Keywords

1 Mitochondria: The Powerhouses of the Cell

Oxidative phosphorylation (OXPHOS) takes place in mitochondria, and thus these organelles are crucial for the regeneration of ATP from ADP and inorganic phosphate in respiration [112]. They are the ‘powerhouses of the cell’ [84, 115]. In addition, several other essential metabolic pathways take place in these organelles, such as the β-oxidation of fatty acids [66], the formation of iron-sulfur centers [76], the urea cycle, as well as the biogenesis of pyridines, nucleotides and phospholipids [116]. Mitochondria take part in the cellular Ca2+ homeostasis [61, 127], and they play an important role during the progression of programmed cell death referred to as apoptosis [25, 42, 70].

Especially in the last 2 decades, it has become clear that mitochondria also play a crucial role in a number of human diseases [36, 94], including diabetes mellitus [93], cancer [12, 15, 86], neurodegenerative diseases such as Parkinson’s and Alzheimer’s [72, 106, 128], and several others. Further, mitochondrial (dis-) function has been linked to the cellular aging processes, characterized by impaired levels of oxidative phosphorylation and increasing amounts of reactive oxygen species [88, 115].

2 Mitochondrial Heterogeneity Between Different Species

The term mitochondrium [6] is derived from the Greek words mitos, which stands for fiber or thread, and chondros, which means grain or corn. Put together, they can be translated as a thread-like grain; thus, the word already indicates the heterogeneity of the mitochondrial morphology, which has been known since this organelle was first described [68].

In the last 3 decades, most studies investigating the morphology and dynamics of mitochondria have relied on various forms of far-field fluorescence microscopy imaging mitochondria that were stained with specific fluorescent dyes or tagged with fluorescent proteins [62]. Mitochondrial morphology and dynamics have been studied in many eukaryotic organisms, ranging from monocellular yeasts [63, 95, 96] to higher multicellular eukaryotes, including plants [3, 77, 122] and mammals [7, 27, 71, 75, 108].

Examples of diverse mitochondrial shapes in different organisms are shown in Fig. 1. The spherical mitochondria in the guard cells of a tobacco plant (Fig. 1b) vary considerably from the tubular mitochondria in budding yeast (Fig. 1a) or from the complex mitochondrial networks of a cultivated human cancer cell (Fig. 1c). Mitochondria in plant cells often do not form a continuous network, and they are frequently located next to chloroplasts. Indeed, plant mitochondria differ substantially from mitochondria of other eukaryotes in a number of aspects, including different strategies for genome maintenance, genetic decoding, gene regulation and organelle segregation [9, 80, 104, 105]. In the plant Arabidopsis thaliana, the mitochondrial proteome encompasses about 3,000 proteins [10], whereas about 1,000 different mitochondrial proteins were predicted for budding yeast [85, 103] and about 1,500 for human cells [85].

Fig. 1
figure 1

Different mitochondrial morphologies in fungi, plant and mammalian cells. Fluorescence micrographs of the budding yeast Saccharomyces cerevisiae (a), a guard cell of the tobacco plant Nicotiana tabacum (b) and a cultured human (U2OS) cell (c). In a and b the mitochondria (green) were labeled by the expression of the green fluorescent protein (GFP) targeted to the mitochondrial matrix. In c the mitochondria (green) were immunostained by an antibody against the mitochondrial outer membrane protein Tom20. The chloroplasts (b) (red) were visualized using their strong autofluorescence. The nuclei (a, c) (blue) were highlighted with DAPI

Full size image

3 Mitochondrial Heterogeneity Between Different Cell Types

In most cultivated mammalian cells mitochondria form a complex, more or less connected network [39]. These mitochondria have a similar set of proteins and fulfill similar metabolic needs; still, their actual shape may vary considerably: Whereas the mitochondria of HeLa (derived from a human cervical carcinoma) and SHEP (derived from a human neuroblastoma) cells build up a highly interconnected, dense meshwork of slightly curled mitochondrial tubules; the mitochondria of RCC-MF (human renal cell carcinoma) cells are much less interconnected, and the straight mitochondrial tubules appear to radiate from the nucleus (Fig. 2a). However, a more spherical grain-like morphology has been observed in a few cell types [2, 20]. Moreover, in addition to these overall shape differences, also more subtle but characteristic distinctions in the diameter of the mitochondrial tubules of different cell types have been reported [31].

