Probe choice in single-molecule microscopy requires deeper evaluations than those adopted for less sensitive fluorescence microscopy studies. Fluorophore characteristics can alter or hide subtle phenomena observable at the single-molecule level, wasting the potential of the sophisticated instrumentation and algorithms developed for this kind of advanced applications. The three typical groups of fluorophores are fluorescent proteins, organic dyes and quantum dots; here their advantages, drawbacks and use in single-molecule microscopy are discussed. Some requirements are common to all applications, such as high brightness and photostability, specific and efficient labeling, controlled stoichiometry, no perturbation on the system. Other requirements depend on the specific type of single-molecule technique; some of them are here described with their specific requirements for probe choice.
Three important classes of single-molecule techniques are single-molecule FRET (smFRET), single-molecule tracking (SMT, also called single-particle tracking (SPT)), and single-molecule localization microscopy (SMLM) (Figure 1).
FRET is a powerful approach for studying intramolecular and intermolecular interactions (Figure 1A). At the single-molecule level, it allows observation of rare states and characterization of detailed kinetics. FRET consists in an energy transfer from an excited fluorophore (the donor) to another one (the acceptor) when they are close enough (typically in the nanometer range). A spectral overlap between donor emission and acceptor absorption is necessary for an efficient transfer. Other requirements are: low cross-talk between the channels (low donor emission leakage in the acceptor emission window and minimum direct acceptor excitation), high photostability (low photobleaching and intensity fluctuations), and, mostly in the single molecule case, similar quantum yields and detection efficiencies of donor and acceptor [1][2].
SMT consists in the reconstruction of the trajectories of single molecules (Figure 1B). An essential requirement is the ability to follow the molecules for enough time, so the probe must be resistant to photobleaching. Moreover, intensity fluctuations are not desirable because they cause interruptions in tracks. An accurate track reconstruction also requires a high time resolution: higher SNR allows shorter integration times per frame and a larger number of sample points for reconstructing tracks [3][4].
In SMLM, individual fluorescent molecules are localized with resolutions of the order of nanometers or tens of nanometers, starting from diffraction-limited images, thanks to the activation of small subsets of molecules per cycle [5][6] (Figure 1C). The fundamental requirement is a mechanism that ensures stochastic transitions between “Off” and “On” states, which must lead to sparse activations in order to have only one On-dye in a diffraction-limited area. This requires a low On/Off duty cycle (defined as the fraction of time a fluorophore spends in the On state) and long-lived dark states. Other requirements are high photon counts per switching event and generally a high number of switching cycles [7][8].
Figure 1. Single-molecule applications in microscopy. (A) smFRET. Top: example of conformational study: donor (D) and acceptor (A) probes label two sites of the same molecule. When the sites are distant, no FRET occurs: under D excitation, only its emission is detected; when the sites are near enough (typically a few nm), FRET occurs: when D is excited, emission of A is detected. Middle: example of corresponding single-molecule intensity traces. (B) Single-molecule tracking. Positions of labeled molecules are determined at different time points and their trajectories are reconstructed. Examples of tracks reproduced with permission from [9] (© 2013 The Company of Biologists Ltd.), where TrkA receptors were labeled with Quantum Dots. Multimodal tracks include periods of slow, fast and drifted motions. (C) Single-molecule localization microscopy (SMLM). Top, left: in conventional diffraction-limited images, resolution is around 200–250 nm, so that PSFs of close emitters overlap. Top, right: SMLM exploits activation of small subsets of probes at a time to have nonoverlapping point spread functions (PFSs), localize them individually and reconstruct images with 10–50 nm resolution. In the bottom part of each panel, specific probe requirements for each technique are highlighted.
