Increased supraorganization of respiratory complexes is a dynamic multistep remodelling in response to proteostasis stress

An immediate cellular response to any kind of stress is to shut down energy-expensive metabolic processes. For example, protein synthesis is highly energy expensive and is globally reduced in response to stress (Lane and Martin, 2010). Nevertheless, using energy for cap-independent translation of stress chaperones, refolding of unfolded proteins by ATP-consuming chaperones or degradation of misfolded proteins by the ubiquitin–proteasome system (UPS) remains indispensable. Therefore, preserving optimal energy-production machineries and investing available energy in adaptive mechanisms is essential to survive stress and to mount defence responses.

Oxidative phosphorylation is the most efficient energy generation pathway in cells. Four respiratory complexes (RCs), NADH–ubiquinone oxidoreductase (complex I; CI), succinate–ubiquinone oxidoreductase (complex II; CII), ubiquinone–cytochrome c oxidoreductase (complex III; CIII) and cytochrome c oxidase (complex IV; CIV), generate an electrochemical gradient across the inner mitochondrial membrane to convert ADP to ATP by ATP synthase (complex V; CV). RCs are composed of proteins of dual genetic origin and assemble in the mitochondrial matrix. Multiple nuclear-encoded RC subunits associate with mitochondria-encoded subunits into subcomplexes, followed by formation of holocomplexes (Fernández-Vizarra and Zeviani, 2015; Guerrero-Castillo et al., 2017; Signes and Fernandez-Vizarra, 2018; Vidoni et al., 2017). Cellular protein quality control capacity prevents accumulation of unassembled subunits and controls fidelity of RC assembly by selectively preserving founding subcomplexes for on-demand assembly. We have recently reported that unassembled RC subunits are aggregation-prone proteins (Rawat et al., 2019). Interestingly, despite aggregation of subunits, the electron transfer capacity of individual RCs remains unperturbed during short-term proteasome inhibition.

RCs are functionally independent in terms of substrate utilization. However, a large proportion of CI, CIII and CIV cooperate as interacting enzymes. Such supraorganizations of RCs are known as ‘respiratory supercomplexes’ or ‘respirasomes’ (Gu et al., 2016; Guo et al., 2018b; Schagger and Pfeiffer, 2000). CII is predominantly present as tetramer, but contradicting evidence for CII acting as part of respiratory megacomplexes also exists (Guo et al., 2018a; Jha et al., 2016). CV exists as dimeric and oligomeric complexes along the edges of cristae (Bezawork-Geleta et al., 2018; Wittig and Schägger, 2009). Association of individual RCs into supercomplexes is dynamic and depends on the metabolic status of the cell (Acín-Pérez et al., 2008; García-Poyatos et al., 2020; Guarás et al., 2016; Lapuente-Brun et al., 2013; Moreno-Lastres et al., 2012; Porras and Bai, 2015).

Studies of the plasticity of RCs have been largely restricted to blue native PAGE (BN–PAGE) followed by western blotting, which reports a distribution of 1–2 subunits per complex. The switch of RC subunit organization from subcomplexes to holocomplexes and supercomplexes is not resolved in these experiments (Balsa et al., 2019; Greggio et al., 2017). Complexome profiling combines BN–PAGE and mass spectrometry to provide extensive insights into RC biosynthesis (Heide et al., 2012; Saada et al., 2009; Wessels et al., 2009). Here, we established an iBAQ (intensity-based absolute quantification) and SILAC (stable isotope labelling by amino acids in cell culture) based two-dimensional complexome profiling protocol for comprehensive mapping of the stress-responsive relative distribution of each and every RC subunit between different RC assemblies. Using this analysis strategy, we find that formation of high molecular weight supercomplexes (SCs) is favoured over free individual holocomplexes (HCs) and subcomplexes (SubCs) during proteostasis stress. We identify that this dynamic self-remodelling; referred to as increased supraorganization of respiratory complexes (iSRC), is an exclusive capacity of RCs among the mitochondrial protein complexes. The mechanism is fine-tuned via partial destabilization and re-stabilization of SCs. The core composition of CI and CIII subunits in the destabilized SCs remains stable, whereas CIV subunits are comparatively flexible. We propose that the quinary interactions at the partially disintegrated SCs during early proteostatic stress facilitate their reassembly to ensure functional protection of these multimeric enzymes.

We have recently reported RC subunits form reversible aggregates in Neuro2a cells during short-term (8 h) proteasome inhibition by MG132; a widely used proteasome inhibitor with mild non-specific effects on calpains and cathepsins at very high concentrations (Kisselev and Goldberg, 2001; Rawat et al., 2019; Tsubuki et al., 1996). Interestingly, mitochondrial morphology was not perturbed, membrane potential and CI- and CII-dependent electron transfer capacity was not impaired, and mitochondrial import of RC-subunits was not blocked in these cells. However, levels of reactive oxygen species were increased by 2-fold and an ∼10% drop in oxygen consumption by CIV was observed (Rawat et al., 2019).

RC subunits inside mitochondria may have three fates during proteasome inhibition. They may remain soluble as individual subunits and as part of RCs, precipitate into the insoluble fraction or be degraded by other mitochondrial proteases or via mitophagy. To investigate, we performed SILAC-based quantification of mitochondrial fractions of Neuro2a cells after MG132 treatment (5 µM) for 8 h and 24 h. Identifying mitochondria-encoded RC subunits by SDS–PAGE and mass spectrometry is challenging due to their high hydrophobicity, aggregation propensity and low abundance (Attardi and Ching, 1979; Wang et al., 2012). Nevertheless, in addition to the nuclear-encoded subunits, we were able to identify several mitochondria-encoded membrane-associated subunits for all RCs. Abundance of RC subunits remained unaltered or slightly increased in the total and soluble fraction after 8 h of MG132 treatment, confirming that in general mitochondrial import was not impaired and a large fraction of subunits inside mitochondria remained soluble (Fig. S1A; see Table S1A–C for mitochondrial total, soluble and insoluble fractions, respectively). Simultaneously, the abundances of several CI, CII and two CIII subunits were increased in the insoluble pellet, suggesting no active turnover for these unstable aggregation-prone subunits inside mitochondria at 8 h of MG132 treatment. Exceptionally, the presence of CIV subunits was not increased, but rather was reduced, in the insoluble fraction at 8 h of treatment. Depletion of many RC subunits in the soluble fraction at 24 h of MG132 treatment was prominent, as indicated in the heatmap by increasing blue intensity (Fig. S1A). Several CIV subunits were depleted in the total, soluble and insoluble mitochondrial fractions at 24 h of MG132 treatment, indicating multiple possibilities including lack of import and/or degradation of unstable subunits by mitochondrial proteases (Ruan et al., 2017), or active autophagy in Neuro2a cells during prolonged proteasome inhibition (Rawat et al., 2019).

