Restricted diffusion of OXPHOS complexes in dynamic mitochondria delays their exchange between cristae and engenders a transitory mosaic distribution

Mitochondria provide the vast majority of the cellular energy by oxidative phosphorylation (OXPHOS). Their role in calcium homeostasis, in the process of aging, their significance for cell life and death and their participation in the progress of diverse diseases emerged only recently, though (Beal, 2005; Lu, 2009). Mitochondrial fusion and fission dynamics (Bereiter-Hahn and Jendrach, 2010) play a crucial role in these processes. In general, the balance between fusion and fission determines the mitochondrial morphology and facilitates the exchange of metabolites, mtDNA and proteins (Nakada et al., 2001; Ono et al., 2001; Malka et al., 2005; Twig et al., 2006). Also the maintenance of a healthy mitochondrial population is directly linked to ongoing fusion/fission (Twig et al., 2008; Campello and Scorrano, 2010). The immediate outcome of mitochondrial fusion and fission such as the redistribution of membrane compounds as well as the problem of concerted membrane fusion and division have to still be worked out in the field, though.

Mitochondria possess two membranes: the outer (OMM) and the inner (IMM) mitochondrial membrane. The inner membrane harbors one fifth of total mitochondrial proteins, of which the protein complexes of the respiratory chain and the mitochondrial F₁F_O ATP synthase provide 55%. NADH dehydrogenase (NADH:ubiquinone oxidoreductase; CI) (Harmon et al., 1974), succinate dehydrogenase (CII), cytochrome bc₁ complex (cytochrome c reductase; CIII), cytochrome c oxidase (cytochrome c:oxygen oxidoreductase, CIV) and the F₁F_O ATP synthase (complex V) constitute the oxidative phosphorylation (OXPHOS) system. Multiple intrusions of the inner mitochondrial membrane leaflet enlarge the surface of the IMM many fold (Mannella et al., 1994; Perkins et al., 2010). As a consequence, the inner mitochondrial membrane is basically composed of two parts: the inner boundary membrane (IBM) facing the OMM and the intruding cristae membrane (CM), separated by the so called cristae junctions (CJ). Probably, the CM is the principal site of oxidative phosphorylation (Gilkerson et al., 2003). Beside ATP synthesis, maintenance of a membrane potential by the OXPHOS is essential for mitochondrial and cell survival. To guarantee this, different levels of mitochondrial quality control are established: (1) mitochondrial proteases degrade dysfunctional proteins (Arnold and Langer, 2002); (2) damaged organelles lose their ability to fuse and are decomposed via a process called mitophagy (Scherz-Shouval and Elazar, 2007; Kurz et al., 2008; Twig et al., 2008; Mouli et al., 2009); and (3) fusion and fission dynamics provide a mechanism to avoid the accumulation of damaged proteins by facilitating the regular remixing of mitochondrial compounds (Ishihara et al., 2003; Arimura et al., 2004). Despite this, the distribution and function of mitochondria within a cell is not necessarily homogeneous (Bereiter-Hahn and Voeth, 1998; Collins et al., 2002; Busch et al., 2006; Muster et al., 2010) and protein and membranous structures might be unevenly distributed (Wikstrom et al., 2009; Wurm et al., 2011). To further dissect this, we here analyzed the distribution of OXPHOS complexes in recently fused mitochondria by means of super-resolution fluorescence and immunoelectron microscopy (IEM). Our data suggest that during mitochondrial fusion and fission cristae are predominantly preserved resulting in patchy mitochondria at first. Restricted mobility of OXPHOS complexes in cristae membranes engenders this transitory mosaic composition of recently fused mitochondria.

Recently, we showed that mitochondria displayed a patchy distribution of OXPHOS complexes for at least some hours in fused cells (Muster et al., 2010) (Fig. 1). We here asked whether the patches from fluorescence images correspond to cristae with primarily unmixed composition of OXPHOS complexes. This question is firsthand related to the dynamics of the CM in the context of mitochondrial fusion and fission. Little is known about how CMs conduct in these processes. Certainly, cristae rearrange in the immediate fusion zone as observed by correlative microscopy (Bereiter-Hahn and Vöth, 1994; Busch et al., 2006), and cristae also remodel after apoptotic stimuli (Frezza et al., 2006). However, the (re-)organization of cristae more distal from the fusion site is still concealed. In general, two scenarios are possible: (i) single cristae are preserved when mitochondria fuse; or (ii) cristae are completely rebuilt and recomposed to form new hybrid cristae (Fig. 1). To dissect this, cristae have to be characterized by their OXPHOS complex composition, which can be done by means of IEM. Therefore, OXPHOS complexes from different mitochondria have to be distinguished, which is not possible for endogenous complexes. For the same reason photoswitchable proteins – which are useful for analyzing spreading in single cells – are not suitable for IEM analysis of mitochondrial dynamics. Thus, cells expressing OXPHOS complexes with different tags were fused and mixing was analyzed on the mitochondrial level according to an established method (Ishihara et al., 2003).

To visualize individual OXPHOS complexes, complex I was tagged with monomeric eGFP (eGFP_m) C-terminally at the 30 kDa subunit, CII was tagged with eGFP_m at the subunit B (SDHB) and CIII at the subunit 10 (6.4 kDa subunit). CIV was tagged with DsRed–HA at the Cox8a subunit that faces the IMS, and complex V was labeled at the γ-subunit. The successful assembly of the tagged subunits into complexes was shown before (Muster et al., 2010; Sukhorukov et al., 2010). For immunodetection, antibodies against the eGFP_m- or HA-tagged OXPHOS complexes were used, and the secondary antibody was probed with a gold nanoparticle (Fig. 1; supplementary material Fig. S1). Combinations of 6 nm/12 nm and 12 nm/20 nm gold were used. Control cells without labeled OXPHOS complexes were negative in IEM.

