Loss of TRAP1 increases OXPHOS due to an anaplerotic increase in glutamine uptake and metabolism
The gene TRAP1 was disrupted in HEK293T, HCT116, A549, and UMUC3 cells using the CRISPR/Cas9 technology and the workflow presented in Additional file 1: Figure S1a. To confirm that the TRAP1 knockout (KO) resulted in an increase in mitochondrial respiration, the cellular oxygen consumption rate (OCR), which is a measure of mitochondrial respiration, was measured in real time in WT and KO HEK293T and HCT116 cells (Fig. 1a, Additional file 1: Figure S1b). Similar to what we had found with mouse adult fibroblasts (MAFs) [3], the KO increases mitochondrial OCR (Fig. 1b) and OXPHOS-linked ATP production (Fig. 1c) in HEK293T cells grown in medium with all standard carbon sources. An analysis of the energy profile of these cells further showed that although the glycolytic potential of KO cells remained similar to the WT cells (baseline and stressed), the KO made these cells more “aerobic” and dependent upon OXPHOS under normoxic conditions when compared to the WT cells (Fig. 1d). Note that while both HEK293T and HCT116 KO cell lines exhibited increased OCR (Fig. 1a, Additional file 1: Figure S1b), the impact of the KO on OCR is not comparable across the two cell lines, probably because of their different metabolic preferences [20]. The increase in mitochondrial respiration could be suppressed in both HEK293T and HCT116 KO cells by re-introducing TRAP1, but not by overexpressing EGFP directed to the mitochondrial matrix with a TRAP1 mitochondrial targeting signal (MTS) (Fig. 1e, f). The mitochondrial EGFP construct (mitoEGFP) primarily served as a control to verify that overexpression of an unrelated protein in mitochondria did not affect the OXPHOS function. Also note that there is always a slight but statistically insignificant dip in mitochondrial respiration due to transient transfection toxicity (Fig. 1e, f).
We next wanted to identify the differential use of carbon sources underlying this respiratory dysregulation. In the central carbon metabolism, mitochondrial respiration is primarily driven by the three major carbon sources glucose (Glc), pyruvate (Pyr), and glutamine (Gln), all present in a standard growth medium. The OCRs of WT and KO cells incubated separately with each of the three carbon sources were therefore determined (Fig. 1g–i).
When grown only on glucose as the primary carbon source, an uptake assay with the fluorescent tracer 2-NBDG showed that HEK293T KO cells have a lower Glc uptake than WT cells (Fig. 1j). Consistent with this, they display a reduced OCR (Fig. 1g) and rate of extracellular acidification (ECAR), caused by lactate secretion, a measure of the glycolytic flux (Fig. 1k).
To maintain a minimal glycolytic rate and to promote pyruvate oxidation in mitochondria, WT and KO cells were grown overnight in a medium containing galactose and pyruvate (Gal + Pyr) as the only carbon sources [21]. Under these conditions, the ECAR profile tends to mimic the OCR profile because the carbon sources are primarily used for OXPHOS and the acidification comes from the carbonic acid produced with the CO2 released by OXPHOS (compare Fig. 1l with Additional file 1: Figure S1c, and panels d and e of Additional file 1: Figure S1e). Real-time respiration monitoring showed that the basal OCR in both HEK293T (Fig. 1l, h) and HCT116 KO cells (Additional file 1: Figure S1d) is decreased, indicating an overall decrease in the assimilation of pyruvate into the tricarboxylic acid (TCA) cycle. A separate OCR analysis with only pyruvate as the carbon source gave similar results demonstrating that this outcome was not due to a galactose-induced artifact (Additional file 1: Figure S1f). In contrast, OCR analysis with only Gln as the primary carbon source (Fig. 1m, i; Additional file 1: Figure S1 g) indicated a metabolic preference of KO cells for Gln. This may compensate for the reduced Glc or Pyr metabolism and indicate an anaplerotic shift, which is the replenishment of TCA cycle intermediates diverted to various biosynthetic pathways [22], in this case by the increased utilization of Gln. Similarly to Pyr alone, the ECAR profiles with only Gln mimicked the OCR profiles of both HEK293T and HCT116 cells, which indicates that Gln is also primarily metabolized in mitochondria in both cell types (Additional file 1: Figure S1 h, i).
