【文献解读】随机光学重建显微镜助力线粒体动力学的STORM成像（申请试拍）

智能文字提取功能测试中

随机光学重建显微镜（STORM）可应用于超微线粒体结构的可视化。本文作者开发了一种新的策略来捕获线粒体动态使用压缩传感STORM算法以下的原始数据预处理的噪声校正的主成分分析和K因子图像分解。使用STORM显微镜与邻近二硫醇蛋白质靶向探针，可视化线粒体动力学是可实现的空间和时间分辨率为45 nm和0.8 s，值得注意的是，动态线粒体微管收缩〜 746 nm在1.2 s的监测。标记的缀合物被观察为分布在线粒体外膜上的簇（半径，~90 nm），据我们所知尚未报道。这种策略是有前途的定量分析细胞内的行为低于光学衍射极限。线粒体的特点是高度互联的网络和动态变化。线粒体网络的形成和维持由单个线粒体的融合和微管之间的连接介导。融合-分裂平衡的动态变化和线粒体管状网络的形成决定了线粒体的整体形态和生物学行为。此外，在一个被称为动态微管化的过程中，小管伸展以与其他线粒体的膜融合。这些动态事件促进正常线粒体功能的维持，例如线粒体物质的交换、DNA完整性、呼吸能力、对细胞应激的响应和细胞凋亡。然而，线粒体动力学的破坏导致异常发育和人类疾病，包括神经退行性疾病和癌症。因此，线粒体动力学的调查是至关重要的，使更深入地了解与线粒体功能障碍有关的疾病。荧光显微镜被广泛用于研究活细胞中的单个线粒体。然而，使用常规光学显微镜成像线粒体亚结构受到光学衍射极限的限制。由于超分辨率成像的出现，使超细结构可视化已成为可能。在这项工作中，研究者进行了STORM成像的活细胞中的线粒体动力学，并进行了定量分析的目标下STORM超分辨率显微镜。一种新的荧光标记适合于STORM成像的开发选择性地标记VDPs的外线粒体膜上的简单的共培养与细胞。同时，研究者采用了一个温和的照明激光功率，以获得尽可能多的单帧发射极定位，并减少光漂白和光毒性。STORM成像的时间分辨率可以通过降低激光功率和提高CCD帧频来实现。为了避免背景噪声的干扰，噪声校正的主成分分析（NC-PCA）和K因子图像分解的组合被开发来处理原始数据之前，使用CSSTORM算法重建。使用这种策略，研究者实现了线粒体动力学的STORM成像，包括融合，分裂和线粒体微管化。此外，在STORM成像下，通过层次聚类分析定量评估线粒体外膜上的VDP分布。Cy 5荧光染料是STORM显微镜的理想的光闪烁候选者，并且1，3，2-二硫砷杂环戊烷是用于VDPs共价结合的有用单元。哌嗪间隔基用作连接基（图1A）。根据先前描述的方案16制备荧光探针1。在将其应用于细胞成像之前，研究了1的光谱性质。首先，乙醇中的探针1在649 nm处显示最大吸收，摩尔消光系数为2。1 ± 0.3 × 105和670 nm处的发射最大值（图1B）。还在水溶液中检测了相同的参数，获得了相似的结果。与罗丹明B相比，在乙醇中获得0.28的荧光量子产率，在水溶液中获得0.18的荧光量子产率。此外，使用自制的dSTORM系统研究了1的单个分子的光闪烁行为。在不存在成像缓冲液和硫醇7，17的情况下，将1的样品直接用激光束（656 nm，0.8kW·cm 2）照射，以显示具有高光子数（约100 nm）的显著的光闪烁荧光。3000个光子/像素）和低的开关占空比（约0.001）。0.0013）。在合适的开关频率下，光闪烁的持续时间持续超过900秒，而“开启”状态的平均光闪烁周期约为88.2ms（图1C-F）。图1. 目标化合物1在没有成像缓冲液的情况下的化学结构和光学性质(A)目标化合物1的化学结构;（B）1在乙醇溶液中的吸收（黑色）和发射（红色）光谱;（C）在不存在含有β-巯基乙醇（BME）和葡萄糖氧化酶（GLOX）的STORM成像缓冲液的情况下，在656 nm下用0.8kW.cm-2的激光功率激发，在数十秒内可逆和自发荧光闪烁;（D）探针1的单分子荧光的时间轨迹;（E）开-关占空比值被计算为使用100秒的滑动窗口在许多分子上平均的在开状态中花费的时间的分数，平均占空比在400-600秒之间测量，由黄色框指示;（F）探针1的荧光开启时间的分布，从直方图的单指数拟合（红色曲线）导出88.2ms的指示平均持续时间。为了进一步验证在CSSTORM定位之前通过NC-PCA+ K因子处理的去噪，我们将通过CSSTORM获得的重建结果与单分子拟合（insight 3软件）算法的重建结果进行了比较，如图2所示，使用具有从公开可用的EPFL网站18获得的高密度标记STORM数据的微管蛋白。首先使用NC-PCA算法对数据集进行去噪，然后使用K因子算法进行锐化。通过CSSTORM（图2A（a-c）和单分子拟合（图2A（d））获得了微管蛋白的类似结构，并且CSSTORM仅使用500个相机帧，甚至少至200个帧就表现出更大的结构连续性;然而，在CSSTORM（图2A（例如，i-k，m-o））和insight 3算法（图2A（h，l，p））之间的比较中，当使用100帧时观察到差的连续性。图2A（a-d）中的放大视图示出了在使用NC-PCA+ K因子算法去噪时明显锐化的结构（图2A（g）），其保持了与insight 3结果的最大结构一致性（图2A（h））。在NC-PCA+ K因子处理后，七个微管是可区分的，并且在原始数据的CSSTORM处理中或仅对于K因子处理，在相邻微管之间不可避免地出现模糊的噪声峰，如图2B（a-b）中的虚线圆所示。傅立叶环相关（FRC）是一种定量成像分辨率测量工具，已用于进一步评估图像分辨率的改善。图2. 使用具有高密度标记的微管蛋白的实验数据集（来自EPFL）验证NC-PCA去噪算法(A)原始数据（第1列）、K因子（第2列）和NC-PCA+ K因子预处理（第3列）的CSSTORM结果和单分子拟合结果（第4列）的比较。行24，行1的放大视图，分别使用500、200和100帧重建。比例尺，（d）中为2 μm，（h、l、p）中为500 nm;（B）沿着（e-g）中所示的黄色虚线的分布图。在用PBS、PBS-乙醇或乙醇溶剂洗涤后，探针不能从线粒体中除去（图S2 B、S2 C、S2 D、S5和S6）。此外，探针在我们的实验条件下未显示细胞毒性（图S7），这有利于活细胞超分辨率成像（图3）。图3. 超分辨率成像探针1的线粒体膜贴上住海拉细胞固定海拉细胞(A)和(B)(A)广角荧光图像,和STORM图像(B和c)通过CSSTORM de-noised CSSTORM算法;线粒体(d, e)放大的图像单面板c CSSTORM和de-noised CSSTORM,分别;(f, g在B和e)外膜的宽度的分析;蓝色柱状图显示了沿着线路横断面(d, e)在面板和面板B (d);(B)广角成像(B);风暴(a、c、d)使用200帧图像通过CSSTORM, EMCCD在250赫兹的频率为64×128像素的原始数据采集。为了确定1在活细胞线粒体中的确切位置，进行STORM超分辨率成像以确定由1标记的线粒体的超细结构。将探针1（1 μM）与HeLa细胞一起孵育1小时，无需进一步处理，重要的是，该程序比使用Alexa647的程序更简单。随后，在成像之前用戊二醛固定细胞样品或通过STORM系统用于活细胞成像。如图3A（a）所示，与不能提供线粒体的精细结构的宽视场图像不同，超分辨率图像显示了详细结构，其在活HeLa细胞（图3A（c））或固定细胞（图3B）中使用200帧通过CSSTORM（图3A（b））和去噪CSSTORM重建。从图3A（c）中的虚线方块所示的感兴趣区域的放大图像，去噪的CSSTORM分辨率明显高于CSSTORM的分辨率（图3A（d）），并且在图3A（e）中清楚地观察到线粒体膜上的离散颗粒分布，表明探针1选择性地标记线粒体膜上的VDP。通过高斯拟合计算CSSTORM和去噪CSSTORM的成像分辨率，示出108.7和45的半高全宽（FWHM）。如图3A（f，g）中所示。使用固定的HeLa细胞，STORM进一步支持1与线粒体膜上的VDPs的共价结合。当HeLa细胞与1孵育，用戊二醛固定并用PBS洗涤时，通过去噪CSSTORM算法获得线粒体的类似超分辨率图像（图3B（a，c，d）），并且高成像分辨率（图3B（e）中的30.2nm）使得能够研究线粒体外膜上的VDP分布以及它们的动力学。STORM进一步应用于评估动态线粒体变化，包括线粒体融合、分裂和微管化。通过CSSTORM或去噪CSSTORM算法使用200个相机帧以20帧的步长进行重建以获得0.8s的时间分辨率来获得演示动态的视频。视频结果清楚地显示，与宽视场图像（图4A）和CSSTORM图像（图4 B）相比，去噪CSSTORM图像的分辨率明显更好。在通过NC-PCA+ K因子方法预处理后，去噪CSSTORM分析显示出明显的噪声抑制，并且实现了线粒体膜上的VDP的超细成像（图4C、）。此外，可以观察到线粒体的典型动力学以量化线粒体融合、分裂和成管的速度，这可能反映了生物活动中的线粒体状态20、21。捕获单个线粒体微管化和微管拉伸（图4C中的虚线正方形），并且不同的时间点显示在图4D中（1-6）。在管收缩期间的距离为~746 nm，并且在1.2s内完成（图4 E）。