Characterization of the refolded Cox13 protein
In this study, Cox13, with a predicted mass of 15 kDa, was recombinantly overexpressed in Escherichia coli. Inclusion-body protein precipitate was refolded in the detergent dodecylphosphocholine (DPC). From a size-exclusion column (SEC), the Cox13-DPC complex eluted in a single peak (Additional file 1: Figure S1A), indicating that it was present in a single oligomeric state with an apparent mass of 55–60 kDa (see calibration inset in Additional file 1: Figure S1A). This size is consistent with a dimer of Cox13 (2×15 kDa) in a DPC micelle (~25 kDa [21]). The expected molecular mass and purity were confirmed by SDS-PAGE, where the Cox13 monomer runs at an apparent mass of 16 kDa and a fraction consistent with a dimer was also observed (Additional file 1: Figure S1B). Far-UV circular dichroism (CD) spectroscopy revealed that α-helix is the dominant structural element in refolded Cox13 protein (Additional file 1: Figure S1C). The Cox13-DPC complex was further analyzed by a two-dimensional (2D) heteronuclear single quantum correlation (HSQC) NMR experiment, and was found to have good spectral characteristics, and the expected dispersion of resonances typical for an alpha-helical protein [22, 23]. A transverse relaxation optimized spectroscopy (TROSY) type 15N-1H HSQC spectrum recorded at 900 MHz demonstrated that conducting solution NMR-based structural studies on Cox13 was feasible (Fig. 1).
NMR resonance assignment
The backbone chemical shifts assignment of Cox13 was completed to 93% using a combination of six TROSY-type heteronuclear triple-resonance experiments with a [U-15N, 13C]-labeled sample (Additional file 1: Figure S2). The secondary structure of Cox13 in DPC micelles was determined using backbone secondary chemical shifts [24], which clearly demonstrate that Cox13 contains four α-helices (Fig. 1c): α1 (A29–A39), α2 (T48–E79), α3 (P93–F106), and α4 (D112–R123), which are separated by regions lacking well-defined secondary structure. Resonance assignments were extended using the combination of side-chain correlation experiments and nuclear Overhauser effect (NOE) experiments on differently labeled samples. Specifically, the assignment of approximately 79% of side-chain resonances was achieved (Additional file 1: Figure S3). The NOE cross-peak assignments were obtained by an iterative procedure using a combination of manual and automatic approaches. A total of 1324 intramolecular NOE distance restraints were extracted from the 15N and 13C-edited NOESY-HSQC spectra. Moreover, using a mixed sample with 50% [U-15N, 13C]-labeled protein and 50% unlabeled protein, 15N, 13C-filtered/edited NOESY experiments were carried out to detect NOE cross-peaks that are due to intermolecular interactions [25] (Additional file 1: Figure S4). The presence of intermolecular NOE cross-peaks in combination with the SEC profile indicated that Cox13 forms a dimer under our experimental conditions. Altogether, the collected data allowed us to determine a well-defined Cox13 dimer structure.
Solution structure of Cox13 dimer in DPC micelles
A final ensemble of the 15 lowest-energy structures calculated using the program CNS [26] representing the Cox13 dimer in DPC micelles is depicted in Fig. 2a. Summaries of the experimental constraints and structural statistics are given in the supplementary information (Additional file 1: Tables S1 and S2). Overall, the solution structure of the Cox13 dimer is characterized by the dimeric transmembrane (TM) entity, composed of the TM helix from each monomer, forming a two-helix bundle (Fig. 2b). Apart from the TM region (α2), each Cox13 monomer structure also contains three short helices, the α1-helix at the N terminus, and the α3- and α4-helices at the C terminus. The helices are connected by short- or medium-length loops. These helices do not have well-defined positions in the structure (Fig. 2), but a few distance restraints between the α3 and α4 helices were found, indicating that they interact, at least transiently. The high helicity of the Cox13 dimer structure is consistent with the CD spectrum and secondary chemical shift analysis (Additional file 1: Figure S1 and Fig. 1c).