The discovery of the green fluorescent protein (GFP) in the early 1960s revolutionized biological investigations, and in the last 30 years, it has seen extensive use [10][11]. Fluorescent proteins (FPs) from different species and several mutants have expanded the color palette of this kind of fluorescent reporter [12][13]. One of the advantages of FPs is their intrinsically specific labeling with controlled stoichiometry, thanks to the fact that their DNA sequence is genetically fused to the one of the molecule of interest. The only impact of aspecific labeling could arise from the partial degradation of the chimeric protein. Moreover, the system can be used in many contexts, such as cells, tissue, yeasts, and bacteria, with better biocompatibility compared to the other kinds of probes. At the same time, there are some downsides. FPs are less bright and photostable than organic fluorophores or Qdots. They are also larger than organic fluorophores (even if smaller than Qdots), so they are more likely to introduce perturbation in some processes. Indeed, FPs are ~25 kD in size, while organic fluorophores have average sizes typically less than 1 kD [13]. Another critical point is that many FPs tend to oligomerize [14]. This tendency may depend on some experimental conditions, such as concentration, so an assessment of the phenomenon in the context of interest can be required to avoid artifacts [15][16][17][18].
In particular, for smFRET, the most common pair consists of a cyan-emitting variant (cyan fluorescent protein, CFP) and a yellow-emitting variant (yellow fluorescent protein, YFP) [19]. However, some limitations are associated with this FRET pair and multiple factors must be considered to allow quantitative measurements: overlap of donor and acceptor emissions, dependence of YFP properties on environmental conditions (e.g., pH), complex photokinetics such as reversible photobleaching or photoconversion of YFPs into CFP-like FPs, low brightness of CFP, and dimerization tendency [19][20][21]. Advanced versions of CFP and YFP have been developed to improve these factors, such as eCFP, Cerulean, mTurquoise, mCerulean3, mTFP1, Aquamarine for the first; EYFP, mVenus, mCitrine, sEYFP, YPet, for the second [19][20][22][23][24]. These reduced some of the mentioned weaknesses, although only partially. Green donors (GFP and enhanced versions eGFP, Clover, mNeonGreen) with red acceptors (such as mCherry, monomeric DsRed, RFP or mRuby) have been proposed and used as well [18][21][25]. This approach allows for lower induced autofluorescence, less phototoxicity, greater separation of spectra and resistance to environmental changes [20][22]. However, red FPs typically have low brightness, with a FRET emission that can be too weak even compared to the donor emission tail [22]. Recently, there have been efforts to develop new versions of red-emitting FPs to make up for the lack of bright and stable red fluorescent proteins, especially for FRET applications [26][27][28].
Live-cell SMLM was enabled by the development of photoactivatable fluorescent proteins, which are still widely used in particular for photoactivated localization microscopy (PALM) [29][30][31][32]. Employment of FPs was also favored by the ease of performing intracellular labeling in live cells, not possible with labeled antibodies or with many membrane-impermeable organic dyes and dye conjugates [33][34]. Several types of FPs suitable for SMLM have been engineered, including irreversible photoactivatable FPs (PA-FPs), photoconvertible or photoshiftable FPs (PS-FPs) and reversible PA-FPs (also called photoswitchable FPs) [5][35][36][37][38]. A summary of the photoswitching properties of selected FPs and an overview of their SMLM applications are available in [39].
FPs have been employed in SMT as well [40][41], even if their limited brightness and photostability usually lead researchers to prefer other kinds of probes for better SNR, time resolution and total observation time, especially for studies on molecules with faster diffusion. FPs are used in a specific type of SPT based on photoactivated localization microscopy, sptPALM, where the trajectories of a few photoactivated FPs are followed [32][42][43][44][45][32, 42–45].
Organic dyes, in comparison to FPs, have higher brightness, less photobleaching, wider available spectral range and smaller size [46]. The main weaker aspect compared to FPs is that labeling is not genetically encoded, so it requires additional steps such as chemical coupling reactions; these can be laborious, require controls of labeling efficiency and lead to aspecific labeling and interactions [47][48][49]. Extensive washing is usually necessary to try to remove unbound dyes, but this must be mild for preserving cell adhesion and functioning, so it is usually not fully effective. Moreover, the time needed for washing makes time-dependent assays difficult.
