Structural characterization of DcuS-PAS_C

The primary sequence of DcuS-PAS_C is given in Figure 2a. Attempts to solve its three-dimensional structure by solution-state NMR and X-ray crystallography thus far were unsuccessful. Alternatively, we have previously shown that protein precipitation allows for ssNMR-based structural studies under native-like conditions²³. To reduce a potential influence of the sample conditions, we prepared two isotope-labeled samples of DcuS-PAS_C using different precipitation protocols. The first sample (referred to as U-PAS_C) was uniformly ¹³C- and ¹⁵N- labeled and precipitated using PEG. The second sample (R-PAS_C) was uniformly labeled with ¹³C and ¹⁵N, but contained for reverse labeling^{24, 25} the six amino acids phenylalanine, isoleucine, lysine, leucine, arginine and valine in natural abundance (that is, U[¹³C,¹⁵N\(F,I,K,L,R,V)]) and was precipitated by changing the pH to the isoelectric point. Reverse labeling simplifies the structural analysis of two-dimensional (¹³C,¹³C) and (¹⁵N,¹³C) ssNMR spectra while still enabling the comparison between different sample preparation conditions.

Figure 2: Experimental results obtained on isolated DcuS-PAS_C.

(a) DcuS-PAS_C primary sequence and secondary structure according to PSIPRED²⁹. (b) ¹³C-¹³C spin-diffusion spectrum of U-PAS_C. Circles indicate C-C peak position based on the predicted secondary structure (according to ref. 30; color code as in a, yellow circles represent random coil secondary structure). (c) ¹³C-¹³C spin-diffusion spectrum of R-PAS_C (black). A simulated spectrum based on the peak positions according to shiftX³² of one modeled structure is superimposed in green. Specific residues and amino acid types used during the analysis are labeled. (d) Summary of ssNMR peak position (left) and volume (right) analysis mapped on the average ROBETTA structure. The radius of the ribbon is scaled according to the residue-specific r.m.s. deviation between all five predicted structures. Orange residues (left) probably occur in multiple configurations, and dark gray residues show isolated single strong, sequential cross peaks. For residues given in dark gray (right), experimental signal intensities are in line with structural predictions. In both cases, experimental cross-peak signals for red residues were missing. (e) Difference between experimental and predicted cross-peak signal intensities for a selected set of amino acid types and secondary-structure elements, assuming a structural model of the entire PAS_C domain (black) or without the N-terminal helix (white). Note that experimental cross-peak amplitudes were normalized using amplitude values seen for alanine (C-C) cross peaks in -strand conformations, which occur twice in the structural model. (f) CHHC spectrum obtained on R-PAS_C. Long-range contacts that are unambiguous based on the ROBETTA models are indicated. (g,h) Superposition of the five lowest-energy structures of DcuS-PAS_C, back calculated from an extended chain (see text and Supplementary Figures 4 and 5 for details). Residues identified in f are highlighted in green. The N-terminal cap (residues 208–229) is not shown in g and is indicated in gray in h.

Full size image (95 KB)

As described elsewhere²⁶, a de novo structural analysis of ssNMR DcuS-PAS_C samples would require, in the first place, sequential resonance assignments along the protein sequence. However, the three-dimensional structures of PAS domains are highly conserved despite low sequence identity²⁷. To facilitate the ssNMR-based structural analysis, we therefore generated comparative models that resulted from use of the fully automated ROBETTA server²¹. In general, PAS domains comprise four structural attributes²⁸: (i) an N-terminal cap (1); (ii) a PAS core (1,2,2,3,4); (iii) a helical linker connecting the PAS core and -scaffold (5); and (iv) the -scaffold (3,4,5). The modeled structures of DcuS-PAS_C converged to a specific PAS-fold with an -helical N-terminal cap. The orientation of the N-terminal cap differs most strongly among the five modeled structures (see Supplementary Fig. 1a online for an alignment of the modeled structures). In the context of ssNMR experiments, such structural disorder can lead to signal attenuation and changes in resonance frequency and may impact the ssNMR measurement of through-space distances. We hence analyzed our ssNMR data—resonance frequencies, cross-peak amplitudes and proton-proton distances detected indirectly—with close reference to the predicted secondary and tertiary structure.

