Completing the canvas: advances and challenges for DNA-PAINT super-resolution imaging

Abstract

Single-molecule localization microscopy (SMLM) is a potent tool to examine biological systems with unprecedented resolution, enabling the investigation of increasingly smaller structures. At the forefront of these developments is DNA-based point accumulation for imaging in nanoscale topography (DNA-PAINT), which exploits the stochastic and transient binding of fluorescently labeled DNA probes. In its early stages the implementation of DNA-PAINT was burdened by low-throughput, excessive acquisition time, and difficult integration with live-cell imaging. However, recent advances are addressing these challenges and expanding the range of applications of DNA-PAINT. We review the current state of the art of DNA-PAINT in light of these advances and contemplate what further developments remain indispensable to realize live-cell imaging.

Super-resolution imaging

Over the past decade insights in wave optics have enabled the development of fluorescence super-resolution microscopy, granting researchers the ability to image with a resolution beyond the diffraction limit [1]. The high resolution allows structures to be visualized at the molecular scale, thereby unraveling the complexity of cells. Super-resolution imaging can be achieved by a variety of techniques, including stimulated emission depletion (STED) [2], photo-activated localization microscopy (PALM) [3,4], and stochastic optical reconstruction microscopy (STORM) [5], which rely on a universal working principle – namely limiting the number of simultaneously emitting fluorophores within a diffraction-limited region. Although many fluorophores may be present within a nanometer-sized sample, only a few are excited at a given moment. This restricted excitation and identification can be realized in two distinct manners: reversible saturable optical fluorescence transitions (RESOLFT) [6] and SMLM (see Glossary) [7], although recently they have been combined into a single method [8].

Glossary.

Association rate (on-rate,k_on)

the number of times a particular imager strand binds to a docking strand per second per mol. A typical ~8 nt DNA imager strand has a k_on of ~2 x 10⁶ M⁻¹ s⁻¹. This scales with the affinity of an imager strand for its docking strand and the number of binding sites on a docking strand

Binding frequency (f_b)

the number of times a target molecule hybridizes with an imager strand per second; f_b is inversely proportional to the unbound time τ_u, the average time between subsequent binding events on a single target molecule, $f_{\text{b}}=\frac{1}{{\tau }_u}=k_{\text{on}}.C$

Binding time (τ_b)

the average duration for which an imager strand remains hybridized with a docking strand, ${\tau }_{\text{b}}=\frac{1}{k_{off}}$ Generally τ_b $({\tau }_{\text{b}}=\frac{1}{k_{off}})=\sim 2.$

Dissociation rate (off-rate, k_off)

the rate at which an imager strand dissociates from the docking strand; k_off is inversely proportional to the binding time

DNA-based point accumulation for imaging in nanoscale topography (DNA-PAINT

a single-molecule localization microscopy (SMLM) method that relies on repetitive and transient binding of fluorescently labeled DNA imager strands to their complementary docking strands that are conjugated to a molecular target. The attainable resolution is <5 nm

Docking strand

a DNA sequence that serves as the landing site for the imager strands and is attached to the point of interest in DNA-PAINT imaging. A single docking strand can have multiple imager strand binding sites

Fluorogenic probe

an imager strand that only emits fluorescence when hybridized with a docking strand and not while freely diffusing in solution

Förster resonance energy transfer (FRET)

event in which a donor fluorophore in the excited state transfers energy to an acceptor fluorophore via dipole–dipole coupling. The typical range is 1–10 nm. The efficiency equals the acceptor intensity divided by the sum of the intensities of both donor and acceptor

Imager strand

a fluorescently labeled DNA sequence (~8 nt) that is complementary to the docking sequence and transiently binds to it

Localization

a datapoint consisting of one or several frames that is used to determine the center position of a fluorescence signal through Gaussian fitting

Localization precision

a metric to quantify the deviation in the estimated position of multiple subsequent localizations of a single fluorescent molecule. This metric scales with the square root of the number of photons

Multiplexing

the concept of probing various distinct targets in a single experiment while having the capacity to distinguish the signal from each

Permissive strand concentration (c)

the maximum concentration of fluorescently labeled imager strands at which the signal-to-noise-ratio (SNR) is still sufficient to achieve super-resolution. A typical value in conventional DNA-PAINT is 10 nM

