DNA-PAINT adaptors make for efficient multiplexing

Abstract

Multiplexed super-resolution imaging offers a route to spatial proteomics; however, time-efficient mapping of many protein species has been challenging. Two recent works in Cell highlight SUM-PAINT and FLASH-PAINT, methods that leverage adaptor DNA strand design to combine advances in multiplexing with increases in speed of label exchange. These advances permit unbiased omics-style analyses to advance biological insights from super-resolution images.

Multiplexed super-resolution imaging offers a route to spatial proteomics; however, time-efficient mapping of many protein species has been challenging. Two recent works in Cell highlight SUM-PAINT and FLASH-PAINT, methods that leverage adaptor DNA strand design to combine advances in multiplexing with increases in speed of label exchange. These advances permit unbiased omics-style analyses to advance biological insights from super-resolution images.

Main text

In a human cell, the proteins encoded by over 17,000 genes exhibit unique spatial relationships.¹ The spatial arrangement of specific proteins in proximity to one another contributes to and, in some contexts, dictates their function. Elucidating the spatial proteome of the cell—in particular, at the molecular scale—is therefore expected to lead to fundamental understanding of complex cellular functions, but measuring it remains a formidable challenge. Super-resolution fluorescence microscopy aims to address this challenge by highlighting specific target proteins and revealing their positions at the molecular scale. The entire spatial proteome of a cell or tissue can in theory be measured through multiplexed labeling and imaging.

DNA technology has enabled the multiplexed super-resolution method of point accumulation for imaging in nanoscale topography (Exchange-PAINT), a form of single-molecule localization microscopy. In Exchange-PAINT, fluorescent single-stranded DNA (ssDNA) oligomers, or imager strands, transiently bind to complementary sequences, or docking strands, attached to the target proteins (Figures 1A and 1B). These sparse single-molecule binding events are collected over time, and their localizations are combined to reconstruct a super-resolution image. Multiplexed imaging is performed by washing out the imager strands and adding a new imager strand against another target protein.

Figure 1.

Schematic comparison of multiplexed DNA-PAINT methods

(A) Initially, all protein targets are labeled to display an ssDNA docking strand. Each shape and color combination corresponds to a different protein species.

(B) Exchange-PAINT localizes one protein target at a time through transient binding of the corresponding imager strand to the docker. Following imaging, imager strands are washed out and interchanged to label the next target.

(C) SUM-PAINT hybridizes adaptors to a subset of docking strands, up to six targets at a time. These adaptors bridge the unique and multiplexable docking strands to the limited imager strand options of speed-optimized PAINT. Targets are sequentially imaged and then the adaptors are extinguished through toehold-mediated strand displacement. Subsequently, the next round of hybridization can begin.

(D) In FLASH-PAINT, adaptor and imager strands for a target are both applied simultaneously. The adaptor binds transiently to the target’s unique docking strand. The adaptor links the docker to the imager, and the target is imaged. The adaptor is removed through the addition of an eraser, designed to stably bind to the adaptor and thus prevent it from interacting with the docker. This can be done simultaneously with the addition of the next adaptor while the same imager strand is used for all targets.

(E) A multicolor super-resolution image is formed from the many rounds of DNA-PAINT imaging. These data can now be subjected to omics-style analyses to identify protein organization patterns.

Figure created using BioRender (https://biorender.com).

Due to the slow binding kinetics of most DNA oligonucleotides and the need for extensive wash-out steps to remove prior rounds of imager strands, the data for an individual protein target in Exchange-PAINT can take several hours to acquire.² In practice, this bottleneck has prevented Exchange-PAINT from multiplexing beyond nine in situ targets.³ Imaging strand design was subsequently improved for speed-optimized PAINT by creating multiple overlapping binding sites on the docking strand, thereby increasing the on-rate.⁴ However, only six different imaging strand sequences can recognize these overlapping docking sites.⁴ Another approach to increase on-rate uses fluorogenicity, where in the unbound state, the imager strand self-quenches the fluorescence of the dye but not following binding. The combination of large probe length with the need to mismatch certain base pairs to achieve suitable off-rates currently limits these fluorogenic imagers to only two sequences.⁵ These improvements to Exchange-PAINT’s acquisition speed thus come at the cost of limitations to the imaging strand design, which consequently limit the number of protein targets in an experiment.

In a recent issue of Cell, two methods are presented that offer improved feasibility of high-speed, multiplexed Exchange-PAINT: secondary label-based unlimited multiplexed DNA-PAINT (SUM-PAINT) and fluorogenic labeling in conjunction with transient adapter-mediated switching for high-throughput DNA-PAINT (FLASH-PAINT).^6,7 These methods recover the multiplexing capability of Exchange-PAINT and are compatible with previous improvements to PAINT probe design that increase imaging throughput.^4,5,8 As before, proteins of interest are tagged with unique orthogonal ssDNA docking strands. What is new in SUM-PAINT and FLASH-PAINT is that a “transient adapter,” or “secondary label,” ssDNA molecule binds both the docking strand and the imager strand (Figures 1C and 1D). The advantage of using an adapter is that it decouples the on- and off-rates of the fluorophore binding. Thus, the off-switching rates can be boosted through either toehold-mediated strand displacement, hybridization with an “eraser” of the adaptor, or hybridization of the adaptor’s docking site for the imager, without interfering with the fast on-rates of fluorogenic or speed-optimized imager strands. Following one ∼1- to 10-min incubation in FLASH-PAINT, or two 15-min incubations in SUM-PAINT, the adapters between different target docking strands are interchanged. While this is similar to the rate of target exchange in Exchange-PAINT, the time required to image each label is on average reduced by a factor of ∼5 as in speed-optimized PAINT.

