<i>In situ</i> submitochondrial mapping in living plants via Hessian Structured Illumination Microscopy

In situ sub-mitochondrial compartments illumination in living Arabidopsis roots via HIS-SIM. Over the past few decades, the conceptualization of mitochondria has expanded from their traditional role as energy-producing factories to that of multifaceted regulators governing cellular metabolism, signaling, and cell fate decisions (Monzel et al., 2023). Structurally, this organelle is defined by three primary compartments: the outer mitochondrial membrane (OMM), the inner mitochondrial membrane (IMM), and the mitochondrial matrix (MM). Determining the precise subcompartmental localization of mitochondrial proteins is often a decisive first step toward inferring function and constructing mechanistic hypotheses. However, mitochondria are dynamic organelles with closely apposed outer and inner membranes, and they are also highly light-sensitive and susceptible to swelling due to phototoxicity, thereby impeding the native ultrastructure and compromising the precise localization of proteins at the subcompartment level (Lee et al., 2024). Conventional approaches, such as immunofluorescence, biochemical fractionation, and immunogold labeling, are constrained by the diffraction limit, the need for fixation or disruptive preparation, or a lack of nanoscale spatial context in vivo (Christopher et al., 2021). Super-resolution microscopy (SRM) can overcome the diffraction limit but typically involves significant trade-offs (Teixeira et al., 2024). Stimulated Emission Depletion Microscopy (STED) and Stochastic Optical Reconstruction Microscopy (STORM) can achieve 20- to 60-nm resolution, but generally require high photon flux, long acquisition times, or extensive optimization to mitigate photobleaching and phototoxicity. Airyscan, a confocal-based point-scanning approach using a multi-element detector, provides c. 1.7-fold improvement over confocal resolution (c. 140-nm lateral and 400-nm axial), and is compatible with live-cell imaging (Huff, 2015; Korobchevskaya et al., 2017). By comparison, Structured Illumination Microscopy (SIM) represents a camera-based area-imaging method that typically provides c. 100-nm lateral and c. 300-nm axial resolution and is comparatively photon-efficient and well-suited for live-cell imaging (Demmerle et al., 2017). Although SIM has been widely applied in mammalian cell lines, its application in intact multicellular tissues including both animals and plants remains challenging, primarily because SIM relies heavily on the precise projection of a high-contrast illumination pattern onto the sample (Patwary et al., 2018). The recently developed Hessian/High Intelligent and Sensitive Structured Illumination Microscopy (HIS-SIM) technique generated artifact-minimized super-resolution images with < 10% of the photon dose used by conventional SIM, achieving 88-nm spatial resolution at 188 Hz (Huang et al., 2018). Its high sensitivity permits the use of sub-millisecond excitation pulses followed by dark recovery intervals to reduce photobleaching of fluorescent proteins, enabling high-speed super-resolution imaging over short time windows (i.e. 61 200 consecutive raw images captured within c. 1 min) as well as long-term live-cell time-lapse imaging over hour-long durations (Huang et al., 2018). In addition, sparse deconvolution can further enhance effective resolution and signal-to-noise ratio by applying a related idea by incorporating prior information that biological structures are spatially organized rather than random pixel-to-pixel fluctuations, allowing 60- to 85-nm spatial resolution at frame rates up to 564 Hz (Zhao et al., 2022). These properties make HIS-SIM especially relevant for intact plant roots, where mitochondria must be visualized within densely packed, walled, and optically heterogeneous multicellular tissues while preserving native cellular context, a key consideration for applying SRM to plant samples (Shaw et al., 2019). By contrast, diffraction-unlimited approaches generally trade this live-imaging compatibility for higher nominal resolution: STED relies on high-intensity depletion light; notably, even dedicated live-cell mitochondrial STED implementations have emphasized phototoxicity constraints and often required specialized dyes and/or substantial dark-recovery intervals (Schermelleh et al., 2019). To date, however, a dedicated workflow for applying this technique to visualize mitochondrial subcompartments in living plants has not been established. Here, we demonstrate that HIS-SIM, coupled with a standardized stepwise reconstruction workflow, enables robust, in situ imaging of morphology-robust mitochondrial subcompartments in living Arabidopsis roots. The differentiation at c. 100 nm resolution allows easy handling of the determination of protein localization at submitochondrial