To confirm the increased Gln uptake and utilization by KO cells, indicated by the OCR experiments, a quantitative flux tracing experiment was performed. For this, isotopically labeled Gln (13C-Gln) was added in addition to unlabeled Glc and Pyr as carbon sources (Additional file 2: Figure S2a-c and Additional file 3: Table S1 for absolute quantitation of metabolites; for 13C tracing in metabolites, see the NEI area tab in Additional file 4: Table S2). For the quantitation of metabolites, we focused on the ones with increased 13C abundance in KO cells. Both HEK293T and A549 KO cells exhibited a significant increase in total Gln and glutamate concentrations (Fig. 2a), further confirming that KO cells prefer Gln even in the presence of the other two major carbon sources (Glc and Pyr). This is also associated with an increase in the levels of traced TCA cycle intermediates (Fig. 2b) indicating that KO cell metabolism is indeed anaplerotic: the increased Gln uptake and utilization allows the replenishment of TCA cycle metabolites. This correlates with an increased sensitivity of the KO cells to the glutaminase inhibitor CB-839 (Fig. 2c). We further extended the metabolomic comparison to 42 different quantitated metabolites (Additional file 2: Figure S2 in conjunction with the NEI area tab in Additional file 4: Table S2) and also observed a notable increase in 13C-traced reduced glutathione (GSH) in both HEK293T and A549 KO cells (Fig. 2d). This may indicate an adjustment to cope with increased reactive oxygen species (ROS), which are often associated with increased OXPHOS [3, 23].
Full-length TRAP1 but not its ATPase activity is essential to regulate OXPHOS
We next investigated which parts and functions of TRAP1 are necessary to rescue the metabolic phenotype of KO cells. With our initial set of truncation mutants, we set out to test whether any of the three major domains of TRAP1, which is the N-terminal ATPase domain (N), the middle (M) or C-terminal (C) domains, or combinations, thereof could be sufficient. We designed a custom construct to express TRAP1 variants with a C-terminal HA tag and an N-terminal TRAP1-MTS to ensure that proteins are directed into the mitochondrial matrix (Additional file 5: Figure S3a). mitoEGFP was used as a control (Additional file 5: Figure S3b). As mentioned previously, this construct was used to test whether overexpression of an unrelated protein in mitochondria might non-specifically disrupt the OXPHOS function (Figs. 1e, f and 3a–d). All TRAP1 truncation mutants as well as the full-length protein were expressed with some exhibiting bands corresponding to precursor proteins with uncleaved MTS and to shorter ones due to N-terminal cleavage (Additional file 5: Figure S3c). The TRAP1 truncation mutants were then overexpressed in the HEK293T KO cells to determine OCR profiles in the presence of all three carbon sources (Fig. 3a, c). Once again, the OCR data with the mitoEGFP controls confirm a slight reduction in mitochondrial respiration due to transient transfection toxicity (Figs. 1e, f, and 3a, c). However, the slightly lower OCR of cells transfected with the control plasmid expressing mitoEGFP was still significantly higher when compared to the OCR of cells transfected with the WT TRAP1 expression plasmid (Fig. 3b, d). None of the TRAP1 truncation mutants were able to suppress the KO OXPHOS phenotype to WT levels (Fig. 3 b, d). This indicates that a full-length TRAP1 protein is essential for normal OXPHOS regulation.
Since TRAP1 is a paralog of HSP90, a molecular chaperone that is well known to be dependent on its ATPase cycle [24, 25], we speculated that the ATPase activity of TRAP1 might be required for OXPHOS regulation. To test this, we generated a panel of point and truncation mutants that affect this enzymatic activity. Note that our numbering includes the 59 amino acids of the MTS. The following ATPase activity mutants were tested: the double point mutant E115A/R402A with a 10-fold reduced ATPase activity relative to WT (Additional file 5: Figure S3d), the 30-fold hyperactive ATPase mutant ΔStrap, and the moderately activated (2.5-fold) ATPase single point mutant D158N [14]. To our surprise, all ATPase mutants are able to suppress the OXPHOS phenotype of the KO cells, reducing the OCR to WT levels (Fig. 3e–i). Similar results were obtained when the OCR analysis was done with cells in a medium with only Gln as the carbon source (Additional file 5: Figure S3e). We further confirmed the ATPase independence of the complementation by performing a separate real-time OCR analysis with murine cells comparing KO MAFs stably expressing either WT or the single point mutant E115A of human TRAP1 (Fig. 3j). Note that the mutant E115A was designed by analogy to the yeast HSP90 E33A mutant, which has been reported to be able to bind to ATP, but to be defective for ATP hydrolysis [24, 26]; E115A, similarly to the single mutant mentioned above, binds ATP, but is defective for ATP hydrolysis [15]. Thus, the ability to hydrolyze ATP, at least as well as WT TRAP1, is not essential for the regulation of OXPHOS by TRAP1.