Fig. 2
figure 2

Mitochondrial heterogeneities between different mammalian cell types. a The overall mitochondrial morphologies differ between three human carcinoma cell types (HeLa cervical carcinoma, SHEP neuroblastoma, RCC-MF renal cell carinoma). Mitochondria were immunostained with an antiserum against the mitochondrial protein Tom20; the nuclei were labeled with DAPI. Shown are representative images taken with an epi-fluorescence microscope. b The mitochondrial membrane potential (MMP) reflects the functional status of mitochondria. The mean MMP varies significantly between different mammalian cell types (from [58], with permission)

Full size image

Morphologic shape differences in genetically similar cells are not restricted to closely related mammalian cell lines, but have also been described in other kingdoms. For example, in most plant tissues, the mitochondria exhibit spherical structures of uniform diameter. However, in certain cell types within the vascular tissue, the shapes range from sausage-shaped to long worm-like forms [78].

In addition to these morphological variations, Huang et al. reported on different mitochondrial membrane potentials in various mammalian cell types (Fig. 2b). The mitochondrial membrane potential is a key indicator of cellular viability, as it reflects the pumping of protons across the inner membrane during the process of electron transport and oxidative phosphorylation. The mean MMP in the analyzed cell types ranged from −112 ± 2 mV in fibroblasts to −87 ± 2 mV in SH-SY5Y cells, suggesting specific adaptations to the energy demands of the respective cells [58].

Hence, there can be large differences in the shapes of the mitochondria of different cells. Some of these differences may be attributed to different functional tasks of the respective cells, whereas other shapes are less obvious to comprehend. Obviously, cells have a wide repertoire of potential mitochondrial morphologies that may be adapted to the specific metabolic state of the cell.

4 Mitochondrial Shape Changes in a Cell Over Time

4.1 Structural Adaptations to the Cellular Energy Demands

The main physiological role of mitochondria is to generate ATP. Frequently, a connection between mitochondrial structure and the bioenergetical requirements of the cell has been suggested [5, 13, 109, 131]. For example, in budding yeast cells (S. cerevisiae), the volume of the mitochondrial reticulum is increased up to three-fold after changing from a fermentable (glucose) to a non-fermentable (glycerol) carbon source (Fig. 3a) [34, 125]. When glucose is available in large amounts in the budding yeast, respiration is repressed and ATP is primarily produced via glycolysis (without involvement of mitochondria), a phenomenon also known as the “Crabtree effect” [24, 26, 29, 37]. Therefore, only under non-fermentable conditions are mitochondria required for ATP production. This is a clear example of mitochondrial shape adaptations to different energy needs.

Fig. 3
figure 3

Mitochondrial shape adaptations to different conditions. a The mitochondrial network of the budding yeast S. cerevisiae adapts to different carbon sources in the growth medium. Shown are 3D reconstructions of the mitochondria of living cells labeled with the green fluorescent protein (GFP). The cells were grown in a fermentable (glucose) or a non-fermentable (glycerol) growth medium. Images were taken with a multifocal multiphoton 4Pi-confocal microscope (from [34], with permission). b Electron tomographic reconstructions of rat liver mitochondria reveal changes in the mitochondrial inner membrane topology associated with the orthodox-condensed transition. The mitochondrial outer membrane is shown in red, the inner boundary membrane in yellow and the cristae in green. The depicted mitochondria have diameters of 1,500 nm (left) and 500 nm (right) (from [82], with permission). c Mitochondria in unchallenged cultivated mammalian cells (U2OS) adopt a tubular shape, forming an interconnected network (left). Upon induction of apoptosis (10 μM actinomycin D for 12 h), the network fragments, resulting in numerous small mitochondria with a spherical shape (right). For visualization, the mitochondria were labeled with an antiserum against the mitochondrial protein Tom20 and imaged with a confocal microscope

Full size image

Likewise, it has been reported that in drought-stressed spinach leaves, a decrease of the mitochondrial volume in parenchyma cells can be observed [138]. This reduced mitochondrial volume has been suggested to be caused by glucose starvation deriving from decreased photosynthetic activity.