First, a distribution profile of OXPHOS complexes in the different IMM microcompartments CM, CJ and IBM was compiled. OXPHOS complexes were visualized by immunogold labeling on cryosections. Generally, the arrangement of cristae in HeLa mitochondria was rather regular, with >90% of cristae perpendicular to the longitudinal axis of the respective mitochondrion (Fig. 2A). The mean width of single cristae was rather constant with 17±2 nm (±s.d.), and the distance between two adjacent cristae was about 51±15 nm in average (Fig. 2A′). For allocation of the probed OXPHOS complexes in the IMM microcompartments, a typical crista with a length according to the average length (413 nm) was delineated. The preferential distribution of all complexes in the CM microcompartment is obvious (Fig. 2B–F). CI showed a rather homogeneous distribution in the middle part of the CM (Fig. 2B). Similar longitudinal distributions were obtained for all complexes, whereas the absolute distance of the gold labels to the crista membrane varied as predicted from the likely position of the tagged subunit.

To dissect the distribution of the OXPHOS complexes in the different microcompartments IBM, CJ and CM quantitatively, a window was moved alongside the IMM. Within this sliding window, particles were counted and respective protein localization profiles were generated as recently described (Rabl et al., 2009) (supplementary material Fig. S2). All five investigated OXPHOS complexes were predominantly located in the CM-microcompartment (Fig. 2G). The comparison of the distribution profiles clearly shows an overlap of CI, CIII, CIV and CV (Fig. 2G): 80% of CI coincided with 80% of CIII, CIV and CV. Analysis of the CJ region revealed that in contrast to the other complexes CI hardly was found in the immediate CJ region, while CII showed the highest abundance in close proximity to the CJ (65% of the molecules were found within the proximal region of the CJ region). Also the number of CII molecules decreased distal from the CJ (Fig. 2E,G). CV was homogeneously distributed within the CM including the CJ. We also plotted the relative distribution of CV under consideration of the true length of the respective crista and found a homogeneous distribution of CV along the CM (supplementary material Fig. S2C).

Taken together these data show that (1) OXPHOS complexes were predominantly found in the CM microcompartment; (2) the distribution profiles of CI, CIII, CIV and CV along the CM were similar; and (3) CII had a significant quota of molecules localized in the IBM.

Next, cells with differently labeled OXPHOS complexes were fused and analyzed at different time points after cell fusion. The mixing of OXPHOS complexes in mitochondria was quantified by calculating the Pearson cross correlation coefficient r_P for both fluorescence distributions alongside the respective mitochondrion as described earlier (Muster et al., 2010). The overview fluorescence images already show that the number of fused mitochondria increased with time: 4 h after cell fusion the area covered by yellow mitochondria was generally larger than in a specimen analyzed 2 h after fusion (Fig. 3A,D). Also the corresponding r_P - values in exemplary mitochondria were higher (r_{P 4h, fusion} = 0.42 versus r_{P 2h, fusion} = 0.26).

We next asked how an imbalance in fusion/fission would influence the distribution pattern. Overexpression of hFis1 resulted in fragmented mitochondria due to an increase in fission (James et al., 2003, Yoon et al., 2003). Hardly any mixing of mitochondrial compounds was observed in these cells (supplementary material Fig. S3). Opposite, it is known that certain stresses such as starvation can induce elongated mitochondria (Tondera et al., 2009). This stress-induced mitochondrial hyperfusion (SIMH) likely results from a shift of mitochondrial dynamics towards relative more fusion than fission events. Probably, the number of active Drp1 molecules acting at mitochondria is lowered, resulting in a reduction of fission (Gomes et al., 2011; Rambold et al., 2011). We suspected that unopposed fusion would increase the patchiness of mitochondria, since more and more fusion events enhance the amount of added mitochondrial material.

To induce SIMH, the cells were incubated under serum and amino acid deprivation in PBS for 2 hours after PEG-induced fusion. This form of starvation induced a relative higher mitochondrial fusion rate in surviving cells with a tendency to an overemphasized mitochondrial network (Fig. 3G,H). Barely at the fusion zones of the cells expressing CI-DsRed–HA and CIII-eGFP_m, respectively, the mitochondria fused and became hybrid, while in the peripheral zones mitochondria formed an interconnected network without mixing of mitochondrial compounds (Fig. 3G). These observations resemble the situation in aging cells, where increased fusion (relative to fission) was combined with a reduced exchange of mitochondrial material (Mai et al., 2010). A detailed view of the fusion area shows that the mitochondria sparsely mixed their components resulting in larger patches within mitochondria and red and green mitochondria existing next to each other (Fig. 3H). A line scan for fluorescence distribution in a mitochondrion where green as well as red fluorescence is detectible (dotted line in Fig. 3H) revealed a Pearson cross correlation coefficient of r_{P 2h,SIMH} = 0.18, indicating very low mixing (Fig. 3I). The mean Pearson cross correlation coefficient for fusion areas in SIMH cells was r_{P 2h,SIMH} = 0.22±0.14 (±s.d.), whereas the mean values for the normal fused cells were r_{P 4h,fusion} = 0.47±0.1 for 4 hours after fusion and r_{P 2h,fusion} = 0.38±0.17 for 2 hours after fusion (Fig. 3J), indicating less mixing in hyperfused mitochondria.

