Individual cristae within the same mitochondrion display different membrane potentials and are functionally independent

1Mitochondria utilize nutrients and molecular oxygen to generate a membrane potential (ΔΨm) across the inner mitochondrial membrane (IMM). The energy available for ATP synthesis is directly derived from ΔΨm (Mitchell, 1961; Mitchell & Moyle, 1969); therefore, depolarization directly translates to decreased energy availability for ATP synthesis.

2Classical studies suggested that the ΔΨm was homogeneous along the IMM. Data supporting the equipotential model are (i) mitochondria labeled with ΔΨm‐dependent dyes show a homogeneous signal along a single mitochondrion visualized under a low resolution microscope, indicating that the ΔΨm is likewise homogeneous all along the organelle (Amchenkova et al, 1988; Skulachev, 2001); and (ii) an elongated mitochondrion stained with a ΔΨm‐dependent dye appears to instantaneously lose its ΔΨm following laser‐induced damage to a small (≤ 0.5 μm2) region, suggesting that a mitochondrial filament is analogous to a power cable, where, if one part is compromised, the voltage will simultaneously collapse across its entire length (Amchenkova et al, 1988; Skulachev, 2001; Glancy et al, 2015). These conclusions were drawn after imaging mitochondria with ΔΨm‐dependent dyes performed with light microscopes lacking sufficient spatial resolution to visualize the ultrastructure of the IMM. Furthermore, previous studies lacked the temporal resolution to determine whether laser‐induced depolarization leads to an instantaneous collapse of ΔΨm across the whole organelle.

3The IMM consists of subcompartments called cristae and inner boundary membrane (IBM) (Palade, 1953). Cristae are invaginations protruding into the mitochondrial matrix, whereas the IBM runs parallel to the outer mitochondrial membrane (OMM). Cristae and IBM are connected via narrow tubular or slit‐like structures, known as crista junctions (CJs). In recent years, studies show that components of the electron transport chain (ETC) are confined to the lateral surfaces of the cristae rather than equally distributed along the IMM (Vogel et al, 2006; Wilkens et al, 2013). Moreover, dimers of F 1 F 0 ATP Synthase assemble in rows along the edges of the cristae (Dudkina et al, 2005; Strauss et al, 2008; Davies et al, 2011). The CJs can be kept in a closed state by oligomers of the inner‐membrane dynamin‐like GTPase, OPA1 (Frezza et al, 2006; Pham et al, 2016), as well as components of the mitochondrial contact site and cristae organizing system (MICOS complex) (John et al, 2005; Rabl et al, 2009; Barrera et al, 2016; Glytsou et al, 2016).

4These findings provide a conceptual framework, where protons pumped by the ETC across the cristae membrane appear first in the cristae lumen (Busch et al, 2013; Pham et al, 2016). However, unless the ΔΨ of the crista membrane is kept more negative compared to its neighboring IBM, protons would not remain in the cristae. This consideration implies that differences in ΔΨm would exist between the cristae membrane and the IBM.

5To establish whether the distribution and structural properties of OXPHOS complexes are functionally significant, it would be critical to directly visualize and quantify the ΔΨm in relation to the IMM in living cells. If the ΔΨm is uniform from one end of a mitochondrion to another, the ΔΨm would be equal at any point along the IMM, supporting the equipotential model. If, however, the ΔΨm stems from cristae functioning as independent and heterogeneous compartments, the ΔΨm would vary substantially along the IMM—between cristae and IBM, as well as between different cristae. Testing such hypotheses, nonetheless, has remained virtually unfeasible, because the only way to resolve the IMM has been with the electron microscope, which requires freezing or fixation of mitochondria and therefore precludes any direct measurement of structure and ΔΨm.

6To overcome this limitation, we developed a novel approach for imaging the IMM at high spatiotemporal resolution in living cells, using the LSM880 with Airyscan as well as STED microscopy. Staining active mitochondria with various dyes, we verified that we can resolve cristae from IBM. We then used various ΔΨm‐dependent dyes to explore how the intricate architecture of the IMM relates to the most basic mitochondrial function—the ΔΨm generated by the electrochemical gradient of protons.

