124-Color Super-resolution Imaging by Engineering DNA-PAINT Blinking Kinetics

Optical super-resolution techniques reach unprecedented spatial resolution down to a few nanometers. However, efficient multiplexing strategies for the simultaneous detection of hundreds of molecular species are still elusive. Here, we introduce an entirely new approach to multiplexed super-resolution microscopy by designing the blinking behavior of targets with engineered binding frequency and duration in DNA-PAINT. We assay this kinetic barcoding approach in silico and in vitro using DNA origami structures, show the applicability for multiplexed RNA and protein detection in cells, and finally experimentally demonstrate 124-plex super-resolution imaging within minutes.


Stochastic binding simulations
DNA-PAINT blinking traces were simulated using the stochastic reaction simulation tool COPASI 1 . Simulations were carried out as described earlier 2 . In brief, we simulated structures displaying 40 binding sites of 8 nt length, or 40 sites of 10 nt length and structures displaying 120 sites of 8 nt or 120 sites of 10 nt length. All simulation parameters were experimentally obtained. For 8 nt and 10 nt binding sites, a mean binding duration of 0.4 s and 5 s was determined, respectively. With our current imaging buffer constitution, an association rate of 2.9 • 10 ' (Ms) ,was used. Imager concentration was set to 80 pM, and the integration time to 30 ms with 1500 s total acquisition time. For each DNA origami design, 50 structures were simulated separately. The data from the four simulations were pooled, and clustered using the HDBSCAN algorithm, with 'min_cluster_size' set to 15, and 'min_samples' set to 1. (Supplementary  Figure 1). DNA origami design, assembly and purification DNA origami structures were designed using the design module of Picasso 3 . Folding of structures was performed using the following components: p7249 M13 single-stranded DNA scaffold (0.01 µM), core staples (0.5 µM), biotin staples (0.5 µM), modified staples (each 0.5 µM), 1× folding buffer in a total of 20 µl for each sample. Annealing was done by cooling the mixture from 65 ºC to 25 ºC in 2 hours in a thermocycler. Structures were purified either using PEG-precipitation 4 (40 versus 120 binding sites and 4 corner origami), or by running the samples on a 1.5% agarose gel (1.5% agarose, 0.5× TA buffer, 12.5 mM MgCl2, 1× SYBR Safe), cutting out bands containing the folded structures and purifying them using Freeze 'N Squeeze Columns (spun for 5 min at 1,000 ×g) (124 color imaging) (Supplementary Figure 9).

Cell culture
HeLa cells were used for CHC and PMP70 imaging. For RNA-FISH experiments A-549 cells were used. All cells were grown in high glucose (4.5 g/l) DMEM supplemented with GlutaMAX TM , 1 mM sodium pyruvate and 10% FBS. Cells were seeded into 8-wellchambered cover glasses and grown to approximately 70% confluency.

Design of RNA-Fluorescence in situ Hybridization probes
RNA-FISH probes were designed against the mRNA sequence of the longest transcript variant of each gene (Supplementary Notes 3 and 4). FASTA sequences were taken from the NCBI Genome Browser. We used the Stellaris â Probe Designer version 4.2 with a masking level of 5 to get 40 probe strands for each target. These probes were then elongated on the 3'-end with DNA-PAINT handle sequences for DNA-PAINT imaging (Supplementary Note 2).

Hybridization of RNA-FISH probes
Cell media was aspirated and cells were rinsed with 1× PBS. Cells were fixed with 4% formaldehyde in 1× PBS for 10 min at room temperature, then washed two times with 1× PBS and 4 mM VRC. Permeabilization was carried out with 70% (v/v) ethanol for 6 hours at 4 ºC. Before hybridization, cells were incubated in wash buffer supplemented with 4 mM VRC for 10 minutes at room temperature. Hybridization: 300 μl of the hybridization solution containing 12.5 nM of probes was added to the cells. Hybridization was carried out in a sealed chamber at 37 ºC for 16 hours. Washing: Chambers were rinsed once then washed twice for 30 min each at 37 ºC in wash buffer.

