). This demonstrates that channel opening by ribosomes requires a specific SecY-ribosome interaction. The addition of ribosomes to just lipid bilayers did also not increase current, demonstrating that ribosomes are free of any pore-forming contaminant (Fig. 3a, 1 st black trace). In line with the recently discovered ability of ribosomes to open the SecYEG channel at low membrane potentials 39 , channel activity exclusively appeared upon ribosome addition. Figure 3a shows how three such ribosome-bound YidC-SecYEG channels close one after the other (blue trace). Co-reconstitution of YidC with a 3-fold excess of SecYEG yielded a mixture of larger SecYEG-and smaller SecYEG-YidC channels ( Fig. 3; lower black trace). We obtained the single channel conductivity g by plotting the single channel current as a function of the applied voltage ( Fig. 3b and Table 1). The g-value of ribosome-activated SecYEG channels was 1.6 ± 0.1 pS/mM, which is in perfect agreement with previously obtained results 39 . However, for the SecYEG-YidC complex we determined a significantly lower g-value of 0.9 ± 0.2 pS/mM, which is in line with a protrusion of YidC into the SecYEG lumen and verifies the crosslinking data between YidC and SecY-residue 191.
Crosslinking between SecY's residue 191 and YidC was only observed under in vitro conditions when the YidC-SecYEG complex was in its resting state. When actively engaged in protein transport, i.e. bound to translating RNCs, crosslinking to 191 was not observed, suggesting that YidC was expelled to the rim of the complex. This movement of YidC in the presence of RNCs should also be reflected by an increased g-value -provided the pore is not blocked by the nascent chain. In contrast to the experiments in Fig. 3, we applied a salt gradient across the membrane to test for ion selectivity of the complex, which would manifest itself in a reversal potential, ψ r . As a side effect, the salt gradient drives the fusion of proteoliposomes with planar bilayers prompting us to use these fusion events for reconstituting the SecYEG or the YidC-SecYEG complexes into the planar bilayers. Since fusion only occurs when the channels in the proteoliposomes conduct the osmolyte, we did not observe any channel activity in the absence of FtsQ-RNC (Fig. 4a, 2 nd black line). We did also not observe channel activity by just adding FtsQ-RNCs to bare bilayers (Fig. 4a, 1 st black line). In the presence of FtsQ-RNC, proteoliposome fusion served to insert open channels into the planar bilayer. Binding of FtsQ-RNCs or high concentrations of non-translating ribosomes opened the reconstituted SecYEG channels (Fig. 4a, black and red traces) and FtsQ-RNCs opened also the SecYEG-YidC channels (Fig. 4a, green trace). Under the conditions of the experiment, we did not observe channels that could be attributed to YidC alone. However, in complex with ribosomes or RNCs, YidC shows channel activity. These channels are distinct from both SecYEG and SecYEG-YidC channels, e.g. their single channel conductivity amounts to only 1.3 pS/mM when measured under the same conditions as the traces from Fig. 4 (to be reported elsewhere).
We obtained g of the FtsQ-RNC-YidC-SecYEG-complex by recording the single channel current at different membrane potentials (Fig. 4b). The addition of FtsQ-RNCs to the YidC-SecYEG complex resulted in an increased g-value of 1.75 ± 0.1 pS/mM, which was identical to the g-value of the FtsQ-RNC activated SecYEG complex (Fig. 4b). This further supports our hypothesis of a substrate-induced relocation of YidC from the channel interior to the rim. One possible explanation for identical g values is that YidC is no longer part of the channel wall and only contacts SecY residues at the outer surface of the channel. However, this is at odds with the observed change in the anion (Cl − ) to cation (K + ) permeability ratio r (P Cl /P K ). For the RNC FtsQ -YidC-SecYEG complex, we measured ψ r values of −15 ± 3.9 mV, while the FtsQ-RNC-SecYEG complex showed values of ψ r −6.7 ± 1.1 mV. The corresponding r values of 5.1 ± 3.15 for the FtsQ-RNC-YidC-SecYEG complex and 1.9 ± 0.2 for the FtsQ-RNC-SecYEG complex indicate that the pore-lining residues are different in the two complexes. It appears likely that a positively charged YidC residue lines the pore thereby augmenting ion selectivity.
