Mitochondria are double membrane-bound eukaryotic organelles that perform critical cellular functions including producing the bulk of cellular energy, acting as hubs for synthesis of various biomolecules, and driving the apoptotic response. Mitochondria produce the majority of the cell’s ATP through oxidative phosphorylation (OXPHOS), which involves five multi-protein complexes (complexes I-IV and the F₁-F₀ ATP Synthase, or complex V) found within the inner mitochondrial membrane (IMM) (Hatefi, 1985; Rich and Maréchal, 2010; Hock et al., 2020). Mitochondria evolved from endosymbiotic prokaryotes, and therefore have their own genome, known as mitochondrial DNA (mtDNA), and their own ribosomes, known as mitoribosomes. Most protein coding genes in mtDNA were lost or transferred to the nucleus early in eukaryotic evolution, and numerous machineries have evolved to import and process these >1,000 proteins (Adams and Palmer, 2003; Roger et al., 2017; Rath et al., 2021). Following synthesis in the cytosol, nuclear encoded proteins are imported into the mitochondria through translocases of the inner and outer membranes, TIM and TOM respectively (Jackson et al., 2018; Pfanner et al., 2019). In humans, 13 proteins are encoded by mtDNA, all hydrophobic transmembrane subunits of the OXPHOS complexes that are translated on mitoribosomes and co-translationally inserted into the IMM (Ott and Herrmann, 2010; Thompson et al., 2018).

Complexes I-IV act as an electron transport chain with electrons from metabolic sources such as the tricarboxylic acid (TCA) cycle passing from complex to complex in a series of redox reactions. The energy from this process is used to pump protons from the mitochondrial matrix to the intermembrane space (IMS), creating a proton gradient used by complex V to synthesize ATP (Jonckheere et al., 2012). Mitochondrial respiratory complexes I, III and IV are found together in higher-order complexes, so called supercomplexes (SCs), that have well defined stoichiometry (Schagger and Pfeiffer, 2000; Letts et al., 2016; Wu et al., 2016; Guo et al., 2017). These include complex I and III (I/III₂), complex III and IV (III₂/IV), and complex I, III and IV (I/III₂/IV), the latter referred to as the respirasome and considered the dominant form.

Complex III (CIII) or the cytochrome bc ₁ complex is at the center of the OXPHOS system, using electrons received from complexes I, II and other sources via ubiquinone to reduce cytochrome c, while also pumping protons into the IMS (Crofts et al., 2008; Fernandez-Vizarra and Zeviani, 2015). Mammalian CIII forms an obligate homodimer, with each subunit composed of 9 nuclear encoded proteins and one mtDNA-encoded protein, cytochrome b (MT-CYB) (Schagger et al., 1986). Complex III biogenesis begins with the synthesis of mtDNA-encoded MT-CYB on mitoribosomes and its co-translational insertion into the IMM. Two small membrane bound subunits UQCRB and UQCRQ then associate with MT-CYB (Hildenbeutel et al., 2014), while in parallel to this, peripherally associated subunits UQCRC1 and UQCRC2 form a tetramer of two UQCRC1-UQCRC2 dimers which is thought to coalesce on the nascent MT-CYB assembly along with additional membrane bound subunits CYC1, UQCR10 and UQCRH to yield a dimeric intermediate of CIII called pre-CIII₂ (Stephan and Ott, 2020). Finally, UQCR11 and catalytic subunit UQCRFS1 and incorporated to form the mature, active complex (Smith et al., 2012; Fernandez-Vizarra and Zeviani, 2018; Ndi et al., 2018). Although complexes I and IV are structurally independent from CIII, complete complex III assembly is critical for their biogenesis (Acin-Perez et al., 2004; Protasoni et al., 2020). Although the molecular details of this necessity are not yet fully understood, loss of CIII leads to inhibition of the final step of Complex I assembly; addition of the catalytic NADH dehydrogenase module (N-module) (Protasoni et al., 2020). Mutations in MT-CYB and 5 of the 9 nuclear encoded subunits in Complex III, along with mutations in many Complex I and IV subunits and assembly factors required for the biogenesis of OXPHOS complexes, impact complex and result in mitochondrial disease, a debilitating metabolic disorder affecting at least 1 in 5,000 live births (Fernandez-Vizarra and Zeviani, 2015; Gorman et al., 2016; Hock et al., 2020). For the majority of mitochondrial diseases there are no effective treatments, thus a nuanced understanding the molecular impact of dysfunctional assembly will be critical to the future development of therapeutics (Gorman et al., 2016).

