Switchable Fluorophores for Single-Molecule Localization Microscopy

Abstract

The past decade has witnessed an explosion in the use of super-resolution fluorescence microscopy methods in biology and other fields. Single-molecule localization microscopy (SMLM) is one of the most widespread of these methods and owes its success in large part to the ability to control the on-off state of fluorophores through various chemical, photochemical, or binding-unbinding mechanisms. We provide here a comprehensive overview of switchable fluorophores in SMLM including a detailed review of all major classes of SMLM fluorophores, and we also address strategies for labeling specimens, considerations for multichannel and live-cell imaging, potential pitfalls, and areas for future development.

Graphical Abstract

1. Introduction

Fluorescence microscopy is a workhorse technique in biological research due in large part to its ability to probe the distributions of specific molecules or structures in fixed and living specimens. Despite its widespread use in biology, however, fluorescence microscopy has long suffered from poor spatial resolution that obscures key details of interest. The limited spatial resolution results from the diffraction of light into the far field (i.e., at a distance of many wavelengths from the emitter) and is referred to as the diffraction limit of light microscopy or Abbe’s diffraction limit, in honor of Ernst Abbe. Nearly 150 years ago, Abbe articulated the inverse relationship between spatial resolution and numerical aperture¹, as summarized by the equation d ≈ λ/2NA, where d is spatial resolution, λ is the wavelength of light, and NA is numerical aperture of the microscope objective lens. In practical terms, this means that objects closer than 200-250 nm are unresolved by traditional approaches to light microscopy. Although electron microscopy and atomic force microscopy can achieve much better spatial resolution, these techniques generally have a poor ability to report on distributions of specific molecules, lack compatibility with living specimens, and/or are limited to measurement of surface features.

The 2014 Nobel Prize in Chemistry recognized the development of optical microscopy techniques that circumvent the 200-250 nm diffraction limit of light and that now allow researchers to image distributions of molecules with a resolution of 100 nm or better. The prize was awarded to Stefan Hell for the development of stimulated emission depletion microscopy (STED), to W.E. Moerner for the optical detection of single molecules, and to Eric Betzig for the development of single-molecule localization microscopy (SMLM) which will be the main focus of this review.^2–4 These techniques and related subsequent developments are now widely utilized in studying diverse problems in biology and other disciplines at most major research institutions worldwide, a testament to the ability of transformative methodologies to stimulate discovery and innovation.

Previous reviews have discussed super-resolution methods and/or SMLM in general,^5–13 including some that have provided guides/protocols for SMLM^14–18 and others that have focused on fluorescent probes for SMLM.^19–25 Here, we provide a comprehensive and up to date review of fluorophores and probes that enable SMLM, including a history of the field, a review of the major classes of SMLM fluorophores, strategies and considerations for sequential localization, a discussion of key methods for specimen labeling, a discussion of multichannel and live SMLM, cautionary notes about some potential artifacts or pitfalls, and we also discuss future directions. We use the acronym SMLM, throughout, to acknowledge the work of many groups who have contributed to this field and in consideration that the initial four publications of the field from 2006 used different names/acronyms (i.e., STORM²⁶, PALM²⁷, fPALM²⁸, PAINT²⁹), with subsequent variations introducing many more names to be described in a later section.

2. Basics of SMLM

2.1. SMLM Principle

In SMLM, individual fluorophores on a specimen are detected sequentially and localized with low position uncertainty in order to build up a high-resolution image of the labeled specimen (Figure 1). The fluorophores may be introduced through the use of antibodies, fluorescent proteins, small molecules, or other labels, and fixed or living specimens may be imaged using the procedure. The raw data typically consist of movies containing thousands of frames, where each frame contains a sufficiently sparse subset of molecules to allow detection of individual molecules. Automated analysis software is then used to determine the positions of the individual molecules in all frames of the movie. Because each molecule emits a large number of photons, the position of each molecule may be determined with a low uncertainty (typically in the range 20-50 nm, FWHM (full width at half maximum)). From a list of the known molecule positions, software is used to render a high resolution SMLM image.

Figure 1.

Basic principle of SMLM. Scale bar indicates the ~250 nm diffraction limit of visible light.

2.2. History of SMLM

Multiple lines of work converged in the first few years of the new millennium, resulting in several independent publications of the basic SMLM idea within the span of ~four months in 2006.^26–29 Below is a brief account of some of the key developments leading up to the breakthrough year for SMLM. Rather than a strictly chronological account, this story has been organized according to five themes that include: the optical detection of single molecules and photoswitches; the localization of multiple overlapping emitters; single-molecule FRET (Förster resonance energy transfer) and single-molecule tracking; the pursuit of improved spatial resolution; the SMLM breakthrough year; and analogs of SMLM used in diverse imaging modalities

2.2.1. Optical detection of single molecules and photoswitches

Single molecules were first optically detected by frequency-modulation absorption spectroscopy in 1989 by Moerner et al.³⁰ This breakthrough was followed by two key papers in 1990 where single molecules were detected by fluorescence in crystals at low temperature by Orrit et al.³¹ and at room temperature in solution by Shera et al.³² It was evident from the first fluorescence-based studies and from others published shortly thereafter that fluorophores exhibit frequent transitions between distinct states which in some cases could be driven by light as a photoswitch.^31–37 In 1997, a noteworthy paper from Dickson et al. reported the observation of reversible photoswitching of individual fluorescent protein molecules (Figure 2).³⁸ Around this same time, it was also shown that other fluorescent emitters such as quantum dots, conjugated fluorescent polymers, and individual photosynthetic reaction centers each exhibited switching or blinking behavior.^39–41 Although reversible photochromic effects^42,43 have been known for many decades prior to the optical detection of single molecules, it was nevertheless surprising to observe fluctuations and/or switching behavior among such broad classes of chromophores.

Figure 2.

On-off photoswitching of YFP. A single YFP molecule was illuminated with 488 nm light prior to odd-numbered frames, creating a nonfluorescent state. Illumination of nonfluorescent YFP with 405 nm light prior to even-numbered frames returned the molecule to the fluorescent state. The process could be repeated many times. Adapted by permission from Springer Customer Service Centre GmbH: Springer Nature, Ref. 38, Copyright 1997.

In addition to the 1997 single-molecule paper by Dickson et al.³⁸, researchers began to discover that fluorescent proteins exhibited a range of intriguing photoswitching properties, including ensemble studies which reported light-dependent fluorescence enhancement as well as green to red or red to green switching behavior.^44–48 Robust photoactivatable or photoswitchable fluorescent proteins were engineered and/or selected for beginning with photoactivatable GFP⁴⁹ and Kaede⁵⁰ in 2002; these were soon followed by kindling FP⁵¹, Dronpa⁵², and many others.^53–57 In the intervening years, the palette of photoactivatable or photoswitchable fluorescent proteins has grown rapidly and is discussed further, below, as well as in several reviews on the subject.^25,58–66

2.2.2. Localization of several overlapping emitters

In 1995, Betzig proposed a general strategy to circumvent the diffraction limit of light microscopy by sequentially identifying distinct features labeling a structure of interest that are packed more densely than the diffraction limit.⁶⁷ While the proposal, which focused on spectrally distinguishable emitters, resembled super-resolution methodology previously proposed by Burns et al. in 1985 for localizing spatially overlapping point emitters with distinct spectral properties⁶⁸, Betzig clearly articulated how detection of individual fluorophores could achieve spatial resolution that surpasses Abbe’s diffraction limit. Experiments by van Oijen et al. in 1998 were able to detect up to seven spectrally distinct emitters within a diffraction limited region by using these schemes.⁶⁹ Several years later, in 2004, two reports of super-resolution methods based on localization by sequential photobleaching were able to identify up to ~5 emitters per diffraction-limited region with identical spectra;^70,71 this was followed in 2005 by work from Lidke et al. proposing and demonstrating the identification of two blinking quantum dots with identical spectra within a diffraction-limited region.⁷²

2.2.3. Single-molecule FRET and single-molecule tracking

By the late 1990s, single-molecule FRET had been developed and began to be applied to the study of molecular interactions and dynamics of proteins and nucleic acids, etc.^73–75 In short order, the fluorophores Cy3 and Cy5 became popular choices for single-molecule FRET, and in 2005 two groups independently published observations of reversible photoswitching exhibited by the fluorophore Cy5 in deoxygenated solutions containing a thiol, a solution commonly used for single-molecule FRET studies.^76,77 While this photoswitching or blinking behavior had been previously observed by other researchers as a nuisance in single-molecule FRET studies, the two reports in 2005 recognized that the remarkable photoswitching properties of Cy5 could be potentially useful and suggested applications as memory devices or short-range probe of intermolecular distance.

In parallel with the above, the use of fluorescence microscopy had proliferated rapidly among biologists due to technical advances in instrumentation, fluorescent probes, imaging methodology, and data analysis.⁷⁸ Biophysical experiments studying motor proteins and other dynamic processes had long before developed strategies for detecting few nanometer displacements of relatively large objects such as metal nanoparticles or beads.⁷⁹ By 2003, Yildiz et al. extended the concept to few nanometer localization of single fluorescent molecules using FIONA (fluorescence imaging with one nanometer accuracy, Figure 3), work that was particularly influential in the development of SMLM methods.^80–82

Figure 3.

Trajectories of the motor protein Myosin V translocating along immobilized actin filaments, where the abrupt vertical jumps indicate individual stepping events. This was achieved by labeling Myosin V dimers on one head with a Cy3 fluorophore and then localizing the fluorophores to within a few nm uncertainty. From Ref. 80, adapted with permission from AAAS.

2.2.4. The pursuit of improved resolution

In the decades preceding the development of SMLM, many innovative methods were developed for improving the spatial resolution of fluorescence microscopy. Confocal microscopy progressed quickly in the late 1980s and has become the gold standard against which subsequent methods are compared, generally achieving a spatial resolution of ≥200 nm along the transverse dimension and ≥500 nm along the axial dimension.⁷⁸ One of the earliest super-resolution methods was near-field scanning optical microscopy (NSOM), developed in the mid 1980s^83–86, which scanned a small aperture or tip over a specimen and enabled measurement of surface features at <100 nm lateral spatial resolution.^87,88 Dual objective lens confocal microscopy, also known as 4pi microscopy, was first reported by Hell et al. in 1994⁸⁹ and was further developed by Gustafsson et al. with <100 nm axial resolution.⁹⁰ Stimulated emission depletion (STED) microscopy was developed by Hell and colleagues in the 1990s, published first as a concept in 1994⁹¹ and then realized experimentally by Klar et al. in 1999 and in 2000 with better than 100 nm lateral spatial resolution.^92,93 Structured illumination microscopy was developed by Lanni et al.⁹⁴, Heintzmann et al.⁹⁵, and Gustafsson et al.⁹⁶ during the 1990s and enabled up to a twofold improvement of spatial resolution over the diffraction limit along the lateral and axial dimensions. All of these methods were further improved in subsequent years, and some are commonly used by biologists to this day. In any event, by the early 2000s it was clear that improving the spatial resolution of fluorescence microscopy was a critically important task for biology and the problem had captured the attention of a large community of researchers.

2.2.5. The SMLM breakthrough year

With the advent of robust photoactivatable or photoswitchable fluorophores, relatively mature hardware and software for localization of single fluorophores down to a few nm, published concept or demonstrations for super-resolution microscopy including some which used point-emitters, and a growing interest in developing tools to improve the spatial resolution of fluorescence microscopy, the stage was set for rapid progress. This progress became evident within the span of a few months in late 2006, when four independent groups published experimental demonstrations of SMLM including STochastic Optical Reconstruction Microscopy (STORM) from the group of Xiaowei Zhuang, PhotoActivated Localization Microscopy (PALM) from Eric Betzig and Harald Hess and collaborators, Point Accumulation for Imaging in Nanoscale Topography (PAINT) from the group of Robin Hochstrasser, and fluorescence Photoactivation Localization Microscopy (fPALM) from the group of Samuel Hess.^26–29 All four of these papers utilized serial detection of multiple emitters with identical spectral properties within a diffraction-limited area, in contrast with earlier proposals or demonstrations. STORM, PALM, and fPALM achieved sequential detection by sequentially activating single photoswitchable fluorescent proteins or organic fluorophores bound to the specimen^26–28 whereas PAINT utilized transient binding of hydrophobic fluorophores that are activated upon binding to a lipid membrane.²⁹ Notably, PAINT built on earlier work from the same group using diffusing probes that activate upon binding.⁹⁷ Together, these methods formed the key foundations of SMLM that would be further developed and elaborated by growing community of researchers.

2.2.6. Diverse analogs of SMLM

It is interesting to observe with the advantage of hindsight that point-by-point acquisition and reconstruction methods resembling SMLM are used in diverse imaging modalities. In a first example, low electron dose electron microscopy methods from as far back as 1978 utilize low illumination current such that individual electrons are detected sequentially, individually localized, and accumulated to reconstruct an image.⁹⁸ A similar procedure has recently become popular in cryo-electron microscopy methods which use ‘super-resolution mode’ for sub-pixel localization of individual electrons incident upon a direct detection camera.^99,100 In a second example, medical radioisotope imaging (i.e., positron emission tomography and single photon emission computed tomography) also uses sequential detection of individual decay events with sub-pixel localization, although these data require considerable additional processing to reconstruct the original distribution of a radioactive contrast agent.¹⁰¹ In a final example, ultrasound imaging has now adopted the SMLM approach for sequential localization of microbubbles in order to achieve super-resolution ultrasound images of tissue vasculature.^102–105

2.3. Alphabet Soup

Single-molecule localization microscopy in many ways built off of earlier single-molecule techniques bearing colorful acronyms including Fluorescence Imaging One Nanometer Accuracy (FIONA)⁸⁰, Defocus Orientation and Position Imaging (DOPI)¹⁰⁶, Single-molecule High Resolution Colocalization (SHREC)¹⁰⁷, Single-molecule High Resolution Imaging with Photobleaching (SHRImP)⁷⁰, Nanometer-Localized Multiple Single-molecule microscopy (NALMS)⁷¹.

The proliferation of new names or acronyms continued unabated with the advent of SMLM in 2006, with four papers published under four different names: Stochastic Optical Reconstruction Microscopy (STORM)²⁶; PhotoActivated Light Microscopy (PALM)²⁷; Point Accumulation for Imaging in Nanoscale Topography (PAINT)²⁹; and fluorescence PALM (FPALM)²⁸. These were shortly followed by many others, including: PALM with Independently Running Acquisition (PALMIRA)¹⁰⁸; Ground Stated Depletion with Individual Molecule Return (GSDIM)¹⁰⁹; direct STORM (dSTORM)¹¹⁰; Single Particle Tracking PALM (sptPALM)¹¹¹; Photoactuated Unimolecular Logical Switching Attained Reconstruction (PULSAR)¹¹²; Spectral Precision Distance Microscopy (SPDM)¹¹³; interferometric PALM (iPALM)¹¹⁴; Reversible Photobleaching Microscopy (RPM)¹¹⁵; Nanometer Accuracy by Stochastic Catalytic reactions (NASCA),¹¹⁶ Single-molecule Active Control Microscopy (SMACM)¹¹⁷; Chemically Improved Resolution for Optical Nanoscopy (CHIRON)¹¹⁸; super-resolution power-dependent active intermittency PAINT (SPRAI-PAINT)¹¹⁹; Binding-Activated Light Microscopy (BALM)¹²⁰; Bleaching/blinking assisted Localization Microscopy (BaLM)¹²¹; Complementation Assisted Light Microscopy (CALM)¹²²; Enzymatic Turnover Activated Localization Microscopy (ETALM)¹²³; Tracking and Localization Microscopy (TALM)¹²⁴). While these various reports have contributed a range of strategies for detection, switching mechanisms, or other innovations, the barrage of acronyms can be bewildering to nonspecialists.

2.4. Requirements for SMLM fluorophores

In SMLM, many individual points (referred to as localizations) must be collected sequentially in order to outline a structure of interest with sufficient detail (Figure 4), a conceptually similar requirement to fulfilling the Nyquist sampling criterion. In that case, the best achievable resolution cannot be less than twice the average distance between neighboring localizations, or in other words, for D dimensions (where D is typically 2 or 3), the maximum spatial resolution is 2/(localization density)^1/D.¹²⁵ Due to the stochastic nature of the samples in SMLM, which are not, in general, evenly spaced, Legant et al.¹²⁶ have suggested that the localization density should be exceeded by a factor of ~5 fold in SMLM in order to reliably achieve the nominal Nyquist resolution, with the caveat that smaller structures or less dense proteins/molecules would not require such high densities (see Section 9.4). Additionally, the position of each localization must be known with low uncertainty.

Figure 4.

Impact of density of localizations on resolution achievable by SMLM. A high density of localizations (top row) resolves even the finest details on all of the test grids, while lower densities of localizations (middle and bottom rows) only resolves details about the larger grids. Reproduced by permission from Springer Customer Service Centre GmbH: Springer Nature, Ref. 125, Copyright 2008.