Photoswitchable fluorophore

fluorescent dye that can cycle between a dark state and a bright state, where excitation and fluorescence emission are only possible in the latter state. Photoswitching is typically induced by illumination with another wavelength than the excitation wavelength

Single-molecule localization microscopy (SMLM)

a classification of various super-resolution methods such as PALM, STORM, and PAINT. Superresolution is achieved by spatiotemporal separation of fluorescence emission of single fluorophores, which allows Gaussian fitting to each single molecule, thereby drastically reducing the uncertainty in fluorophore position

The key concept to achieve super-resolution with SMLM is the switching of fluorescent probes between on and off states, such as in PALM [3,4] and STORM [5] (Box 1). An alternative approach for SMLM is point accumulation for imaging in nanoscale topography (PAINT) [9], in which repetitively and transiently binding fluorescent probes are used. In the first demonstration of PAINT in 2006, a lipophilic stain bound nonspecifically to the membrane of large unilamellar vesicles (LUVs) [9]. Recently, DNA nanotechnology has revolutionized PAINT imaging via DNA-based PAINT (DNA-PAINT) [10]. DNA-PAINT uses short fluorescently labeled oligonucleotides that can bind transiently to their complementary labeled targets to achieve blinking.

Box 1. Principles of super-resolution microscopy.

The class of reversible saturable optical fluorescence transitions (RESOLFT) represents the deterministic avenue of super-resolution microscopy and relies on controlled and selective illumination of many fluorophores within a region. Stimulated emission depletion (STED) is a prominent example of this approach and uses a depletion laser to selectively suppress excited fluorophores on the edge of a region of interest. Upon illumination with the excitation laser, signal is collected only from the non-depleted fluorophores in the center of the region [2]. This reduces the effective point-spread function of the laser below the diffraction limit to achieve super-resolution. Although STED has the benefit of being compatible with conventional fluorophores, complex illumination setups are required [2].

Single-molecule localization microscopy (SMLM) uses widefield illumination and relies on stochastic cycling between the bright and dark states of the fluorophores. By ensuring that a sufficiently small fraction of the molecules is in the bright state at each moment, emission is collected from a single molecule within each diffraction-limited area. This subsequently allows high-precision fitting, thereby achieving a highly accurate localization for each single molecule. The cycling of fluorophores can be achieved by using photoswitchable fluorophores, a principle that underlies photoactivated localization microscopy (PALM) [3,4] and stochastic optical reconstruction microscopy (STORM) [5]. For these techniques, the blinking behavior of photoswitchable fluorophores is controlled through a low-level or pulsed activation beam [3–5]. The localization precision of a single molecule increases with the number of detected photons $(\sigma =\frac{1}{\sqrt{N}})$ [82], while the number of blinking cycles a single fluorophore can undergo is limited.

The early days of DNA-PAINT primarily focused on bringing the resolution down to the molecular level, and, having achieved this, recent developments have improved other aspects of the technique while exploring the plethora of potential applications (Figure 1A). For example, DNA-PAINT has been implemented to measure piconewton forces in living cells, allowing the simultaneous quantification of mechanical force and visualization of cellular structures, thereby bridging the gap between structural biology and mechanobiology [11]. It is also increasingly being used in the medical realm as both a more accurate and versatile tool to monitor biomarkers for disease diagnosis [12], and to study patient histology at the highest detail [13]. Preceding these applications is a wide range of ongoing developments that greatly expand the versatility, applicability, and ease-of-use of DNA-PAINT. As novel multiplexing strategies and advances in acquisition speed are addressing a crucial limitation of lengthy acquisition times, progress in labeling probes and alternative PAINT methods also pave the way for live-cell imaging. We discuss these advances and contemplate remaining challenges before the DNA-PAINT canvas is completed and live-cell imaging can be realized.

Figure 1. The concept of DNA-PAINT.

(A) Timeline indicating the three main phases of the DNA-PAINT field: the development and improvement of the technique, recent advances that optimize and functionalize DNA-PAINT, and future progress for novel applications and live-cell imaging. (B) Transient binding of short dye-labeled DNA oligonucleotides (imager strands) to the complementary target sequence (docking strands) causes an increase in fluorescent signal (ON) and is detected as a localization event. (C) Computational simulation comparing diffraction limited imaging (left) and DNA-PAINT super-resolved imaging (right) of DNA origami nanostructures. The DNA origami was designed to have 12 docking sites arranged in a 20 nm grid pattern (inset in the DNA-PAINT image). The simulation was performed with Picasso Software [16]. Scale bars are 100 nm.

