Insights into glycan import by a prominent gut symbiont

In Bacteroidetes, one of the dominant phyla of the mammalian gut, active uptake of large nutrients across the outer membrane is mediated by SusCD protein complexes via a “pedal bin” transport mechanism. However, many features of SusCD function in glycan uptake remain unclear, including ligand binding, the role of the SusD lid and the size limit for substrate transport. Here we characterise the β2,6 fructo-oligosaccharide (FOS) importing SusCD from Bacteroides thetaiotaomicron (Bt1762-Bt1763) to shed light on SusCD function. Co-crystal structures reveal residues involved in glycan recognition and suggest that the large binding cavity can accommodate several substrate molecules, each up to ∼2.5 kDa in size, a finding supported by native mass spectrometry and isothermal titration calorimetry. Mutational studies in vivo provide functional insights into the key structural features of the SusCD apparatus and cryo-EM of the intact dimeric SusCD complex reveals several distinct states of the transporter, directly visualising the dynamics of the pedal bin transport mechanism.


Introduction 38
The human large intestine is home to a complex microbial community, known as the gut 39 microbiota, which plays a key role in host biology 1-3 . One such role is to mediate the breakdown transporters (TBDTs) known as SusCs (we propose to re-purpose the term "Sus" for 54 saccharide uptake system rather than starch utilisation system) 7,12,13,15 . SusC proteins are 55 unique amongst TBDTs in that they are tightly associated with a SusD substrate binding 56 lipoprotein [15][16][17] (Fig. 1a). Recently we showed that SusCD complexes mediate substrate 57 uptake via a "pedal bin" mechanism 15,17 . The SusC transporter forms the barrel of the bin, 58

A FOS co-crystal structure reveals SusCD residues involved in ligand binding 97
In the B. theta levan PUL, the periplasmic enzymes are GH32 exo-acting fructosidases that 98 release fructose from the imported β2,6 FOS ( Fig.  1a) 19 A previous crystal structure of Bt1762-63 obtained in the absence of levan substrate revealed 116 a dimeric (SusC2D2) closed state in which the SusC TonB-dependent transporter (Bt1763) 117 lacked the plug domain as a result of proteolytic cleavage 15 . We have now obtained a structure 118 without substrate using a preparation that did not suffer from proteolysis (Fig.  1b,

left panel;; 119
Supplementary Table 1). Interestingly, while this structure is very similar to that reported 120 earlier, density for the plug domain is weak but clearly present. While this suggests that the 121 plug has been ejected from the barrel in the majority of transporter molecules in the crystal, 122 the relatively poor fit of the density with the native plug domain suggests increased dynamics 123 of the in situ plug domain in the subset of transporters that contain a plug (Extended Data Fig.  124 1).

126
To provide further insight into glycan recognition and transport by SusCD complexes, we next 127 used the same protein preparation to determine a co-crystal structure using data to 3.1 Å 128 resolution with b2,6-linked FOS. The FOS were generated by partial digestion of levan by 129 Bt1760 endo-levanase, followed by size exclusion chromatography (SEC) and analysis by 130 thin-layer chromatography (TLC) and mass spectrometry (MS;; Methods). In this structure, 131 containing FOS with a wide range of sizes (~DP15-25), the plug domain is present with normal 132 occupancy, suggesting that it is more stable in the presence of substrate (Fig. 1b, middle panel  133 and Extended Data Fig. 1). Like the oligopeptide ligands in the RagAB and Bt2261-64 134 structures 15,17 , the FOS is bound at the top of a large, solvent-excluded cavity formed by the 135 Bt1762-63 complex. Density for seven b2,6-linked fructose units can unambiguously be 136 assigned in the structure and this was designated as the primary binding site (Fig.  2a). The 137 bound oligosaccharide is compact and has a twisted, somewhat helical conformation. The 138 ligand makes numerous polar contacts with side chains of residues in both Bt1762 and Bt1763 139 ( Fig. 2b). For Bt1762 (SusD) these residues are D41, N43, D67, R368 and Y395, and for 140 Bt1763 (SusC) T380, D383, D406 and N901. In addition, prominent stacking interactions are 141 present between the ring of fructose 2 (Frc 2) and W85 of Bt1762. Interestingly, a β2,1 142 decoration is present in the bound ligand at Frc 4, and the branch point interacts with the 143 extensive non-polar surface provided by the vicinal disulphide between Cys298 and Cys299 144 of Bt1762 (Fig.  2b).

