Three-dimensional imaging of mitochondrial cristae complexity using cryo-soft X-ray tomography

Mitochondria are dynamic organelles that change morphology to adapt to cellular energetic demands under both physiological and stress conditions. Cardiomyopathies and neuronal disorders are associated with structure-related dysfunction in mitochondria, but three-dimensional characterizations of the organelles are still lacking. In this study, we combined high-resolution imaging and 3D electron density information provided by cryo-soft X-ray tomography to characterize mitochondria cristae morphology isolated from murine. Using the linear attenuation coefficient, the mitochondria were identified (0.247 ± 0.04 µm−1) presenting average dimensions of 0.90 ± 0.20 µm in length and 0.63 ± 0.12 µm in width. The internal mitochondria structure was successfully identified by reaching up the limit of spatial resolution of 35 nm. The internal mitochondrial membranes invagination (cristae) complexity was calculated by the mitochondrial complexity index (MCI) providing quantitative and morphological information of mitochondria larger than 0.90 mm in length. The segmentation to visualize the cristae invaginations into the mitochondrial matrix was possible in mitochondria with MCI ≥ 7. Altogether, we demonstrated that the MCI is a valuable quantitative morphological parameter to evaluate cristae modelling and can be applied to compare healthy and disease state associated to mitochondria morphology.


Scientific Reports
| (2020) 10:21045 | https://doi.org/10.1038/s41598-020-78150-3 www.nature.com/scientificreports/ biochemical processes in healthy and diseased cells 10 . It has been well established that structure-related dysfunction in mitochondria can lead to age-associated diseases in cardiopathies 11 and neuronal disorders 12 . Mitochondria and their internal structures are challenging to image using optical microscopy due their small size and their dynamic and sensitive behavior towards the many stresses inherent in living cells 13 . Optical fluorescence based microscopies, such as STED in living cells reached 50 nm resolution 14 while PALM 15 and STORM reached molecular-scale resolution of 20 nm 16 . The transmission electron microscopy (TEM) combined with tomography, allowed to extract information about a few microns in the mitochondria depth 17 . However, the organization of OMM and IMM inside a cell cannot be fully observed within the narrow depth of few microns provided by electrons. The light and electron-based techniques can provide high spatial resolution, but the volumetric information extracted is poor, due to the low penetration depth. The latter techniques also frequently require labeling and/or fixation procedures in sample preparation, which can induce to artifacts. On the other hand, the use of X-ray based imaging can provide information about thick and large area biological samples and under cryogenic conditions to avoid the radiation damage, for instance the modalities of X-ptychography of vegetal tissues 18 and soft X-rays of whole cells 19 . Mitochondria has been already explored by X-ray imaging to be identified in the cellular context of wild-type single cells 19,20 , in cancer cells 21,22 and isolated from the cells aiming to unveil its internal structure 23 . However, there are no reports of X-ray imaging-based analysis focusing on extracting 3D morphological and quantitative information about the OMM and IMM organization.
Soft X-ray tomography (SXT) emerged in the 90´s allowing to image whole cells without staining and chemical fixation 24 . The soft X-rays have higher penetration power in hydrated specimens as compared to electrons and produce bright-field images with higher resolution than light. The specimen illumination in the so-called "water window" of the electromagnetic spectrum (284-543 eV ) leads to specific elemental absorption lines in which organic material has ten times more contrast than water 25 . In this spectral region, the carbon and nitrogen main components attenuate the light transmission with an order of magnitude greater than does water 26,27 . This attenuation follows the Beer-Lambert law, being linear with thickness and a function of the biomolecular species present at each point in the specimen 28 . Consequently, it is possible to image thick and whole cells and to resolve intracellular structures, without staining agents 27,29 . Moreover, imaging specimens, in cryogenic conditions, with soft X-ray microscopy enables to maintain the specimens in their natural state and minimizes radiation damage during the data collection 30 .
In this work, we present a novel strategy for studying the ultrastructure of isolated mitochondria using cryo-soft X-ray tomography (SXT). Mitochondria structure were identified by the data analysis based on the 3D electron density provided by linear attenuation coefficient (LAC) 27 . The volume fraction and surface area of cristae and mitochondrial matrix compartments were accurately determined, and the cristae modelling was given by mitochondrial complexity index (MCI) 31 . It provided a quantitative parameter to characterize mitochondrial morphology which can be applied to reveal the dissimilarities between health and diseased mitochondria.