Single-molecule localization microscopy with DNA-PAINT

DNA-PAINT uses base-pairing between short fluorescently labeled DNA oligonucleotides [10]. A target is labeled with a short DNA docking strand while the complementary, fluorescently labeled imager strands diffuse freely in solution. Upon hybridization, an increase in fluorescence intensity is observed (ON) for several hundreds of milliseconds, after which the imager strand unbinds (OFF) and leaves the docking strand unoccupied (Figure 1B). As imager strands bind and unbind, the pool of imaged fluorophores is continuously replenished, eliminating concerns over the photon-budget in DNA-PAINT. In recent work, however, photo-induced depletion of docking strands has been observed [14], which implies that the binding and unbinding cycle of imager strands is finite. Furthermore, DNA-PAINT allows high target specificity and programmability because the length and sequence of imager strands can be tuned [15,16]. Another advantage over other SMLM methods is that the fluorophore choice is unrestricted because photoswitchable fluorophores are not required. These advantages have expanded the use of DNA hybridization beyond the field of DNA-PAINT to other imaging methods such as spectroscopy [17], STED [18–20], structured illumination microscopy (SIM) [19,20], and STORM [19]. Likewise, relying on DNA hybridization to measure colocalization has allowed the determination of target proximity unconstrained by the optical resolution [21–23].

Early developments of DNA-PAINT to improve both the localization precision of single molecules and the signal-to-noise ratio (SNR) have allowed discrete molecular imaging with <5 nm spatial resolution [24] (Figure 1C). Despite the high resolution of DNA-PAINT, the quantification of the absolute number of target molecules remains a challenge, especially in densely packed clusters. An attractive approach for the quantification of these complexes is quantitative DNA-PAINT (qPAINT). qPAINT relies on the predictability of DNA hybridization, since the imager strand association rate (on-rate, k_on) linearly increases with the number of docking strands, thereby reporting on the number of molecular targets within a region of interest [25]. For conventional qPAINT, a calibration step is required, and this might not be possible in complex heterogeneous biological samples. Recently, calibration was made redundant with the development of localization-based fluorescence correlation spectroscopy (lbFCS) which employs a post-imaging algorithm that is capable of autocorrelation [26].

Advances in visualizing distinct species through multiplexing with DNA-PAINT

Novel advances have focused on multiplexing which is the visualization of multiple distinct molecular species within a single sample [27–31]. The number of dimensions through which multiplexing is achieved has recently expanded to include sequence, kinetic, and spectral barcoding.

In sequence-based multiplexing, orthogonal sequences are employed to label distinct targets, and this strategy is the working principle of Exchange-PAINT [28]. With sequence multiplexing, the level of multiplexing is only limited by the number of orthogonal sequences that can be designed. However, because only a single type of fluorophore is used, and pseudocolors are assigned to each orthogonal DNA sequence, imaging must take place in sequential imaging cycles (Figure 2A). The acquisition time thus scales with the number of structures, making the imaging of a large number of structures in a single sample a long process. To reduce the time between sequential imaging cycles, the washing step can be eliminated by adding 'quencher' strands before each new imaging round. These quencher strands are complementary to the imager strands from the previous round, and, upon hybridization, the quencher strands prevent binding to the target and eliminates background fluorescence [32].

Figure 2. Multiplexing with DNA-PAINT.

(A) In sequence-based multiplexing, different targets within a sample can be imaged sequentially. Each point of interest (POI) in a target sample is labeled with a unique docking sequence (1, 2, ..., N) and, in a first round, the imager strand for POI 1 is introduced. After obtaining sufficient localizations, the imager strand for POI 1 is washed away and the next imager strand can be introduced. This cycle can be repeated for N number of cycles, and pseudocolors are assigned to each imaging round. (B) Kinetic multiplexing can achieve its discernibility by varying the length of the hybridized duplex and the subsequent difference in the dissociation rates and thereby binding times of the imager strands (top). Alternatively, by having a distinct number of binding sites in a docking strand, the difference in binding frequency (bottom) adds another layer of multiplexing. (C) Spectral multiplexing requires either orthogonal imager strands that are each labeled with a unique fluorophore to probe various targets in parallel (top) or a varying distance between a donor and acceptor Förster resonance energy transfer (FRET) pair that results in a different FRET efficiency (bottom).

