High-Speed Super-Resolution Imaging Using Protein-Assisted DNA-PAINT.

Super-resolution imaging allows for the visualization of cellular structures on a nanoscale level. DNA-PAINT (DNA point accumulation in nanoscale topology) is a super-resolution method that depends on the binding and unbinding of DNA imager strands. The current DNA-PAINT technique suffers from slow acquisition due to the low binding rate of the imager strands. Here we report on a method where imager strands are loaded into a protein, Argonaute (Ago), which allows for faster binding. Ago preorders the DNA imager strand into a helical conformation, allowing for 10 times faster target binding. Using a 2D DNA origami structure, we demonstrate that Ago-assisted DNA-PAINT (Ago-PAINT) can speed up the current DNA-PAINT technique by an order of magnitude, while maintaining the high spatial resolution. We envision this tool to be useful for super-resolution imaging and other techniques that rely on nucleic acid interactions.

Single-molecule localization microscopy techniques allow researchers to image cellular structures that are not visible through diffraction-limited microscopy methods. Most single-molecule localization techniques rely on the stochastic blinking of a fluorescent signal, by using photoswitchable fluorophores as in photoactivated localization microscopy (PALM)[1] and (direct) stochastic optical reconstruction microscopy ((d)STORM).[2] An alternative approach to achieve stochastic blinking is through fluorescent probes that transiently bind their target, as in point accumulation in nanoscale topography (PAINT).[3−5]

In DNA-PAINT, a fluorophore is attached to a short DNA oligonucleotide, called an imager strand, that specifically binds to a complementary target DNA sequence, called a docking strand.[6] The stochastic blinking of signals is achieved through binding and unbinding of the incoming imager strands to the docking strands and is imaged using total internal reflection fluorescence (TIRF). By changing the length and sequence of an imager strand, one can tune the on- and off-rates of the imager and adjust the specificity. This allows for high multiplexing capabilities since the number of probes is only limited by the number of orthogonal DNA sequences. Furthermore, compared to conventional super-resolution techniques, DNA-PAINT comes with the advantage that imager strands are continuously replenished from the solution, and thus photobleaching is circumvented during the imaging process.

A critical limitation of DNA-PAINT, however, is the low binding rate of DNA, which is typically in the order of 106 M–1 s–1. Given this binding rate, obtaining images with a high spatial resolution (5 nm) usually takes several hours.[7−9] Shorter acquisition times can be achieved by increasing the concentration of the imager strand. However, single-molecule binding events become unresolvable from the background of unbound imager strands, even when TIRF is used. To reduce this acquisition time, DNA-PAINT was recently combined with a single-molecule Förster resonance energy transfer (smFRET).[10,11] This, however, comes at a cost of reduced spatial resolution due to reduced energy transfer efficiency and due to limited choice of dyes. Here we describe an alternative approach, in which protein-assisted delivery of imager strands is demonstrated to speed up the acquisition time 10-fold and only to require a single fluorescence channel.

Argonaute proteins (Agos) are a class of enzymes that utilize a DNA or RNA guide to find a complementary target, either to inactivate or to cleave it. In eukaryotes, an RNA guide directs Ago to complementary RNA targets for post-transcriptional regulation.[12] Ago proteins initially bind their target through base pairing with the seed segment of the guide (nucleotides 2–7 for human Ago).[13−15] Crystal structures have revealed that Ago preorders this seed segment into a helical conformation, allowing for the formation of a double helix between guide and target, and hence effectively prepaying the entropic cost of target binding.[16,17] This results in binding rates that are near diffusion limited (∼107 M–1 s–1).[18−21] In prokaryotes, there is a broad diversity of Agos with respect to the identity of their guide (RNA/DNA) and their target (RNA/DNA).[22,23] Some well-characterized prokaryotic Ago nucleases (TtAgo, CbAgo) use DNA guides to target single-stranded (ss)DNA.[24,25]

Here we describe a new DNA-PAINT method based on protein-assisted delivery of DNA imager strands, which allows for faster acquisition of super-resolved nanostructures. We use a wild-type Ago protein from the bacterium Clostridium butyricum (CbAgo) to speed up the kinetic binding of DNA imager strands. CbAgo reshapes the binding landscape of the imager strand, resulting in a 10-fold higher binding rate compared to conventional DNA-PAINT. Ago-PAINT can be implemented without additional complexity while retaining the programmability and specificity of DNA-PAINT, due to the favorable targeting feature of CbAgo.[25,26] We determine the spatial resolution of Ago-PAINT through the use of 2D DNA origami structures and show that Ago-PAINT generates super-resolution images of diffraction-limited structures at least 10-fold faster than conventional DNA-PAINT.