The requirements for a high density of localizations and a low position uncertainty in turn place very specific requirements on fluorophores for SMLM. First, the fluorophores must be able to be switched between a bright and a dark state, with most molecules in the dark state, and the contrast between these states should be very high. For instance, in order to detect one bright fluorophore in a diffraction-limited volume out of a background of 100 dark fluorophores, the ratio of brightness of the on and off states, referred to as the on-off contrast ratio, should be greater than ~100:1. Although the on-off contrast ratios are seldom reported, many of the popular SMLM fluorophores have on-off contrast ratios much greater than ~1000:1. Second, the fluorophores must have a low duty cycle, or fraction of time spent in the bright state prior to bleaching, regardless of whether the fluorophore is reversibly photoswitchable or irreversibly photoactivatable. In the case where there are 100 fluorophores in a diffraction-limited volume, the fluorophores must spend much less than 1% of their time in the bright state in order to detected individually (i.e., must have a duty cycle of <0.01) in order to avoid a very high incidence of double/multiple localizations (see Section 9.3). The relationship of duty cycle to on/off rates is shown in Figure 5 and, notably, is applicable whether using reversible switching or irreversible activation followed by bleaching. It should be noted that the duty cycle of a fluorophore is in general dependent upon illumination conditions and chemical environment; researchers commonly illuminate a specimen with violet light (~405 nm) during the acquisition of a SMLM movie in order to acquire localizations more quickly (i.e., increasing the duty cycle above tabulated value) while taking care to avoid excessive double/multiple localization artifacts. Third, the fluorophores must be bright enough to allow single-molecule detection and they must emit a large number of photons during an on burst, where position uncertainty scales roughly as the inverse square root of the number of detected photons.^80,81,127 The ideal SMLM fluorophore would therefore have a low duty cycle, a high on-off contrast ratio, a large number of emitted photons per burst, as well as other important but less fundamental properties that include the number of switching cycles, the utility of the fluorophore in labeling structures of interest, and low phototoxicity when imaging living specimens.

Figure 5.

A) Simplified analysis of reversible switching, B) irreversible activation, and C) estimate of duty cycle, where k_off is the rate of switching off, k_on is the rate of switching on, k_off,2 is the rate of bleaching for a reversible switch, DC is the duty cycle, t_on is the average time spent in the on state prior to bleaching, and t_off is the average time spent in the off state. Notably, DC is approximately the ratio of the on and off rates for both reversible switching and irreversible activation.

2.5. Fluorophore switching based on blinking/bleaching/binding of probes

In SMLM, individual fluorophores are imaged on a dark background and positions are determined through analysis of the distribution of fluorescence in the on state. A set of related methods uses analysis of two or more subsequent frames to determine distributions of fluorophores either by observing the abrupt appearance/disappearance of a signal on top of a nonzero background or through statistical analysis of fractional fluctuations of the signal. In some cases, photoswitching properties that may not be suitable for observing individual molecules by SMLM may still be useful for these related methods. Similarly, hardware requirements for these methods may in some cases be relaxed.

The activation or bleaching of a single fluorophore in the presence of other emitting fluorophores which do not change may be detected by subtracting consecutive frames (Figure 7). A fluorophore which appears abruptly due to photoactivation or binding will produce a bright peak in a difference image, while a fluorophore which disappears abruptly due to bleaching or unbinding will produce a valley. Although the presence of a high background increases localization uncertainty,⁸¹ individual fluorophores may nonetheless be detected with sub-diffraction resolution provided that one or fewer transitions occur per diffraction-limited area between consecutive images.

Figure 7.

Schematic illustration of photobleaching based SMLM wherein consecutive frames (F_n) are subtracted to localize individual bleached molecules. A similar strategy can be used to identify individual activated molecules in consecutive frames.

Two reports published in 2004 introduced the acronyms SHRImP (single-molecule high-resolution imaging with photobleaching) and NALMS (nanometer-localized multiple single-molecule microscopy) and used sequential photobleaching to localize a handful of fluorophores per diffraction limited area.^70,71 On the same theme, generalized SHRImP (gSHRImP), QDB3, and bleaching/blinking assisted localization microscopy (BaLM) were published in 2011 for the detection of both bleaching and blinking events through difference analysis of consecutive frames.^121,129,130 A more computationally intensive approach termed 3B Analysis (Bayesian analysis of blinking and bleaching) concurrently determines the positions of relatively large ensembles of molecules from image stacks consisting a few hundred frames based on blinking and bleaching signals observed within the stack.¹³¹ SOFI (super-resolution optical fluctuation imaging) and photobleaching microscopy with nonlinear processing (PiMP) instead use statistical methods to analyze spatially and temporally varying fluorescence fluctuations within the specimen in order to extract sub-diffraction images.^132–134 Given the large variety of ways to extract sub-diffraction-limit information from the fluctuating fluorescence emission of a specimen, it is likely that fluorophores which are unsuitable for one method may find good application in another.

3. Switchable fluorophores for SMLM

A wide range of fluorophores has been utilized for SMLM, including synthetic organic fluorophores, fluorescent proteins, and quantum dots. These fluorophores rely upon a diverse set of switching mechanisms to convert between fluorescent and nonfluorescent forms so that many individual fluorophores may be sequentially detected in an image series within a diffraction limited volume for SMLM. We review the main fluorophore categories, below, with an emphasis on organic fluorophores spanning a range of chemical classes.

3.1. Major classes of organic fluorophores

To accompany the discussion, below, on organic fluorophores for SMLM, we have prepared a supporting document containing the structures of most of the organic fluorophores we discuss in this paper together with a summary of the key switching reactions (see Supporting Information).

3.1.1. Cyanines

Cyanine dyes are perhaps the most popular fluorophore class for SMLM, particularly the red-absorbing fluorophore Cy5 and its close structural relatives such as Alexa Fluor 647. The photoswitching behavior of Cy5 was first reported by Bates et al. and Heilemann et al. in 2005, where both groups detected the switching behavior through observation of single molecules transitioning reversibly between fluorescent and nonfluorescent states (Figure 8A).^76,77 The reports showed that red light could be used to switch Cy5 to a long-lived nonfluorescent state and that shorter wavelength light could be used to activate nonfluorescent Cy5 back to a fluorescent state; Heilemann et al. utilized strong visible light for activation, while Bates et al. utilized much weaker visible light to excite Cy3 “activator” dyes that were positioned within 1-2 nm of Cy5 molecules and could activate dark-state Cy5 molecules.^76,77 Photoswitching in both cases was achieved in a solution commonly used for single-molecule FRET microscopy which contains a primary thiol such as β-mercaptoethylamine (MEA) or β-mercaptoethanol (BME) as well as an enzymatic oxygen scavenging system (see also Section 3.4).^76,77

Figure 8.

Photochemical and chemical reactions switch cyanine dyes between fluorescent and nonfluorescent states. A) Photoswitching of a single Cy5 molecule in a deoxygenated solution containing a thiol. The Cy5 molecule was continuously illuminated with 633 nm light, switching off at ~500 ms. When intense 488 nm light also illuminated the sample at time ~5 sec, the Cy5 fluorescence was recovered and the molecule repeatedly switched between bight and dark states. B) Chemical quenching of an ensemble of Cy5 molecules with the phosphine TCEP and photoactivation with brief pulses ultraviolet (UV) light (gray arrows). C) UV light-induced activation of reduction-quenched Cy3B. Reproduced with permission from Ref. 77, Copyright 2005 American Chemical Society. Reproduced with permission from Ref. 135, Copyright 2013 American Chemical Society. Reproduced by permission from Springer Customer Service Centre GmbH: Springer Nature, Ref. 136, Copyright 2012.

The first application of Cy5 photoswitching to SMLM was reported in the original STORM paper in 2006 by Rust et al., where the Cy3-Cy5 dye pair was utilized to image circular RecA-coated plasmid DNA.²⁶ The same group later showed that Cy5.5 and Cy7 also exhibit good switching properties and that each of the reporter molecules (Cy5, Cy5.5, and Cy7) may be paired with a number of activator molecules (e.g., Alexa Fluor 405, Cy2, Cy3) in order to create a palette of activator-reporter pairs for multicolor super-resolution microscopy (Figure 9A-B).^137,138 Bock et al. showed in 2007 that Cy5 could be used for SMLM without an activator dye;¹³⁹ this “activator-free” approach was later substantially expanded upon by Heilemann et al. and other researchers using a number of cyanine or non-cyanine fluorophores and is often referred to as dSTORM (or direct STORM).^109,110,140 Importantly, Cy5 (and its close structural analog Alexa Fluor 647) is still considered one of the best overall fluorophores of any class for SMLM due to its particularly suitable photoswitching properties including a high photons per localization, a low duty cycle, and reasonably large number of switching cycles.

Figure 9.

SMLM imaging using cyanine dyes. A) SMLM zoom-in view of a cell immunostained for microtubules (green) and clathrin-coated pits (red) with activator-reporter fluorophore pairs. Alexa Fluor 647 was the reporter fluorophore used on both structures but Cy2 was used as an activator dye on microtubules and a Cy3 was used as an activator dye on clathrin-coated pits. SMLM images of cells immunostained for B) microtubules and mitochondria or C) tyrosinated (T) and detyrosinated (DT) microtubules using multi-reporter imaging with Alexa Fluor 647 and Alexa Fluor 750. D) SMLM image of microtubules grown in vitro, and directly labeled with Cy3B (green, with partial overlay of SMLM image on magenta conventional image); inset shows cross-sectional profile of zoomed in region (red box) with the successful resolution of the inner microtubule walls (~16 nm). In A-C, images were acquired under conditions where fluorophores can reversibly switch on and off, whereas D was acquired using reductive caging as described in the text. From Ref. 137, adapted with permission from AAAS. Adapted with permission from Ref. 138, Copyright 2012 John Wiley and Sons. Adapted with permission from Ref. 135, Copyright 2013 American Chemical Society. Adapted by permission from Springer Customer Service Centre GmbH: Springer Nature, Ref. 136, Copyright 2012.

Dempsey et al. used single-molecule imaging and mass spectrometry to study the structure and mechanism of formation of the dark state of Cy5.¹⁴¹ They found that upon illumination with red light, thiolate anions added covalently to the bridge of Cy5, disrupting the conjugation and shifting the absorption to the ultraviolet (Figure 10A).¹⁴¹ The site of addition to the Cy5 polymethine bridge was identified as most likely the γ-carbon of the bridge (Figure 10A).¹⁴¹ It was later shown by Vaughan et al. through mass spectrometry and NMR spectroscopy that the phosphine TCEP (tris 2-carboxyethyl phosphine) adds reversibly to the γ-carbon of the bridge of Cy5 even in the absence of illumination with red light (Figure 8B, Figure 10A), further supporting the idea that the γ-carbon of the bridge of Cy5 is also the site of addition of the thiolate anion.¹³⁵ This TCEP-based switching approach also provided an improved two-color imaging approach using Cy5 and Cy7 (or close structural relatives of Cy5 and Cy7 such as Alexa Fluor 647 and Alexa Fluor 750) by boosting the relatively low photon output of Cy7 approximately 3-fold compared to thiol-based imaging.¹³⁵

Figure 10.

Schematic illustration of switching reactions of cyanine fluorophores. A) Cy5 is switched off by BME in the presence of intense red light or by TCEP (without illumination) to form a non-fluorescent adduct in which a thiolate anion or the phosphine TCEP is bound to the γ position of the polymethine bridge, switching it to a nonfluorescent state. Illumination with ultraviolet light dissociates the adduct, returning Cy5 to its bright form. B) Cy3B and other cyanines may be chemically reduced by NaBH₄ to a non-fluorescent form which is photoactivatable.

Unlike the fluorophores Cy5, Cy5.5, and Cy7 which have polymethine bridges that are either 5 or 7 carbons long, the fluorophore Cy3 does not photoswitch well in deoxygenated, thiol-containing solutions.^128,141 This may reflect a lack of access to the three-carbon bridge of Cy3 or differences in the rates of other photochemical steps. Instead, the rigidified fluorophore Cy3B (Figure 10B) was found to photoswitch reasonably well in deoxygenated solutions, whether or not a thiol was included in the solution,¹²⁸ implying that Cy3B may use a different mechanism of photoswitching such as a long-lived radical as has been shown to be the case for some rhodamine fluorophores.¹⁴² Glembockyte and Cosa recently showed that Cy3B exhibits remarkably efficient geminate recombination of photo-generated radical ion pairs in the presence of selected reducing agents (particularly for BME) which may help in the development of optimized imaging cocktails for SMLM (see also Section 3.4).¹⁴³ Regardless, the fluorophores Cy3B, Cy5, and Cy7 and their close structural relatives exhibit good photoswitching properties that make the cyanine class of fluorophores some of the most heavily used in SMLM.¹²⁸

Hydrocyanines are cyanines that have been rendered nonfluorescent through chemical reduction and that, under the right circumstances, can be activated back to a fluorescent state.^144–146 This class of molecules, whose first example dates back at least to Richard Kuhn in 1932,¹⁴⁴ has been used in various applications including as sensors for electron beams¹⁴⁵ or reactive oxygen species¹⁴⁶ as well as in SMLM.¹³⁶ Hydrocyanines are typically formed through net addition of a hydride anion to C_α of the cyanine polymethine bridge (Figure 10A). Vaughan et al. developed an in situ method for “reductive caging” in which the cyanines Cy3, Cy3B, Cy5, or Cy5.5 could be reduced to hydrocyanines in situ through brief treatment of a fixed and labeled specimen with dilute aqueous sodium borohydride.¹³⁶ The hydrocyanines were then photoactivated in an imaging cocktail (see Section 3.4) that suppresses bleaching and blinking, yielding 4-20 times more detected photons per localization (Figure 8C) and a very low resulting position uncertainty (Figure 9D). Although the improved photon yield resulted in greatly improved localization, the method was only able to activate ~12-40% of fluorophores and did not exhibit an improvement in duty cycle over thiol-based switching.¹³⁶ Lehmann et al. have extended this approach to two or three color SMLM with a range of other cyanine and non-cyanine fluorophores.¹⁴⁷ Michie et al. have developed a new rigidized Cy5 variant called Cy5B which has a recovery yield of ~38%, which is an approximately six-fold improvement over its non-rigidized analog.¹⁴⁸

3.1.2. Rhodamines

Rhodamines are a popular class of fluorophores for many applications in fluorescence microscopy due to their high quantum yield, large extinction coefficient, and high photostability, and several rhodamines are commonly used in SMLM. The relatively easy synthesis of rhodamines also makes them good candidates for further innovation by synthetic chemists.

Several popular, commercially available rhodamine fluorophores exhibit good photoswitching in deoxygenated solutions containing a thiol, including ATTO 488, Alexa Fluor 488, and Alexa Fluor 568.^109,128,140 Thus, a number cyanines and rhodamines (as well as other fluorophores) can be switched well using a single “imaging cocktail” (Section 3.4). Unlike the red and far-red cyanines such as Cy5 and Cy7, however, many rhodamines are photoswitched to a dark state that has been shown by electron paramagnetic resonance spectroscopy to consist of a long-lived radical (Figure 11, upper left).¹⁴² The radical state is believed to be generated by the photo-reduction of its triplet state by thiol in solution.¹⁴²

Figure 11.

Reversible switching reactions of rhodamine fluorophores. Illumination of a fluorescent rhodamine (green, lower left) in a deoxygenated solution containing a thiol leads to the generation of a nonfluorescent, long-lived radical (upper left) while chemical reduction leads to the generation of a non-fluorescent, reduced rhodamine (lower right). Oxidation or illumination with ultraviolet light can return the radical or reduced rhodamine back to the fluorescent state. The inset diagram illustrates the thiol-induced photo-reduction of the rhodamine triplet state to a radical species (F^•) which is followed by oxidation back to the ground state. Adapted from Ref. 142 with permission of The Royal Society of Chemistry (RSC) on behalf of the Centre National de la Recherche Scientifique (CNRS) and the RSC.

Rhodamines are also good candidates for reductive caging and show very good photon-recovery yields by UV illumination (e.g., 66% for ATTO 488) compared to cyanines (12–40% recovery).^136,147 The net addition of a hydride anion to the central carbon bridge of the xanthene group breaks its conjugation (Figure 11, lower right),¹⁴⁷ thereby quenching fluorescence emission. UV illumination removes the hydride and restores the caged rhodamine back to its fluorescent form prior to imaging and bleaching for SMLM. Lehmann et al. screened a variety of rhodamine dyes for reductive caging and found that ATTO 488, ATTO 514, and ATTO 532 also have good recovery yields of ~30-60%.¹⁴⁷

One of the most well-known reactions of rhodamines is that their absorption/emission can be reversibly switched off through formation of a spiro-ring on the central carbon of the xanthene (Figure 12, top).¹⁴⁹ An intramolecular nucleophile such as a carboxylate, hydroxy, or amide, is generally required to form the spiro-ring and break the conjugation of the xanthene chromophore. Changes of pH or the concentration of certain metal ions or small molecules can shift the equilibrium in either direction by coordination effects or by specific reactions,^150,151 while tuning electron density on the xanthene ring can also shift the equilibrium.¹⁵² This spiro-ring equilibrium reaction has been widely exploited in the development of various rhodamine-based sensors or switches¹⁵⁰ and has been adapted for use in SMLM.¹⁵³

Figure 12.

Schematic illustration of switching reaction of rhodamine fluorophore via ring opening and intramolecular spirocyclization reactions (top). Examples of spiro-rhodamine dyes used for SMLM are shown (bottom).