Effort has been made to develop alternative multiplex approaches that can detect multiple species in a single round of imaging. One such approach is kinetic fingerprinting, which is able to probe different species simultaneously. Multiplexing is achieved by varying both the binding time and binding frequency for different species (Figure 2B). Whereas the former is tuned by the number of basepairs that are formed between docking and imager strand, the latter is modulated by the number of binding site repeats on a docking strand. The two approaches can thus be varied combinatorially. The concept was demonstrated with fourfold multiplexing; however, to achieve higher levels of multiplexing, different dyes had to be integrated. This allowed 124-fold multiplexing on DNA origami constructs [31], but reaching this level of multiplexing requires up to 44 sequence repeats, and this might not be possible in more complex systems where the labeling efficiency is lower and the number of labeling sites is limited.

A third dimension of multiplexing exploits the spectral properties of dyes. Multiplexing by using different dyes is the most easily implemented approach (Figure 2C, top panel), but it is inherently limited by the number of distinguishable dyes. To minimize color crosstalk, the dyes are typically excited sequentially at different wavelengths. The number of excitation lasers required for spectral multiplexing was recently reduced by frequency modulation, allowing the detection of five different dyes [33].

To circumvent this constraint and still allow spectral multiplexing, Forster resonance energy transfer (FRET) between a donor and an acceptor fluorophore has been used in correlative FRET multiplexing. By varying the position of the donor fluorophore on the imager strand, the separation between the dye pair will alter, and different FRET efficiencies (E) will be obtained (Figure 2C, bottom panel). However, since FRET efficiency is bounded between 0 and 1, so far only 3-FRET efficiencies could be distinguished [29,30].

The kinetic and spectral multiplexing strategies have enabled the detection of several targets of interest in a single imaging round, thereby increasing the imaging speed compared to conventional Exchange-PAINT. For these approaches, however, the overall low binding frequency is still an intrinsic limitation. In the next section we discuss the most recent advances in acquisition speed that bring the acquisition time for super-resolution imaging with DNA-PAINT down from multiple hours to only a few minutes.

Advances in the acquisition speed of DNA-PAINT

A long acquisition time, rooted in the requirement to collect many photons to pinpoint the center location of a fluorophore, is a fundamental limitation of all SMLM techniques [34–36]. Because distinct targets within a diffraction-limited region should not blink simultaneously to be super-resolvable, each individual docking position is unoccupied most of the time, causing an acquisition time up to several hours [10,16,24,37]. The acquisition time of DNA-PAINT is affected by the number of required localizations, the number of docking positions within a diffraction-limited area, and the binding frequency.

DNA-PAINT uses imager strands ~8 nt in length which have an association rate of ~2 x 10 ⁶ M⁻¹s⁻¹ under standard DNA-PAINT experimental conditions [10], but this parameter has a wide range depending on buffer composition, strand length, and sequence. The freely diffusing imager strands contribute to background intensity, thus their concentration (c) is limited by the minimal required SNR and their concentration typically varies between 0.5 and 10 nM [15]. Recent advances have focused on accelerating image acquisition through increasing the binding frequency (f_b) of a target molecule either by enhancing the permissive strand concentration or by increasing the association rate of individual imager strands (f_b = concentration x k_on).

Increasing the permissive imager strand concentration

The constraint on imager strand concentration originates from the fact that the imager strands are not fluorogenic probes; that is, both the hybridized and the freely diffusing imager strands emit photons, with the latter increasing background signals. Acceleration methods for DNA-PAINT have focused on alleviating this concentration constraint by reducing the detected background intensity. To this end, approaches have been designed in which fluorescence from freely diffusing imager strands is not detected, either through various implementations of FRET or photoactivation [38–40].

In FRET-PAINT, for example, donor labeled imager strands bind to an acceptor labeled docking strand, allowing energy transfer between them (Figure 3A, left panel). By detecting only the acceptor fluorescence, donor labeled imager strands do not contribute to background signal and their concentration can be increased to 1200 nM, consequently reducing the acquisition time to less than 1 minute [40].

Figure 3. Methods to increase the binding frequency of DNA-PAINT.