Because only partial dysfunction of RCs was observed, despite accumulation of a fraction of RC subunits in the insoluble mitochondrial pellets, we performed BN–PAGE followed by western blotting of the soluble mitochondrial extract to investigate the structural and functional integrity of soluble RCs during proteasome inhibition (Jha et al., 2016). Probing for the CI subunit Ndufs1 and the CIV subunit MtCo1 revealed the presence of CI and CIV in multiple RC assemblies with different masses (Fig. 1A). The band at ∼800 kDa in the Ndufs1 blot indicated the CI HC, while the upper bands represented CI SCs (Fig. 1A). The bands in the ∼350 kDa region and above in the MtCo1 blot were attributed to CIV HC and CIV SCs, respectively. Compared to the Hsp60 loading control, the abundance of Ndufs1 in the CI HC and CI SCs remained similar at 8 h, whereas it was reduced at 24 h of MG132 treatment (Fig. 1A). Similarly, a reduction of MtCo1 levels was observed for the CIV HC and CIV SCs at 24 h MG132 treatment. Depletion of many CI and CIV subunits was also observed in the mitochondrial soluble fraction at 24 h (Fig. S1A). Interestingly, despite this depletion, Ndufs1 abundance in CI SCs was relatively increased over that in CI HCs at 24 h of MG132 treatment (Fig. 1B). However, distribution of MtCo1 between CIV HC and CIV SCs was very dynamic between experiments.

Ndufs1 and MtCo1 bands in SubCs (below ∼800 kDa and ∼350 kDa, respectively) were not detected in western blots (Fig. 1A). Furthermore, probing western blots for a single subunit is unable to provide a comprehensive distribution profile of RC subunits during stress and can only partially report on the plasticity of RCs. Therefore, we established an iBAQ-and-SILAC-based complexome profiling protocol to obtain subunit-wise distribution map across SCs, HCs and SubCs. BN–PAGE gels were divided into 20 slices according to Coomassie staining, keeping the HC and SC bands separate, and quantitative mass spectrometry was performed (Fig. 1C). In this 2D complexome profiling strategy, SILAC ratios report increases or decreases in abundance of RC subunits within the same BN–PAGE gel slice due to MG132 treatment. Simultaneously, relative iBAQ intensities of isotopic peptides across the slices provide distribution profiles of SCs, HCs and SubCs (Fig. 1C).

First, we calculated slice-wise absolute abundance of RC subunits in unstressed Neuro2a cells, using a previously published iBAQ algorithm (Schwanhäusser et al., 2011). Slice 7 (∼800 kDa) contained CI subunits with moderate iBAQ values suggesting the presence of fully assembled CI HC (Fig. 1D; Table S2A). CI interacts with the CIII dimer (CIII₂) to form a supercomplex [SC(I+III₂)], or with both CIII₂ and CIV monomers or dimers to constitute larger supercomplexes [SC(I+III₂+IV), (I+III₂+IV₂), etc.] (Jha et al., 2016; Milenkovic et al., 2017). Higher cumulative intensity of CI and CIII subunits in slice 5 indicated that the majority of CI in Neuro2a cells was associated with CIII (Fig. 1D; Table S2A,B). CI, CIII and CIV subunits were also identified in slice 4 (∼1000 kDa), suggesting the presence of SC(I+III₂+IV) but in lesser abundance than SC(I+III₂) (Fig. 1D; Table S2A,B). CI, CIII and CIV subunits were either absent or present in low intensities in the highest molecular weight slices (slices 1–3), suggesting the absence of further high molecular weight SCs in Neuro2a cells (Fig. 1D).

CV subunits were most abundant in slice 9 (∼700 kDa), corresponding to the CV HC. CV also forms dimers and oligomers (He et al., 2018) and has been previously identified in higher molecular weight regions of BN–PAGE gels (Müller et al., 2016). We also identified CV subunits in high molecular weight assemblies (Fig. 1D; Table S2B). Abundance of both CIII and CIV in slice 9 suggested the presence of SC(III₂+IV) (Fig. 1D). CIII and CIV subunits were also identified in slice 11, representing dimeric HCs (Jha et al., 2016; Lapuente-Brun et al., 2013). Monomeric CIV HC was found in slices 14 and 15, whereas the CII HC was identified in slice 16 (Fig. 1D). Slices 9–15 contained multiple Q- and P_D-module components of CI, indicating the presence of intermediate CI SubCs (Guerrero-Castillo et al., 2017). Core subunits of the CI N module (Ndufv1, Ndufv2 and Ndufs1) were found in slices 16–20 (∼150–250 kDa) along with several Q- and P_D-module components, suggesting multiple small SubCs (Fig. 1D). CIII and CV SubCs were also present in slices 13–20, as was evident from the abundance of their constituting subunits (Fig. 1D) (Fernández-Vizarra and Zeviani, 2015; Vidoni et al., 2017). Taken together, the relative distribution of iBAQ intensities across the BN–PAGE gels suggested that most of the CI subunits were present within SCs, whereas CIII and CIV subunits were also abundant in HCs (Fig. 1E). CII and CV subunits were distributed between HCs and SubCs, although 25.4% of CV subunits were also found to be part of multimeric organizations.

Multiple N- and Q-module CI SubCs of molecular weight less than ∼700 kDa (below slice 9 of the BN–PAGE gels) were found to be reduced in abundance upon MG132 treatment (Fig. 2A). Small P_D-module SubCs that ran below slice 16 (molecular weight below ∼250 kDa) were also reduced in abundance, whereas intermediate P_D SubCs (between slices 9–16) were comparatively increased in abundance at 8 h of treatment (Fig. 2A; Table S2C). Uqcrc1, Uqcrc2 and Cyc1 together form a CIII SubC (Fernández-Vizarra and Zeviani, 2015). Both Uqcrc1 and Uqcrc2 were reduced in abundance in slices 19 and 20. CIV SubCs composed of Cox4i1 and Cox5a and of Cox5b and Cox6c (Signes and Fernandez-Vizarra, 2018) were depleted in slices 19 and 20. The CV stator stalk SubC composed of Atp5f1 (also known as ATP5PB), Atp5l (also known as ATP5MG) and Atp5i (also known as ATP5ME) (Fujikawa et al., 2015) in slices 17 and 18 was also reduced in abundance. CII is known to be constituted through independent maturation of Sdha, Sdhb and an Sdhc–Sdhd dimer (Signes and Fernandez-Vizarra, 2018). Sdha, Sdhc and Sdhd in slice 19 were reduced in abundance following MG132 treatment (Fig. 2A; Table S2D).

SILAC ratios for all CII subunits were increased in slices 15 and 16 at 8 h of MG132 treatment, suggesting consolidation of the CII HC (Fig. 2A). Similarly, CI subunits were either increased in abundance or remained unchanged in HCs and SCs found in slice 7 and above. Most of the CI and CIII subunits composing SC(I+III₂) at slice 5 were slightly increased in abundance or remained unchanged at 8 h of MG132 treatment. CI, CIII and CIV subunits were prominently increased in abundance at slice 4 in the 8 h sample, suggesting increased abundance of SC(I+III₂+IV) (Fig. 2A).

