Few to single molecule imaging of fluorescent probe molecules can provide information on the distribution, dynamics, interactions and activity of specific fluorescently tagged proteins during cellular processes. Unfortunately, these imaging studies are made challenging in living cells because of fluorescence signals from endogenous cofactors. Moreover, related background signals within multi-cell systems and intact tissue are even higher and reduce signal contrast even for ensemble populations of probe molecules. High-contrast optical imaging within high-background environments will therefore require new ideas on the design of fluorescence probes, and the way their fluorescence signals are generated and analysed to form an image. To this end, in the present review we describe recent studies on a new family of fluorescent probe called optical switches, with descriptions of the mechanisms that underlie their ability to undergo rapid and reversible transitions between two distinct states. Optical manipulation of the fluorescent and non-fluorescent states of an optical switch probe generates a modulated fluorescence signal that can be isolated from a larger unmodulated background by using OLID (optical lock-in detection) techniques. The present review concludes with a discussion on select applications of synthetic and genetically encoded optical switch probes and OLID microscopy for high-contrast imaging of specific proteins and membrane structures within living systems.
Recent advances in optical microscopy and the design of synthetic and genetically encoded optical probes have enabled high-resolution multiscale analysis of the distribution, interactions and activity of specific proteins within living cells, tissues and organisms [1–7]. For example, it is now possible to detect the location and activity of individual protein molecules in vitro , within fixed cells [9–12] and within living bacteria [13,14], and for specialized cases within mammalian cells [15,16]. Single-molecule imaging in mammalian cells is usually made very difficult, however, because of fluorescent signals arising from endogenous molecules that may greatly exceed the signal from individual probe molecules . Indeed the autofluorescence signal within a typical mammalian cell is equivalent to some 10000 molecules of fluorescein [or GFP (green fluorescent protein)] [17–19]. Since many of the fluorescent cofactors and matrix proteins that make up autofluorescence are essential, one could argue that this background constitutes a fundamental limitation for high-contrast imaging of individual and ensemble populations of optical probes within living cells and tissue . In our view, improvements to image contrast within these high-background environments require a paradigm shift in the way that we design fluorescent probes and the way that fluorescence signals from these probes are generated and analysed to form an image. The present review discusses recent progress towards this goal with an emphasis on fluorescent optical switch probes. Several excellent reviews have been written on the application of contrast-enhancing probes for optical super-resolution microscopy [1,4,11,20,21], and so discussion in the present review is limited to applications of optical switches for the imaging of ensemble populations of probe molecules within living cells and tissue.
ADVANTAGES OF IMAGING INDIVIDUAL PROTEIN MOLECULES WITHIN A LIVING CELL
Imaging few copies of a fluorescently tagged protein within a living cell has several advantages over ensemble imaging, two of which are considered here. First, imaging individual protein molecules allows the investigator to generate spatio-temporal information on the dynamics, behaviour, interactions, activity and function of the protein during a cellular process and over the entire cell cycle [1,13,14]. Secondly, detailed analysis of the emission arising from individual probe molecules provides a means to generate super-resolution images of that protein within the cell [11,20,21]. Consider, for example, a study aimed at understanding the role of an ABP (actin-binding protein) during a cell protrusion event. These molecular events are usually confined to the plasma membrane, or within specialized compartments that enclose a miniscule volume (~10−15 l) that envelopes some 600 molecules of a soluble cytosolic ABP present at a concentration of 1 μM. In principle, the distribution, dynamics and interactions of these 600 molecules could be quantified by imaging a GFP–ABP protein. Ideally, the number of GFP–ABP molecules used should be a fraction of the number of endogenous protein molecules, perhaps as few as ten molecules in a single protrusion. Now, while the detection of a handful of fluorescently labelled protein molecules is easily realized in vitro , the same measurement in a living cell is made very difficult because of signals from cell fluorescence (Figure 1A). Several approaches have been used to reduce the contribution of autofluorescence to live-cell imaging microscopy, for example by imaging within bacteria, which have a very small volume and consequently low numbers of endogenous fluorophores [13,14]. In addition, TIRF (total internal reflection fluorescence) microscopy  can greatly improve signal contrast as it excites only those fluorophores within ~100 nm of the surface (as illustrated in Figure 1B, discussed below). These few- to single-molecule imaging studies coupled with approaches to improve image contrast allow the investigator to track the private life of individual protein molecules as they move around the bacterium or close to the plasma membrane, and to ascertain where, and for how long, they interact with other proteins or exert their activity, and how these properties relate to the regulation of the specific cellular processes [1,13,14]. Unfortunately, the imaging of single-few probe molecules in healthy mammalian cells is not really feasible using widefield or confocal microscopy, although realizing this capability is likely to significantly improve our understanding of the molecular basis of cell signalling and cellular proteomics. In our view, the best approach to meet this challenge is to develop new imaging technologies and probes that can overcome the problem of autofluorescence within a mammalian cell, which is approximately three orders of magnitude larger than within a bacterium. [13,17].