Results and discussion
Mitochondria identification by LAC in SXT images. When analyzing images from X-ray imaging, it is important to verify if radiation dose is causing sample damage 27 . The cryo-preservation is a fundamental step since it increases the radiation dose which the sample can withstand. Imaging the specimens at a cryogenic temperature at 110 K prevents radiation damage for doses of up to 10 10 Gy 32 . The total adsorbed dose in our experiments was 4.57 × 10 7 Gy, calculated according to Eq. 1 (Methods session) and considering mitochondria mass density and the experimental conditions. This dose is similar or lower compared to previous cryo-SXT dose values reported in the literature 24,29 . Moreover, we did not find compromised integrity of the structures or sample changes during the tomographic acquisition that could hamper further analysis.
Depending on the micro zone plate (MZP) used as objective we obtained 60 nm and 35 nm resolution 19 data, which allowed us to identify mitochondria by observing the high-density cristae. The image contrast was enhanced after considering the effect of the point spread function (PSF) of the optics 33 generating sharp edges and better defined mitochondrial cristae when compared with the non-linear reconstructions ( Supplementary  Fig. 1). Thus, the PSF reconstructions had a great importance to ensure accurate and realist measurements.
Since the absorption of X-rays follows the Beer-Lambert's law, each voxel contains quantitative information about attenuation allowing us to calculate the linear attenuation coefficient (LAC) (µm -1 ) for the pixel size and distinguish the different structures contained in the glass capillary. Although the mitochondria isolation process resulted in contamination with other intracellular structures (Fig. 1a), the LAC and the high resolution obtained with the cryo-SXT allowed to identify and to perform quantitative analysis with the mitochondria within the whole capillary.
We calculated the 3D LAC for the 102 mitochondria (Fig. 2a). The average LAC for the mitochondria contained in the capillary was 0.247 ± 0.04 µm −1 . The 3D geometrical parameters for isolated mitochondria (Fig. 2b) was calculated showing surface areas of 1.66 ± 0.75 µm 2 , volumes of 0.11 ± 0.07 µm 3 , lengths of 0.90 ± 0.2 µm and widths of 0.63 ± 0.12 µm. Despite of the mitochondria population heterogeneity in terms of shape, the aspect ratio presented a normal distribution with peak at 0.7, showing a tendency to spherical morphology.
In the spectral region, known as the "water window", between K-edges of carbon (E = 284 eV, λ = 4.4 nm) and oxygen (E = 543 eV, λ = 2.34 nm), water is relatively transparent whereas carbon rich molecules are one order of magnitude more strongly absorbing 25 . The reconstructed volume of the imaged samples in SXT results in a 3D density that describes the local LAC of the sample and indicates if the chemical composition of the structure provides sufficient contrast to identify biological features without any label 27 . Cell structures with a high water content, for example vacuoles, will be less absorbing and, consequently, will present less contrast than the higher carbon content organelles, such as lipid droplets 34 20 . Here, we obtained mitochondrial LAC value of 0.247 ± 0.04 µm −1 and this lower value can be explained by the low carbon content of PBS, medium where the isolated mitochondria were resuspended. The cytoplasm is rich in biomolecules and consequently more strongly absorbing and gives higher values of LAC for mitochondria measured inside the cell. Therewith, we presented a focused study on isolated mitochondria population with statistical analysis of geometry and LAC, unveiling the macromolecule 3D morphology.
Morphological and quantitative characterization of features related to mitochondria structure. We selected 5 mitochondria (M1 to M5) from the capillary pool, identified based on LAC, where the resolution achieved allowed the observation of the internal structures (Fig. 1c). www.nature.com/scientificreports/ Machine learning combined with watershed algorithm resulted in the separation between what here we call the cristae (comprising the OMM, IMM and IMS) and mitochondrial matrix (Fig. 3). As the resolution did not allow us to distinguish the separated OMM, IMM and IMS, we named the compartment containing those 3 structures as cristae as it can gives the notion of cristae configuration. The cristae correspond to 79 ± 0.04% of the total volume, while the matrix corresponds to 21 ± 0.04% (Table 1). Despite of the difference in M1 to M5 total volumes, the cristae to matrix ratio is maintained.
The matrix-to-cristae ratio distribution was compared using virtual slicing of length and width parameters (Fig. 3), indicating that, at this sampling, there is no pattern that could be established between these two compartments.
The scope of this study was to unveil the 3D mitochondrial cristae morphology using SXT to extract quantitative information. Thus, we applied the MCI modelling (Eq. 2, Methods) as previously applied to quantify the morphological complexity of smooth muscle mitochondria branches 31 . Cristae complexity was evaluated considering the parameters of the matrix surface area and volume to be sensitive to internal features. As more complex and higher area are generated by the invaginations, the more complex the structure will be and higher values for MCI will be obtained. Note that segmentation of mitochondria with MCI ≥ 7 (M1, M2 and M3 in Table 1) allowed a better visualization of the cristae into the mitochondrial matrix than mitochondria with lowers values of MCI (M4 and M5). As M4 and M5 have smaller dimensions (length and width), their structure could not be determined as the measurements were performed at the limits of resolution of the SXT. This data highlights the role of 3D high-resolution imaging to improve the structural and quantitative knowledge of important biomolecules such as the mitochondria.
Our study was performed using isolated murine mitochondria. It is known that mitochondria morphology depends on cytoskeleton organization and it may change if the cell is attached or in suspension. The same argument is valid for isolated mitochondria and it needs to be taken in consideration when studying mitochondrial structures. In most forms of microscopy, depth-of-field is linked to spatial resolution. The higher the spatial resolution the smaller depth of field 37 . In the SXT set up, the use of high spatial resolution optics (35 nm) results in a depth of field limited to imaging cells that are a few microns thick, for example yeasts, protozoan and bacteria. www.nature.com/scientificreports/ In our work, the mitochondria were isolated from vascular smooth muscles cells which are typically 100 µm in size and 5 mm maximum thick (determined by atomic force microscopy) limiting the possibility of imaging the mitochondria in the adherent culture. Thus, considering the current X-ray microscopes available, imaging and quantifying mammalian organelles is more straightforward if they are extracted from their cellular environment. The other advantage to using isolated organelles is the capability to collect statistically significant information from a single tomogram. The extracted organelles used in this study are derived from several cells and artifacts are excluded by collecting information of different mitochondria in a single sample. High resolution microscopy, such as X-ray based techniques, are time consuming and scientists depend on the short experiment times granted at external synchrotron facilities. www.nature.com/scientificreports/