CHC and PMP70 Immunostaining
Cell medium was aspirated and cells were fixed with 3% paraformaldehyde, 0.1% glutaraldehyde, and 0.3% Triton X-100 in PBS for 10 min at room temperature, then washed three times with PBS. Free aldehyde groups were reduced using 1 mg/ml sodium borohydride in PBS for 7 min, followed by three washing steps with PBS for 5 min. Cells were blocked and permeabilized with blocking buffer for 90 min. Cells were stained with primary antibodies, anti-CHC goat and anti-PMP70 mouse (both diluted 1:100), in blocking buffer overnight at 4 ºC. Cells were washed three times with PBS for 5 min. Cells were incubated with DNA-conjugated secondary antibodies (anti-goat-P12-8 nt and anti-mouse-P13-9 nt (Supplementary Table 8) diluted in blocking buffer (1:200) for 1 hour at room temperature before finally washing the cells three times in PBS for 5 min.

Super-resolution microscope setups
Custom TIRF Setup. Fluorescence imaging was carried out on an inverted Nikon Eclipse Ti microscope (Nikon Instruments) with the Perfect Focus System, applying an objective-type TIRF configuration with an oil-immersion objective (Apo SR TIRF 100×, NA 1.49, Oil). Two lasers were used for excitation: 561 nm (200 mW, Coherent Sapphire) or 640 nm (150 mW, Toptica iBeam smart). The laser beam was passed through cleanup filters (ZET561/10 or ZET642/20, Chroma Technology) and coupled into the microscope objective using a beam splitter (ZT561rdc or ZT647rdc, Chroma Technology). Fluorescence light was spectrally filtered with an emission filter (ET600/50m and ET575lp or ET705/72m and ET665lp, Chroma Technology) and imaged on an electron-multiplying charge-coupled device (EMCCD) camera (Andor iXon Ultra 897) or sCMOS camera (Andor Zyla 4.2) without further magnification, resulting in an effective pixel size of 160 nm (EMCCD) or 130 nm (sCMOS after 2×2 binning). Our custom TIRF setup was used for Figures 1d and 2f (EMCCD). Spinning Disk Confocal Setup: DNA-PAINT imaging of RNA-FISH samples was performed using an Andor Dragonfly Spinning Disk Confocal system (Andor) based on an inverted Nikon Eclipse Ti2 microscope (Nikon Instruments) with the Perfect Focus System, using an oil immersion objective (Plan Apo 100×, NA 1.45, Oil). For excitation, a 561 nm laser (2 W, MPB) was used. The laser beam was passed through a beam conditioning unit (Andor Borealis) for reshaping the beam from a Gaussian profile to a flat top profile. Next, the beam was coupled into the Andor Dragonfly spinning disk unit, passed through the multi-pinhole disk with a pinhole size of 40 µm and from there coupled into the objective lens. Excitation and emission light was spectrally split using a beam splitter (CR-DFLY-DMQD-01). Fluorescence light was spectrally filtered with an emission filter (TR-DFLY-F600-050) and imaged on an sCMOS camera (Andor Zyla 4.2 PLUS) without further magnification, resulting in an effective pixel size of 130 nm (sCMOS after 2×2 binning). The field of view was 1024 × 1024 pixels which is equivalent to 133.12 µm × 133.12 µm when taking the pixel size into account. The disk speed was set to 6000 rpm and an excitation field stop of 13.3mm × 13.3mm was applied. The Spinning Disk Confocal setup was used to acquire the image in Figure 2b. Figure 1d. Imaging was carried out using an imager strand concentration of 75 pM (P3-Cy3B). 50,000 frames were acquired using the EMCCD camera at 30 ms integration time with the EM gain set to 300, the pre-amp gain set to 3, and the readout bandwidth set to 17 MHz. Laser power (@561 nm) was set to 2.5 mW (measured before the back focal plane (BFP) of the objective). Figure 1f. Images were acquired with an imager strand concentration of 2 nM (P3-Cy3B imager). 150,000 frames were acquired using the EMCCD camera at 30 ms integration time with the EM gain set to 300, the pre-amp gain set to 3, and the readout bandwidth set to 17 MHz. Laser power (@561 nm) was set to 500 mW (measured at the back focal plane (BFP) of the objective).