TRAP1 primarily interacts with other mitochondrial chaperones and OXPHOS-associated proteins
While HSP90 has an exhaustive list of clients and co-chaperones [13, 27,28,29,30], the interactome of its mitochondrial paralog remains poorly characterized [6]. After ascertaining that a full-length TRAP1 is essential for OXPHOS regulation, we wondered which proteins interact with TRAP1 and whether these might explain its role in OXPHOS regulation.
We carried out an immunoprecipitation mass spectrometry (IP-MS) experiment with WT TRAP1 and the ATPase mutants E115A/R402A and ΔStrap overexpressed in HEK293T cells (Additional file 6: Figure S4a; Additional file 7: Table S3). To refine this list of identified proteins, the protein interactors were first filtered for validated mitochondrial proteins and then by limiting the dataset to proteins with 4 or more identified unique peptides. This yielded a list of 81 proteins common to WT TRAP1 and the 2 ATPase mutants; we took these to represent the most probable TRAP1 interactors (Additional file 8: Table S4). This list primarily contains other mitochondrial chaperones (for example GRP75, CH60, and PHB, which are also known as mtHSP70/mortalin, HSP60, and prohibitin, respectively), OXPHOS complex subunits (ATP synthase, complexes I and IV), channel/carrier proteins (TOM/TIM complexes, VDACs), and other mitochondrial enzymes (YMEL1, FAS, ECHA). It is noteworthy that, while we could detect the previously reported TRAP1 interactors SDHA [4, 31], COX4, ATPB, and NDUA9 [19], we did not see others including cyclophilin D [32], PINK1 [33], c-Src [3], HTRA2 [34], and SIRT3 [19] (Additional file 7: Table S3). This may be due to the differences in cell lines, relative affinities, interactor-directed IPs, or other experimental details. More unexpectedly, we did not find any enzymes directly involved in Gln metabolism, such as glutaminase, glutamine synthase, and glutamate dehydrogenase. Note that as a consequence of a decline in Glc and Pyr metabolism, the fluctuating ADP to ATP ratios in KO cells may act as a potent activator of glutaminase to fuel the TCA cycle [35, 36]. ADP has been reported to be the strongest nucleotide activator of glutaminase [35], but ATP, both at low and high concentrations, also stimulates glutaminase activity [36].
For further analysis, we used the total peptide spectral matches (PSM, a metric based on the total number of identified peptides for a given protein), to standardize and to compare the data from IPs with WT and mutant TRAP1. Once standardized to WT, interactors of individual TRAP1 mutants could be compared among themselves and as a ratio to the respective TRAP1 versions (set to 100). It is striking that TRAP1-interacting proteins segregate into two major groups based on how much protein was pulled down with WT or mutant TRAP1 (Fig. 4a, Additional file 8: Table S4). Quantitatively, the mitochondrial chaperones GRP75 (mtHSP70), CH60 (HSP60), and PHB2 are the main TRAP1 interactors while all other interactors segregate into the second less abundant group (Fig. 4a, inset).
Consistent with what has been observed for yeast HSP90 by a two-hybrid screen [37], most of the TRAP1 interactors, except the major mitochondrial chaperones mtHSP70 (GRP75) and HSP60 (CH60), have a preference for binding the TRAP1 mutant E115A/R402A, which has a tenfold reduced ATPase activity and might therefore accumulate in the ATP-bound conformation (Fig. 4b, Additional file 8: Table S4). This preference for the ATP-bound state could also be seen when low and hyperactive ATPase mutants were individually compared to WT TRAP1 (Additional file 6: Figure S4b, c).
Taken together, these results show that while the ATPase activity of TRAP1 can vary greatly without affecting OXPHOS regulation and interaction with other mitochondrial chaperones, TRAP1 ATPase activity is inversely correlated with binding to other TRAP1 interactors.