level within intact multicellular tissue of living plants. The preserved native cellular context and mitochondrial architecture are instrumental in enabling more accurate functional inference and mechanistic exploration of novel mitochondrial proteins. To enable reliable in situ assignment of mitochondrial proteins to distinct subcompartments in living plant tissues, we established a standardized HIS-SIM workflow for Arabidopsis roots (Fig. 1a,b; Supporting Information Table S1). We benchmarked the workflow using established fluorescent markers targeting three subcompartments: UBP27–YFP (OMM), CLS–YFP (IMM), and ScCOX4-SP–YFP (MM). Briefly, Arabidopsis roots were directly mounted on glass slides with coverslips for observations. HIS-SIM equipped with a ×100/1.5 NA oil immersion objective (Olympus) was used for image acquisition. A 2D-SIM dataset that consisted of nine phase-shifted raw frames per focal plane was obtained for an individual region of interest. The dataset was then subjected to Wiener deconvolution for background removal, defocus elimination, and artifact removal. Then, Gaussian filtering was applied for noise reduction. Finally, sparse deconvolution was conducted to further improve the resolution and contrast (See details in the Materials and Methods section; Table S1). The stepwise strategy, Wiener reconstruction followed by sparse deconvolution, is applied to act as a conservative regularization pipeline that suppresses high-frequency noise while preserving resolvable structural information (Patwary et al., 2018; Zhao et al., 2022). As shown in Fig. 1(c), relative to wide-field imaging, Wiener reconstruction improved the delineation of mitochondrial boundaries, while sparse reconstruction further increased effective spatial resolution, resulting in enhanced separation and more precise spatial definition of distinct subcompartment marker signals. During observation, we identified two distinct mitochondrial morphologies in root cells: the rod-shaped and the spherical-shaped (Fig. 1d). All three subcompartment markers displayed a characteristic spatial signature in both morphologies, highlighting an important consideration for live plant mitochondrial imaging: Organelle morphology can influence the apparent distribution of matrix vs membrane signals, underscoring the necessity for morphology-robust imaging approaches. Plant mitochondria are highly dynamic and can shift between discrete particles and transient networks through fission and fusion, processes linked to development and stress responses (Arimura, 2018). Thus, the ability to resolve subcompartment features across diverse shapes is a practical prerequisite for connecting ultrastructural remodeling to physiological states in living plant tissues. To determine whether HIS-SIM can distinguish between different mitochondrial subcompartments in plant tissues, especially the OMM and the inner boundary membrane of IMM, which are separated by a narrow distance of c. 20 nm (Fig. S1) (Kuhlbrandt, 2015), we conducted colocalization analysis of the markers described previously with MPPα, a plant mitochondrial electron transport chain complex III (mETC III) subunit that is associated with the IMM (Klusch et al., 2023). In both rod-shaped and spherical mitochondria, MPPα-RFP was encapsulated by the OMM marker, whereas it forms a membrane-like layer surrounding the MM marker (Fig. 2a,c). In comparison, the MPPα-RFP signal exhibited an overlapping pattern with the IMM marker in both types of mitochondria (Fig. 2a,c). In our observations, the consistent ordering of peaks (OMM marker outside of MPPα, the latter of which seals the matrix) across mitochondria indicates that HIS-SIM can recover a reliable radial stratification, even if it cannot directly measure the true nanometer-scale membrane gap. To further quantify the extent of resolution that can differentiate the submitochondrial markers with MPPα, we conducted line profiles of GFP/RFP fluorescence. The result showed a peak-to-peak separation between MPPα-RFP and the OMM marker ranging from 97 to 162 nm, whereas MPPα-RFP peaks overlapped closely with the IMM marker (Fig. 2b,d). It should be noted here that, although the physical intermembrane space between the OMM and IMM is c. 20 nm, the peak-to-peak distances measured in our study should not be interpreted as the physical membrane spacing. Instead, they reflect the apparent separation of fluorescent marker distributions after 2D projection, optical blurring, and image reconstruction. IMM can be subdivided into inner boundary and cristae membranes, and proteins including respiratory-chain components are not uniformly distributed across these subdomains (Gilkerson et al., 2003; Vogel et al., 2006), resulting in heterogeneous labeling density. Therefore, the larger OMM-MPPα peak separation observed here could be better interpreted as marker separability