Different rhodamine skeletons show different sensitivities to environmental variations which can greatly influence the open/close equilibrium. For example, Lukinavičius et al. showed that the red-shifted dye Si-rhodamine (SiR, absorption/emission = 645 nm/661 nm) with a carboxylic acid group at the 2-position of the phenyl ring (SiTMR in Figure 12, shown in its nonfluorescent spiro-ring form) is quenched while free in solution but becomes ~5x brighter when a ligand conjugated to the dye binds to a reactive protein tag such as SNAP.¹⁵⁴ SiTMR thus exhibits decreased background signal while improving cell permeability because the favored spiro-ring form of the dye masks charges on the zwitterionic, open form of the dye.¹⁵⁴ At the same time, the bound, open form of the dye exhibits suitable photoswitching properties for SMLM, allowing, for instance, high quality imaging of histone proteins in the nuclei of living specimens without interference from unbound probes.¹⁵⁴

Recently, Lavis’s group has made substantial improvements to the fundamental spectroscopic and photoswitching properties of rhodamine fluorophores.¹⁵⁵ For instance, they discovered that the replacement of the classical N,N-dialkyl groups with four-membered azetidine rings substantially improved the fluorescence quantum yield of several rhodamine dyes (e.g., Figure 13A, R=H, JF₆₄₆ (Janelia Fluor)).¹⁵⁶ By further adjusting the substituents on the azetidine ring, they were able to fine-tune the ring-closed/open equilibrium,¹⁵² resulting in a series of novel fluorogenic/chromogenic ligands such as JF₆₄₆-Halo and JF₆₃₅-Halo, which provide a 21-fold and 113-fold absorbance enhancement, respectively, upon binding a HaloTagged protein (Figure 13B). Binding of these probes to HaloTagged proteins also stabilizes their ring-open (bright) form and with much higher on-off contrast than that of the N,N-dialkyl equivalent ligand while retaining good switching properties (Figure 13C).

Figure 13.

A) Structure of two example fluorogenic/chromogenic HaloTag ligands JF₆₄₆-Halo and JF₆₃₅-Halo. B) Absorbance enhancement of JF₆₄₆-Halo and JF₆₃₅-Halo after binding with HaloTag protein. C) No wash fluorescence imaging of COS7 cells expressing HaloTag-histone H2B fusion and then labeled with JF646-Halo (i), JF₆₃₅-Halo (ii) and SiTMR-Halo (iii). Scale bar, 15 μm. Adapted by permission from Springer Customer Service Centre GmbH: Springer Nature, Ref. 152, Copyright 2017.

Beyond providing a means for reducing background signal, the open-closed equilibrium of spiro forms of rhodamines can, themselves, form the basis of the on-off switching chemistry for SMLM. For example, spiro-lactams are an easily formed nonfluorescent form of rhodamine fluorophores (RSA-1 and RSA-2 in Figure 12) which in general exhibit higher stability than spiro-lactones, although acid or certain other analytes can shift the equilibrium toward the ring-open state.¹⁵³ The increased stability of spiro-lactams makes them suitable to caging a wider range of rhodamine dyes compared to spiro-lactones, and interestingly, by proper tuning of substituents on the amide moiety, the spiro-lactam ring can be designed to be opened by UV light irradiation before spontaneously converting back to the ring-closed form (Figure 12, top). For example, Fölling et al. incorporated a 4-aminophthalimide substituent (RSA-1) such that electron-withdrawing groups on the phthalimide, together with its original phenylacetyl, stabilized the charge on the amide nitrogen atom by delocalization and improving the photoactivation process.^157–159 The resulting dye RSA-1 can be activated with ultraviolet light (~310-380 nm) in order to open the spiro-lactam ring such that the dye becomes fluorescent, emitting up to 900 detected photons per molecule, before returning to the spiro-lactam form. Fölling et al. also demonstrated that two-photon activation can extend this type of rhodamine spiro-lactam for deep 3D imaging with axial optical sections of 0.5-1 μm (Figure 14A-C).¹⁵⁹ Further optimizations were performed, either by fluorination, sulfonation or other functionalizations¹⁶⁰ on xanthene or phenyl-amide moieties,^158,161 or by introducing other new rhodamine frameworks,¹⁵⁸ to improve photoactivation and tune the optical properties of dyes for multicolor imaging (Figure 14D-H).^158,162 Lee and co-workers hypothesized that extension of the conjugation area on the phenyl-lactam moiety will further lower the energy required to photoactivate the ring-closed isomer¹⁶³ which is favorable for retaining the fluorescence from already activated isomers. Inspired by this, RSA-2 was developed (Figure 12), showing more efficient photoactivation in response to 405 nm illumination. The pyridinium-derived activated ester linker furnished a convenient handle for labeling of the specimen of interest. Using RSA-2 as photoswitchable fluorescent marker, they quantified the distribution of cell stalk lengths at a mixed population of C. crescentus cells (Figure 14I).¹⁶³

Figure 14.

SMLM images obtained using various photoswitchable spiro-lactam rhodamines. A) A 3D SMLM image of a 5 μm silica bead whose surface has been stained with PC-Rh590 and imaged in 16 layers. B) The middle focal plane from A, and C) an associated line profile showing a lateral FHWM of ~30 nm. D-H) Multicolor images of mixtures of silica beads labeled with photoswitchable spiro-lactam rhodamine fluorophores SRA545 (blue), SRA577 (green), and SRA617 (red). I) Montage of SMLM images of C. crescentus cells labeled with RSA-2. Scale bars, 1 μm for B, D and I, 200 nm for E-H. Adapted with permission from Ref. 159, Copyright 2008 John Wiley and Sons. Adapted with permission from Ref. 162, Copyright 2008 American Chemical Society. Adapted with permission from Ref. 163, Copyright 2014 American Chemical Society.

Rather than use light to switch caged fluorophores, the equilibrium between ring-closed (nonfluorescent) and ring-open (fluorescent) forms may be tuned in order to allow passive on-off switching that is suitable for SMLM. Uno et al. have achieved this by preparing and optimizing a series of rhodamine derivatives bearing various intramolecular nucleophiles at the 2-position of the phenyl ring.¹⁶⁴ Among these, HMSiR (Figure 12) exhibited good switching properties in fixed or living specimens with ~2,600 detected photons per switching event and a duty cycle of ~0.01 (Figure 15A).¹⁶⁴ Although passive switching lacks the convenient adjustability of on/off rates achievable by photoswitching-based approaches, passive switching avoids the use of possibly toxic reagents or near-ultraviolet illumination for live-cell imaging. SR imaging of HMSiR labeled RecA filaments by SMLM revealed a circular structure (Figure 15B-C) with a feature width (51.4 ± 3.2 nm, FWHM) that was comparable to that achieved with thiol-based Cy5 photoswitching in deoxygenated solutions (47.2 ± 2.0 nm, FWHM).

Figure 15.

Use of the spontaneously-blinking fluorophore HMSiR for SMLM. A) Single-molecule fluorescence time traces of antibody-bound HMSiR in the absence of thiol and an oxygen-scavenging system. B) Conventional and C) SMLM image of circular, plasmid DNA decorated with HMSiR-labeled RecA filaments. D) Cross-sectional profile of yellow boxed-region in C, showing an apparent ~52 nm width. Scale bars, 500 nm. Adapted by permission from Springer Customer Service Centre GmbH: Springer Nature, Ref. 164, Copyright 2014.

Belov et al.¹⁶⁵ reported a novel caging strategy for rhodamines by using a diazoketone group to form nonfluorescent spirocyclic rhodamines, termed Rhodamine NN dyes (Figure 16). Unlike spiro-lactams, which after activation spontaneously return to their ring-closed nonfluorescent form, the irradiation of non-fluorescent Rhodamine NN with UV light converts the rhodamine to the fluorescent spiro-open form while also undergoing subsequent chemical steps that degrade the caging group (fluorescent products are shown in Figure 16).¹⁶⁵ Thus, the uncaged rhodamine fluorophores are locked in the ring-open until they photobleach. The flexible diazoketone caging group can be incorporated into a variety of rhodamine and carbopyronine fluorophores in order to span a wide spectral range.^166,167

Figure 16.

Photo-activation reactions of rhodamine derivatives by photo-cleavage of diazoketone on spiro-ring (top). Examples of diazoketone caged rhodamine dyes, PA-JF₅₄₉, PA-JF₆₄₆ and their corresponding Halo- and SNAP tag ligands (bottom).

Grimm et al. recently reported improved photoactivatable rhodamine NN dyes¹⁶⁸ based on combination of diazoketone caging¹⁶⁵ with the azetidine-rhodamine dyes JF₅₄₉ and JF₆₄₆.¹⁵⁶ The resulting dyes, such as PA-JF₅₄₉ and PA-JF₆₄₆, exhibit good caging/uncaging properties and good cell permeability that are well suited for SMLM and single-molecule tracking experiments (Figure 16 bottom).¹⁶⁸ Upon photoactivation, PA-JF₅₄₉ and PA-JF₆₄₆ each generate fluorescent products with spectroscopic properties consistent with uncaged rhodamine dyes (Figure 17A). The photoconversion showed improved yields when attached to Halo- or SNAP-tag proteins. Compared with the genetically encoded fluorescent protein mEos3.2 and the commercially available TMR-HaloTag probe, PA-JF₅₄₉-Halo also shows higher detected photons per particle per frame, longer tracks in single-molecule tracking experiments, and higher calculated localization precision in SMLM (Figure 17B-C). Grimm et al. used PA-JF₅₄₉-Halo, together with the spectrally independent far-red Si-rhodamine fluorophore PA-JF₆₄₆, to perform single-molecule tracking and SMLM in live embryonic stem cells expressing histone H2B–SNAP-tag and HaloTag–Sox2 (Figure 17D).

Figure 17.

A) PA-JF646-Halo can be photoactivated, and when bound to a Halo-tagged protein, the photoactivation yield is improved. Comparison of B) photons per localization per frame and C) track length (number of frames per localization) for PA-JF₅₄₉-Halo and mEos3.2 showing that PA-JF₅₄₉-Halo is brighter and produces longer tracks. D) Single-molecule tracking and SMLM in a live embryonic stem cell expressing histone H2B–SNAP-tag and HaloTag-Sox2 labeled with PA-JF₆₄₆-SNAP and PA-JF₅₄₉-Halo, respectively; the apparent diffusion rate of Sox2 is lower for the fraction of Sox2 that colocalizes with H2B than for the fraction that does not colocalize (H2B data not shown here). Adapted by permission from Springer Customer Service Centre GmbH: Springer Nature, Ref. 168, Copyright 2016.

Another well-known rhodamine caging strategy in combination with formation of a spiro-ring is the introduction of 2-nitrobenzyl derivatives as photo-cleavable caging groups (Figure 18),^169–172 a caging strategy originally developed by Mitchison with caged fluorescein in 1989.¹⁷³ Among these, Wysocki et al. recently established an improved strategy to functionalize rhodamine (and fluorescein) dyes with 4,5-dimethoxy-2-nitrobenzyl (DMNB) groups on the aniline moiety of the xanthene ring (Rh_Q8, Figure 18).¹⁷⁴ A key advance in their work was the use of reduced fluorophore leuco-dye intermediates in order to improve the efficiency of rhodamine acylation with the DMNB caging groups.¹⁷⁴ UV irradiation can cleave the DMNB group in order to restore the dye to its fluorescent form and this caging strategy can also be applied to other structurally similar fluorophores, such as silicon-rhodamine,¹⁷⁵ fluorescein,¹⁷⁴ carborhodamine, and carbofluorecein.¹⁷⁶ However, the use of two caging groups (e.g., Rh_Q8 in Figure 18) potentially increases the irradiation time required to achieve full uncaging and also leads to the possibility of incomplete uncaging and an associated variation in the emission signal.^170,177 Banala et al.¹⁷⁸ caged rhodamine 110 with one DMNB group (BG-cRhod1 in Figure 18) while the other aniline group was modified with a functional linker for conjugation with benzylguanine (BG) to react with SNAP-tag fusion proteins. Photo-cleavage of the DMNB group by illumination with ultraviolet light recovers the fluorescence emission,¹⁷⁷ leading to a large fluorescence enhancement.

Figure 18.

Photo-switching of 4,5-dimethoxy-2-nitrobenzyl (DMNB) group caged rhodamine derivatives by photo-cleavage reaction. Example dyes Rh_Q8 and BG-cRhod1 used for SR-imaging by SMLM are shown at bottom.

3.1.3. Spiropyrans and oxazine-auxochromes

Spiropyrans (SPs) are classic photochromic molecules which have attracted interest for their use in SMLM.^179,180 Similar to the on-off switching behavior of spirocyclic rhodamines, SPs can exist in a spiro-closed form (SP) or a spiro-open form called the merocyanine (MC) form. Additionally, illumination with suitable wavelengths of light can shift the equilibrium between the SP and MC forms (Figure 19 top).^179,181 The spiro-cyclization is formed from an internal nucleophile under visible light irradiation, which converts the C_α carbon on the bridge from sp² to sp³ hybridization and breaks the original π-conjugation to either generate a nonfluorescent product¹⁸² or to generate a spectrally blue-shifted emissive product.^183,184 UV or blue light will cause photo-elimination of the nucleophile and recovers the π-conjugation to generate MC, while visible light illumination accelerates the spiro-ring-close conversion to dark SP state again. The SP-MC conversion shows good reversibility by UV/visible light irradiation, and similar to the rhodamine spiro-lactam system, there are no requirements for special fluorescence quenching agents or a strict oxygen-free environment. Hu et al.¹¹² have utilized SP-MC photoswitching for SMLM, in order to demonstrate the imaging of several nanoparticles within a diffraction-limited volume (Figure 20). In their work, illumination with 532 nm light shifted the equilibrium to a nonfluorescent SP state while at the same time exciting fluorescence from only those molecules remaining in the MC form. As with some other photoswitches, illumination with UV light provides a convenient option for activating fluorophores. Multiple modification options enable functionalization of various kinds of particles with SPs allowing investigation of their structure or organization at sub-diffraction-limit resolution.^185–188

Figure 19.

Photoswitching reactions of spiropyran (SP, top) and oxazine-auxochromes (OA, bottom).

Figure 20.

70-nm spiropyran nanoparticles as imaged by conventional fluorescence microscopy (top) and SMLM (bottom). Reproduced with permission from Ref. 112, Copyright 2008 American Chemical Society.

Oxazine-auxochromes (OAs) are fluorophore dyads that, like SPs, exhibit intramolecular spirocyclic ring-open/close reactions that induce a large spectroscopic shift or fluorescence on/off switching suitable for SMLM.^189–191 The oxazine moiety of the dyad acts as a molecular switch that can modulate the fluorescence of a range of appended fluorophores. Deniz et al. and Petriella et al. reported OAs in which the appended fluorophores include coumarin, pyrene, fluorene, BODIPY, or cyanine fluorophores and all but the oxazine-cyanine dyads existed primarily in the ring-closed form. ^189–193 Among these, the oxazine-coumarin OA-1 (Figure 19 bottom) exhibited a good fluorescence quantum yield and reversible switching sufficient for SMLM on nanoparticles¹⁸⁹ and immunolabeled cells.¹⁹¹ Based on their modular and easily tunable design, OAs are important photoswitchable compounds that with further development may emerge as powerful tools for SMLM.

3.1.4. 1,2-dithienylethenes

1,2-dithienylethene (DTE) fluorophores are another classic class of photochromic compounds with excellent fatigue resistance (many switching cycles before degradation), fast response, good thermal stability, and large spectral shift between the two isomers, which are in general desirable characteristics for SMLM.^194–196 DTEs undergo photo-isomerization reactions between ring-open and ring-closed states accompanied with de-coloration and coloration (Figure 21) that occurs differently from that of rhodamine-spiro dyes or spiropyran cyclization. For the two latter dyes, UV irradiation cleaves one covalent bond on the spiro-ring, inducing a ring-opening reaction, and restoring the π-conjugation system to the colored state. For DTE dyes, the ring-open form shows no absorption in the visual region. UV irradiation enables the formation of a C-C bond between two thiophene moieties to produce the ring-closed form of DTE which typically absorbs in the visible range and for a subset of DTE variants is fluorescent.^{197, 198}

Figure 21.

Photoswitching of DTE based on light-induced ring-opening/closing.

DTEs are useful as reversible photoswitchable fluorescence quenchers in quencher-reporter dyads. This is achieved by linking a DTE molecule to a fluorescent reporter molecule, such as naphthalimide,¹⁹⁹ perylenemonoimide,²⁰⁰ perylenebisimide²⁰¹ or tetraphenylethene,²⁰² whose fluorescence emission spectral range of ~500 nm to ~700 nm overlaps well with the absorption spectrum of the closed-ring form of DTE. Photo-isomerization of DTE switches the linked fluorophore or aggregation-induced emission luminogen^202,203 between on/off states based on intramolecular energy transfer, and this on/off switching facilitates SMLM.^199,200,202 Strategic modification can provide DTEs with good emission properties,^197,198 enabling their use in SMLM.