Loss of TRAP1 has a minor impact on mitochondrial and total cellular proteomes
We speculated that the absence of TRAP1 might destabilize some of its interactors or lead to a compensatory transcriptional or post-transcriptional up- or downregulation of other proteins. We used 2 separate approaches to identify such proteome changes. First, we performed a quantitative stable isotope labeling by amino acids in cell culture (SILAC) MS analysis comparing WT to KO UMUC3 cells. Almost 50% of the mitochondrial proteome (507 proteins) could be detected, of which 200 were detected in all replicates (Additional file 9: Table S5). For these 200 proteins, we found little variations comparing KO to WT cells when the minimum significant fold change is set to 2 (p < 0.05) (Fig. 4c). Even with a cutoff of 1.5-fold, only a few alterations in the mitochondrial proteome could be seen (Fig. 4c, Additional file 9: Table S5). With the notable exception of PHB2 (when a 1.5-fold change is set as a threshold), most of the mitochondrial proteins including those predicted to interact with TRAP1 (especially the subunits of the ATP synthase complex highlighted by the analysis in Fig. 4b), show no significant up- or downregulation in UMUC3 KO cells (Additional file 9: Table S5). Thus, TRAP1 KO does not have a significant impact on the portion (about 15–20%) of the mitochondrial proteome that our SILAC analysis could capture.
Second, we did a label-free quantitation (LFQ) MS analysis of the total cellular proteome with WT and KO HEK293T and HCT116 cells cultured with the 3 different cocktails of carbon sources (Glc + Pyr + Gln, Gal + Pyr only, Gln only; Additional file 10: Table S6). We reduced the initial list of 4578 proteins to 3679 proteins by using as a criterion the identification of at least 4 unique peptides per protein (Additional file 11: Table S7). The comparison of the LFQKO/LFQWT ratios for these proteins from cells cultured in a medium with all 3 carbon sources did not reveal any significant changes (Additional file 6: Figure S4d, e). Although a few proteins were observed outside the 2-fold limit, they were not consistent across HEK293T and HCT116 cells and therefore failed to correlate with the loss of TRAP1. The LFQ ratio profiles turned out to be similar for media with other combinations of carbon sources (Additional file 11: Table S7).
In toto, all three MS experiments indicated that while TRAP1 interacts with multiple mitochondrial proteins, its loss does not have much of an impact on the mitochondrial or cellular proteomes.
TRAP1 forms an oligomeric complex
Our IP-MS experiment suggested that TRAP1 associates with a number of proteins of the mitochondrial matrix in a manner independent of its own ATPase activity. To explore this further, we decided to separate mitochondrial extracts made with a non-ionic detergent from HEK293T cells on clear native polyacrylamide gels (native PAGE) capable of resolving molecular complexes between 1 MDa and 240 kDa (Fig. 5a). For the following experiments, we chose clear native rather than blue native PAGE [38] because with the latter, although possibly better suited for membrane-associated complexes, there is always the risk that the surface coating with the negatively charged Coomassie dye affects the integrity or stability of protein complexes. Overall, despite the slightly poorer resolution compared to blue native gels, clear native gels have been demonstrated to yield largely comparable results, notably for mitochondrial complexes [39]. We expected the migration of complexes with a protein such as TRAP1 with a pI of 6.40 in a separating gel at pH 8.8 to be reasonably well correlated with molecular weight and size. When blotted for endogenous TRAP1, a single molecular complex of ~ 300 kDa could be seen, which is absent from KO cells (Fig. 5a). However, the molecular weight of the detected complex was not exactly what was expected if a TRAP1 dimer was in a complex with mtHSP70, HSP60, or even both proteins. Moreover, looking at overexpressed WT or ATPase mutant TRAP1 side by side, we found that the E115A/R402A mutant forms a complex of the same size as WT TRAP1 whereas the hyperactive ATPase mutant (ΔStrap) seems to form a slightly larger or conformationally different, more slowly migrating complex (Fig. 5a).