The energy requirements of the cell may also influence the inner structure of the mitochondria. Depending on the ADP concentration, the architecture of the inner membrane can change between a condensed and an orthodox state [43]. Electron tomography revealed that in the condensed state, the matrix is compacted, and the cristae form large compartments with multiple tubular connections to the peripheral region as well as to each other. In the orthodox state, the matrix is expanded, and the cristae tend to be tubes or short flat lamellae with one or two openings in the peripheral region of the inner membrane (Fig. 3b) [82]. It has been suggested that this morphological transition could result in the elimination of diffusion bottlenecks inside large intracristal compartments that would otherwise reduce the efficiency of ATP production [82, 83].

4.2 Fusion and Fission

In healthy cells, mitochondrial fusion and fission events are in an equilibrium so that their relative rates determine the average size of the individual mitochondrial tubules and the degree of network connectivity [7, 18, 19, 56, 63, 95]. In the fission yeast Schizosaccharomyces pombe, the equilibrium is shifted during mitosis to fission, resulting in highly fragmented mitochondria [65]. A similar observation has been made in budding yeast cells undergoing meiosis [41]. A tempting explanation for these temporal fragmentations is the conversion of a low copy number organelle into a high copy number one, thus increasing the chance to distribute a sufficient number of mitochondria for each daughter cell to commence the next cell cycle [65, 79, 130].

Another process where excessive fission takes place is apoptosis. Upon induction of the cell death program, the mitochondrial network disintegrates, yielding numerous and smaller mitochondria (Fig. 3c) [17, 64, 137]. Whether this fragmentation of the mitochondrial network has a functional relevance for the ongoing cell death program or if it is just a by-product is still under debate [16, 38, 73, 98, 123, 137]. Apoptosis is also accompanied by a remodeling of the mitochondrial inner membrane [120, 129, 135, 136].

5 Mitochondrial Heterogeneity Within a Single Cell at a Certain Time

5.1 Morphological Heterogeneity

The distribution and shape of the mitochondria in many cultured mammalian cells are rather heterogeneous. Although they may disperse throughout the whole cytosol, frequently a pronounced tendency of aggregation around the nucleus can be observed (Figs. 1, 2a) [7, 20, 39]. But are the mitochondria of a cell luminally continuous? To address this question, Collins and coworkers performed FRAP (fluorescence recovery after photobleaching) experiments with different cell types including HeLa, PAEC, COS-7, HUVEC, cortical astrocytes and neuronal cells expressing a fluorescent protein in the mitochondrial matrix [20]. It was observed that the fluorescence in the bleached regions did not recover to >10% of its initial value 1 h after irradiation. These data suggest that in the analyzed cells, the mitochondria were largely disconnected, possibly indicating functional heterogeneities of the mitochondria within a single cell.

5.2 Functional Heterogeneity

For pancreatic acinar cells, the existence of three distinct groups of mitochondria with different functions was shown [97]: perigranular mitochondria, perinuclear mitochondria and peripheral mitochondria near the basal plasma membrane. Photobleaching experiments indicated that these three groups are not luminally connected and respond differently to cytosolic Ca2+ signals. Therefore, it was suggested that they participate in the local regulation of Ca2+ homeostasis.

Differences between mitochondria within a single cell were not only revealed by their response to Ca2+ levels, but also by their membrane potential. The membrane potential is a potent functional readout of mitochondrial activity and can be monitored by using positively charged, lipophilic fluorescent dyes [62, 74, 92, 113], including JC-1 (tetrachloro-1,1,3,3-tetraethylbenzimidazol-carbocyanine-iodide) [124] and TMRM (tetramethylrhodamine-methyl-ester) [35]. JC-1 is a green fluorescent monomer at low membrane potentials and forms orange/red fluorescent aggregates at high membrane potentials. The appropriateness of JC-1 has been controversially discussed, because it provides only a qualitative readout on the membrane potential, and the staining efficiency is concentration and salt dependent [92, 126, 133]. Still, using JC-1, it was demonstrated that within a single HeLa cell, mitochondria exhibiting green fluorescence coexist with red fluorescing mitochondria (Fig. 4) [20]. Heterogeneities in the membrane potential were also conclusively reported using TMRM allowing quantitative studies [14, 28, 30, 126, 132].