The Cox13 dimer interface is well-defined from inter-molecular NOEs and its dimeric TM core is mainly formed by hydrophobic interactions, which include the residues W51, V58, A62, L65, T66, F72, and V73 (Fig. 2c and Additional file 1: Table S2). In addition, an interaction between N69 in each monomer appears to stabilize the dimer formation. Notably, residue P61 induces a kink in the TM helix (Fig. 2d), which is presumably involved in forming the bow-shaped structure observed for the Cox13 in the full III2IV2 supercomplex structure [7].
In summary, the solution structure of the Cox13 dimer in DPC micelles comprises a well-defined TM bundle induced by the interaction between the TM α2-helix from each monomer, while the positions of the other helices are less defined. NOEs were, however, found between the α3 and α4 helices indicating an interaction between them.
To better understand the position of Cox13 in micelles, we carried out paramagnetic relaxation enhancement (PRE) experiments by recording 2D TROSY-HSQC spectra in the presence of either the detergent-soluble 16-doxyl stearic acid (16-DSA) or water-soluble gadodiamide. From data obtained with the latter compound, the most affected (and thus solvent-exposed) regions of Cox13 is the random coil following the TM helix, as well as the C-terminal unstructured residues (Additional file 1: Figure S5). In comparison, the α3 and α4 helices have slightly higher intensity ratios than the flanking loops and are thus only partially solvent-exposed. 16-DSA data supports the gadodiamide results for the aforementioned regions by showing opposite trends in intensity ratios. As expected, detergent-embedded 16-DSA has a large effect on the residues of the transmembrane helix, but also interestingly on some residues in the N-terminal loop and on the α3 and α4 helices. The results thus indicate an interaction of α3 and α4 with detergent in the micelle-water interface. Together with gadodiamide’s overall modest attenuation of the signal intensity for the N-terminal loop and α1 helix, the results indicate that large parts of Cox13 fold back on and interact with the micelle. We conclude that the interaction with detergent contributes to the stability of the non-transmembrane helices α1, α3, and α4.
Comparisons to Cox13 and homologs in full CytcO structures
The Cox13 subunit was recently resolved in a cryo-EM structure of the III2IV2 supercomplex from S. cerevisiae [7], which shows that it is localized at a peripheral position without contact with the Complex III dimer (Fig. 3a). A comparison of our NMR structure to the Cox13 structure in the III2IV2 complex (Fig. 3b) shows a similar overall topology, with a relatively well-aligned TM region, but significant differences in the overall structure. The largest differences are in the soluble domains, where the region occupying the intermembrane space (IMS) is mostly disordered in the supercomplex, whereas it folds into two short helices in our NMR structure. However, this is a very flexible region of the protein, which in the III2IV2 complex makes contacts with Cox3 and Cox12, discussed further below (and see [27]). Yeast Cox13 is in part homologous with bovine subunit VIa and is similarly positioned in relation to other CytcO subunits. In the crystal structure of the bovine Complex IV dimer, subunit VIa is located at the dimer interface [9], but there are no major differences to the subunit VIa structure in the monomeric CytcO form [28]. Comparing yeast Cox13 to the bovine subunit VIa (Fig. 3b), the largest differences are seen at the matrix-facing N-terminal region, where Cox13 has an organism-specific extension which is not present in subunit VIa.
In the yeast cryo-EM supercomplex structure, the Cox13 TM helix is significantly longer than in our NMR structure, and the TM region makes contact with Cox3, and the recently resolved (for specific conditions) respiratory supercomplex factor 2 (Rcf2) [27]. This applies to some of the residues that form the dimer interaction surface in our NMR structure (Fig. 2c), e.g., N69 and V73, which are in contact with Cox3, while V58 and W51 interact with Rcf2. It is interesting to note that the Rcf2 protein, which has been suggested to regulate supercomplex formation and CytcO activity [29,31,31], thereby could influence the propensity for Cox13 to induce dimerization of the CytcO, which then could lead to a “string” of supercomplexes, as suggested earlier [32].
Influence of nucleotides on the Saccharomyces cerevisiae CytcO catalytic activity
For both yeast and bovine CytcO, Cox13 (or equivalent) has been suggested to house an allosteric ATP binding site, but at different locations; yeast Cox13 was proposed to house such a site in the IMS [20] whereas in mammalian (bovine) mitochondria, subunit CoxVIa (equivalent to Cox13), was suggested to contain an allosteric nucleotide-binding site at the matrix-facing N terminus [16].