Sodium carbonate extraction is a ∼30 year old experimental technique used to determine whether a specific protein is an integral membrane protein, peripherally associated with the membrane, or a soluble protein. In the original method, cellular isolates containing membranes are solubilized in 0.1 M sodium carbonate, pH 11.5 (Fujiki et al., 1982a; Fujiki et al., 1982b). The alkaline carbonate solution disrupts protein-protein interactions without disrupting the membrane. Ultracentrifugation is then used to generate a pellet fraction, containing integral membrane proteins, and a supernatant fraction containing soluble and membrane-associated proteins. These fractions can then be analyzed by immunoblotting with specific antibodies.

The mitochondrial inner membrane is one of the most densely packed eukaryotic membranes with a high protein: lipid ratio. As a result of this, the transmembrane domains (TMDs) of mitochondrial membrane proteins are less hydrophobic compared to ER membrane proteins (von Heijne, 1986). This, along with the unique lipid content of mitochondrial membranes may explain the observation that mitochondrial proteins with single, moderately hydrophobic TMDs are susceptible to carbonate extraction at pH 11.0 (Kim et al., 2015). Thus, sodium carbonate extraction alone may yield false negatives as to the presence of TMDs in mitochondrial proteins, and that proteins may be present in both the pellet and supernatant fractions in different proportions, depending on their level of membrane integration (Fujiki et al., 1982a). One approach to overcome this is the use of multiple carbonate solutions with increasingly lower pH, as at lower pH peripheral membrane proteins are partially retained in the pellet fraction (Lytovchenko et al., 2014).

Here we describe a technique combining sodium carbonate extraction with quantitative mass spectrometry (SCE-MS) applicable to routine use in molecular investigations. We used SCE-MS to generate membrane association profiles for over 840 mitochondrial proteins in human embryonic kidney (HEK293T) cells, grouping them into 6 distinct clusters based on their resistance to sodium carbonate extraction across multiple pH conditions. By coupling SCE-MS with stable isotope labelling of amino acids in cell culture (SILAC), we were able to track changes in the membrane integration of mitochondrial proteins in gene-edited knockout cell-lines lacking different CIII subunits. We find that loss of single pass membrane-spanning subunit UQCR10 leads to defective complex assembly. SCE-MS revealed that while the majority of CIII subunits remained membrane associated, peripherally associated core-module subunits UQCRC1 and UQCRC2 became more soluble, consistent with the destabilization and detachment of the core-module from the membrane bound CIII intermediate. Additionally, we found that arrest of CIII assembly, either through loss of UQCR10 or UQCRC1, leads to partial loss of the Complex I N-module as has been previously shown in CIII deficient cells (Protasoni et al., 2020). Using SCE-MS we found that while a population of N-module remained associated with Complex I, N-module subunits had increased solubility relative to other CI subunits, suggesting the modules attachment to the complex becomes unstable in the presence of a structural CIII defect. Thus, we see SCE-MS as a novel and unbiased means to assess the destabilization of protein complexes; a concept often cited to describe the impact of an assembly defect, but generally without a clear molecular definition.

To assess the membrane association of the mitochondrial proteome in an unbiased manner, we adapted a sodium carbonate extraction protocol for use with quantitative proteomics workflows (Figure 1A). This included the use of stable isotope labelling using amino acids in cell culture (SILAC) to allow the impact of perturbations (such as CRISPR/Cas9 knockout of a protein of interest) to be quantitatively assessed, as discussed in subsequent results sections. In order to validate our protocol, we first analyzed sodium carbonate extraction data for only control HEK293T mitochondria, separated from the dataset in silico and treated in a similar way as would be done in a label free quantification (LFQ; see Materials and Methods for more detail). SDS-PAGE and immunoblotting using antibodies against mitochondrial proteins with varied localizations and membrane integration was also performed to establish baseline profiles for proteins with known behavior (Figure 1B). At pH 11.5, outer membrane channel VDAC2 and peripheral IMM electron carrier cytochrome c (CYCS) are found exclusively in the pellet and supernatant fractions respectively as expected (Ma et al., 2014;Letts et al., 2016). Complex I N-module/matrix arm subunit NDUFV1 and assembly factor NDUFAF1, both peripherally associated with the membrane (Stroud et al., 2016), along with import translocase components TIMM50 and mt-HSP70 are found in both fractions at pH 11.5. At pH 9.5, a reduction in carbonate extraction is observed for each protein, with only cytochrome c and mt-HSP70 found in equal proportions in both the pellet and supernatant fractions, possibly due to their localization in both soluble and peripherally membrane associated pools (Kang et al., 2018;Timón-Gómez et al., 2018). Using SCE-MS we were able to quantify the relative abundance of 844 mitochondrial proteins (Rath et al., 2021) from control HEK293T cells at both pH 9.5 and 11.5 (Figure 1C; Supplementary Table S1). This represents 74% of the known mitochondrial proteome. Under both extraction conditions the majority of proteins with transmembrane domains annotated in UniProt (The Uniprot Consortium, 2021) were detected at greater abundance in the pellet fraction. Proteins without annotated TMDs were found most enriched in the supernatant extracted at pH 11.5, but were more evenly distributed between supernatant and pellet fractions following extraction at the lower pH, as expected.

