Structural insights into complex I deficiency and assembly from 1 the disease-related ndufs4 -/- mouse 2

Abstract

like the N-module, these two subunits were not found to co-migrate with the rest of the Association of assembly factor NDUFAF2 with ndufs4 -/complex I 3 2 2 The structure of the class 3 ndufs4 -/complex I matches that from class 2 closely, except that 3 2 3 assembly factor NDUFAF2 is bound in the place occupied by subunit NDUFA12 in the wild- type complex ( Figure 5A), and the whole of the adjacent NDUFS6 subunit is absent, including the Zn-containing C-terminal domain present in class 2 (see Table 1 for a 3 2 6 comparison of complex I compositions). The presence of NDUFAF2 is consistent with our 3 2 7 complexomic analyses and mass spectrometry on the isolated enzyme, although the particle 3 2 8 classification data suggest that only ~14% of the population contains NDUFAF2, while the 3 2 9 other ~86% (classes 1 & 2) contain NDUFS6 instead: NDUFS6 and NDUFAF2 were not (in the wild-type enzyme) through a common three-strand β-sheet and adjacent short helix; 3 3 2 however, NDUFAF2 also contains a long 'diagonal' helix that is not present in NDUFA12, 3 3 3 which crosses the region occupied by the small central helix of NDUFS6 in the wild-type 3 3 4 enzyme ( Figures 5B and 1C). A 35-residue unresolved stretch of NDUFAF2 is followed by a β-hairpin. Finally, the low resolution map from ndufs4 -/kidney matches best to the heart 3 3 9 class 3 map, with density for the NDUFAF2 diagonal helix clearly visible, and no density 3 4 0 observed for NDUFS6 ( Figure EV5). The 12% of particles in class 1 of the ndufs4 -/complex were found to have an additional both contained substantially more peptides (and greater sequence coverage) for ACADVL 3 5 2 (Table S1). Considering the sequence similarity between the two proteins, we expect a 3 5 3 similar number of peptides from each if they are present in equimolar quantities, so these data suggest ACADVL is present in greater abundance than ACAD9. We thus chose to model the (Tyr, Trp and Phe) vary between the two proteins subsequently confirmed that ACADVL is 3 5 7 the major species present. The resolution achieved was sufficient to model the two FAD cofactors in the ACAD dimer interacting with core subunit ND1, and the other less deeply embedded, approaching subunit 3 6 6 NDUFA3. However, the position of the ACAD dimer is flexible relative to complex I and 3 6 7 their interface is not well resolved. Fitting both models into a composite map showed the 3 6 8 closest proximity residues to be ACADVL-Thr514 (monomer B) with NDUFA3-Ile34, and overlap spatially with the NDUFAF2 bound to class 3 we do not expect them to be mutually 3 7 5 exclusive, so it is likely that the expected fourth class that contains both has simply been 3 7 6 missed due to its low population (predicted ~1250 particles or 1.8%).

7 7
The location of ACAD binding in our map ( Figure 6A) is surprising: ACAD proteins are 2008), but in our structure the ACAD dimer appears to have 'slipped' around, into a position presence of a similarly-bound ACAD could be identified in the maps from the ndufs4 -/-3 8 2 kidney enzyme, which was not treated with a cross-linker, or in any of our structural data on ( Figure 6D). Therefore, we conclude the ACAD dimer is artefactually bound and stabilized in 3 8 6 the ndufs4 -/enzyme by the cross-linking procedure that was applied to stabilize the N- A, Lys556 and Lys557) and NDUFA8 (Lys105) are modelled ∼4 and ∼14 Å apart, 3 9 0 respectively, consistent with reaction with the BS 3 cross-linker ( Figure 6E). Continuous density resembling that expected for ubiquinone-9 was identified along the length 3 9 3 of the ubiquinone-binding channel in the class 2 ndufs4 -/enzyme ( Figure S5A). The density 3 9 4 feature observed is substantially longer than the BS 3 cross-linker molecule, and so we 3 9 5 conclude it originates from bound ubiquinone, even though the density for the ubiquinone 3 9 6 headgroup is poorly defined, precluding a confident model being built. Similar but less 3 9 7 contiguous densities were also observed in class 3, whereas the densities observed in class 1, 2020)), were fragmented and could not be interpreted. Although the class 2 ndufs4 -/- ubiquinone headgroup is not resolved, the residues around its binding site match closely to 4 0 1 the residues in the wild-type reference model ( Figure S5B). Comparison to the models for  with those that genuinely result in interrupted-assembly intermediates. However, as well as  Our kinetic assays revealed substantially lower NADH:ubiquinone oxidoreductase activity  6 (the penultimate, 4Fe-4S cluster in NDUFS8) ( Figure 1D). Furthermore, the small drop in proposed to increase electron leakage to O 2 , then further oxidative damage to the enzyme in a rate of H 2 O 2 production remaining two orders of magnitude slower than the rate of turnover.  The decreased catalytic activity may also result from the altered dynamics and conformational landscape that likely results from loss of the structural junction between 4 7 0 NDUFS4, NDUFS6 and NDUFA12, which pins N-module subunit NDUFS1 to NDUFA9. The NDUFA9 C-terminal domain, buttressed by the NDUFS7 C-terminus at the base of the these may change the ubiquinone-9/10 binding dynamics. It is possible that the same effect 4 7 7 explains the unexpected retention of ubiquinone-9 in the ndufs4 -/enzyme ( Figure S5). deactive state in our ndufs4 -/complex I, so these subtle changes may also alter the active-to- changes that occur during catalysis is required to further evaluate these proposals. In the Y. lipolytica ndufs4 -/strain, complex I levels decreased to ~40% of wild type, similar 4 8 8 to here, but the subunits of the N-module plus NDUFS6, NDUFA12 and NDUFAF2 were all 4 8 9 found to stay associated with the intact complex on BN-PAGE (Kahlhöfer et al, 2017).
Subsequently, a 4 Å-resolution cryo-EM map (PDB: 6RFS) showed that NDUFS6 and 4 9 1 NDUFA12 were bound as normal, but no density for assembly factor NDUFAF2 was  Table 1). Thus, unlike the mouse ndufs4 -/enzyme, the Y. NDUFA11 and the C-terminus of ND5 (part of the transverse helix and TMH16), and has 4 9 7 lost structural integrity for ND6-TMH4 and adjacent loops, as well as for the N-terminus of 4 9 8 NDUFS2. NDUFA11 has been reported to associate in the final stages of complex I assembly the same space as the loop in NDUFS6 that connects its two domains together, so that both 5 0 8 cannot bind together.

