【Nat.Methods】400微米活体超分辨!多模态自适应光学显微镜(MOSAIC系统)实现从分子到器官深层活体成像
文章标题:A multimodal adaptive optical microscope for in vivo imaging from molecules to organisms
通讯作者:Eric Betzig, Wesley R. Legant, Srigokul Upadhyayula
文章链接:doi.org/10.1038/s41592-026-03066-1
文章概要
引言
传统光学显微镜在优化特定成像任务时,往往受限于固有的光学与样本处理权衡,缺乏多功能性,且多细胞环境中的样本诱导光学校差会导致成像性能严重下降。为了克服这一瓶颈,研究团队开发了多模态自适应光学显微镜(MOSAIC)。该系统在紧凑的空间内,高度复用了同一套硬件与软件资源,将晶格光片、无标记斜照明、超分辨率和双光子成像等多种尖端技术无缝集成。通过引入直接波前传感的自适应光学技术,MOSAIC能够动态矫正组织深处的复杂像差,在大幅降低设备成本与空间占用的同时,首次实现了在同一标本内跨越空间与时间尺度的多模态关联研究。

Fig. 1: Large field-of-view dynamic imaging with 3D lattice light-sheet or 2D label-free oblique illumination microscopy.
a, LLSM image from Supplementary Video 2 of LLC-PK1 cells expressing Calnexin–mEmerald (ER) and H2B–mCherry (nuclei). Maximum-intensity-projections (MIPs) show xy (top) and xz (bottom) views of the 1,000 × 750 × 10 µm3 volume. Scale bar, 50 µm. b, Zoomed-in views of cell division events from a. Magenta panel (left), volume rendering of a metaphase cell having a large ER protrusion; blue panel (middle), four points in nominal cell division from metaphase to telophase; yellow panel (right), six points during a rare tripolar mitotic event. Scale bar, 10 µm. c, Label-free OI imaging at 1 Hz (Supplementary Video 3) captures HeLa cell lamellipodial ruffling and replication of contaminating bacteria. Scale bar, 10 µm. Right, Magnified view (magenta box) of three bacterial division events (arrows). Scale bar, 2 µm. d, Tiled label-free imaging over a 1,218 × 975 µm2 field of live U2OS cells at 1 Hz. Scale bar, 200 µm. Right, zoomed-in view (white box) of a single dividing U2OS cell at four points, showing condensed chromosomes (arrows). Scale bar, 10 µm (right).
主要实验及结论
研究团队通过多场景生物实验验证了MOSAIC的卓越性能。在细胞层面,系统以高时间分辨率连续24小时无创获取了49TB的三维数据,捕捉到罕见的三分裂有丝分裂细节;利用超分辨率光片成像模式将横向分辨率拓展至180纳米,动态观测到内质网小管的伸缩与线粒体裂变。在病理组织中,系统仅耗时8.3小时便完成了4倍膨胀阿尔茨海默病海马体组织的37TB高精度数据采集,精准揭示了髓鞘气球样变等病理特征。

Fig. 2: Multimodal super-resolution imaging of subcellular dynamics.
a, Nuclear imaging (SPY 505) and single-particle tracking with LLSM, visualizing molecular trajectories and diffusion dynamics of SOX2 (HaloTag-PA-JF646) in mouse embryonic stem (mES) cells at 50 Hz. Scale bar (left), 5 µm. Scale bar (top right), 2 µm. The probability density histogram of log10 diffusion coefficients is binned from −3 to +3 in 0.1-unit increments. b, Top: LLS-SIM xy MIP of hTERT-RPE1 cells expressing ER (StayGold-ER, gray) and Golgi (β4Gal-HaloTag9 labeled with JFX549, orange) markers, from Supplementary Video 6. Insets, corresponding xy and yz OTFs. Scale bar, 25 µm. Bottom, volume renderings from boxed regions above showing ER and Golgi organization. Scale bar, 2 µm. c, Comparison of widefield and 3D-SIM imaging in hTERT-RPE1 cells with mitochondria (COX8a-StayGold, cyan) and Golgi marker (orange). Scale bar, 25 µm. Insets, 3D-SIM OTFs. d, Timelapse 3D-SIM (Supplementary Video 7) captures mitochondrial and Golgi dynamics during cell division (white box from c). Scale bar, 5 µm. e, Correlative optical microscopy applying multiple imaging modalities (widefield, 3D-SIM, OI, LLS and LLS-SIM) to a dividing hTERT-RPE1 cell with labeled mitochondria (blue–green) and Golgi (orange), from Supplementary Video 8. Scale bar, 10 µm. f, Timelapse sequence of the correlative imaging shown in e. Scale bar, 10 µm.

Fig. 3: Volumetric imaging with nanoscale resolution.
a, Two-color 3D DNA-PAINT of the mitochondrial marker TOMM20 and the nuclear envelope protein marker Lamin A/C in U2OS cells. Left to right, overview MIP of a 180 × 200 × 17 µm3 FOV (scale bar, 20 µm), zoomed-in 3D rendering of a single cell (boxed region, 28 × 23 × 5.4 µm3), its xz orthoslice (scale bar, 2 µm) and the close-up of nuclear invaginations (boxed region, 4.7 × 4.5 × 5.4 µm3). See also Supplementary Video 9. b, ExLLSM MIP overview of a 2,000 × 2,375 × 98 µm3 human hippocampal tissue section from a patient with AD after 4× expansion. NF-200 (blue) and MBP (yellow) label neurofilaments and myelin sheaths, respectively. Scale bar, 500 µm. c, Zoomed view from b showing neurofilament and myelin sheath ballooning. Scale bar, 20 µm. d, Nanoscale structure of axon and myelin sheath blebs. Scale bar, 2 µm. e, Zoomed-in region from b highlighting clustering of NF-200 protein. Scale bar, 20 µm. f, Nanoscale structure of an individual NF-200 cluster. Scale bar, 2 µm. Scale bars throughout represent pre-expansion dimensions.