Imaging fluorescence probes within high-background environments
The second advantage of imaging few molecules of a fluorescent protein is that the fluorescence signals can be used to resolve the location of protein molecules in a sample with a high- or super-resolution [8–12]. In particular, PALM (photoactivation localization microscopy), STORM (stochastic optical reconstruction microscopy) [9,10] and RESOFT (reversible saturable optical fluorescence transitions)  generate high-resolution images of a specific protein in a cell by high-contrast imaging of fluorescence signals that arise from individual protein molecules. Descriptions of these techniques are covered in detail elsewhere [11,20,21]. Although impressive, the super-resolution imaging capabilities of both PALM and STORM are limited to the detection of immobile molecules over a reasonably long (0.1–20 s) timescale. This factor most probably explains why most optical super-resolution imaging studies are conducted on fixed cells that also facilitate the removal of endogenous fluorescence by chemical treatments and washings. Given the need to use fixed cells, one could argue that optical super-resolution microscopy actually provides few advantages for protein localization over more established electron microscopic techniques. On the other hand, an obvious benefit in the case of PALM is that optical probes are genetically encoded and better integrated into the cellular machinery [10,21,24]. On the other side of this argument, and a view that we support, is that once it becomes possible to detect fluorescence signals from single-probe molecules within healthy living mammalian cells, then optical super-resolution imaging microscopy will provide an unprecedented view into the private lives of individual and ensemble collections of proteins during cell processes. In our view, the starting point for these advances will be to develop new optical probes and detection systems for high-contrast fluorescence imaging in living cells [2,25,26].
OPTICAL PROBES FOR HIGH-CONTRAST IMAGING OF PROTEINS WITHIN LIVE CELLS
In the following section we discuss various approaches to improve signal contrast for fluorescence imaging with particular emphasis on the design and performance of synthetic and genetically encoded reversible optical switches.
Non-linear optical control of fluorescence
An optical switch exists in one of two distinct structural states. Ideally, these states will correspond to a non-fluorescent and fluorescent form of the probe. For the purpose of the present review, we classify optical switch probes according to whether they undergo a single (irreversible) transition from the non-fluorescent to the fluorescent state, or multiple (reversible) transitions between these two states, as indicated in Figures 1(C) and 1(D) respectively. Non-linear optical control of the quantum yield of fluorescence from the optical switch is achieved by controlling the populations of the two states in the system via orthogonal light-driven reactions. Contrast enhancement can be realized from a simple optical switch by recording and analysing fluorescence signals from the optical switch immediately before and after a fluorescence photo-activation event. This method is used in PALM to detect individual molecules of a photo-activated protein .
Irreversible optical switches
Excitation of a caged fluorophore  with near-UV light, typically 365 nm, triggers an irreversible excited state reaction that cleaves a covalent bond, releasing the protection group to generate a high quantum yield fluorophore. Mitchison and colleagues [28,29] showed that photoactivation of caged resorufin and caged fluorescein attached to actin or tubulin respectively, could be used to fluorescently highlight a subpopulation of these proteins at specific sites in the cell. Subsequent imaging of the distribution of these photoactivated proteins allowed his group to track their behaviour and quantify their diffusion constants during a physiological response [28,29]. Moreover, by limiting the uncaging reaction to a narrow slit, they were able to dramatically improve the contrast of the fluorescence signal against autofluorescence. Quantitative analysis of time-dependent changes in the distribution of resorufin–actin fluorescence following photoactivation was used to determine rates of actin filament dynamics at specific loci in the cell during a motile response . In spite of the great improvement in signal contrast made possible by using photoactivation of caged fluorescein and caged resorufin, as well as other caged fluorophores , this class of irreversible optical switch is rarely employed in cell biology. This lack of interest is somewhat surprising as caged fluorescein, rhodamine 110 and resorufin have excellent radiative decay rates that allow for high photon output per unit time, an important parameter for PALM . Genetically encoded photoactivatable proteins such as paGFP (photoactivated GFP) and red-shifted variants [24,31] are the preferred choice of irreversible optical switch probe for PALM, no doubt owing to the ease with which they can be targeted to specific proteins in a cell or tissue, and the ability to generate mutant forms with specific photochemical or photophysical properties .
Reversible optical switches
Reversible optical switch probes may exist in two distinct structural states, but unlike caged fluorophores and photoactivatable fluorescent proteins, transitions between their two states are rapid and reversible [2,25]. Reversible optical switching occurs in a number of small molecule chromophores [26,32], nanoparticles  and within genetically encoded proteins [21,34–36]. However, many small-molecule optical switch probes are most often found to undergo irreversible excited state reactions involving a chemical change . The best-described reversible optical switch probes are diheteroarylethenes (Figure 2A) , spiroamido-rhodamine (Figure 2B) [39,40], thioindigo (Figure 2C), azobenzene (Figure 2D)  and spiropyrans (Figure 3A) [2,24,25,42–44].
