Automated highly multiplexed super-resolution imaging of protein nano-architecture in cells and tissues

Understanding the nano-architecture of protein machines in diverse subcellular compartments remains a challenge despite rapid progress in super-resolution microscopy. While single-molecule localization microscopy techniques allow the visualization and identification of cellular structures with near-molecular resolution, multiplex-labeling of tens of target proteins within the same sample has not yet been achieved routinely. However, single sample multiplexing is essential to detect patterns that threaten to get lost in multi-sample averaging. Here, we report maS3TORM (multiplexed automated serial staining stochastic optical reconstruction microscopy), a microscopy approach capable of fully automated 3D direct STORM (dSTORM) imaging and solution exchange employing a re-staining protocol to achieve highly multiplexed protein localization within individual biological samples. We demonstrate 3D super-resolution images of 15 targets in single cultured cells and 16 targets in individual neuronal tissue samples with <10 nm localization precision, allowing us to define distinct nano-architectural features of protein distribution within the presynaptic nerve terminal.

Experimental workflow for multiplex experiment in U2OS cells shown in Figure 2a SI file, p. 28 Supplementary Table 2 Experimental workflow for the multiplex experiment in the medial nucleus of the trapezoid body shown in Figure 3c SI file, p. 29 Supplementary Table 3 All antibodies and other labels used in this work SI file, p. 30 Supplementary Table 4 maS 3 TORM components SI file, p. 31 Supplementary Table 5 Exact number of experiments, samples, and selections for all quantified analyses  Table 4). For emission (blue background), the home-built dSTORM system is equipped with four lasers: 405 nm (#1), 488 nm (#2), 561 nm (#3), and 661 nm (#4). The 405 nm, 488 nm, and the 561 nm laser beams are merged using two dichroic mirrors (#5 and #6). The 661 nm laser beam that is primarily used for dSTORM measurements is conjoined with all the other laser beams by the dichroic mirror (#7). Subsequently, the excitation light is widened by two lenses forming a telescope and is reflected onto the sample using another dichroic mirror (#8). The emission beam (red background) is collected by the objective (#9; yellow background) and passes a tube lens followed by the cylindrical lens (#10) that can be used for 3D imaging. At the focal point of the tube lens a slit (#11) is placed, followed by another lens parallelizing the beam. Different emission filters to block excitation light are mounted in a filter wheel (#12). The dichroic filter (#13) splits the emission light into two beams with one beam passing and the other being reflected at the inner edge of a prism (#14) resulting in two channels that are focused onto the camera (#15). The microscope is further provided with an infrared beam-based active focus stabilization system (green background). The incoming beam of the infrared diode (#16) is coupled into the objective by the dichroic mirror (#17). The beam is reflected at the edge between the glass surface of the sample and the sample medium. Using a movable mirror (#18), the reflected beam is projected onto a quadrant diode (#19). To correct for focus drift, the quadrant diode is linked to a controller (not shown) that, in turn, initiates movement of a piezo-driven stage (#20) wherein the objective is mounted.  The Chronos plugin continuously checks for the text file. Once the desired file has been found (3), the path is opened and the respective list of commands is imported in Chronos (4) and executed by the robot (5; photograph at the lower right provided by CTC Analytics AG with permission to publish). After successful execution, Chronos receives a response (6) and the plugin saves a file containing the string "done" to the exchange folder (7). The Experiment Editor monitors the exchange folder (8), finds the "done" file (9), and jumps to the next module (10). If the next module is a microscope-related module, the command is transferred to the Microscope Control software (11) and to the hardware components of the microscope (12). After the command has been executed, a feedback is given to the Microscope Control (13) and to the Experiment Editor (14) that proceeds to the next module (15). (b) Experiment Editor modules for the robot ensure maximal flexibility by using variables. In this example the preparation of a simple module for the addition of variable volume (magenta highlighted fields) of solutions from variable washing stations (yellow highlighted fields) to the sample (see upper part of panel 3) is described. First, within the Chronos software (Screenshot with permission of Axel Semrau GmbH & Co KG) a list of commands called "Sample list", in turn, consisting of smaller command blocks called "Analysis methods" is created (upper part of panel 1) and saved as a command list template in the cml format (lower part of panel 1). Second, values that are intended to be specified within the Experiment Editor GUI are substituted by specific variables (panel 2). Third, when the robot-related module gets executed, the variables in the pre-generated cml template file are automatically substituted by the parameters specified by the user (panel 3) and saved. A file containing the path to the customized command list is saved in the exchange folder as described in (a).