如通过高斯拟合计算的，中空管直径（图4F中放大的虚线圆）为51.8nm，并且膜宽度的FWHM值为46.9nm和43.5nm（图4G），与先前报道的结果22相当。图4. 基于超分辨率成像的线粒体微管化分析(A)线粒体宽视野图像;（B，C）通过CSSTORM和去噪CSSTORM算法分析的活HeLa细胞中线粒体的超分辨率图像;（D，E）线粒体微管运动，（D1-D 6）不同颜色代表不同时间点线粒体的运动状态，（E）合并D1-D 6的整个微管收缩过程;（F）（C）中放大的单个线粒体和伸展小管;⑹线粒体小管的宽度分析;蓝色直方图示出了沿着（F）中的虚线圆圈中标记的线的横截面轮廓，其中FWHM值从高斯拟合（红色和绿色）获得。研究VDPs在线粒体膜上的分布可以提供对生物学功能的深入了解。我们使用开源软件Localization Microscopy Analyzer（LAMA）23基于基于密度的分层聚类算法来分析VDP的形态簇，即一个基于密度的算法发现集群在大型空间数据库中的噪声（DBSCAN）和排序点，以确定聚类结构（OPTICS）。考虑到压缩感测算法的像素化而不直接返回分子坐标列表，通过识别非零网格点的簇，首先将CSSTORM结果转换为分子列表作为LAMA的输入。选择感兴趣的区域（128×128像素的超分辨率网格图像）用于蛋白质聚类分析（图5A）。观察半径设定为19 nm，在DBSCAN中最小簇尺寸设定为6，并且在OPTICS 24中噪声水平设定为13.1%。总共，分别通过DBSCAN和OPTICS检测到32个和36个簇，并且在图5（B，E）中示出为簇群的2D分布。使用DBSCAN或OPTICS的结果进行基于DBSCAN的层次聚类分析，以通过多边形形成来提取单个簇的形态信息。具有16个假色的形态聚类分析结果总结在图5（C，F）中。最后，一个集群人口分布与集群大小约。如图5（D、G）所示，获得了90 nm，表明这是分析活细胞中蛋白分布的一种有前景的策略。图5. 使用DBSCAN（B，C，D）和OPTICS（E，F，G）通过蛋白质聚类分析确定线粒体膜VDPs的分布(A)STORM超分辨率图像（比例尺500 nm）;（B，E）聚类总体的二维分布;（C，F）基于DBSCAN的形态聚类分析的视觉表示;（D，G）簇半径的分布。在该方法中，应用新的STORM探针共价结合线粒体外膜中的VDPs，这实现了STORM超分辨率成像，而不需要成像缓冲液和硫醇。使用由NC-PCA+ K因子预处理组成的噪声校正方法通过CSSTORM成像获得线粒体动力学分析。去噪CSSTORM能够实现线粒体的超细结构成像，在固定细胞中的空间分辨率高达30.2 nm，在活细胞中为45.7，时间分辨率为0.8 s。利用线粒体膜上1-VDP复合物的动态变化来显示线粒体动力学，包括融合、分裂和微管化，特别是ca.746 nm（1.2 s）。将线粒体外膜上的VDPs分布进一步分析为蛋白质聚集（约100%）。90 nm），以更深入地了解其生物学功能。新开发的策略将是有用的线粒体功能的进一步研究。在本研究中，研究者主要借助STORM超助力线粒体动力学的STORM成像，目前在国内，随机光学重建显微镜STORM已成功实现商用，有需要STORM成像技术进行实验研究的专家老师们，请文末填写问卷，即可预约获得 iSTORM 超高分辨率显微成像系统试拍服务~力显智能现已发布的超高分辨率显微成像系统 iSTORM，成功实现了光学显微镜对衍射极限的突破，使得在20纳米的分辨率尺度上从事生物大分子的单分子定位与计数、亚细胞及大分子复合物结构解析、生物大分子生物动力学等的研究成为现实，从而给生命科学、医学等领域带来重大突破。STORM Imaging of Mitochondrial Dynamics using a Vicinal-Dithiol-Proteins-Targeted Probe Bingling Chen#a， Wanjun Gong# a， Zhigang Yang a ， Wenhui Pan， Peter Verwilst， Jinwoo Shin，Wei Yan， Liwei Liu ， Junle Qu a and Jong Seung Kim*，a .b “ Key Laboratory of Optoelectronic Devices and Systems of Ministry of Education and Guangdong Province， College of Physics and Optoelectronic Engineering， Shenzhen University， Shenzhen 518060，China Chemistry Department， Korea University， Seoul 02841， Korea # These author s contributed equally to this work * Corresponding emails： zhgyang@szu.edu.cn； j lqu@szu.edu.cn； jongskim@ rea.ac.kr Abstract： Stochastic optical reconstruction microscopy (STORM) i s a promising method for the visualization of ultra-fine mitochondrial structures. However， this approach is l imited to monitoring dynamic i ntracellular events owing to its low temporal resolution. We developed a new strategy to capture mitochondria l dynamics using a compressed sensing STORM algorithm following raw data pre-treatments by a noise-corrected principal component analysis and K-factor image factorization.Using gS STORM microscopy with a vicinal-dithiol-proteins targetingprobe，，V visualizing mitochondrial dynamics was attainable with spatial and temporal resolutions of 45 nm and 0.8 s，notably， dynamic mitochondrial tubulation retraction of ~746 nm in 1.2 s was monitored. The labeled conjugate was observed as clusters (radii，~90 nm) distributed on the outer mitochondrial membranes， not yet reported as far as we know. This strategy is promising for the quantitative analysis of intracellular behaviors below the optical diffraction l imit. Keywords： STORM； Mitochondrial Imaging； VDP Target i ng； Cyanine； Live cell imaging Indroduction Mitochondria are characterized by highly interconnected networks and dynamic changes. The formation and maintenance of mitochondrial networks are mediated by the fusion of individual mitochondria and junctions between tubules. Dynamic changes in the fusion-fission balance and mitochondrial tubular network formation determine the e overall morphology y and biological behaviors SO of mitochondria.Furthermore， in a process referred to as dynamic tubulation， tubules stretch to fuse with the membranes of other mitochondria. These dynamicc events facilitate the maintenance of [normalmitochondrial functions， such