Since the amount of fused mitochondria with mixed composition depends on the number of fused cells in a syncytium, the time and the position within the syncytium, a quantitative analysis to determine the quota of fused-mixed versus original mitochondria is not apposite. Rather the individual analysis of fused-mixed mitochondria in more detail is appropriate to learn about the redistribution of IMM proteins and IMM dynamics during fusion and fission.

Next we analyzed the distribution pattern of OXPHOS complexes in cristae of fused mitochondria by means of IEM. Gathering of cristae composed of only one species of labeled OXPHOS complexes in fused mitochondria indicates preservation of cristae rather than remodeling of cristae. So, the analysis of cristae compositions will in addition elucidate inner membrane dynamics. We determined the distribution of labeled OXPHOS complexes in cristae in mitochondria of fused HeLa cells by IEM (Fig. 1). Fused cells were identified by the existence of at least two nuclei. Areas of at least three to five cristae that exclusively displayed only one type of gold particles were characterized as a region of preserved cristae. Single cristae with mixed composition were identified by the presence of a smaller and a larger gold particle on one crista. By screening hundreds of single mitochondria, the following pattern of distribution evolved: preserved regions with one species of labeled OXPHOS complex were repeatedly found. These parts were habitually alternating with preserved regions with the complementary OXPHOS complex of different origin indicating generally non-mixed parts. These non-mixed parts retained their original state within fused mitochondria (Fig. 4A–D, green and red arrowheads). In some fused mitochondria we found larger regions composed of several cristae with mixed composition (colored yellow), flanked by regions with cristae that harbor only one type of OXPHOS complex (Fig. 4A, arrowheads; supplementary material Fig. S4Aa). These conserved regions constituted up to 25% of the mitochondrial area. So, mixed cristae were embedded in regions where only one species of tagged OXPHOS complex was present. Within one cell, also original non-fused mitochondria were found next to fused mitochondria (Fig. 4C, green asterisk). Rarely, also mitochondria with good mixing and numerous hybrid cristae side by side were identified (Fig. 4E). Projection of the distribution of different OXPHOS complexes related to single cristae on the longitudinal axis of the respective mitochondria generated an outline that conclusively corresponded to the patterned dispersion found in fluorescence images before (supplementary material Fig. S4). In SIMH cells fused mitochondria with mixed OXPHOS complex composition were definitively less frequent as already expected from the fluorescence analysis. Also the number of hybrid cristae with mixed composition in single mitochondria was reduced (Fig. 4F,G). The overall labeling degree was comparable with normally fused cells, though.

For comparison, the distribution of OXPHOS complexes in cells coexpressing two differently tagged fusion proteins was determined by IEM. From previous fluorescence microscopic analyses we expected a more homogeneous distribution of OXPHOS complexes in coexpressing cells (Muster et al., 2010). Indeed, we found significantly more hybrid cristae with mixed composition in mitochondria of coexpressing cells compared to fused mitochondria (Fig. 4H,I; supplementary material Fig. S4B). This observation was independent of the particular OXPHOS complex combination: CI-DsRed–HA transiently cotransfected with CII-eGFP_m showed numerous hybrid cristae (Fig. 4H), which was also the case for cells stable coexpressing CI-DsRed–HA and CIII-eGFP_m (Fig. 4I). In addition, projections of stable cotransfected cells emphasized the well-mixed hybrid composition of the cristae (supplementary material Fig. S4B).

We next determined the number of hybrid cristae per mitochondrion that univocally were probed with both sizes of gold particles. This quantification confirmed that the relative amount of hybrid cristae with mixed composition was significantly lower (33±23%) in fused cells than in cells coexpressing OXPHOS complexes (46±27%, P = 3.9×10⁻²). SIMH cells possessed even less hybrid cristae (5±11%, P = 1.3×10⁻⁸; Fig. 5A). The quota of preserved cristae in single mitochondria of fused cells was higher (49%) than in coexpressing cells (46%). Eighteen percent of the cristae in mitochondria of fused cells were not labeled compared to only 8% within those of coexpressing cells. Since labeled OXPHOS complexes are diluted out by non-tagged OXPHOS complexes from the fusing partner cells, this is reasonable. To quantify the internal mixing degree, the distances between nearest neighbor gold particles of different sizes were determined in fused and hyperfused mitochondria (Fig. 5B). Indeed, the OXPHOS mixing was less in SIMH mitochondria reflected by significantly larger distances between gold particles of different size (6 nm/12 nm; 12 nm/20 nm; P-value of 9.6×10⁻⁴). The mean value for gold particle distances in SIMH cells was 200±140 nm (±s.d.) compared to 130±100 nm (±s.d.) in normally fused cells. 80% of the measured distances for the SIMH cells ranged from 70 nm to 390 nm compared to a range from 40 nm to 280 nm in normal fused cells. Obviously, the mingling of OXPHOS complexes from different origin is handicapped in SIMH cells.

We next analyzed the mixing of OXPHOS complexes in single hybrid cristae that derived from mitochondrial fusion/fission dynamics. The distance between different gold particles (respective different OXPHOS complexes) is an indicator for the mixing degree: If hybrid cristae in fused cells develop by the spreading and mixing of previous separated OXPHOS complexes from different cristae, this will be reflected in the distance between OXPHOS complexes from different origins. If the model is right, then the distance between small and large gold particles in cristae in fused cells should be larger than in cristae from coexpressing cells. In coexpressing cells, a uniform dispersion of OXPHOS complexes and thus a short distance between large and small gold particles is expected.