Results

For high-quality super-resolution images, a PAINT-based method requires more than five transient binding events per localization spot, each with a dwell time of at least several hundreds of milliseconds.[7−9] A typical 8-nt DNA-PAINT imager strand exhibits an on-rate of ∼106 M–1 s–1 and a dwell time (=1/off-rate) of ∼1 s.[9] DNA-PAINT experiments use an imager strand concentration between 1 and 10 nM. This range is chosen to be high enough to obtain a sufficient number of binding events during the acquisition time, but not too high to avoid cross-talk localization between structures.[7]

We determined the on- and off-rates of Ago-PAINT imager strands and compared these to the on- and off-rates of conventional DNA-PAINT with the same imager strands using a smFRET assay (Figure ). Acceptor (Cy5)-labeled ssDNA targets were immobilized through biotin–streptavidin conjugation on a PEGylated quartz slide. Next, either donor (Cy3)-labeled 8-nt DNA-PAINT imager strands or Ago-PAINT imager strands (CbAgo loaded with a Cy3-labeled guide) were injected, and their interactions with the immobilized target strand were probed using TIRF microscopy (Figure A). The assay was designed to give a high FRET signal upon specific binding of either DNA imager strand or Ago-guide complex to the complementary target (Figure B,C). The Cy3 position was picked the same as in previous studies with CbAgo, to prevent any photophysical artifacts from occurring.[26,27] The time between the introduction of the imager strands and the first binding event is the apparent arrival time (which is the inverse of the on-rate, kon). The duration of the FRET binding events is the dwell time (Figure B).

Figure 1

Single-molecule FRET assay to quantify binding kinetics Ago-PAINT vs DNA-PAINT. (A) A schematic of the single-molecule FRET assay with the target strand immobilized on a PEGylated surface through biotin–streptavidin conjugation. The green and red stars indicate the Cy3 and Cy5 dye, respectively. Binding of the Ago-guide complex or ssDNA probe to the ssDNA target results in a high FRET signal. (B) Representative traces of ssDNA binding (top) and Ago-complex binding (bottom). The dashed line indicates the time point at which the Ago-guide or DNA is iintroduced inside the microfluidic chamber. (C) A schematic of the sequences used for Ago-PAINT and DNA-PAINT. Upon binding, both constructs will give rise to a high FRET signal. (D) Dwell-time histogram (Δτ) of ssDNA (sequence shown in Figure C). Maximum likelihood estimation (MLE) gives 1.1 ± 0.2 s as the parameter for a single-exponential distribution (blue line). Number of data points: 1029. (E) Dwell-time histogram (Δτ) of Ago (sequence shown in Figure C). MLE gives 1.2 ± 0.2 s as the parameter for a single-exponential distribution (blue line). Number of data points: 696. (F) Cumulative binding event plots of DNA-PAINT (black) and Ago-PAINT (orange) vs time. A single exponential fit is used for DNA-PAINT (red line) and Ago-PAINT (orange line). Errors in panels D–F are determined by taking the 95% interval of 105 bootstraps.

For a comparison between Ago-PAINT and DNA-PAINT, we designed an 8-nt DNA-PAINT imager strand (Figure C) and found that, under our experimental conditions, the average dwell time of this imager strand is 1.1 ± 0.2 s (Figure D). Next, we sought to find an Ago-PAINT guide with a similar dwell time. The first nucleotide of an Ago guide is embedded within the protein structure (Figure S1A).[16,17] Therefore, we determined the dwell time of Ago-guide complexes with different numbers (N) of base pairing with the target starting from the second nucleotide onward (Figure S1B). A guide with N = 5 (nt 2–6) base pairing to the target exhibited a comparable dwell time of 1.2 ± 0.2 s (Figure E). We observed that, for Ago-PAINT, the apparent arrival time is influenced by the number of base pairs that are formed between the guide and its target. For N = 5 or larger, the on-rate reaches a saturated value (kon = 8.3 ± 0.8 × 107 M–1 s–1) (Figures 1F and S1C). Those values are 10 times higher than the typical on-rates for an 8-nt DNA-PAINT imager strand, 8.7 ± 0.8 × 106 M–1 s–1 (Figure F).