Fig. 4
figure 4

Heterogeneity in the functional status of mitochondria within a single cell. Heterogeneity in the mitochondrial membrane potential is revealed by the dye JC-1 in HeLa cells. JC-1 is a green fluorescent monomer at low membrane potentials and forms orange/red fluorescent aggregates at high membrane potentials. Thus, the different colors of the fluorescence emission indicate the heterogeneity of the mitochondrial membrane potential in a single cell (from [20])

Full size image

These findings demonstrate that even in cells with morphologically rather similar mitochondria, pools of functionally distinct mitochondria may exist. Hence, presumably not unexpectedly, functional heterogeneities within the mitochondria were also observed in nerve cells, which exhibit a pronounced spatial and functional asymmetry. For example, in one of the largest nerve terminals in the central nervous system of mammals, the calyx of Held [45], two mitochondrial subpopulations were described: a small mitochondrial population with complex geometries located near the presynaptic membrane called the mitochondria-associated adherens complex (MAC) and a large mitochondrial pool with a simpler architecture, which was not preferentially located near presynaptic membranes [110]. The MAC-forming mitochondria were suggested to play a central role in high rate, temporally precise neurotransmission. Further, electron tomography revealed that these MAC-forming mitochondria have a specialized ultrastructure exhibiting a polarized cristae architecture in that cristae junctions were aligned with the cytoskeleton and occurred at higher density in the mitochondrial membrane that faces the presynaptic membrane [102].

These data conclusively indicate that, at least in some cell types, morphologically and functionally distinct mitochondrial subpopulations exist. However, very little is known about whether such functional differences are due to different protein distributions within the mitochondria.

6 Nanoscopy of Protein Distributions in Mitochondria

Almost all that we know about the inner architecture of mitochondria comes from electron microscopy and electron tomography data. These techniques are very powerful for dissecting the membrane architecture of the organelles, but are generally less well suited to study the distributions of proteins, requiring specific labeling, ideally in living cells. For these challenges, fluorescence microscopy is generally the method of choice.

However, the wave nature of light imposes a seemingly fundamental limit to the attainable resolution of light microscopes. According to Abbe, the resolution limitation is ultimately rooted in the phenomenon of diffraction [1]. Because of diffraction, focusing of light always results in a blurred spot [11], whose size determines the resolution. Thus, the highest achievable resolution with objective lenses and visible light is ~180 nm in the imaging plane. When using a single lens, the resolution along the optical axis is inescapably worse; even confocal or two photon fluorescence microscopes, which stand out in their ability to provide 3D images by optical sectioning, can only distinguish fluorescent objects if their axial separation is at least 500–800 nm.

For this reason, it has been impossible to resolve protein distributions within the inner membrane of unaltered mitochondria using light microscopy [134]. Likewise, the protein density in mitochondria is apparently so high that also protein complexes in the mitochondrial matrix or in the outer membrane have proved to be non-resolvable by conventional light microscopy. In fact, not very long ago, obtaining a spatial resolution sufficient to resolve inner-mitochondrial features with an optical microscope that uses lenses and focused visible light was considered unfeasible.

6.1 Concepts to Overcome the Diffraction Barrier

In recent years, a number of ‘nanoscopy’ or ‘superresolution’ fluorescence microscopy techniques have been invented to fundamentally overcome the diffraction barrier. A number of excellent and exhaustive reviews describing the theory as well as the practical details are available [47, 48, 57, 99]. Therefore, we give here only a brief overview of the concepts of the various nanoscopy schemes.

Stimulated emission depletion (STED) microscopy [53] and ground state depletion (GSD) microscopy [52] were the first concrete and viable physical concepts to fundamentally overcome the limiting role of diffraction in a lens-based optical microscope. In brief, STED and GSD use a selected pair of bright (fluorescent) and dark (non-fluorescent) fluorophore states to restrict the bright state to subdiffraction dimensions. To this end, optical transitions are utilized that allow one to transiently switch off the ability of the dye to fluoresce by confining the dye to a dark state. The transition is effected with a light intensity distribution featuring a zero, switching the fluorescence off everywhere except at the zero where the fluorophore is still allowed to be bright. Moving the zero across the specimen switches the signal of adjacent features sequentially on and off, allowing their separate registration. This allows one to image fine structures that would otherwise be blurred in a conventional diffraction-limited image. Hence, the spatial confinement of molecular states allows the elimination of the resolution-limiting effect of diffraction without eliminating the diffraction.