Labeling can be performed using antibodies chemically conjugated with fluorescent dyes (direct or indirect immunostaining [50]). One of the main limitations of this labeling method arises from the relatively large size of antibodies (150 kDa, ~10 to 15 nm), which nullifies the advantage of organic dyes’ small size. To reduce the possible hindrance, it is possible to use smaller Fab (fragment antigen-binding) fragments (55 kDa, ~5 to 6 nm), or nanobodies (15 kDa, ~4 nm) [51][52][53]. Fab fragments have been used for single-molecule fluorescence microscopy in living cells, e.g., in labeling CD59 receptors with Alexa 647 [54] or CD36 receptors with Cy3 [55].
Another class of strategies to perform fluorescent labeling with organic dyes is based on the genetic insertion of a protein/peptide tag in a protein to be labeled. The tag is recognized and bound by specific chemical motifs, which are bound to or are part of the dyes. A specific and strong conjugation between the dye and a short peptide tag is often achieved by exploiting enzymes that catalyze the covalent bond between them or by exploiting click chemistry. Extensive reviews about protein/peptide tags useful for fluorescent labeling can be found in [56][57][58][59]. The most used approaches, especially for single-molecule imaging, are based on SNAP-tag [60], CLIP-tag [61], HaloTag [62][63][64], and ACP and related tags [65][66] (Figure 2).SNAP-tag is a 20 kDa (30-residue) self-labeling protein tag, engineered to react efficiently with benzylguanine (BG) derivatives, resulting in a covalent link (Figure 2A). Several popular organic dyes suitable for single-molecule applications are commercially available from New England Biolabs (Ipswich, MA, USA), as ready-to-use BG substrates, e.g., Alexa, Atto, or Dyomics dye derivatives. Additional ones can be prepared via a reaction of BG-NH2 (New England Biolabs) with N-hydroxysuccinimide ester (NHS) forms of the dyes (example of protocols is available in [67]), typically commercially available for dyes of practically all companies. SNAP-tag technology can be used for both extra- and intracellular labeling: in the first case, using non-cell-permeable substrates such as SNAP-Surface® (New England Biolabs), for studying, e.g., membrane receptors [68][69]; in the second case, using cell-permeable substrates such as SNAP-Cell® (New England Biolabs) or Janelia Fluor dyes, a quite recent family of cell-permeable fluorophores for single-molecule imaging [70] (protocols for the synthesis of SNAP-tag ligands of Janelia Fluor and subsequent labeling can be found in [71]).
CLIP-tag is another 20 kDa self-labeling protein tag engineered from SNAP as an orthogonal labeling system, accepting benzylcytosine (BC) derivatives as substrates (Figure 2B). By using the two tags, it is possible to obtain simultaneous and orthogonal two-color labeling with two different dyes [72].
Figure 2. Most popular tag-based labeling strategies. (A) SNAP-tag. A benzylguanine—derivative of the label reacts with the self-labeling SNAP-tag fused to the protein of interest (image from https://www.addgene.org accessed on 20 October 2022). (B) CLIP-tag. A benzylcytosine—derivative of the label reacts with the self-labeling CLIP-tag fused to the protein of interest (image from https://www.addgene.org accessed on 20 October 2022). (C) Halo-tag. A chloroalkane—derivative of the label reacts with the self-labeling Halo-tag fused to the protein of interest (POI) (image from [66] used with permission from Promega Corporation), TEV: Tobacco Etch Virus. (D) ACP-tag. The enzyme ACP- (or SFP-) synthase catalyzes the attachment of a Coenzyme A (CoA)-label to the ACP- (or MCP-) tag fused to the protein of interest (image from https://www.addgene.org accessed on 20 October 2022).