Data presentation as a box-and-whisker plot shows that 80% of the measured distances for fused cells (CI+CIII) cover the distance range from 35 nm to 290 nm, while in coexpressing cells 80% of the measured distances range from 39 nm to 160 nm (CI+CIII) and from 20 nm to 90 nm (CI+CII), respectively (Fig. 5C). In mitochondria of fused cells with immunogold labeled CI and CIII the mean distances between 6 nm and 12 nm sized gold particles was 123±99 nm (±s.d.; n = 65). In contrast, the mean distance between 6 nm and 12 nm gold particles on single cristae of coexpressing cells was significantly smaller: 81±49 nm (n = 65) for cells coexpressing CI and CIII (P = 2.8×10⁻³). Interestingly, in cells coexpressing CI and CII the mean distance was even smaller with 50±28 nm (n = 65; Fig. 5C). The observed difference in the distances between CI+CII compared to CI+CIII in cotransfected cells was also significant (P = 1.8×10⁻⁵), indicating a more homogeneous distribution of CII. In sum, (1) the percentage of identified cristae with mixed composition in cotransfected cells was higher than in fused cells; (2) SIMH mitochondria possessed less hybrid cristae than normally fused mitochondria; and (3) in single hybrid cristae in fused mitochondria the mixing of different complexes was sparse.

The data so far strongly support the model of a preponderant preservation of cristae and their composition during mitochondrial fusion and fission cycles. The patterned distribution of OXPHOS complexes and the low amount of hybrid cristae with mixed composition also indicate that protein diffusion apparently is no doorway to overcome this heterogeneity. To prove this experimentally we applied single molecule tracking and localization microscopy (TALM) and determined the spatiotemporal behavior of OXPHOS complexes in mitochondria in live cells (Appelhans et al., 2012).

We analyzed the diffusion behavior of CII and CV. CII was chosen because our results indicated a specific distribution profile of CII combined with better mixing behavior compared to the other OXPHOS complexes. CV is specific in its role to shape cristae curvature (Strauss et al., 2008; Davies et al., 2012), probably by dimerization and oligomerization. Thus a mobility analysis within the cristae membrane microcompartment is tantalizing.

For tracking, respective subunits were tagged C-terminally with the HaloTag7 and covalently labeled with HaloTag-ligand conjugated with tetramethylrhodamine (HTL-TMR). Substoichiometrical labeling enabled the visualization of single molecules. Trajectory maps of CII in mitochondria were generated from time series (1000 frames) by reconnecting single molecules from frame to frame by use of the multiple-target tracing algorithm (Sergé et al., 2008). Single trajectories were plotted on an averaged image of the raw data image series (Fig. 6A). To determine diffusion coefficients, step lengths were plotted, and diffusion constants were determined from grading the obtained jump size histograms by a mixture model of increasing complexity. The final complexity was validated by applying a χ²-test (Fig. 6B). By analyzing the jump size distributions, three populations characterized by different diffusibilities could be gauged: (1) mobile CII freely diffusing along mitochondria (dark blue curve); (2) a less mobile fraction (light blue curve); and (3) rather immobile CII (red curve). To prove whether the position of the tag had an effect on the stability and mobility of the complex, subunit SDHD was labeled in a control, but no difference of the diffusion coefficients was found (supplementary material Fig. S5).

For tracking, CV-Halo was specifically labeled with HTL-TMR at subunit-γ. Here, 3000 subsequent image frames with single molecules were recorded for further analysis. The generated trajectory map was overlaid with an averaged image of the raw data series (Fig. 6C). Immediately, the difference in the course of trajectories with respect to the longitudinal axis of the mitochondria is visible (Fig. 6A,C). In this example, the jump size distribution diagram revealed three sub-populations of complex V with different mobility, respectively: (1) a mobile fraction with a diffusion coefficient of D_app = 0.070 µm²/s (indicated by the dark blue fitting curve); (2) a less mobile fraction with an apparent diffusion constant D_app = 0.019 µm²/s (light blue curve); and (3) an immobile fraction with an apparent diffusion constant D_app = 0.005 µm²/s. For more detailed analysis, selected trajectories of CV were plotted and their course was evaluated with respect to the position within a mitochondrion (Fig. 6E). About 60% of the CV trajectories showed a restricted course perpendicular to the longitudinal axis of the respective mitochondrion (depicted in red). Blue trajectories represent free diffusion or a course along the longitudinal axis of the mitochondrion.

As we recently demonstrated, the orthogonal trajectories most likely reflect the movement of OXPHOS complexes in CM. In contrast, we assigned the molecules with the free respective longitudinal movements to diffusion in the IBM. The trajectories of most of the CV fell in either of these categories. In rare cases (estimated <1%), a change in course from orthogonal to random (light blue trajectories) or vice versa was found. When the perpendicular part of the course overlaid with the perpendicular course of a different molecule, we assigned this part to be restricted diffusion within one crista and the overall course to reproduce a transition between CM and IBM microcompartment. Analysis algorithms to identify these specific events for automatic quantification are not yet developed. We next asked whether differences in diffusion constants could be assigned to different localizations of molecules (IBM versus CM). So, histograms of single trajectories of CV molecules were constructed using their diffusion constants calculated from their mean square displacements (MSD, supplementary material Fig. S6). The frequency plot shows clearly two subpopulations: Subpopulation A with a mean D_app = 0.056 µm²/s and subpopulation B with a mean D_app = 0.008 µm²/s. These subpopulations were independently plotted in two trajectory maps. In the trajectory map of the molecules with higher mobility, trajectory courses were rather fortuitous, but some perpendicular courses were also present not unlike a possible confinement in cristae (supplementary material Fig. S6A). In contrast, for the majority of reduced mobile molecules the displacement was quasi-punctual or was restricted in perpendicular courses (supplementary material Fig. S6B). Although the tendency that the diffusion coefficient is related to the localization within a definite microcompartment is obvious, an univocal assignment of molecules by MSD analysis to different localizations (IBM, CM) is limited due to some inaccuracy deriving from the 2D projection of 3D trajectories as well as the imperfect structure of mitochondria (e.g. also bended cristae or regions with no cristae exist).