ssNMR resonance frequencies. Figure 2b compares an experimental ¹³C-¹³C spin-diffusion spectrum of U-PAS_C to the expected C-C peak positions (given as colored circles), assuming a secondary structure as predicted by PSIPRED²⁹ (peak positions are taken from ref. 30). Additional cross-peak signals in the experimental spectrum arising from side chain correlations were not included in the prediction. In general, C and C resonance frequencies are sensitive to the local secondary structure of a protein³¹. The remarkable agreement between predicted and measured cross-peak signals strongly supports the validity of the predicted secondary structure. To reduce spectral overlap, we recorded the same ¹³C-¹³C spin-diffusion spectrum using the R-PAS_C sample (Fig. 2c). This approach facilitates the analysis of spectral regions characteristic for tyrosine, proline, glutamic acid and glutamine (C-C) correlations. Overall, the signal pattern agrees with data seen for U-PAS_C (Supplementary Fig. 1b), suggesting that the two precipitation protocols lead to a similar three-dimensional fold of DcuS-PAS_C. On the basis of the comparative models, a prediction of the peak position using the program shiftX³² is possible (Fig. 2c, green). The good agreement between the experimental spectra and the predicted correlation pattern indicates the equivalence of a statistical/PSIPRED and shiftX/ROBETTA analysis in our case. In total, only the three residues Ile217 (Fig. 2b), Tyr215 and Pro214 (Fig. 2c) are not present at the predicted positions. These residues relate to the complete set of resolved correlations expected for the N-terminal helix. For further analysis, we recorded a ¹³C-¹³C spectrum revealing sequential correlations³³ (Supplementary Fig. 2a online). Knowledge of the amino acid sequence and the secondary structure readily led to pairwise assignments in several segments of the protein (Fig. 2d, left, indicated in dark gray). On the other hand, multiple peaks for a single correlation could be identified for a few residues. Such peak splitting is in general indicative of polymorphism—that is, the occurrence of a residue in different configurations—and this was observed for Ser310, Asn311 and Gly312 (Fig. 2d, orange; see also Supplementary Fig. 2b). Notably, previous molecular dynamics simulations^{34, 35} identified the same protein region connecting 4 and 5 as the segment with highest flexibility (reflected by the C positional shift) among the investigated PAS domains and with the exception of the N-terminal cap. Notably, the sequential correlation spectrum (Supplementary Fig. 2a) also revealed weak helical proline C-C cross signal, suggesting residual -helical propensity in the N-terminal cap.

ssNMR cross-peak amplitudes. Several sections of the two-dimensional ssNMR spectra can be attributed to signals stemming from certain residue types and secondary-structure elements. Even in the presence of spectral overlap, we can hence compare experimentally determined ssNMR amplitudes to values predicted on the basis of the structural model, assuming an ssNMR line width measured experimentally. Using this strategy, one finds major differences in the region characteristic for helical glutamic acid and glutamine (C-C) cross signals in difference spectra (Supplementary Fig. 3a,b online). The selected pattern of unlabeled amino acids in combination with the primary PAS_C sequence enables the investigation of this spectral region in R-PAS_C (see also Supplementary Fig. 3c). The difference between predicted and measured peak volumes is largest for glutamic acid and glutamine residues in -helical backbone conformations (Fig. 2e, Gln+Glu, black bar) and drastically decreases assuming a mobile and/or unstructured N-terminal helix (Fig. 2e, white bars). Accordingly, residues shown in red in Figure 2d (right) are missing in the ssNMR spectra, whereas experimental signal intensities for protein residues given in dark gray are in line with structural predictions. We note that the agreement between predicted and observed cross-peak intensities can be further improved assuming local disorder, for example, in -helical segments of the PAS core.

ssNMR proton-proton distances. As an independent means to study the structure of DcuS-PAS_C, we conducted a series of CHHC correlation experiments³⁶ on R-PAS_C and U-PAS_C that indirectly probe tertiary structure via short proton-proton contacts (Fig. 2f and Supplementary Fig. 4a,b online). For the fully labeled sample, we used a short ¹H-¹H contact time (t_HH) of 100 s to detect only the shortest contacts (<3 Å). Because the number of visible contacts is highly reduced in the reverse-labeled sample, a longer contact time of t_HH = 400 s could be used. The resolution of the spectra combined with the limited resonance assignment accuracy based on shiftX for larger fractions of the protein still complicates unambiguous de novo assignments of specific long-range contacts. However, assuming the validity of the modeled PAS fold, there are long-range distance correlations in isolated regions that are unambiguous based on the predicted fold. The highlighted correlations (Fig. 2f) indicate that DcuS-PAS_C precipitates adopt a stable overall fold that is consistent with large parts of the modeled structure.