146
We also determined a co-crystal structure of Bt1762-63 with shorter β2,6 FOS (~DP6-12) 147 using data to 2.69 Å resolution (Supplementary Table 1 and W483 in the barrel wall, and with H169 and E170 in the plug domain (Fig.  2c). The fit to 153 the density is better for a 3-mer with a β2,1 decoration compared to a b2,6-linked 4-mer, 154 suggesting the transporter may have some specificity for FOS with a β2,1 decoration, or 155 alternatively, that Erwinia levan contains extensive β2,1 decorations such that most of the 156 levanase products are branched. The relatively small size of the co-crystallised FOS, 157 combined with the relative orientation and the large distance between FOS1 and FOS2 (> 20 158 Å;; Fig. 1c), makes it highly plausible that there are two ligand molecules in the Bt1762-63 159 cavity. The co-crystal structure with the longer FOS also shows some density at the secondary 160 site, but it is of insufficient quality to allow model building, perhaps due to the lower resolution.   and is colored black, with sidechains displayed as sticks. The equivalent region in the closed 213 structure is not visible and is therefore assumed to be disordered and likely protrudes from the 214 barrel, leaving the Ton box accessible to TonB. The visible density for the N-terminus starts 215 at residue 96 of the "closed" plug and at residue 84 for the "open" plug. Cryo-EM figures were 216 made with ChimeraX 52 . 217 218 The established model of TonB-dependent transport 24 assumes that extracellular substrate 219 binding to a site that includes residues from the plug domain induces a conformational change 220 of hinge1 (Methods) caused little to no growth defect (Fig.  4b), which is surprising given that 270 Bt1762-63Dhinge1 expression is barely detectable (Figs. 4d,e). By contrast, the Dhinge2 strain 271 showed a complete lack of growth during the 24 h monitoring period, but expression of this 272 mutant was also very low (Figs. 4b,d,e). Surprisingly, a strain in which both hinges were 273 deleted (Dhinge1&2), grew similarly to the DSusD strain, i.e. after a ~8 hr lag phase (Fig.  3b).

275
TonB box and N-terminal extension mutants: SusC-like proteins are predicted to be TonB-276 dependent transporters (TBDTs), but direct evidence for this is lacking. We therefore 277 examined the importance of the putative TonB box located at the N-terminus of Bt1763. In  Supplementary  Tables  3  and  4). The NTE 304 structure shows a well-defined core of an Ig-like fold with a 7-stranded barrel (Fig.  5c, left  305 panel). The N-terminus (including the His-tag) and the C-terminus, corresponding to the Ton 306 box, are flexibly unstructured, as evidenced from their random coil chemical shifts and the 307 absence of long-range NOEs ( Fig.  5b  and  Fig.  5c Interestingly, the structure of the FoxA STN in complex with the CTD of TonB shows that the 329 STN is also composed of a small barrel with seven elements, some of which are helical instead 330 of strands (Fig.  5c) 27 . This similarity suggests that, like the STN, the NTE might interact with a 331 protein in the periplasmic space. In both domains, the Ton box is separated from the domain 332 body and will thus be accessible to binding by the C-terminal domain of TonB. One possibility 333 for a role for the NTE could be to provide interaction specificity for the multiple TonB orthologs 334 present in the B. theta genome. saccharide displaying no affinity for the transporter (Fig. 6, Extended Data Fig. 8 and  357 Supplementary Table  5). For the larger FOS, affinity increased from DP5 to 6 (Kd ~30 and 17 358 µM, respectively) and plateaued at DP8 (tube 174, T174 Kd ~1 µM), with Bt1762-63 binding 359 to all FOS between DP8 and at least DP13-14 (T115) with similar affinity. These data are in 360 broad agreement with the co-crystal structures, which show well-defined density for 7 fructose 361 units in the primary binding site, suggesting that these provide the bulk of the binding 362 interactions. Surprisingly, no binding was detected for the FOS in SEC fractions T114 and 363 T113, despite these fractions having similar MS profiles to T115 with a broad range of 364 oligosaccharides present (Fig.  6  and  Extended  Data  Fig.  7). The average MW of the FOS in 365 tube T114 (Mn >2666) is larger than that of T115 (Mn >2193;; Extended Data Fig.  7), and it may 366 be that this increase in average size is enough to preclude binding to the transporter. 367 Furthermore, based on the co-crystal structure we can see that at least some, and perhaps 368 all, of the bound Erwinia levan-derived FOS has a b2,1 decoration, which may influence 369 binding to Bt1762-63. However, it was not possible to identify b2,1 decorations in the TLC or 370 MS analysis. Thus, T115 FOS might contain significantly more branched species than T114 371 and this could explain a higher affinity for the T115 fraction. Taken together, however, these 372 data indicate there is both an upper and lower size limit for FOS binding to the Bt1762-63 373 transporter in vitro, with the lower limit being DP5 and the upper limit ~DP15. In addition to 374 wild-type Bt1762-63, we also measured binding of ~DP9 FOS to the Bt1762(W85A)-63 variant 375 (Fig.  6). Surprisingly, no binding is observed for the mutant, even though the Bt1762W85A-63 376 strain grows as well as wild type on levan (Fig.  3a), suggesting that FOS binding by Bt1762 is 377 not essential for Bt1762-63 function in vivo. The complexity of the binding pattern in the spectrum is consistent with polydispersity of the 398 T114 and T115 fractions (Extended Data Fig.  9). More useful insights were obtained with the 399 T159 sample, which consists mainly of FOS with 8-10 fructose units (Fig.  7 and Extended 400 Data Fig.  7b). These medium-chain oligosaccharides bind preferentially to the intact SusC2D2 401 dimer rather than to the SusCD monomer such that no ligand-free dimer was evident in the 402 spectrum, potentially suggesting some kind of cooperativity for ligand binding in the dimer. 403 Interestingly, the relative proportions of protein-bound FOS mirrored their abundance in the 404 T159 sample (Fig. 7b), supporting the similar affinities of FOS with 8-10 fructose units for 405 Bt1762-63 as measured by ITC ( Fig. 6