In order to classify the mitochondrial proteome based on extraction properties, we performed hierarchical clustering on our SCE-MS data (Figure 2A; Supplementary Table S2) and used this to define six distinct clusters that could be readily visualized using principal component analysis (PCA; Figure 2B; Supplementary Table S2). Cluster 1 contains 276 proteins that are resistant to membrane extraction at both pH 9.5 and 11.5. This includes hydrophobic proteins such as mitochondrial transporter family SLC25A members and all detected mtDNA encoded proteins. Cluster 2 contains 44 proteins that are resistant to extraction at pH 9.5 but are partially susceptible to extraction at pH 11.5. Cluster 3 contains 20 proteins that are resistant to extraction at pH 9.5 but very susceptible at pH 11.5. Cluster 4 contains 128 proteins that are moderately resistant to extraction at pH 9.5 but highly susceptible at pH 11.5, including 21 subunits of the small subunits of the mitoribosome, 16 of the large subunit and 6 out of the 9 detected subunits of the N-module of complex I. Cluster 5 contains 278 proteins that are partially resistant to extraction at pH 9.5 but very susceptible at pH 11.5. Cluster 6 contains 88 proteins that are susceptible to extraction at both pH 9.5 and 11.5. A total of 10 proteins quantified in this analysis did not segregate into a cluster.

To further validate our dataset, we cross-referenced the number of predicted TMDs present in proteins found within each cluster. Predictions were drawn from UniProt (The Uniprot Consortium, 2021), which annotates TMDs automatically based on the TMD being identified by both TMHMM (Krogh et al., 2001) and Phobius (Käll et al., 2004) algorithms with substantial overlap. Nearly every protein with two or more annotated TMDs was found in cluster 1 (Figure 2C), along with the majority of single TMD proteins. The remaining proteins with single TMDs are distributed across the other clusters, with a bias toward clusters 2 and 4. Of the 276 proteins in cluster 1, 73 had no TMD annotated in Uniprot (Figure 2D). Using a number of additional bioinformatics tools for predicting TMDs, including TMPred (Hofmann, 1993) and HMMTOP (Bioinformatics Toolkit, RRID:SCR_010277) (Tusnády and Simon, 2001), as well as TMHMM (RRID:SCR_014935) (Krogh et al., 2001) and Phobius (RRID:SCR_015643) (Käll et al., 2004) without the overlap constraint used by UniProt, we identified an additional 30 proteins that had TMDs identified by at least two algorithms. A number of these proteins have been experimentally determined to be genuine membrane proteins, including complex I subunit NDUFB2 through high resolution structural studies (Zhu et al., 2016) recently identified complex III assembly factor OCIAD1 through classical carbonate extraction studies (Le Vasseur et al., 2021) and pro-survival protein FUNDC2 through mutation and localization analysis (Ma et al., 2019).