Molecular structures and reversible reactions between the two states of some optical switch probes
Reversible optical switching of NitroBIPS between the SP and MC states
Ideally, one of the two states of a reversible optical switch will have a fluorescence quantum yield that tends to zero and reach a maximum value in the fluorescent state [24,25,32]. The reversible optical switches identified above are known to undergo excited bond-cleavage or photo-isomerization reactions that lead to unique absorption (and emission) characteristics for each state. However, it is worth noting that most of the spectroscopic and optical switching properties of optical switch probes have been determined within organic solvents [37,44,45], and less frequently in water or within biological systems [2,24,32,46,47]. This solvent preference is most probable because these probes exhibit their best switching properties in apolar solvents, besides being poorly soluble in aqueous environments . Interestingly, we showed that the quantum yield for these photochemical transitions increased for probe molecules that are attached to or associated with proteins or within a biological membrane [2,24,25]. The efficiency of optical switching and fluorescence emission of the fluorescent states of the optical switch probes shown in Figures 2 and 3 is also sensitive to the immediate solvent environment and through subtle chemical modifications to the core photochromic scaffold. Consider, for example, the MC (merocyanine) state of NISO (naphthoxazine), which has virtually no fluorescence emission, yet the quantum yield for the MC-to-SP (spiro) transition is very high . On the other hand, the MC state of NitroBIPS has a measurable fluorescence emission and somewhat lower quantum efficiency for MC–SP transition . In spite of the low quantum yield of MC fluorescence, this signal is still a very useful indicator for analysing the amount of MC in a system during optical manipulation of the NitroBIPS probe [2,25]. Moreover, this MC fluorescence can be used to image ensemble populations of a tagged protein or membrane-associated molecules within living systems [2,32]. Select applications of NitroBIPS for high-contrast imaging in living cells are discussed later in the present review. First, however, we introduce other classes of reversible optical switch probes that have been used for high-contrast imaging of single molecules in cell systems.
Optical switches for STORM
The mechanisms for reversible optical switching between the two states of the Cy3–Cy5 (indocarbocyanine/indodicarbocyanine) probe (Figure 2E) used in STORM is unique and widely different from that found for NitroBIPS (Figure 3A) and other probes listed in Figure 2. For example, transitions between the two states of a STORM probe are brought about via sequential excited-state chemical reactions involving a Cy5 probe, an organic mercaptan, and one or more closely associated excited-state molecules of Cy3 [9,20,48,49]. A single cycle of optical switching of Cy5 fluorescence in a Cy3–Cy5 probe complex begins with the chemical conversion of all Cy5 molecules in the field into a non-fluorescent state. This dark background condition is achieved through an excited state reaction that ultimately generates a Cy5–thiol adduct; unlike Cy5, this adduct has no visible absorption band and consequently has no red fluorescence when excited at 650 nm . A sparse population of these non-fluorescent Cy5-adducts in a field-of-view are converted back into the fluorescent Cy5 state via a poorly defined excited-state reaction that is accelerated by 535 nm excitation of a closely associated Cy3 probe. The number of fluorescent Cy5 probes generated in this Cy3-triggered reaction depends on the energy of the 535 nm light, providing a means to control the population of fluorescent Cy5 probes in the field-of-view. The sparse population of reactivated Cy5 probes is then illuminated with 632 nm and the Cy5 fluorescence signal from single molecules is imaged. At some point during this phase of the optical-switching cycle, the Cy5-excited state will react with a different molecule of the thiol, generating a dark background. Multiple cycles of 632 nm and 535 nm irradiation of the sample are used to stochastically control the Cy5 fluorescence from thousands of individual probe molecules in the sample. The fluorescence intensity distribution from individual Cy5 molecules is used to super-localize the position of the probe molecule as detailed in [9,20,49]. A STORM image of Cy5 probes in a sample has a considerably higher spatial resolution compared with images obtained by using widefield or confocal microscopes , and this capability is creating a tremendous level of excitement and enthusiasm in the field of cell biology. Although STORM has been shown to dramatically improve the resolution of optical microscopy in fixed-cell preparations, as we outlined above the experimental conditions required for stochastic switching of Cy5 fluorescence are not quite physiological. For example, the concentration of thiolate required to bring about a collision during the lifetime of the Cy5-excited state (3 ns) is approx. 1 M at pH 7 or a few millimolar at pH 10 . Finally, since Cy5 is easily photobleached by excitation with 632 nm, STORM imaging requires anoxic conditions that might have an impact on cell health and signalling pathways [9,49].