In its initial demonstration, STED microscopy was realized as a point-scanning system (Fig. 5), whereby the excitation focus was a normal confocal spot and the STED focus resembled a doughnut, featuring a light intensity zero in the center [67]. Figure 5 shows a typical experimental focal intensity distribution of the excitation spot (blue), overlapped with a STED-spot (red) featuring a central intensity zero. Saturated depletion inhibits the fluorescence everywhere except at the very center of the focal region.

Fig. 5
figure 5

Schematic drawing of a point-scanning STED microscope. Excitation and STED are accomplished with synchronized laser pulses focused by a lens onto the sample, sketched as blue and red beams, respectively. Fluorescence is registered with a detector. A phase plate is placed in the light path of the STED beam to create a ring-shaped focus featuring an intensity zero in its center. Measured intensity distributions in the focus are shown on the right. The diffraction limited excitation focus is overlapped with the ring-shaped STED focus. Saturated depletion confines the region of excited molecules to the zero, leaving an effective focus of subdiffraction dimensions

Full size image

Later, the concept of STED and GSD microscopy was expanded to photoswitching molecules, including synthetic organic molecules and reversibly photoswitchable fluorescent proteins (RSFPs). This family of approaches was named RESOLFT, standing for reversible saturable/switchable optical linear (fluorescence) transitions [46, 50, 51]. The RESOLFT concepts are purely “physical” or “physicochemical” concepts, because the subdiffraction resolution is a direct consequence of the molecular transition employed. Because the position of the zero is defined with the RESOLFT concepts, e.g., it is defined where the molecules are “on” and where they are “off,” these concepts operate with any number of molecules, ranging from single to many (Fig. 6a).

Fig. 6
figure 6

Concepts to overcome the diffraction barrier. To resolve image details that are closer than the diffraction limit, all far-field fluorescence nanoscopy concepts realized so far switch the fluorophore between two distinguishable states, a bright state A and a dark state B, to construct subdiffraction images. a In the targeted readout mode, a spatial light intensity distribution I(x, t) having a zero intensity point in space switches the molecules such that one of the states (here: A) is confined to subdiffraction dimensions. The image is assembled by scanning the zero over the sample and recording adjacent features sequentially in time. To parallelize the recording procedure, the zero can also be line shaped or an array of zero lines. As the zero is translated across the object, the molecules undergo several times the transition B → A → B, which explains the need for reversible transitions A ⟷ B. b The stochastic readout mode detects single fluorophores from a random position within the diffraction zone. To this end, a molecule is transferred to a state A that is able to emit m ≫ 1 photons in a row, while the neighboring molecules remain in the dark state B. The distance between molecules in state A should be larger than the diffraction limit. The detection of m ≫ 1 photons allows the calculation of the coordinate of emission from the centroid of the diffraction fluorescence spot formed on a camera. After the recording, the molecule is switched off to B in order to allow the recording of an adjacent molecule. If it is sufficient to record a single picture, the stochastic readout requires each molecule to cycle only once B → A → B (adapted from [49], with permission)

Full size image

The concept of switching is also essential in more recent far-field fluorescence nanoscopy approaches referred to here as superresolution microscopy by single-molecule switching and localization, which differ from the RESOLFT concepts by the fact that they switch molecules stochastically in space and utilize mathematics to assemble the image (Fig. 6). This family of approaches has been initially implemented independently by several groups and named photoactivated localization microscopy (PALM) [8], fluorescence photoactivated localization microscopy (FPALM) [55] and stochastic optical reconstruction microscopy (STORM) [111].

The operating principle of these concepts is to start with the vast majority of labels in an inactive (dark) state, not contributing to the fluorescence (Fig. 6b). A small fraction (≪1%) is then stochastically transferred to the fluorescence state so that the single molecules can be individually imaged and localized to give nanometer-level precision coordinates. After the coordinates have been recorded, the bright fluorophores are then removed (e.g., by bleaching, thermal relaxation, or otherwise) so that a new subset of the fluorophores can be transferred into the fluorescent state and recorded to obtain an additional set of molecular coordinates. This process is repeated thousands of times until a sufficient number of molecular coordinates is recorded. Importantly, the molecular coordinates do not have infinite localization accuracy, but the localization accuracy depends on the number of emitted photons from the individually localized single molecule. Finally, a composite single-molecule nanoscopy image of all these coordinates is created.