Supplementary Figure 5 | Defining bleaching and elution conditions.
For the following control experiments, U2OS cells were initially (before the 1 st imaging round) stained with primary antibodies against Tom20 (from rabbit) and Alexa 647 labeled secondary antibodies against rabbit. The 2 nd imaging round was performed after elution/and or bleaching and the 3 rd round after cells were re-stained with secondary antibody against rabbit. (a) Comparison of different elution buffers. Tom20 signal in the initial imaging round; signal remaining after bleaching and treatment with three buffers in the second round; signal that can be retrieved after re-staining. The latter shows that treatment with 3. For the full-frame analysis of data, rendered images were split into foreground (Tom20 signal) and background pixels using Ilastic software. Foreground areas were exported as binary masks and loaded in ImageJ together with the rendered images. The average foreground pixel intensity per signal-positive area was calculated using a custom-written script in ImageJ. Bars represent mean values. For sample numbers, see Supplementary Table 5. Source data are provided as a Source Data file.

Supplementary Figure 6 | Control experiments for bleaching-and elution-mediated signal removal. (a)
Six test experiments evaluating the cross-talk between consecutive multiplexed imaging rounds. U2OS cells were stained with a primary antibody against Tom20 from rabbit (rb) and a secondary anti-rabbit antibody conjugated to Alexa Fluor 647. Three STORM imaging rounds were carried out for each condition, see Supplementary Note 1 for a detailed description. The dashed line reflects the average signal intensity normalized to the initial staining (obtained by quantifying localizations within selected Tom20-positive regions).
Bleaching was either induced by the STORM acquisition itself or by the intended photobleaching procedure (see Methods). The secondary antibody was reapplied and the percentage of signal retrieval was evaluated. To measure the influence of a competing primary antibody (test condition 6), rabbit Fibrillarin antibody was applied. Each experiment was carried out for 3 independent sample dishes with 3 cells analyzed per dish. For Tom20 signal quantifications, 3 rectangular selections in mitochondria-positive areas were analyzed per cell (b). Background signal was determined in cell area lacking mitochondria (3 selections/cell) and subtracted from Tom20 signal to quantify specific Tom20 signal. Bars represent mean ± SD.
For statistical analysis, all bars corresponding to imaging round 2 were compared to round 2 of the first test condition. In analogy, all bars corresponding to imaging round 3 were compared to round 3 from the first test condition. Asterisks indicate statistical significance (*** corresponding to p < 0. (g) Nup133-Ypet expressing U2OS cells that underwent ten rounds of bleaching and elution were stained with anti-GFP nanobodies (also recognizing Ypet). Despite the very low epitope density (four epitopes per nuclear pore subunit), individual nuclear pore complexes are still apparent in the two exemplary panels and insets representing magnifications of boxed regions in imaging round 11. Data points (a, c) and bars (e, f) represent mean values. Scale bars correspond to 10 µm (b, d) and 500 nm (g). For sample numbers, see Supplementary Table 5. Note that in addition to representative images in b and d another 2 experiments were carried out, with similar designs and outcomes. Experiment shown in g has been carried out only once. Source data are provided as a Source Data file. distance between corresponding localizations from adjacent STORM frames 3 . NeNa analysis resulted in a mean NeNa distance of 7.5 ± 2 nm (mean ± SD) for all imaging rounds, indicating a very high localization precision. For this analysis we used the NeNa algorithm of LAMA software 4 . This was also confirmed by a decorrelation analysis 2 resulting in a high estimated average spatial resolution of ~28 nm for the experiment in cells (b) shown in Figure 2a and ~29 nm for the experiment in tissue (c) shown in Figure 3c. (d) Analysis of the nuclear pore complex (NPC) metrics based on the Nup133 signal from the multiplex experiment in U2OS cells (Fig 2a) results in an average diameter of 103 ± 8 nm, which is within the range described in literature 5 . For this analysis, using a custom-written ImageJ script, the center of mass of an NPC was found (1) and a line selection with a width of 30 nm was laid through the NPC (2) and rotated around the center of mass in 30° steps resulting in six line selections (3). Six line plots were generated (4) per NPC, their peaks were Gauss-fitted (5) by a custom-written Matlab code and the mean distance between the two peaks was calculated and averaged among 10 NPCs tested (6). For this, the inner calyx border is first localized within a defined window; this fit is very reliable as the distance between the line profile origin and the inner border is nearly constant. As different calyces and calyx stretches have variable thickness, the second (outer) border is identified by two intensity minima: a first one that directly follows the first border, and a subsequent one (green arrows in panel 6). The strong intensity peak in between those minima is identified as the second border, and its position approximated by a Gaussian function. (7) Mean calyx thickness is determined from aligned and scaled profiles. (8) All profiles of WGA and proteins of interest are averaged, scaled, and aligned according to the mean calyx thickness. (b) For the analysis of AZ-specific protein distribution, two line profiles in AZ-negative (AZ(-)) and two profiles in AZ-positive (AZ(+)) presynaptic regions were drawn per calyx. (c) For colocalization analysis, line profiles were truncated to specifically cover only the presynaptic area and converted to area selections (1). Using a custom ImageJ plugin, all possible combinations of targets underwent Pearson's colocalization analysis (2). All data were imported to Matlab to generate colocalization matrices (3), which were imported to Excel where they were averaged and finalized. (a-c) Scale bars correspond to 1 µm.  Figure 6a) We thoroughly assessed the performance of re-staining, bleaching, and elution using Tom20 staining as a test case. For each control condition (Supplementary Figure 6a), three sequential dSTORM imaging experiments (with three acquisitions per experiment and three ROIs per acquisition) were conducted, from which we quantified the fluorescence intensity by determining the number of single-molecule localizations (Supplementary Figure 6b).