In contrast to short-term proteasome inhibition (8 h), abundance of CI HC and CI SCs was reduced at 24 h MG132 treatment (Fig. 2A). Subunits in CIII and CV HCs at slice 11 and 9 remained unchanged or marginally reduced in abundance. Subunits of CII and a few CIV subunits displayed increased abundance in HCs found in slices 15–17 and 14–15, respectively. The abundance of most of the CIII and CIV subunits remained unchanged in slice 9, suggesting protection of SC(III₂+IV) (Fig. 2A).

A significant increase in the relative distribution of CI subunits into SCs with increasing time of MG132 treatment was confirmed as per the iBAQ intensity profile of isotopic peptides (Fig. 2B; Fig. S2A). Abundance of the CI HC was found to be slightly increased at 8 h of MG132 treatment, but dropped with time. Simultaneously, a reduction in the abundance of CI subunits in SubCs was prominent upon MG132 treatment. The increased abundance of SCs and drop in abundance of SubCs was most prominent for the N and Q modules (Fig. S2B), which protrude out from the inner mitochondrial membrane (IMM) towards the hydrophilic environment of matrix (Letts et al., 2016). The membrane-resident P module was comparatively unchanged.

Increased incorporation of CII and CV subunits into HCs due to proteasome inhibition was observed with a concomitant drop in abundance of their SubCs (Fig. 2B; Fig. S2A). On the other hand, the relative abundance of CIII HC was significantly reduced, with simultaneous increase in abundance of SCs (Fig. 2B). The association of CIII and CIV subunits into SC(III₂+IV) identified in slice 9 was increasingly stronger with MG132 treatment (Fig. S2A). Incorporation of CI, CIII and CIV subunits into SCs in slices 4 and 5 was also increased. Thus, despite reduction in the abundance of RC subunits in SCs during longer proteasome inhibition (24 h), as depicted by the SILAC heatmap, favoured distribution of soluble subunits into SCs at the expense of subassemblies was revealed by iBAQ analysis. Hereafter, we term this phenomenon increased supraorganization of respiratory complexes (iSRC).

Next, we performed in-gel CI and CIV activity assays to investigate the function of the multimeric RC enzymes (Fig. 2C). CI HC activity was greater than that of the control already by 1 h of MG132 treatment, suggesting better substrate utilization upon stress (Fig. 2C,D; Fig. S2C). SC(I+III₂) was more abundant than SC(I+III₂+IV) (Fig. 1E) but its activity was not increased, except at 8 h MG132 treatment (Fig. 2C,D). The intensity of the SC(I+III₂+IV) activity band gradually increased with increasing duration of stress. Maximum activity was achieved by 8 h, when increased integration of RC subunits into SCs was also observed (Fig. 2C,D). Activity band intensities for CI HC and SC(I+III₂+IV) remained higher than those of control unstressed cells, even at 24 h of MG132 treatment (Fig. 2C,D). Compared to CI activity, CIV enzyme activity was marginally increased in SC(III₂+IV) (Fig. 2C,D). A slight drop in activity for both the CIV HC and SC was also observed at 4 h of MG132 treatment. Taken together, the functional analyses suggested that, despite extensive remodelling of RC organization (Fig. 2A), CI and CIV HC and SCs remained optimized to protect catalytic activities during prolonged proteasome inhibition.

Many stress-triggered proteome reorganization mechanisms are instantaneous. For example, small stress granules emerge in the cytoplasm within 3–4 min of stress and then progressively fuse with each other to form bigger granules (Nadezhdina et al., 2010). Similarly, increased CI activity at SC(I+III₂+IV) after 1 h of MG132 treatment indicated immediate upregulation of SCs after stress. Therefore, we performed complexome profiling experiments after 1 and 4 h MG132 treatment in Neuro2a cells to investigate the kinetics of iSRC. Surprisingly, unlike the effects of 8 h MG132 treatment (Fig. 2B), increase of SCs was not apparent at 1 and 4 h of treatment. Instead, distribution of RC subunits into CIII and CV SCs was reduced (Fig. 3A; Fig. S3A and Table S3A–D). The abundance of CI HC was reduced, with a concomitant increase in the abundance of SubCs (Fig. 3A). Distribution of CI subunits was marginally increased in Q- and P_D-module SubCs (Fig. S3B).

Similar to the changes observed in the iBAQ data, the SILAC heatmap for slices 4–7 suggested that the abundance of CI HC and CI SCs was reduced during early stress; more prominently at 4 h MG132 treatment (Fig. S3C). The abundance of SubCs remained unaltered or was increased below slices 13–14 at 1 h of MG132 treatment. P_D-module SubCs in slices 14–19 and Q-module SubCs in slices 19 and 20 were increased in abundance at 4 h treatment (Fig. S3C). The abundance of CIII HC in slices 11 and 12 was also increased, although CIII SCs and SubCs were reduced in abundance at 4 h (Fig. S3C). Overall, a decrease in abundance of CIV subunits was observed throughout the SILAC data (Fig. S3B). Interestingly, CIV subunits were also found to be depleted in the total mitochondrial fraction (Fig. S1A). Slice-wise line profiles of SILAC fold changes (Fig. S3D) indicated a drop in abundance of CI subunits in the SCs and HC at 4 h of MG132 treatment, whereas subunits within SubCs remained comparatively stabilized. With time, CI SCs increased in abundance at the expense of SubCs. Similarly, the CII HC in slice 16 and SCs between slices 12–15 were initially destabilized and then re-stabilized at the expense of SubCs after longer proteasome inhibition. CIII subunits were not much destabilized at 4 h treatment, whereas CIV subunits were depleted at that time point. The profile of CV subunits remained unchanged after 4 h MG132 treatment. Overall, high molecular weight RCs were destabilized, while low molecular weight SubCs were protected during early stress. Despite this partial disintegration, CI and CIV HC and SCs were optimized to protect RC function by 1 and 4 h of stress (Fig. 2C,D).

Furthermore, the timecourse iBAQ data revealed disproportionate stoichiometry of individual RCs within different SCs at early time points of MG132 treatment (Fig. 3B). SC(I+III₂+IV) in slice 4 was most dynamic. The abundance of CI and CIII increased proportionately in slice 4 at 8 h of MG132 treatment, but the fold increase in CIV abundance was ∼30% in excess (Fig. 3B). This suggested that, although a large population of SCs remained at (I+III₂+IV) stoichiometry, a subpopulation of SCs with increased CIV units emerged at this stage. Disproportionate CIV distribution at 1 and 4 h was also observed in SC(III₂+IV) at slice 9 (Fig. 3B).