7Previous studies show that components of OXPHOS are unevenly distributed between the cristae and IBM (Vogel et al, 2006; Wilkens et al, 2013), suggesting the possibility of ΔΨm heterogeneity along the IMM within a single mitochondrion. To develop an approach for the imaging of ΔΨm associated with cristae and IBM, we first sought to determine whether we could resolve the compartmentalization of the IMM in living cells. To address this question, we incubated various cell types with 10‐N‐nonyl acridine orange (NAO), a fluorescent probe that preferentially binds cardiolipin but also shows some affinity for other phospholipids found in mitochondria, such as phosphatidylethanolamine (PE) and phosphatidylinositol (PI) (Leung et al, 2014). Imaging mitochondria from living HeLa, L6, and H1975 cells with the LSM880 equipped with Airyscan technology, we resolved intramitochondrial structures, typically perpendicular to the long axis of the mitochondrion, resembling cristae, as observed in electron micrographs (Fig 1A). Accordingly, the high resolution of the Airyscan‐based microscopy allowed separation of cristae structures, IBM, as well as dimmer regions, appearing to be matrix (Fig 1B). To verify that they were matrix, we used matrix‐targeted DsRed to label the matrix in H1975 cells and stained their IMM with NAO (Fig 1C). Airyscan imaging confirmed that the dimmer regions of NAO fluorescence within the mitochondria showed the strongest matrix‐DsRed signal and vice versa (arrowheads). To confirm that we correctly identified the cristae structures in cells stained with NAO, we tested whether the pattern of NAO labeling was changed in cells with disrupted cristae structure. As a model, we used HeLa cells with Crispr/Cas9‐mediated KO of Mic13 (Fig EV1A), which destabilizes CJs and disrupts cristae structure (Fig 1D) (Anand et al, 2016; Guarani et al, 2015). Compared to control HeLa (Fig 1A and B), Mic13‐KO mitochondria showed a substantial decrease in the number of perpendicular structures, supporting their identification as cristae. As a second model of cristae perturbation, we examined H1975 cells with stable KD of PTPMT1 through lentiviral transduction encoding shRNA (Fig EV1B). PTPMT1 is a mitochondrial phosphatase, essential for biosynthesis of phosphatidylglycerol, a precursor of cardiolipin. Deletion of PTPMT1 has been shown to result in severe derangement of the IMM (Zhang et al, 2011). Our Airyscan imaging of sh‐Scramble (sh‐Scr) control mitochondria shows structures closely resembling normal cristae (Fig 1E), whereas sh‐PTPMT1 mitochondria display a variety of deformed structures (Fig 1F) analogous to cristae perturbations that were previously observed in electron micrographs of PTPMT1‐deficient models (Zhang et al, 2011). Overall, these data demonstrate that Airyscan technology can resolve mitochondrial ultrastructure in living cells. We subsequently used this approach to measure ΔΨm at the different compartments along the IMM and determine the level of heterogeneity in ΔΨm within the individual mitochondrion.

8The power‐cable model of the ΔΨm assumes an electrical continuity along the IMM without any electrical resistance (Skulachev, 2001). However, the possibility of heterogeneity in ΔΨm along the IMM could not be investigated until now. To image ΔΨm along the IMM, we stained HeLa cells with NAO and TMRE (Farkas et al, 1989; Loew et al, 1993). Remarkably, TMRE appeared to align with the IMM in a non‐homogeneous manner, where the most‐intense TMRE signal colocalized with NAO at cristae (Fig 2A). To substantiate this observation, we looked at L6 (rat myoblast) cells, which presented the same heterogeneous pattern as HeLa cells (Fig 2B).

9When excited by a 488‐nm laser, NAO has an emission spectrum that is limited to green wavelengths. However, in these experiments, we needed to excite NAO with the 488‐nm laser while simultaneously exciting TMRE with the 561‐nm laser, resulting in NAO being exposed to both 488‐ and 561‐nm lasers. The observed alignment of TMRE signal with the membrane staining by NAO raised the possibility that exciting NAO with the 561‐nm laser could result in red light emission and thus be wrongly detected as TMRE. To address this possibility, we imaged cells stained with NAO alone and excited simultaneously with 488‐ (NAOEX) and 561‐nm (TMREEX) lasers (Fig 2C, top row). We found that emission of NAO after excitation with the 561‐nm laser (TMREEX) was undetectable. After adding TMRE to the cells, initially stained with NAO alone, and then exciting with the 561‐nm laser using the same power, we observed the appearance of strong signal in the red channel, with most‐intense pixels colocalizing with NAO at cristae (Fig 2C, bottom row).

10To further validate our findings with TMRE and NAO, we used two additional dyes, MitoTracker Green (MTG) and Rhodamine123 (Rho123). MTG covalently binds to various proteins embedded in the cristae membrane and, as such, is considered a ΔΨm‐independent dye, although its initial sequestration in mitochondria depends on ΔΨm (Presley et al, 2003). Rho123 is a ΔΨm probe, which partitions to mitochondria in a transient way, indicating changes to ΔΨm (Ward et al, 2000; Duchen, 2004). We found that MTG colocalized with TMRE, showing a similar heterogeneous pattern (Fig EV2A). Then, we examined the partitioning of Rho123 and found that it shows the most‐intense signal associated with cristae (Fig EV2B).