(A) Conventional DNA-PAINT suffers from a comparably high fluorescence background signal from the imager strands in solution, and this limits their maximum concentration. Approaches that reduce background signal can thus increase the permissive concentration of imager strands, thereby causing an equal acceleration in binding frequency and acquisition speed. FRET-PAINT (left) blocks donor emission from the imager strands in solution and only detects acceptor emission. In fluorogenic DNA-PAINT (middle) a quencher is fused to the imager strand to quench fluorescence signal in solution; upon binding, the imager strand linearizes and fluorescence signal can be detected. Photoactivatable DNA-PAINT (right) uses photoswitchable fluorophores that are in the dark state while in solution and become activated only upon the UV illumination (purple) near the surface. (B) The association rate at which a particular imager strand binds to a target molecule can be increased by electrostatic screening, for example by increasing the magnesium concentration in the buffer (left) or by increasing the number of bindings sites in a docking strand (middle). The rate also increases as secondary structures in the imager strand are removed through sequence design in which complementary bases are avoided to prevent self-interactions (top right), or by Ago-PAINT which reduces the entropic barrier of hybridization through preforming the imager strand (bottom right). (C) Comparison of the various acceleration methods on working principle, acceleration performance, and compatibility with multiplexing approaches. Acceleration is defined as relative to conventional DNA-PAINT [10], with 1 dot = 1–4×, 2 dots = 5–9×, 3 dots = 10–19×, 4 dots = 20–100×, and 5 dots = >100×. Approaches indicated by an asterisk (*) have been integrated for up to 100x acceleration [36]. For an acceleration method to be compatible with a given multiplexing approach, both the acceleration and the level of multiplexing must be as high as when used separately.

In another scheme, fluorogenic DNA-PAINT employs imager strands that contain a dye and a matching quencher linked to opposite ends of a single imager strand [41] (Figure 3A, middle panel). In the unbound state, the imager strands coils, placing the dye and quencher in close proximity and causing quenching of fluorescence signal. However, when bound to the docking strand, the imager strand is linearized and fluorescence emission is detectable. Using this design, the probes become practically fluorogenic and the associated increase in permissive imager strand concentration accelerates image acquisition by 26-fold.

Lastly, with photoactivatable DNA-PAINT, imager strands are chemically reduced, and activation with UV illumination is required before photon emission [42], effectively integrating the concept of photoswitchable fluorophores (Figure 3A, right panel) with DNA-PAINT [3–5]. Through evanescent illuminations, only the imager strands that are close to the surface – those that are hybridized with a docking strand – are activated, alleviating background signal and allowing a higher imager strand concentration.

Increasing the imager strand association rate

The imager strand association rate is largely dependent on the sequence of the imager strand because the freely diffusing imager strands can coil up into secondary structures, which is one of the main causes of their comparatively low association rates. A fivefold speed increase has been accomplished by optimizing the imager strand sequence [46] (Figure 3B). First, the formation of secondary structures that decelerate binding was prevented by refraining from including complementary bases within an imager strand. In addition, the sequence was chosen such that the free energy of the hybridized duplex resulted in optimal binding times, which were as short as possible, to limit simultaneous binding in dense regions, but long enough to collect sufficient photons. Alternatively, by varying the ethylene carbonate concentration in the buffer, the probe dissociation rate (off-rate, k_off), and thereby the binding time, can be tuned by an order of magnitude [47]. Buffer composition can also affect the imager strand association rate through increased electrostatic screening and variations in the magnesium concentration allow a twofold change [46]. In addition, the approach of protein-assisted DNA-PAINT [48] reduces the entropic barrier of hybridization by preforming the imager strand. The concept was first demonstrated with the Argonaute (Ago) protein – a naturally occurring protein that uses an RNA or DNA guide strands to bind to complementary RNA or DNA targets [49]. Ago-assisted DNA-PAINT (Ago-PAINT) can facilitate a 10-fold acceleration and has the major advantage of being sequence-independent.

Optimization of the docking strand sequence also increases the imager strand association rate, which was shown recently when a docking strand with repeated DNA binding site motifs produced a predicted 100-fold acceleration [36]. The increase in the number of binding sites on a single docking strand led to an equal linear increase in binding frequency and acquisition speed, and the concept has been verified repeatedly [31,47,50]. Furthermore, by using periodic binding motifs with partial overlap, the required docking strand size was minimized. Although one might expect the repeated and spatially distinct binding motifs on a single molecule to blur signal and reduce resolution, this has been shown not to be the case [50].