as the einterchange of mitochondria l materials， DNA integrity， respiratory capacity， response to cellular stress，and apoptosis. However， disruptions in mitochondrial dynamics lead to abnormal development and human diseases， including neurodegenerative disorders and cancers.Therefore， investigations of mitochondrial dynamics are critical to enable a deeper understanding of diseases related to mitochondrial dysfunctions. Fluorescence microscopy is widely used to investigate single mitochondria in live cells°.However， i maging mitochondrial substructures using conventional optical microscopy is limited by the optical diffraction limit. Since the emergence of super-resolution imaging， i t has become feasible to visualize ultra-fine structures’. Stochastic optical reconstruction microscopy (STORM) has an outstanding imaging ability for biological objects below 200 nm in size. Fluorescent probes composed of fluorophores (e.g.，Alexa647) and antibodies are favored for STORM imaging in fixed cells’，10. However，live cel l imaging by STORM remains challenging owing to the lack of appropriate probes and low temporal resolution. In particular， direct STORM has rarely been used to evaluate mitochondrial dynamics. Smart fluorescent probes that selectively target mitochondrial membrane-bound proteins would facilitate the imaging of mitochondrial membrane structures using STORM. Mitochondria possess abundant vicinal dithiol proteins (VDPs) with two proximal cysteine residues in their primary structures； they play essential roles in maintaining mitochondrial redox homeostasis， antioxidant defense， and redox signaling.Fluorescent arsenicals contain a 1， 3，2-dithiarsenolane moiety capable of selectively binding to VDPs by forming covalent bonds. They can label mitochondrial VDPs i n situ， providing a basis for the application of STORM to mitochondria in l ive cells.However， STORM i nvolves sparse lighting and dilute label i ng with fluorescent probes during sample preparation， and thousands of frames are needed to reconstruct a super-resolution image ， resulting in poor temporal resolution13，14. Temporal resolution can be improved bby reducing the number of f rames. Compressed sensing STORM (CSSTORM) is an li -norm minimization-based multi-molecule localization algorithm and is more suitable for dense fluorescence than single-molecule fitting. It can substantially shorten t he time required for signal acquisition by cutting the number of required frames， and the temporal resolution can be f urther improved by increasing the acquisition f rame rate of the electron multiplying charge coupled device (EMCCD)1.However， the rapid signal acquisition will increase background noise， thereby increasing interference in image reconstruction. For the application of STORM to l ive cell mitochondrial dynamics， the development of smart probes targeting mitochondrial membrane VDPs with an improved reconstruction algorithm for noise suppression is a promising strategy. In this work， we performed STORM imaging of mitochondrial dynamics in l ive cells，and carried out the quantitative analysis of the t arget under STORM super-resolution microscopy. A new fluorescent marker suitable for STORM i maging was developed to selectively label VDPs on the outer mitochondrial membrane by simple co-culturing with cells. Meanwhile， we employed a moderate i lluminating laser power to obtain as much single-frame emitter localiztion as possible and to reduce photobleaching and phototoxicity. The temporal resolution of STORM imaging can be achieved by reducing the laser power and increasing the CCD frame rate. To avoid interference from background noises， a combination of noise-corrected principal component analysis (NC-PCA) and K-factor image factorization were developed to treat the raw data prior to reconstruction using the CSSTORM algorithm. The