SecDF depletion and the proton motive force stabilize the interaction of YidC with the channel interior of SecY. The interaction between SecY and YidC was reported to be dependent on the SecDFYajC complex 27, 41 , but SecDFYajC depletion had only a minor effect on the crosslink between YidC and the lateral gate of SecY 20 . To analyze whether SecDF influences YidC's contact to the channel interior, we employed the conditional SecDF depletion strain E. coli BL325, which carries secDF under arabinose promoter control 9 . We prepared INV from E. coli BL325 cells expressing SecY(I91pBpa), grown either in the presence of arabinose or in the presence of glucose. In INV from glucose grown cells, SecD and SecF were undetectable by western blotting using α-SecD or α-SecF antibodies (Fig. 5a), while the cellular levels of SecY or YidC did not significantly change (Fig. 5a). When these INV were UV-exposed, a 90 kDa YidC crosslink product was detectable in both types of ψ r (mV) r  Table 1. Translocation channel diameter d and selectivity r of SecYEG and SecYEG-YidC complexes. The reversal potential ψ r was obtained by plotting the single channel current as a function of membrane voltage (Fig. 3b). Anion to cation permeability ratios r (mean ± standard deviation) were calculated according to Eq. 1.
 g is the single-channel equivalent conductance, which is defined as the ratio of the single channel conductance g and the average ionic strength of solution in both the cis and trans compartments and d is the pore diameter calculated according to Eq. 2.
INV (Fig. 5b). In SecDF-depleted cells, the crosslink product was only slightly weaker, which confirms the previous conclusion that the SecY-YidC interaction persists even in the absence of SecDF 20 . SecDF was suggested to be involved in pmf-dependent steps during protein transport. We thus also analyzed the effect of the protonophore CCCP (Carbonylcyanid-m-chlorophenylhydrazon) which collapses the proton gradient across the cytoplasmic membrane. The addition of CCCP reduced the amount of the SecY-YidC crosslinking product both in SecDF-containing and in SecDF-depleted cells (Fig. 5b). Thus, although the pmf appears to stabilize YidC's interaction with SecY's lateral gate, the effect is not drastically different in the presence or in the absence of SecDF.
The stabilizing effect of the pmf on the SecY-YidC crosslink was also observed in E. coli BL21 cells in vivo. The SecY-YidC crosslink products were significantly weaker in the presence of CCCP for residues 91 and 278 (Fig. 5c). For residue 191, we only found a weak in vivo SecY-YidC crosslink product, but this crosslink also seemed to be CCCP sensitive (Fig. 5c). SecDF depletion had no significant effect on the interaction of YidC with the surface of SecY's lateral gate (position 91 of helix 2B) 20 and this was also observed for position 127 within helix 3 of the lateral gate (Fig. 6a). It should be noted that we had previously observed two additional crosslink products for position 127 in vivo, which were by mass spectrometry also identified as SecY-YidC crosslinking products 20 . These products were not visible here, which is probably related to differences between E. coli BL21(C43) used previously and strain BL325 used here. We then analyzed whether SecDF influenced the interaction of YidC with the channel interior by performing in vivo crosslinking from residue 191, deep inside the Sec channel. As shown before for wild type cells (c.f. Fig. 5c), the insertion of pBpa into this residue only produced a very weak in vivo crosslinking product in SecDF-containing BL325 cells (Fig. 6a). In the absence of SecDF, however, a significant SecY(191pBpa)-YidC crosslink product was observed (Fig. 6a).
Whether the depletion of SecDF indeed promotes the interaction of YidC with the SecY channel interior was further confirmed by in vitro crosslinking. We isolated INVs of BL325 cells expressing either SecY(191pBpa) or SecY(278pBpa) from arabinose-or glucose-containing cultures and UV exposed them. For both position 278 and I191, we observed a significant increase of the SecY-YidC crosslink product in the absence of SecDF (Fig. 6b). Thus, both in vivo and in vitro, YidC has preferred access to the channel interior, in the absence of SecDFYajC. In contrast to previous reports 25,27,41 , our data indicate that SecDF is not required for the SecY-YidC interaction and that YidC even has preferred access to the channel interior when SecDF is missing. This discrepancy is probably explained by the fact that we used a conditional SecDF depletion strain, while SecDF-overexpressing strains were often analyzed for a possible role of SecDF on SecY-YidC interaction 10,25 . Intriguingly, the overexpression of SecDF induces a prlA4 phenotype 11,23 , which is connected with a weakened association between the subunits of the SecYEG translocon 42,43 and a SecYEG mediated ion leak 44 . We therefore analyzed whether the SecY-YidC interaction was influenced by the prlA4 allele of SecY. pBpa insertion at position 91 of wild type SecY confirmed the 90 kDa SecY-YidC crosslinking product in INV (Fig. 6c). The same crosslinking product was observed when pBpa insertion at position 91 was combined with the prlA300 allele of SecY, which carries a mutation within the plug domain 45 . However, combining the pBpa insertion at position 91 with the prlA4 allele significantly increased SecY-YidC crosslinking (Fig. 6c). This was even more significant, because the expression level of SecY(prlA4-I91pBpa) was significantly lower than the SecY(I91pBpa) or SecY(prlA300-I91pBpa) expression levels (Fig. 6c, lower panel). The prlA4 allele also displayed a slightly reduced electrophoretic mobility, which explains why the prlA4 SecY-YidC crosslink product migrated slightly higher. These data demonstrate that the prlA4 mutation of SecY stabilizes YidC's interaction with SecY's lateral gate. This probably explains why the SecY-YidC interaction is stronger 10,27 in SecDF overexpressing cells that display a prlA4 phenotype 11 .