Interestingly, cryo-electron microscopy revealed two distinct architectures for SC(I+III₂+IV); in the ‘tight’ structure, CIV is attached to both CI and CIII, whereas the ‘loose’ structure refers to contact between CI and CIV only (Letts et al., 2016). Two separate CIV monomers associated with (I+III₂) in situ have also been reported in mammalian cells (Davies et al., 2017). Our data indicated that CIV is the most dynamic RC in terms of SC constitution. However, relative distribution profiles for CI, III and IV absolutely overlapped in slices 4, 5 and 9 at 24 h of MG132 treatment, suggesting that CI, III and IV subunits were relatively increased in abundance and consolidated at SC(I+III₂+IV), (I+III₂) and (III₂+IV), as per steady-state stoichiometry (Fig. 3B). Thus, in contrast to rigid super-quaternary structures with fixed stoichiometry, we observed plasticity and inequivalence of constituent RCs suggesting ‘quinary’ SC organizations at 1 and 4 h (Gierasch and Gershenson, 2009; Hingorani and Gierasch, 2014; McConkey, 1982). Resembling the cooperative unfolding and refolding model for protein folding (Englander and Mayne, 2017; Ribeiro et al., 2018), these dynamic SC ensembles interconvert at multiple stages of assembly or disassembly to re-establish the steady-state stoichiometry of SC(I+III₂+IV) and (III₂+IV) with time.

It is proposed that during RC biogenesis, CI, CIII and CIV subunits are first incorporated into HCs and then integrated into SCs (Fig. 3C) (Guerrero-Castillo et al., 2017). This mechanism might remain consistent for iSRC. Alternatively, SubCs might directly integrate into SCs (Fig. 3C). Only 12.4% of CI subunits were found within HCs in Neuro2a cells, suggesting it to be a transient CI assembly that rapidly integrated into SCs or dissociated into SubCs (Fig. 1E). The CI HC was reduced in abundance at 1 h MG132 treatment, with a concomitant increase in abundance of SubCs, resulting in decreased CI HC:SubC ratio (Fig. 3A,D). The CI SC:HC ratio was increased simultaneously, as the relative distribution of CI subunits in SCs remained unchanged. A further drop in CI HC abundance was observed at 4 h (Fig. 3A). Simultaneously, abundance of CI SubCs increased further, and the HC:SubC ratio decreased sharply (Fig. 3D). The abundance of CI SC and HC were proportionately increased at 8 h (Fig. 3E). The CI SC:HC ratio remained unchanged; however, an increased HC:SubC ratio suggested incorporation of SubCs into the HC (Fig. 3D). Integration of CI HC into SCs increased at 24 h of MG132 treatment, with a concomitant drop in abundance of CI HC (Fig. 3E).

In contrast to CI, only 5% of CIII subunits were found in SubCs (Fig. 1E).

Increase in the CIII HC:SubC iBAQ ratio at 1 and 4 h MG132 treatment suggested increased incorporation of subunits into HC (Fig. 3A,F). A decrease in the CIII SC:HC ratio indicated simultaneous disintegration of SCs. The CIII SC:HC ratio was increased at 8 h of MG132 treatment, whereas the HC:SubC ratio remained close to that of the control (Fig. 3F). Conversely, unchanged SC:HC and HC:SubC ratios suggested that the relative distribution of CIV subunits between SubCs, HCs and SCs remained similar to that in the control at both 1 and 4 h of MG132 treatment (Fig. 3G). Both SC:HC and HC:SubC ratios were increased for CIV at 8 h, indicating synchronized incorporation of subunits into higher assemblies. Thus, iSRC followed a two-step SubC to HC to SC mechanism similar to that occurring during biogenesis; however, each complex showed distinct integration kinetics.

To probe the dynamics of RC organization further, we performed limited proteolysis of mitochondria-enriched fractions from MG132-treated cells and blotted for the core N-module subunit Ndufs1. Ndufs1 is a CI peripheral arm component, which forms a SubC with Q-module core subunit Ndufa2 at early stages of CI biogenesis (Guerrero-Castillo et al., 2017). Both N and Q modules protrude out towards the matrix, and the N module displays a lateral movement in respirasome structures (Gu et al., 2016). The abundance of Ndufs1 was also found to be increased in the mitochondrial insoluble fraction after 8 h MG132 treatment, confirming its instability upon stress (Fig. S1A). As plotted in Fig. 3H, Ndufs1 was also more accessible to protease digestion in MG132-treated fractions compared to in control unstressed cells. Rapid digestion of Ndufs1 was observed at 1 h MG132 treatment, as evident from a reduction in band intensity after 5-min protease treatment, suggesting maximum conformational flexibility at that time. Taken together, the kinetics experiments revealed that iSRC was not an instantaneous proteome remodelling. Rather, RC subunits underwent extensive reorganization before their increased incorporation into SCs.

Stress-responsive proteome reorganizations are generally reversible. Stress granules disappear within hours of stress recovery (Kedersha and Anderson, 2007). Protein aggregates also resolubilize after withdrawal of heat stress (Wallace et al., 2015). Similarly, aggregates of overexpressed CI subunit Ndufa5 emerged at 8 h MG132 treatment and disappeared after MG132 withdrawal (Fig. 4A). The intensity of CI activity bands also decreased by 16 h after MG132 withdrawal; but did not return to the pre-stress state (Fig. 4B,C); in particular, SC(I+III₂+IV) remained at an increased functional state.

Decrease in the abundance of SubCs and simultaneous increase in HC and SC abundance was a signature of iSRC after 8 h MG132 treatment (Fig. 2). This trend was reversed during recovery. As per iBAQ assessment of isotopic peptides, distribution of subunits in CI, CII and CIV SCs was restored to normal equilibrium, whereas CIII SCs and multimeric CV were reduced compared to levels in unstressed cells (Fig. 4D; Fig. S4A, Table S4A,B). Increase in CII HC levels and decrease in CIII HC observed during iSRC (Fig. 2) was reversed after MG132 withdrawal (Fig. 4A; Fig. S4A). The SILAC data also confirmed comparatively more abundant SubCs than HCs and SCs during recovery (Fig. S4B; Table S4C,D). Abundance of subunits in the CIII HC (slices 11 and 12) was increased, whereas the abundances of SCs were reduced (Fig. S4B). CIII SubCs were also increased in abundance during early recovery. As depicted in the heatmap, abundances of higher-order RCs were more reduced at 16 h recovery compared to 8 h recovery (Fig. S4B). Increased catalytic activity of SC(I+III₂+IV) at 16 h recovery (Fig. 4C) suggested the continued presence of functionally optimized SCs, despite reduction in quantity (Fig. S4B, slice 4).