We purified wildtype S. cerevisiae CytcO with an engineered His-tag on Cox13 to ensure the presence of the subunit (as Complex IV in yeast shows a fraction lacking Cox13 [19]). We also purified the Cox13Δ-CytcO (with a Flag-tag on Cox6) in order to assess the influence of the Cox13 subunit on catalytic turnover. We observed no significant effect on the maximum catalytic activity of CytcO in the absence of Cox13, similar to previous results [20], but the relative activity (between wildtype and Cox13Δ-CytcO) varied between preparations and with buffer conditions (Additional file 1: Figure S6). ADP had small stimulatory effects for both variants. High phosphate concentrations had a small stimulatory effect in wildtype CytcO, which was increased in Cox13Δ-CytcO (Additional file 1: Figure S6). Addition of ATP also increased turnover activity in both wildtype and the Cox13Δ-CytcO (Additional file 1: Figure S6), and tentative fits indicate that ATP binds less tight (with tentative Kd in the mM range, qualitatively similar to a previous study [20]) and with a higher maximum activity in Cox13Δ-CytcO (Additional file 1: Figure S6BC). These (and previous [20]) data indicate a complex behavior with more than one ATP binding site on CytcO. In the previous study [20], the results were interpreted in terms of two regulatory ATP binding sites in CytcO, where one is stimulatory and one is inhibitory and located on Cox13.
The data presented above were not easily interpreted, and in the bovine CytcO, it has been shown that the interaction between the electron donor cyt. c and CytcO is highly dependent on the presence of anions, including ATP and phosphate [33, 34], which could complicate titrations such as those described above. Therefore, we monitored the CytcO activity as a function of added yeast cyt. c in both the absence and presence of 5 mM ATP using a low-Pi buffer for both the wildtype and the Cox13Δ-CytcO (Fig. 4 and Additional file 1: Figure S7). In the wildtype CytcO, we fitted the data to two Kms for cyt. c with values of 1.3±0.6 μM and 20±10 μM (Fig. 4a), consistent with a previous study that also found two Kms ([35], but see also [36]). In the Cox13Δ-CytcO, the higher Km interaction with cyt. c is very similar, i.e., 20±10 μM, while the lower Km is lowered to 0.2±0.1 μM. In the presence of 5 mM ATP, there are two clear effects in the wildtype CytcO; first the low Km cyt. c binding site is clearly inhibited (or lost) (similar to the effect observed in bovine CytcO [33]), and the data can be fitted with only one Km of 8.6±0.3 μM (Fig. 4b and S7). Moreover, the overall maximum activity is stimulated to 208±2% (Fig. 4b). In Cox13Δ-CytcO in the presence of 5 mM ATP, the low Km interaction with cyt. c is similarly lost with only one Km at 5.8±0.3 μM (Fig. 4b). Also here, the maximum activity is increased, by a larger factor to 236±3%. Our data shows that the overall affinity for cyt. c is lowered in the presence of ATP in both wildtype and Cox13Δ-CytcO, but that the increase in the maximum rate is higher in Cox13Δ- than in wildtype CytcO. The stimulation of activity in the presence of phosphate is also higher in Cox13Δ-CytcO than in wildtype (see Additional file 1: Figure S6A). It is clear that there are several effects of ATP binding to CytcO and that removal of the Cox13 subunit impacts at least one of these effects; the maximum stimulation of activity. However, the removal of Cox13 has complex effects on the CytcO activity, discussed further below. There is a difference between our results for the effect of ATP on yeast CytcO/cyt. c interactions and those obtained for the bovine (or horse) CytcO/cyt. c interactions where ATP more clearly inhibits CytcO activity at low cyt. c [33, 34]. The reason for this difference is presently unknown, but could be related to the observation that in addition to binding CytcO, ATP binds and affects also horse cyt. c (see e.g. [37]), something that was suggested to not occur in yeast cyt. c [38].
Interaction of Cox13 with ATP and ADP
We addressed the question of whether or not ATP binds to Cox13 by NMR and CD titration experiments. We first studied ATP/Cox13 interaction by CD up to a molar ratio of ~20:1 (corresponding to 2.5 mM ATP and 135 μM protein), but measurements at higher ratios were prevented by ATP’s high absorbance. For the ATP/Cox13 ratios possible to assay in CD experiments, slight changes in secondary structure were observed (Additional file 1: Figure S9B). The changes indicate a small increase in helicity, as the absolute intensities of the peaks at 208 and 222 nm increased.