Discussion
The association of YidC with the SecYEG translocon in E. coli was first shown by co-purification 10 and was confirmed by in vivo crosslinking, which demonstrated that YidC crosslinked to the lateral gate of SecY 20 . The initial model of a sequential hand-over of a TM from SecY to YidC 12,19 assumed that YidC binds to an emerging TM outside of the SecY channel. However, we found that YidC is essentially part of the channel and thus the protein conducting channel is not only formed by SecY but rather by a SecY-YidC complex. Protrusion of YidC into the pore is demonstrated i) by crosslinks between YidC and SecY's pore ring residues, ii) by the reduced single channel conductance of the translocation pore in YidC's presence, and iii) by a YidC-induced increase in anion selectivity.
According to our conductivity measurements, the SecY pore ring forms an opening of approximately 8.4 ± 0.5 Å in the ribosome bound state (Table 1). Upon accommodation of a nascent chain, the pore ring assembly is partially disrupted by movement of TM10 46 , but ensemble measurements show no major plug movement 46,47 . In contrast, single molecule measurements reveal conductivity, indicating that the plug moves out of the lumen in a fraction of channels and/or for a fraction of time 40 . Whatever the status of the plug, movements of TM2, TM7 and TM10 generate a continuous conduit including the pore and the lateral gate [46][47][48] . This allows substrate TMs to exit the channel, and probably also allows YidC to enter it. In vivo crosslinking is unsynchronized and covers YidC's interaction with SecYEG channels that harbour a nascent chain and with empty SecYEG channels. This is different in vitro, where only a minority of SecYEG channels is likely to contain a nascent chain. The YidC crosslink product to residue 191 deep inside the channel was clearly detected, but only under in vitro conditions and this result suggests that YidC moves out of the channel interior in the presence of a nascent chain. This was demonstrated by the loss of the crosslink product in the presence of a nascent membrane protein and by conductivity measurements that revealed an increase in the YidC-SecYEG channel lumen from 6.2 ± 0.6 to 8.1 ± 0.5 Å in the presence of a stalled FtsQ chain. In the proteoliposomes used for the electrophysiology measurements, the orientations of SecYEG and YidC are difficult to control and they likely contain the proteins also in a mixed orientation. However, as the ribosome binding site of SecY is on the cytosolic phase of the membrane, ribosomes/RNCs will only open those SecY channels that are in the correct orientation. Those in the opposite orientation will not bind ribosomes and are therefore silent in the electrophysiology experiments. YidC is also predominantly inserted into proteoliposomes in an inside-out orientation 49 . We also expect that a physiological interaction between SecY and YidC is only possible if both proteins are in their correct orientation. This is in line with our observation that only SecYEG-YidC channels are observable in a 1:1 stoichiometry. Only when an excess of SecYEG is reconstituted into the bilayer, we observe both smaller SecYEG-YidC and larger SecYEG channels.
Both our electrophysiological and crosslinking data are most easily explained by the assumption that one or more of the YidC TMs is already intercalated at the lateral gate of the empty SecY. There is up to now no structural information about a SecYEG-YidC complex, therefore it is currently unknown which part of YidC reaches into the channel. The first TM of YidC has not been crystallized so far because it is like the N-terminal part of the periplasmic domain disordered 50 . This probably indicates that the first TM is not in contact with the five TM-core of YidC, which would make it suitable for intercalating into the lateral gate. This would probably maintain YidC in an orientation in which its hydrophilic substrate binding groove would face the lateral gate and thus allow contact to substrates emerging from the Sec translocon. During co-translational protein insertion, the YidC residues serve to alter the hydrophobicity of the channel interior thereby modifying the release probability of TMs into the membrane. Since this kind of thermodynamic partitioning 51, 52 is known to be susceptible to the exact location of amino acids in the pore and to groupings of hydrophobic or hydrophilic residues in it 53 , it must also be sensitive to the actual hydrophobicity of the wall lining residues. The release of some weakly hydrophobic TMs is thus promoted by rendering the pore less hydrophobic 19,54,55 . YidC's hydrophilic groove with a conserved positively charged arginine residue in the membrane center 50 would work nicely for that purpose. The involvement of this arginine residue in the RNC FtsQ -YidC-SecYEG pore may also explain the observed increase in anion selectivity as compared to the RNC FtsQ -SecYEG complex. This particular function of the arginine residue during SecYEG-mediated membrane insertion might also be the reason why deleting it had no effect on the YidC-only membrane insertion pathway in E. coli 56 .