More recently, Roubinet et al. reported the synthesis and use of new hydrophilic DTEs for SMLM of biological specimens using antibody-conjugated DTEs. In this approach, DTEs are directly used as a photoswitch where they switch transiently between fluorescent and non-fluorescent states.²⁰⁴

3.1.5. Dicyanodihydrofurans

Dicyanodihydrofuran (DCDHF)-based fluorophores have large extinction coefficients, high fluorescence quantum yields, and good photo stability;^205–207 through combination with caging and activation strategies, DCDHFs have been used in SMLM. DCDHFs are strong electron acceptors when combined with a conjugated unit (e.g., benzene, thiophene, alkene, styrene) and when linked to an electron donor (e.g., amine, hydroxyl) a typical D-π-A (donor—acceptor) system will be constructed. This electron pull-push character offers DCDHFs a bright fluorescence emission with adjustable spectral properties through the use of various substitutions, although these and other properties are relatively sensitive to the local environment (e.g., viscosity, polarity, pH).^208,209

Several methods have been developed to switch DCDHFs off and on for use in SMLM. The DCDHF D-π-A system can be blocked by replacing the amine with azido group, a mildly electron-withdrawing group, to produce a non-fluorescent, caged DCDHF molecule. Generally, the aryl azide groups are able to be photoactivated to release one nitrogen molecule and give an aryl nitrene, which is prone to undergo rearrangement to seven-membered azepine heterocycles.²¹⁰ However, in DCDHFs, a high electron-deficiency will stabilize the aryl nitrene and preferentially convert the nitrene to an amine or amino derivative upon reaction with the surrounding molecules.^210,211 The photoconversion process of azide to amine restores the D-π-A system as well as its fluorescence emission (Figure 22).^212,213 This can be a general strategy to construct photoswitching in other D-π-A fluorophores.²¹¹ The photoactivated process in DCDHFs system is irreversible, like the above-mentioned DMNB or diazo caged rhodamines; once activated, the fluorophore cannot revert to its dark state. This irreversible uncaging property is useful in SMLM since it can result in more photons being collected in a single burst rather than several separate small bursts (that in general cannot be linked to the same molecule) so that a superior localization precision can be obtained for each detected molecule. Lee et al. reported an azido-chloroalkane-DCDHF probe and used it to label Halo-tagged proteins (see Section 6.2) and to image protein distributions by SMLM.²¹³

Figure 22.

Example DCDHF fluorophores (top) and fluorogenic reaction of azido-DCDHFs based on photoactivated conversion of caging groups.

3.1.6. Summary of commonly used organic fluorophores

A wide range of organic fluorophores for SMLM are commercially available or have been reported in the literature, and we believe it is critical for the field that SMLM fluorophore developers report key SMLM fluorophore properties including photons/localization, duty cycle, and number of cycles. Along these lines, Dempsey et al. compared 26 commercially available organic fluorophores for SMLM using single-molecule measurements of photoswitching properties with dyes on antibodies as well as SMLM imaging of standard specimens.¹²⁸ They identified superior fluorophores in each of four spectral bands for low-crosstalk multichannel imaging (ATTO 488, Cy3B, Cy5 or Alexa Fluor 647, DyLight 750 or Alexa Fluor 750) and showed that all exhibited good performance in a deoxygenated solution containing a thiol (see Section 3.4 on imaging cocktail).¹²⁸ Bittel et al. showed that unconjugated fluorophores embedded in polyvinyl alcohol film behave similarly to antibody-conjugated fluorophores in an aqueous solution, thus paving the way for a linker-free fluorophore evaluation that could save time-consuming functionalization steps when evaluating newly synthesized fluorophores.²¹⁴ That study evaluated several of the same fluorophores evaluated in the Dempsey et al. paper as well as some that were not, including BODIPY FL.

We have included here a summary of switching properties of selected organic fluorophores (Table 1) which shows duty cycle, number of detected photons per localization, and number of switching cycles before bleaching, by thiol-,¹²⁸ TCEP-based switching,¹³⁵ or reductive caging^136,147 methods. For example, using thiol-based switching, many fluorophores have a good duty cycle, but Alexa Fluor 647 also emits a large number of photons per localization and exhibits a high number of cycles, and Alexa Fluor 647 therefore remains the most popular choice to date. On the other hand, the fluorophore Cy3B works quite well reductive caging for approximately one photoswitching event and a very high number of detected photons per localization.

Table 1.

Switching properties of selected organic fluorophores for SR-imaging by SMLM.

Organic dye	λex/ (nm)	λem (nm)	λon/off (nm)	ε (M−1cm−1)	QY	Duty cycle	# Photons	# Cycles	Refs
Alexa Fluor 647a,f	650	665	405/647	239,000	0.33	1.2×10−3	5,202	26	128
Alexa Fluor 647c,f	650	665	405/647	239,000	0.33	1.3×10−4	~2400	NR	135
Alexa Fluor 750a,f	749	775	405/750	240,000	0.12	1×10−4	703	6	128
Alexa Fluor 750c,f	749	775	405/750	240,000	0.12	4×10−4	2800	NR	135
ATTO 488a,f	501	523	405/488	90,000	0.8	2.2×10−3	1,110	49	128
ATTO 488b,g	501	523	405/488	90,000	0.8	2×10−3	~104	~1	128,136
BODIPY FLa,g	503	512	405/488	80,000	0.97	1.8×10−2	756	3.4	214,215
Cy3Ba,f	559	570	405/561	130,000	0.67	4×10−4	2,057	5	128
Cy3Bb,g	559	570	405/561	130,000	0.67	3×10−3	~105	~1	128,136
Amino-DCDHFe,i	594	613	407	54,100	0.39	NR	2.3×106	1	212
DTE (4-Et)e,f	488	582	488	37,300	0.45	4×10−4	200-300	14	204
Dyomics 654a,f	654	675	405/647	220,000	NR	1.8×10−3	3,014	19	128
DyLight 750a,f	752	778	405/750	220,000	NR	2×10−4	749	6	128
ER-Tracker Reda,h	587	615	405/561	64, 300	NR	3×10−4	820	28	216
HMSiRe,g	650	672	NR	100,000	0.39	NR	2,600	≥20	164
PA-JF549d,h	561	571	405	101,000	0.88	1×10−5	1.9×104	1	156,168
PA-JF646d,h	637	664	405	152,000	0.54	1.6×10−6	NR	1	156,168
OA-1 (open)e,j	532	650	355/532	83,000	0.02	NR	222	NR	189
SiRa,h	645	662	NR/640	100,000	0.39	NR	630	NR	154
SP (open)e,j	532	665	UV/532	52,000	0.17	NR	1.8×105	NR	112,217

Note: λ_ex (excitation wavelength); λ_em (emission wavelength); λ_on/off (wavelengths for switching the fluorescence on and off, respectively); ε (extinction coefficient); QY (fluorescence quantum yield); duty cycle (fraction of time spent in the on state prior to photobleaching); # Photons (number of detected photons per localization event); # cycles (number of switching cycles); Refs (journal references for published data).

^{‘a’, ‘b’, ‘c’, ‘d’, and ‘e’}indicate the composition of the imaging cocktail: a) deoxygenated + thiol; b) deoxygenated + MVAA (methyl viologen and ascorbic acid) where fluorophores had been previously reduced with NaBH₄; c) deoxygenated + TCEP + MVAA; d) deoxygenated; and e) no additives to the aqueous buffer.

^{‘f’, ‘g’, ‘h’, ‘i’ and ‘j’,}indicate the specimen used for fluorophore characterizations: f) fixed cells; g) antibody adsorbed to coverglass; h) live cells; i) fluorophores immobilized in PMMA (for quantum yield measurements) or gelatin (for other measurements); j) fluorophores on nanoparticle^s. ‘N.R.’ indicates the parameter was not reported.

3.2. Fluorescent proteins

FPs are a powerful class of fluorophores which have been widely utilized in SMLM since its inception. Recent articles have recently reviewed FPs for SMLM^{19,59,65,218,219} and we provide here a relatively brief summary with a few key details. A major benefit of FPs is that they are genetically encoded and do not require addition of exogenous ligands, thus enabling labeling of target proteins with high specificity and high efficiency (potentially stoichiometric). FPs are also compact, with molecular weights near ~30 kDa and a size of ~3-4 nm across, so there is minimal linkage error between the chromophore and the target. In terms of switching properties, FPs for SMLM tend to be substantially dimmer than the preferred organic fluorophores, although many of the preferred FPs have duty cycles in the range of 10⁻⁴-10⁻⁶ which are over 10-100 times lower (i.e., better) than that of the preferred organic fluorophores (Table 1). Finally, many FPs tend to dimerize or aggregate, leading to artifacts in protein localization, and researchers have invested considerable time to design probes that are sufficiently monomeric.

Based on their switching properties, FPs for SMLM may be divided into three classes: irreversible photoactivatable FPs (PA-FPs), irreversible photoconvertible FPs (PC-FPs), and reversible photoswitchable FPs (PS-FPs). The representative photoswitching mechanisms are illustrated in Figure 23 and a summary of photoswitching properties of selected FPs are shown in Table 2.

Figure 23.

Representative photo-modulation mechanisms of FPs (only the FP chromophores are shown). A) Photoactivatable FPs (PA-FPs) such as PA-GFP convert irreversibly from a native dark from to a green fluorescent form by photo-induced decarboxylation of a glutamate residue (pink arrow) near to the chromophore. B) Photoconvertible FPs (PC-FPs) such as EosFP convert irreversibly from a green fluorescent form to a red fluorescent form upon illumination with violet or ultraviolet light. The photochromic shift occurs due to cleavage of the peptide backbone which extends the conjugation of the chromophore and red-shifts the absorption and emission spectra by 50–60 nm. C) Photoswitchable FPs (PS-FPs) such as Dronpa undergo reversible cis–trans isomerization, where the cis form is fluorescent and the trans form is dark. Note that certain FPs such as PAmCherry ²²⁰ and IrisFP ²²¹ may use a combination of these switching mechanisms.

Table 2.

Photo-switching properties of selected FPs for SR-imaging by SMLM.

FPs	λex (nm)	λem (nm)	λon/off (nm)	ε (M−1cm−1)	QY	on/off contrast	Duty Cycle	Photons/ localization	# Switching Cycles	clustering	Ref.
Dendra2	490/553	507/573	405/488	45,000/35,000	0.5/0.55	300	4×10−6	686	2.7	-	56,62,63,240
Dreiklang	511	529	340/412	83,000	0.41	75	N.R.	400-700	>160	N.R.	61,237,241
Dronpa	503	518	405/488	95,000	0.85	20	8×10−4	262	60-70	-	25,63,236,242
mEos2	506/573	519/584	405/573	56,000/46,000	0.84/0.66	200	3×10−6	1000	2.8	+	62,63,227,228
mEos3.2	507/572	516/580	405/572	63,400/32,200	0.84/0.55	N.R.	3×10−6	1000	2.4	-	62,63,229
mMaple	489/566	505/583	380/566	15,000/30,000	0.74/0.56	400	2×10−6	1000	3.4	+	62,63,228
mMaple3	489/566	506/583	405/561	N.R.	N.R.	N.R.	6×10−7	675	2.8	-	63
PAmCherry1	564	595	~370	18,000	0.46	4000	8×10−6	725	1	+	62,63,223
PAGFP	504	517	405	17,400	0.79	200	1.3×10−3	313	N.R.	-	49,63

Note: λ_ex (excitation wavelength); λ_em (emission wavelength); λ_on/off (wavelengths for switching the fluorescence on and off, respectively); ε (extinction coefficient); QY (fluorescence quantum yield); on/off contrast (ratio of intensity of the emissive form to the dark form); duty cycle (fraction of time spent in the on state); photons/localization (number of detected photons per localization event); # switching cycles (number of switching cycles); ‘N.R.’ indicates the parameter was not reported;‘+’ indicates clustering or aggregation was observed in the ClpP assay by Wang et al.⁶³ (‘-’ indicates no clustering).

The first PA-FP, PA-GFP, was reported by Patterson et al.⁴⁹ Illumination at 405 nm shifts its absorption peak from 400 nm to 504 nm, with a 200-fold fluorescence enhancement when excited at 488 nm (Figure 23A, Table 2). van Thor et al. used X-ray crystallography to determine that the photoactivation process results from the light-induced decarboxylation of a glutamate residue near to the chromophore which irreversibly converts PA-GFP from a dark, neutral cis form to a brightly fluorescent, anionic cis form (Figure 23A).²²² The photoactivated form of PA-GFP shows good stability at physiological temperature and pH environment,⁴⁹ accompanied by good overall brightness and fusion protein compatibility. PA-GFP, as well as other FPs, were utilized by Betzig et al. in their original 2006 paper.²⁷

PA-FPs can implement fluorogenic emission in the green (PA-GFP), red (PA-TagRFP,²²³ PA-mCherry1²²⁰ and PA-mRFP-1⁵⁴) and far-red (PA-mKate²²⁴) channels, allowing multi-color SR imaging by SMLM. Subach et al. implemented two-color SMLM imaging using PA-GFP tagged clathrin light chain (CLC) and PAmCherry1 tagged transferrin receptor (TfR) in COS-7 cells to show the nanoscale distributions of CLC and TfR (Figure 24A).²²⁰ PA-mKate shows far-red shifted fluorescence emission and may be used in three-color SR imaging when paired with other green and red emitting PA-FPs. Gunewardene et al. performed three-color SMLM in cells co-expressing PAmKate-TfR, PAmCherry1b-actin, and Dendra2-hemagglutinin (HA, Figure 24D).²²⁴ Their results revealed correlations between these membrane proteins and membrane-associated actin structures: TfR and HA are both correlated with actin but are either uncorrelated or anticorrelated with each other. Like photoactivatable organic dyes, the best PA-FPs have several advantages including a low fluorescence background, a low duty cycle, and in some cases a single burst of photons. However, PA-FPs are invisible prior to photoactivation so that it may be difficult for researchers at the microscope to select transformed cells exhibiting the phenotype of interest. This issue is addressed by using PC-FPs.

Figure 24.

Multicolor SMLM using fluorescent proteins. A) Two-color SMLM in COS-7 cells showing (i) PA-GFP-tagged clathrin light chain (green), (ii) PAmCherry1-tagged transferrin receptor (red), and (iii) overlay. B) Two-color SMLM of mitochondria in EpH4 cells showing outer membrane labeled with (i) PAmCherry1-Lk-BclXl (red), (ii) inner membrane labeled with BCS1L-Lk-rsKame (green), (iii) two-channel overlay, and (iv) intensity profile along white line in iii. C) Two-color SMLM of tdEos-tagged paxillin (green) and PS-CFP2-tagged zyxin (red) in an HFF-1 cell reveals little overlap despite appearing to co-localize using conventional microscopy. D) Three-color SMLM image of Dendra2-hemagglutinin (cyan), PAmCherry1b-actin (yellow), and PAmKate-transferrin receptor (magenta) in fixed HAb2 mouse fibroblast cells. Scale bars are all 500 nm except as indicated otherwise. Adapted by permission from Springer Customer Service Centre GmbH: Springer Nature, Ref. 220, Copyright 2009. Adapted with permission from Ref. 231, Copyright 2014 National Academy of Sciences, U. S. A. Adapted with permission from Ref 232, Copyright 2007 National Academy of Sciences, U. S. A. Adapted from Ref. 224, Copyright 2011, with permission from Elsevier.

PC-FPs are generally photoconverted from one fluorescent state to another by violet or blue light irradiation. The photoconversion process is attributed to the photo-cleavage of the peptide backbone on histidine, which creates a new double bond formation between its C_α and C_β atoms and thereby extends the π-electron conjugation from the chromophore to the histidine imidazole ring (Figure 23B).^225,226 For example, EosFP⁵⁷ exists initially in a green fluorescent state and upon illumination with UV or near-UV light will convert to a red fluorescent form (Figure 23B), permitting on/off and off/on switching in green and red channels, respectively. The activation light (~400 nm) is well separated in wavelength from the excitation wavelength (~510 nm for the green channel and ~570 nm for the red channel), which reduces the risk of unintended photoconversion.

Many PC-FPs are popular in SMLM including EosFP, mMaple, Dendra and subsequent variations, due their high on-off contrast, low duty cycle, and relatively large number of detected photons per localization (Table 2).^{56,62,63,227–230} mEos2 and mEos3.2 were engineered to have a reduced tendency to dimerize or aggregate while still retaining the spectroscopic properties of EosFP.^57,227,229 mMaple2 also shows minimal aggregation with good switching properties while exhibiting a superior efficiency of expression/folding/maturation.^63,228 mEos4 was developed to be able to survive fixation conditions used for electron microscopy (EM) and thus nicely enables correlative SMLM and EM.²³⁰ A major limitation of PC-FPs, however, is their dual-channel emission property; by converting from a green fluorescent form to a red one, two fluorescent channels are occupied, creating serious challenges for multi-channel imaging with PC-FPs.

PS-FPs show reversible on/off photoswitching between fluorescent and nonfluorescent states. This process is reminiscent of reversible thiol-based photoswitching of organic dyes, however, special reagents (e.g., a deoxygenated solution containing a thiol, Section 3.4) are not generally required to switch PS-FPs. Dronpa⁵⁰ and its variants ^233,234 are among the best known examples of PS-FPs. Intense irradiation with ~488 nm light converts Dronpa from an anionic cis form that is fluorescent to a neutral trans form that is nonfluorescent, while illumination of the nonfluorescent form with ~405 nm light converts it back to the fluorescent form (Figure 23C). This deactivation/activation cycle can be repeated many times and allows Dronpa’s application to single-molecule localization techniques and SMLM.^232,235,236 For instance, Rosenbloom and co-workers paired the Dronpa variant rsKame with PAmCherry1 to capture two-color SMLM images of the inner and outer mitochondrial membranes (Figure 24).²³¹ Other Dronpa variants, like bsDronpa,²³⁴ Padron²³⁴ and rsFastlime²³³ have been developed which show markedly improved photostability (8.5 to 34-fold) compared with Dronpa.