Next, we wanted to demonstrate the usefulness of SCE-MS in identifying changes in membrane integration following a perturbation. We used the CRISPR/Cas9 system to generate knockout cell lines where the expression of complex III subunits UQCRC1 and UQCR10 was disrupted. Two individual clones for each knockout were validated by sequencing of targeted alleles (see Materials and Methods). To determine changes in OXPHOS complex assembly, we analyzed mitochondria from two clones of each knockout using Blue Native (BN) PAGE and immunoblotting (Figure 3A). Mitochondria from both knockouts retained a ∼150 kDa subassembly containing CYC1 that was more abundant in the UQCRC1^KO than UQCR10^KO (Figure 3A, compare lanes 2–5 with 1; subassembly indicated by Ŧ). Cells lacking UQCR10 retained a CIII₂-like assembly containing CYC1 and UQCRC1 (compare lanes 4–5 with 1). No signal for complex III subunit UQCRFS1 was observed in either UQCRC1^KO or UQCR10^KO clones (compare lanes 7–10 with 6), indicating that UQCRFS1 is not assembled into CIII in either knockout cell line in agreement with existing assembly models that describe incorporation of that subunit to be a late stage assembly event (Fernandez-Vizarra and Zeviani, 2018), and despite UQCRFS1 levels being only slightly reduced in UQCR10^KO (see Figure 3D and Figure 4A; UQCRFS1 is decreased by a fold-change of ∼2 in UQCRC1^KO ). Interestingly, the CIII₂-like complex in UQCR10^KO appeared to migrate slightly faster than wildtype CIII₂ (Figure 3A), leading us to speculate that it could represent the pre-CIII₂ intermediate (Fernandez-Vizarra and Zeviani, 2018). In the absence of peripheral core-module subunit UQCRC1, no supercomplex assemblies were observed, with CI migrating in two major sub-assemblies (compare lanes 22–23 with 21; assemblies indicated by † and ‡). In the absence of single-spanning membrane subunit UQCR10, similar CI sub-assemblies were observed, in addition to a supercomplex-like structure, albeit at a lower abundance than in control cells (compare lanes 24–25 with 21). In both UQCRC1^KO and UQCR10^KO, assembly of holo CIV appears to be unperturbed, however higher order CIV assemblies are mostly absent, with the exception of a low abundance complex migrating at a similar molecular weight to the supercomplex in UQCR10^KO. Given the reported role of CIII in CI and CIV assembly and function (Protasoni et al., 2020) and the unexpected impact of the loss of UQCRC1 and UQCR10 on CI seen in Figure 3A, we wanted to determine if CI and CIV activity was retained in our knockouts. Surprisingly, despite the loss of catalytic complex III subunit UQCRFS1 from all CIII-containing complexes and sub-assemblies (Figure 3A) and thus the loss of overall respiration via OXPHOS (Figure 3B), partial complex I in-gel activity was observed in the remnant supercomplex-like structure in UQCR10^KO (Figure 3C, compare lanes 4–5 with 1) and in the lower molecular weight assemblies seen in both knockouts (Figure 3C). Despite the low abundance band seen in immunoblotting, no Complex IV activity was observed in the supercomplex-like structure found in UQCR10^KO, with the only activity found holo CIV (Figure 3C). Given the reproducible phenotype between multiple independent sequence-verified clones complementation experiments were not performed. Clones 1 of UQCRC1^KO and UQCR10^KO were used for all subsequent experiments.

We first focused our SCE-MS analysis on UQCR10 knockout mitochondria. Assembly of the dimeric pre-CIII₂ occurs with the addition of membrane subunits UQCR10, CYC1 and UQCRH and peripherally associated core-module subunits UQCRC1 and UQCRC2 (Stephan and Ott, 2020). We reasoned that loss of UQCR10 may result in accumulation of a soluble core-module observable through SCE-MS, demonstrating how the technique can provide additional insights into CIII assembly. First we assessed the baseline steady-state levels of proteins in UQCR10^KO and controls using quantitative mass spectrometry of isolated mitochondria, as baseline data would be critical to interpretating SCE-MS data since some proteins may have an overall reduced cellular abundance due to perturbed assembly. Knockout and control HEK293T cells were subjected to SILAC labeling using light or heavy (¹³C₆ ¹⁵N₂ lysine and ¹³C₆ ¹⁵N₄-arginine) amino acids, and the relative abundance of mitochondrial proteins in the knockouts compared to HEK293T was determined. In total, we quantified relative changes in abundance of 749 mitochondrial proteins (Supplementary Table S3). While UQCR10 itself was not detected in UQCR10^KO mitochondria (Supplementary Table S3), OXPHOS subunits including most CIII subunits were only minimally reduced in abundance, with only UQCRC2 reaching our significance threshold of 1.5 fold change (Figure 3D). The non-specific trend in reduced abundance of CIII and supercomplex subunits is also evident using topographical heatmaps (Stroud et al., 2016) (Figure 3E) where the fold change for individual proteins observed in steady state proteomics is mapped to the cryo-electron microscopy structure of the human CI/CIII₂/CIV supercomplex (Gu et al., 2016).