In both works, the methods were benchmarked against known structures and compared with existing methods. Using DNA origami structures with arrays of direct and adapter docking sites, both teams compared their methods with conventional DNA-PAINT. Interestingly, the dark time between imager strands binding to a single molecule was shorter when using the adapter strategy, meaning faster acquisitions leading to reduced total imaging time. Both teams were able to resolve 5-nm distances between single docking sites in DNA origami structures. Adaptor-mediated labeling, through SUM-PAINT, was also validated through imaging the nuclear pore complex, which is known to take on a cylindrical geometry of ∼100-nm radius, with an 8-fold symmetry. Two copies of the protein Nup96 incorporate into a larger protein complex, of which 16 are arranged into two rings at the top and bottom of the cylinder, tens of nanometers apart. The authors confirmed lateral distances between copies, although vertical distances showed a discrepancy compared with previous works.^9,10

The methods were applied to capture multiplexed super-resolution images of a wider variety of proteins in cells. Schueder et al. imaged whole human cells with four targets and 1-micron-thick optical sections with up to 13 targets,⁷ including proteins enriched in primary cilia, mitochondria, the endoplasmic reticulum, the Golgi, and lysosomes. Two morphologically different cilia were examined, revealing that the scaffold protein septin 2 and the cargo transporter Ift88 differed in abundance and distribution, potentially indicating distinct functions. In another example, protein trafficking to the Golgi was disrupted by brefeldin A treatment, and proteins normally present in the medial and trans cisternae were found to be depleted. Additionally, inter-organelle contact areas were measured across an entire cell and found to range between 0.1 and 0.2 μm².

By iterating through five sets of six secondary labels, Unterauer et al. collected 30-target images of 1-micron-thick optical sections of neurons.⁶ These encompassed synaptic, cytoskeletal, and organellar targets. Features below the diffraction limit of light such as periodic spectrin rings, hollow clathrin vesicles, single neurofilaments, and trans-synaptic nanocolumns were visualized. Synapses were evaluated based on morphometric features such as density of protein species, co-localization, distance, and shape. The unsupervised machine learning approach, uniform manifold approximation and projection (UMAP), identified clusters of synapses within this feature space, representing excitatory, inhibitory, and a new “mixed” synaptic subtype. The authors report that this rare subtype comprises ∼6% of synapses in cell culture and 1.3% in P50 mice and resembles an excitatory glutamatergic presynapse, with an inhibitory postsynapse. Further synaptic diversity was reflected in heterogeneous expression of different synaptic proteins, possibly reflecting maturation state. The presented examples demonstrate the value of multiplexed imaging for characterizing cellular states that could be rare and/or require simultaneous detection of many protein species in individual cells to identify those states.

A notable difference between SUM-PAINT and FLASH-PAINT is the design of the adaptor. In SUM-PAINT, a 20-nt docking sequence stably binds the adaptor, while in FLASH-PAINT, a 10-nt docking sequence transiently binds the adaptor. The transient binding of the FLASH-PAINT adaptor allows for faster signal extinction, reducing the time to change between adaptors in the sample. The adaptors in SUM-PAINT require more time to fully interchange, but the stability of their binding lets multiple adapters hybridize to different protein targets simultaneously without being washed away during the exchange of imager strands. This way, six different speed-optimized imager strand designs can be used sequentially following each round of hybridization (Figure 1C). The choice of strategy to use in each experiment will depend on the imager design. Since there are just two fluorogenic imager strand designs, it is more time efficient to use them with the FLASH-PAINT transient adaptor (Figure 1D).

Ultimately SUM-PAINT and FLASH-PAINT synthesize multiple key advances in Exchange-PAINT, simplifying multiplexing of accelerated single-target imaging. Minimal solution changes are needed to exchange one protein target for another, making it feasible to fully automate the process with a flow chamber setup. Particularly exciting is the demonstration that data obtained, combined with machine learning-based omics analysis, can identify new biology in an unbiased manner (Figure 1E). However, these methods are limited by a key issue common to fluorescence microscopy, in that they require protein targets to be labeled to be detected. Although both teams show resolution of locations 5 nm apart on DNA origami, antibodies, widely used for labeling, are ∼12 nm in size. Antibodies are also limited in affinity and specificity and may be unable to penetrate protein-dense structures to bind their targets. Genetically encoded labels are smaller in size. For example, HaloTag is ∼4 nm in size.¹¹ But these labels are limited in orthogonal options for multiplexing, and when used in combination with overexpression, they no longer capture endogenous protein distributions. Genome-edited knockins are a powerful tool but remain challenging to produce. A wider range of nanobodies to specific targets would address some of these issues and improve the precision of DNA-PAINT-based techniques.

Furthermore, PAINT experiments rely on bathing the sample in a reservoir of unbound fluorescent molecules generating background noise. Fluorogenic probes can mitigate this issue by reducing the fluorescence of unbound fluorophore.⁵ Samples that are too thick to be imaged with total internal fluorescence (TIRF) or highly inclined and laminated optical sheet (HiLo) illumination may require confocal or lattice light sheet microscopes for optical sectioning.^12,13 Lastly, we note that imaging throughput remains slow—the 30-channel neuron image required 30 h to capture, despite the massive improvement over the estimated 800 h it would have taken with Exchange-PAINT. In regards to resolution, these methods may be further improved by combining them with resolution enhancement by sequential imaging (RESI) or minimal fluorescence photon fluxes (MINFLUX).^10,14 Such tools will be powerful for discerning the complex heterogeneity of biological processes in different contexts. One can imagine a future where large spatial proteomic datasets become a valuable open science resource, allowing researchers from around the world to investigate interactions in common cell types, thus providing the basis for exciting insights.