Finally, we compared the diffusion coefficients of CII and CV (Fig. 6F). For CII and CV, a three fraction fit best matched the distribution of jump sizes respectively in the different experiments (n≥300 mitochondria from 42 cells in n≥3 independent preparations and at least 30,000 trajectories). It clearly evolved that complex II overall is more mobile than complex V: the mobile fraction of CII had an apparent diffusion constant of D_app = 0.140±0.035 µm²/s (±s.d.; s.e.m. = 0.005 µm/s), while the mobile fraction of complex V was characterized by D_app = 0.082±0.014 µm²/s (s.d.; s.e.m. = 0.002 µm²/s). The diffusion coefficient of the less mobile fraction of CII was D_app = 0.039±0.013 µm²/s (s.e.m. = 0.002 µm²/s), while it was D_app = 0.019±0.007 µm²/s (s.e.m. = 0.001 µm²/s) for CV. A small immobile fraction (≤15%) with D_app = 0.006±0.001 µm²/s (CII) and 0.004±0 (CV), respectively was also found.

Single molecule fluorescence microscopy allows the localization of single molecules with a precision of 15 nm in living cells (Appelhans et al., 2012). We combined FPALM (Hess et al., 2006) in combination with TALM (Appelhans et al., 2012) to perform dual-color super-resolution imaging of OXPHOS complexes in fused mitochondria (Fig. 7A). One OXPHOS complex was tagged with paGFP, the second with the Halo-tag, which was posttranslational labeled with HTL-TMR. Rendered images (from 3000 single processed frames) show the distribution of fluorescence-tagged OXPHOS complexes with high resolution. Complex II-halo/TMR distribution shows a broader signal in the cross section, which is visible as a red border in mixed parts. These are probably the CII molecules located in the IBM microcompartment. The green Complex I signal is smaller in the cross section, because CI almost exclusively is found in the cristae membranes (Fig. 2B). Different states of fused mitochondria were found with respect to their OXPHOS complex distribution pattern. State I comprises mitochondria that clearly show areas in the size of several hundreds of nanometers harboring one type of labeled OXPHOS only, alternated by areas with OXPHOS complexes of different origin (Fig. 7A). Right at the border between these alternating red- and green areas, small yellow areas are visible, indicating overlap of red- and green fluorescence signals (Fig. 7A, yellow arrowheads). This strongly indicates colocalization of OXPHOS complexes, probably in the same CM. In state II, mitochondria possessed larger areas with colocalizing complexes, sometimes in the middle of a long mitochondrion, sometimes at one of the ends (Fig. 7A, state II). Mitochondria with extensive areas of homogeneous distribution of both OXPHOS complexes were also found, in some cases with a red or green part at one end (Fig. 7A, state III). We attribute this to a recent fusion of an original mitochondrion to an already well mixed mitochondrion. Altogether dual color super-resolution image analysis completed the model obtained by distribution analysis of OXPHOS in IEM micrographs of fused cells. This method provides a higher set of single molecule data than IEM but does not allow the direct visualization of underlying structures of microcompartments. Thus, single molecule fluorescence analysis has to be seen as a complementary method to IEM.

Mitochondria fuse and divide continuously, but little is known about the reformation of the inner mitochondrial membrane in these processes. Heterogeneous distribution of IMM complexes in recently fused mitochondria indicates that spatiotemporal constraints exist retarding homogeneous remixing. Here, we dissected this observation on the molecular and ultrastructural level by IEM and super-resolution tracking and localization microscopy. First, a distribution profile for all five OXPHOS-complexes in the different IMM compartments IBM, CJ and CM of mammalian HeLa cells was compiled. Complex I (NADH-ubiquinone-oxidoreductase), complex III (cytochrome c reductase) and complex IV (cytochrome c oxidase) showed a similar distribution profile along the cristae membrane. This is in line with previous observations derived from activity staining and immunolabeling of EM sections of mitochondria (Perotti et al., 1983; Gilkerson et al., 2003; Vogel et al., 2006), which designated the cristae membrane as the site of oxidative phosphorylation.

F₁F_O ATP synthase was homogeneously distributed along the cristae membranes. The enrichment of F₁F_O ATP synthase in positively curved membrane parts of cristae was observed before (Parsons, 1963; Strauss et al., 2008; Bieling et al., 2010; Truta-Feles et al., 2010), and ATP synthase dimerization probably a main determinant of cristae shaping (Paumard et al., 2002; Gavin et al., 2004; Minauro-Sanmiguel et al., 2005; Strauss et al., 2008; Davies et al., 2012). Unfortunately, the limited degree of immunogold labeling does not render the explicit confirmation of F₁F_O ATP synthase dimers possible. Beside its known localization in positively curved membranes, the presence of F₁F_O ATP synthase in negatively curved membranes such as the CJ are was rather new. Presumably, F₁F_O ATP synthase complexes in the negatively curved CJ are not dimerized but exist as single, less rigid F₁F_O ATP synthase complexes (Rabl et al., 2009). Of all five OXPHOS complexes, only F₁F_O ATP synthase and complex II (succinate dehydrogenase) would fit into a negatively curved membrane, since the intramembrane parts of CI, CIII and CIV, are too large to be integrated in a strongly negatively curved membrane of roughly 40 nm, especially when assembled in supercomplexes (Althoff et al., 2011). Indeed, hardly any complex I was found in the CJ region.

