Meiqi Li

Stomatin is a member of the SPFH (Stomatin, Prohibitin, Flotillin, and HflK/C) protein family, and has been proposed to scaffold functional membrane microdomains (FMMs) on the plasma membrane. Functionally, stomatin has been reported to associate with diverse partner proteins and is implicated in various cellular processes. However, the mechanisms by which it organizes FMMs and exerts its regulatory roles remain unclear. Here, we characterize cryo-EM structures of the stomatin complex, revealing a hexadecameric membrane-anchored assembly that resembles an inverted cup. Sixteen SPFH1 domains insert into the cytosolic leaflet, defining an isolated circular membrane microdomain. Proteomic analyses reveal numerous proteins associated with stomatin, including solute carrier transporters, Rab GTPases and integrins. We further show that stomatin oligomerization is required for integrin-regulated directional cell migration. Together, these findings provide a foundation for understanding the basic role of stomatin in FMM organization and its versatile functions in various physiological and pathological processes.

The dual crises of electronic waste proliferation and inequitable access to scientific instrumentation demand transformative solutions. Here we present a low-cost, modular spectrometer constructed from repurposed DVDs (as diffraction gratings), discarded webcams (as detectors), and Lego® bricks (as reconfigurable optomechanical frames), which achieves performance comparable to commercial units (resolution Δλ ~1.24 nm; R ² > 0.99 for calibrated emission lines) while reducing production costs by 99%. The system’s design leverages circular economy principles, converting e-waste into a precision tool capable of quantifying environmental contaminants. The design also supports hands-on spectroscopy education, offering a scalable model for sustainable STEM pedagogy. This work demonstrates a practical pathway to reduce e-waste and promote equitable science education through accessible, high-performance instrumentation.

Deep learning has demonstrated remarkable success in augmenting fluorescence imaging under photon-limited conditions. However, existing restoration networks are typically devised for training with augmented patches far smaller than the full-view raw data, an overlooked aspect that compromises fidelity and noise-resistance due to the loss of global statistics. To address this limitation, we propose a large-patch network (LargePNet), which synergizes the large effective receptive field provided by shallow ultra-large-kernel convolutions and the nonlinear representation capabilities of deep networks through scale separation. It effectively and efficiently leverages large-view global information for restoration. Directly trained with large-view images, LargePNet shows contrasting advantages over state-of-the-art small-patch networks, with 0.5-2 dB higher peak signal-to-noise ratio across eight representative restoration tasks, involving implementations for single-image, video, and volumetric fluorescence data. For full-view processing, LargePNet generally holds around 4-fold and 20-fold higher computational efficiency compared to advanced convolution-based and Transformer-based networks, respectively. The assistance of LargePNet helps achieve 30-hour-long fluorescence imaging to monitor cytoskeleton dynamics, and hour-long tri-color super-resolution imaging to investigate organelle interaction, showcasing its advancement in live-cell imaging.

The crowded intracellular milieu shapes organelle architecture and dynamics, yet nanoscale heterogeneity in its physicochemical properties remains difficult to visualize with conventional fluorescence anisotropy (FA) imaging. Here, we develop fluorescence anisotropy structured illumination microscopy (FA-SIM), which employs orthogonal-polarization structured illumination with dual-angle detection to achieve ∼100-nm resolution and quantitative FA retrieval with 0.56% relative error, representing over 20-fold higher accuracy than conventional FA imaging. With low phototoxicity, FA-SIM enables dual-color, hour-long quantitative super-resolution imaging in cells. Using viscosity standards, defined nanoparticles and small-molecule drug binding assays, we validate FA-SIM as a quantitative reporter of rotational mobility and molecular interactions. In cells, FA-SIM resolves nanoscale crowding heterogeneity, correlates anisotropy landscapes with condensate dynamics, and uncovers a radial crowding gradient across the microtubule network and mitotic spindle. Long-term dual-color imaging further resolves coordinated actin–microtubule remodeling and associated microenvironmental changes. By enabling quantitative, super-resolved mapping of intracellular physical properties in living systems, FA-SIM provides a powerful platform for investigating the physical regulation of cellular organization and dynamics in health and disease.