0 9
Assembly pathway Work on the assembly of both mammalian and yeast complexes I has led to a model in which (and perhaps NDUFS6) may also be required to stabilise the N-module, before NDUFAF2 is  the N-terminal domain of NDUFS6 (without NDUFS4, NDUFA12 or NDUFAF2, in class 2).

2 1
We do not observe the N-module bound alone. However, while the N-terminal domain of NDUFS6 binds between N-module subunits NDUFS1 and NDUFV2 and QP-subcomplex Arg20 to Tyr106) is bound (predominantly) to subunits NDUFS7, ND1 and NDUFS8 on the heel of the complex, whereas its C-terminal domain (modelled from Glu141 to Glu161) is 5 2 7 located on NDUFS1 in the N-module ( Figure 5B). There is no sturdy connection between 5 2 8 them, since residues 107 to 140 are disordered -and indeed this stretch of the sequence is firmly attached to the heel, its intrinsically disordered region allows the C-terminus to search 5 3 3 the surrounding space, capture the N-module, and then restrict its freedom to promote on the mature wild-type complex, but lacks the flexibly-linked β-hairpin anchor to NDUFS1.

3 7
Thus, our class 3 NDUFAF2-bound structure likely represents the first N-module-containing 5 3 8 species on the assembly pathway.

3 9
Our structures show that the three subunits required to complete the mature protein (NDUFS4, is ordered in our class 3 ndufs4 -/structure, but disordered in the structure of Y. lipolytica its C-terminus replaces the NDUFAF2 C-terminus on NDUFS1 ( Figure 5C). The connecting loop, which is ordered in NDUFA12 and interacts with the NDUFS4 β2-β3 loop and 1C), which must thus bind either after or along with it. Our class 2 ndufs4 -/structure, representing the majority of the particles in the heart complex, and containing the N-terminal can be assigned, not to structural effects, but to the near-complete absence of NDUFA12 in ndufs4 -/mouse tissues, embryonic fibroblasts and NDUFS4-mutated Leigh syndrome patient glutamatergic vestibular neurons, that are affected in Leigh syndrome (Gella et al, 2020).

6 3
Why NDUFA12 is also absent, as well as NDUFS4, is currently unclear, as well as 5  6  4 unfortunate, since it compounds the ndufs4 -/defect and its pathological consequences by 5 6 5 preventing the final stages in assembly. with decreased complex I catalysis (which results in decreased ATP production) causing 5 6 9 increased steady-state NADH and O 2 levels, which together may result in increased ROS 5 7 0 production. The discovery that a hypoxic environment mitigates the ndufs4 -/mouse Hirst, 2006), rather than from an increased molecular-level propensity of the ndufs4 -/-5 7 7 complex I to produce ROS. However, this mechanism suggests that hypoxia should benefit all complex I-linked mitochondrial diseases that similarly compromise the rate of catalysis, not be specific to a particular subset (including the ndufs4 -/family) or a specific genetic structures have also provided some intriguing hints for additional subtle molecular changes, including the presence of bound ubiquinone, changes to the active-deactive transition, and the 5 9 1 substantially decreased catalytic activity, which we cannot currently explain through hypoxia-rescue, and the balance of enzyme synthesis, assembly and degradation that occurs through complex feedback loops in mitochondrial homeostasis.