We next turned to SCE-MS, which was performed as outlined in Figure 1A using SILAC labelled UQCR10^KO and control cell lines (Supplementary Figure S1, Supplementary Table S4). To account for the overall decreased steady-state abundance of some proteins seen in Figure 3D we also measured proteome changes in a “Total” fraction containing sodium carbonate solubilized material prior to ultracentrifugation (Supplementary Figure S1, left panels), using this data to normalize pellet and supernatant SCE-MS results (Figure 3F shows normalized data at pH 9.5. Normalized data for pH 11.5 can be found in Supplementary Figure S2, and raw ratios in Supplementary Figure S1). Following carbonate extraction at pH 9.5, the normalized pellet fraction had no significant changes in the levels of OXPHOS subunits. However in contrast, CIII subunits UQCRC1 and UQCRC2 were found to be relatively more abundant in supernatant in UQCRC10^KO. Similar results were observed following extraction at pH 11.5, however under these conditions the effect was reversed, with UQCRC1 and UQCRC2 being less abundant in the pellet fraction of UQCRC10^KO (Supplementary Figure S2B). To more easily visualize changes in membrane association of CIII subunits we compared the proteome wide changes to membrane association across both extraction conditions, assigning protein solubility in the knockout to the previously identified clusters (Figure 2B). While in the knockouts most proteins fell into the same clusters as in control cells, the clustering of core-module subunits UQCRC1 and UQCRC2 changed from clusters 2 to 5 in the knockout (Figure 3G, compare filled to empty squares) indicating altered membrane association. Also notable is UQCRFS1, the last assembled subunit (Fernandez-Vizarra and Zeviani, 2018), which moved from cluster 2 to 5. Cluster 5 contains proteins that are partially resistant to extraction at pH 9.5 but very susceptible at pH 11.5 and predominantly includes soluble matrix proteins such as those found in the mitoribosome and various tRNA ligases, along with proteins having mixed peripheral and soluble localization such as mt-HSP70 (Kang et al., 2018; Timón-Gómez et al., 2018) (Supplementary Table S2). These data, taken together with our BN-PAGE analysis that shows UQCRC1 remains associated with CIII₂ in the UQCR10^KO under native-electrophoresis conditions (Figure 3A, compare lanes 14–15 with 11), suggest that the interaction between core module subunits UQCRC1 and UQCRC2 and the nascent CIII membrane assembly is destabilized in the absence of UQCR10, evidenced by the increased resistance of UQCRC1 and UQCRC2 to extraction at pH 9.5.

A concern we had over the validity of our SCE-MS data is the impact of the carbonate extraction procedure on protein complex stability. Our knockouts harbor sub-assemblies with different protein composition to that found in control cells. Should the presence of carbonate lead to selective partial breakdown of these complexes but not the comparable control complex, this would potentially lead to artefactual results caused by altered extractability of these breakdown sub-complexes. An example sub-assembly would be the CIII₂-like complex lacking UQCRFS1 and UQCR10 that we suspect may represent pre-CIII₂ (Figure 3A, compare lanes 4–5 with 1), as in the above experiments we are directly comparing the extractability of UQCRC1 and UQCRC2 from this and wildtype holo-CIII₂ (along with other assemblies). To address this, we analyzed membrane fractions from our control and knockout cells extracted at both pH 9.5 and 11.5 using BN-PAGE and immunoblotting (Supplementary Figure S3). While the relative abundance of some sub-assemblies changed under different conditions as expected based on our SCE-MS data, their migration on BN-PAGE did not. We also did not observe any sub-complexes unique to either carbonate extraction condition. Taken together, we conclude that the presence of sodium carbonate does not alter the composition of the complexes and sub-assemblies being discussed in this study.