The PS-FPs, Dreiklang²³⁷ and reversibly switchable enhanced GFP (rsEGFP)²³⁸ reported by Brakemann, Grotjohann and co-workers also have good fatigue resistance, allowing photoswitching for more than 160 and 1000 cycles, respectively. Among them, Dreiklang shows a larger number of detected photons per localization than Dronpa (Table 2). It is worth mentioning that the fluorescence excitation of Dreiklang is decoupled from its photoswitching. Light irradiation around its excitation maximum (511 nm) does not activate dark Dreiklang, thereby avoiding unintended deactivation or activation and enabling control over the length and probability of a photoswitching cycle. However, Dreiklang is sensitive to temperature, exhibiting poorer fatigue resistance and duty cycle at higher temperature.

There are yet other classes of fluorescent proteins, including some which exhibit a combination of photoconversion and photoswitching behaviors. For instance, mIrisFP is photoconvertible between green and red fluorescent states while both these two states exhibit reversible photoswitching between on and off forms.²²¹ This dual PC/PS mode of mIrisFP enables the combination of pulse-chase experiments with SMLM to implement dynamic studies of protein complexes.²³⁹

Although FPs are generally dimmer compared with the preferred organic dyes,¹²⁸ their high compatibility and specificity of labeling makes them particularly useful for living specimens.⁵⁹ Finally, it should be mentioned that expression of exogenous protein tags raises several issues. For instance, overexpression or oligomerization/aggregation may perturb the target protein or specimen, and their use may be impacted by species-specific codon usage, maturation time of the fluorescent tag, and changes in the expression levels of the fusion protein (see Section 9.2).⁶³

3.3. Fluorescent quantum dots

Quantum dots (QDs) are nanoscale semiconductors whose fluorescence emission spectra are tunable by varying particle size and composition (Figure 25).²⁴³ Compared with most organic fluorophores or fluorescent proteins, QDs generally exhibit intense brightness²⁴⁴ and resistance to photobleaching,²⁴⁵ both of which are valuable general properties for fluorescence imaging, but face challenges due to their relatively large size and limited cell permeability for live-cell applications. While spontaneous QD blinking is a well-known phenomenon that had been reported approximately a decade before the first SMLM publications³⁹, the duty cycle of spontaneously blinking QDs tends to be quite high and the switching rates rather heterogeneous, making the fluorophores unfavorable for SMLM. Nonetheless, Lidke et al.⁷² in 2005 and Lagerholm et al.²⁴⁶ in 2006 both demonstrated the identification of two QDs within a diffraction-limited region.

Figure 25.

Quantum dot emission spectra are tunable through control of A) particle size and B) particle composition. Reproduced with permission from Ref. 243, Copyright 2015 American Chemical Society.

Spontaneously blinking QDs have been used with good results for super-resolution optical fluctuation imaging (SOFI, Section 2.5),¹³³ however, SOFI attains improved spatial resolution through statistical analysis of signal fluctuations (an ensemble measurement) rather than requiring observation of individual fluorophores. In the original SOFI publication,¹³³ Dertinger and co-workers reported ~55 nm resolution (FWHM) using commercially available quantum dots consisting of CdSe cores surrounded by ZnS shells. Subsequent researchers were able to tune the blinking properties either by synthesis of QDs with a thinner ZnS shell²⁴⁷ or by addition of ascorbic acid in solution in order to increase the blinking rate.²⁴⁸ Through a variation of SMLM called QDB3 (Section 2.5), Wang et al. were able to use the spontaneous blinking of quantum dots to study cell surface receptors on breast cancer cells.¹³⁰

Hoyer and co-workers exploited stochastic ‘blueing’ (or blue-shifting) of QDs to enable use of QDs in a low duty cycle mode that is suitable for SMLM.²⁴⁹ In QD blueing, the CdSe cores gradually oxidized from the outside inwards such that the emission wavelength shift to shorter and shorter wavelengths over time until the CdSe core is oxidized and becomes nonluminescent. By this means, a detection band between the excitation wavelength and the unoxidized QD emission allows observation of individual QDs as they are stochastically blue-shifted but before irreversible bleaching. Tehrani et al.²⁵⁰ and Xu et al.²⁵¹ extended the application of QD blueing to multi-color SMLM and 3D-SMLM. As an illustration in Figure 26, QD565 and QD705 were used for studying the morphology and localization of microtubules and mitochondria. Their work also demonstrated that the rate of QD blueing can be tuned and optimized by modulation of laser intensity and the mounting medium (oxygen concentration), resulting in better localization data for SR image reconstruction. Alternative strategies for producing photoswitchable QDs with low duty cycle that are potentially well-suited to SMLM include using chemical caging,²⁵² light-driven modulation,²⁵³ or FRET when decorated with a photoswitchable organic fluorophore.²⁵⁴

Figure 26.

Two-color SMLM using QD blueing (stochastic photoconversion). A-B) Widefield and corresponding SMLM fluorescence image of a fixed HepG2 cell immunostained for microtubules with QD565 (cyan) and for mitochondria with QD705 (red). C-F) Zoom-in views of boxed regions from A and B. Scale bars are 2 μm (A-B) and 500 nm (C-F). Reproduced with permission from Ref. 251, Copyright 2015 American Chemical Society.

3.4. Imaging cocktails

Fluorophores are sensitive to their chemical environment, and it is common in SMLM to use various “imaging cocktails” in order to facilitate or enhance photoswitching properties of a given fluorophore. While a range of imaging cocktails have been reported in the literature, including several fluorophore-specific examples described above, we will confine our comments to a few limited examples below that are of particular importance or special interest.

Dempsey et al. showed that a wide range of commercial organic fluorophores photoswitch well in deoxygenated solutions containing a thiol and a buffer (e.g., 100 mM Tris pH 7.4).¹²⁸ Deoxygenation is often achieved enzymatically by addition of a mixture of glucose oxidase and catalase²⁵⁵, along with glucose (the substrate of glucose oxidase), to convert glucose and dissolved oxygen into gluconic acid while lowering the concentration of dissolved oxygen. Several alternative oxygen scavenging systems have been reported in the SMLM and single-molecule fluorescence literature, including the use of other enzymes such as protocatechuate deoxygenase, galactose oxidase, pyranose oxidase, and the proprietary enzyme Oxyrase, the latter two of which are notable for not leading to acidification of the imaging solution.^256–259 Oxygen may be depleted nonenzymatically, as well, for instance by using a solution containing methylene blue and a thiol.²⁶⁰ Similarly, a range of thiols have been used, including 2-mercaptoethanol, 2-mercaptoethylamine, cysteine-methyl-ether, and others.^{76,77,128,141} Live-cell SMLM with photoswitchable organic fluorophores requires additional care, particularly in formulation of oxygen scavenger systems, because the high concentrations of glucose (~10%) used with the popular glucose oxidase and catalase oxygen scavenger system can cause osmotic shock. For example, Jones et al. and Shim et al. used a modified oxygen scavenger system containing ~2% glucose for live-cell SMLM.^216,261 Conveniently, reduced glutathione is present in living mammalian cells at a concentration of ~0.5-10 mM²⁶² and provides a thiol for photoswitching.

Many other imaging cocktails have been used with photoswitchable SMLM fluorophores. Olivier et al. showed that addition of the triplet state quencher cyclooctatetraene to deoxygenated solutions containing a thiol could boost the number of photons detected per localization for Alexa Fluor 647 by up to 3.5×, although it was not clear whether the low duty cycle was maintained under these conditions. In a separate report, Olivier et al. also showed that the proprietary mounting agent VECTASHIELD, alone, was able to produce good photoswitching from several cyanine dyes, including Alexa Fluor 647, without requiring additional reagents (e.g., enzymes or thiols), although its use is apparently incompatible with 405 nm activation of dark-state fluorophores due to background fluorescence from the violet-colored solution.²⁶³ Lee et al. and Klehs et al. found that heavy water (D₂O) moderately boosted photon yields or emission rates for a range of cyanine and oxazine fluorophores in SMLM.^264,265 The use of TCEP, together with an oxygen scavenger system and a mixture of methyl viologen and ascorbic acid, was described above for the switching of the cyanine dyes Cy5 and Cy7 (Section 3.1.1, Figure 10A, Figure 9C).¹³⁵

While fluorescent proteins tend to be less sensitive to their chemical environment than small organic fluorophores, several imaging cocktails have been reported in the literature in order to tune fluorescent protein switching properties for SMLM, including: the boosting of photon yields of several fluorescent proteins by up to ~50% by imaging in heavy water²⁶⁶; the chemical caging of mCherry and some other red fluorescent proteins using a thiol and its subsequent imaging (uncaging) with or without the thiol present²⁶⁷; and the photoconversion of eGFP to a red absorbing/emitting form when illuminating eGFP in the presence of riboflavin.²⁶⁸ Although specialized imaging cocktails are not widely used in SMLM with fluorescent proteins at present, this area seems ripe for further exploration given the wide range of fluorescent proteins and associated sensors that have been created over the past two decades.

In summary, a deoxygenated solution containing thiol and buffer works well with SMLM for a broad variety of photoswitchable organic fluorophores while specialized imaging cocktails are not generally required for SMLM with fluorescent proteins, spontaneously-blinking switches such as HMSiR, or binding-based switches used in DNA-PAINT (see Section 4). Live-cell imaging, on the other hand, requires additional care.

4. Fluorophore switching by binding-based methods (PAINT, DNA-PAINT, etc.)

The visualization of sequential binding of fluorescent probes is known as PAINT, or Point Accumulation for Imaging by Nanoscale Topography, is a SMLM method with unique properties compared to those in which the fluorescent probes begin bound. PAINT was first reported in 2006 by Sharonov et al.²⁹ and was based on related work published during the prior year by the same group.^97,269,270 In its initial implementation, PAINT was used to image large unilamellar vesicles (LUVs) with the fluorophore Nile red, which is weakly fluorescent in water but becomes much brighter in a lipid environment. Transient binding events of Nile red to LUVs created brief flashes similar to blinking behavior in photoswitchable or photoactivatable fluorophores and over time were able to provide super-resolution information on the vesicle positions and diameters (Figure 27). Nile red has also been used to study supported lipid bilayers and cell membranes,^29,119 while another turn-on probe, merocyanine 540, has been used to study phase segregation in supported lipid bilayers.²⁷¹

Figure 27.

SMLM imaging using PAINT (Point Accumulation for Imaging by Nanoscale Topography). A) Conventional fluorescence image and B) corresponding SMLM image of vesicles bound to a coverglass acquired by imaging flashes of light caused by binding of individual molecules of the fluorophore Nile red. Reproduced with permission from Ref. 29, Copyright 2006 National Academy of Sciences, U. S. A.

The PAINT approach for SMLM, in which fluorophores are initially diffusing in solution and bind stochastically to a specimen during acquisition of a stack of single molecule images, has been extended to a wide range of probes and specimens. In universal PAINT, or uPAINT, small proteins of interest were probed with nonblinking fluorophores bound to a ligand or antibody with affinity for the target protein.²⁷² This allowed the tracking of membrane receptors over a relatively long time in live cells and resembles sptPALM experiments performed using photoactivatable or photoswitchable fluorescent proteins.¹¹¹ In interface PAINT, or iPAINT, Aloi et al.²⁷³ coupled a photoactivatable rhodamine fluorophore¹⁶⁷ to polyethylene glycol in order to develop an interface-binding probe for visualization of solid/liquid, liquid/liquid, and liquid/air interfaces at nanometer scale resolution. Kiuchi et al. developed small, protein-based PAINT probes, which they called IRIS probes, in order to avoid the use of bulky antibodies which tend to broaden labeled structures of interest.²⁷⁴ For instance, the actin-binding probe LifeAct²⁷⁵ was coupled to ATTO 488 and was used to image actin in permeabilized cells, and a number of other protein fragments were created for use as IRIS probes starting from a careful review of known protein-protein interactions.²⁷⁴ Similar sequential binding approaches have been used for nonbiological specimens including to record images of polymeric nanogratings and to map out electric field hot spots near nanoparticles.^276,277

The SMLM variant DNA-PAINT uses reversible binding of diffusing fluorescently labeled single-stranded DNA probes to image structures that are decorated with complementary DNA strands.²⁷⁸ Unbound probes diffuse in solution, appearing as a background haze, while immobilized probes exhibit a bright spot. The rates of binding and unbinding are easily tuned, where the binding rate is primarily controlled by adjusting the imager strand concentration and the unbinding rate is primarily controlled by the length and base pair content of double-stranded DNA formed upon hybridization. Virtually any bright, photostable fluorophore may be used in DNA-PAINT, since it doesn’t require fluorophore photoswitching or photoactivation, and the number of localizations attainable is only limited in principle by the amount of time the user is willing to wait. DNA-PAINT has been used to image DNA origami structures as well as cellular specimens labeled with single-stranded DNA-antibody conjugates (Figure 28, Figure 39).^279–285

Figure 28.

SMLM imaging using DNA-PAINT. A) DNA origami structure designed to spell ‘Wyss!’ within an area of ~50 nm × 75 nm. B) Representative SMLM image of a single Wyss! DNA origami and C) a class average obtained from a collection of about 80 imaged origami. Adapted by permission from Springer Customer Service Centre GmbH: Springer Nature, Ref. 281, Copyright 2016.

Figure 39.

Four-color Exchange-PAINT image of a cell that has been immunostained for tubulin, mitochondria, peroxisomes, and the Golgi apparatus. Scale bar is 5 μm. Adapted by permission from Springer Customer Service Centre GmbH: Springer Nature, Ref. 279, Copyright 2014.

PAINT, and several variations, thereof, have delivered promising results for SMLM but have typically faced a tradeoff between speed of imaging and high background signal. The binding rate in PAINT, or k_on, is a diffusion-limited process whose rate primarily controlled through choice of concentration of the probe, while the unbinding rate, or k_off, is controlled through the strength of the interaction of the probe with its target. Speeding up the data acquisition rate requires an increase in the diffusing probe concentration along with a decrease in the affinity of the probe for its target (e.g., for DNA-PAINT, a shorter imager strand). However, doing so provides less time during which to acquire signal from the bound probe, while also increasing the background signal resulting from the higher concentration of diffusing, unbound probes. In many implementations of PAINT, acquisition frame rates of ~20 Hz were used, which can be prohibitive when many frames of data are desired.

One strategy to decrease background signal with PAINT and DNA-PAINT is to utilize illumination of a thin volume. When a specimen is situated very close to the coverglass (<200 nm), total internal reflection illumination may be used to confine the excitation volume to be limited to a 100-200 nm thick layer above the coverglass. Imaging of specimens which are located substantially higher than 200 nm requires widefield illumination of a much larger thickness which in turn leads to more background signal. Legant et al.¹²⁶ reported a potential solution to the imaging of thick specimens by combining PAINT with lattice light sheet (LLS) microscopy²⁸⁶ using commercially available and novel fluorescent probes (Figure 29).²⁸⁶ In this arrangement, the LLS excited an approximately 1 μm thick plane while repeatedly scanning thick samples (~5-20 μm thick) back and forth along the axial dimension of the detection path. A frame acquisition rate of ~40 Hz was used, although it required relatively long overall acquisition times of several days per image; shorter acquisition times would be possible with thinner specimens. This work used both commercially available fluorescent probes (BODIPY TR methyl ester) as well as custom-synthesized rhodamine probes (AzepRh, JF₆₄₆-Hoechst) that had suitable affinity for membranes (BODIPY TR, AzepRh) or DNA (JF₆₄₆-Hoechst) to enable PAINT-based SMLM.

Figure 29.

A) Volume rendering of an entire dividing LLCPK1 cell imaged by LLS-PAINT. The plasma membrane was labeled with a wheat germ agglutinin-Alexa Fluor 555 probe and intracellular membranes were labeled with BODIPY TR methyl ester. Bounding box is 50 μm × 27 μm × 20 μm. B) Cutaway section through the cell in A revealing various labeled membrane structures and histones (mCherry-H2B, imaged by diffraction-limited imaging prior to LLS-PAINT). Adapted by permission from Springer Customer Service Centre GmbH: Springer Nature, Ref. 126, Copyright 2016.

A second strategy to decrease background signal would be to utilize turn-on probes that emit little to no signal when freely diffusing but which become much brighter when transiently bound. Nile red and merocyanine 540 exhibit behavior of this sort but are limited to very specific membrane targets. In DNA-PAINT, it was reported that the fluorescence of ATTO 655 on unbound imager strands is partly quenched relative to the bound state and exhibits 70% increase of fluorescence brightness upon binding.²⁷⁸ If a quenching strategy could be developed in DNA-PAINT that would enable much higher turn on ratios of, for instance 1:100, it would help reduce background levels and/or speed up data acquisition rates. Turn on ratios of more than 1:10,000 were achieved with the fluorogen malachite green when bound to engineered fluorogen-activating single chain antibody tags,²⁸⁷ and with further engineering these²⁸⁸ or similar approaches may help improve background and or frame rates in PAINT. Auer et al. recently implemented DNA-PAINT with a FRET-based scheme designed to reduce background signal in which an antibody-tethered acceptor fluorophore would only emit signal when a diffusing donor-labeled oligonucleotide bound to a complementary DNA sequence in proximity to the acceptor fluorophore.²⁸³

5. Fluorophore switching by catalysis, electrochemistry, or complexation

In addition to (photo)chemical reactions and binding/unbinding processes, stochastic switching of fluorophores can be achieved using catalytic reactions (enzymatic or otherwise), electrochemical reactions, or metal ion complexation.^{118,289–296} The timescale of switching can be tuned by controlling kinetic rates of the enzymatic, catalytic or electrochemical reactions, or by changing the concentration of the metal ion for complexation. These switching methods are compatible with SMLM of biological specimens and can be extended to other regimes including the mapping of reactivity patterns on single nano-catalysts.