Fig. 4: Observing cellular and subcellular dynamics within zebrafish embryos.
a, Comparison of LLSM imaging in a zebrafish xenograft system (55 × 183 × 50 µm3) showing actin-labeled MDA-MB-231 human breast cancer cells (magenta) within the zebrafish vasculature (green), both without AO (left) and with AO correction plus deconvolution (right). Insets, corresponding Fourier spectra (at gamma = 0.3) of the magenta channel. Scale bar, 10 µm. b, Timelapse imaging of a capturing cancer cell dynamics and vascular damage during extravasation. Scale bar, 20 µm. c, Zebrafish tail fin volume (216 × 272 × 37 µm3) at 66 h post-amputation showing plasma membranes and nuclear histones. Scale bar, 50 µm. d, Cellular and subcellular events during the initial stages of regeneration after amputation (Supplementary Video 11), showing: extracellular vesicle release from a cell adjacent to the cut site (yellow box); anchoring fibril dynamics in the epidermal basement membrane (red box); a mesenchymal cell fusion event (gray box); and a transiently trapped red blood cell during remodeling of the caudal vascular plexus (green box). Scale bar, 5 µm. e, Visualization of cell cycle state across the fin, based on the cytoplasmic-to-nuclear fluorescence ratio of CDK biosensor DNA Helicase B (DHB), in segmented and computationally separated cells (216 × 173 × 37 µm3 pre-separation, inset). Peripheral cells have the highest fraction in G2 (Supplementary Video 12).
在活体多细胞生物成像中,自适应光学技术展现出核心优势。通过对斑马鱼胚胎进行多区域拼接校正,系统完美恢复了空间频率,清晰记录了人类乳腺癌细胞在血管中的游走、外渗及血管损伤。在活体小鼠大脑功能显微成像实验中,系统成功对抗了颅窗和脑组织像差,在皮层400微米深处清晰解析出单个树突棘结构,并将钙信号信噪比显著提升,捕获到比未校正状态多出2.5倍的树突棘钙瞬变事件,全面证实了其活体研究价值。

Fig. 5: Super-resolution imaging in vivo.
a, MIPs of raw zebrafish membrane images with LLS-SIM illumination without (top) and with AO (bottom). Ticks show the expected positions of the LLS pattern excitation maxima. Scale bar, 5 µm. b, Orthoslices (180-nm thick) in xy, xz and yz from an AO-LLS-SIM reconstruction (61 × 57 × 40 µm3) in the eye of a 14 hpf zebrafish embryo expressing mitochondrial (magenta) and plasma membrane (cyan) markers. Scale bar, 10 µm. Top left inset, Fourier spectrum of reconstructed mitochondria (Supplementary Video 14). c, AO-LLS-SIM MIPs of orthogonal slabs (3-µm thick) in the hindbrain from a 56 × 56 × 40 µm3 volume (inset) in a 14 hpf zebrafish with mitochondria in RGB colors and membrane in gray. Mitochondria in each cell are color-coded by the ratio of total mitochondrial length to cell volume. Scale bar, 10 µm. d, Cutaway view of segmented mitochondria in four different segmented cells. Scale bar, 4 µm. e, Cutaway view of one cell from the volume, showing mitochondrial rearrangements to the daughter cells during division. Scale bar, 4 µm. f, Top, ISM MIP views before and after AO correction of a 0.324 µm xy orthogonal slab within a larger 336 × 319 × 84 µm3 image volume spanning brain, muscle and notochord in a membrane-labeled 7 dpf zebrafish. Scale bar, 50 µm. Bottom, zoomed-in views comparing muscle and neural progenitor cells imaged by ISM with and without AO (Supplementary Video 16). g, AO-ISM MIP (354 × 332 × 16.3 µm3) of a dorsal-mounted, membrane-labeled, 7 dpf zebrafish. Scale bar, 50 µm. h, Volume rendering with xy and yz orthoslices through a neuromast. Scale bar, 10 µm. i, Timelapse AO-ISM showing xy orthoslices of a migrating cell in the neuromast. Scale bar, 10 µm.
总结及展望
尽管MOSAIC显微镜展现出了前所未有的成像能力,但它依然是一台极具操作门槛的高复杂度仪器。为了应对高分辨率和超快成像带来的海量数据挑战(每小时产出高达4TB数据),研究团队正在全力开发多模态机器学习四维基础模型。展望未来,研究团队提出了建立“细胞观测台”的宏伟愿景。这类集显微成像、高性能计算和专家团队于一体的集约化创新中心,将通过托管共享数据集与预训练AI模型,为全球生物学家提供即开即用的高价值数据分析产出,从而彻底攻克大规模活体成像的操作与数据瓶颈,解锁细胞生物学的核心运行规律。