A second method how docking sequence design can increase the imager strand association rate is by including a spacer between the target and the binding sequence [47]. Incorporating a polymer spacer has been shown to increase the association rate by 60%, and this has been ascribed to a reduction in steric hindrance between the imager strand and the target molecule [47]. Figure 3C compares the relative acceleration of different techniques and their compatibility with multiplexing approaches. Several approaches have been integrated to accelerate acquisition speed in a synergistic manner [36,47], and we anticipate that new combinations will further reduce the acquisition time.

All speed-optimization approaches discussed here increase the binding frequency, leaving room for improvement of other aspects governing the acquisition time. Notably, the duration and number of localizations can be optimized by increasing fluorophore brightness [43], reducing fluorophore bleaching [44], and developing more advanced analysis algorithms [45].

Next-generation PAINT probes towards live-cell imaging

Despite tremendous advances in the field of DNA-PAINT, the sensitive and dynamic environment makes high-resolution imaging in living cells very challenging. In addition, the extended total acquisition time restricts the range of biological samples that can be measured. Therefore, state-of-the-art DNA-PAINT concepts (Figure 4) are instead typically validated in more controllable fixed environments on cellular structures, such as microtubules, mitochondria, and nuclear pore complexes [28,36,40,41,46,51–54]. Furthermore, nucleic acids are routinely visualized in fixed cells with fluorescence in situ hybridization (FISH), but the long sequences required to attain a given target specificity are difficult to unite with the transient binding required for DNA-PAINT [55]. Recently, DNA-PAINT has been used to visualize, multiplex, and quantify short RNA (sRNA) fragments with 10 nm resolution inside fixed cells [56]. To ensure sufficient specificity despite these short targets, the incorporation of locked nucleic acid (LNA) bases was vital because this increases stability, specificity, and hybridization efficiency [57]. The static environment in fixed cells ensures that labeled targets are immobile over the course of an experiment. In addition, the membrane is permeabilized, and this eases the removal of interfering proteins and oligonucleotides and allows the introduction of staining molecules. Live-cell imaging does not benefit from these simplifications, and nucleic acid imager strands may be rapidly degraded after they are successfully introduced inside the cell via perfusion. These challenges have until recently limited live cell imaging with DNA-PAINT to surface proteins [51,58].

Figure 4. Probe design to label cellular target molecules for (DNA-)PAINT imaging.

Super-resolution imaging of cellular target structures requires labeling with DNA docking strands. DNA docking strands are attached to antibodies/nanobodies, affimers, or SOMAmers (slow off-rate modified aptamers) and are introduced into fixed cells to label cellular targets. Alternatively, a protein or peptide backbone, rather than a DNA backbone, may be used to create the blinking events for PAINT imaging. Short peptide docking sequences are conjugated to an antibody in Peptide-PAINT and introduced into the cell, or are intracellularly expressed with LIVE-PAINT.

An additional complication is that an arbitrary DNA imager strand will have thousands of complementary binding sequences with cellular DNA and RNA, which results in an abundance of false positives and elevated background levels, thus reducing resolution. These challenges are surmounted by the recent approach using left-handed DNA (L-DNA) for transient binding instead, which is non-natural and thus cannot hybridize with cellular nucleic acids [59].

In fixed cells, DNA docking strands could be successfully linked to antibodies that bind to intracellular targets, or genetically fused tags [60]. However, the use of an antibody resulted in a link-age error – the distance between position of the fluorophore and the actual target position [61] – of at least 10 nm [62–65] (Figure 4, left panel). This error not only introduces a localization bias but also reduces the maximum labeling density owing to steric hindrance, and possibly impedes imaging of denser cellular structures that are impermeable for the probe [61]. Nanobodies (single-domain antibodies) do not suffer from these problems and have reduced the linkage error to 4 nm while achieving a resolution of 20 nm on various organelles in fixed cells [66]. However, the number of nanobodies that have sufficient affinity for endogenous proteins is limited, making this approach challenging. Therefore, proteins of interest must be genetically tagged with epitopes that can be recognized by the available nanobodies [54,66].