One puzzling observation of the current study is that the interaction of YidC with the channel interior is more pronounced in the absence of the accessory SecDF complex. SecDF probably functions as a pmf-powered chaperone system that facilitates late steps during protein translocation 22 . So far, SecDF has been mainly associated with SecA-dependent steps of protein transport 9, 26, 57 , but its contribution to membrane protein integration is largely unknown. In a stoichiometric SecYEG-YidC-SecDFYajC complex, SecDF were crosslinked to SecE, SecG and to the N-terminal half of SecY (helices 1-5) 25 . Thus, the SecDF binding site on SecYEG is very close or even partially overlaps with the binding site for YidC, which involves helices 2, 3, 7 and 8 20 . The contact of YidC with these helices is not significantly different in the absence or presence of SecDF 20 , but the observed crosslink probabilities suggest that YidC appears to have easier access to the channel interior when SecDF is depleted. This could reflect possible competition, in which SecDF disturbs YidC's contact to the pore ring residues. However, competition is difficult to envision considering that SecDF are low abundant proteins with approx. 30 copies per E. coli cell compared to approx. 300-400 SecYEG copies and 3000 YidC copies 58 . A more likely explanation is provided by SecDF's ability to facilitate transmembrane proton transport. SecDF could convert the proton gradient into an increment in transmembrane potential that in turn, may force conformational changes in SecYEG. This is in line with the enhanced co-translational insertion of the SecYEG-and YidC-dependent membrane protein CyoA in the presence of SecDF 25 and the observation that SecDF influences SecYEG-YidC complex formation, although SecDF is not part of this complex 15 . Finally, this explanation is also supported by our previous observation that SecYEG-SecA complexes with a stalled translocation intermediate cease to conduct ions at high membrane potentials 40 . Assuming that the ion-conducting empty SecYEG-ribosome complex 39 behaves similarly, this would explain YidC's reduced access probability to the channel interior.
We conclude that YidC facilitates membrane protein insertion by forming a heterotetrameric channel with SecYEG. The complex maintains the membrane barrier to ions at rest and opens up upon ribosome or RNC binding (Fig. 7). Although higher (physiological) membrane potentials are likely to close the channel for the passage of ions, the nascent chain may still be released through the lateral gate, which by virtue of its reduced hydrophobicity when YidC is intercalated, favors membrane partitioning of TMs that the SecYEG pore would otherwise preferentially release into the aqueous environment.

Methods
In vivo and in vitro pBpa cross-linking, protein purification. For in vivo cross-linking, E. coli BL21 and E. coli BL325 harboring the plasmids pSup-BpaRS-6TRN and pTrc99a-SecY(His)EG were grown at 30 °C in minimal medium in the presence of 1mM pBpa as previously described 20,29 . After harvesting, the cells were washed once with 50 mM triethanolamine acetate (TeaOAc) pH 7.5 and incubated on ice for 30 minutes under a UV lamp in PBS buffer (137 mM NaCl, 2.7 mM KCl, 10 mM Na 2 HPO 4 , 1.76 mM NaH 2 HPO 4 , pH 7.6). Cells were lysed in a French pressure cell and membranes were prepared and solubilized with 1% DDM (n-Dodecyl β-D-maltoside) (Thermo Scientific, Dreieich, Germany). SecY was further purified via Talon ® Affinity Resin (Clontech, Mountain View, CA) and its cross-linking partners were identified by immune detection. To perform the in vivo crosslinking in SecDF depleted cells, E. coli BL325 containing pSup-BpaRS-6TRN and pTr-c99a-SecY(His)EG with the amber stop codon at the indicated positions within SecY were grown overnight on LB medium containing 0.2% arabinose. After a washing step, these cells were used to inoculate a 1l culture containing pBpa and either 0.2% arabinose (SecDF + ) or 0.2% glucose (SecDF − ). Cells were grown to an OD600 of 1.0 and The empty SecYEG-YidC complex allows YidC to crosslink to the three indicated positions (red) at SecYEG's pore ring (yellow) or the lateral gate. Ribosome binding is required to elicit the formation of a transmembrane channel (cross-section in blue) that allows ion permeation at low membrane potentials. Right Panel: Upon insertion of a nascent chain (green), YidC is expelled to the outer rim of the SecYEG pore, thereby increasing the cross-section (ion conductivity) of the SecYEG channel. Crosslinks to the channel interior (I191) is sterically constrained in the presence of a nascent chain while crosslinks to the peripheral residues 91 and 278 are still observable. In order to release the nascent chain, the lateral gate must open, i.e. conceivably the contact between SecYEG and YidC is further weakened. This is in line with crosslink data showing conformational changes at the SecY(I91)-YidC interface 20 . In summary, YidC facilitates the partitioning of a nascent membrane protein into the lipid environment by reducing the hydrophobicity of the lateral gate.