A variety of synthetic substrates based on fluorescein,²⁹⁷ rhodamine,²⁹⁸ rhodol,²⁶¹ and DCDHF¹²³ fluorophores have been used to develop novel enzymatic fluorogenic reactions, due to the convenient incorporation of enzyme substrate groups on these fluorophores. For instance, Lee and co-workers¹²³ developed a SMLM method based on enzymatic activation of DCDHF fluorophores which had been caged by a nitro group. In this scheme, nitroreductase (NTR) enzymes³⁰⁰ could be targeted to structures of interest such that diffusing nonfluorescent DCDHF substrate molecules could become reduced, activated, and imaged with high intensity excitation light that was used to localize and photobleach activated DCDHF fluorophores before they diffused very far from the NTR enzyme (Figure 30).¹²³ Adjusting the concentration of nitro-DCDHF or slowing the enzymatic rate using inhibitors allows tuning of the rate of activation to produce a higher or lower density of fluorophores. A similar approach has been used in live eukaryotic cells by Halabi et al. who developed novel, non-fluorescent diazoindanone probes and imaged their activation to fluorescent rhodol fluorophores by endogenous cellular esterases in order to map enzymatic activity using SMLM.³⁰¹

Figure 30.

A) Fluorogenic reactions of nitro-DCDHF based on enzyme-catalyzed conversion of caging groups. B) Conventional fluorescence image (and corresponding bright-field image, inset) of B. subtilis labeled with nitro-DCDHF. C) SMLM image of blue box area in B, showing localization information of the membrane and punctate spots. D) Temporal progression was revealed for the localizations in the punctate spots (blue dashed box area in C). E) SMLM image of green-boxed region in B. F) Enzyme diffusing was demonstrated in green dashed box area of E. Scale bars: 1 μm (B, C, E); 0.5 μm (D, F). Adapted from Ref. 123 with permission of The Royal Society of Chemistry.

The redox conversion between resazurin and resorufin has been used for a variety of catalytic or electrochemical switching experiments.^{293,296,302–305} Resazurin is a small nonfluorescent, organic molecule that can be reduced to resorufin, which is fluorescent, on an electrode or via N-deoxygenation by NH₂OH in the presence of a gold catalyst. Xu et al. have used gold nanoparticles to reduce diffusing resazurin molecules to their fluorescent resorufin form and were able to show that the catalytic activity at sharp corners of gold nanoparticles was higher than that of edge regions which in turn was higher than surface facets (Figure 31).³⁰⁴

Figure 31.

A) Catalyst-enhanced conversion of nonfluorescent resazurin to fluorescent resorufin. B) SMLM image showing a density map of activated resorufin molecules relative to a triangular gold nanoplate. Adapted with permission from Ref. 304, Copyright 2013 American Chemical Society.

Rather than use a turn-on strategy, reversible binding of the fluorescence-quenching ion Cu²⁺ in close proximity to a fluorophore has been used to modulate emission of the fluorophore between emissive and nonemissive states.^118,289,306 Tuning of the concentration of Cu²⁺ affords some control over the duty cycle of the fluorophore. In addition to Cu²⁺, other paramagnetic metal ions such as Ni²⁺, Co²⁺ or Mn²⁺ may also be candidates for metal ion complexation switching.

6. Sample labeling

One of the key advantages of fluorescence microscopy is the ability to visualize structures of interest with high molecular specificity and low background signal. This is achieved through a variety of schemes for labeling biological specimens that span the range of covalently or noncovalently binding small molecules, fluorescent proteins, antibodies, oligonucleotides, proteins or peptides. All of these approaches have been used with good results in SMLM, and in many cases, distinct advantages of the various labeling schemes have been powerful tools in enabling key experiments. In this section, we provide an overview of methods for sample labeling. We explicitly focus here on strategies to label specimens with organic fluorophores since labeling with fluorescent proteins is relatively standardized.

6.1. Immunostaining

Immunostaining is perhaps the most widespread technique used to label proteins with organic fluorescent dyes for SMLM in fixed specimens. The basic steps of immunostaining consist of fixation to preserve the structure(s) of interest, labeling of the specimen with dye-conjugated antibodies, and removal of unbound antibodies through washing.⁷⁸ In practice, a great deal of optimization is often required to produce good immunostained specimens including the choice of fixative, antibodies, detergent, blocking agents, wash stringency, and a host of optional, but often important additional steps, such as pre-fixation extraction by treatment with detergent for studies of the cytoskeleton, reduction with sodium borohydride to quench (glutaraldehyde) fixation-induced fluorescence background, and postfixation with glutaraldehyde to help avoid antibody dissociation from the specimen. Practitioners most commonly use indirect immunofluorescence in which nonfluorescent “primary antibodies” bind a target on the specimen and are then labeled with a fluorescent “secondary antibody” that binds the primary antibody. While some labs conjugate fluorophores to (secondary) antibodies in house in order to achieve a specific dye/antibody ratio or another key property, many commercial sources of antibodies also work well. The fluorescent secondary antibodies can be mixed and matched for multichannel imaging^137,138 and many practical guides are available.^{14,15,17,307–309} The use of indirect immunofluorescence with primary/secondary antibodies often limits the maximum number of channels to ~3 due to the requirement that each primary antibody is antigenically distinct for unique binding by secondary antibodies.

Importantly, the finite size of antibodies (~10 nm) leads to a spatial broadening of structures of interest that is measurable by SMLM. For instance, the binding of primary and secondary antibodies to microtubules, which have an outer diameter of ~25 nm, produces a broadened structure by SMLM that has a full-width at half maximum of ~50-60 nm and contains a ‘hollow’ structure due to the primary antibody and microtubule itself which are surrounded by the fluorescent secondary antibody layer.^{135–137,310} In some cases, particularly when there is bright signal from fluorescently-labeled primary antibodies without the need for amplification by secondary antibodies, the secondary antibody can be omitted and the amount of label-induced broadening may be reduced.³¹⁰ The use of direct immunofluorescence with very small antibodies, such as camelid nanobodies which are ~2-3 nm in size, can further reduce label-induced broadening to a negligible amount (Figure 32)^310,311 and nanobodies against an increasing variety of endogenous proteins and have been used in SMLM to probe, for example, tubulin,³¹¹ several components of the nuclear pore complex,³¹² and vimentin.³¹³ Pleiner et al. reported anti-immunoglobulin G (IgG) nanobodies as compact secondary antibodies which also help to minimize antibody-induced broadening compared to full IgG-based secondary antibody labeling.³¹⁴

Figure 32.

Antibody broadening of features for immunostained microtubules in fixed Ptk2 cells expressing tubulin-YFP. A) Indirect immunostain of tubulin (left), direct immunostain with anti-GFP nanobody on cells expressing tubulin-YFP (middle), and direct immunostain of YFP on cells expressing tubulin-YFP using an anti-GFP antibody (right). B) Cross-sectional profiles for boxed regions in A showing substantially narrower profile using the nanobody. C) Illustration of linkage error using either indirect immunofluorescence with full immunoglobulin G antibodies (left) or direct immunofluorescence with nanobody-based probes (right). Scale bars in A-C are 500 nm. Adapted by permission from Springer Customer Service Centre GmbH: Springer Nature, Ref. 310, Copyright 2012. Adapted by permission from Springer Customer Service Centre GmbH: Springer Nature, Ref. 309, Copyright 2015.

6.2. Protein fusion tags for labeling with organic fluorophores

Rather than use bulky antibodies to target organic dyes to proteins of interest, it is possible to express the protein of interest labeled with a protein tag that can covalently or noncovalently bind a small organic group.^24,315–319 These methods also have the potential advantage of compatibility with living specimens, unlike antibody staining which in generally is only used for fixed specimens. Many tags and techniques have been developed for labeling of specific proteins with organic fluorophores, such as FlAsH-tag,³²⁰ TMP-tag,³²¹ SNAP-tag,³²² CLIP-tag,³²³ and Halo-tag³²⁴ (Figure 33). These protein tags are genetically encodable and have a small size, as do fluorescent protein tags, but allow researchers to harness the (often) superior optical properties of organic fluorophores for live- or fixed-cell imaging. Each has their own challenges regarding specificity, efficiency of labeling, speed of labeling, and their potential for perturbation of the specimen.

Figure 33.

Schematic illustration of representative techniques for targeting organic fluorophores to a tagged protein of interest, including A) FlAsH-tag (tetra-Cys tag); B) TMP-tag; C) SNAP-tag, CLIP-tag, and Halo-tag (Nu=nucleophile, F=fluorophore).

Biarsenical probes that bind tetracysteine motifs genetically introduced into proteins (Figure 33A) have been used in SMLM, both with the original fluorescein bisarsenic “FlAsH” probes^320,325 as well as novel bisarsenical cyanine probes³²⁶. Figure 33B illustrates the interaction between E. coli dihydro-folate reductase (eDHFR) and trimpethoprim (TMP), which binds eDFHR with high affinity (K_D ~1 nM) and high selectivity. This enables the use of TMP-linked fluorophores, in which a protein of interest that is fused with the eDHFR tag is labeled with a TMP-fluorophore conjugate.³²¹ Various fluorophores have been used with TMP-tags, including the cell-permeable fluorophore ATTO 655 for labeling of eDHFR-tagged human histone H2B to study histone core protein dynamics by SMLM in live HeLa cells.³²⁷

Several covalent protein labeling strategies based on protein tags that react with specific small molecule ligands have been used in SMLM, including SNAP-tag,³²² CLIP-tag,³²³ and Halo-tag (Figure 33C).³²⁴ SNAP-tag and CLIP-tag are modified alkylguanine transferase DNA repair proteins which can react specifically and efficiently with benzylguanine (BG) or chloropyrimidine substrates (SNAP-tag) or with O²-benzylcytosine substrates (CLIP-tag) by forming a stable, covalent thioether bond. These have been used in several live-cell super-resolution microscopy studies including by SMLM and STED.^{154,178,261,328} For instance, Jones et al. performed live-cell SMLM on clathrin coated pits using SNAP-tagged clathrin together with Alexa Fluor 647-BG, a top dye for SMLM (Figure 40).²⁶¹ Importantly, since Alexa Fluor 647 is cell-impermeant, Jones et al. used electroporation to deliver the fluorophore to cells,²⁶¹ although special measures are not necessary when using cell-permeant fluorophores. Halo-tag is another broadly used protein tag for specific labeling and imaging of proteins of interest. With HaloTag, the terminal chloride on the HaloTag ligand undergoes nucleophilic attack by Asp106 of the HaloTag protein and results in the covalent linkage of the HaloTag to the HaloTag ligand and any associated cargo (Figure 33C).³²⁴ Like SNAP-tag, HaloTag has been used in a number of live-cell SMLM studies on organelle or protein function, particularly with cell-permeable fluorophores that require no electroporation or other special procedures for intracellular delivery.^{154,213,328,329}

Figure 40.

Live-cell SMLM reveals nanoscale features and dynamics. A) 3D SMLM images of clathrin-coated pits (CCPs) labeled with Alexa Fluor 647 in live BS-C-1 cells that express clathrin light chain tagged with SNAP (see Section 6.2). B) Two-channel, 3D SMLM image of a CCP with clathrin light chain (magenta) labeled as above and with its cargo transferrin (green) directly labeled with Alexa Fluor 568. C) 2D SMLM image of a cultured hippocampal neuron labeled with Dil showing changes to the dendritic structure including a growing spine or filopodium (green arrowhead), an extending filopodium (blue arrowhead), and a retracting filopodium (magenta arrowhead). Scale bars 100 nm (A, B), 1 μm (C). Panels A-B were acquired in 30 sec, while the snapshots in panel C were obtained in 10 sec. Adapted by permission from Springer Customer Service Centre GmbH: Springer Nature, Ref. 261, Copyright 2011. Adapted with permission from Ref. 216, Copyright 2012 National Academy of Sciences, U. S. A.

Small peptide tags, together with enzymatic labeling techniques or reactive proteins, have also been applied nicely in SMLM. Howarth et al. have engineered biotin ligase to covalently label a 15-mer amino acid tag (named an acceptor peptide) with biotin for downstream detection via readout with fluorescent streptavidin.³³⁰ This general procedure has been refined with the use of monomeric streptavidin, produced recombinantly and fluorescently labeled in vitro, for studies of synaptic adhesion proteins by a variety of super-resolution microscopy methods, including SMLM.³³¹ In a different approach, Zakeri et al. adapted the Streptococcus pyogenes adhesion protein FbaB from to form a new tag system termed SpyTag/SpyCatcher that is based on FbaB’s spontaneous ability to form an intramolecular isopeptide covalent bond.³³² In this system, FbaB was split into a small 13-mer amino acid peptide named SpyTag and a 128-mer amino acid protein domain named SpyCatcher that are able to rapidly form a covalent bond in solution. Chamma et al. demonstrated the use of SpyTag/SpyCatcher with SMLM for imaging of SpyTagged actin, keratin, and clathrin proteins in fixed specimens that were labeled with a SpyCatcher-Alexa Fluor 647 conjugate.³³³

Finally, tagging of proteins of interest with classic, small epitope tags such as HA, myc, FLAG, and V5 provides a convenient means for downstream antibody labeling, although these are generally limited to use with fixed cells. Epitope tagging can be particularly useful when working with proteins for which good antibodies are unavailable or when densely tagging structures for efficient downstream detection. Viswanathan et al. have further advanced this capability by engineering highly antigenic GFP-like tags which utilize GFP as a protein scaffold on which to display 10-15 copies of a classic epitope tag; the resulting tags, termed ‘spaghetti monster’ fluorescent proteins, have enabled bright immunolabeling of key proteins and structures with antibodies for SMLM and other fluorescence-based experiments.³³⁴ While classic short epitope tags are generally read out using IgG-based antibodies due to a lack of available nanobodies, a recent report described a nanobody against BC2-tag, a new epitope tag originally derived from beta-Catenin, and its use with SMLM on BC2-tagged actin, tubulin, and vimentin in fixed cells.^335–337

6.3. Small molecule affinity probes

Small molecules are another important class of probes that can label specific organelles, structures, or (in some cases) proteins, by virtue of the intrinsic affinity of the fluorophore or through affinity conferred by an appended group. For instance, cell permeant, cationic fluorophores label mitochondria in living cells due to their negative potential, octadecane-conjugated fluorophores target the plasma membrane due to their hydrophobicity, and phalloidin-fluorophore conjugates bind filamentous actin. While such affinity-based probes lack the generality of genetically encoded protein tags, they are nonetheless important tools in SMLM due to their sometimes very intense stains (e.g., MitoTrackers) or their specificity for one protein conformation over another (e.g., phalloidin probes bind polymerized, not monomeric, actin).

Table 3 provides a list of some affinity-based probes used for SMLM with some example images in Figure 34. For example, the staining of filamentous actin can be achieved much more densely with a fluorescent phalloidin conjugate than with bulkier anti-actin antibodies or actin-affinity-based proteins such as LifeAct, and Xu et al. used phalloidin-Alexa Fluor 647 staining of actin together with a novel dual-objective microscope to capture highly detailed images of actin in fixed specimens.³⁴³ In live specimens, Shim et al. used various established organelle tracker fluorophores to visualize membrane bound organelles in live cells by SMLM, including the use of MitoTracker Red to monitor mitochondrial fusion and fission events and ER-Tracker Red to monitor dynamics of the endoplasmic reticulum.²¹⁶

Table 3.

Example affinity stain probes used in SMLM.

Target	Basis of Affinity	Recognition Mechanism	Examples used in SMLM
Mitochondria	Positive charge of dye and/or ligand	Mitochondria negative potential in live cells attracts cationic, membrane permeable dyes	MitoTracker Orange,216 MitoTracker Red,216,338 MitoTracker Deep Red,216 Rhodamine 123,338 TMRE338, RhoB-HMSiR339
Plasma Membrane	Appended hydrophobic group (e.g., octadecane)	Hydrophobic group partitions dye to plasma membrane; uptake may follow	DiI/DiD/DiR216, DiI-C6-HMSiR339
Endoplasmic Reticulum (ER)	Lipophilic framework or glibenclamide (containing sulfonylurea group)	Lipophilic affinity or binding between sulfonylurea and its potassium channel receptor enriched on ER	ER-Tracker Red,216 Nile Red338,340, Cer-HMSiR339
Golgi	Appended ceramide or sphingomyelin group	Interaction with endogenous lipids and cholesterol	Cer-HMSiR339
Lysosomes	Appended weakly basic amine linker	Protonation-induced accumulation in acidic environment of lysosomes	LysoTracker Red,216 LysoTracker Green338
DNA	Polycyclic, aromatic, and planar	Intercalation, π- π stacking	Picogreen,120,341 YOYO-1120,342
Filamentous actin	Appended phalloidin group	Phalloidin binds interface of polymerized actin subunits	Phalloidin-Alexa Fluor 647343

Figure 34.

Use of affinity-based small molecule labels for SMLM. A) 3D SMLM image of axons in fixed, cultured neurons where filamentous actin is labeled with phalloidin-Alexa Fluor 647 and reveals a periodic membrane-associated cytoskeletal structure. B) SMLM image of mitochondria and the endoplasmic reticiulum in a live BS-C-1 cell labeled using MitoTracker Red and ER-Tracker Red. C) SMLM image of DNA in a live U2OS cell labeled with Picogreen. Scale bars are 500 nm for B, and 2.5 μm for C. From Ref. 344, adapted with permission from AAAS. Adapted with permission from Ref. 216, Copyright 2012 National Academy of Sciences, U. S. A. Adapted with permission from Ref. 341, Copyright 2012 John Wiley and Sons.