The need for simpler labels that do not require genetically encoded protein tags has pushed the development of affimer labeling [67] and slow off-rate modified aptamers (SOMAmers) [51], both of which use small (<30 kDa) target-specific probes to which a DNA docking strand is attached. Affimers are naturally occurring proteins that have been screened for target protein affinity and that have a DNA docking strand fused to one of their amino acids [68] (Figure 4, middle left panel). Their use has been validated on intracellular targets in live cells, yielding ~15 nm resolution [67]. By contrast, SOMAmers are DNA structures that contain a region of modified bases with hydrophobic residues to increase target affinity and specificity (Figure 4, middle right panel). SOMAmers have achieved an impressive resolution of ~8 nm, but the limited ability of SOMAmers to bind to intracellular targets might confine future applications [51]. Both affimers and SOMAmers are limited by unpredictable target-binding affinity, necessitating laborious high-throughput screening and selection to find suitable probes.

A forthright approach to overcome some of the mentioned challenges inherent to DNA oligonucleotides is to replace them with proteins or peptides (Figure 4, right panel). In protein-PAINT, synthetic cell-permeable fluorophores are added extracellularly and, upon cell entry, the fluorophores transiently bind to genetically encoded protein tags that are fused to target proteins [69]. More recently, the heterodimeric E/K coiled-coil peptide pair has been used for transient and tunable binding in vitro with peptide-PAINT, where the docking peptide was conjugated to the target protein via secondary antibodies [70]. This peptide counterpart of DNA-PAINT has a roughly doubled association rate because the electrostatic interactions are less repulsive than for DNA, accelerating imaging acquisition. Furthermore, peptide-PAINT labels more efficiently and has a smaller linkage error because the docking strand is genetically fused to the protein of interest and antibodies or nanobodies are not required. These advances set the stage for live-cell imaging with LIVE-PAINT, which relies on similar peptide–protein interactions [71] (Figure 4, bottom middle). In this approach, the imager peptides were also genetically encoded and endogenously expressed inside living yeast, circumventing extracellular introduction. Owing to the limited predictability and specificity of peptide interactions, these alternative backbones have not yet been widely adopted for PAINT imaging, but this may change in the near future as coiled-coil interactions are becoming increasingly programmable [72].

Challenges for live-cell imaging

Novel variations of DNA-PAINT have expanded the super-resolution imaging toolbox, enabling research in previously uncharted directions. Advances have facilitated a resolution down to the molecular level [24] and spectrally unrestricted multiplexing [27–31]. Although the lengthy acquisition time has traditionally been considered to be the Achilles’ heel of DNA-PAINT, it has now been reduced by several orders of magnitude to the point where super-resolution images can be acquired within several minutes [36,38–42,46–48,50]. If the approaches for speed optimizations perform well inside cells, a crucial obstacle to live-cell imaging will be surmounted. That the underlying SMLM super-resolution concept of DNA-PAINT is in principle compatible with living systems was shown when another SMLM approach, STORM, was used to image living eukaryotic cells [7]. Other important advances have also been made to adapt DNA-PAINT probes to cell imaging. Strategies that use peptides [70,71] or proteins [69] have successfully eliminated problems such as probe introduction and degradation, and intracellular target labeling has been demonstrated with affimer- [67] and aptamer-based [51] approaches. Nevertheless, several outstanding challenges remain.

Two key obstacles to DNA-based imaging inside living cells are the stability of the DNA and potential non-specific interactions with cellular nucleic acids. The photo-induced depletion of DNA docking strands can be minimized by using a lower excitation power [14], by increasing the spacing between the fluorophore and the docking strand [14], or by increasing the number of imager binding sites in a docking strand [50]. Furthermore, increased DNA stability against DNases may be achieved by protecting the imager strands with Ago-PAINT [48], but, to permit live-cell imaging, a smaller, truncated version of the protein might be required [73]. Alternatively, oligonucleotides may be protected through chemical modifications of the DNA, for example through the use of LNA as in sRNA-PAINT [56,74]. The use of LNA also reduces non-specific interactions with other nucleic acids, while L-DNA eliminates any interaction with cellular nucleic acids [56,59].

In addition, the labeling of targets of interest in living cells continues to be one of the biggest challenges in the super-resolution community [31,52,53,56]. Currently, most approaches in fixed cells rely on the use of docking sequences labeled with antibodies that bind specifically to a target protein. However, this may not be suitable for live-cell imaging owing to the challenge of introducing these sizeable antibodies inside the cell. Several chemistry-based approaches have been developed which rely on the incorporation of unique functional groups via unnatural amino acids [75,76] or self-labeling protein tags, thereby reducing the size of the probe that needs to be introduced into the cell (e.g., SNAP [77], HALO [78], and FGE [79] tags). Although most of these labeling methods require engineering of the target molecules and thus cannot be directly applied to unperturbed cells or tissues, we envisage that these strategies will be important for early proof-of-concept experiments. We invite biophysicists and chemists to further develop protocols for efficient and specific labeling strategies to boost super-resolution for live cell imaging.