Following the manual identification of correlations based on the predicted structure and the previously described approach of comparative analysis of CHHC and CC spectra³⁷, we developed an automated protocol to assemble a list of constraints that are most consistent with all ssNMR data. In total, 64 distance constraints including 3 intraresidue, 16 sequential, 13 medium range (|i–j| < 5) and 32 long-range constraints were used. The result of a conventional X-PLOR structure calculation based on these constraints is shown in Figure 2g,h (see Methods, ref. 37 and Supplementary Figs. 4 and 5 online for more details).

Analysis of ssNMR peak positions, amplitudes and CHHC correlations hence led to a self-consistent view of the three-dimensional fold of DcuS-PAS_C: a well-defined core, predominantly formed by the -sheet, and an unstructured or mobile N-terminal part involving mainly residues 208–229. Disordered protein regions identified by ssNMR correlate with PAS segments showing enhanced r.m.s. deviation values according to the ROBETTA models. Notably, these results were obtained irrespective of the two sample precipitation conditions and structure-prediction algorithms.

Structural characterization of DcuS-(PAS_P-TM_1,2-PAS_C)

To study the effect of domain interactions and membrane anchoring, we performed ssNMR measurements on the 39.6-kDa multidomain DcuS construct shown in Figure 1 (DcuS-(PAS_P-TM_1,2-PAS_C)) after reconstitution in E. coli lipids. Figure 3a shows a ¹³C-¹³C spin-diffusion spectrum of uniformly ¹³C- and ¹⁵N-labeled DcuS-(PAS_P-TM_1,2-PAS_C), referred to as U-DcuS. The spectrum is characteristic for a well-folded protein with well-dispersed ssNMR signals and a ¹³C line width below 1 p.p.m. As exemplified in Figure 3b for the alanine C-C cross peaks, the peak positions assigned for the sensory domain in solution²² (brown circles) fit well to the experimental spectrum of the membrane-embedded construct. In addition, signal sets annotated in green and blue in Figure 3b can be readily correlated with residues in the PAS_C and transmembrane domains. A more detailed analysis (Supplementary Fig. 6a,b,d online) suggests that the structure of both PAS domains is largely conserved. Within the given resolution, two assignments (Ser45 and Ser66 C-C) seem to be markedly changed (Fig. 3c). Ser45 is the first N-terminal residue found in solution of isolate PAS_P. Ser66 is part of the loop following the N-terminal helix in the periplasmic domain. The observed change for Ser45 suggests that the connection of the terminal ends to the transmembrane parts indeed modifies the local structure. The changes for Ser66 could be explained by a different orientation of the N-terminal helix or by dimer formation, which is expected according to X-ray data on the homologous domain of CitA³⁸, especially affecting the N-terminal helix and the following loop. However, ssNMR signals of the remaining residues highlighted in Figure 3c again fit well to the spectrum.

Figure 3: ssNMR results obtained on DcuS-(PAS_P-TM_1,2-PAS_C).

(a) ¹³C-¹³C spin-diffusion spectrum of membrane-embedded U-DcuS. Close up of the alanine C-C (b) and serine C-C (c) spectral region. Resonance assignments as reported for soluble DcuS-PAS_P²² are shown as brown circles. Green and blue circles are predicted peak positions for residues in the PAS_C and TM_1,2 domains, respectively.

Full size image (71 KB)

To reduce spectral overlap and to facilitate spectroscopic comparison to results obtained for isolated DcuS-PAS_C, additional ssNMR spectra were recorded using a sample of DcuS-(PAS_P-TM_1,2-PAS_C) with the same reverse labeling as for R-PAS_C; we refer to this sample as R-DcuS. Reduced overlap in R-DcuS facilitates sequential assignments for several residues throughout the protein sequence (Supplementary Fig. 6b,d). For example, secondary chemical shifts and amplitudes for residues assigned in transmembrane regions TM₁ and TM₂ (Ser26, Ala27, Gly190, Met191, Gly197, Thr198 and Cys199) are in line with a well-ordered -helical configuration within the membrane (Fig. 4a).