11To further verify that the staining patterns of TMRE depend on ΔΨm, we used a previously described method to influence ΔΨm: Continuous exposure of TMRE‐stained mitochondria to the 488‐nm laser results in robust and rapid depolarization and repolarization, a phenomenon known as flickering (Duchen et al, 1998). We reasoned that the smaller portion of TMRE bound to the membrane in a ΔΨm‐independent manner would remain during the flickering event and reveal the level of noise. Moreover, if the differences in TMRE fluorescence intensity (FI) between cristae and IBM depend on ΔΨm, we would expect that the differences between the brightest and dimmest pixels would markedly decrease during depolarization. Conversely, following repolarization, we would expect these differences in pixel intensities to return. Figure 3A shows an example of a mitochondrion from a HeLa cell stained with MTG and TMRE (arrowheads), where we initially observed the heterogeneous patterns of TMRE; however, at ~ 9 s, the mitochondrion depolarized, and the heterogeneous staining pattern of TMRE was lost. Notably, at ~ 16 s, this mitochondrion repolarized and exhibited nearly the same TMRE heterogeneity as before the depolarization. Quantification of the changes in ΔΨm during the flickering phenomenon demonstrates that ~ 85% of TMRE signal was lost during depolarization (Fig 3B). Moreover, the differences between the brightest and dimmest areas (cristae and matrix, respectively) are attenuated during transient depolarization but are reestablished following restoration of ΔΨm. These data support that the heterogeneous staining patterns of the TMRE are due to differences in ΔΨm.

12The Nernst equation can be used to quantify ΔΨm by acquiring the FI of ΔΨm‐sensitive probes (e.g., TMRE). The FIs of the probes at different subcellular compartments can be used to extrapolate the differences in concentrations of the probe, which are needed to calculate the difference in ΔΨm between compartments (Ehrenberg et al, 1988; Farkas et al, 1989; Loew et al, 1993; Wikstrom et al, 2007; Twig et al, 2008). We used the average TMRE FI of the mitochondria as a reference point to calculate the ΔΨm of the different compartments, an approach similar to that employed in multi‐electrode ECG. We found the voltage at cristae to be significantly higher than at IBM (Fig 4A and C). These data indicate that the hetero‐potential along the IMM consists of at least two basic segments—the cristae and the IBM. Representing pixel intensities in pseudo‐color as a LookUp Table (LUT), where white and blue correspond to the highest and lowest ΔΨm, respectively, it is apparent that the voltage associated with cristae (arrowheads) is generally higher than IBM, emphasizing the electrochemical discontinuity between these contiguous regions of the IMM (Fig 4B).

13We next explored whether the ΔΨm of primary cells would follow the same pattern. We stained mitochondria in primary hepatocytes with TMRE and found ΔΨm heterogeneity similar to that in HeLa cells, indicating that discrete electrochemical domains also exist along the IMM of differentiated cells with strong mitochondrial oxidative function (Fig 4D–F).

14To further confirm the ΔΨm differences between cristae and IBM observed using Airyscan imaging, we determined whether the ΔΨm would display the same heterogeneity by super‐resolution microscopy (e.g., STED). Using living HeLa cells stained with TMRM, we found a nearly identical pattern in the heterogeneity of ΔΨm, where, notably, cristae ΔΨm significantly exceeds that of IBM (Fig 4G–I). Altogether, these data demonstrate that the voltage associated with cristae is significantly higher than that of IBM, which is consistent with the higher concentration of ETC components associated with cristae membranes.

15F 1 F 0 ATP Synthase is primarily localized to the rims of cristae (Dudkina et al, 2005; Strauss et al, 2008; Davies et al, 2011). If cristae maintain ΔΨm that is distinct from the IBM, then inhibiting the consumption of the proton gradient by F 1 F 0 ATP Synthase would further increase the difference between IBM and cristae. To assess this hypothesis, we inhibited F 1 F 0 ATP Synthase with oligomycin and measured the difference in the ΔΨm between the cristae and IBM (ΔΨCr‐IBM). We found that not only did oligomycin increase the total ΔΨm (Fig 5A and C), as expected (Farkas et al, 1989), but it also increased the ΔΨCr‐IBM, hyperpolarizing the cristae (Fig 5B and C). Conversely, we tested whether treatment with FCCP, a protonophore that shuttles protons across the IMM, would diminish the ΔΨCr‐IBM. Consistent with previous studies (Farkas et al, 1989; Loew et al, 1993), we determined that FCCP decreased the total ΔΨm (Fig 5A and C) and resulted in mitochondrial fragmentation (Fig EV3A and B) (Duvezin‐Caubet et al, 2006; Ishihara et al, 2006; Griparic et al, 2007). Furthermore, we observed that treatment with FCCP decreased the ΔΨCr‐IBM (Fig 5B–D).

16To determine the role of CJs and cristae architecture in the generation of ΔΨCr‐IBM, we used cellular models in which proteins controlling the formation of cristae and CJs are perturbed. To date, the most notable of these are the family of MICOS‐complex proteins (Rabl et al, 2009; Harner et al, 2011; Hoppins et al, 2011; von der Malsburg et al, 2011; Zerbes et al, 2012; Barbot et al, 2015; Friedman et al, 2015; Guarani et al, 2015; Barrera et al, 2016; Hessenberger et al, 2017; Rampelt et al, 2017; Wollweber et al, 2017) and the GTPase, Opa1 (Frezza et al, 2006; Cogliati et al, 2016). To determine whether the MICOS complex and/or Opa1 are required to preserve the ΔΨCr‐IBM, we examined ΔΨCr‐IBM in cells with either Opa1 or MICOS‐complex subunits deleted.