Extended information relating to control experiments assessing signal removal after each staining round (Supplementary
First, we determined the fluorescence intensity in three consecutive dSTORM experiments performed with identical imaging conditions (data set #1 in Supplementary Figure 6a). The number of localizations of the second and third experiments were normalized with respect to the first experiment. As expected, dSTORM imaging decreased fluorescence intensity in subsequent imaging experiments by photobleaching some of the fluorophore labels (see Supplementary Figure 6c for graphical illustration).
Second, we performed an experiment with three consecutive dSTORM acquisitions, but labeled again with the same fluorophore-labeled secondary antibody prior to the third imaging round (data set #2). We found similar fluorescence intensities as in the first data set, despite re-labeling with the secondary antibody. This result tells us that the secondary antibody labeling reached saturation.
Third, we recorded three dSTORM data sets, yet now photobleached the remaining fluorescence signal after the first acquisition by applying high irradiation intensities (data set #3, see Methods). The second imaging round showed us that bleaching was efficient since the fluorescence intensity dropped to background level. Prior to the third experiment, we relabeled the sample again with the same fluorophore-labeled secondary antibody. We found a recovery of the fluorescence intensity, which we explain by photo-induced unbinding of the secondary antibody through high-intensity illumination of the sample during the photobleaching step. We find similar ratios of recovery (about 45%) as reported in the literature 6 .
Fourth, we explored the efficiency of elution (data set #4, see Methods). After the first imaging round, we eluted the antibodies, and we observed a decrease in fluorescence intensity. A third imaging round shows a decreased intensity as observed in condition one and two.
Fifth, we explored the efficiency of both photobleaching and elution after the first imaging round (data set #5). After the initial staining, we first exposed the sample to the elution buffer followed by a photobleaching step. We found the fluorescence intensity to drop to background level. Re-staining with the same secondary antibody recovers fluorescence signal (to 8.6%), which we attribute to photo-unbinding of the secondary antibody 8 .
Sixth, we show that the combination of photobleaching and elution is highly efficient, if another primary antibody (other than Tom20; in this case Fibrillarin) competing for the same secondary antibody is introduced before the last acquisition round (data set #6). This would represent the typcial workflow of a multiplex experiment.
In summary, for optimal performance of our approach, we recommend to design a re-labeling experiment such that (i) both elution and photobleaching steps are integrated in the workflow, and that (ii) the labeling sequence is designed such that the species of secondary antibodies changes every imaging round. See Supplementary Note 2 for a detailed description of experimental design.