6.2 Nanoscopy on Mitochondria

As described above, several studies using conventional light microscopy or electron microscopy have provided ample evidence on morphological and functional heterogeneities of mitochondria within single cells. However, due to the previous lack of appropriate technologies, it is still largely unclear whether these heterogeneities also reflect heterogeneities in the sub-mitochondrial distribution of proteins. Until the advent of the nanoscopy concepts detailed above, it had proved to be very challenging or even impossible to address such questions because many mitochondrial protein complexes could not be resolved with conventional microscopy due to the limited resolution. For example, the TOM complexes, which are the primary import pores for nuclear encoded mitochondrial proteins, proved to be so densely packed in the mitochondrial outer membrane that it required STED microscopy to reveal individual TOM clusters in the mitochondrial outer membrane (Fig. 7a) [32]. Likewise, STED microscopy in combination with co-localization algorithms was used to determine quantitatively the degree of co-localization between hexokinase-I and each of the three isoforms of the human voltage-dependent anion-selective channel (hVDAC). The nanoscopy data showed that the degree of co-localization between hVDAC and hexokinase-I is isoform-specific, suggesting a more complex interplay of these proteins than previously anticipated [91].

Fig. 7
figure 7

STED microscopy reveals sub-mitochondrial protein distributions on the nanoscale. a Comparison of images taken with a (diffraction-limited) confocal microscope (left) and a STED microscope (right). The mitochondria of a PtK2 (rat kangaroo kidney) cell were labeled with an antibody against the outer membrane protein Tom20. In case of the STED image, individual Tom20 clusters are resolved, whereas they are blurred in the confocal case. b Two-color isoSTED images of mitochondria in Vero (African green monkey) cells allow distinguishing between proteins localized in the outer mitochondrial membrane (Tom20, left) and proteins of the mitochondrial matrix (Hsp70, right). c With the isoSTED approach, it is possible to reveal the arrangement of cristae, here in mammalian PtK2 cells (left). The cells were labeled with antibodies against the inner membrane protein complex F1F0ATP-synthase. The brackets indicate parallel cristae arrangements, and the arrows indicate regions devoid of cristae. Alterations in the cristae structure could be observed when depleting the protein mitofilin (right), which controls cristae morphology (modified from [117] and [118], with permission)

Full size image

Recently, the implementation of STED with opposing lenses, called isoSTED microscopy, has enabled the recording of the interior of mitochondria with a 3D resolution of better than 50 nm in all room directions [117]. Using this approach, Schmidt et al. analyzed the distributions of various proteins within these organelles (Fig. 7b) [117]. By labeling the F1F0ATPase, a protein complex that lines the inner membrane, these authors could also delineate the flow of the inner membrane in the mitochondria of intact cells, revealing heterogeneities in the cristae arrangements (Fig. 7c) [118].

These data conclusively demonstrate that with the now widely available nanoscopy approaches, it is possible to analyze sub-mitochondrial protein distributions. We expect that in the near future these approaches will be utilized to correlate functional heterogeneities with the nanoscale distribution of mitochondrial proteins, providing new insights into the heterogeneity of mitochondria on an additional level.

7 Quantitative Image Analysis of Mitochondria

The pronounced heterogeneity of mitochondria on the functional and morphological level not only between individual cells, but also within them, immediately prohibits a biologically meaningful analysis of subtle mitochondrial differences based on individual images. A plethora of computational tools has been developed to analyze large image data sets quantitatively [4, 21–23, 30, 54, 59, 60, 69, 81, 89–91, 100, 114, 119] and have been reviewed expertly [33, 40, 44, 87, 101, 107, 121]. The availability of nanoscopy/superresolution techniques to a growing scientific community will undoubtedly further increase the need for quantitative data analysis and at the same time will elicit new challenges, including among others the fact that with these techniques the attainable optical resolution is similar to the size of the used fluorescent labels.