17Firstly, we examined the effect of deleting Mic13 on ΔΨCr‐IBM in HeLa cells. While Mic13‐KO cells showed a marked decrease in the number of cristae, we identified some regions of IMM that maintained cristae structures (Fig 6A; arrowheads). Mic13‐KO mitochondria showed decreased heterogeneity of TMRE staining along the IMM and ΔΨCr‐IBM was significantly diminished, compared to control HeLa cells (Fig 6A and B). To further address the role of the MICOS complex in regulating the mitochondrial hetero‐potential, we examined cells in which Mic60 or Mic10 were deleted (Fig EV1C). In these cells, we observed a significant diminishment of heterogeneity of TMRE FI along the IMM (Fig 6C), as well as a drop in ΔΨCr‐IBM (Fig 6D and E).

18We next studied the effects of Opa1 deletion on ΔΨCr‐IBM. Various studies have demonstrated that not only does Opa1 associate with components of the MICOS complex (Barrera et al, 2016; Glytsou et al, 2016), but it appears to have an independent function as a molecular staple, holding the cristae in a closed configuration (Frezza et al, 2006). We hypothesized, therefore, that deletion of Opa1 (Fig EV1D), like deletion of components of the MICOS complex, would disrupt the electrochemical boundaries between the cristae and the IBM, thereby equilibrating the potential along the IMM. Our data indicate that deletion of Opa1 indeed results in a significant loss in ΔΨCr‐IBM (Fig 7A and B). Altogether, these data indicate that cristae architecture and CJs are essential for the formation of the electrochemical boundaries that allow for ΔΨm of cristae to remain different from that of IBM.

19Intriguingly, in ~ 25% of Mic10 KO cells, we observed structures that appeared to have detached from the IBM and become hyperpolarized (Fig 6C, lowest row, arrow). To test whether these hyperpolarized structures in Mic10 KO cells had no membrane continuity with the IMM, we performed laser‐induced depolarization of mitochondria that included such vesicles. If these hyperpolarized vesicles were unanchored from the IMM, we would expect that they would remain polarized despite the collapse of ΔΨm elsewhere in the mitochondrion. Figure 8A shows that such a hyperpolarized structure (arrowhead) maintained its ΔΨvesicle for at least 15 s after the rest of the mitochondrion had depolarized. We determined that the ΔΨvesicle‐IBM of such vesicles was significantly higher than cristae that maintained attachment to the IBM (ΔΨCr‐IBM for Control cristae = 10.11 mV vs. ΔΨvesicle‐IBM for Mic10 KO vesicles = 25.85 mV; P = 0.0005; N = 3 independent experiments).

20We further hypothesized that deletion of both Opa1 and Drp1 (Fig EV1D) would result in an extreme case of electrochemical discontinuity of the IMM. Indeed, our imaging of Opa1&Drp1 DKO MEFs demonstrates the formation of strongly hyperpolarized vesicles moving haphazardly within the matrix (Fig 8B; Movie EV1 shows time series of hyperpolarized vesicles). Compared to control MEFs that had a ΔΨCr‐IBM of 10.67 mV, the hyperpolarized vesicles in the Opa1&Drp1 DKO MEFs showed a ΔΨvesicle‐IBM of 39.55 mV (P = 0.0014; N = 3 independent experiments).

21Hyperpolarized vesicles were quite rare in control cells, suggesting that they are formed in response to bioenergetic dysfunction and/or impaired cristae formation.

22Previous studies suggest that a single mitochondrion constitutes a single bioenergetic unit (Amchenkova et al, 1988; Skulachev, 2001; Glancy et al, 2015). A typical experiment would involve laser‐induced depolarization at one tip of a mitochondrion, which would appear to instantaneously result in a complete collapse of the ΔΨm, leading to the conclusion that the entire organelle functions as one electrochemical unit. These time‐lapse experiments determined ΔΨm at intervals of 5 s or longer after laser‐induced depolarization and were employing microscopes that allowed for single mitochondrion, rather single crista, resolution (Amchenkova et al, 1988; Skulachev, 2001). A 5‐s interval, however, is enough time to generate and propagate a soluble signaling molecule and/or changes in membrane architecture that could induce depolarization of independent cristae structures in an entire mitochondrion. Thus, to test whether laser‐induced depolarization reflected (i) an instantaneous collapse in voltage due to the mitochondrion functioning as an uninterrupted electrochemical conduit; or (ii) a gradual loss of voltage due to a chemical signal and/or structural changes propagating through the organelle, we re‐ran the same experiments as previously described but acquired images ~ 20× more rapidly. Visualizing mitochondria from rat myoblasts (L6) using Rho123 (Fig 9A; Movie EV2 shows time series), we exposed a small mitochondrial region (~ 0.5–1 μm2) to a rapid, high‐power pulse of the 2‐photon (2‐P) laser (arrow, white box). We then observed mitochondria depolarizing in a wavelike (i.e., gradual) manner: The ΔΨm‐dependent dye dispersed more rapidly near the original site of laser stress, prior to dispersing along areas farther away (arrowheads follow the loss of ΔΨm over time). The Rho123 signal is displayed as a multicolored LUT, where pixel intensity is color‐coded on a scale where the most‐intense pixels appear white (highest ΔΨm) and the least‐intense pixels appear blue (lowest ΔΨm). Examining the images immediately following laser‐induced depolarization (~ 0.30, ~ 0.45, and ~ 0.60 s), we found that the mitochondrion loses ΔΨm first in proximity to the site of laser stress, because the pixels become more green and blue, whereas the distal site remains polarized, as indicated by the white and red pixels. In Fig 9B, we plotted the pixel intensity of the mitochondrion in 9A as a function of distance from the top to the bottom at different time points. In the image immediately preceding laser‐induced depolarization (red line), the peaks and valleys occupy a similar range of intensities from one end of the mitochondrion to the other. Following depolarization induced by the 2‐P laser pulse, Rho123 intensity proximal to the site of laser stress drops substantially, while the signal intensity at the distal end remains almost unchanged (green line). These asymmetrical changes in FI are visible at later time points (blue and purple lines), until ~ 2.1 s, when the mitochondrion has lost virtually all of its ΔΨm (dark purple line).