Next, the influence of ATP on 15N-labeled Cox13 was assayed by monitoring chemical shift changes in 1H-15N-HSQC spectra. Spectra were collected at six different concentrations of ATP (with the same protein concentration as in the CD experiments) resulting in ligand/protein ratios of 2:1, 5:1, 10:1, 25:1, 50:1, and 100:1 (Fig. 5a and Additional file 1: Figure S8), the highest ratio corresponding to [ATP]=13 mM. Higher ATP/Cox13 concentration ratios induced protein aggregation, possibly due to larger conformational changes in Cox13 induced by ATP. From the incremental addition of ATP to Cox13 in the NMR titrations, we observed gradual shift changes in resonance positions. A chemical shift perturbation (CSP) analysis across the sequence (Fig. 5a) revealed that the most affected residues are located in two distinct regions of the Cox13 structure; in the loop region immediately following the TM-helix (R81, H83, K85), and in the beginning of α3 (R94, Y96). A residue at the C terminus was also significantly affected by ATP addition (I125). Examples of CSPs as a function of the ligand/protein ratio are shown in Fig. 5b for four of the most affected residues. The higher CSPs suggest that these residues are in direct contact with ATP or are indirectly affected by the environmental changes produced by ATP binding, e.g., through conformational rearrangement of the protein. We note that although full saturation has not been achieved, the data indicates that at the highest concentration we see an onset of saturation for all residues. A global Kd of 15±2 mM was readily fitted to the data for all residues that displayed large CSPs, indicating that the changes in chemical shifts are consistent with one single binding event (Fig. 5).
To investigate the corresponding effects, those on ATP upon binding of the Cox13 protein, changes in 31P NMR spectra of ATP were monitored in a titration series of ATP to unlabeled Cox13. Significant 31P chemical shift changes were observed for the terminal, negatively charged phosphate group of ATP, with only minor changes for the central phosphate and no effect on the innermost one, indicating that electrostatic interactions are important for binding (Fig. 5c and Figure S9).
We also titrated ADP to Cox13 in the same way as for ATP, and the results are shown in Additional file 1: Figure S10. Overall, chemical shift changes suggest that ADP interacts with Cox13 in a similar manner as ATP, with the same residues being affected by the nucleotide. Also the fitted Kd values were very similar (17±2 mM for ADP vs 15±2 mM for ATP). Similarly as for ATP, the terminal phosphate in ADP was the most affected part of the molecule upon addition of Cox13. However, Cox13 was not observed to precipitate together with ADP at the final titration step, possibly indicating a difference in the interaction between Cox13 and the two nucleotides at higher ligand concentrations.
In order to model how Cox13 interacts with ATP in our studies, the CSP data were used to direct docking of ATP to the Cox13 NMR structure. Due to the NMR ensemble’s large conformational heterogeneity, good binding scores were obtained in several Cox13 C-terminal structural arrangements (Fig. 6). The top-scoring models had extensive electrostatic contacts in common, between the negatively charged phosphate groups of ATP and the positively charged side-chains or polar backbone amide groups of residues in the loop preceding the α3 helix, e.g., R81, H83, and K85. The cluster with the best HADDOCK score (−91 ± 4) had well-defined electrostatic (−290 ± 40 kcal mol-1) and Van der Waals (−22 ± 4 kcal mol-1) energies. Three top-scoring models from this cluster are displayed in Fig. 6a. Although the top-scoring models never made simultaneous, direct contact with ATP using all residues with large CSPs, good overall docking scores could be obtained where different combinations of perturbed residues participated. This indicates either that the interaction can occur in several ways or that some residues with large CSPs are indirectly affected by ATP binding. Furthermore, it is possible that Cox13 forms other contacts with ATP in addition to those residues identified by NMR and 1H-15N-HSQC analysis. Our data is therefore most consistent with the presence of several possible Cox13-ATP complexes (for the isolated Cox13 subunit), but that the interactions are confined to the C-terminal region of Cox13, including a few charged residues.