Purification of FtsQ-RNCs. Elongation-arrested 176 amino acid-long nascent chains consisted of the 101 N-terminal amino acids of the membrane protein FtsQ 60 . Between amino acids 27 and 28 of FtsQ, a calmodulin binding peptide was inserted that was flanked on both sides by Gly -Gly spacers. A TEV cleavage site and the stalling sequence SecM represented the C-terminus of the nascent chain. We used Bad22 as the vector backbone for expression in his-tagged -ribosome producing JE28 E. coli cells 59 . The JE28 cells were grown to an OD of 1.0 in 2xYT medium with kanamycin (30 mg/l) and ampicillin (100 mg/l). RNC expression was induced by arabinose Reconstitution of SecYEG and YidC/SecYEG into planar bilayers. "Solvent-free" planar lipid bilayers were folded by raising the level of two adjacent aqueous solutions over the dividing 100-200 μm-wide aperture in a Teflon septum with E. coli polar lipid extract (Avanti Polar Lipids, Alabaster, AL) monolayers on top 61 . Fusion of proteoliposomes containing either reconstituted SecYEG or YidC/SecYEG to the free standing planar lipid membranes (Fig. 4) was facilitated by raising the osmotic pressure in one of the compartments (cis) of the Teflon chamber 39,40 . The cis compartment also contained the proteoliposomes and the ribosomes/RNCs. SecYEG and YidC were present in proteoliposomes in a 1:1 molar ratio (2.5 nM each). The ribosome concentration was 150 nM and the FtsQ-RNC concentration 5 nM. The second compartment (trans) contained only the buffer. It consisted of 50 mM K-HEPES, pH 7.5 and 150 mM KCl. Protein reconstitution into planar bilayers in symmetrical salt conditions (Fig. 3) was performed as described before 44 . In brief, a mixture of proteoliposomes and empty lipid vesicles was added to the cis compartment. The trans compartment solely contained empty lipid vesicles in buffer. After allowing the formation of monolayers on top of the suspension, planar bilayers were folded as described above.
Single ion channel measurements. Ag/AgCl reference electrodes were immersed into the buffer solutions on both sides of the planar bilayers. The command electrode of the patch clamp amplifier (model EPC9, HEKA electronics, Germany) was immersed into the cis compartment and the ground electrode into the trans compartment. The recording filter for the transmembrane current was a 4 pole Bessel with −3dB corner frequency of 0.1 kHz. The raw data we attained were analyzed using the TAC software package (Bruxton Corporation, Seattle, WA). Gaussian filters of 12 Hz were applied to reduce noise.
Calculation of channel ion selectivity. The anion (Cl − ) to cation (K + ) permeability ratio r (P Cl /P K ) is calculated from ψ r (Fig. 3)  where + K cis , + K trans , − Cl cis , and − Cl trans indicate the ion concentrations on the two sides (termed cis and trans) of the membrane. We account for the osmotic water flow within the unstirred near-membrane layers which concentrates the solution on one hypoosmotical side and dilutes it on the hyperosmotical side of the membrane. For membranes with conductivities in the range of tens of pA, this effect does not usually exceed 10% 62 . Therefore we assumed the bulk KCl gradient of 450 to 150 mM to correspond to a 230 mM gradient across the lipid bilayer proper.
Estimating pore diameter. We used channel conductance to obtain a rough estimate of pore diameter as previously described 44 : where d, l = 3 nm, and σ = 3.7 S/m correspond to channel diameter, channel length and conductivity of the solution in the pore, respectively.