HaloTag is a self-labeling protein tag that reacts specifically with chloroalkane (CA)-derivatized fluorophores (Figure 2C). Like SNAP and CLIP, it leads to specific and covalent linkage. It is larger than SNAP and CLIP, being 33 kDa and having 297 residues. Complete protocols for HaloTag-based labeling (generating a HaloTag fusion protein, generating Halo fluorophore ligands starting from their NHS forms, labeling) are provided by Promega [73]. Protocols for single-molecule tracking inside the nucleus of living cells exploiting HaloTag labeling are available in the literature as well [74]. Some ready-to-use dyes for HaloTag labeling are available from Promega (including some for intracellular labeling, e.g., tetramethylrhodamine (TMR), Oregon Green, Janelia Dyes, and others for labeling on the cell membrane, e.g., Alexa 488 and 660).
Another group of tags has been engineered from the acyl-carrier protein (ACP). These are enzyme-mediated labeling tags: phosphopantetheine transferase (PPTase) enzymes catalyze the covalent conjugation of a phosphopantetheine-dye to a serine residue in the tag (Figure 2D). The first developed tags were the peptidyl-carrier protein (PCP) and acyl-carrier protein (ACP, also with its mutant MCP) domains (80–120 residues, ~9 kDa in size and so smaller than SNAP, CLIP and Halo Tag) [65][75][76]. These tags were further optimized to obtain even shorter tags, such as the YBBR (11 residues), S6 (12 residues) and A1 (12 residues) [64][77][78]. In this group of tags, couples allowing orthogonal labeling with two different probes were identified, i.e., ACP-MCP or the shorter tags A1-S6. Orthogonality is achieved because two different enzymes, Sfp and AcpS, have to be used for the labeling [78]. Organic dyes in phosphopantetheine form can be obtained via conjugation with coenzyme A (CoA), obtainable from the maleimide reactive form of the dye reacting with CoA via a thiol-maleimide reaction. Protocols for production of PPTase enzymes AcpS and Sfp, synthesis and purification of CoA–dye conjugates, and expression and labeling of tagged proteins can be found in [79][80]. CoA-conjugated probes are not cell-permeable, so labeling of ACP-derived tags in living cells can be performed only on the cell membrane.
Organic dyes have been widely employed in single-molecule applications. Different couples of organic dyes have been used for smFRET, like the very popular Cy3/Cy5 or Atto 550/Atto 647N, but also Alexa 555/Alexa 647, Atto 532/Atto 647N, Atto 488/Atto 647N, Alexa 488/Alexa 594, Alexa 488/Alexa 555, Alexa 488/Atto 647N [2][81][82][83][84][85][86][87]. Förster radii of couples of Atto dyes can be found at [88], and of couples of Alexa dyes can be found at [89]. Even multicolor smFRET has been achieved with organic dyes [90][91][92][93].
Organic dyes were used in many more SMT experiments than FPs. They allowed tracking a variety of biomolecules on the cell membrane and in different cell compartments, in order to study several processes, among which an established example is diffusion and interactions of receptors [72][94]. Their current levels of brightness and photostability limit simultaneous multicolor studies or registration of very long trajectories (for these purposes, Qdots are more often exploited, as discussed in the following). Tracking duration is typically limited to about 10 s. However, there is a continuous effort to improve the photostability of organic dyes. It is known that their photodegradation can be caused by both oxygen-dependent and oxygen-independent processes and that the triplet state plays a fundamental role, although higher excited states can be involved as well. The exact mechanisms are not fully understood and depend on the specific conditions. Approaches for improving photostability are based on deoxygenation methods or on the use of additives in the medium such as antioxidants, triplet state quenchers, oxidizing and/or reducing agents [95].