principle and implementation of the algorithms were further described in the Supplemental Information (Figure S14-S17). Using this strategy， we achieved STORM imaging of mitochondrial dynamics including fusion， fission， and mitochondrial tubulation. Furthermore， the VDP distribution on the outer mitochondrial membrane was quant i tatively evaluated by hierarchical clustering analysis under STORM imaging. Results and Discussion Cy5 fluorescent dye is an ideal photoblinking candidate for STORM microscopy and 1，3， 2-dithiarsenolane i s a useful unit for the covalent binding of VDPs. A piperazine spacer was used as the l inker (Figure 1A). Fluorescent probe 1 was prepared according to previously described protocols6(see Supporting Inforamtion for details). Prior to i ts application in cell i maging， the spectral properties of 1 were studied. First， probe 1 in ethanol showed absorption maxima at 649 nm with a molar extinction coefficient of 2.1±0.3×105 and emission maxima at 670 nm， respectively (Figure 1B). The same parameters have also been tested in aqueous solution and similar results were obtained (Table S1 in Supplemental Information). A fluorescence quantum yield of 0.28 in ethanol and 0.18 in aqueous solution was obtained compared to Rhodamine B.Additionally， the photoblinking behavior of a single molecule of 1 was i nvestigated using the home-built dSTORM system. In the absence of an imaging buffer and thiols’， a specimen of 1 was directly illuminated by a laser beam (656 nm，0.8kW·cm ) to show marked photoblinking fluorescence with a high photon number (ca. 3000photons per pixel at each frame of the fluorescent state i n the “on”cycle) and a low on-off duty cycle (ca. 0.0013). The mechanism of photoblinking under l aser i llumination was further discussed in the Supplemental Information (Supplementary Scheme 2). The duration of photoblinking lasts for more than 900 seconds with a suitable switching frequency， while the average photoblinking cycle of“on”state was approximately 88.2ms (Figure 1C-F and Figure S1). To evaluate the analytical results of the CSSTORM reconstruction after NC-PCA de-noising and K-factor sharpening， a fluorescent probe was placed on a coverslip to produce photoblinking under laser i llumination at 656 nm. The blinking dataset was analyzed by NC-PCA using different major principal components， as shown in Figure S17. Comparisons of 1-3 to 1-6 major principal components showed that the fewer principal components chosen for signal reconstruction， the more evident the noise reduction and the higher the potential for signal distortion. 1-6 major principal components were empirically employed to balance noise reduction and the fidelity of the signal in the NC-PCA de-noising process for super-resolution image reconstruction，based on the validation experiments of the NC-PCA algorithm by using simulation data (Figure S15-S16). When implementing the K-factor algorithm， the parameters M = 54(number of factors) and h=9 (number of harmonics) were set manually， and the k value was set automatically， gradually reducing according to the number of factors. The detailed workflow is presented in Figure S14. 2 0 100 200 300 400 500 600 700 800 900 Time (s) Time(s) Fluorescence-on time (ms) Figure 1. Chemical structures and optical properties of target compound 1 without imaging buffer. (A) Chemical structure of target compound 1； (B) Absorption (black)and emission (red) spectra of 1 in ethanol solution； (C) Reversible and spontaneous f l uorescence bl i nking over tens of seconds i n the absence of a STORM i maging buffer containing B-mercaptoethanol (BME) and glucose oxidase (GLOX)， excited at 656 nm with a laser power of 0.8 kW.cm；(D) Time traces of single-molecule fluorescence of probe 1； (E) The on-off duty cycle value was calculated as the fraction of time spent i n the on state averaged over