Figure 4: Graphical illustration of the ssNMR experimental results on DcuS-(PAS_P-TM_1,2-PAS_C).

(a) Schematic of the multidomain, membrane-embedded, 351-residue fragment. Color coding reports on the consistency of ssNMR signal sets in U-DcuS and R-DcuS compared to values observed or predicted for the isolated domains. Dark and light blue residues have isolated inter- or intraresidue correlations, respectively, where expected. Yellow residues are in overlapping regions, but are consistent with the spectra. Peak positions for residues indicated in purple are not present at the expected position (see also Supplementary Fig. 6a,b). Gray and white residues are not labeled in R-DcuS or are not incorporated in experimental or modeled data, and are not considered here. (b) Structural model built from the isolated domains. Blue and purple residues (as in a) are highlighted on the left structural model. Results of the volume analysis are mapped on the right structural model. Red residues are unstructured and/or mobile and do not contribute to the ssNMR signal. Dark gray residues are fully present. Orientation of the periplasmic domains of both monomers is adjusted according to the X-ray structure of the homologous CitA dimer³⁸. The relative orientation of the modeled transmembrane domains results from the bent terminal helices of the DcuS-PAS_P solution structure. (c) Peak-volume–specific analysis of R-DcuS. Differences between experimental and predicted cross-peak signal intensities are given (analogous to Fig. 2e), considering the entire DcuS-(PAS_P-TM_1,2-PAS_C) construct (black), the construct without the N-terminal helix of the PAS_C domain (white) and, in addition, the construct without the DcuS N terminus (gray).

Full size image (97 KB)

In addition, a comparative analysis of the spectra of U-DcuS and R-DcuS (Supplementary Fig. 6a,b) suggests that signals assigned for Asn177 (PAS_P domain) in solution are shifted in the membrane-embedded construct, indicative of structural alterations in the PAS_P-TM₂ connecting region. Correspondingly, Ser45, Ser66 and Asn177 are not consistent with solution-state NMR data on PAS_P (Fig. 4a, purple).

In a structural model (Fig. 4b), chemical shift perturbations found for the PAS_P-TM₁ connecting region, which is strongly bent in solution, suggest a different relative orientation of the transmembrane domains to each other, which is in agreement with crystallographic data on CitA³⁸. As in Figure 4a, blue and purple residues are highlighted on the left-handed structure in Figure 4b. Although the existence of the other domains in the ssNMR spectra interferes with a further investigation of the characteristic peak positions described for the individual domain, the large abundance of glutamic acid and glutamine in the N-terminal part of the PAS_C domain still enables the analysis of the peak amplitudes. We note that DcuS-(TM₂-PAS_C) and DcuS-(PAS_P-TM_1,2) constructs formed no stable membrane-integral product and could not be used for ssNMR studies. Indeed, a large difference between measured and predicted peak volume is found for helical glutamic acid and glutamine residues in DcuS-(PAS_P-TM_1,2-PAS_C) (Fig. 4c, black bars) if secondary-structure elements are used as indicated in Figure 4a. The most plausible explanation for the observed difference found for glutamic acid and glutamine residues is structural disorder in PAS_C. In fact, a contribution of PAS_C, as seen in the isolated domain (Fig. 2e, white bars), leads to good agreement between measured and predicted signal intensities (Fig. 4c, white bars), suggesting that the disorder found for the N-terminal helix of the cytoplasmic PAS_C domain is not restrained by the connection to the transmembrane helix. Notably, a similar discrepancy is found for threonine and serine residues with resonance frequencies typically found for coil or -strand conformations. These differences can be readily attributed to the presence of a disordered N terminus, which is equivalent to the absence of the first 20 residues plus the 20-residue histidine tag (Fig. 4c, gray bars) in our ssNMR spectra. As in Figure 2d (right), results of the volume analysis are mapped on the structural model (Fig. 4b); red residues are disordered and do not contribute to the ssNMR signal, whereas dark gray residues are fully present.