SC dissociation was also observed during early stress (1 and 4 h MG132); however, we found subtle differences during stress recovery. CI Q module was redistributed between HC and SubCs during early stress (Fig. S3B). However, during recovery, the CI N module was increased in abundance, while the P_D module was prominently decreased in abundance in SCs (Fig. S4C). Abundance of CI HC was reduced during early stress (Fig. 3A). In contrast, CI HC was restored to equilibrium during recovery (Fig. 4D). However, CI SC:HC and HC:SubC iBAQ ratios remained similar during early stress and during the recovery period (Fig. 4E). CIII SC:HC and HC:SubC ratios reached close to equilibrium at 16 h recovery (Fig. 4F). Distribution of CIV subunits between SubCs, HC and SCs maintained equilibrium during early stress, whereas the HC:SubC ratio decreased during recovery (Fig. 4D–G). Although the abundance of CIII was marginally under-proportion in SC(I+III₂+IV) and (I+III₂), relative changes in the overall distribution for RCs suggested restoration of steady-state equilibrium during recovery (Fig. 4H). A disproportionate relative increase in CIV abundance at slice 9 suggested extensive reorganization of SC(III₂+IV). Thus, the mechanism of iSRC was distinct for individual RCs in terms of kinetics of integration into SCs during stress and dissociation during recovery.

Although plasticity of SCs in the IMM has been proposed to be mediated via quinary interactions (Blaza et al., 2014; Cohen and Pielak, 2016), assembly factors critically contribute to formation of individual RCs and SCs. Many CI-assembly factors that transiently associate with SubCs have been described (Guerrero-Castillo et al., 2017). Among them, Timmdc1, Ndufaf3 and Ndufaf4 have been proposed to be associated with Q- and P_P-module SubCs and were identified in slices 13 and 14 (∼480 kDa) and in slice 20 (∼66 kDa). We observed partial depletion of these assembly factors with increasing duration of MG132 treatment, and observed that Ndufaf4 abundance was increased after 16 h recovery (Fig. 5A,B; Table S5A,B). P_P-module-assembly factors Acad9, Tmem186 and Tmem126b were distributed close to CI HC in slices 6–9 (∼750 kDa) and in intermediate SubCs (slices 12 and 13, ∼500 kDa; Fig. 5A) and were decreased in abundance at 24 h MG132 treatment (Fig. 5B). Only Tmem126b was found to increase in abundance during recovery. Ecsit and Ndufaf1 also contribute to form P_P-module SubCs. They were identified across the same gel slices but remained comparatively unchanged in abundance with MG132 treatment. Ndufaf2 plays a role in the final stages of CI HC formation and was prominently increased in abundance at 8 h MG132 treatment in slice 8, suggesting a possible involvement in improving CI HC assembly. CIII assembly chaperones Uqcc1 and Uqcc2 (Fernández-Vizarra and Zeviani, 2018) were found to have increased abundance in slices 15–20 during stress, where CIII SubCs were found (Fig. 5B). Abundance of the CIV-assembly factors Higd1a (Vidoni et al., 2017) and Lrpprc (Mourier et al., 2014) remained unchanged in slices 14 and 15, along with the CIV HC. Supercomplex assembly chaperone Cox7a2l (SCAF1) (Lobo-Jarne et al., 2018) was identified with CIV SCs in slices 4 to 9 of the BN–PAGE gels and showed depletion following MG132 treatment. Interestingly, Cox7a2l abundance was increased during recovery at slices 10 and 11 where the CIV dimer was identified (Fig. 5B; Table S5A,B). Thus, our data revealed that the ‘quinary-state’ RC organizations (Powers and Balch, 2013; Roth and Balch, 2013) associate and dissociate with an array of assembly chaperones at different stages of stress to stabilize their soluble forms and to favour stage-specific stoichiometry (Fig. 5C).

In addition to RCs, many other protein complexes inside mitochondria are related to energy production. For example, eight Krebs cycle enzymes are postulated to form a supramolecular complex via weak non-covalent interactions and operate together at the mitochondrial matrix to generate chemical energy in the form of ATP, NADH and FADH₂ (Bulutoglu et al., 2016; Wu and Minteer, 2015). Among these, citrate synthase (Cs), mitochondrial malate dehydrogenase (mMdh, also known as MDH2) and aconitase (Aco, also known as ACO2) form a homogeneous structure in vitro termed as a metabolon (Bulutoglu et al., 2016). These proteins were identified together between slices 14 and 19 of the BN–PAGE gels (Fig. 6A). Although aconitase was found to be precipitated into the mitochondrial insoluble fraction at 8 h MG132 (Fig. S5A; Table S7A), we did not see any deviation in the overall relative distribution of this metabolon during stress, except at 24 h MG132 treatment (Fig. 6B). The deviation at 24 h treatment was largely contributed by Cs, which was increasingly detected in slices 14 and 16 at this timepoint (Table S6A). Interestingly, CII also contributes to the Krebs cycle and was detected in the same gel slices as the metabolon (slices 15–19). Oligomeric CII at slices 15–17 was found to be increased in abundance during proteasome inhibition, at the expense of its SubCs in slices 18 and 19 (Fig. S2). The other Krebs cycle enzymes, namely Idh2 (isocitrate dehydrogenase 2), Idh3 (isocitrate dehydrogenase 3), α-Kgdh (alpha ketoglutarate dehydrogenase, also known as OGDH) and Scs (succinyl-CoA synthetase) were distributed in different slices of the BN–PAGE gels, suggesting that their proposed supraorganization (Bulutoglu et al., 2016; Wu and Minteer, 2015) was not detected in our experiments (Fig. S5B; Table S6B).

Pyruvate dehydrogenase complex (PDC) supplies acetyl-CoA to the Krebs cycle by catalyzing the production of acetyl-CoA from pyruvate. Mammalian PDC is a very large supraorganized complex of ∼7–10 MDa and therefore is not readily identifiable by standard protocols of BN–PAGE (Henderson et al., 2000; Smolle et al., 2006). We inconsistently identified some subunits of the PDC in slice 1 (Fig. 6C). However, two subunits of the PDC SubC E1 were always quantified between slices 14 and 19. Unlike RC SubCs, the relative abundance profile of PDC-E1 did not deplete with time but remained unchanged during the course of stress (Fig. 6D; Table S6C). Interestingly, multiple PDC subunits also precipitated in the mitochondrial insoluble fraction at 8 h MG132 treatment (Fig. S5C; Table S7B).