under in vivo imaging conditions rather than a direct estimate of the OMM-IMM distance. The optical resolution that can be achieved following our pipeline reached 80 nm based on a nanoruler assay (Fig. S2). Therefore, the peak-to-peak distances ranging from 97 to 162 nm provide a quantitative summary of marker separability achieved in vivo. Several factors likely contribute to the observed spread, including local curvature, which changes the projection of 3D geometry into the imaging plane, and mitochondrial orientation relative to the optical axis. Importantly, the lower end of this range is within the expected performance envelope of linear SIM in biological samples (Demmerle et al., 2017), whereas the upper end is consistent with additional broadening from tissue scattering and labeling geometry in plant samples (Shaw et al., 2019). Together, these results consistently support the IMM association of MPPα and validate the feasibility of HIS-SIM for visualizing mitochondrial sub-compartments and discriminating OMM from IMM under native conditions in intact plant tissue. From the biological perspective, the capacity to separate OMM- and IMM-associated signals in living roots provides a new approach to examine how mitochondrial organization is remodeled during development and stress. Genetic impairment of mitochondrial fission compromises tapetal mitochondrial dynamics and pollen maturation (Arimura, 2018; Izumi, 2019), whereas drought, chilling, and oxidative stress differentially alter the plant mitochondrial proteome and mitochondrial dysfunction activates no apical meristem (NAM), arabidopsis transcription activator factor 1/2 (ATAF1/2), and cup-shaped cotyledon (CUC2)-mediated retrograde signaling (Taylor et al., 2005; Ng et al., 2013). In root stress responses, Fe-toxicity tolerance requires S-nitrosoglutathione reductase to limit nitrosative and oxidative cytotoxicity (Li et al., 2019), and this pathway has recently been connected to ANAC044-dependent mitochondrial stress signaling and stem-cell death in root meristems (Yan et al., 2025). These studies suggest that mitochondrial responses involve not only changes in organelle morphology but also dynamic reorganization of proteins across the OMM, IMM, and matrix. By enabling low-photon, live-tissue separation of submitochondrial signals, HIS-SIM provides a means to resolve protein-level dynamics in intact roots and to link submitochondrial protein organization with developmental and stress-response signaling. All Arabidopsis thaliana materials used in this study are Columbia-0 background. Mitochondrial subcompartment marker lines (UBP27–YFP (outer membrane), CLS–YFP (inner membrane), and ScCOX4–SP–YFP (matrix)) (Pan et al., 2014) were kindly provided by Prof. Ronghui Pan (Zhejiang University). To conduct colocalization analysis, the coding sequence of MPPα (AT3G16480) was amplified using 2×Phanta Flash Master Mix (P510-01; Vazyme, Nanjing, Jiangsu, China). PCR product was fused into a pCAMBIA1300-based vector linearized by EcoR I and Sal I restriction enzymes using a one-step cloning kit (C115-01; Vazyme) to generate the 35S:MPPα-RFP expression construct. The recombinant plasmid was transformed into Agrobacterium tumefaciens GV3101 (pSoup) and introduced into Arabidopsis using the floral-dip method (Zhang et al., 2006). Positive transgenic lines were screened by hygromycin resistance followed by RFP fluorescence determination and were used for crossing with the mitochondrial subcompartment marker lines. Arabidopsis seeds were surface-sterilized with 10% sodium hypochlorite and rinsed 3–4 times with deionized water. After stratifying at 4°C for 2–3 d, seeds were sown on Yamagami medium (Li et al., 2017) supplied with 1% (w/v) sucrose and solidified with 1.5% (w/v) purified agar (01056-15; Nacalai Tesque, Kyoto, Japan). Seedlings were grown under a 16 h : 8 h, light : dark photoperiod at 22°C for 6–7 d before imaging. For living plant samples, before imaging, Arabidopsis seedlings were mounted in water on a microscope slide, with the cover slip physically suspended above the roots by two parallel strips of tape affixed to the slide. The prepared slide was immediately used for observation and imaging. Imaging was performed on a commercial HIS-SIM system (Guangzhou CSR Biotech Co. Ltd, Guangzhou, Guangdong, China) in a 2D-SIM configuration. During preview, the intensity of the structured-illumination pattern was set to 20% of the full power. After selecting the region of interest, pattern intensity was set to 60–80% of the full power with an exposure time of 20 ms per frame for imaging a 2D-SIM dataset that consisted of nine phase-shifted raw frames per focal plane for each channel (Huang et al., 2018). Excitation and emission wavelengths were set as follows: 488 and 525–550 nm for YFP; 561 and 600–660 nm for RFP. For each slide, imaging was completed within 5–10 