23In previous studies, laser‐induced depolarization experiments were often performed on mitochondria in fibroblasts, which tend to be remarkably elongated. Therefore, fibroblasts represent an optimal system to test the model of the mitochondrion as an “electrical wire” or “power cable”, where loss of ΔΨm in one place would lead to simultaneous depolarization across the whole organelle (Amchenkova et al, 1988; Skulachev, 2001). Staining mitochondria in human fibroblasts with MTG and TMRE, we performed laser‐induced depolarization experiments (Fig 9C). Loading of MTG depends on ΔΨm, but it is retained in mitochondria even after depolarization, because, once inside the organelle, it binds covalently to thiol moieties of various proteins (Presley et al, 2003). Partitioning of TMRE to mitochondria, on the other hand, depends entirely on ΔΨm. Following 2‐P laser pulsation (arrow, white box), there is greater and greater loss of TMRE over time, while the signal from MTG remains. Figure 9D, which shows the TMRE channel alone, demonstrates that, even though the mitochondrion begins to depolarize proximal to the site of the initial laser stress, it maintains its ΔΨm, distally, for a period of time. To quantify this asymmetry in ΔΨm, we measured TMRE FI immediately after 2‐P laser pulsation at sites distal to the initial perturbation (≥ 10 μm from white box) vs. proximal (≤ 1 μm from white box; Fig 9E).

24Quantification shows that in the seconds following laser‐induced depolarization, mitochondria tend to remain significantly more polarized at sites distant from targeted laser stress, compared to sites close to it. The time gap of seconds for the propagation of the depolarization wave does not support the notion that the IMM functions as a single electrochemical conduit. We next determined the frequency that mitochondria appeared to depolarize in a wavelike vs. instantaneous manner. Time‐lapse imaging at approximately 300 ms/frame, we were significantly more likely to observe wavelike instead of instantaneous depolarization (Fig 9F).

25Together, using different cell lines and mitochondrial dyes, our data suggest that the ΔΨm is organized into multiple, disparate electrochemical domains along the length of the IMM. It further suggests that the laser‐induced depolarization phenomenon propagates within the mitochondrion through the diffusion of a signal and/or structural changes rather than by the membrane acting as a continuous electrochemical element.

26If cristae possess a measure of functional autonomy, then it is possible that different cristae within a single mitochondrion could maintain significantly different membrane potentials (ΔΨCr) from each other. To determine whether one crista within a single mitochondrion has higher ΔΨCr compared to a neighboring crista, we first defined the level of stability of ΔΨCr of individual cristae and then asked if the difference between two cristae is larger than the ΔΨCr fluctuation observed over time in a single crista. To explore this possibility, we performed time‐lapse imaging of mitochondria stained with MTG and TMRE, and we examined the stability of ΔΨCr generated by different cristae over time. The membrane potential of each crista was calculated by referencing it to the IBM to produce a value of ΔΨCr‐IBM. Figure 10A shows an example of four different cristae—“A–D” (arrowheads)—that we followed for a period of ~ 10 s. Figure 10B shows that the average ΔΨCr‐IBM over time of Cristae C and D is more than double that of Cristae A and B, indicating that cristae maintain different membrane potentials. To further examine their relative functional stability over time, we quantified the standard deviation of ΔΨCr‐IBM within each crista over time and compared it to the standard deviation of ΔΨCr‐IBM between cristae over time. Our data indicate that, over a period of ~ 10 s, there are significantly larger differences between cristae within a single mitochondrion than the variability recorded from any individual crista over time (Fig 10C). To further investigate the extent to which cristae could exhibit functional autonomy, we studied time‐lapse images of mitochondria stained with MTG and TMRE and discovered instances where mitochondria would undergo partial depolarization (Fig 10D–F; arrowhead, arrows; [Link], [Link], [Link] show time series of partial depolarization from E, F). Remarkably, instances of partial depolarization of a single mitochondrion revealed cristae that remained polarized despite the depolarization of adjacent cristae. Analyzing the specific frames showing partial depolarization, we quantified the ΔΨCr‐IBM of the polarized cristae vs. depolarized cristae and determined that the polarized cristae maintained a higher ΔΨCr‐IBM than neighboring cristae that appeared to have depolarized (Fig 10G). Altogether, our data demonstrate that cristae display a degree of functional autonomy, highlighting a new role of these critical structures as independent bioenergetic units.