6.4. FISH probes

A key sub-class of non-genetically-encoded affinity-based probes are oligonucleotide probes used in various FISH (fluorescence in situ hybridization) experiments. FISH probes are designed to hybridize with nucleic acids for the detection of specific sequences of DNA or RNA using a range of fluorophores including the popular photoswitchable fluorophore Alexa Fluor 647. These probes are typically comprised of DNA oligonucleotides but sometimes higher affinity variations are used such as peptide nucleic acids or locked nucleic acids. For example, fluorescent oligonucleotides have been used to label: Xist RNA in mouse embryonic fibroblast nuclei for the study of X chromosome inactivation (Figure 35A-C)³⁴⁵; telomere conformation and factors impacting formation of T-loop structures in which the 3’ terminal overhang of telomeres invades the double-stranded region of the telomere (Figure 35D)³⁴⁶; chromosome conformations^347,348; and mRNA distributions in yeast for simultaneous detection of 32 different gene transcripts.³⁴⁹ FISH is an important method for SMLM and will no doubt remain so for years to come.

Figure 35.

SMLM using labeling of nucleic acids by fluorescence in situ hybridization (FISH). A) Conventional fluorescence image of a tetraploid mouse embryonic fibroblast nucleus stained with the general nuclear stain Hoechst 33342 (blue) and FISH probes against Xist RNA (magenta), revealing two inactivated X-chromosomes (zoom-in views, right). B-C) 3D SMLM images of Xist RNA from zoom-in views of A, where distinct Xist puncta are evident at a range of heights (encoded in color). D) SMLM image of telomeres detected using FISH in chromatin from a mouse splenocyte. Adapted with permission from Ref. 345, Copyright 2015 National Academy of Sciences, U. S. A. Adapted from Ref. 346, Copyright 2013, with permission from Elsevier.

6.5. Metabolically incorporated probes

Rather than utilize probes that bind covalently or noncovalently to targets on a biological specimen, some labeling strategies utilize a cell’s metabolic machinery to incorporate exogenous probes. Many of these strategies utilize a two-step procedure in which small molecules containing a “click chemistry” group such as an azide or alkyne can be incorporated into specific molecules or structures and are then incubated with a fluorophore containing a ligand that can specifically react with the click chemistry group on the metabolically introduced label.

Jiang and co-workers³⁵⁰ introduced to living cells monosaccharides functionalized with azide or alkyne groups, conjugated to them a cyanine dye bearing an alkyne or azide group, and studied their dynamics on the membranes of live cancer cells by single-molecule tracking and by SMLM. They were able to observe differences in the mobility of N-linked sialic acids and O-linked N-acetyl galactosamine and they also captured SMLM images of membrane nanotubes in various cancer cell lines.

In order to label actively replicating nucleic acids, 5-ethynyl-2-deoxyuridine (also known as EdU, an alkyne modified nucleobase) can be incubated with a cell and later coupled to a fluorophore using copper-catalyzed azide alkyne cycloaddition.^351–353 Zessin et al. utilized EdU labeling of cells to incorporate Alexa Fluor 647 onto chromosomal DNA, revealing a detailed chromatin map.³⁵²

An expanded genetic code (Figure 36) has been used to incorporate fluorophores directly to non-canonical amino acids on a specific residue of a protein of interest.³⁵⁴ For example, Uttamapinant et al. incorporated an amino acid bearing a bio-orthogonal reactive group BCNK (bicycle[6.1.0]nonyne-lysine) into various proteins of interest (actin or tubulin) and labeled it with the fluorophore SiR using inverse-electron-demand Diels-Alder cycloaddition with a tetrazine-SiR conjugate; the resulting labeled specimens were imaged by live-cell SMLM.³⁵⁵

Figure 36.

Use of an expanded genetic code to introduce a non-canonical amino acid into a specific protein of interest at a specific site. Blue box and orange arrow indicate complementary functional groups for biorthogonal reactions and red box indicates the fluorophore (F). Adapted with permission from Ref. 356, Copyright 2012 John Wiley and Sons.

7. Multichannel SMLM

Many clever strategies have been utilized to date to image distinct structures or molecules in SMLM, although most researchers either use spectrally distinct fluorophores or use sequential imaging of the same fluorophore. Some may use a combination of these two basic approaches and many exploit unique switching properties of the fluorophores.

The most conceptually straightforward approach for multichannel imaging is to utilize fluorophores which are spectrally well-resolved from each other. For instance, the particularly good SMLM fluorophore Cy5 (or its close relative Alexa Fluor 647) has been paired with many fluorophores for two-color imaging, including ATTO 488, ATTO 520, Cy3B, Cy7 (and their close relatives), mEos2, rsFastLime, and others.^{128,135,138,139,261,357,358} This type of multichannel approach is commonly used by researchers, but requires a careful sub-pixel alignment of the two or more spectral channels and the use of an instrument with the necessary laser lines at a sufficient power for imaging.

Bates et al. introduced activator-reporter fluorophore pairs for multichannel imaging¹³⁷ which evolved naturally from their earlier work in developing the Cy5 photoswitch and in developing SMLM.^26,76 In this approach, a photoswitchable “reporter” fluorophore such as Cy5 was paired with a bluer-absorbing “activator” fluorophore (e.g., Alexa Fluor 405 or Cy3) on antibody molecules, where distinct sets of antibody molecules were labeled with different activator molecules but the same reporter. The activator-reporter phenomenon has not been investigated mechanistically but appears to be unique to pentamethine and heptamethine cyanine reporters with thiol-based (but not phosphine-based) switching and works with a wide range of blue-shifted activator dyes.^135,137 Labeling with activator-reporter pairs is complemented by use of a custom illumination series as follows. Intense red light is used to initially switch reporter molecules to a nonfluorescent ‘off’ state, then illumination with weak light on resonance with one set of activator dyes is used to activate a subset of reporter fluorophores, and then the activated fluorophores are imaged and switched off again with intense red light. The process is then repeated with different activator wavelengths and then the entire cycle is repeated hundreds or thousands of times to build up a large number of localizations. During analysis, the reporter molecules are assigned to a specific channel based on the known activation wavelength that preceded the localization. Nonspecific activation may lead to substantial crosstalk in this activator-reporter approach, particularly when one activator-reporter pair is much more abundant than another, but the effect can be minimized through use of a crosstalk subtraction algorithm.³⁵⁹ Although activator-reporter pairs are not widely used at present, they have nonetheless enabled many important studies (Figure 37).^{344,359–361}

Figure 37.

Conventional image (A) and corresponding SMLM image (B) of synapses in mouse accessory olfactory bulb imaged using activator-reporter pairs. Scale bars are 1 μm for A and B. Adapted from Ref. 359, Copyright 2010, with permission from Elsevier.

Fluorescent reporter molecules with overlapping, but distinct emission spectra are also used for multicolor SMLM in two or more channels. In one approach, the fluorescent image is split by a dichroic mirror into two different spectral bands such that the fluorescence emitted by a given fluorophore is split between two bands with a characteristic ratio.¹⁶² When individual fluorophores are imaged in this way, the detected ratio of fluorescence in the two bands can be analyzed and compared to control data to determine the fluorophore’s identity. Up to four spectral channels have been imaged in this way, with good discrimination between fluorophores whose emission spectra may differ by as little as 20 nm.^{147,216,362–364} In a second approach, a dual-objective SMLM instrument was used to capture images of single molecules in one detection path together with the measurement of single-molecule spectra in the other detection path; up to four-channel imaging was demonstrated, with excellent discrimination of fluorophores whose emission spectra was separated by as little as 10 nm (Figure 38).³⁶⁵ The concurrent detection of a spectrally dispersed image with a non-dispersed image may also be performed using a single objective lens with a custom detection path that splits the image into two beams, one of which is dispersed spectrally.^{340,366–368} Spectrally-resolved SMLM is also useful in combination with sensor fluorophores like Nile Red, whose emission spectrum reports on local hydrophobicity, to measure membrane polarity in living specimens.^340,368

Figure 38.

SMLM using A) spectrally-resolved detection for four-channel imaging. The fluorophores Dyomics 634, DyLight 650, CF660C, and CF680 were used to immunolabel peroxisomes, vimentin filaments, microtubules and the outer mitochondrial membrane, respectively, in a fixed PtK2 cell. Although the fluorescence emission spectra overlapped for the four fluorophores B), concurrent spatial and spectral detection allowed discrimination of each fluorophore from the others for multichannel imaging. Adapted by permission from Springer Customer Service Centre GmbH: Springer Nature, Ref. 365, Copyright 2015.

The activator-reporter approach and the overlapping emission spectra approaches each have several benefits: only one powerful laser is required for the instrument; the distinct channels in single detection path instruments are easily registered, not requiring careful sub-pixel alignment. A notable drawback of these approaches, however, is that they divide up the maximum number of achievable localizations among all the channels, often resulting in a reduced density per channel. Consider the scenario where the duty cycle of a particular fluorophore is sufficient to support ~200 fluorophores in a diffraction-limited region without excessive double-localization artifacts. In this case, the fluorophores can be used to label one structure very densely, or four structures with lower density per structure and thus worse resolution per channel than for a single channel. Because spectrally overlapping fluorophores must also be detected individually within a diffraction-limited region, they also face the challenge of finite fluorophore density; additionally, for experiments with different types of fluorophores, researchers should use the lowest duty cycle fluorophores on the densest structures to avoid being limited to the properties of the fluorophore with the highest duty cycle.

Sequential imaging of different proteins or structures using the same fluorophore has been implemented in a variety of ways. For instance, Tam et al.³⁶⁹, Valley et al.³⁷⁰, and Yi et al.³⁷¹ developed procedures consisting of multiple cycles consisting of: labeling the specimen with the fluorophore Alexa Fluor 647 (or Cy5); imaging by SMLM; and removal of signal through elution, chemical reduction, and/or photobleaching. The SMLM images obtained in each round may then be assigned to a given channel of a composite image. These approaches nicely leverage the robust photoswitching behavior of Alexa Fluor 647 but require a lengthy overall procedure which includes multiple rounds of immunolabeling. The method dubbed Exchange-PAINT from Jungmann et al.²⁷⁹ (Figure 39) minimized these delays by using a single round of immunolabeling with DNA-conjugated antibodies prior to imaging followed by imaging rounds consisting of sequential rounds of incubation with fluorescently labeled (single-stranded) imager DNA strands. Sequential imaging approaches using DNA-conjugated antibodies and imager DNA strands enable scaling to many channels by exchange or to larger depths by refreshing photobleached probes and are an area of active development.^284,372,373

8. Live SMLM

Imaging of live specimens by SMLM faces the key challenges of labeling the living specimen, recording a series of frames rapidly enough to avoid blurring due to sample or molecular motion, and avoiding perturbation of the sample. In this section, we will review these key challenges and then we will highlight a few studies which have pushed the envelope for live-cell SMLM beyond those which have already been mentioned, above.

The labeling of living specimens is relatively straightforward using fluorescent proteins. Many early live-cell SMLM experiments utilized various photoactivatable or photoconvertible FPs and focused on the study of relatively slow processes, including: the study of the nanoscale distribution and motion of viral membrane proteins in eukaryotic cells using PA-GFP and EosFP;^111,374 the study of focal adhesion dynamics using EosFP;¹²⁵ and the study of bacterial cytoskeletal proteins using YFP (under conditions that lead to photoswitching of YFP).³⁷⁵ The labeling of living specimens with cell-permeant small molecule probes such as MitoTracker or ER-Tracker has been enabling with experiments for which switchable, specific small molecule probes are available.^216,338 A broader set of SMLM experiments in live cells has been enabled by the use of transgenically expressed (reactive) protein tags such as TMP, SNAP, and Halo that can target a specific protein, together with the use of cell-permeant fluorophores such as ATTO 655, tetramethyl rhodamine, and SiR .^154,327,362 The development of photoactivatable and cell-permeable organic fluorophores such as PA-JF₅₄₉ and PA-JF₆₄₆ has been particularly useful for SMLM-style photoactivated particle tracking experiments (sptPALM)¹¹¹ due to the high brightness and high photostability of these probes, and they have been used in, for instance, several studies of chromatin and transcription factor dynamics.^168,376,377

The second challenge with live-cell imaging is that many localizations must be rapidly recorded in order to observe the cell with a spatial and temporal resolution that captures the features and dynamics of interest. This leads to an inevitable tradeoff between spatial resolution, which is limited in part by the number of frames in an image stack, and temporal resolution, which is limited by the time required to record the image stack. Rapid charge coupled device and scientific complementary metal on semiconductor cameras have achieved frame rates of ~500-3200 Hz for SMLM,^216,261,378 which is impressive but still restricts imaging to relatively slow processes.

A third challenge with live imaging is avoiding perturbation of the specimen. In the case of SMLM, where researchers typically use relatively high illumination intensity (≥1 kW/cm²), phototoxicity due to illumination of the sample is a major concern, particularly when imaging with relatively short wavelengths (<500 nm).³⁷⁹ Some fluorophores, upon illumination, can produce toxic substances such as reactive oxygen species, although SMLM fluorophores have not been systematically tested in this regard, and other modes of phototoxicity, for instance due to excitation endogenous chromophores, also likely occur. Finally, for experiments in which an imaging cocktail is used, researchers must exercise care to avoid or minimize toxicity (see Section 3.4). At a minimum, researchers should perform control measurements to make sure their phenotypes of interest are not perturbed by the experimental conditions.

We would like to highlight four recent studies employing live-cell SMLM. Jones et al.²⁶¹ used live-cell 3D SMLM to resolve the nanoscopic structure of clathrin-coated pits (Figure 40 A-B). Living cells expressing clathrin light chain fused to a SNAP tag (see Section 6.2) were labeled with Alexa Fluor 647, where electroporation was used to deliver the otherwise cell-impermeant fluorophore to the cytosol of eukaryotic cells. In some experiments the cells were also incubated with Alexa Fluor 568-labeled transferrin. Using these specimens, the researchers were able to capture a series of snapshots that revealed the nanoscale morphology of the clathrin coat enclosing the transferrin cluster—features unobservable by conventional fluorescence microscopy methods (Figure 40B). In another study, Shim et al.²¹⁶ used various commercially available organelle or membrane tracker fluorophores such as Dil (see Section 6.3) to label the plasma membrane of a living, cultured hippocampal neuron, and clearly resolved the extension and retraction of filopodia or dendritic spines (Figure 40C).

Takakura et al.³⁸⁰ used novel conjugates of the spontaneously-blinking silicon rhodamine fluorophore HMSiR¹⁶⁴ (Figure 12, Figure 15A) for long time-lapse SMLM of membrane-bound organelles in live cells. They found that the environmentally-sensitive fluorophore HMSiR, when localized to hydrophobic environments such as the membranes of the endoplasmic reticulum (ER), or the cell plasma membrane, exhibits spontaneous blinking with a very low duty cycle (up to a 50-fold improvement over that of HMSiR in a hydrophilic environment) while still emitting ~1,500 detected photons per localization. Additionally, they used a powerful two-step labeling procedure to attain highly dense labeling: in step 1, cells were incubated with a targeting group that localized to a particular region of the cell (e.g., endoplasmic reticulum, plasma membrane) and that was functionalized with a trans-cyclooctene group; in step 2, cells were incubated with a HMSiR fluorophore that is functionalized with tetrazine, where the tetrazine and trans-cyclooctene groups react rapidly and selectively with one another in order to generate the probe, in situ. Together, this approach has enabled hundreds of snapshots to be recorded over tens of minutes in living specimens with a high spatial resolution (Figure 41). The extension of these so-called HIDE (high density, environment-sensitive) probes to multiple channels may benefit from a recently reported³⁸¹ green-absorbing, spontaneously-blinking fluorophore (HEtetTFER, spectrally orthogonal to HMSiR), although further developments may be required to optimize a second orthogonal coupling strategy to efficiently and selectively couple two different fluorophores with their respective targeting groups.

Figure 41.

High density environmentally sensitive (HIDE) probes for live cell SMLM based on the spontaneously-switching fluorophore HMSiR (red, see also Figure 12). A) HIDE probes are synthesized in vivo in a two-step reaction between a trans-cyclooctene-functionalized targeting ligand R that labels an organelle or region of a cell and the tetrazine-functionalized fluorophore HMSiR. The targeting ligand R and labeling conditions can be chosen to label the Golgi, endoplasmic reticulum (ER), mitochondria, or the plasma membrane. B) A SMLM image of the ER in a live HeLa cell acquired in 2 sec using the HIDE probe Cer-HMSiR with partial conventional fluorescence overlay (lower left). This is one snapshot from a series of ~750 spanning ~25 min. Features of the ER are indicated in lowercase including ER sheets (s), ER tubules (t), and fenestrated ER sheets (f). C-D) Several 2 sec snapshots across 25 minutes of acquisition (2 sec per frame) for the boxed regions in B showing dynamic behavior of the ER including budding (yellow arrowheads), looping (green arrowhead), fusion (blue and magenta arrowheads), and reversible transition between tubular and sheet-like morphologies (white arrowheads). E) Kymograph along purple line in B showing the line profile (vertical) as a function of time. Scale bar 500 nm. Adapted by permission from Springer Customer Service Centre GmbH: Springer Nature, Ref. 380, Copyright 2017.