min to minimize the effects of potential physical stress or hypoxia, and all of the processes were conducted in a room with the temperature controlled at 23°C. To determine the optical resolution that can be achieved by our HIS-SIM platform, the commercial SIM 80B-brightline (https://www.gattaquant.com/) single color nanoruler immobilized on a coverslip was purchased and used. After putting it on the microscope with the coverslip side facing the objective, sample observation and imaging processes were consistent with the plant samples described previously. Raw image datasets were processed through a stepwise reconstruction pipeline consisting of (1) generalized Wiener reconstruction, (2) Gaussian denoising, and (3) sparse deconvolution. The algorithmic description of each step is given later so that the pipeline can be reproduced in any SIM-reconstruction environment: In this framework, the different shifted spectral components are combined using the system optical transfer function (OTF), while a constant regularization term (Wiener parameter, w) suppresses amplification of high-frequency noise. In our analysis, we used w = 0.5, which corresponds to the ‘Image Denoise L4’ preset in HIS-SIM IMAGER (Guangzhou CSR Biotech Co. Ltd). To reduce out-of-focus contributions and suppress OTF sidelobes, an attenuation function of the form g(k) = 1 − a·exp(−k2) was applied (Wen et al., 2021). The attenuation parameter was set to a = 0.5, corresponding to moderate suppression of low-frequency background while preserving structural information. This corresponds to the software-specific setting ‘Defocus elimination L3’. Periodic reconstruction artifacts were removed using a frequency-domain notch filter N(k) = 1 − exp(−k2/β2) (Wen et al., 2021), with β = 30. This step selectively attenuates residual structured noise without affecting the main signal band. This corresponds to the software-specific setting ‘Artifacts remove L3’. The Wiener-reconstructed image was convolved with an isotropic Gaussian kernel of radius 4 pixels to suppress residual high-frequency noise. To further improve effective resolution and contrast, sparse deconvolution was performed using a Landweber iterative scheme, which enforces continuity and sparsity priors to further improve effective resolution and contrast (Zhao et al., 2022). For the spot-like mitochondrial signals analyzed here, the parameters were as follows: effective numerical aperture = 3.0; image fidelity = 300; axial continuity = 0.1; sparsity weight = 15; sparse iterations = 100; no background term; no upsampling; and 10 Landweber iterations. These settings correspond to the ‘Spot-shaped sample/Sparse Level L3’ preset in MicroscopeX FINER (Guangzhou CSR Biotech Co. Ltd). The corresponding software-specific presets used in the his-sim imager and microscopex finer packages are listed in Table S1. The line profile of fluorescence intensity for colocalization analysis was conducted by applying the ‘Plot profile’ function under the ‘Analyze’ menu from Fiji (Schindelin et al., 2012). Gray values across the line of interest were used for plotting. Apparent peak-to-peak separation was defined as the distance between the neighboring fluorescence peaks of two mitochondria subcompartments marker proteins. Transmission electron microscopy (TEM) sample preparation and observation were performed according to a previously described method with minor modifications (Xie et al., 2019). Seven-day-old Arabidopsis seedlings were used for TEM analysis of mitochondrial ultrastructure in root maturation zone cells. Root segments were fixed in 2.5% glutaraldehyde for 4 h, followed by postfixation with 1% osmium tetroxide in 0.1 M phosphate buffer (pH 7.0) for 2 h at room temperature in the dark. Samples were dehydrated through a graded ethanol series and acetone, embedded in Epon 812 resin, and polymerized at 60°C. Ultrathin sections of c. 70 nm were prepared using an ultramicrotome, mounted on copper grids, and double-stained with uranyl acetate and lead citrate. Sections were examined using a transmission electron microscope at an accelerating voltage of 80 kV, and representative mitochondrial ultrastructural images were We Ronghui Pan of and for the and transgenic lines. We CSR Biotech Co. Ltd, for in HIS-SIM analysis and This is by the of to to and the for the to and and the and performed the and analyzed the and the with contributions from all All in this study are in this or in and Table S1. Raw images of this study have been on Fig. of plant mitochondrial Fig. of optical resolution using Table for imaging and reconstruction of mitochondria subcompartment marker proteins in Arabidopsis roots. is not for the or of any Supporting Information supplied by the than should be to the The is not for the or of any information supplied by the than should be to the corresponding for the The remains with to in and in any