27ΔΨm is the main driving force for proton re‐entry through F 1 F 0 ATP Synthase into the mitochondrial matrix (Mitchell, 1961; Mitchell & Moyle, 1969). In this study, we provide evidence, for the first time, that cristae maintain higher ΔΨm than IBM and that each individual crista within the same mitochondrion can maintain ΔΨm distinct from neighboring cristae. Using the LSM880 with Airyscan and STED microscopy, we directly visualized the relationship of the ΔΨm to the IMM in respiring mitochondria. Our observation that cristae and IBM have heterogeneous ΔΨm is based on differential partitioning of ΔΨm‐dependent probes to the different segments of the IMM. We provide several lines of evidence to support that this differential partitioning of the ΔΨm‐dependent probes is reflecting ΔΨm and not an artifact of dye binding to membranes: (i) We repeated the observation with the unique ΔΨm‐dependent probes, TMRE, TMRM, and Rho123 (Farkas et al, 1989; Loew et al, 1993; Duchen, 2004). If partitioning to cristae were based on non‐Nernstian binding, it is unlikely to occur with different dyes. (ii) ΔΨCr‐IBM increased in response to oligomycin and decreased in response to FCCP. (iii) Partitioning to the cristae was decreased in models lacking MICOS‐complex subunits or Opa1. (iv) Analyzing flickering events, heterogeneity of TMRE FI along the mitochondrion was markedly decreased during instances of depolarization and reestablished following restoration of ΔΨm. This confirms that TMRE signal stemming from ΔΨm‐independent binding of TMRE to the cristae in our experimental system was negligible. (v) The data relating to ΔΨCr‐IBM obtained by Airyscan technology were confirmed by STED microscopy. The imaging was performed in different laboratories with different batches of HeLa cells. (vi) We provided data to confirm the same observations in MEFs, HeLa cells, Hap1 cells, L6 myoblasts, H1975 cells, primary mouse hepatocytes, and human fibroblasts. Altogether, this evidence strongly supports the conclusion that cristae and IBM are sufficiently separated, thus being electrochemically insulated from each other, to allow for different membrane potentials to co‐exist along the IMM. A limitation of this methodology is the lack of dyes that are completely independent of ΔΨm or cardiolipin content. As a result, the NAO and MTG staining of IBM appeared weaker than that of cristae, precluding ratiometric imaging. This limitation only pertains to the differences in ΔΨm of cristae vs. IBM but would not affect conclusions related to the ΔΨm differences between different cristae within the same mitochondrion. Moreover, the six points listed above support that the TMRE FI differences we observed between cristae and IBM indeed reflect ΔΨm differences.

28After confirming that the heterogeneous partitioning of ΔΨm‐dependent probes was in fact due to differences in ΔΨm, we examined the possibility that different cristae within the same mitochondrion could function as unique electrochemical domains and thus as independent bioenergetic units. We observed that, when imaging at high temporal resolution, laser‐induced depolarization of elongated mitochondria did not tend to result in a simultaneous collapse of ΔΨm across the entire organelle but rather in a rapid, wavelike depolarization. This discrepancy from earlier studies can be attributed to the fact that previous assays involving laser‐induced damage did not image the mitochondria rapidly enough after perturbing a small region of the organelle—for example, acquisition of a single frame expended 5–10 s (Amchenkova et al, 1988; Skulachev, 2001). Critical delays, either in the time it takes to acquire a single frame of a time series (Amchenkova et al, 1988) or in the duration of the perturbation itself (Glancy et al, 2015), will increase the likelihood of missing possible spatial differences in depolarization (originating close to the site of laser‐induced damage). Our study suggests that, unless imaging at ≥ 1 frame(s) per second, observing partial depolarization events is improbable. The tendency for one mitochondrion to exhibit localized depolarization indicates that a single organelle does not resemble an electrical wire (Fig 11A and B) but rather appears to function more like a configuration of interconnected batteries (Fig 11C and D).

