Extending the spatiotemporal resolution of super-resolution microscopies using photomodulatable fluorescent proteins

In the past two decades, various super-resolution (SR) microscopy techniques have been developed to break the diffraction limit using subdiffraction excitation to spatially modulate the fluorescence emission. Photomodulatable fluorescent proteins (FPs) can be activated by light of specific wavelengths to produce either stochastic or patterned subdiffraction excitation, resulting in improved optical resolution. In this review, we focus on the recently developed photomodulatable FPs or commonly used SR microscopies and discuss the concepts and strategies for optimizing and selecting the biochemical and photophysical properties of PMFPs to improve the spatiotemporal resolution of SR techniques, especially time-lapse live-cell SR techniques.


Introduction
Fluorescence microscopy using genetically encoded°u orescent proteins (FPs) plays a key role in elucidating biological processes as well as in vivo dynamics in a minimally invasive manner. However, because of the di®raction limit, it is a challenge to visualize objects with sizes smaller than 200 nm in the lateral direction and 500 nm in the axial direction. To overcome this problem, several superresolution (SR) techniques have been developed to extend the di®raction-limited spatial resolution by as much as an order of magnitude. [1][2][3][4][5][6][7] Most of these SR techniques use light-controllable FPs whose°u orescence emission is modulated by light irradiation with speci¯c wavelengths. We refer to these light-controllable FPs as photomodulatable FPs (PMFPs). Compared with organic dyes, FPs enable easy and genetically speci¯c labeling of both¯xed and living cells. Although some unique characteristics of organic dyes and other°uorophores, such as their brightness and photostability, can be superior to those of PMFPs, the sophisticated approaches required to deliver them into biological cells and to decrease unspeci¯c labeling limit their application mainly to¯xed samples and make the imaging of living cells a challenge. There are three classes of PMFPs: photoactivatable FPs (PAFPs), photoconvertible (also called photoswitchable) FPs (PCFPs), and reversibly photoswitchable FPs (RSFPs). 8,9 PAFPs can be activated from a non°uorescent (dark) state to a°uorescent state, whereas PCFPs undergo a conversion from one color to another. In contrast to PAFPs and PCFPs, RSFPs can be reversibly photoswitched between the active and inactive states. In this review, we focus on the recently developed PMFPs for commonly used SR microscopies and discuss the concepts and strategies for optimizing and choosing the biochemical and photophysical properties of PMFPs to enhance the spatiotemporal resolution of SR techniques, especially time-lapse live-cell SR imaging.

Breaking the Di®raction Limit with Photomodulatable FPs
The image of an in¯nitely small object under a light microscope consists of a central spot surrounded by a series of higher-order di®raction rings. The size of the central spot (also called the Airy disk) is equal to the di®raction limit, d, which is given by the equation d ¼ 0:61 /NA, where is the emission wavelength and NA is the numerical aperture of the objective. 10 Fluorophores within the di®raction limit remain indiscernible from each other if their°u orescent signals are simultaneously recorded. In the past decades, several SR techniques have been developed to break the di®raction limit. The main principle of these techniques is to prevent the simultaneous excitation or emission of all the°u orophores in the sample using nonuniform sub-di®raction excitation or emission within the di®raction spot, successively changing the excitation or emission location and recording time-sequential images. If digital cameras are used, the image within one di®raction-limited zone occupies several pixels. The subdi®raction excitation or emission locations are sequentially moved across the image¯eld to ensure that all the°uorescent molecules within one di®raction-limited zone will be recorded. The timesequential information on the camera pixels, either the intensity and/or°uctuation of each pixel or the centroid of the single-molecule°uorescent distribution, is used to reconstruct an SR image using di®erent algorithms in di®erent SR techniques. We refer the readers to the recent excellent reviews for in-depth descriptions of these techniques. [11][12][13] Generally, there are two strategies of nonuniform subdi®raction excitation or emission: the stochastic emission or the patterned excitation. The¯rst strategy uses PMSFs whose emissions are stochastically controlled by the illumination light so that one, several or an ensemble of molecules within one di®raction-limited zone are recorded at a time. The (f)PALM/STORM 1-3 and single-moleculebased variations [14][15][16][17] utilize this strategy to improve resolution by localizing the position(s) of a single°u orophore molecule or overlapping°uorophores, respectively, whereas the super-resolution optical°u ctuation imaging (SOFI) technique 18 uses the intensity°uctuation of pixels resulting from the capture of sequential images for a cross-correlation analysis and the SR reconstruction. The second strategy, used in stimulated emission depletion (STED), 7 ground state depletion (GSD) microscopy 19 and structured illumination microscopy (SIM) 6 SR microscopies, relies on the patterned sub-di®raction excitation (point pattern or line pattern) to reduce the size of an ensemble of excited°uorophores. During the imaging of the subdi®raction pools at each time point, the signal from the molecules outside of the subdi®raction area in the di®raction-limited area will contribute to the background and signal noise and decrease the resolution. Thus, the near-¯eld scanning optical microscopy (NSOM) 20 that also uses this strategy has a smaller excitation spot size and a lower sur-rounding°uorescence background within one diffraction-limited zone compared with other SR techniques. Both conventional FPs, such as GFP, and PMFPs can be used for this strategy. However, PMFPs require orders of magnitude lower light intensity by using nonlinear subdi®raction excitation 5,7 and further improve spatial resolution in saturation-based SR techniques (NL-SIM 21 and RESOLFT 4 ). Notably, our recently developed PSFP Skylan-NS and patterned activation NL-SIM (PA NL-SIM) make possible practical noninvasive time-lapse live-cell imaging with very high spatiotemporal resolution. 22 For both strategies and their related SR techniques, three key characteristics to be discussed later are very important for improving the spatiotemporal resolution: (1) the size of the subdi®raction excitation pattern (spot or line), (2) the°uorescence signal of the subfocus location, and (3) the signal-tobackground ratio (SBR). We will focus on recently developed PMSFs and discuss their photochemical properties that are related to resolution improvement in commonly used SR techniques, especially live-cell SR techniques.