After ensuring the localization of OXPHOS complexes in the cristae microcompartment we used them as markers to resolve the cristae reorganization during mitochondrial fusion and fission dynamics. Distribution and localization analysis of OXPHOS by IEM showed that fused mitochondria possess roughly one third true hybrid cristae with remixed OXPHOS composition. The majority of cristae displayed a preserved OPXHOS composition, though. In elongated SIMH mitochondria, the number of identified hybrid cristae was lower. Dual color super-resolution images of fused mitochondria confirmed these results and underscore a piecewise or mosaic configuration of fused mitochondria with respect to the inner membrane composition. Together, our data suggest that cristae are maintained during fusion and fission and hybrid cristae with mixed OXPHOS complex distribution are only successively formed. We propose the following working model (Fig. 7B): first, mitochondria of different origin fuse. Whether cristae reform at the fusion sites is still unreadable, but they are maintained in the original arms. Restricted diffusion of OXPHOS complexes supports this preservation. Hardly, OXPHOS complexes move between existing cristae. As a consequence, hybrid cristae with mixed composition predominantly exist at the fusion sites flanked by cristae with a preserved composition of OXPHOS complexes (Fig. 7B, early state). After several rounds of fusion and fission a mosaic mitochondrion comprising alternating preserved zones (red and green parts) and parts with mixed composition (yellow) at the former fusion sites evolves (Fig. 7B, intermediate state). In the light microscope this generates the patchy appearance of mitochondria. By time and corresponding to repetitive fusion/fission cycles, mitochondrial heterogeneity decreases. Elongated SIMH mitochondria possess larger patches of conserved regions and less hybrid cristae. This can be explained when a relative increase in fusion results in the addition of further mitochondrial material without true mixing and underlines the significance of intermediate fission events for the merging process. Only by undergoing balanced fusion/fission mitochondria progressively reach a well-mixed state with respect to inner membrane protein distribution (Fig. 7B, late state).

A possible alternative to explain the mosaic composition of mitochondria is while the OMM is fused completely, the inner membrane is not (Liu et al., 2009). This would result in singular mitoplasts surrounded by a common OMM (Skulachev et al., 2004). Our EM micrographs gave no evidence for transient fusion or mitoplasts, though. The mitochondria we investigated displayed normal shape and matrix appearance, and a regular arrangement of cristae with no segmentation of the IMM. We thus exclude that incomplete inner membrane fusion is the reason for the mosaic appearance.

The explanation we provide here is that diffusion and thus spreading of OXPHOS complexes in the IMM microcompartments is restricted. Single particle trajectory analysis revealed that complex V particles showed confined diffusion in cristae membranes and punctual immobility (supplementary material Fig. S6B, box), while transitions between the CM and IBM or CM and CM compartments were only rarely observed. It was for longer suggested that cristae junctions are diffusion barriers (Sukhorukov and Bereiter-Hahn, 2009), a discussion that recently found support by the finding that a large protein complex exists in the CJ region (Harner et al., 2011; Hoppins et al., 2011; von der Malsburg et al., 2011; Alkhaja et al., 2012). It is likely that this MINOS/MitOS complex with a size of ∼1200 kDa constitutes a diffusion barrier for the transition of proteins between the CM and IBM microcompartment.

How supercomplex formation influences diffusibility of OXPHOS complexes, is not clear, yet. OXPHOS complexes I, III and IV assemble in supercomplexes of different stoichiometry (Schägger and Pfeiffer, 2000; Eubel et al., 2004; Dudkina et al., 2005; Schäfer et al., 2006; Wittig et al., 2006; Althoff et al., 2011), while dimers of F₁F₀ ATP synthase often organize in row like structures in cristae membranes (Davies et al., 2011). According to the Saffman-Delbrück model, the translational mobility is independent of the amount by which the particle sticks out of the membrane sheet (Saffman and Delbrück, 1975) and diffusion coefficients of membrane domains are only logarithmically related to domain radii suggesting little influence (Petrov and Schwille, 2008). In the future, diffusion analysis of single membrane proteins and complexes hopefully will help to elucidate these relations further.

Localization and mobility analysis of succinate dehydrogenase (complex II) revealed its exceptional role within the OXPHOS complex ensemble. Complex II has a dual role in mitochondria: it is part of the OXPHOS system, but is also a key membrane complex of the citric acid cycle. Complex II does not translocate protons and thus not necessarily need to be confined to cristae. Indeed, a substantial amount of molecules were found within the IBM microcompartment in line with previous results showing a strong activity staining in this membrane part (Kalina et al., 1969; Bertoni-Freddari et al., 2001). Mobility of CII was significantly higher than of complex V and thus the mixing of OXPHOS complexes with complex II was best. This finding endorses our model that the homogeneity of dynamic mitochondria is determined and limited by the diffusion ability of inner membrane proteins.