and Supplementary Table 5). At the higher FOS 406
concentrations, binding of more than one FOS molecule per SusCD transporter was observed 407 (Fig.  7b), confirming the observation from our co-crystal structure that more than one ligand 408 molecule can be present in the binding cavity, at least for the relatively small FOS.

410
Finally, we wanted to confirm the upper FOS size limit in vivo by using testing growth of a 411 strain lacking the surface endo-levanase BT1760 against FOS of different sizes as the sole 412 carbon source. The D1760 strain was previously reported to lack the ability to grow on levan 19 , 413 which would provide another indication that the Bt1762-63 complex cannot import high 414 molecular weight substrates. Surprisingly, however, the growth rate of the D1760 strain on 415 levan from several different sources was similar or only slightly slower than that of the wild 416 type strain (Extended Data Fig.  10). PCR of the ∆1760 cells taken from stationary phase of 417 the cultures confirmed the deletion of the BT1760 gene from the cells, indicating the phenotype 418 was not due to contamination with wild-type strain (Extended Data Fig. 10). These data 419 suggest that all the levans tested contained enough low DP FOS to allow growth without 420 needing digestion by the surface endo-levanase. It was therefore not possible to determine 421 an upper FOS size limit of the Bt1762-63 importer in vivo. native MS data, it is likely that this total mass would comprise several individual molecules, 456 rather than one large molecule. As there are unlikely to be large structural differences among 457 SusCD-like systems, we suggest ~5 kDa as a general total size limit for these transporters, 458 which is consistent with recent data for the archetypal Sus 18 .

460
Our structures that show FOS in the principal binding site at the Bt1762-63 interface raise an 461 important question: how is ligand occupancy relayed to the plug domain, and how does this 462 lead to increased accessibility of the Ton box? This key issue is likely unique to SusCD 463 systems, in particular those SusCs without the long plug loop present in e.g. Bt2264 15 that is 464 able to contact ligand in the principal binding site and relay binding site occupancy directly to 465 the plug domain (Extended Data Fig. 11). In Bt1763, the smallest distance between the visible 466 part of the substrate in the principal binding site and the plug is 15 Å, and so an optimal-sized 467 FOS (in terms of binding affinity) of ~DP8-12 could not contact the plug directly. The presence 468 of a second substrate molecule at the bottom of the binding cavity (Fig.  1b) might be a way to 469 overcome this problem, implying a mechanism in which the binding cavity "bin" is filled first via 470 two or more substrate binding-release cycles to provide plug contacts, that collectively 471 increase accessibility of the Ton box and binding to the CTD of TonB.