To demonstrate the use of Ago-PAINT for super-resolution imaging, we designed a rectangular two-dimensional DNA origami structure of 76 nm × 80 nm (Figure and Figure S2). The DNA origami structure has four docking sites that are spaced 61 nm × 68 nm apart (Figure A). To achieve optimal Ago binding to the DNA origami docking strands, we introduced a polyT linker between the target sequence of Ago and the DNA origami structure (Figure A, right panel). As our previous observation with CbAgo[26] suggested that the protein occupies a footprint of around 20 nt, we made the length of polyT to be 30 nt.

Figure 2

Ago-PAINT enables the same localization precision as conventional DNA-PAINT. (A) Left: A schematic design of the 2D-DNA origami structure. The orange honeycombs indicate the approximate locations of binding sites. Right: 3D representation of the imaging scheme with the docking strand sequence. The green star indicates the position of the Cy3 dye labeled on aminomodified thymine. (B) A representative super-resolution image showcasing binding sites of the 2D-DNA origami structures using Ago-PAINT. Bottom: Super-resolution reconstruction of the four-corner origami structures of the top panel. (C) A summed image of 220 origami structures visualized through the use of DNA-PAINT. (D) A summed image of 219 origami structures made through the use of Ago-PAINT. The concentration of the imager strand was 1 nM for both DNA-PAINT and Ago-PAINT. (E) Fitting of a cross-sectional intensity histogram from the yellow encircled area in panel C to a Gaussian (blue line) shows a localization precision of 10.6 nm. (F) Fitting of a cross-sectional intensity histogram from the yellow encircled area in panel D to a Gaussian (blue line) showing a localization precision of 9.5 nm. Scale bars in panel B indicate 500 nm (top) and 50 nm (bottom three). Scale bars in panels C and D indicate 100 nm.

Next, we sought to compare the localization precision of Ago-PAINT and DNA-PAINT. We tested our Ago-PAINT approach by injecting guide-loaded Ago into our flow cell in which DNA origami structures were immobilized. A super-resolution image could be reconstructed from the Ago-PAINT data, which revealed four detectable spots on the origami structures as expected from our assay design (Figure B). We determined the localization precision by selecting 220 origami structures for DNA-PAINT and 219 structures for Ago-PAINT and created a sum image using the Picasso analysis software[7] (Figure C,D). The localization precision was determined by plotting the cross-sectional histogram of one of the four binding sites of the summed DNA origami structure. For DNA-PAINT, this resulted in a localization precision of 10.6 nm (Figure E), and for Ago-PAINT, we found a localization precision of 9.5 nm (Figure F). The histogram demonstrates that Ago-PAINT delivers the same quality of localization precision when compared to the DNA-PAINT approach. Nearest neighbor analysis[28] reconfirms that the localization precision is similar for both Ago-PAINT and DNA-PAINT (Figure S3). Additionally, we probed the possibility of using different linker lengths for Ago-PAINT imaging. When we tested DNA origami structures with longer linkers (50 thymines or 100 thymine nucleotides), we found that this did not affect the localization precision of Ago-PAINT (Figures S4 and S5), showing that Ago-PAINT is compatible with various linker lengths (≥T30).

Finally, we compared the speed of super-resolution imaging through Ago-PAINT with the conventional DNA-PAINT approach using the 2D DNA origami structures as a testing platform. We evaluated the quality of a super-resolution image after each time point for both Ago-PAINT and DNA-PAINT (Figure A). The overall resolution of a single-molecule localization microscopy image is dependent on the number of localizations per docking strand. Therefore, to quantify the speed of imaging, we plotted the standard error of the localization precision as a function of frame number (Figure B), where we took the sigma values from Figure E,F as the localization precision. We observed that the standard error of the localization precision for Ago-PAINT is smaller than that of DNA-PAINT at each time point, indicating that super-resolved images of identical resolution will be obtained 10× faster through Ago-PAINT compared to DNA-PAINT. This result is further supported by the intensity vs time traces, which shows that our Ago-PAINT method results in more binding events compared to the DNA-PAINT approach, under similar conditions with DNA concentrations of 1 nM (Figure C–E and Figure S6). The on-rates for both Ago-PAINT (kon = 4.4 ± 0.1 × 107 M–1 s–1) and DNA-PAINT (kon = 6.6 ± 0.1 × 106 M–1 s–1) on our DNA-origami structure (Figure E) are similar to the on-rates that we found in our single-molecule experiments (Figure F).