The mitochondrial ribosome is another big protein complex inside mitochondria and contains rRNA as important structural component. Studies on kinetics of mitoribosome assembly suggest rapid exchange of newly synthesized subunits with the existing mitoribosomes (Bogenhagen et al., 2018). Multiple subassemblies of mitoribosomal proteins (MRPs) composed of old and newly synthesized proteins also exist inside mitochondria. Excess MRPs are degraded by the proteasome (Sung et al., 2016). We found many MRPs to be increased in abundance in the insoluble fraction after proteasome inhibition (Fig. S5D, Table S7C). The molecular masses of the fully assembled mitoribosome and its 39S and 28S subunits are ∼3 MDa (slice 3 and above), ∼1.6 MDa (slices 4 and 5) and ∼1 MDa (slices 7 and 8), respectively (O'Brien, 2003; Sharma et al., 2003). We did not consistently observe intact mitoribosomes, or 39S or 28S subunits in our SILAC-based BN–PAGE experiments, possibly due to the absence of RNase inhibitors in the protocol (Fig. 6E). Although many subassemblies were detected, we did not find uniform relative abundance profiles of these assemblies between experiments, suggesting their variable stability during fractionation. Despite this discrepancy in biochemistry, composition of these identified assemblies remained unchanged throughout the stress period, as indicated by the relative abundance profiles of the isotopic peptides. The only exception was a 39S subassembly at ∼700 kDa that showed reduced abundance at 1 and 4 h of MG132 treatment compared to the levels in the untreated control (Fig. 6F; Table S6D,E). Thus, stress-related dynamic changes in the distribution of relative abundance were consistent and reproducible for only RC assemblies and was not observed for other mitochondrial protein complexes, despite increased aggregation propensity of their subunits.

Increased flexibility and protease sensitivity of Ndufs1 at early stages of proteasome inhibition prompted us to investigate iSRC in other protein-destabilizing stresses. Active degradation machinery continuously degrades unfolded proteins in heat-stressed cells (Raychaudhuri et al., 2014). Therefore, a decrease in Ndufs1 level in total and soluble mitochondrial fractions of heat-stressed Neuro2a cells (42°C for 2 h) could be due to degradation of unfolded Ndufs1 (Fig. 7A). In agreement with this, we also observed depletion of subunits of CI HC and SCs in the SILAC heatmap following complexome profiling of heat stress (Fig. S6A, Table S8C,D). Ndufs1 protein levels were restored in the mitochondrial fraction following 2 h recovery (37°C for 2 h after heat stress; Fig. 7A). Similarly, decreasing abundance of SCs and HCs depicted in the heatmap suggested partial restoration of RC subunits during recovery, possibly with the assistance of heat shock chaperones (Fig. S6A). CI activity within SCs was also protected in heat-stressed cells (Fig. 7B).

iBAQ redistribution of RC subunits was indicative of stabilization of SCs. Increase in abundance of CI SCs and oligomeric organization of CV was observed with a concomitant drop in abundance of their SubCs (Fig. 7C; Fig. S6B, Table S8A,B). Similar to MG132 stress, N and Q modules within CI SCs were prominently increased in abundance during heat stress (Fig. S6C). Strikingly, we did not observe simultaneous increases in CIII and CIV abundance suggesting a non-steady-state stoichiometry in SCs during heat stress. Thus, iSRC in heat-stressed cells was limited to CI and CV, but protected SC function despite overall depletion of subunits (Fig. S6A).

Finally, we investigated iSRC in HEK293T cells. SC organizations vary in different cell types depending on factors such as source organism or tissue, differential expression of RC subunits and assembly factors, and proteostasis capacity inside and outside mitochondria. For example, larger SCs including (I+III₂+IV₂), (I+III₂+IV₃) and (III₂+IV₂) were absent in Neuro2a cells (Fig. 1D). In contrast, these high molecular weight SCs, including respiratory megacomplex (I₂+III₂+IV₂), are reported to be present in the human embryonic kidney cell line HEK293T (Guo et al., 2018b). Despite differences in SC organization, primary amino acid sequences of RC subunits are nearly identical in mouse and human cells (Fearnley and Walker, 1992) and are similarly prone to aggregation. For example, mouse Ndufa5 shares ∼83% sequence identity with human Ndufa5 and found to form aggregates upon proteasome inhibition when overexpressed in HEK293T cells (Fig. 7D; Fig. S7A).

The distribution profile of RCs in HEK293T cells was different compared to those in Neuro2a cells (Fig. 1E; Fig. S7B). CI and CIII subunits were mostly present in high molecular weight SCs, including the megacomplex (I₂+III₂+IV₂) in slices 2–4. In comparison to levels in Neuro2a cells, the abundance of CI and CIII subunits in HCs (slices 7 and 11, respectively), and in SC(I+III₂) in slice 5, was reduced. CIV abundance was higher in the four slices with highest molecular weight, although CIV HC remained highly abundant, as in Neuro2a cells, in slices 13 and 14. Similarly, the CII HC and CV HC were abundant in slices 16 and 9, respectively (Fig. S7B, Table S9A,B).

iSRC remained consistent in HEK293T cells. Increasing abundances in the high molecular weight slices of the SILAC heatmap suggested the formation of stable and consolidated high molecular weight SCs in HEK293T cells, even after 24 h MG132 treatment (Fig. S7C). Abundance of the megacomplex (I₂+III₂+IV₂) was markedly increased in slice 2. Multiple intermediate CI SubCs also remained stabilized in slices 8–10. The abundance of CII HC remained unchanged in slice 16, but the abundances of SubCs were reduced in lower molecular weight slices (Fig. S7D, Table S9C,D). CIII SubCs in slices 12–20 were increased in abundance at 8 h MG132 treatment; however, they were depleted after 24 h of treatment, when most of the CIII subunits accumulated in SCs in slices 2 and 3 (Fig. S7C). CIV HC and SCs were found to be increased in abundance in slice 14 and above (molecular weight greater than ∼480 kDa); more prominently after 24 h of MG132 treatment (Fig. S7C). Increases in abundance of higher-order assemblies of CV were not apparent in the SILAC data heatmap. Rather, the abundance of CV subunits remained unchanged at 8 h MG132 treatment, or were slightly reduced throughout the BN–PAGE gel. Abundances of CV SubCs below slice 9 of the gel (molecular weight less than ∼720 kDa) were highly reduced at 24 h MG132 treatment, although higher assemblies were comparatively protected (Fig. S7C).

Relative isotopic iBAQ distribution of RC subunits suggested that CI SCs, particularly the megacomplex (I₂+III₂+IV₂) in slice 2, were significantly increased in abundance following MG132 treatment (Fig. 7E; Fig. S7B, Table S9A,B). Simultaneously, CI SubCs were reduced in abundance with increasing treatment time. Similar to effects in Neuro2a cells, increased abundance of SCs and reduced abundance of SubCs was prominent for N- and Q-module subunits at 24 h MG132 treatment (Fig. S7D). Abundances of CII and CV HCs were increased in slices 16 and 9, respectively. Simultaneously, the presence of CV subunits in SubCs was significantly reduced with treatment time (Fig. 7E). CIII and CIV SCs were increased in abundance at slices 2 and 3, predominantly at 24 h MG132 treatment, at the expense of HCs (Fig. 7E). An increase in SC:HC iBAQ ratios for CI, CIII and CIV was not observed at 8 h, but only after 24 h MG132 treatment in HEK293T cells, indicating slower iSRC kinetics compared to those in Neuro2a cells (Fig. 7F).