Super-resolution imaging has revolutionized our ability to visualize biological structures at subcellular scales. However, deep-tissue super-resolution imaging remains constrained by background interference, which leads to limited depth penetration and compromised imaging fidelity. To overcome these challenges, we propose a novel imaging system, confocal² spinning-disk image scanning microscopy (C²SD-ISM). It integrates a spinning-disk (SD) confocal microscope, which physically eliminates out-of-focus signals, forming the first confocal level. A digital micromirror device (DMD) is employed for sparse multifocal illumination, combined with a dynamic pinhole array pixel reassignment (DPA-PR) algorithm for ISM super-resolution reconstruction, forming the second confocal level. The dual confocal configuration enhances system resolution, while effectively mitigating scattering background interference. Compared to computational out-of-focus signal removal, SD preserves the original intensity distribution as the penetration depth increases, achieving an imaging depth of up to 180 μm. Additionally, the DPA-PR algorithm effectively corrects Stokes shifts, optical aberrations, and other non-ideal conditions, achieving a lateral resolution of 144 nm and an axial resolution of 351 nm, and a linear correlation of up to 92% between the original confocal and the reconstructed image, thereby enabling high-fidelity super-resolution imaging. Moreover, the system’s programmable illumination via the DMD allows for seamless realization with structured illumination microscopy modality, offering excellent scalability and ease of use. Altogether, these capabilities make the C²SD-ISM system a versatile tool, advancing cellular imaging and tissue-scale exploration for modern bioimaging needs.

In fluorescence microscopy, a persistent challenge is the defocused background that obscures cellular details and introduces artifacts. Here, we introduce Dark sectioning, a method inspired by natural image dehazing for removing backgrounds that leverages dark channel prior and dual frequency separation to provide single-frame optical sectioning. Unlike denoising or deconvolution, Dark sectioning specifically targets and removes out-of-focus backgrounds, stably improving the signal-to-background ratio by nearly 10 dB and structural similarity index measure of images by approximately tenfold. Dark sectioning was validated using wide-field, confocal, two/three-dimensional structured illumination and one/two-photon microscopy with high-fidelity reconstruction. We further demonstrate its potential to improve the segmentation accuracy in deep tissues, resulting in better recognition of neurons in the mouse brain and accurate assessment of nuclei in prostate lesions or mouse brain sections. Dark sectioning is compatible with many other microscopy modalities, including light-sheet and light-field microscopy, as well as processing algorithms, including deconvolution and super-resolution optical fluctuation imaging.

Studying organelles’ interactome at system level requires simultaneous observation of subcellular compartments and tracking their dynamics. Conventional multicolor approaches rely on specific fluorescence labeling, where the number of resolvable colors is far less than the types of organelles. Here, we use a lipid-specific dye to stain all the membrane-associated organelles and spinning-disk microscopes with an extended resolution of ~143 nm for high spatiotemporal acquisition. Due to the chromatic polarity sensitivity, high-resolution ratiometric images well reflect the heterogeneity of organelles. With deep convolutional neuronal networks, we successfully segmented up to 15 subcellular structures using one laser excitation. We further show that transfer learning can predict both 3D and 2D datasets from different microscopes, different cell types, and even complex systems of living tissues. We succeeded in resolving the 3D anatomic structure of live cells at different mitotic phases and tracking the fast dynamic interactions among six intracellular compartments with high robustness.

Fluorescence microscopy is indispensable in biological and medical research for its ability to provide detailed visualization of tissue and cellular structures. However, wide-field fluorescence microscopy often suffers from out-of-focus blur and background emission, degrading image clarity and axial resolution. To address this limitation, we propose multiscale convolutional gated UNet (MG-UNet), a novel deep neural network designed to enhance wide-field fluorescence images by learning the high contrast and sharpness features of confocal microscopy. MG-UNet employs multi-scale convolutional gate modules that use a combination of convolutional filters at different scales for coordinate encoding, improving computational efficiency while preserving spatial information. The model is tailored for two-dimensional image restoration and extends to lightweight three-dimensional applications by integrating spatial-channel transformation operators. Rigorous validations on both simulated and experimental data demonstrate that MG-UNet outperforms state-of-the-art models in both 2D and 3D fluorescence microscopy tasks, achieving superior image quality with lower computational costs compared to standard UNet, underscoring its potential for advanced biological imaging.

Mitochondrial membrane environmental dynamics are crucial for understanding function, yet high‐resolution observation remains challenging. Here, HBimmCue is introduced as a fluorescent probe localized to inner mitochondrial membrane (IMM) that reports lipid polarity and membrane order changes, which correlate with cellular respiration levels. Using HBimmCue and fluorescence lifetime imaging microscopy (FLIM), IMM lipid heterogeneity is uncovered across scales, from nanoscale structures within individual mitochondria to mouse pre‐implantation embryos. At the sub‐organelle level, stimulated emission depletion (STED)‐FLIM imaging highlights nanoscale polarity variations within the IMM. At the sub‐cellular and cellular level, reduced IMM lipid polarity is observed in damaged mitochondria marked for lysosomal degradation and distinct IMM lipid distributions are identified in neurons and disease models. Additionally, metabolic dysfunction associated with oocytes aging and metabolic reprogramming from zygote to blastocyst is detected. Together, the work demonstrates the broad applicability of HBimmCue, offering a new paradigm for investigating lipid polarity and respiration level at multiple scales.