Reversible optical switches based on spiroamido-rhodamine have been introduced for super-resolution imaging microscopy (Figure 2B) [39,40]. These probes exist at room temperature in their non-fluorescent lactam state, but undergo an excited state-driven, intramolecular reaction that generates the fluorescent ring-opened rhodamine . This fluorescent state returns rapidly to the non-fluorescent lactam in a thermally driven reaction. The optical switching and fluorescence emission properties of spiroamido-rhodamines are in principle ideal for applications in single-molecule imaging and optical super-resolution microscopy . First, the non-fluorescent lactam state is strongly favoured at room temperature, which overcomes the need to pre-irradiate the sample before beginning an imaging study. Secondly, a sparse population of the fluorescent open form is generated in nanoseconds following irradiation of the field with a short pulse of UV light . Thirdly, the number of fluorescent molecules in the field can be controlled by varying the energy of the near-UV light, or else by using a wavelength at the red-edge of the lactam absorption band . Fourthly, once activated, the fluorescent state returns to the non-fluorescent lactam in a rapid, thermally driven reaction. Although attractive in design, this class of optical switch probe should be further improved for applications in live-cell imaging of individual and ensemble populations of labelled proteins. The most obvious improvements would be to: (i) shift the absorption spectrum of the lactam state from <320 nm to ~400 nm; (ii) increase the quantum yield for the ring-opening reaction; (iii) stabilize the fluorescent open state in order to increase the number of collected photons before the ring closes; and (iv) incorporate a means to optically turn off the fluorescent state of the switch at a longer wavelength than that used to excite the fluorescence.
Azobenzene is widely used as an optically driven molecular actuator . Azobenzene exists at room temperature either in the extended trans-state or the compact cis-state (Figure 2D). Transitions between the trans- and cis-states are driven by near-UV light (trans-to-cis) and visible light (cis-to-trans). Although of limited utility for optical imaging, the ability to optically manipulate pendant ligands on the azobenzene switch has been used to great effect in the control of engineered ion-channel proteins in neuronal cells grown in culture, in tissue slices and even within organisms [41,50]. In most of these studies one end of the azobenzene is immobilized to a specific site on the protein via maleimide coupling to a cysteine residue while the ligand, appended to the other end of molecule, is used to control the activity of the ion channel [41,51]. Unfortunately, azobenzene has a poor two-photon absorption cross-section and the small molar absorption coefficient of the cis-state and modest quantum efficiency of the excited-state transitions require relatively high-energy illumination of the cell or tissue.
NitroBIPS and NISO
The intramolecular bond-breaking and bond-forming reactions within the reversible optical switch NitroBIPS that bring about excited state transitions between the SP and MC states are shown in Figure 3(A). Optical switching between the SP and MC state proceeds with low efficiency in water and in other highly polar solvents [26,37,43–45,52]. While this property would ordinarily limit the usefulness of NitroBIPS for studies in biological systems, these transitions can proceed with higher quantum yield for NitroBIPS probes on proteins [2,25]. Importantly, these transitions occur with high fidelity for NitroBIPS probes that are specifically attached to a unique site on a protein  or when bound within a membrane . This improvement most probably occurs in response to removing the NitroBIPS probe from the bulk water. Specific details of the spectroscopic, photophysical and functional properties of the SP and MC states, and factors affecting transitions between these states are reviewed in the following section.
As with the spiroamido-rhodamines, the thermodynamically stable form of NitroBIPS is the non-fluorescent SP state . A single cycle of optical switching of NitroBIPS begins with the excitation of SP using a short pulse of 365 nm light, or irradiation with 720 nm (two-photon excitation; Figure 3). The high quantum yield for the SP-to-MC transition results in complete conversion of the SP state into the MC state when using a single 100 ms pulse of 365 nm or a single scan of 720 nm light . Moreover, in the case of two-photon (720 nm) excitation, the reaction is complete within any given pixel during the 2 μs pixel dwell time of the laser [2,32,53]. Consequently, the time required for optical switching is not limited by the photochemistry, but rather by the tolerance of the cell to the power of the radiation.
The MC state of NitroBIPS has a strong absorption S0–S1 band centred at ~550 nm with a ground-state dipole moment of 20 D . This strong MC-dipole is stabilized within a protein by specific dipolar interactions . This property has two important consequences for optical switching of NitroBIPS in proteins; first, the quantum yield for optical switching between the SP and MC states in a protein conjugate is higher compared with bulk water ; secondly, the thermally driven MC-to-SP reaction has a time constant of 3000 s for MC attached to a protein compared with 170 s for MC in ethanol, allowing for all-optical control of the SP and MC states .