Importance of ‘quinary states’ of protein complexes has long been perceived. However, extensive characterization of such intracellular interactions is limited due to technical challenges. Here, we developed an iBAQ-and-SILAC-based two-dimensional complexome profiling strategy to investigate the plasticity of RCs in proteostasis-stressed cells. Our protocol reduces the biochemical artefacts of mitochondrial fraction preparation and increases the accuracy of quantification of quinary states of RC ensembles by mapping each and every RC subunit. We identify a subtle but consistent, organized and reversible equilibrium shift of RC subunits from smaller assemblies to superquaternary structures (iSRC) during proteostasis stress. Simultaneously, catalytic activity of different RCs is protected at supraorganizations. Our data suggest that this is not an instantaneous remodelling of proteins originating from free monomeric RC subunits and SubCs in a single step. Rather, high molecular weight SC assemblies partially disintegrate at early stages of stress. CIV, with multiple tissue-specific and developmentally regulated subunit isoforms (Sinkler et al., 2017), is the most flexible RC in these destabilized SCs. These transiently populated asymmetric intermediates promote native-like quinary interactions to re-establish and consolidate the steady-state stoichiometry, avoiding large energy expenditure (Cohen and Pielak, 2016). Similar to early stress, the stress recovery period represents another dynamic state of RC organization. In situ structural analysis at both these stages is expected to reveal novel dynamics of functional and conformational crosstalks (Fig. 5C).

As proposed previously, plasticity of SCs during normal growth conditions might be driven by only proximity-based weak interactions in the crowded environment of the IMM (Blaza et al., 2014). On the other hand, we find dynamic association and dissociation of assembly chaperones with different RC ensembles, suggesting contribution of these helper proteins in the unidirectional shift of RCs towards the stabilization of SCs during stress. It is also possible that some RCs or SubCs are normally associated with chaperones at steady state. These RC ensembles become free from chaperones during stress and therefore are increasingly engaged into SC formation. Titration of chaperones is well known to trigger remodelling of proteins and protein complexes, as exemplified by trimerization of the stress-responsive transcription factor Hsf1 (Raychaudhuri et al., 2014).

Recently, it has been shown that proteasome inhibition protects the RC-assembly factor COA7 from degradation and improves CIV function in models of mitochondrial leukoencephalopathy (Mohanraj et al., 2019). Similarly, endoplasmic reticulum stress has been reported to stimulate SC formation (Balsa et al., 2019). Our data also suggest that triggering iSRC by mild proteostasis stress might ameliorate oxidative phosphorylation deficiency in mitochondrial diseases. For this, understanding the stage-specific mechanistic involvement of assembly chaperones is necessary. On the other hand, many protein complexes assemble co-translationally to prevent aggregation of free subunits (Kamenova et al., 2019; Shiber et al., 2018). Nuclear-encoded RC subunits do not enjoy that opportunity because they are imported from the cytoplasm before their assembly inside mitochondria. Therefore, a considerable fraction of free RC subunits and smaller SubCs always exist inside mitochondria with aggregation-prone interaction surfaces exposed. Controlled and specific integration of these free subassemblies into HCs and SCs might offer an indirect fitness benefit during proteostasis-stress by preventing non-functional interactions and subsequent aggregation. However, other mitochondrial protein complexes and their subcomplexes do not partially disintegrate and reintegrate, rather their composition remains stable throughout the stress period. Thus, stress-responsive remodelling may not be a universal strategy to prevent aggregation of free subunits of any protein complex. This confirms the uniqueness of RCs in terms of capacity to self-remodel and suggests that the amino acid sequences of RC subunits have evolved for this purpose. Finally, our results establish the utility of our two-dimensional complexome profiling protocol over conventional protocols for analysis of RC plasticity and suggest that this strategy could prove a powerful approach to investigate spatiotemporal details of many dynamic protein interactions explaining a myriad of biological functions.

Ndufa5 was PCR-amplified from Neuro2a cDNA and cloned into pcDNA4/TO EGFP (Thermo Fisher Scientific) using the restriction enzymes KpnI and XhoI, as described previously (Rawat et al., 2019).

Neuro2a and HEK293T cells were obtained from ATCC, maintained at the CSIR-CCMB cell culture facility and tested for contamination regularly. Briefly, cells were maintained in Dulbecco's Modified Eagle's Medium (Gibco) supplemented with 10% fetal bovine serum (FBS) (Gibco) and 90 U/ml penicillin (Sigma-Aldrich), 50 µg/ml streptomycin (Sigma-Aldrich) at 37°C and 5% CO₂. Transfection of cells was performed with Lipofectamine 3000 reagent (Invitrogen) according to the manufacturer's protocol, for 24 h. For aggregation kinetics, random fields were captured using a FLoid Cell Imaging Station (Thermo Fisher Scientific) at the different timepoints. A Student's t-test was performed for significance calculation.

For SILAC-based mass spectrometry, cells were grown in SILAC DMEM (Thermo Fisher Scientific) supplemented with 10% dialysed FBS (Gibco), 90 U/ml penicillin (Sigma-Aldrich), 50 µg/ml streptomycin (Sigma) and either light [L-lysine 2HCl/L-arginine HCl (Lys0/Arg0)], medium [L-lysine 2HCl (4,4,5,5-D4)/L-arginine HCl (¹³C₆) (Lys4/Arg6)] or heavy [L-lysine 2HCl (¹³C₆, ¹⁵N₂)/L-arginine HCl (¹³C₆, ¹⁵N₄) (Lys8/Arg10)] isotopes of lysine and arginine (Thermo Fisher Scientific).

Mitochondria-enriched fraction preparation from cultured cell lines was performed according to Schägger (1995) and Acín-Pérez et al. (2008), with some modification. Briefly, ∼10 million cells were pooled together and the pellet was washed twice with PBS. The cell pellet was frozen at −80°C and homogenized in a Dounce homogenizer with about 10 cell-pellet volumes of homogenizing buffer, buffer A (83 mM sucrose, 10 mM MOPS pH 7.2). After adding an equal volume of buffer B (250 mM sucrose, 30 mM MOPS pH 7.2), nuclei and unbroken cells were removed by centrifugation at 1000 g for 5 min at 4°C. The mitochondria-enriched fraction was collected from the supernatant by centrifuging at 12,000 g for 5 min and washed twice under the same conditions with buffer C (320 mM sucrose, 1 mM EDTA, 10 mM Tris-HCl pH 7.4). The pellet was resuspended in buffer D (1 M 6-aminohexanoic acid, 50 mM Bis-Tris-HCl pH 7.0), and the concentration was estimated using Pierce Coomassie Plus (Bradford) Assay Reagent (Thermo Fisher Scientific).