Organic dyes are used in SMLM as well, in particular in stochastic optical reconstruction microscopy (STORM) [96]. The principle of STORM is similar to that of PALM, but in this case the photoswitching of fluorophores is random and requires special buffers, which are not suitable for living cells [7][33]. The most used formulations consist of deoxygenated solutions containing a thiol, where deoxygenation is achieved by enzymatic (e.g., via glucose oxidase and catalase, galactose oxidase, Oxyrase) or nonenzymatic (e.g., via methylene blue and a thiol) approaches; different types of thiols have been used (e.g., β-mercaptoethylamine, cysteine, glutathione) [141]. These buffers work for several but not all dyes, and the effects of the different buffer components on the photochemical mechanisms are not well understood yet. Different dyes can require different conditions for optimal photoswitching, either in buffer composition or additive concentrations [39]. A summary of the photoswitching properties of selected organic fluorophores and an overview of their SMLM applications are available in the literature [39]. A recent field of research focuses on the development of spontaneous blinking dyes, which require no special buffers nor additives and not even high excitation power to perform SMLM; therefore, they can allow more flexible and cell-friendlier studies [97][98][99]. Another way to perform SMLM with organic dyes without special buffers is the method of point accumulation for imaging in nanoscale topography (PAINT) [100][101]. Instead of using photoswitching, this technique exploits stochastic and transient binding to the target of freely diffusing probes to visualize small subsets of molecules at a time. This allows more freedom in the dye choice. The most recent approach is DNA-PAINT, where fluorescently labeled DNA oligonucleotides diffuse freely in solution and can bind transiently to targets labeled with complementary DNA strands [102][103][104].
Quantum dots (QDs) have the great advantage of being brighter than organic dyes (and FPs), thanks to much higher extinction coefficients; moreover, they are not affected by photobleaching. The main limitation concerns their relatively large size. Functionalized QDs usually are 15–50 nm in size (diameter), bigger than organic dyes (<1 nm) and even than FPs (about 4 nm) [105][106].
The most used methods for labeling with QDs exploit antibodies or the streptavidin-biotin interaction (Figure 3A,B). QDs conjugated with streptavidin or with different kinds of antibodies are commercially available in ready-to-use formulations (e.g., by Thermofisher).
Figure 3. Labeling of membrane molecules with quantum dots. (A) Example of labeling based on antibodies. Left: a membrane molecule is recognized by a primary antibody, a biotinylated Fab fragment of the secondary antibody binds to the primary antibody, and streptavidin quantum dot (QD) binds to biotin. Right: QD is directly conjugated with the primary antibody. (B) Labeling of biotinylated molecules with streptavidin QDs. Labeling of a phospholipid (left) and a lipid raft component, the ganglioside GM1 recognized by a biotinylated cholera toxin B subunit (CtxB, right), are reported as examples. (A,B) are reprinted from [107] with permission (copyright 2018, Elsevier). (C,D) represent examples of hindrance caused by QD labeling. (C) Visual comparison between YBBR-tagged-NGF (nerve growth factor) labeled with organic dye and biotinylated NGF labeled with streptavidin QD (only backbone represented for NGF and tag, reprinted from [108]). (D) Labeling by QDs causes a slowing down of p75NTR receptors compared to labeling by smaller organic dyes, with a maximum in the reported measured diffusion coefficient distributions that decreases from 0.71 to 0.2 µm2/s (reprinted from [94]).
Alternative strategies for covalent labeling with QDs still need improvement and, in general, are not usually employed.
Since extracellular targets are accessible to QD labeling, many SMT studies exist on various membrane molecules, such as receptors [109][110], ion channels [111][112] and lipids [113][114], resulting in longer tracks in comparison to organic dyes [106]. On the other side, the internalization of QDs and QD conjugates into the cytoplasm of living cells for intracellular imaging is challenging. Several methods have been tested, e.g., electroporation, microinjection, liposome fusion, and cell-penetrating peptides, but these approaches often cause QD aggregation and limited stability within the cytoplasm [115][116].
An intermittent behavior for QD emission (blinking) has often been reported [117], which could cause interruptions in track reconstruction. However, several strategies have been developed to reduce QD blinking, so the phenomenon has been improved over the years [118][119]. Moreover, tracking algorithms can afford the problem of temporary missing spots: especially in conditions of high SNR, as in the case of QDs, they allow for obtaining reliable tracking solutions [120][121].