Site-selective compared with random labelling of NitroBIPS probes
Site-selective labelling of NitroBIPS on a protein is achieved by using standard coupling chemistry (maleimide)  or by covalent linking of NitroBIPS to the SNAP-tag using a benzylguanine suicide substrate . Ideally, these conjugates should contain a single NitroBIPS probe that is linked to a unique residue on the protein . This requirement is necessary because MC engages in strong specific dipolar interactions with polar groups on proteins that can affect both the MC absorption and the quantum yield for the MC-to-SP transition . However, many studies in the literature describing NitroBIPS conjugates use proteins or nanoparticles that are randomly labelled with the optical switch [47,54,55]. There are potential problems associated with randomly labelled NitroBIPS probes, including the formation of MC charge-transfer complexes and the presence of heterogeneous MC–protein interactions, the strongest of which may have low quantum yields for optical switching. Interestingly, previous studies have shown that the MC state of NitroBIPS serves as a high quantum yield probe for single nanoparticle-based super-resolution imaging within living cells [54,55]. It should be said at this point that the requirement for a high quantum yield of switching between the two states of a probe is not quite the same for single molecule and ensemble populations of a probe molecule. For example, a low quantum yield is acceptable for single-molecule manipulation of an optical switch probe, as eventually one can usually guarantee that an excited state transition will occur during a given period of time simply by ensuring a saturating condition for the irradiation. On the other hand, optical switching of an ensemble population of probe molecules within the same period requires far higher quantum yields for the transitions, especially for quantitative conversions between the two states [2,53].
It is important, in our view, to demonstrate that optical switching of a probe on a protein or a nanoparticle proceeds with high fidelity. This property is shown as a scheme in Figure 3(B). In this idealized study, the populations of the SP and MC states of NitroBIPS are precisely controlled between two extreme values (0% and 100%) over multiple cycles of optical switching by irradiating the sample with a defined sequence of UV and visible light pulses (Figure 3B). The actual state of the switch in the sample is monitored in real-time by measuring the MC fluorescence signal. The extent of the conversion between the SP and MC states depends primarily on the quantum yields for their respective transitions and the energy of the irradiation pulses [2,32,53]. Multilabelling of NitroBIPS on proteins and nanoparticles can affect the fidelity of optical switching studies because the MC–MC complexes that form on the protein or nanoparticle may exhibit different absorption spectra and quantum efficiencies for optical switching compared with single probes specifically labelled to a protein or nanoparticle [25,26]. An experimental demonstration of an optical switching waveform for a NitroBIPS–actin conjugate injected within a living Xenopus embryo is shown in Figure 3(C). These data show that each cycle of the waveform has an identical MC fluorescence intensity profile (Figure 3C). This constant profile, together with the finding that the maximum fluorescence intensity is similar over the 10 cycles of optical switching (Figure 3C), suggests that the excited-state transitions between the SP and MC states occur with high fidelity and with a low level of fatigue and photobleaching . The profile of optical switching between the SP and MC states over a single cycle can be shaped by varying the power of the 365 nm and 535 nm light. For example, it is possible to generate a quasi-square waveform profile by irradiating the SP state with a single high-intensity light pulse at 365 nm, which converts all SP molecules in the field into the MC state, whereas quantitative conversion of MC into the SP state is realized by using a single high-intensity pulse of 543 nm light. On the other hand, the saw-tooth waveform shown in Figure 3(C) is obtained by illumination energies of the 365 nm and 543 nm light sources that are compatible with live-cell imaging .
Highly fluorescent optical switches
As detailed earlier, the decay of the excited-state of MC in NitroBIPS occurs via competing processes of fluorescence, photochemistry, photobleaching and transitions to dark states [25,53]. Consequently, the quantum yield for fluorescence can never be as high as TMR (tetramethylrhodamine) and Cy3, which are validated probes for single-molecule imaging. The best way to improve the quantum yield for fluorescence emission in optical switch probes would be to isolate the fluorescing and photochromic units in a bi-functional probe. These optical switch probes are composed of a highly fluorescent donor probe and a highly efficient optical switch with the latter undergoing FRET (Förster resonance energy transfer), with concomitant quenching of the donor emission in only one of its two states [46,53,56,57]. For example, Mao et al.  showed the intensity from GFP or fluorescein acting as a donor, and a closely associated NitroBIPS acceptor on the same protein could be modulated between the no-FRET and FRET states by optical control of the SP and MC states in the conjugate. This optical control scheme was used to generate a modulated GFP fluorescence signal within a GFP–NitroBIPS fusion protein and served as a basis to increase the sensitivity of the FRET method in living cells . Improvements to the design and performance of optically switchable FRET probes now allow for high-sensitive detection of optical switch probes in living cells. In particular, Petchprayoon et al.  introduced a new class of optical switch that integrates a highly fluorescent TMR and NISO , a highly efficient optical switch (Figure 4A) in the same molecule. The NISO moiety in TMR–NISO undergoes rapid and reversible excited-state driven transitions between a colourless SP state and a coloured MC state in response to irradiation with 365 nm and >600 nm light. This design feature allows for orthogonal control of the switching and fluorescence imaging functions of the probe (see ). The high degree of spectral overlap between TMR emission and MC–NISO absorption and their close proximity in the probe molecule (<1 nm) ensures that FRET will occur with high efficiency in the MC state, resulting in the effective extinction of TMR fluorescence. On the other hand, virtually no FRET occurs between TMR and the colourless SP state .