Figure 2b
. DNA-PAINT microscopy was carried out using 10 nM of P3-Cy3B in imaging buffer. 40,000 frames were acquired using the EMCCD camera at 200 ms integration time and a readout bandwidth of 540 MHz. Laser power (@560 nm) was set to 500 mW resulting in a power of 18.3 mW at the sample plane. This can be translated to an intensity of 103.27 W/cm 2 at the sample plane. Figure 2f. DNA-PAINT imaging of protein samples was carried out using the following imager strands: P12-Cy3B (250 pM) and P13-Cy3B (50 pM) in imaging buffer. 80,000 frames were acquired using the EMCCD camera at 30 ms integration time with the EM gain set to 300, the pre-amp gain set to 3, and the readout mode to 17 MHz. Laser power (@561 nm) was set to 8 mW (measured before the back focal plane (BFP) of the objective). Figure 3c. Imaging was carried out using the following imager strands: P1-Atto655 (20 nM), P2-Cy3B (20 nM) and P3-Atto488 (20 nM), in Buffer B. 15,000 frames were acquired using the EMCCD camera at 100 ms integration time with the EM gain set to 300, the pre-amp gain set to 3, and the readout mode to 17 MHz. Laser power was set to ~25 mW (@488 nm), ~30 mW (@561 nm), and ~30 mW (@642 nm, all measured before the back focal plane (BFP) of the objective). For all imager strand sequences see Supplementary Table 1.

Image analysis
Raw fluorescence data was subjected to spot-finding and subsequent super-resolution reconstruction using the 'Picasso' software package 3 . Analysis of DNA origami data Automated structure selection: After super-resolution reconstruction, structures were automatically selected using Picasso's 'Pick similar' function with the following settings: Pick radius: 320 nm; Standard deviation: 2. Filtering: After automated selection, picked 'spots' were further processed in order to remove unspecific binding events from specific ones originating from DNA origami locations. To achieve this, we implemented a multi-step filtering procedure. First, in order to remove non-repetitive binding events (e.g. imager strands non-specifically adsorbing to the surface), we fitted the mean frame value of binding events (from all picked spots) throughout the whole image acquisition. The rationale behind this step is that repetitive, correct picks (i.e. containing DNA origami structures) will yield a mean frame value of roughly half the number of total frames in the acquisition (gaussian distributed), while non-repetitive events will in most cases not last throughout the whole image acquisition time frame, leading to a mean frame value that is outside this distribution. We chose the mean of the distribution and set a cut-off value at +/-two times the standard deviation for filtering. Next, to also filter out structures with a non-repetitive blinking behavior, but with most events occurring around the mean frame value, we plotted the standard deviation of the mean frame values and used a cut-off value of 2000, and all data below this threshold were disregarded. (see Supplementary Figures 3 and 4 for results). Cluster Analysis: After filtering, the data was analyzed using an HDBSCAN 6 clustering algorithm. 'Min_cluster_size' was set to 15, and 'min_samples' was set to 1, in the case of the four origami species (Supplementary Figure 5 and 6), and to 20 and 3, respectively, in the case of the 4-corner origami structures ( Supplementary Figures 7 and 8). Barcode Identification of 124 origami structures: First, all structures from all three acquisition rounds were aligned. Every structure exhibits a distinct kinetic blinking information in each of the three channels. This information was extracted for each spectral channel. The distribution of the number of binding events in each channel shows four separated clusters (see Figure 3b). After assigning every picked structure to one of the clusters in each channel, barcodes were identified. Analysis of RNA-FISH data Single mRNA species were manually selected using Picasso's pick tool with a pick diameter of 520 nm. Kinetic information was extracted as for the DNA origami case. Localizations were linked allowing a gap size of 2 frames between localizations to obtain a list of single binding events. The total number of binding events calculated from each structure was plotted to obtain a histogram of binding frequencies.

Analysis of protein data
Approximately 200 protein clusters were manually selected using Picasso's pick tool with a pick diameter of 240 nm. Kinetic information was extracted as for the DNA origami case. Localizations were linked allowing a gap size of 15 frames between localizations to obtain a list of single binding events. A mean binding time (i.e. blinking duration) was calculated from all events per pick.

Supplementary Figures
Supplementary Figure 1 | Cluster detection in simulated data. Data points from four individual stochastic binding simulations (for details see Online Methods) were clustered using HDBSCAN 6 . Data points plotted according to binding time and blinking frequency before clustering (left), and data plotted after clustering (middle). Individual colors were assigned to each of the resulting four populations. As each population was simulated individually, by comparing the results of the clustering algorithm to the original data, we were able to acquire the success rate of the clustering algorithm we used. The clustering resulted in a positive assignment rate of more than 92% in the case of all four populations (right). Picks are rejected (gray) based on the following metric: More than 2´ standard deviation of the mean. (b) For filtering of structures that show nonrepetitive binding, however whose bindings are clustered around the mean frame, we plotted the standard deviation of the mean frame. Using a cut-off value of 2000 (red dashed line) all data below this threshold were disregarded (gray).