QDs were particularly preferred over organic dyes for multicolor SMT studies because they allow for easier implementation of the microscopy setup. Indeed, QDs have excitation spectra that increase towards the UV, such that probes emitting in different colors can be excited with one single wavelength; moreover, contrary to organic dyes, QDs have relatively narrow and symmetric emission spectra, without a tail at longer wavelengths, so many different channels can be better separated without spectral overlap [122][123][124][125][126].
However, some studies highlighted perturbations of biomolecule behavior caused by QD labeling, e.g., QD labeling based on biotin–streptavidin conjugation caused a decreased diffusion coefficient for p75NTR receptors compared to labeling with small organic dyes on S6-tag, as observed with SMT on the membrane of living cells [94] (Figure 3D). Hindrance of receptor mobility has also been observed on B cell receptors [127]. A study employing single-molecule imaging and tracking of YBBR-tagged neurotrophins labeled with a small organic dye revealed that their signaling endosomes contain clusters of NGF made of 2–8 dimers [108], while a previous study with streptavidin-QD labeling of biotinylated NGF detected a single NGF dimer per vesicle [128]. The discrepancy has been explained on the basis of the different sizes and steric hindrance of the labeling (Figure 3C): QD labeling introduces a volume up to 70 times larger than NGF; using smaller organic dyes might allow accommodating a physiologically higher number of clustered NGF molecules in the same vesicle [108]. The impact of more cumbersome labeling may depend on the considered system and the measurements of interest. Abraham et al. demonstrated that, even if QDs labeling caused a general slowdown, it allowed detecting large differences in receptor mobility, such as changes caused by destruction of the actin cytoskeleton, on par with organic dyes [127]]. The impact of QD-labeling complexes is likely more evident on small, fast-diffusing and flexible molecules [94], while others can be less or not significantly influenced. Indeed, as mentioned before, several SMT studies have been performed using QD labeling and obtained relevant results [109–113][109][110][111][112][113].
The application of QDs in FRET and smFRET is not as widespread as for organic dyes or FPs [2][105][129][130][2, 105, 129, 130]. The typical configuration involving QDs is based on a QD as donor and a different kind of probe (often an organic dye) as acceptor [131][132]. Indeed, the employment of QDs as donors benefits from their high extinction coefficient for efficient donor excitation, their broad excitation spectra for efficient excitation/emission light separation, and their narrow and symmetric emission for cross-talk minimization. On the contrary, their use as acceptors is very challenging: their high extinction coefficient and broad excitation spectrum make it nearly impossible to avoid direct acceptor excitation upon excitation of the donor fluorophore.
QDs are less used than FPs or organic dyes in SMLM [5]. QDs typically exhibit intrinsic blinking; however, exploiting this property for STORM has proven difficult. Indeed, QDs show a small Off/On time ratio, which makes it hard to create the desired large population of Off-probes. As a result, multiple emitters overlap in a diffraction-limited volume, limiting the ability to localize single contributions [133][134]. Moreover, QDs cannot be photoactivatable/switchable [134].
QDs are the most used inorganic nanoparticles in single-molecule microscopy. Of note, other kinds of nanoparticles have been used or are being developed, especially in SMT. These include upconversion nanoparticles (UCNPs), polymer dots (PDots), fluorescent nanodiamonds (FNDs). These are even brighter and more photostable than QDs. Tracking time ranges from tens of minutes up to hours with localization accuracies of a few nm [135]. A recent study exploited UCNPs to perform single-particle tracking on molecular motors in live cells, achieving 2.4 nm localization accuracy with 2 ms time resolution [136]. However, several challenges still limit the use of these powerful probes, mainly due to large sizes (UCNPs have sizes of 40–50 nm) and complex surface chemistries. As discussed for QDs, these features make it hard to obtain efficient and specific labeling or intracellular delivery and may induce perturbation on some labeled molecules, and these problems are even more severe than in the case of QDs [135].