Optical switching of TMR fluorescence in a highly fluorescent optical switch
Optical control of the SP and MC states of the NISO probe in TMR–NISO was shown to generate highly modulated TMR fluorescence signals via a FRET-based mechanism for probes in solution and within living cells (Figure 4B) . A closer inspection of the intensity trace of TMR fluorescence over several cycles of optical switching (Figure 4C) shows that the intensity profile is uniform for each cycle and oscillates between a TMR fluorescence value close to zero (MC state) and a maximum (SP state) with a time constant of 200 ms. This result strongly suggests that transitions between the two states of the TMR–NISO switch are governed by defined quantum yields for the SP-to-MC and MC-to-SP transitions. Moreover, the small variation in the values of the maximum and minimum fluorescence intensity found over many cycles of optical switching suggests that these transitions occur with little photobleaching or occupancy of dark states. Since TMR is a validated probe for single-molecule imaging, the TMR–NISO optical switch should prove useful for high-contrast imaging of proteins at the level of few to single molecules. Highly fluorescent donor probes have also been linked to NitroBIPS and used to control the fluorescence from molecules of green-emitting donor probes, such as GFP, by a photochromic FRET mechanism [46,53,56,57]. In another type of integrated FRET-based optical switch, a donor probe is used whose absorption spectrum closely overlaps with that of a closely linked MC state of NitroBIPS . Although this class of probe is shown to exhibit sensitivity for single-molecule imaging and super-resolution imaging, the spectral overlap does not allow for orthogonal control of the optical switching and fluorescence functions. This complication most probably accounts for the modest modulation of donor fluorescence as measured for ensemble populations of probe molecules .
Genetically encoded optical switches
Studies have shown that introducing specific mutations in GFP and mCherry-like fluorescent proteins can generate proteins harbouring an optically switchable chromophore, each having a unique fluorescence quantum yield and/or absorption spectrum [34–36]. These probes are usually used for single-molecule super-resolution imaging of tagged proteins  and for OLID (optical lock-in detection) imaging microscopy . The attraction of using genetically encoded optical switches resides in the ability to target the probe to almost any protein in a cell within an organism. Moreover, mutations within the parent protein can be used to shift the action spectra for absorption, fluorescence and photochemistry, as well as providing a means to change the direction of optical switching between the two states and their quantum yields [35,36].
Dronpa, a genetically encoded photochromic protein, undergoes optically driven transitions between a green-emitting fluorescent cis-state and a non-fluorescent trans-state (Figure 5) . Dronpa is a far better fluorophore than it is an optical switch, the quantum yield for cis-Dronpa fluorescence is 0.85, whereas the yield for the cis-to-trans reaction is reportedly too low to measure from experiments . In spite of these widely different quantum yields, Dronpa is an effective optical switch probe for OLID imaging . A single cycle of reversible optical switching is achieved by irradiating the image field with two sequential scans at 800 nm (two-photon, ~45 mW at sample), or a ~100 ms pulse of 365 nm light; both excitation modes trigger the trans-to-cis transition. The reverse transition is realized by irradiating the field with five to ten scans at 488 nm using a laser power at the sample of ~70 μW, which is typical for imaging fluorophores in living cells . Optical switching between the two states of Dronpa can be repeated over many cycles and is only limited by cumulative occupancy of non-switchable dark states [2,34]. While these dark states tend to reduce the number of Dronpa molecules in the sample that can undergo optical switching, normalization of the cis-Dronpa intensity over the course of an optical switching reveals that the intensity profile of optical switching for each cycle in a study is constant, i.e. the quantum yield for these transitions does not change for the population of functional probe molecules that survive each cycle .