Structure-function relationship in the cytoplasmic PAS domain

The function of DcuS, or the capability for autophosphorylation and transcriptional activation of target genes, can be tested in vivo by measuring the expression of the dcuB'-'lacZ reporter gene. Expression of dcuB'-'lacZ depends on DcuS and reports the functional state of DcuS in the DcuSR two-component system^{5, 39}. We performed a mutagenesis study to investigate the functional role of the cytosplasmic PAS_C domain for gene expression (Fig. 5a). The most prominent effects were found for replacement of Asn248 by alanine, aspartic acid or glycine residues and of Asn304 by an aspartic acid residue, which caused full (constitutive) activity of dcuB'-'lacZ expression without addition of C₄-dicarboxylates; in contrast, the wild type proteins or proteins with mutations in other regions of PAS_C (including N248S) required C₄-dicarboxylates for induction of the reporter.

Figure 5: Functional effects of mutations in the cytoplasmic PAS domain and their potential structural consequences.

(a) In vivo activation (dcuB'-'lacZ expression levels) of E. coli IMW260 (dcuS deficient) and dcuS on plasmid (wild type or mutated). Bars represent expression of dcuB'-'lacZ in IMW260 with a control plasmid and a plasmid encoding wild-type (DcuS⁺) and mutant forms of DcuS after activation without (gray) and with (black) fumarate. (b) Two versions of the first ROBETTA model (blue and green) are aligned according to the dimeric crystal structure of AvNifL⁴⁰ (transparent yellow). Mutation sites that show activation in the absence of stimulus are shown in red. (c) Asn248 forming hydrogen bonds to the N-terminal helix. (d) Asn304 is found in close proximity to a lysine side chain of the other monomer. (e) Model of PAS_C-mediated signal transduction in full-length DcuS. On the periplasmic side, two monomeric solution-state NMR structures of DcuS-PAS_P²² (white) are aligned according to the X-ray structure of dimeric CitA³⁸ (red). Dimer disassembly of the PAS_C domains (blue and green) upon activation/mutation substantially alters their C-terminal extensions and impacts the adjacent kinase domains (from TM0853, PDB 2C2A)⁴⁴. The phosphate acceptor His260 is highlighted in purple.

Full size image (95 KB)

How can these findings be related to our spectroscopic view of a disordered N-terminal helix? A structural analysis of the dimeric crystal structure of the PAS domain of Azotobacter vinelandii NifL sensor protein (AvNifL)⁴⁰, which was identified by ROBETTA to be most homologous to DcuS-PAS_C, offers some insight. We aligned two versions of the modeled structure of DcuS-PAS_C to the dimeric crystal structure of AvNifL-PAS⁴⁰ (Fig. 5b–d). In the dimeric model of DcuS-PAS_C, the N-terminal helix has a crucial role in formation of the dimer interface, and the two mutation sites (indicated as red sticks on both monomers) that lead to C₄-dicarboxylate–independent activation are found near the -helical N-terminal cap. Residue Asn248 is conserved and in both proteins forms intramolecular hydrogen bonds to the N-terminal helix that to a large extent determine the orientation of this helix (Fig. 5c and Supplementary Fig. 5d,e). Removal of these interactions, as occurs with the N248A, N248G and N248D mutations, leads to protein activation; this is not observed for the N248S mutation, which can partially restore hydrogen bonding. Notably, residue Asn248 is the most conserved residue in PAS domains²⁸ and within PAS_C of the DcuS/CitA family. The corresponding residue (Asn34) of the aerotaxis sensor Aer of E. coli is also required for signal transduction⁴¹, although Aer is a chemotaxis sensor with a completely different domain composition. According to the arrangement shown in Figure 5b,d, the second mutation site (Asn304) of one monomer reveals close intermolecular proximity to the side chain of Lys232 of the second monomer. Because Lys232 is not conserved in the homolog, crystallographic data of a possible contact are missing. However, the effects of an N304D mutation could very well be explained by an intermolecular interaction with the Lys232 side chain.