29Our high‐ or super‐resolution imaging of the IMM stained with ΔΨm‐dependent dyes is consistent with a model of cristae as individual, interconnected batteries. The ΔΨm is not homogeneous along the IMM but shows significant differences between the cristae and IBM, demonstrating electrochemical discontinuity between the otherwise physically connected membranes. The relatively more negative ΔΨ of the cristae membrane is expected to exert a stronger electrical force, keeping protons proximal to the cristae membrane. Indeed, our study can provide an explanation for heterogeneous distribution of protons along the cristae membrane and between the cristae and the IMS (Busch et al, 2013; Rieger et al, 2014; Pham et al, 2016). Following deletions of various proteins involved in regulating cristae architecture, including CJ formation, we observed a decrease in the difference between the ΔΨm at the cristae membranes relative to the IBM. These data suggest that opening of CJs tends to equalize the ΔΨm between cristae and IBM by a mechanism that requires further study. The generation of hyperpolarized vesicles, following different genetic perturbations that eliminate CJs, reinforces our general observation that increasing compartmentalization of the IMM, in turn, intensifies the electrochemical gradient. Lastly, our time‐lapse imaging experiments show that the ΔΨCr‐IBM of an individual crista is less variable over time compared to the ΔΨCr‐IBM among different cristae within the same mitochondrion. Furthermore, we observe that, during transient depolarization, some cristae can maintain polarity despite the collapse in ΔΨm of adjacent cristae. In summary, our data support a new paradigm where each crista maintains its own ΔΨm that is different from both the IBM and from neighboring cristae.

30This study raises interesting questions as to why mitochondria organize the ΔΨm in this way. One advantage, for example, could be related to the fact that ΔΨm constitutes the main energy available to drive protons through F 1 F 0 ATP Synthase to produce ATP. As such, the localization of F 1 F 0 ATP Synthase at the cristae rims appears to be advantageous in terms of proximity to the batteries. Another possible advantage could be compartmentalization of ΔΨm in each crista may serve as a safeguard mechanism restricting the impact of localized damage. In the case of the equipotential model, where the inner membrane of the entire mitochondrion represents a single capacitor, a breach in membrane integrity in one crista would cause a collapse in voltage in all cristae and compromise the function of the whole organelle. If, on the other hand, the IMM could maintain numerous, discrete electrochemical gradients, like a group of batteries, then failure of one or more would not invariably jeopardize the entire mitochondrion. This may be of particular relevance in cells harboring a highly interconnected mitochondrial network as opposed to cells with less elongated and/or branched mitochondria. Furthermore, the hetero‐potential model suggests that cristae with higher ΔΨCr‐IBM could compensate for cristae with impaired function.

31Hyperpolarized and depolarized IMM potentials are associated with different states of respiration. While both uncoupling and an increased rate of ATP synthesis dissipate ΔΨm, a decrease in ATP synthesis may result in hyperpolarization and increased ROS production. The hetero‐potential model of the mitochondrion allows for different cristae to serve different functions. In this model, some cristae could be more dedicated to ATP synthesis, whereas neighboring cristae could play a role in ROS signaling. The hetero‐potential model further allows for the consideration that different cristae may engage in primarily complex II vs. complex I respiration, which are associated with different membrane potentials and could be driven by different fuels.

32In conclusion, our study identifies a new parameter of mitochondrial function: a mitochondrial hetero‐potential arising from the compartmentalization of the ΔΨm along the IMM. These findings have wide‐ranging implications for human health, because many diseases and medical complications—for example, dominant optic atrophy (Amati‐Bonneau et al, 2008; Zanna et al, 2008), ischemia‐reperfusion injury (Birk et al, 2013), and even aging (Daum et al, 2013)—are associated with severe perturbations in cristae structure. Pathogenic changes to cristae architecture may undermine the autonomy of cristae bioenergetics, leading to the conversion of the IMM into an equipotential membrane. Consequently, damage inflicted to any isolated region may impact the bioenergetics of the entire mitochondrion, rendering the cell more vulnerable to mitochondrial toxicants and metabolic stress. Restoring the architecture of the IMM, therefore, could represent a viable approach to reestablishing the disparate electrochemical gradients that are the basis for normal mitochondrial function. This study emphasizes the need to directly explore the relationship between cristae structure and function to further elucidate the etiology of such diseases and, in turn, foster more effective therapies for treating them.

33H1975 (sh‐Scramble and sh‐PTPMT1) cells were grown in RPMI‐1640 (31800‐022), supplemented with sodium bicarbonate, Pen/Strep, sodium pyruvate, HEPES, and 10% FBS, and cultured in 5% CO2. Patient‐derived fibroblasts were grown in DMEM (12100‐046), supplemented with sodium bicarbonate, Pen/Strep, sodium pyruvate, HEPES, and 10% FBS, and cultured in 5% CO2. L6, MEF, and HeLa cells were cultured in DMEM (12100‐046), supplemented with sodium bicarbonate, Pen/Strep, sodium pyruvate, HEPES, and 10% FBS, and cultured in 5% CO2. Hap1 cells were cultured in Iscove's modified Dulbecco's medium (IMDM) (I3390‐500 ml), supplemented with 20% FBS, Pen/Strep, and 2 mM Glutamax. Primary cultures were isolated from unmodified C57BL6/J males and females, 12–28 weeks old, procured from Jackson laboratories and bred in our facility. Mice were provided with water and food ad libitum, and housed 2–5 mice per cage, 12‐h light:dark cycle and at a room temperature of 22–24°C.