Optical switching of Dronpa fluorescence
Switchable mCherry proteins
mCherry, a monomeric genetically encoded red-emitting protein, was introduced by Shaner et al. . Stiel et al.  showed that certain mutations in mCherry generated proteins whose fluorescence quantum yield could be modulated with alternate irradiation with ~400 nm and >550 nm light respectively. For one of these mutants (rsCherryRev), irradiation of the non-fluorescent state with 405 nm generates the fluorescent state, whereas the reverse transition is effected by irradiating the fluorescent state at >550 nm. Interestingly, the two states of a different mCherry mutant undergo the inverse transitions with the same wavelengths of light . Unlike Dronpa, where transitions between the cis- and trans-states are associated with excited state cis–trans isomerizations and accompanied by a large change in their absorption spectra , the fluorescent and non-fluorescent states of rsCherryRev have identical absorption and could reflect the effects of 405 nm/>550 nm modulation of interactions between the chromophore and the protein matrix. However, very recently a variant of rsCherry was described whose absorption spectra shifts between the two states of the switch . This feature now offers the potential to extend OLID–FRET  for high-contrast imaging of protein interactions using a completely genetically encoded probe system.
Other approaches to improve image contrast
TIRF microscopy is used to significantly reduce autofluorescence signals in living cells . This improvement results in the most part from the use of an excitation field that penetrates ~100 nm into the sample and thereby limits the number of background molecules that are excited (Figure 1B). TIRF is widely used to image single molecules in vitro  and optical probes close to the plasma membrane within living cells. For example, Kusumi and colleagues  used TIRF microscopy to image single molecular events in a receptor-activated regulation of a signalling pathway. In particular, they detected a change in FRET efficiency between a probe pair associated with the formation of a GFP–Rac1 (donor) and Cy3-labelled PtdIns5-kinase (acceptor) complex during growth-factor-mediated activation of the signalling pathway.
Nanoparticles, NIR (near-IR) probes and lifetime-resolved imaging
Bright fluorescent nanoparticles including Qdots are often touted as probes for single-particle imaging in living cells [61–64]. Nanoparticles are especially suitable for this purpose because of their remarkably high molar absorption coefficient, high quantum yield for fluorescence emission and photostability . In spite of these advantages and more than 20 years of imaging studies using Qdots, there are a few examples in the literature that actually show a benefit to using a Qdot compared with a conventional fluorescent probe to improve signal contrast . The poor record of using nanoparticles as probes for imaging studies in living cells most probably results from difficulties in delivering and targeting these large probes to specific proteins in a cell and their toxicity and blinking properties.
Image contrast can also be improved by employing probes whose absorption and emission occurs in the NIR region of the spectrum , and by using long-lived fluorophores [67,68]. In the case of the NIR dyes, the improvement results from an absence of NIR-absorbing species in the cell whereas for long-lived excited-state probes, it is possible to use gated detection to reduce contributions from the shorter-lived autofluorescence. NIR probes offer great potential for few-single molecule imaging within living cells, although research is still needed to increase their quantum yields for fluorescence emission and reduce their high rate of photobleaching . Lifetime-resolved imaging of long-lived fluorescence probes provide remarkable improvements in signal contrast [67,68]; however, because of their long excited-state lifetimes (μs to ms), fewer photons are collected per unit time compared with a prompt fluorescent probe, such as TMR.
OLID imaging microscopy
OLID is similar in principle to other lock-in detection methods that are widely used in the physical sciences and electronics. In particular, a signal exhibiting a known time-dependent change (the reference waveform) is used as a lock-in detection template to extract a modulated signal component from background noise. The reference waveform may be the time-dependence of the change in a physical or electrical signal, e.g. pressure or voltage [69,70]. In OLID, the reference waveform is generated by recording the fluorescence emission of an optical switch probe over many cycles of optical switching with 365 nm and visible light . OLID is a new imaging technique developed by our group that greatly improves image contrast of a fluorescent optical switch probe. Moreover, by using the coloured state of the optical switch as an acceptor probe in FRET with a high quantum yield donor such as GFP, it was shown by Mao et al.  that OLID–FRET can detect as little of 1% of a protein complex in a living cell.
Digital cross-correlation analysis
The modulated signal generated from manipulating an optical switch between the fluorescent and non-fluorescent states is unique and can be isolated from the unmodulated background emission by using a digital cross-correlation approach . This discrimination is realized by fitting the measured signal within each pixel of an image over several optical cycles of switching to the reference waveform of the optical perturbation . The reference waveform is obtained from an average of the fluorescence intensity from a region of five by five pixels recorded over two or more cycles of optical switching. An internal reference waveform is obtained from a region of the image having the highest depth of intensity modulation (Figure 6B), whereas external reference waveforms are obtained from the intensity of an optical switch probe attached to a micron-sized over several cycles of optical switching. Next, a cross-correlation analysis is conducted between the intensity of each pixel or group of pixels in the image and the reference waveform over several cycles of optical switching and a correlation coefficient is calculated for each pixel. The value of the cross-correlation coefficient is displayed on a pixel-by-pixel basis to generate a correlation image . The correlation image thus maps the level of correlation between the raw intensity and the reference intensity profile over multiple switching cycles, normally for two to six cycles. It is noteworthy that the value of the correlation coefficient is independent of the raw pixel intensity, but is determined by the persistence of the modulated signal in a region over the course of the optical switching study. This means that a zero value for the correlation coefficient must be a region where the signal is completely dominated by unmodulated background fluorescence. A low value of the correlation coefficient could arise from an optical switch probe that diffuses during the course of optical switching. In contrast, a high value of the correlation coefficient reflects that the signal contains a population of immobile optical switch probes. Thus a region of an image emitting both modulated (immobilized switch) and unmodulated (background) signals will have higher contrast in the correlation image compared with the intensity image . The effectiveness of OLID imaging microscopy in improving image contrast has been demonstrated in living cells and tissues by using both synthetic and genetically encoded optical switches.