To determine what the 300-kDa TRAP1 complex contains, we expressed a TRAP1-GST fusion protein and GST alone as a negative control and applied the workflow described in Additional file 12: Figure S5a for a GST-pulldown MS analysis. Upon setting the cutoff for an interactor at a minimum of 11 unique peptides, no mitochondrial chaperone could be detected in the excised gel piece. Apart from TRAP1, only proteins that were also co-purified with GST alone could be identified (Additional file 12: Figure S5b; Additional file 13: Table S8). Hence, the high-molecular weight TRAP1 complex (~ 400 kDa in the case of TRAP1-GST) only contains TRAP1-GST. The TRAP1 interactors mtHSP70 and HSP60 may not be sufficiently stably bound to remain associated during native gel electrophoresis. The sizes of the TRAP1 and TRAP1-GST complexes are consistent with TRAP1 forming a stable tetramer or a dimer of dimers. We were concerned that the specific mitochondrial lysis conditions could contribute to generating this unexpected TRAP1 complex; however, we observed the same complex independently of whether we prepared the mitochondrial extract without or with reducing agent and without or with any one of 3 different detergents (Additional file 12: Figure S5c).
Our results showing the existence of a previously unreported TRAP1 oligomeric complex, in all likelihood a TRAP1 tetramer, were quite surprising considering that structural [10, 15] and crosslinking [40] studies had only reported TRAP1 to exist as a dimer. To determine whether the dimer and tetramer co-exist at a steady state in mitochondria without crosslinking, we compared endogenous TRAP1 to our panel of full-length TRAP1 proteins with different tags using a clear native gel analysis capable of resolving complexes from 480 to ~ 120 kDa (Fig. 5b). We expected homodimers to migrate at the level of the 146 kDa or between the 146- and 242-kDa marker bands. Although all protomers were well expressed (Fig. 5b, lower panel with SDS gels), we did not observe any band that could correspond to TRAP1 dimers at steady state, neither with endogenous TRAP1 nor upon overexpression of TRAP1 (Fig. 5b).
We next set out to confirm the existence of TRAP1 tetramers with two orthogonal methods comparing endogenous mitochondrial TRAP1 to recombinant human TRAP1, which we purified from Escherichia coli. We reasoned that a biochemical analysis such as a blue native PAGE [38], different from what had previously been done with recombinant TRAP1, might allow us to corroborate the existence of TRAP1 tetramers with TRAP1 from an entirely different source and devoid of all other proteins present in a mitochondrial extract. Thus, we compared the migration of endogenous TRAP1 present in a mitochondrial extract of HEK293T cells with that of small amounts of purified recombinant TRAP1 by blue native PAGE. Remarkably, in both cases, some fraction of TRAP1 migrated as a large complex consistent with tetramers (Fig. 5c). Whereas the majority of recombinant TRAP1 molecules migrated as a faster complex consistent with homodimers, the ratio was more or less inverted for TRAP1 from a mitochondrial source. The slight differences in migration between TRAP1 from the two sources might be due to technical reasons relating to the vastly different amounts of total protein loaded on the gel or to post-translational modifications of the mitochondrial protein not present in TRAP1 purified from bacteria.
We further employed a single-particle cryo-electron microscopy (cryo-EM) with the same recombinant material as an additional method to demonstrate the existence of a TRAP1 tetramer and to visualize its conformation. The N-terminally closed state of TRAP1 was stabilized using the non-hydrolyzable ATP analog AMPPNP. A total of 665 micrographs were collected, from which 192,583 particles were selected. Reference-free 2D class averages revealed both TRAP1 dimer and tetramer populations (Fig. 5d). While the TRAP1 dimer can adopt different orientations, the TRAP1 tetramers were captured in a single view; this very strongly preferred orientation for the TRAP1 tetramer prevented us from pursuing its 3D structure. Nevertheless, the cryo-EM data clearly showed that TRAP1 can exist as a tetramer, even though cryo-EM conditions predominantly showed the dimer (~ 80%). As expected, the conformations for both TRAP1 dimer and tetramer shown in Fig. 5d are the closed state. Interestingly, our current model suggests that a TRAP1 tetramer might be formed by the orthogonal association of 2 dimers (Fig. 5e). These experiments neither support nor rule out the existence of TRAP1 tetramers in the apo (without nucleotide) and open (for example, ADP-bound) states, which are too flexible to be readily visualized using cryo-EM. It is noteworthy that the relative proportion of tetramers versus dimers for the same recombinant protein preparation is consistent between the two methods we have employed, that is blue native PAGE and cryo-EM. For endogenous mitochondrial TRAP1, the blue native gel indicates that the tetramer, even under these specific experimental conditions, may be the predominant form (Fig. 5c), an oligomeric form that might be preserved and further favored by our clear native PAGE method (see the “Discussion” section).