We next turned our attention to the UQCRC1 knockout. To obtain a baseline of proteome changes in the knockout, we performed steady state quantitative mass-spectrometry on isolated mitochondria, as described above (Supplementary Table S3). Loss of UQCRC1 led to a significant decrease in the levels of subunits present in CI, CIII and CIV, particularly the subunits of the core module of CIII, UQCRC1 and UQCRC2, and of the N-module of Complex I (Figure 4A). We again used topographical heatmaps to interpret the data in the context of the CI/CIII₂/IV supercomplex (Figure 4B) (Gu et al., 2016), which clearly visualized the greatest impact of UQCRC1 loss to be on the core module of CIII and the N-module of Complex I. While this is consistent with recent reports that dysfunctional CIII assembly prevents addition of the N-module to Complex I (Protasoni et al., 2020), the molecular underpinnings of this phenomenon are not fully understood. To investigate this further, we performed SCE-MS on UQCRC1^KO as described above. In mitochondria from this knockout we observed a significant increase in the solubility of several CI subunits, with the greatest increase seen in N-module subunits (Figure 4C, right panel). Also moderately increased in solubility were the CIII subunits UQCRC2 and UQCRB. As done for UQCR10^KO, we used a PCA-based approach incorporating both extraction conditions to better understand the implication of proteome solubility changes seen in the UQCRC1^KO. While most proteins quantified in UQCRC1^KO could be assigned to the same clusters as in control cells (Figure 2B), the clustering of Complex I N-module subunits changed (Figure 4D, compare filled to empty squares) indicating altered membrane association. In the control HEK293T dataset, whereas most N-module subunits appeared at the intersect of cluster 3 and 4, these proteins migrated toward cluster 5 in UQCRC1^KO (Figure 4D). Complex III subunits were largely unchanged, with the exception of UQCRFS1, which like in UQCR10^KO moved from cluster 2 to 5. Similar trends in respect to N-module solubility were observed, though with less severity, for UQCR10^KO (Figure 3G). Despite N-module subunits being increasingly soluble in UQCRC1^KO and given the results of Protasoni and colleagues (2020) we were surprised by the relatively large amount of catalytic N-module subunit NDUFV1 still migrating with other Complex I subunits in UQCRC1^KO (Figure 3A), and moreover, CI-linked NADH dehydrogenase activity, an enzymatic reaction performed by the N-module, retained in UQCRC1^KO (Figure 3C). Interestingly, Protasoni and colleagues (2020) also observed low amounts of Complex I activity in their model of CIII dysfunction, however complexome profiling, a technique where lanes from BN-PAGE gels are excised into ∼60 slices and the protein distribution obtained by MS analysis, showed that the majority of holo-Complex I lacked associated N-module subunits. To better understand our results, we performed complexome profiling on the UQCRC1^KO (Figure 4E; Supplementary Figure S4; Supplementary Table S5) which accurately recapitulated observations we made using traditional BN-PAGE and immunoblotting (Figure 3A). A closer inspection of the dataset revealed multiple high molecular weight Complex I sub-assemblies in UQCRC1^KO, including a dominant assembly containing all Complex I subunits, and at least two assemblies lacking the N-module (Figure 4F, † indicates assembly containing N-module, ‡ the assembly lacking N-module observed in Figure 3A). Taken together with our SCE-MS and in-gel activity data, we conclude that while the severe structural defect in CIII assembly found in UQCRC1^KO leads to the association between the N-module and the Complex I membrane assembly being destabilized, the N-module still retains an ability to associate with the complex. We suspect the differences in N-module association between our study and Protasoni et al. (2020) can be attributed to cell type and other confounders (see Discussion), and while our data broadly supports their conclusions it raises additional questions as to the molecular nature of Complex I N-module assembly.

Sodium carbonate extraction is technique commonly used to identify integral membrane proteins. The technique has been adapted for use in both immunoblotting and radiolabeled in vitro import based assays (Stojanovski et al., 2007) and in addition to being useful in the characterization of newly identified proteins, is commonly used to gain mechanistic insights into membrane protein complex biogenesis through the use of mutant and gene-edited cell lines where the assembly process is blocked at specific steps. The use of multiple pH extraction conditions and control proteins with known localization additionally allows proteins with peripheral membrane association to be assessed (Lytovchenko et al., 2014). Here we describe a technique that combines sodium carbonate extraction with quantitative proteomics applicable for routine use in the study of mitochondrial complex assembly. Sodium carbonate extraction has previously been coupled with mass-spectrometry (Vögtle et al., 2017) however in that study the technique was combined with several other biochemical fractionation protocols to develop a steady state map of the Saccharomyces cerevisiae mitochondrial proteome. Vögtle and colleagues utilized SILAC multiplexing to compare crude and highly pure yeast mitochondria, allowing novel mitochondrial proteins to be confidently identified (Vögtle et al., 2017), but making their protocol not applicable to routine comparative analysis. Using SCE-MS we profiled the membrane association of 844 mitochondrial proteins in the commonly used human HEK293T cell line, roughly 75% of the human mitochondrial proteome (Rath et al., 2021). We applied bioinformatic tools to this dataset in order to cluster proteins based on their membrane association profiles, identifying 6 well defined clusters with decreasing membrane association. Clusters were validated against proteins with known membrane association and readily separate proteins with genuine membrane integration via TMDs (Cluster 1) such the polytopic SLC25A family of solute carriers found in the IMM (Palmieri, 2013) from truly soluble proteins (Cluster 6). One interesting outcome of this analysis is the large proportion of proteins that behave as peripherally associated membrane proteins. We identified four different clusters, accounting for almost 60% of proteins quantified, with varying levels of extraction at both pH 11.5 but partial resistance to extraction at pH 9.5, suggesting at least some peripheral association with the membrane. Our dataset, available in Supplementary Table S2, can be used as a general reference for the membrane association of a protein of interest in HEK293T cells.