Another hurdle for live-cell imaging is the variation in cellular content (e.g., protein concentration) among different cells in a single sample, which precludes uniform and up-front labeling of cellular targets. One outcome might be Action-PAINT [80], in which cellular targets are first probed and then labeled after visualization, allowing the labeling to be tuned to the composition of each individual cell. In Action-PAINT, the imager strands in the first round are chemically modified such that they can be rapidly crosslinked to a complementary docking strand upon UV illumination [81]. In addition, these imager strands contain a sequence that functions as a new binding site for a sub-sequent round of DNA-PAINT imaging with new imager strands. In this second round, only the user-selected cellular components that were labeled through crosslinking in the first round are imaged with DNA-PAINT. With Action-PAINT, cellular targets can thus be labeled with high-resolution after visualization on a per-cell basis.

Concluding remarks

The field of DNA-PAINT has seen tremendous advances in multiplexing, acquisition speed, and resolution in vitro; however, it will be challenging to achieve the same performance in living cells (see Outstanding questions). We envisage that live-cell imaging with DNA-PAINT will first be demonstrated in its most primitive form. Once a capable methodology has been developed, we expect that more sophisticated concepts, such as multiplexing and quantitative analyses, can be implemented with relative ease. These concepts are subject to the same barriers as conventional DNA-PAINT and have already been demonstrated in fixed cells. As soon as live-cell imaging with DNA-PAINT becomes a routine experiment, elemental aspects such as the dynamics of intracellular protein localization and protein interactions might be addressed, and the cellular concentrations of proteins and nucleic acids may be quantified in real time with super-resolution, answering fundamental questions about the rate and regulation of translation and transcription.

Outstanding questions.

Three main approaches for multiplexing with DNA-PAINT have been developed that distinguish targets by sequence, binding kinetics, or spectral emission. Combined, a level of multiplexing on the order of 100 has been reached. What distinguishing characteristics will newer multiplexing strategies underpin, and how many fold multiplexing will these strategies allow?

Several affinity-based approaches have been developed to label a target molecule with a docking strand, relying on antibodies, nanobodies, SOMAmers, or affimers, but finding such probes necessitates laborious screening. How will the increased predictability and programmability of protein interactions affect the development of probes for DNA-PAINT? Will we be able to rationally design these probes for high labeling efficiency and minimal off-target binding?

In living systems, imager strand degradation and non-specific binding hinder quantitative analyses, reduce resolution, and produce false localizations. What new synthetic constructs such as LNA and L-DNA will be designed to address these challenges?

DNA-PAINT methods are primarily used in static environments, but, as the challenges inherent to the imaging of dynamical systems are overcome, we will gradually approach live-cell imaging. To what extent could one use DNA-PAINT to also observe dynamic events in the cell, and what would its spatiotemporal resolution be?

Genomics, transcriptomics, and proteomics have revolutionized the way in which biological science is carried out. Will live-cell imaging with DNA-PAINT similarly transform the field of structural biology and cellular (co)localization?

DNA-PAINT has found applications in the medical realm as a detector of bio-markers and as a nanoscale instrument to measure mechanobiological forces. Relying on the advances in recent years, what other applications will future variations of DNA-PAINT give rise to?

Highlights.

In recent years the performance, utility, and ease-of-use of DNA-based point accumulation in nanoscale topography (DNA-PAINT) have greatly improved by increasing compatibility with existing technology and diversifying the usable probes.

Multiplexing with DNA-PAINT has become possible, which allows users to probe many targets simultaneously and paves the way for high-throughput methods.

Advances have alleviated the limiting factors that govern the binding frequency, thereby accelerating imaging acquisition and enabling DNA-PAINT imaging to be performed in several minutes.

The complex environment in living cells creates numerous challenges for standard DNA-PAINT imaging. These challenges are being addressed through improved probe design such as using modified DNA nucleotides or amino acid-based backbones for the imager and docking probes.