Conclusion
In summary, we demonstrated that SXT 3D density information combined with the high-resolution optics is a powerful methodology to distinguish morphological features and extract accurate quantitative information about the mitochondrial architecture. The tens nanometers resolution imaging and the 3D density mapping based on the LAC of all the molecules present in the sample, allowed us to characterize mitochondria structure by soft X-ray microscopy. MCI modeling allowed us to characterize the morphology and to quantify the mitochondrial cristae complexity considering 3D parameters and providing cristae invagination structure. We envision that this methodology can be used to better understand cellular process associated to mitochondrial structure morphology such as apoptosis in cardiac and neurodegenerative diseases.  Fig. 1). Custom designed capillaries were manually pulled to about 5 µm diameter tips and 1-2 µL of mitochondria solution was loaded using a microloader. The capillaries were plungefrozen in liquid propane at 90 K using an in house developed apparatus to ensure the sample vitrification.

Methods
Cryogenic soft X-rays tomography (cryo-SXT). Capillaries containing the cryo-preserved mitochondria were place in the soft X-ray microscope 38 of XM-2 beamline in the National Center for X-ray Tomography (http://ncxt.lbl.gov) at the Advanced Light Source (http://www.als.lbl.gov) of Lawrence Berkeley National Laboratory (Berkeley, CA). The experiments were carried out in 'water window' illumination (517 eV) using either a 60-nm or 35-nm Fresnel zone plate. This optic defined the maximum spatial resolution of the tomographic reconstruction. Samples were kept in a stream of liquid-nitrogen-cooled helium gas during transferring and data collection to avoid thawing up and radiation damage 35,38 . For each data set, 90 projection images were collected sequentially in rotation axis with 2° increments, with a total rotation of 180°, using exposure times from 200 to 400 ms, depending on the sample thickness. The deposited energy mass due to absorption, D (Gy), was calculated: where F is the photon flux in ph/m 2 /s, µ is the calculated linear coefficient attenuation for mitochondria, considering the formula H 50 C 30 N 9 O 10 S 1 , E is the energy in kg m 2 /s 2 , Et is the total exposure time and ρ is the mass density. The software suite AREC3D was used to align the projection images calculate tomographic reconstructions 39 followed by linear point spread function (PSF) approximation reconstruction 33 . Image analysis. The linear attenuation coefficient (LAC) was calculated and applied for the whole images 27 .
The LAC was used to identify the mitochondria and separate them from other intracellular structures present in the buffer solution. The images segmentation was initially performed using the Trainable Weka Segmentation (TWS) machine learning tool 40