many molecules using a sliding window of 100 seconds， the mean duty cycle was measured between 400-600 seconds i ndicated by yellow box；(F)Distribution of fluorescence-on time of probe 1， the indicated mean duration time of 88.2 ms was derived from the single exponential fit (red curve) of the histogram. To further verify de-noising by NC-PCA+K-factor treatment prior to CSSTORM localization， we compared the reconstruction results obtained by CSSTORM with t hose of the single-molecule f itting (insight3 software) algorithm， as shown in Figure 2， using tubulin with high-density labe l ing STORM data obtained from an openly available EPFL website l 8. The dataset was first de-noised using the NC-PCA algorithm and then sharpened using the K-factor algorithm. Similar structures of tubul i ns were obtained by M (Figure 2A (a-c) and single-molecule fitting (Figure 2A ((d))，.and CSSTORM exhibited greater structural cont i nuity using only 500 camera frames， and even as few as 200 frames； however， poor continuity was observed when using 100frames， in a comparison between CSSTORM (Figure 2A (e-g， i -k， m-o)) and the insight3 algorithm (Figure 2A (h， l ，p)). The magnified view in Figure 2A (a-d) shows markedly sharpened structures upon de-noising using the NC-PCA+K-factor algor i thm (Figure 2A (g))， which maintained the most structural consistency with the i nsight3results (Figure 2A (h)). Seven microtubules were distinguishable after NC-PCA+K-factor treatment， and ambiguous noisy peaks inevitably appeared between adjacent microtubules in CSSTORM treatment of raw data or for K-factor treatment only， as shown by the dashed circles in Figure 2B (a-b). Fourier ring correlation (FRC) is a quantitative imaging-resolution measuring tool t hat has been used to further evaluate improvements in image resolution 19. Image resolution (R) was determined by the inverse of the spatial frequency i ndicated by the intersection of the FRC curve with a given threshold， which was empirically set to 1/7 (Figure S18). NC-PCA+K-factor treatment resulted in t he highest imaging resolution of 39 nm， which was comparable to those obtained by CSSTORM for raw data (66 nm) and K-factor only (62 nm). Figure 2. Validation of the NC-PCA de-noising algorithm using an experimental dataset of tubulins with high-density label i ng (from EPFL). (A) Comparison of CSSTORM results for raw data (column 1)， K-factor (column 2)， and NC-PCA+K-factor pre-treatment (column 3) and single-molecule fitting results (column 4). Row 2-4， magnified view of row 1， reconstructed using 500， 200 and 100 frames， respectively.Scale bar， 2 um i n (d) and 500 nm in (h， l ， p)； (B) Profiles along the yellow dashed l ines shown in (e-g). The biological behavior of probe 1， including colocalization， covalent binding ident i fication， and cytotoxicity， was studied. The results demonstrated that probe 1preferentially targeted mitochondria， demonstrated through the opt i mal merging imaging with Mito-tracker over Lyso-， ER-trackers (Figure S2A， S3 and S4). Moreover，mitochondrial VDPs were covalently labeled by the probe， which was identified by SDS-PAGE and washing-out experiments. TThhee pprroobbee ccaannnnot be removed from mitochondria after washing with PBS， PBS-ethanol， or ethanol solvents， (Figure S2B，S2C， S2D， S5 and S6). Furthermore， the probe showed no cellular toxicity under our experimental conditions (Figure S7)， which is favorable for l ive cell super-resolution imaging (Figure 3， S8). Figure 3. Super-resolution i maging of mitochondrial membranes labelled with probe 1in live HeLa cells (A) and in fixed HeLa cells (B). (A) Wide-field fluorescent image (a)， and STORM i mages (b and c) obtained by CSSTORM and de-noised CSSTORM algorithms； (d， e) magnified images of single mitochondria in panel c by CSSTORM and de-noised CSSTORM， respectively； (f ， g in A and e in