Comparison of these data with our ssNMR data suggests that protein activation of DcuS correlates with increased molecular disorder as observed for the N-terminal helix in the membrane-anchored and precipitated DcuS-PAS_C. In this respect, the data obtained on isolated wild-type DcuS-PAS_C imply that additional factors are necessary to stabilize a dimer interface, as found for AvNifL-PAS. Indeed, recent biochemical results for the system in vivo have shown that DcuS is permanently in the active state, even in the absence of C₄-dicarboxylates, when the fumarate and succinate antiporter DcuB of anaerobic fumarate respiration^{42, 43}, which functions as an additional signal input site of DcuS (A. Kleefeld, J. Bauer, J. Krämer & G. Unden, unpublished data), is lacking. It is hence tempting to assume that DcuS-(PAS_P-TM_1,2-PAS_C) studied here probably reflects the activated state and that mutations in the in vivo case lead to destabilization of a potential PAS_C dimer interface as a prerequisite for activation of kinase activity.

DcuS-PAS_C sample preparation.

We produced uniformly labeled E. coli DcuS-PAS_C (211–325; U-PAS_C) in the E. coli strain Bl21(DE3) in minimal medium with ¹⁵NH₄Cl and ¹³C₆-D-glucose as nitrogen and carbon sources. For reverse labeling (R-PAS_C), this medium was supplemented with unlabeled phenylalanine, isoleucine, lysine, leucine, arginine and valine. Details of expression and purification are described in Supplementary Methods online. The purified protein was dialyzed against 20 mM HEPES, pH 7.0, and 50 mM NaCl, and the final concentration was adjusted to 13 mg ml^-1. U-PAS_C was precipitated from this solution by adding an equal volume of 25% (v/v) PEG 2000 dissolved in the same buffer, followed by concentrating the sample to about half its initial volume in a centrifugal evaporator⁵¹. R-PAS_C was precipitated by gradually adjusting the pH of the protein solution with hydrochloric acid to approximately 5.5.

Histidine-tagged DcuS-(PAS_P-TM_1,2-PAS_C) sample preparation.

The cDNA encoding fragment 1–333 of E. coli DcuS (DcuS-(PAS_P-TM_1,2-PAS_C)) was cloned into vector pET28a (plasmid pMW309) encoding a fusion protein with an N-terminal His₆-tag. Expression was performed in E. coli C43(DE3)⁵² containing plasmid pMW309. For preparation of uniformly and reverse-labeled protein, E. coli C43(DE3) pMW309 was grown in M9 minimal medium⁵³ as described for the DcuS-PAS_C fragment (Supplementary Methods), including the identical amino acids for reverse labeling. Expression and purification of DcuS-(PAS_P-TM_1,2-PAS_C) was similar to the procedure described for full-length DcuS^{5, 37} and is described in detail in the Supplementary Methods. For reconstitution of DcuS-(PAS_P-TM_1,2-PAS_C), liposomes were prepared from E. coli phospholipids (polar lipid extract, 20 mg ml^-1 in chloroform, Avanti Polar Lipids)⁵. Liposomes were destabilized by addition of Triton X-100 at an effective detergent:lipid ratio of 2.5 (ref. 54). Purified DcuS-(PAS_P-TM_1,2-PAS_C) (5.5 mg) was added at a phospholipid:protein ratio of 10:1 (w/w) and mixed by gentle agitation for 10–15 min at 20 °C. For every milligram of Triton X-100, 3 5 mg degassed Bio-Beads were added to remove the detergent. The mixture was incubated for 2 h at room temperature (20–23 °C) with gentle stirring. We then added another 5 mg Bio-Beads per milligram of detergent and incubated the solution overnight at 4 °C. Fresh Bio-Beads were added and incubated for 1–2 h at 20 °C, and the supernatant was removed and centrifugated (300,000g, 50 min). Proteoliposomes were washed twice (300,000g, 10 min) and resuspended in 50 mM TrisHCl, pH 7.7, to a volume of 400 l (5.5 mg protein), frozen in liquid nitrogen and thawed at room temperature for three cycles before freezing and storage at -80 °C.

Genetic methods.