9. Potential artifacts and pitfalls of SMLM

Despite its strong capabilities, SMLM faces several potential artifacts and pitfalls which require special care by practitioners.^382,383 Some of these are common to all fluorescence modalities such as inadequate fixation, nonspecific labeling,³⁸⁴ or overexpression artifacts, while others are more specific to super-resolution methods or SMLM, including double localizations and label linkage errors. In this section we will highlight these concerns.

9.1. Fixation and immunolabeling artifacts

In immunostaining procedures, specimens are typically processed using a protocol consisting of fixation, permeabilization, and antibody staining steps, with optional pre-fixation extraction and post-stain fixation steps, and each may require optimization.^384,385 The choice of fixative often represents a tradeoff between preservation of structure and an ability to label the specimen, in two key ways. For instance, fixation with cold methanol can nicely preserve the microtubule cytoskeleton but will destroy many membrane-bound organelles and tends to flatten cellular specimens. Alternatively, fixation with a mixture of formaldehyde and glutaraldehyde can preserve the structure of cytoskeletal components together with subcellular organelles but the use of glutaraldehyde may inhibit antibody penetration into thick tissues and can severely reduce antibody recognition of antigens for some antigens/antibodies. Formaldehyde-only fixation does a better job preserving antigenicity and penetration into thick tissues but does not preserve some structures sufficiently, such as the microtubule cytoskeleton, which may depolymerizes under these conditions, and the choice of detergent in the permeabilization step must be made carefully to avoid damaging less robustly fixed structures such as mitochondria.^14,15,383 Glyoxal was recently demonstrated as a potential alternative to formaldehyde and/or glutaraldehyde, fixation, and was reported to generally have good penetration and good structure preservation.³⁸⁶ An extraction step, typically consisting of a brief, pre-fixation treatment with detergent, is extremely useful when studying the cytoskeleton in cultured cells since it permeabilizes the membrane and ‘spills’ unpolymerized cytoskeletal proteins that lead to background signal when interested in studying the assembled cytoskeleton.^14,15,383 Many membrane-bound organelles are destroyed or perturbed by extraction, however.

Assuming that the structures of interest are preserved, that the epitopes of interest are preserved, and that antibody penetration into the specimen is sufficient, there still exist several potential pitfalls in antibody labeling including low affinity, low specificity, lot variability, and crosstalk with secondary antibodies. Antibody labeling is idiosyncratic and requires attention to detail.^387,388 Some antibodies seem to provide excellent stains in a wide range of conditions while others can lead to high background signals in some situations and others may never seem to work to target the structure indicated by the manufacturer or publications. When co-staining two different proteins in a cell, sometimes the only effective antibodies available are raised in the same host animal, which would lead to severe crosstalk with indirect (primary/secondary) immunofluorescence; workarounds such as the use of direct immunofluorescence (i.e., fluorescently labeled primaries), sequential immunostaining, or isotype-specific secondary antibodies may remedy the situation but also require careful optimization. Secondary antibodies that are fluorescently labeled and that are designed to detect primary antibodies from different, but closely species (e.g. mouse and rat) may have unwanted crosstalk unless highly cross-adsorbed secondary antibodies are used; when available, fluorescently labeled primary antibodies may be utilized.

Hopefully these brief remarks have emphasized the key point that there is no universal immunostaining protocol. SMLM researchers routinely spend substantial time exploring this parameter space when new systems or proteins are being studied.

9.2. Artifacts with protein fusion tags and fluorescent proteins

Fusing a fluorescent protein, epitope tag for high-affinity immunolabeling, or reactive protein tags such as SNAP or Halo to a protein of interest is an elegant and powerful approach for sample labeling that avoids many of the pitfalls of antibody labeling of endogenous proteins but introduces its own set of risks. Fusion tags have the potential to disturb protein function through the tag itself, through tag-mediated dimerization or aggregation, or through perturbation of protein expression levels for instance by overexpression.^7,25,63,389 As with antibodies, fusion protein labeling is highly idiosyncratic and requires care when working with and validating new constructs and procedures. For many systems, expression of fusion proteins has been typically achieved by transformation with plasmid DNA bearing the fusion protein, but this has the potential problems that high expression levels may be unphysiological or toxic or that without concurrent knockdown such as with siRNA the procedure is unable to report on endogenous, unlabeled proteins.³¹⁹ Some of these problems should be addressable through the use of improved genome editing tools³⁹⁰ which can allow for knockout or knock-in cells with fusion proteins under endogenous promoters.

9.3. SMLM-specific artifacts

The unique point-by-point approach of SMLM lends the method to artifacts.^{9,382,391,392} Besides channel crosstalk incurred through the use of various multi-channel SMLM methods described above, SMLM is particularly sensitive to ‘double localization’ artifacts where two (or more) localizations occur within a diffraction-limited region. As shown in Figure 42, double/multiple localizations can introduce spurious features or substantial distortions and may be easy to overlook for samples containing isolated puncta or filamentous structures where spurious localizations would partly overlay with the true structures.

Figure 42.

Double-localization artifacts in SMLM. A-C) Cartoon illustration, with actual fluorophore distributions shown in black, measured single-localizations shown in magenta and measured double or multiple localizations shown in green. A) Punctate structures produce double localizations toward the center of each punctum and between puncta that are separated by approximately the diffraction limit or less (~250 nm). B) Filaments show double localizations along the length of the filaments, and are easily mistaken for valid, single localizations. Filaments in close proximity, however, reveal spurious localizations where the two filaments approach each other and/or cross. C) A single filamentous structure with abundant double/multiple localizations will appear artificially narrowed. D) Experimental SMLM image of microtubules in a fixed and immunolabeled BS-C-1 cell recorded under conditions that produced many double/multiple localizations; localizations with large spot widths (more likely to be double/multiple localizations) have been rendered in green while others have been rendered in magenta.

Statistically speaking, double/multiple localizations are unavoidable in SMLM, and we can straightforwardly derive a simple estimate for their likelihood. Consider a diffraction-limited region containing N idealized fluorophores which all exhibit perfect, binary switching with the same on/off rates and duty cycle, where duty cycle (also known as the “on fraction”) is defined as DC=t_on/(t_on+t_off). The likelihood that a given fluorophore is off is just 1-DC. From this, it follows that when one of the N fluorophores is on, the likelihood that all of the other N-1 fluorophores will be off is (1-DC)^N-1. Finally, the multiple localization likelihood (MLL), when more than one fluorophore in a diffraction-limited region is on at the same time, is then just MLL=1-(1-DC)^N-1. For example, with DC=0.01 and N=100, we get MLL≈0.63, meaning that 100 fluorophores with a duty cycle of 0.01 in a diffraction-limited region would have multiple localizations ~63% of the time. For DC=0.001 and N=100, the situation is far better with MLL=0.09, or 9% multiple localizations. The use of a low duty cycle fluorophore can help to reduce this effect, as can filtering on the basis of localization brightness and/or shape (e.g., elliptical), or through the use of more sophisticated data analysis algorithms that can detect more than one emitter per fluorescent spot.^{343,393–396} Dai et al. have performed related calculations and simulations on multiple localization artifacts that also include the effect of discrete exposure times²⁸¹ and Sinkó et al. have developed a software tool that can simulate SMLM results including the effects of double/multiple localizations.³⁹⁷

Beyond double/multiple localization artifacts, the short length scales being probed in SMLM or other super-resolution microscopy methods raises additional possible artifacts. First, the linkage between fluorophore and target protein or structure is often many nanometers and this linkage error may result in substantial broadening effects or distortions (see also Section 6.1). Second, multichannel SMLM imaging requires careful sub-pixel registration especially for multicolor approaches where different detection paths are used and aberrations can be substantial.^107,398,399 Third, hindered molecular rotation can induce localization errors due to fixed dipole effects, particularly when using high numerical aperture objective lenses as is standard in SMLM,^{136,400–403} however, custom hardware and has been developed which can mitigate or eliminate these effects.^404–407

9.4. “Resolution”

The assessment of resolution in SMLM is not straightforward, and without delving into a rigorous discussion on quantitative metrics of resolution, we wish to make three points not yet raised in this review.^{126,392,408–410} First, “resolution” in optical microscopy is generally a measure of the minimum distance at which two objects may be distinguished in an image. This basic definition is somewhat at odds with SMLM’s mode of operation, however, since in SMLM, fluorescent labels are already detected individually and are therefore distinguished from one another as a function of both space and time. Second, a high density of localizations is in general sought to support a high spatial resolution in a Nyquist-limit sense (Figure 4), although caveats exist for this sort of analysis. For instance, a sparse distribution of molecules in a sample that is labeled perfectly with fluorophores would have appear to have a low spatial resolution in a Nyquist-limit sense, strictly based on consideration of the density, even if the fluorophore and imaging conditions could have supported a much higher density. Alternatively, a densely-labeled but small object that doesn’t fill a diffraction-limited volume may appear to have a high spatial resolution, but the fluorophore duty cycle may not support that resolution for a larger object of equal density that fills more of the diffraction-limited volume. This apparent dependence of the attainable resolution on the specimen itself is sometimes a source of confusion in the field. Third, localization uncertainty, which also impacts achievable resolution (see Section 2.5), is reported in different ways in the literature. While some researchers refer to the standard deviation (SD) of the localization uncertainty, others refer to the full width at half maximum (FWHM) of the localization uncertainty, where a Gaussian distribution has a FWHM that is 2.35× higher than its SD (Figure 43), and yet others have simply estimated the theoretical localization uncertainty based on the number of detected photons and a few other measured or estimated parameters. Whenever making comparisons of localization uncertainty, it is therefore important to keep in mind which quantity is being considered.

Figure 43.

SMLM studies use different metrics to characterize uncertainty in localization, including full-width at half maximum (FHWM) and standard deviation (SD), and it is important to determine which metric is used when comparing performance in different studies. For a Gaussian distribution, FWHM=2.35×SD.

10. Concluding remarks

SMLM has developed rapidly since its inception in 2006 and within several years has been used by researchers in a wide range of disciplines to gain insights into the nanoscale organization of biological systems. Given the rapid progress in SMLM over the past 12 years, it is tempting to consider its prospects for the future.

Could SMLM one day be used to read out dense distributions of on the few-nm length scale? Recent experiments are promising. Two reports studying DNA origami model systems demonstrated successful resolution of 2D arrays of ~10-50 fluorophores in which adjacent fluorophores were separated by only 5-6 nm. In the first report, Dai et al.²⁸¹ achieved this by detecting a very large number of photons from bound fluorescent probes using DNA-PAINT with optimized imaging conditions (Figure 28). In the second report, Balzarotti et al.⁴⁰⁷ combined photoswitching with a novel patterned excitation approach (MINFLUX) to achieve a similar result with far fewer photons detected per molecule. Pushing to even shorter length scales, Weisenburger et al.⁴¹¹ imaged the fluctuating emission from ~4 spectrally identical fluorophores within a diffraction-limited area in a novel cryogenic fluorescence microscope to attain sub-nanometer localization precision (SD) with purified protein specimens. It is worth noting that fluorophores behave quite differently in vitreous ice at low temperatures (~4 K) compared to their behavior in solution near room temperature (~300 K), including a dramatically lower photobleaching rate; if robust photoswitching or photoactivation could be achieved in this state, it could extend the sub-nanometer localization methodology to larger numbers of fluorophores per diffraction-limited spot. Advancing these ultrahigh-resolution methodologies beyond proof of principle may require further developments, perhaps including new fluorescent probe labeling technology, but there are nonetheless several clear paths forward. Alternatively, multimodal experiments involving SMLM and electron microscopy have made good progress in recent years and potentially offer the benefits of each method with few drawbacks as recently reviewed by Hauser et al. and Kopek et al.^412,413

The prospects for live-cell SMLM are also good based on recent reports. For instance, new probes such as HMSiR probes for tracking the ER (Figure 41) have already substantially advanced live-cell SMLM experiments to enable several hundreds of SMLM snapshots at a time, although the methodology to date has been only shown on select targets. In terms of speed, it is likely that SMLM faces a hard, fundamental limit, which we can briefly consider. An individual localization consisting of an emission burst of ~1000 detected photons would require at least 10,000 fluorescence emission events (only some of which reach the camera) such that a localization could not be detected in a shorter period than about 10,000× the fluorescence lifetime, or ~0.01 ms, assuming a 1 ns fluorescence lifetime.⁷⁸ If 1000 frames are required in order to reconstruct a SMLM image, then doing so would require at least ~10 ms, but probably a factor of 10 or more times longer per SMLM image (~100 ms) given the highly optimistic estimated values and several disregarded complications. Achieving high photon emission rates in such a short period of time will of course require very high laser illumination, raising concerns for phototoxicity that may ultimately impose stricter limitations than any consideration of fluorophore (or hardware) rates.

Beyond pushing the frontiers of spatial and temporal resolution in SMLM, important work remains to be done in making SMLM a workhorse technique that biologists can easily use in their research. Two- or three-channel SMLM with FPs in fixed specimens is still difficult due to a lack of robustly switching, spectrally orthogonal fluorophores, and even one-channel SMLM with FPs would benefit greatly from the development of a switchable FP with a substantial boost in photons. Multichannel SMLM in fixed specimens with organic fluorophores still requires a great deal of optimization in sample preparation and skill to execute well, and it would benefit greatly from improved spectrally orthogonal fluorophores and fluorescent labeling strategies.

This leads to the question of how improve fluorophores for SMLM. Fluorescent protein developers have tended to use a combination of structure-guided design and high-throughput screens to assess the properties of new variants. These have yielded a great many switchable (and non-switchable) FPs, but the screens typically have limited ability to report on single-molecule properties. Recent work from Emanuel et al.⁴¹⁴ may help to move the field forward. They measured the performance of a library of ~60,000 FP mutants using fluorescence microscopy, with one FP mutant expressed per cell, and after sample fixation, read out unique oligonucleotide barcodes that had been previously associated with each mutant FP DNA sequence. This allowed the association of time-resolved measurements by fluorescence microscopy with a FP DNA sequence that could later be re-expressed in cells and assessed in detail. While ensemble FP properties were selected in the work by Emanuel et al., the approach could potentially be adapted to enable evaluation of single-molecule switching properties for libraries of many thousands of FP mutants.

Photoswitchable organic fluorophores have been in some ways slower to develop than FPs, in part since the underlying photochemistry is in many cases not well understood and switching properties are difficult to predict. Consider that Cy5 (and its close structural relative Alexa Fluor 647) was synthesized as a general-purpose fluorophore long before the advent of SMLM, and that it fortuitously possessed superior photoswitching capabilities that on the whole have yet to be surpassed. Substantial effort may be required to better understand the fundamental photochemical processes for the various fluorophore classes used in SMLM in order to enable rationally-guided improvements or computational screening of candidate SMLM fluorophores. A potential alternative strategy to produce improved photoswitchable fluorophores could be to utilize a combinatorial synthetic approach to generate libraries of fluorophores together with a sufficiently high-throughput ensemble and/or single-molecule screen of fluorophore properties.

While SMLM is now enabling biologists to reveal the inner workings of the cell with rich molecular detail at unprecedented resolution, much work remains to be done as the methodology matures into a workhorse tool. In many ways, fluorescent probes have guided researchers to the development of SMLM, and they also hold the key to unlocking SMLM’s full potential.

Figure 6.

Comparison of photoswitching properties of three different fluorophores. A) Single-molecule on-off photoswitching of a representative Alexa Fluor 647 molecule in a deoxygenated solution containing a thiol. Under these conditions, Alexa Fluor 647 exhibits B) an average duty cycle of ~0.0012 and C) an average of ~5200 photons/localization. These switching properties are sufficient to produce high-quality images of hollow clathrin-coated pits in fixed, immunolabeled cells D-F). The fluorophore ATTO 655 has the same duty cycle as Alexa Fluor 647 G-H) but emits ~8× fewer photons per burst I), and the resulting larger position uncertainty obscures the hollow clathrin-coated pits J-L). The fluorophore Cy5.5 has a duty cycle that is ~6× larger that of Alexa Fluor 647 M-N), while having a similar number of detected photons O), and as a result the achievable density of localizations on clathrin-coated pits is insufficient to confidently trace out the hollow structure P-R). Adapted by permission from Springer Customer Service Centre GmbH: Springer Nature, Ref. 128, Copyright 2011.

Biography

Joshua C. Vaughan is an Assistant Professor of Chemistry at the University of Washington, Seattle. He grew up in the San Francisco bay area and received a B.A. degree in Chemistry from Reed College in Portland, OR, where he learned to love the Pacific Northwest. He earned a Ph.D. in physical chemistry at MIT where he worked in the lab of Prof. Keith Nelson on applications of adaptive optics to ultrafast spectroscopy. For his postdoctoral research in Prof. Xiaowei Zhuang’s lab at Harvard he developed tools for super-resolution fluorescence microscopy and studied virus transport. His group at the University of Washington is developing new tools for super-resolution fluorescence microscopy and they are using these tools to understand the molecular scale organization of biological systems in their native cellular or tissue contexts. Outside of work, he enjoys music, sailing, and spoiling his children.

Honglin Li received his B.S. degree in Fine Chemistry at Dalian University of Technology in 2005, and then earned his Ph.D. degree in 2011 under the supervision of Prof. Xiaojun Peng. From 2011 to 2013, he worked as a postdoctoral fellow in Dr. Qian Wang’s group at the University of South Carolina, and then he relocated to Seattle to joined Joshua C. Vaughan’s group at the University of Washington. His research interest is focusing on the development of colorimetric and fluorescent materials for bio-sensing and bio-imaging.