B) Width analysis of the outer membrane； the blue histogram shows the cross-sectional profile along the l ines in (d， e) in panel A and (d) in panel B； (B) Wide-field imaging (b)； (a， c， d) STORM images obtained by the CSSTORM using 200 frames，EMCCD worked at a frequency of 250 Hz for 64×128 pixels raw data acquisition. To identify the exact location of 1 in mitochondria of live cel l s， STORM super-resolution imaging was performed to determine the ultra-fine structure of mitochondria labeled by 1. Probe 1 (1 pM) was incubated with HeLa cells for 1 h without further treatment and importantly this procedure was simpler than that using Alexa647.Subsequently， the cell samples were fixed with glutaraldehyde before i maging or used for live cell imaging by the STORM system. As shown in Figure 3A (a)， unlike the wide-field i mage， which cannot provide the f ine structure of mitochondria， the super- resolution images showed detailed structures， which were reconstructed using 200frames by CSSTORM (Figure 3A (b)) and de-noised CSSTORM i n live HeLa cells (Figure 3A (c)) or in fixed cell s (Figure 3B). From the magnified image of the region of interest， indicated by dashed square i n Figure 3A (C)， the de-noised CSSTORM resolution was evidently higher than that of CSSTORM (Figure 3A (d))， and the discrete granule distribution on t he mitochondrial membrane was clearly observed in Figure 3A (e)， indicating t hat probe 1 selectively labeled VDPs on the mitochondrial membrane. The imaging resolutions of CSSTORM and de-noised CSSTORM were calculated by Gaussian fitt i ng， showing ful l widths at half maximum (FWHM) of 108.7and 45.7 nm， respectively， as shown in Figure 3A (f， g). The covalent binding of1 to VDPs on mitochondrial membranes was further supported by STORM using fixed HeLa cells. When HeLa cells were incubated with 1， fixed with glutaraldehyde and washed with PBS， similar super-resolution images of mitochondria were obtained by the de-noised CSSTORM algorithm (Figure 3B (a， c， d))， and the high imaging resolution (30.2 nm in Figure 3B (e) enabled the study of the distribution of VDPs on the outer mitochondrial membrane as well as their dynamics (Figure S8). STORM was further applied to evaluate dynamic mitochondrial changes， including mitochondrial fusion， fission， and tubulation. Videos demonstrating the dynamics were obtained by the CSSTORM or de-noised CSSTORM algorithm using 200 camera frames for reconstructions with a step length of 20 frames to obtain a temporal resolution of 0.8 s. The video results clearly revealed a markedly better resolution of the de-noised CSSTORM images compared with t hose of wide-field images (Figure 4A) and CSSTORM images (Figure 4B). After pre-treatment by the NC-PCA+K-factor method， the de-noised CSSTORM analysis demonstrated obvious noise suppression (Video S2)， and the ultra-fine i maging of VDPs on mitochondrial membranes was achieved (Figure 4C， Figure S8， and Video S2-S6). Moreover， typical dynamics of mitochondria can be observed to quantify speeds of mitochondrial fusion， fission， and tubulation， which probably reflects the mitochondrial status in biological activities 2021.Single mitochondrial tubulation and microtubule stretching were captured (dashed square in Figure 4C)， and di f ferent time points are shown in Figure 4D (1-6). The distance during the tubulation retraction was ~746 nm， and f inished in 1.2 s (Figure 4E).The hollow tube diameter (magni f ied dashed circle in Figure 4F)， as calculated by Gaussian fitting， was 51.8 nm， and FWHM values for membrane widths were 46.9 nm and 43.5 nm (Figure 4G)， comparable to previously reported results". Mitochondrial fusion and fission dynamics are shown in Figure S9 and Video S3-S8， which have never been achieved by traditional optical microscopy. Figure 4. Mitochondrial tubulation analysis based on super-resolution imaging. (A)Wide-field image of mitochondria； (B， C) Super-resolution images of mitochondria in live HeLa cells analyzed by CSSTORM and de-noised CSSTORM algorithms； (D，E)Mitochondriall tubulation 1movement t .