473
The substrates for the transporter are generated by the combined action of the Bt1760 endo-474 levanase and the Bt1761 surface glycan binding protein (SGBP; ;  Fig.  1a) transiently associating with the SusCD core complex 35,36 . Besides depending on the type of 480 levan 23 , the FOS sizes delivered to Bt1762-63 will depend critically on the binding kinetics of 481 the Bt1760 levanase and on the proximity of Bt1761: a close association between the two 482 would most likely favour production of uniformly-sized FOS of relatively small size which, as 483 we have shown, are preferred substrates. Likewise, a close association between the enzyme 484 and Bt1762-63 will enhance capture of the generated FOS by SusD and subsequent delivery 485 to SusC. With regards to this last step, it is interesting to note that, in contrast to in vitro 486 conditions, the substrate binding function by Bt1762 is not necessary in vivo (Fig.  4  and  Fig.  487  Our structural data provide important clues about the function of SusD and about glycan import 508 in general (Fig.  8). The basis for these clues is the unprecedented observation that closed, 509 but empty transporters lack the entire plug domain. This spontaneous expulsion of the plug is 510 likely to be non-physiological and caused by a loss of lateral membrane pressure due to 511 detergent solubilisation. Nevertheless, it does suggest that Bt1762 lid closure causes 512 conformational changes within the Bt1763 barrel that decrease the "affinity" of the plug for the 513 barrel. This may facilitate the removal of the entire plug domain from the barrel by TonB action, 514 as opposed to local unfolding and formation of a relatively narrow channel as has been 515 proposed for non-Sus TBDTs 38-40 . To prevent plug removal in the absence of substrate 516 resulting in futile transport cycles, we postulate that only the direct contact of substrate with 517 the plug (as observed in the co-crystal structure with FOS2 and the in BT2263/64-peptide 518 complex 15 ) leads to increased accessibility of the TonB box and interaction with TonB (Fig. 8). 519 In our model, the impermeability of the OM, which otherwise would be compromised due to 520 the formation of a very large channel of ~20-25 Å diameter, would be preserved by the seal 521 provided by the closed SusD lid. Upon reinsertion of the plug, the transporter would revert 522 back to its open state (Fig.  8). The most important function of SusD proteins during glycan 523 import may therefore be to provide a seal to preserve the OM permeability barrier. Na2CO3 1 mg/ml, cysteine 0.5 mg/ml, KPO4 100 mM, vitamin K 1 µg/ml, FeSO4 4 µg/ml, 560 vitamin B12 5 ng/ml, mineral salts 50 µl/ml (NaCl 0.9 mg/ml, CaCl2 26.5 µg/ml, MgCl2 20 µg/ml, 561 MnCl2 10 µg/ml and CoCl2 10 µg/ml) and hematin 1 µg/ml. These cultures were supplemented 562  Fig.  2). Invariably, the closed 671 position of BT1762 was associated with an absence of density for the plug domain of BT1763. 672 Global 3D classification was unable to distinguish 'true' closed conformations from those 673 where BT1762 occupied a marginally open state. As a result, a masked 3D classification 674 approach was employed to achieve homogeneous particle stacks. The masked classification 675 was performed without image alignment and, since the region of interest is relatively small, 676 the regularization parameter, T, was set to 20. Intermediate results and further details are 677 provided in Extended Data Fig.  12. Clean particle stacks for the three principle conformational 678 states were subject to multiple rounds of CTF refinement and Bayesian polishing 49 . C2 679 symmetry was applied to both the OO and CC reconstructions. Post-processing was 680 performed using soft masks and yielded reconstructions for the OO, OC, and CC states of 3.9, 681 4.7 and 4.2 Å respectively, as estimated by gold standard Fourier Shell correlations using the 682 0.143 criterion.

684
Model building into cryoEM maps. Comparing the maps to the crystal structure of BT1762-685 BT1763 revealed that their handedness was incorrect. Maps were therefore Z-flipped in UCSF 686 Chimera 52,53 . The reconstruction of the OO state was of sufficient resolution for model building 687 and refinement. Bt1762 and Bt1763 subunits from the crystal structure were independently 688 rigid-body fit to the local resolution filtered map and later subjected to several iterations of 689 manual refinement in COOT 44 and 'real space refinement' in Phenix 45 . The asymmetric unit 690 was symmetrised in Chimera after each iteration. Molprobity 46 was used for model validation. 691 The reconstructions of the OC and CC states were of insufficient resolution to permit model 692 building and refinement owing to low particle numbers and a poor distribution of viewing angles 693 respectively. Instead, the crystal structure of Bt1762-Bt1763 was rigid-body fit to the CC state. 694 The ligand was removed from the model and an inspection in COOT showed that no density 695 extended past Lys213 in the direction of the N-terminus. All residues N-terminal of Lys213 696 were therefore removed from the model before rigid-body fitting. The open state from the OO 697 EM structure and the closed state from the crystal structure (modified as described above) 698 were rigid-body fit to their corresponding densities in the OC state. Rigid-body fitting was 699 performed in Phenix.