Mitochondrial fractions were lysed in NP-40 lysis buffer [50 mM Tris-HCl pH 7.8, 150 mM NaCl, 1% NP-40, 0.25% sodium deoxycholate, 1 mM EDTA and protease inhibitor cocktail (Roche)] at 4°C for 45 min with intermittent vortexing. After centrifugation at 12,000 g for 15 min at 4°C, supernatant was collected as the soluble fraction. The remaining pellet was washed twice with PBS and boiled in 4× SDS loading buffer (0.2 M Tris-HCl pH 6.8, 8% SDS, 0.05 M EDTA, 4% 2-mercaptoethanol, 40% glycerol and 0.8% Bromophenol Blue) for 15 min to obtain the insoluble fraction. The total fraction was prepared by directly dissolving and boiling the mitochondria-enriched pellet in 4× SDS loading buffer for 15 min. The fractions were separated on NuPAGE 4–12% Bis-Tris gels (Invitrogen). The gel was run in MES buffer (100 mM MES, 100 mM Tris-HCl, 2 mM EDTA and 7 mM SDS) at 200 V for 40 min then fixed and stained with Coomassie Brilliant Blue. Preparation of gel slices, reduction, alkylation and in-gel protein digestion were carried out as described by Shevchenko et al. (1996). Finally, peptides were desalted and enriched according to Rappsilber et al. (2003).

Mitochondrial fractions were solubilized by adding 8 g digitonin (Sigma-Aldrich) per gram of mitochondria and incubating for 10 min in ice. The supernatant was collected after a 30 min centrifugation at 13,000 g. 5% Coomassie Brilliant Blue G-250 dye in 1 M 6-aminohexanoic acid was added to a quarter of the final amount of digitonin. Glycerol was added at final concentration of 10% and the sample was loaded onto NativePAGE 3–12% Bis-Tris Protein Gels (Invitrogen). Native PAGE was run using anode buffer (50 mM Bis-Tris HCl pH 7.0) and blue cathode buffer (50 mM Tricine, 15 mM Bis-Tris HCl pH 7.0, 0.02% Coomassie Brilliant Blue G-250) at 100 V until samples entered the separating gel and then at 150 V until the dye front reached one-third of the length of the entire gel. At this point, blue cathode buffer was replaced with dye-free cathode buffer for the remainder of the run at 250 V, until the dye front reached the end of the gel. The gel was incubated overnight at room temperature in freshly prepared CI NADH dehydrogenase substrate solution (5 mM Tris-HCl pH 7.4, 0.1 mg/ml NADH and 0.7 mg/ml Nitrotetrazolium Blue chloride). CIV activity was measured using 0.5 mg/ml Diaminobenzidine and 1 mg/ml cytochrome c in 50 mM phosphate buffer (pH 7.4). Appearance of bands is indicative of CI and CIV activity in the respective gels. The reaction was stopped with 10% acetic acid, washed with water and imaged (Jha et al., 2016). Band intensity was calculated using ImageJ (NIH, Bethesda, MD).

Western blotting after BN–PAGE was carried out similar to a previously reported protocol (McKenzie et al., 2006) with slight modifications. The BN–PAGE gel was equilibrated in soaking buffer (48 mM Tris-HCl, 39 mM glycine, 0.0037% SDS and 20% methanol) for 30 min and transferred onto 0.45 μm PVDF membrane (Bio-Rad) overnight at 20 V, followed by 400 mA for 90 min, using the Mini-Trans Blot cell system (Bio-Rad). The membrane was stained with 0.1% Coomassie Brilliant Blue R-250 dye, 50% methanol and 7% acetic acid and destained with destaining solution 1 (50% methanol and 7% acetic acid) followed by destaining solution 2 (90% methanol and 10% acetic acid), then imaged. Staining was removed using 100% methanol and washed with Tris-buffered saline containing Tween 20. After blocking with BSA, the membrane was probed with appropriate primary and secondary antibodies (Table S10) and imaged using a documentation system (Vilber Lourmat). Band intensity was calculated using ImageJ.

Mitochondria were isolated from SILAC-labelled control and treated cells and pooled together. 100–120 µg digitonin-solubilized mitochondria were loaded onto NativePAGE 3–12% Bis-Tris Protein Gels (Invitrogen) and run as described above. The gel was fixed and stained with Coomassie Brilliant Blue R-250 and cut into 20 slices for mass spectrometry. The lowest protein marker corresponded to 66 kDa. The gels were run until this marker band reached the dye front. Below that, the low molecular weight individual proteins were run out from the gel and thus, slice 20 contained either subunits of close to or above molecular weight 66 kDa, or smaller subcomplexes in that mass range. Preparation of gel slices, reduction, alkylation and in-gel protein digestion were carried out as described by Shevchenko et al. (1996). Finally, peptides were desalted and enriched according to a protocol reported by Rappsilber et al. (2003).

Peptides eluted after desalting were dissolved in 2% formic acid and sonicated for 5 min. Samples were analysed on a Q-Exactive HF (Thermo Fisher Scientific) and were separated on an EASY-Spray PepMap RSLC C18 column (75 μm×15 cm; 3 μm) using a 60 min linear gradient of the mobile phase [5% acetonitrile (ACN) containing 0.2% formic acid (buffer A) and 95% ACN containing 0.2% formic acid (buffer B)] at a flow rate of 300 nl/min. Full-scan mass spectrometry (MS) spectra (from m/z 400–1650) were acquired followed by MS/MS scans of the top 10 peptides with charge states of 2 or higher. The MS-based proteomics data of all these experiments have been deposited to the ProteomeXchange Consortium via the PRIDE partner repository (Vizcaíno et al., 2014) with the dataset identifier PXD014427.

For peptide identification, raw MS data files of individual slices were analysed separately using MaxQuant (Ver. 1.3.0.5) (Cox and Mann, 2008) and searched against Swissprot database of Mus musculus (release 2018.10 with 25,208 entries) or Homo sapiens (release 2019.03 with 42,419 entries) and a database of known contaminants. MaxQuant used a decoy version of the specified database to adjust the false discovery rates for proteins and peptides below 1%. The search parameters included constant modification of cysteine by carbamidomethylation, enzyme specificity trypsin, multiplicity set to 3 with Lys4 and Arg6 as medium label and Lys8 and Arg10 as heavy label. Other parameters included minimum peptides for identification 1, minimum ratio count 1, and the requantify option was selected. iBAQ option was selected to compute abundance of the proteins. Bioinformatics and statistical analysis was performed in the Perseus environment (Ver. 1.5.2.4) (Tyanova et al., 2016).

Digitonin-solubilized mitochondrial fractions (2 μg) from control and treated cells were digested with 100 ng trypsin for the indicated times. Reactions were stopped by adding 4× SDS loading buffer and boiling the samples. The samples were separated using SDS–PAGE and transferred onto 0.2 μm PVDF membrane (Bio-Rad) for 90 min at 300 mA using the Mini-Trans Blot cell system (Bio-Rad). Membranes were probed with appropriate primary and secondary antibodies (Table S10) and imaged using a documentation system (Vilber Lourmat). Band intensity was calculated using ImageJ.

We thank Santosh Kumar for feedback on the manuscript. The proteomics facility and tissue culture facility at CSIR-CCMB provided technical assistance.