，((D1-D6) Different colours s ：represent tt t l he movement status of mitochondria at different time points， (E) Whole tubulation retraction process by merging D1-D6；(F) Magni f ied single mitochondria and stretching tubule i n (C)； (G) Width analysis of mitochondrial tubules； blue histogram shows the cross-sectional profile along the l i ne marked in the dashed circle in (F) with FWHM values obtained from Gaussian fitt i ng (red and green). Studies of the distribution of VDPs on mitochondrial membranes labeled by 1 can provide insight into biological functions.We used the oper n source e software LocAlization Microscopy Analyzer (LAMA)23 to analyze morphological clusters of VDPs based on density-based hierarchical clustering algorithms， i.e.， a density-based algorithm for discovering clusters in large spatial databases with noise (DBSCAN)and ordering point t s s to identify the clustering structure (OPTICS). Considering the pixelation of the compressed sensing algorithm without directly returning a list of molecule coordinates， the CSSTORM results were first converted to a molecule list as input for LAMA by ident i fying clusters of nonzero grid points. A region of interest， a 128×128 pixels super-resolution grid image， was chosen for the protein clustering analysis (Figure 5A). The observation radius was set to 19 nm， the minimum cluster size was set to 6 in DBSCAN， and the noise l evel was set to 13.1% in OPTICS24. In total， 32 and 36 clusters were detected by DBSCAN and OPTICS， respectively， and are shown as 2D distributions of cluster populations in Figure 5 (B， E). A DBSCAN-based hierarchical c l uster analysis was performed using the results of DBSCAN or OPTICS to extract morphologica l information for individual clusters by polygon formation. The morphological cluster analysis results with 16 false colors are summarized in Figure 5(C， F). Finally， a cluster population distribution with a cluster size of ca. 90 nm was obtained， as shown i n Figure 5 (D， G)， indicating that this is a promising strategy for analyses of protein distributions in live cells (Figure S10-S13). Figure 5. Distribution of mitochondrial membrane VDPs determined by a protein clustering analysis-using DBSCAN (B， C， D) and OPTICS (E， F， G). (A) STORM super-resolution image (scale bar 500 nm)； (B， E) 2D distribution of the cluster population； (C， F) Visual representation of the DBSCAN-based morphological cluster analysis； (D， G) Distribution of cluster radii. Conclusion In conclusion， a promising strategy for STORM super-resolution microscopy was developed to carry out mitochondrial dynamics imaging in live cells. In this method， a new STORM probe was applied to covalently bind to VDPs in the mitochondrial outer membrane， which achieved STORM super-resolution imaging without t he need for an imaging buffer and thiols. The mitochondrial dynamics analyses were obtained by CSSTORM imaging using a noise-correction method consisting of a NC-PCA+K-factor pre-treatment. De-noising CSSTORM enabled ultra-fine structural imaging of mitochondria with a spat i al resolution of up to 30.2 nm in fixed cells and 45.7 in l ive cells with a temporal resolution of 0.8 s. The dynamic changes of the 1-VDP complex on the mitochondrial membrane were ut i lized to demonstrate mitochondrial dynamics，including fusion， fission， and tubulation， in particular， dynamic tubulation retraction of ca. 746 nm in 1.2 s. The distribution of VDPs on the outer mitochondrial membrane was further analyzed as protein aggregation (ca. 90 nm) to gain a deeper insight into their biological functions. The newly developed strategy will be useful for further functional studies of mitochondria. Acknowledgements is work was supported by the National Natural Science Foundation of China (61875131/61525503/61620106016/61835009) and Key Project of Department of Education of Guangdong Province (2015KGJHZ002/2016KCXTD007). This work was supported by the National Research Foundation of Korea (NRF) funded by the Ministry of Science and ICT (CRI project no. 2018R1A3B1052702，JSK) Reference 1. J. Nunnari， A. Suomalainen， Cel l 2012， 148，1145-1159. 