One caveat of our dataset that could explain the profiles seen for some proteins is the tendencies for large soluble complexes such as the mitoribosome (subunits of which are predominantly found in Cluster 4) to sediment at the high centrifugation speeds used in our assay (Aibara et al., 2020). This highlights the need for biology and the nature of co-clustering proteins to be considered when interpreting SCE-MS data. For example, the mitoribosome is known to associate with the IMM translocase OXA1 to facilitate co-translational membrane insertion (Itoh et al., 2021) while Cluster 4 also contains subunits found in the membrane associated matrix arm of complex I. Together these lines of evidence suggest the presence of mitoribosome proteins in this cluster may reflect at least some degree of membrane association. To further test our clusters for biological significance we assessed the number of TMDs predicted for each protein. As expected, the majority of proteins with predicted TMDs were resistant to extraction and as such clustered together in Cluster 1. Through bioinformatic analysis of Cluster 1 proteins, we identified 30 proteins with likely TMDs that are not annotated on the Uniprot database. The presence of the other 43 proteins lacking annotated TMDs within Cluster 1 suggests that these proteins may contain transmembrane beta barrels or may be associated with the membrane through post-translational modifications, such as glycosylphosphatidylinositol (GPI) anchors or covalent lipid modifications. While proteins with two or more TMDs are almost exclusively found within Cluster 1, a number of single TMD proteins were found in all other clusters, including two proteins with single TMDs in Cluster 6, which is made up of mostly soluble proteins susceptible to carbonate extraction. One of these, the serine protease HTRA2, contains an N-terminal transmembrane domain that is inserted into the IMM during import, the protein is subsequently cleaved, releasing the soluble domain into the IMS (Challa et al., 2007).

As a test case for SCE-MS as a tool to complement techniques such as BN-PAGE in understanding the assembly pathways of OXPHOS complexes, we decided to investigate the impact of different complex III structural defects on the assembly of complex III and the wider OXPHOS system. While both UQCRC1 and UQCR10 are thought to be incorporated into CIII at the same step (Smith et al., 2012;Ndi et al., 2018), as UQCRC1 is required for the dimerization of CIII (Stephan and Ott, 2020) we hypothesized that UQCRC1^KO would exhibit a more severe CIII defect, more closely mimicking the loss of MT-CYB (Protasoni et al., 2020) that is known to lead to assembly of the Complex I N-module being blocked. We anticipated that the loss of the small membrane subunit UQCR10 would have minimal impact on CIII assembly, but may lead to the presence of a pool of soluble UQCRC1-UQCRC2 core-module. Both situations where chosen as they were predicted to highlight the utility of SCE-MS to observe aspects of OXPHOS assembly defects not accessible through other techniques. In UQCR10^KO, while the abundance of complex III subunits were only moderately altered when observed through steady state quantitative proteomics (Figure 3D), SCE-MS showed UQCRC1 and UQCRC2 to increase in solubility (Figure 3F), migrating from a cluster defined by the presence of membrane proteins to one containing predominantly soluble matrix proteins (Figure 3G). Interestingly, we also found that UQCRFS1 is not present in the CIII intermediate found in UQCR10^KO that we suspect to be pre-CIII₂, and like the core module is found predominantly in soluble clusters (Figure 4D). In the high resolution structure of the mature CIII dimer (Gu et al., 2016), the transmembrane domains of UQCRFS1 and UQCR10 sit next to each other. Taken together with our data, this may suggest that UQCR10 plays a direct role in stabilizing UQCRFS1 incorporation into the complex through a coordinated interaction with the translocase BCS1L (Wagener and Neupert, 2012). As expected, UQCRC1^KO lacks the later stage CIII intermediates (Figure 3A) and has a greater impact on the steady state abundance of CIII subunits (Figure 4A) than UQCR10^KO. Loss of UQCRC1 also strongly impacts the abundance of Complex I and blocks its assembly into the supercomplex (Figure 3A). This is unsurprising as assembly of Complex III is known to be critical to the assembly of Complex I (Acin-Perez et al., 2004; Schägger et al., 2004; Protasoni et al., 2020). This was most recently demonstrated using human cybrids lacking MT-CYB, where it was shown through complexome profiling and other techniques that complex I is blocked at the step where the N-module is incorporated (Protasoni et al., 2020). While our steady state proteomic data for the UQCRC1^KO showed a decrease in the abundance of N-module subunits (Figures 4A,B) SCE-MS suggested the remaining population trended to being more soluble (Figures 4C,D). We performed complexome profiling on UQCRC1^KO to investigate this further, finding multiple holo-Complex I assemblies with and without N-module. Importantly, these results were supported by our in-gel NADH dehydrogenase activity assay (Figure 3C) which confirmed the presence of the N-module on at least one remnant Complex I assembly. Taken together, our data suggests that while Complex I biogenesis strongly relies on the presence of the CIII dimer for efficient assembly, it appears to be able to retain a partial ability to assemble without the structural involvement of CIII. This is not without precedent, as Protasoni and colleagues (2020) also find that 6% of Complex I in their model system remains fully assembled, however unlike in their study we find that the relative amounts of N-module and non-N-module complex I subunits are approximately the same in control and UQCRC1^KO cells. Moreover, Protasoni and colleagues show that restoration of the Coenzyme Q pool through expression of an alternative oxidase in their CIII deficient cell line was able to partly restore complex I activity and re-association of the N-module to holo-Complex I. Such an experiment followed by SCE-MS could provide further insights into the nature of N-module association. Possible reasons for the discrepancy between our results and Protasoni et al. (2020) could be cell line specific differences (HEK293 derived from embryonic kidney in our study, and 143B cybrids derived from an osteosarcoma in Protasoni et al., 2020) and resulting differences in the Q-pool, as potential differences in detergent solubilization and BN-PAGE protocols, as well as the nature of the CIII assembly defect. While it is tempting to speculate that the largest confounder is the latter, our UQCRC1^KO contains an uncharacterized ∼150 kDa CIII sub-assembly containing MT-CYB, CYC1, UQCRB, UQCRQ and UQCR10 (Figure 3A, Supplementary Figure S4) while their cybrid system harbors no comparable sub-assembly, the ∼150 kDa sub-assembly present in our cell line does not appear to be associated with any Complex I sub-assemblies (Supplementary Figure S4). Thus we conclude that while structural association between Complex III assemblies and the nascent Complex I membrane arm is not required for N-module association, its absence may lead to the interaction being more labile, reflected in an increased susceptibility of N-module subunits to carbonate extraction.