Following this, the described heterogeneous spatiotemporal organization of OXPHOS complexes – trapped in the cristae microcompartment – provides an explanation for mitochondrial heterogeneity. Therefore, zones of different membrane potential within single mitochondria (Bereiter-Hahn and Voeth, 1998; Collins et al., 2002; Distelmaier et al., 2008; Wikstrom et al., 2009) could be assigned to local accumulation of functional compromised OXPHOS complexes that are not or only slowly exchanged by remixing. For the future, it would be exciting to identify and characterize these zones more precisely and correlate local membrane potential with the actual distribution and functionality of the present OXPHOS complexes.

The different fluorescent fusion proteins were generated using plasmid cassettes obtained by substitutions of eGFP_m or DsRed–HA into pSEMS-26 m vector from NEB biosciences®. Indicated subunits of OXPHOS complexes were N-terminally fused to fluorescent proteins with monomeric DsRed–HA and monomeric eGFP (DsRed amplified from pDsRed-Monomer-N1 vector from HBD biosciences®; modified monomeric eGFP, a gift from J. Sieber, Göttingen), as described by Muster et al. (Muster et al., 2010). For genetic selection of double transfected cells the neomycin resistance was exchanged for a puromycin resistance in one type of vector. For tracking experiments, respective subunits were cloned N-terminally of the HaloTag7 in the pSEMS-26m-HaloTag7 vector (Lisse et al., 2011). CII-Halo and CV-Halo were then covalently labeled in situ with 1 nM HTL-TMR (HaloTag conjugated with tetramethylrhodamine ligand) for 20 min. Before measurements, the labeling medium was replaced by medium without HTL-TMR.

HeLa cells were cultured in Minimal Essential Medium with Earle’s salts (PAA Lab GmbH) containing 10% (v/v) fetal calf serum (FCS, Biochrom AG), 1% MEM nonessential amino acids (PAA lab GmbH) and 1% 4-(2-hydroxyethyl)piperazine-1-ethanesulfonic acid (HEPES, PAA Lab GmbH) (MEM⁺⁺) at 37°C with 5% CO₂. The cells were transfected with the plasmids described in the text using the method of Muster et al. (Muster et al., 2010). Neomycin (800 µg/ml G418) and puromycin (0.5 µg/ml) resistance were used for selection of stable cell lines. Hyperfusion was induced by starvation stress 30 min after PEG-mediated fusion. After incubation for 2 h in PBS, cells were prepared for IEM. Longer incubation times with PBS turned out to be too stressful for cells.

For the fusion assays, stable expressing cells (CI-eGFP_m and CIII-DsRed; CI-eGFP_m and CV-DsRed; CI-DsRed–HA and CIII-eGFP_m) were sown on 6 cm diameter Petri dishes and co-cultured for 24 h. For fusion, the cells were incubated with prewarmed 40% PEG in PBS (37°C) for 1 min. Afterwards, cells were cultured in medium for 4–7 h before microscopic analysis.

For co-transfection assays, cells stable expressing CI-DsRed–HA and CIII-eGFP_m were generated. Cells stable expressing CI-DsRed–HA were transiently transfected with CII-eGFP_m and cells stable expressing CI-pa-eGFP_m were transiently transfected with CIV-DsRed–HA.

Mixing of OXPHOS in fused cells was followed using an Olympus Fluoview FV1000 cLSM equipped with a 60× objective (UPLSAPO oil, NA 1.35) and two spectral detectors. DsRed was excited with a 559 nm laser diode and emission was recorded in the range of 580–630 nm, eGFP was excited with the 488 nm wavelength of an argon ion laser and emission was collected in the range of 500–560 nm. Measurements were performed in a sequential excitation mode to avoid cross-talk. Z-stacks of mitochondria were taken in certain region of interests (12×12 µm²), with a pixel size corresponding to 30 nm.

Cells were prepared according to previously described methods (Tokuyasu, 1978; Tokuyasu, 1973; Oorschot et al., 2002). In short, cells were fixed with 2% paraformaldehyde and 0.2% glutaraldehyde in 0.1 M phosphate buffer (pH 7.4) at room temperature, embedded as monolayer or pellet in 10% gelatin, mounted on pins, frozen in liquid nitrogen and cut at −120°C (Leica UCT). Grids were washed with PBS, 20 mM glycine in PBS, blocked with 1% BSA in PBS, incubated with the primary antibodies against eGFP_m (rabbit polyclonal to eGFP_m, Abcam, ab6556, 1∶100) and HA (mouse monoclonal to HA.11, Covance, BIOT-101L, 1∶100) for 1 h at room temperature, washed with 0.1% BSA in PBS and incubated with the secondary antibody-gold-complexes (6 nm goat anti-mouse and 12 nm goat anti-rabbit, Jackson Immuno Research, 1∶50, no. 115-195-146 and 111-205-144; 20 nm goat anti-mouse, Abcam, ab27242, 1∶50). Cells were stained in a mixture of 1.6% methylcellulose/0.4% uranyl acetate, and analyzed using transmission electron microscopy (TEM; Zeiss 902A).

Images were processed with ®Bitplane Imaris and ImageJ® (NIH, Bethesda, MD, USA). The datasets were deconvolved using Autoquant software to reduce noise and improve image quality. A stack of 1000 images recording single molecules was collected and averaged to create the equivalent of an epi-fluorescence image. Single trajectories were projected on this pseudo-average image.

TEM micrographs were recorded using iTEM software (SIS). The images were processed afterwards in Adobe Photoshop CS4, making linear adjustments to contrast and brightness. Adobe Photoshop CS4 was also used to color the original micrographs subsequently. To outline single cristae we used a Wacom Intuos®4 drawing tablet. All figures were assembled in Adobe Illustrator CS4 and converted into TIFF using LZW compression.