9 7
Materials and methods 5 9 8 Animals 5 9 9 All procedures were carried out in accordance with the UK Animals (Scientific Procedures) Preparation of mitochondrial membranes from mouse heart and kidney 6 0 7 Heart and kidney tissues were excised from ndufs4 -/and wild type mice, and mitochondria The hearts were diced, washed, then homogenized in 10 mL buffer AT per gram of tissue by 6 1 5 seven to ten strokes of a Potter-Elvehjem homogenizer fitted with a Teflon pestle at ~1,000 6 1 6 rpm. The homogenate was centrifuged (1,000 x g, 5 min), then the supernatant was 6 1 7 centrifuged (9,000 x g, 10 min) to collect the crude mitochondria. The kidneys were diced, added (1 tablet to 50 mL suspension) and aliquots were frozen and stored at -80 °C. suspensions were thawed on ice, sonicated using a Q700 Sonicator (Qsonica, USA), at 65% were visualized using colloidal Coomassie R250, or the gels de-stained with MilliQ water for 6 3 7 in-gel complex I activity assays using 100 μ M NADH and 1 mg mL −1 nitroblue tetrazolium    Germany), and eluted from a Superose 6 Increase 5/150GL size exclusion column (GE 6 7 9 Healthcare, UK) in 20 mM Tris-HCl (pH 7.14 at 20 °C), 150 mM NaCl and 0.05% DDM. heart samples for cryo-EM analyses, 2.5 mM BS 3 was added for 30 min. on ice immediately 6 8 2 before cryo-EM grid preparation. UltrAuFoil gold grids (R0.6/1, Quantifoil Micro Tools GmbH, Germany) were prepared as μ L of complex I solution (2.55 mg mL −1 for kidney CI, 3.82 mg mL −1 for heart CI) was 6 9 0 applied to each grid at 4 °C in 100% humidity and blotted for 9-10s at blotting force setting 6 9 1 -10, before the grid was frozen by plunging it into liquid ethane. The highest quality cryo-6 9 2 frozen grids were identified using a 200 keV Talos Arctica microscope. Datasets for 6 9 3 reconstruction were then collected using a Titan Krios (300 keV) microscope at the UK 6 9 4 National Electron Bio-Imaging Centre at the Diamond Light Source. oxidation. These analyses were performed in triplicate.

0 5
Common peptides amongst the replicates from ndufs4 -/and wild-type samples were identified for these common peptides were used to produce an abundance value for each protein or as a fraction of the wild-type value, with appropriate error propagation.  Supervision, Writing -review and editing. The authors declare that they have no conflict of interest. Cabrera-Orefice A, Yoga EG, Wirth C, Siegmund K, Zwicker K, Guerrero-Castillo S,    Rates of kinetic reactions for ndufs4 -/heart mitochondrial membranes as a proportion of the wild-type rates linking on BN-PAGE of the enzymes purified from heart; wild-type enzyme was not cross-linked (BS 3 0); 1 0 9 2 ndufs4 -/complex I was cross-linked with 0.25 mM BS 3 during solubilization from the membranes, and then run 1 0 9 3 as prepared (BS 3 1); ndufs4 -/complex I was reacted again with 2.5 mM BS 3 after size exclusion 1 0 9 4 chromatography (BS 3 2). The second reaction further increased the stability of the enzyme (compare to the 1 0 9 5 untreated sample in Figure 2F). which appear disordered in ndufs4 -/maps.  complex is shown in Figure 1C. C) Comparison of the structures for NDUFA12 (from wild-type PDB: 6ZR2) 1 1 1 5 and NDUFAF2 (from PDB: 8C2S (mouse) and PDB: 7RFQ (Y. lipolytica)) aligned to one another. protein were normalized to the sum of peptide areas for VDAC1, VDAC 2 and VDAC3, and displayed as the Lys-Lys distances between ACADVL and NDUFA8 that may be responsible for stabilization of ACAD binding  masses of complex I-related bands specific to ndufs4 -/membranes of ~800 (QP subcomplex) and ~200 kDa (N 1 1 4 6 module) were estimated using the migration and masses of known complexes (inset). FeCN. Due to the scarcity of materials, the assays in panels A and B were performed on different membrane red. For proteins labelled in brackets individual peptide scores were below the 95% threshold but their protein 1 1 5 9 scores were above the peptide threshold. Information on protein identification is in Tables S2 and S3.  wild-type heart complex I (PDB: 6ZR2, for NDUFS6, NDUFA7 and NDUFA9) and the class 3 model (for 1 1 6 6 NDUFAF2) from were fitted into each density map using the fit-in-map function in Chimera X. The density 37 the ndufs4 -/heart complex I class 1, 2 and 3 maps; and the ndufs4 -/kidney complex I map, with map thresholds 1 1 7 0 of 0.027, 1.76, 2.18, 1.52 and 1.12, respectively. Suplementary Movie 1. Relative movement between the wild-type and ndufs4 -/heart class 2 models, and 1 1 7 2 motion within the cryo-EM density of ndufs4 -/heart class 2 from 3DVA analyses.