Finally, two additional caveats to SCE-MS introduced by the use of knockout cell lines harboring severe defects in OXPHOS complex assembly should be considered. The first is the potential impact of the carbonate solution on protein complex stability. Since we are directly comparing extractability of proteins from two different complexes, differences in complex stability during the carbonate extraction has the potential to confound results. For example, a sub-assembly unique to a knockout may collapse in the presence of sodium carbonate leading to some subunits being artificially susceptible to extraction from the membrane, whereas the comparable assembly in control cell lines is unaffected. We analyzed membrane fractions from our control and knockout cells extracted at both pH 9.5 and 11.5 using BN-PAGE and immunoblotting (Supplementary Figure S3). This experiment showed that while the susceptibility to extraction changed for some complexes as we expected it would, complexes resistant to carbonate extraction migrated similarly to those solubilized from intact mitochondria. An additional caveat that could be considered is effect of the loss of a protein of interest on mitochondrial lipid composition, which may indirectly alter extractability of proteins. For example, dysfunctional respiratory chain assembly is also known to alter the abundance of cardiolipin (Xu et al., 2019), a mitochondrial specific phospholipid that is critical in the stability of mitochondrial complexes, including OXPHOS complexes (Zhang et al., 2002; Pfeiffer et al., 2003; Mckenzie et al., 2006; Kutik et al., 2008; Gebert et al., 2009). Such alterations could reasonably impact the extractability of complexes between control and knockout cells in a similar way as described above. While measurement of lipid content in our cell lines was out of the scope of our study, we note that while defects in cardiolipin biosynthesis alter the migration and conceivably composition of the pre-sequence translocase of the inner membrane (TIM23 complex) (Kutik et al., 2008), the extractability of Tim23 is not altered (Claypool et al., 2008). Given that the migration of carbonate resistant complexes is not altered in our knockouts (Supplementary Figure S3) we suspect that potential changes in mitochondrial lipid composition would not alter our conclusions.

In summary, SCE-MS is technique that can be used to identify dynamic changes in membrane association on a proteome wide scale. “Destabilization” is an oft-cited concept to describe the impact of an assembly defect on a specific subunit or part of a complex. In most cases it is used speculatively upon observing a reduction in protein abundance, but without other molecular precedent. SCE-MS provides additional insight into “destabilization” by revealing a change in solubility of a protein or population of proteins between two conditions. The quantitative accuracy and high dynamic range of mass-spectrometry relative to techniques such as immunoblotting allow overall reduction in protein abundance to be accounted for, and the global nature of the technique affords the use of 100’s of proteins with known solubility profiles as internal controls. In conclusion, SCE-MS is an easy to implement tool that can add an extra dimension to our understanding of respiratory chain complex assembly.