Figure 3

Ago-PAINT enables fast imaging of super-resolved structures. (A) Snapshots in time for Ago-PAINT (top) and DNA-PAINT (bottom) showing super-resolution images being formed over time. Exposure time: 0.3 s. The same color scale is used for the intensity in all images. (B) Standard error of Ago-PAINT vs DNA-PAINT plotted versus frame number. (C) Representative intensity vs time data trace of DNA-PAINT at 1 nM DNA concentration showing few binding events occurring within 600 s. The raw data trace is taken from a single origami plate. (D) Representative intensity vs the time data trace of the 1 nM Ago-guide complex showing binding events that occur frequently within 600 s. The raw data trace is taken from a single origami plate. (E) Normalized cumulative distribution of dark times (the time between binding events) for DNA-PAINT (black, n = 4870) and Ago-PAINT (orange, n = 5793). A single-exponential growth curve is used to estimate the binding rate for DNA-PAINT (orange) and a linear fit is used for a first order approximation for DNA-PAINT (red). Scale bars in panel A indicate 100 nm.

Discussion

Here we presented a proof-of-concept of Ago-PAINT that allows for rapid super-resolution imaging. We demonstrated that fast Ago-PAINT recording can be used to acquire super-resolution images of nanostructures while retaining the programmability and predictability of DNA-PAINT.

For the visualization of several complex cellular components in a single cell, multiplexing super-resolution is highly anticipated. Recent developments allow for temporal[29] and spectral[30] multiplexing of DNA-PAINT and we believe that these methods can be integrated with Ago-PAINT. Also, in our previous work, we showed that different guide sequences resulted in distinctly different binding kinetics.[25] This kinetic fingerprinting will allow for additional freedom when designing multiplexing Ago-PADelfINT.[29,31] Furthermore, optimization of the imager sequence and imaging conditions allowed for a further increase in acquisition time for DNA-PAINT.[32] Although the binding kinetics or Ago-PAINT are near diffusion limited, we expect that optimization of the guide sequence could further improve the kinetics Ago-PAINT.

In this study, Ago-PAINT experiments are performed with the wild-type CbAgo protein, which substantially increases the probe size compared to conventional DNA-PAINT. However, successful applications of Argonaute proteins for in vivo gene silencing[33,34] hint that our Ago-PAINT approach could be used in cellular super-resolution imaging. While targeting complex cellular structures in cells could be an issue with full-size CbAgo, it is possible to use truncated versions of Ago. Some truncated versions of approximately half the size (short Agos) exist in nature.[22] We speculate that it will be possible to truncate them further as Ago-PAINT only relies on the property of preforming the helix structure of the imager strand. For example, an Ago variant from Kluyveromyces polysporus that contains only the C-lobe was reported to retain almost all of the binding properties of the untruncated version.[35] Furthermore, as the imager strand is loaded and protected inside the protein, degradation of the imager strand is less likely to occur over time, unlike oligos that are rapidly digested.[36]

In this Letter, we demonstrated the use of CbAgo for super-resolution microscopy. While this CbAgo targets ssDNA, Agos from other species can target RNA.[22] For example, the Ago from Marinitoga piezophila (MpAgo)[37,38] targets RNA, and one could harness the property of a high association rate for other single-molecule imaging applications such as RNA sensing. Recently, dTtAgo has been combined with FISH[39] to allow for labeling of genomic loci in fixed cells. We anticipate the use of RNA-guided Agos for a significant speed-up in similar applications for RNA FISH. Lastly, complementary approaches such as DNA-based STED imaging,[40] qPAINT,[41] or cross-linking on a single-molecule target using Action-PAINT[42] could be combined with our Ago-PAINT approach. We envision the use of Ago-PAINT as a general toolkit to speed up many current existing applications that rely on base-pairing interactions.