Set of rules helping the user to design an optimal multiplex experiment
Here, we list some main rules that help to design an optimal multiplex experiment. Depending on the specific scientific question, the experimenter can reorder the rules by prioritizing rules that help to get optimal results for particular structures of interest.
-Stain targets with low epitope density in early staining rounds.
-Use labels with low labelling efficiency in early staining rounds.
-Use high affinity antibodies in late staining rounds.
-Try to avoid elution-based label removal for as many rounds as possible. Bleaching is sufficient as long as not interfering non-antibody labels and antibodies from different species are used.
-Avoid primary antibodies from same species in consecutive rounds.
-For direct labels (such as lectins) that cannot interfere with other labels use bleaching only.
-For primary/secondary antibody labeling use elution only.
-Do not stain targets with high number of epitopes or use labels with high labeling efficiency together with targets with low epitope numbers or low labeling capacity in one round. This is because cross-talk originating from demixing can lower the signal to background ratio for the low-signal target.
-Cytoskeleton components are more sensitive to sample treatment and should therefore be labeled in early staining rounds if very detailed ultrastructural data is required.
-If using primary antibodies from same species in consecutive rounds, alternate secondary antibody labels (Alexa 647 and CF680). This allows to distinguish between cross-talk from fluorophores of remaining secondary antibodies or from remaining primary antibodies.
-If imaging in PAINT (points accumulation for imaging in nanoscale topography) mode, image fluorescent fiducials before application of PAINT dyes. Otherwise fiducial detection might be impeded due to the high background produced by PAINT dyes in solution.
-The best possible correlation of two targets can be achieved by imaging in the same round (channel 1 and 2).

Supplementary discussion of registration precision between different staining rounds for cells or tissues
The serial re-staining approach used here requires a high-fidelity registration of images from each staining round to reliably derive conclusions on the positioning of each protein with regard to all other proteins investigated. We achieved this by imaging fiducial beads positioned between the sample and the coverslip for every staining round. Hence, assuming an invariant position of each bead, the registration procedure should work with similar precision for images acquired in any staining round. Protein positions determined from the first staining round should be equally comparable to those of the second or 10 th staining round. However, if the position of the fiducial beads varies to a small extent during the long duration of a typical multiplexed imaging experiment, deviations will occur, in particular for protein positions determined from imaging rounds far apart, but not much affecting neighboring staining rounds. This relationship is quite evident in Figures 2b and 3d. Noticeably, the registration precision was overall better for super-resolution images acquired in tissue (Figure 3d) compared to cells (Figure 2b). We attribute this to a differential stability of the fiducials in these two imaging situations as detailed in the following. In contrast to the flat 400 nm sections, cultured U2OS cells have a substantial thickness of several micrometers. Due to their thickness and other mechanical properties, cells may not homogeneously adhere to the glass surface, resulting in a slight partial detachment with each additional staining round. This could in turn loosen fiducial beads stuck underneath the cells. Slight changes in cell volume with increasing staining and elution rounds may cause a similar effect. Therefore, to re-establish a focal plane in late imaging rounds comparable to that from the initial round, the focus would have to be placed more and more deeply into the sample, resulting in less focused fiducial beads, and thus, less precise fitting and a higher registration error. Indeed, we once had to manually readjust the focus in round five of the multiplex experiment in cells (Figure 2a). In the much more homogeneous tissue sections, these effects will impact registration precision much less, consistent with the experimental data (Figures 2b and 3d). Hence, we have focused our quantitative analyses on neuronal tissue samples to minimize these complications. Moreover, for the analysis of protein relationships we performed five multiplex experiments with target proteins imaged in a different order. Thereby, we balanced out possible effects arising from misalignment in late compared to early imaging rounds. Future work may include the use of fiducials covalently fixed to the cover slip using nano-engineering procedures.       Localization precision and spatial resolution (Fig. S10a, b, c) Nuclear pore complex metrics (Fig. S10c) Control experiments: repeated bleaching and elution in MNTB tissue (Fig. S9c, f) Control experiments: repeated bleaching and elution in U2OS cells (Fig. S9a, e) Supplementary Line profile analysis of the overall protein distribution at the calyx of Held (Fig. 3f, g; Fig. S13a-d) Line profile analysis of AZ-specific protein distribution at the calyx of Held (Fig. 3h, j; Fig. S13e-o) Colocalization analysis of AZ-specific protein distribution at the calyx of Held ( Fig. 3m; Fig. S14) Control experiment: cross-talk with Tom20 as test case (Fig. S6a) Control experiments: elution and bleaching with multiple targets as test case (Fig. S8b-d) Control experiment: treatment with three different elution buffers (Fig. S5a) Control experiment: three different incubation times for elution (Fig. S5b) Control experiment: four different bleaching times Fig. S5c)