The TRAP1 complex is induced in response to OXPHOS perturbations
Based on the hypothesis that an oligomerized complex might be the functional entity of TRAP1, we checked its levels when OXPHOS is inhibited with a prolonged exposure of HEK293T cells to hypoxia in various media (Fig. 6a). Although the baseline levels of the TRAP1 complex vary in cells adapted to different carbon sources in normoxia (left part of Fig. 6a), we saw a consistent increase in the levels of the TRAP1 complex when cells were placed in hypoxia. It is notable that the maximum increase in the levels of the TRAP1 complex was observed with cells grown in Gal + Pyr medium when they were exposed to hypoxia (Fig. 6a). Cells with this carbon source combination exclusively rely on OXPHOS for respiration (Additional file 1: Figure S1, compare panels d and e). Considering that the ATP synthase is one of the major OXPHOS complexes that is inhibited by prolonged hypoxia [41] and that we had found ATP synthase components to be among the main TRAP1 interactors (see Fig. 4b), we asked whether the inhibition of the ATP synthase complex would affect TRAP1 oligomerization (Fig. 6b). To this end, we compared the levels of the TRAP1 complex from HEK293T cells exposed to hypoxia or to the ATP synthase inhibitor oligomycin under normoxic conditions. Under hypoxic conditions, the induction of the TRAP1 complex is slow and only seems to initiate around 6 h (Fig. 6b). The slow time course may reflect the slow depletion of oxygen from the medium and cells rather than a characteristic of mitochondria or the TRAP1 complex. There is also an overall increase in the levels of TRAP1 protomers in cells exposed to hypoxia (Fig. 6b, middle panel with SDS-PAGE), but this induction does not appear to be HIF1α-mediated (Additional file 14: Figure S6a). In contrast, oligomycin induces a more rapid accumulation of the TRAP1 complex above basal level without a noticeable concomitant increase in total TRAP1 protein levels (Fig. 6b).
All of the experiments presented so far regarding the TRAP1 complex were performed solely with HEK293T cells. We therefore confirmed the existence and inducibility of the TRAP1 complex in four other cell lines: breast cancer-derived cell lines MCF-7 and MDA-MB-134, the prostate cancer cell line PC3, and the colon cancer cell line HCT116. A high-molecular weight TRAP1 complex, which is rapidly further induced in response to ATP synthase inhibition, was readily detected in each cell line (Additional file 14: Figure S6b).
Next, we assessed the impact of inhibitors of the electron transport chain (ETC) on the TRAP1 complex in MCF-7 and HEK293T cells (Fig. 7a and Additional file 15: Figure S7). Both cell lines showed an accumulation of the TRAP1 complex when the ATP synthase was compromised (Fig. 7a and Additional file 15: Figure S7). In contrast to the inhibition of the ATP synthase complex (complex V of the ETC), the inhibition of complexes I or III or both reduced the TRAP1 complex levels in both cell lines (Fig. 7a and Additional file 15: Figure S7). Therefore, we tested whether the inhibition of ATP synthase could override the effects of complex I and III inactivation (Fig. 7b). This was examined at the 3 and 6 h time points with a combination of rotenone + antimycin A and oligomycin + rotenone + antimycin A in parallel. Indeed, the inhibition of ATP synthase was able to override the suppressive effect of the combined inhibition of complexes I and III on the TRAP1 complex in HEK293T cells, as can be most clearly seen at the 6 h time point (Fig. 7b).
Having found that the levels of the TRAP1 complex change upon inhibiting OXPHOS, we wondered what would happen if OXPHOS were upregulated. This question is not trivial to address experimentally as it appears that most cells in culture operate OXPHOS at or close to maximal capacity. We decided to culture HEK293T cells on glucose as the only carbon source and then to force them to divert pyruvate to OXPHOS by blocking its conversion to lactate with a lactate dehydrogenase inhibitor (LDHi) (Fig. 7c). This treatment increased the basal OCR of HEK293T cells by more than twofold compared to the low basal value of cells grown with glucose as the only carbon source (Fig. 7d). When the cells were treated for 2, 4, or 6 h with the LDHi under this condition, we observed a steady increase in the induction of the TRAP1 complex (Fig. 7e). Thus, the TRAP1 complex can be induced both in response to inhibition of OXPHOS at the level of ATP synthase and to an increase of OXPHOS.