OLID imaging of NitroBIPS in a live Xenopus spinal cord explant
Synthetic optical switches
A small molecule lipophilic optical switch, C11-NitroBIPS, was used to image neurons within a Xenopus spinal cord explant . The raw red-fluorescence intensity image of the explant is characterized by strong autoflurescence, most likely arising from yolk particles (Figure 6A). The widely distributed background signal obscures MC fluorescence within neurons. For this particular study, the internal reference waveform was generated from a small area of the cell having the highest depth of fluorescence modulation (Figure 6B) and was used for the cross-correlation analysis. A significant degree of contrast enhancement is evident in a comparison of the raw intensity and correlation images (Figures 6A and 6C). The correlation image, for example, shows that the highest correlation coefficient values are located within the neuronal and glial membranes, and their filopodia-like extensions, even though the raw fluorescence intensity within these regions is low (Figure 6A). An overlay of the fluorescence intensity and correlation coefficient profiles for a line across the image provides a direct means to quantify the improvement to image contrast that results from using OLID imaging, as high as 13-fold for the example shown in Figure 6(D) (black and red traces respectively).
OLID imaging of Dronpa in living cells and tissue
Dronpa can be genetically fused to almost any protein in a living cell or tissue, providing an unrivalled level of convenience and specificity for tagging proteins in cells and organisms compared with synthetic probes. The contrast-enhancing power of OLID was used for intravital imaging of Dronpa–actin expressed within cells in culture and within specific neurons in a mouse hippocampal slice . A dramatic evaluation of OLID imaging of Dronpa is shown in Figure 7, which shows images of Dronpa expressed in specific tissue within living zebrafish larvae . For example, genes encoding Dronpa–actin and Dronpa alone were injected into the single-cell stage of zebrafish embryos and strongly expressed in muscle and neurons respectively at 5 days post-fertilization. OLID imaging of the Dronpa–actin in the muscle (Figures 7C and 7D) and Dronpa in neurons (Figures 7A and 7B) required immobilizing embryos with tricaine. Dronpa–actin and Dronpa fluorescence was modulated in these systems over many cycles of optical switching by irradiation cells with 800 nm and 488 nm light according to an irradiation scheme shown in Figure 5. The modulation intensity of Dronpa fluorescence in these systems is robust and occurs with high fidelity and little fatigue (Figure 7A, inset). The correlation image of Dronpa–actin in muscle (Figure 7D) shows specific structural improvements compared with the intensity image that include better definition sarcomeric organization that clearly defines the paths of individual myofibrils compared with the intensity image (Figure 7C). Zebrafish expressing Dronpa fluorescence in specific neurons also exhibited a strong modulation with the correlation image (Figure 7B) showing significant improvements in the connections between individual neurons compared with the intensity image (Figures 7A).
Optical switching and OLID imaging of Dronpa in live Zebrafish
As we have indicated, developing the ideal optical switch for high-contrast imaging of proteins in living systems will require maximizing the quantum yields for fluorescence emission and photochemistry, while shifting action spectra for transitions between the two states of the switch to the red. Moreover, the advantages of imaging individual molecules of a protein within mammalian cells will most likely require the development of optical switch probes whose emission signal can be detected against a high background, and collected rapidly enough to allow for the imaging of dynamic protein assemblies. This latter point is important because the number of emitted photons detected for a super-resolution image (~6000) is ultimately determined by the rate of fluorescence decay and this under normal levels of illumination might require ~10 ms. During this time, however, a freely mobile spherical protein having a diffusion time of 5×10−7 cm2·s−1 would move the length of a 10 μm cell . Clearly new types of probe are required for these studies whose radiative decay rate exceeds that of conventional organic fluorophores with metal–fluorophore complexes being the most likely candidates .
We thank members of the Marriott and Ehud Isacoff Laboratories at the University of California Berkeley for useful discussions.
Förster resonance energy transfer
green fluorescent protein
optical lock-in detection
photoactivation localization microscopy
stochastic optical reconstruction microscopy
total internal reflection fluorescence
Work in the author's laboratories is supported by the National Institutes of Health [grant numbers R01 GM086233-01 (to G.M.), GM086233-01 (to Y.Y.)].