2. S. Meeusen， J. M. McCaffery， J. Nunnari， Science 2004， 305， 1747-1752. 3.C. Wang， W. Du， Q. P. Su， M. Zhu， P. Feng， Y. Li ， Y. Zhou， N. Mi， Y. Zhu， D.Jiang， S. Zhang， Z. Zhang， Y. Sun， L. Yu， Cell Res.2015，25，1108-1120. 4. D. C. Chan， Annu. Rev. Genet. 2012， 46， 265-287. 5. H. Chen， D. C. Chan， Hum. Mol. Genet. 2009，18，169-176. 6. H. Zhu， J. Fan， J . Du， X. Peng， Acc. Chem. Res. 2016，49， 2115-2126. 7. a) S. A. Jones， S. H. Shim， J. He， X. Zhuang， Nat. Methods 2011，8， 499-508；b)S.-N. Uno， M. Kamiya， T. Yoshihara， K. Sugawara， K. Okabe， M. C. Tarhan， H.Fujita， T. Funatsu， Y. Okada， S. Tobita， Y. Urano. Nat. Chem. 2014，6，681-689；c) A. Small ， S. Stahlheber . Nat. Methods， 2014，11，267-279；d) S. Alexey， R. M.Hochstrasser. Proc. Natl. Acad. Sci. U. S. A. 2006，103，18911-18916；e)R. Iinuma，Y. Ke， R. Jungmann， T. Schlichthaerle， J. B. Woehrstein， P. Yin. Science， 2014， 344，65-69. 8. Y. Doksani， J. Y. Wu， T. de Lange， X. Zhuang， Cell 2013， 155， 345-356. 9. a) M. Bates， B. Huang， G. T . Dempsey， X. Zhuang， Science 2007，317，1749-1753； b) B. Huang， W. Wang， M. Bates， X. Zhuang， Science 2008，319，810-813. 10. S. Shen，X-F. Li ， W. R. Cullen， M. Weinfeld， X. C. Le， Chem. Rev. 2013， 113，7769-7792. 11. C. Huang， Q. Yin， W. Zhu， Y. Yang， X. Wang， X. Qian， Y. Xu， Angew. Chem.Int. Ed. 2011， 50，7551-7556. 12. a) L. Zhu， W. Zhang， D. Elnatan， B. Huang， Nat. Methods 2012，9， 721-723； b)J.Huang， K. Gumpper， Y. Chi， M. Sun， J. Ma， Opt. Lett. 2015，40，2989-2992. 13. S. H. Shim， C . Xia， G. Zhong， H. P. Babcock， J. C. Vaughan， B. Huang， X.Wang， C. Xu， G. Q. Bi， X. Zhuang， Proc. Nat l . Acad. Sci . U.S.A. 2012，109，13978-13983. 14. X. Peng， Z. Yang， J. Wang， J. Fan， Y . He， F. Song， B. Wang， S. Sun， J . Qu， J. Qi，M. Yan， J. Am. Chem. Soc. 2011，133，6626-6635. 15. S.N. Uno， M . Kamiya， T. Yoshihara， K. Sugawara， K. Okabe， M. C. Tarhan， H.Fujita， T. Funatsu， Y. Okada， S. Tobita， Y. Urano， Nat. Chem. 2014， 6，681-689. 16. http：//bigwww.epfl.ch/smlm/datasets/. 17. R. P. Nieuwenhuizen， K. A. Lidke， M. Bates， D. L. Puig， D. Grunwald， S.Stallinga， B. Rieger， Nat. Methods 2013， 10， 557-562.18. a) P. Mishra， D. C. Chan， J. Cell Biol. 2016， 212， 379-387；b) D. Sebastian， M.Palacin， A. Zorzano， Trends Mol. Med. 2017，23，201-215. 19. M. Lavorato， V. R. Iyer， W. Dewight， R. R. Cupo， V. Debattisti， L. Gomez， S. De la Fuente， Y-T. Zhao， H. H. Valdivia， G. Hajnoczky， C. Franzini-Armstrong，Proc. Natl. Acad. Sci. U. S. A. 2017，114，849-858. 20. P.R. Nicovich，D. M. Owen， K. Gaus， Nat. Protoc. 2017，12，453-460. 21. S. Jahirabadkar， P. Kulkarni， Expert Syst . Appl.2014，41，2939-2946. 22. Y. Wang， S. Jia， H. F. Zhang， D. Kim， H. Babcock， X. Zhuang， L. Ying， Optica 2017，4，1277-1284. 23. P.R. Nicovich，D. M. Owen， K. Gaus， Nat. Protoc. 2017， 12，453-460. 24. S. J. Holden， S. Uphoff， A. N. Kapanidis， Nat. Methods 2011，8，279-280. Graphic Abstract A new STORM probe was developed to selectively label vicinal dithiol proteins on the outer mitochondrial membrane， achieved STORM imaging of mitochondrial dynamics in l ive cells. By pre-treating with a noise-correction-principal-component-analysis and a K-factorization algorithm， a compressed sensing algorithm achieved STORM imaging with enhanced temporal resolution. Mitochondrial dynamics， were visualized with ultra-fine details to facilitate the studies on mitochondrial functions in live cells at resolutions beyond the optica l diffraction limit.