34Human H1975 cells expressing shRNA targeting human PTPMT1 were generated using a pLKO.1 vector harboring shRNA (PTPMT1) Clone ID XM_374879.1‐547s1c1 (Sigma‐Aldrich). Lipofectamine 2000 (Thermo Fisher) with PLKO.1 shRNA (PTPMT1) plasmid, lentiviral packaging plasmids pMDLg/pRRE, pRSV‐Rev, and enveloping plasmid pMD2.G (Addgene #12251, 12253, 12259) were used to transfect 293T cells to generate lentiviral particles. Forty‐eight hours post‐transfection, viral media was removed from 293T cells and centrifuged for 5 min at 300 g. Viral supernatant was then passed through 0.45‐μM filter. Target cells were plated at 60% confluency and infected with 1× polybrene and viral supernatant. Viral media was removed and substituted for fresh culture media after 24 h, and incubated for an additional 24 h, followed by puromycin selection. Mic13 KO model was generated through Crispr/Cas9 genome editing (preprint: Kondadi et al, 2019), according to previously described methods (Anand et al, 2016). Mic10 and Mic60 KO cells were made by Horizon (UK) (preprint: Kondadi et al, 2019). Opa1 and Opa1/DRP1 were also generated through Crispr/Cas9 genome editing.

35Cells were plated in CELLview 4‐compartment glass‐bottom tissue culture dishes (Greiner Bio‐One, 627870), PS, 35/10 mm. Dyes (100 nM 10‐N‐nonyl acridine orange, 15 nM TMRE, 5 μM Rho123, and/or 200 nM MitoTracker Green; Invitrogen) were added to cell culture media and incubated 1–3 h prior to live‐cell imaging with the alpha Plan‐Apochromat 100×/1.46 Oil DIC M27 objective on the Zeiss LSM 880 with Airyscan. Prior to image analysis, raw.czi files were automatically processed into deconvoluted Airyscan images using the Zen software. HeLa cells imaged with STED were stained with 50 nM TMRM (Invitrogen) for 30 min before imaging. Leica SP8 LSM, fitted with STED module, was used to perform live‐cell super‐resolution imaging.

36Processed Airyscan images were analyzed using ImageJ (Fiji) software. Briefly, prior to cell cropping and quantification, background was subtracted from all images using a rolling ball filter = 50. After developing analysis protocols, we designed macros for high‐throughput image quantification and analysis. For representative images in figures, we applied the Window/Level function when demonstrating relevant changes in pixel intensities; when specifically comparing cellular structures, we adjusted pixel intensities to optimally demonstrate relevant changes in structure. Images acquired with STED microscopy were deconvoluted using Huygens deconvolution software.

37SDS–PAGE assays were conducted by separating 20–30 μg of protein on precast polyacrylamide NuPAGE gels (Thermo Fisher) and transferred to nitrocellulose membranes in transfer buffer containing 10% methanol. Proteins were probed with the following primary and secondary antibodies: Opa1 1:1,000 (BD Biosciences, 612606); HSP60 1:10,000 (Abcam, ab46798); Drp1/DLP1 1:1,000 (BD Biosciences, 611112); PTPMT1 1:1,000 (Sigma‐Aldrich, HPA043932); β‐actin 1:10,000 (Cell Signaling Technology, BH10D10); TOM20 1:2,000 (Abcam, ab78547); Mic10/C1orf151 1:1,000 (MA5‐26031); Mic13/C19orf70 1:1,000 (ProteinTech 25514‐1‐AP); Mic60/mitofilin 1:1,000 (ProteinTech 10179‐1‐AP); anti‐mouse IgG, HRP‐linked antibody (Cell Signaling Technology, 7076S); and anti‐rabbit IgG, HRP‐linked antibody (Cell Signaling Technology, 7074S). Proteins were detected and imaged using SuperSignal West Femto Maximum Sensitivity Substrate (Thermo Fisher) and ChemiDoc MP Imaging System (Bio‐Rad), respectively. Densitometry was performed using ImageJ (Fiji).

38All statistical analyses were performed on GraphPad Prism and Microsoft Excel. Data sets were subjected to D'Agostino–Pearson omnibus and/or Shapiro–Wilk normality tests to assess whether data were normally distributed. Data were subjected to 2‐tailed Student's t‐tests, and P values < 0.05 were considered statistically significant. Error bars represent SEM, unless otherwise indicated. N = the number of independent experiments. On average, ~ 20 cells were analyzed per independent experiment per condition. Statistical analysis was performed on the averages from independent experiments.

39DMW and MS designed the study, made the figures, and wrote the article. DMW and MS performed all of the Airyscan imaging, Western blotting, as well as data analysis. STB generated the sh‐Scramble and sh‐PTPMT1 cell lines. AKK performed the STED imaging. RA created the Mic13 KO cell line. ASR, AMB, and DBS provided valuable recommendations, contributing to the study design and reagents. OSS and ML provided supervision, designed the study, and wrote the article.

40The authors declare that they have no conflict of interest.