Plasmids and strains are listed in Supplementary Table 1 online. Site-directed mutagenesis was performed with the QuikChange kit (Stratagene) and plasmid pMW181 containing the intact dcuS gene using the primers listed in Supplementary Table 2 online. Plasmids with verified mutations were transformed into E. coli IMW260 with an insertionally inactivated chromosomal dcuS gene and a dcuB'-'lacZ reporter gene fusion. For assays on the function of the dcuS mutations, the bacteria were grown under anoxic conditions in enriched M9 minimal medium with 50 mM glycerol, 20 mM DMSO as an electron acceptor and 20 mM fumarate as effector^{3, 39}. Exponentially growing bacteria (optical density at 578 nm (OD₅₇₈) of 0.5 to 0.8) were assayed for -galactosidase activity⁵³ in four replicates each from four independent growth experiments³⁹.

Solid-state nuclear magnetic resonance.

All ssNMR experiments were conducted using 4-mm triple-resonance (¹H,¹³C,¹⁵N) probeheads at static magnetic fields of 18.8 T and 14.1 T, corresponding to 800 MHz and 600 MHz ¹H resonance frequencies (Bruker Biospin). (¹³C,¹³C) spin-diffusion experiments were measured at effective temperatures between 1 °C and 10 °C. ¹³C-¹³C mixing times of t_CC = 15 ms were used to detect intraresidue spin systems. Sequential resonance assignments were obtained using (¹³C,¹³C) correlation experiments performed under weak coupling conditions³³ using t_CC = 150 ms. Total experiment time for one spectrum was in the order of 1 d (DcuS-PAS_C) to 3 d (DcuS-(PAS_P-TM_1,2-PAS_C)). CHHC experiments were carried out using short cross polarization contact times (100 s for R-PAS_C and 120 s for U-PAS_C) bracketing the proton-proton mixing time (t_HH = 400 s and t_HH = 100 s, respectively). R-PAS_C was measured in the frozen state to increase signal to noise; U-PAS_C was measured at 10 °C. The total experiment time for each spectrum was about 7 d at 800 MHz. SPINAL64 proton decoupling⁵⁵ was applied using radiofrequency fields of 75–90 kHz and magic angle spinning rates between 8 kHz and 12.5 kHz. The effect of sample freezing was investigated for all samples used during this study. Although sensitivity in general increased in the frozen state, no differences were found in line width and overall correlation pattern. (¹³C,¹³C) spectra shown here were measured at effective temperatures around 10 °C (Fig. 2a,c and Supplementary Figs. 1, 2 and 4b) and -2 °C (Fig. 3 and Supplementary Fig. 4a,c). Further details are given in Supplementary Methods.

Comparative modeling and structure calculations.

Primary sequence as present in the DcuS-PAS_C domain samples was submitted to the ROBETTA server. No further manual intervention was done. Only one fragment was used and aligned to PDB 2GJ3 (ref. 40).

Structure calculations were performed with the X-PLOR NIH software package⁵⁶ using the torsion angle dynamics (TAD) protocol⁵⁷. The force field constants for the TAD calculations were taken from parallhdg5.3.pro⁵⁸. Standard values of and backbone dihedral angles for residues predicted by PSIPRED²⁹ as -helix and -strand were included into structural calculation with a tolerance of 25°. For the fully labeled sample, distance restraints were incorporated with an upper limit of 3.0 Å, and, for the reverse-labeled sample, an upper boundary of 4.5 Å was used. The calculation started with 3,000 cycles of TAD at 50,000 K, followed by 3,000 cycles of TAD with increasing values of interatomic repulsion while cooling to 1,000 K, a subsequent 7,000 cycles of molecular dynamics in Cartesian space while cooling to 300 K, and then 1,000 cycles of final Powell energy minimization. The force constants were set to 300 kcal mol^-1, 300 kcal mol^-1, 300 kcal mol^-1 and 150 kcal mol^-1 for the distance restraints, and 100 kcal mol^-1, 200 kcal mol^-1, 250 kcal mol^-1 and 300 kcal mol^-1 for dihedral restraints during the four stages of structure calculations. A soft square potential was used for distance restraints and a square well potential was used for dihedral restraints. See Table 1 for resulting structural statistics.

Table 1: NMR and refinement statistics for DcuS-PAS_C

Full table

Accession codes.

Protein Data Bank: Coordinates for DcuS-PAS_C have been deposited with accession code 2W0N.

Note: Supplementary information is available on the Nature Structural & Molecular Biology website.