To dissect the distribution of OXPHOS complexes in the IMM a representative model of a crista was generated as described (Vogel et al., 2006). The dimensions of this model were obtained by averaging 141 cristae in cryosectioned HeLa cells. The cristae dimensions and the distances of the gold particles to the membranes were measured with iTEM (SIS) and superimposed to the immunogold-labeled cristae (digitalized by Adobe Photoshop CS4 and Adobe Illustrator CS4). The measured distances represent the absolute values of 160 gold particles of each antigen that were plotted to this model. Particles close to badly preserved membranes, those bound to large cristae more than 450 nm distal from the cristae junction or particles which could not be assigned to a membrane within a space of 60 nm radius around a membrane were not counted. The specific, labeled subunits of the different OXPHOS complexes required the adaptation of a radius of tolerance in which a gold particle was assigned to the membrane. This was considered for plotting the distribution ratio diagram of localization between OM/IBM, CM and matrix space. The specific distances of the labeled subunits to the membrane for each of the five complexes were determined according to former published structures (Efremov et al., 2010; Yoshikawa et al., 1998; Zhang et al., 1998; Sun et al., 2005; Watt et al., 2010). Box-and-whisker plot analyses were performed for each of the complexes using Microsoft Office Excel software (Microsoft Corporation, Redmond, WA, USA) and Microcal Origin 6.0 software (OriginLab Corporation, Northampton, MA, USA). Statistical significance was performed on original data and determined by one-way-ANOVA with Microcal Origin 6.0 software (OriginLab Corporation, Northampton, MA, USA). Analysis of the mixing degree in fused and coexpressing cells was performed by the determination of the distances of differently sized antibodies in single hybrid cristae. To compare the mixing in normally fused and SIMH mitochondria, a nearest neighbor analysis was performed because although the degree of labeling in the normal and SIMH specimen was comparable the overall number of hybrid cristae in SIMH cells was low. Thus, an analysis of distances between different labels within one hybrid cristae was limited by the number of hybrid cristae. For the analysis of the relative and absolute distribution of CV alongside cristae, approximately the same set of raw images was used. Starting from the IBM region, the true positions (in shorter and longer cristae) were recalculated to the corresponding position in an average crista (20% distal from CJ in actual cristae is equal to 20% distal from CJ in the average crista).

For single molecule (SM) recording, an inverse microscope (IX71, Olympus) equipped with a TIRF condenser (Olympus), and a back-illuminated EMCCD camera (Andor iXON 897) was used. For excitation of TMR, a solid state laser (561 nm, 200 mW; CrystaLaser) was focused onto 700 µm multi-mode-optical polarization maintaining monomode fiber (KineFlex, Pointsource) and transmitted via the rear illumination port of the microscope. For excitation, the HiLo mode (illumination with a highly inclined and thin beam) was used (Tokunaga et al., 2008). By this, the signal to background ratio can be significantly improved. The estimated intensity in the thin light sheet with a thickness of 9 µm was 25±8 kW/cm² (561 nm, 200 mW, CrystaLaser). A digitally synchronized mechanical shutter controlled exposure times. Laser light reflected from a dichroic mirror (OBS-U-M3TIR 405/488/561, Semrock) passed through a high-numerical aperture objective (150× TIRF objective NA 1.45, Olympus, UAPO). Fluorescence emission from TMR was passed through a bandpass filter (Semrock BrightLine FF01-523/610-25), and was projected on top a back-illuminated EMCCD camera yielding an image pixel size of 107 nm. Equivalent fiber and condenser equipment was used for coupling the 488 nm laser line of a Saphire laser (200 mW, Coherent) for paGFP visualization into the system. The photoactivation of paGFP by the recording 488 nm laser line was sufficient and activation by 405 nm pre-excitation was not necessary. The camera was operated continuously using the Frame Transfer Mode of the CCD allowing full-frame acquisition at 60 Hz. For localization of single emitters, a modified 2D Gaussian mask for approximation was used (Thompson et al., 2002; Gould et al., 2009). Single molecules were tracked using the multiple-target tracing algorithm (Sergé et al., 2008). Multiple deflation loops were performed to ensure identification of all particles. Particle trajectories were recovered based on maximum likelihood estimators (Sergé et al., 2008). For more details see Appelhans et al. (Appelhans et al., 2012).

In dual color super-resolution experiments a combination of fluorescent paGFP and Halo/TMR was used as described recently (Appelhans et al., 2012; Wilmes et al., 2012). paGFP and TMR were excited in parallel allowing a recording rate of 30 Hz. Up to 3000 frames were recorded. Fluorescence emissions from paGFP and TMR were passed through a bandpass filter (Semrock BrightLine FF01-523/610-25). The dual view (Photometrics) was equipped with an according dichroid mirror and filters (Chroma 585 DCXR; HQ 525/50; brightline HC 620/52). Possible crosstalk signals of paGFP signals in the TMR channel were erased by a rendering procedure after single molecule localization allowing only signal intensities above a certain threshold (usually 100 counts). The count to photon conversion factor was 20.5 under our conditions (gain 300, 150× objective/oil, 1.45 NA, 1×). For single molecule localization a 2D fitting algorithm (kindly provided by Sam T. Hess, University of Maine, Orono, ME) was used (Gould et al., 2009).

We thank Achim Paululat and Jacob Piehler for continuous support, Dirk Wenzel for initial technical advice and Christian Ungermann for critically reading the manuscript.