Photophysical and Biochemical Properties of Photomodulatable FPs
Since the discovery of PMFPs, extensive e®orts have been made to engineer FPs with improved properties, with the aim of achieving better image accuracy and precision. The following are some basic properties that determine whether an FP is suitable for use in SR microscopy.

Monomeric properties
This property is very important but often overlooked in SR techniques. In our opinion, it should be the¯rst consideration when choosing an appropriate PMFP for SR imaging as it is not useful to obtain the wrong information from the mislocated arti¯cial structures even if the resolution is extremely high. For a°uorescent protein, it is worth noting that its oligomeric state depends on the concentration, which is related to the local environment in living cells. Many \monomeric" FPs tested in solution or by gel electrophoresis in vitro are not as monomeric as they were expected to be and form oligomers once they are con¯ned in a crowded in vivo environment such as a membrane. An elevated local concentration of the FPs is the main cause as sedimentation velocity experiments show that FPs that are not true monomers gradually form dimers and higher-order oligomers with increased concentration in vitro. 23 For any SR technique, although di®erent target proteins have di®erent expression levels and intrinsic oligomerization tendencies, a monomeric PMFP should be chosen to minimize the risk of disturbing the localization and function of the target protein. Notably, dimeric FPs may cause articial clusters of the target proteins by enhancing the dimerization tendency of the target protein. 24 Studies by our group and others show that the oligomerization of FPs a®ects the target protein localization and even function. 23,24 Fortunately, many truly monomeric PMFPs with very good properties have been developed for common SR techniques by our lab and others ( Table 1).
The brightness was determined as the product of extinction coe±cient and quantum yield.

Brightness
Brighter FPs are always in demand because brightness is a key factor that directly a®ects the imaging resolution. For all SR imaging methods, brighter°u orescence means that more photons can be collected by the detector and a higher SBR can be achieved. Moreover, brighter FPs require a lower light intensity or shorter acquisition time for the same°uorescence signal, therefore signi¯cantly reducing photobleaching and damage to the FPs. Additionally, a shorter acquisition time enables higher temporal resolution. The brightness of a°u orescent protein can be calculated using the formula below: where " ex is the molar absorption coe±cient at a given excitation wavelength and È ex=em is the quantum yield at a given excitation and emission wavelength.
Thus, the photon absorption capability and energy conversion e±ciency of a°uorescent protein determine its bulk brightness. However, for singlemolecule-based technology such as PALM, the number of photons emitted by a single molecule in one frame, instead of bulk brightness, is what really matters. The photon number is the product of the photon emission rate and the exposure time, and the photon emission rate is proportional to the bulk brightness and incident light intensity. From experience, the brightness of a°uorescent protein for single-molecule imaging should be no less than 3 Â 10 4 . For blue and green FPs, this threshold value should be higher, as these proteins may have a higher background signal due to auto-°uorescence.

Photostability
For SR imaging techniques such as STED and RESOLFT, which use a much higher laser intensity (10 4 -10 5 higher than PALM and STORM) than traditional microscopy, the photostability of the FP is one of the indispensable factors that need to be considered. On the other hand, for SR imaging techniques such as PALM and STORM, which need to acquire ten thousand frames to reconstruct one image, high photostability of the FP is also preferable. For a more challenging task such as live cell or 3D SR imaging, the photostability of the candidate FP is the¯rst priority. There are several photostable FPs suitable for 2D super resolution imaging in¯xed cells. However, there is still a great shortage of super photostable FPs, especially red and far red FPs, which are useful in 3D and live-cell imaging.
Photostability can be denoted by 1=2 , which is the time required to photobleach half of the maxi-mum°uorescence. The parameter is dependent on illumination conditions, expression systems and fusion proteins; therefore, it is of great importance to measure the photostability in each experiment. Furthermore, note that for RSFPs, which can be turned on/o® repeatedly, 1=2 is the time it takes for the°uorescence maximum of each cycle to decrease to its half value.

Maturation time
With the help of oxygen, newly translated FP peptides go through several chemical reactions that lead to the maturation of the chromophore and to°u oresce. In this process, oxygen acts like a doubleedged sword because increased contact of the FPs with oxygen will increase both maturation speed and the risk of bleaching. In most cases, fast maturation means more detectable°uorescent PMFPfusion molecules at the time of imaging, a higher signal volume, a short acquisition time and, thus, a better time resolution. This is vital for live-cell imaging, which needs to capture nanosecond to second scale movements. Moreover, FPs with short maturation time, such as mEos3.2, 23 will have a higher labeling density, which is a key determinant of the imaging resolution of PALM. 13 Currently, two methods have been introduced to probe the maturation speed of FPs. One method measures the recovery time of denatured FPs in vitro, whereas the other directly records time intervals of FP expression and°uorescence in vivo. It has been reported that FPs with super folding abilities, such as sfGFP,°u oresce even if their fusion protein is expressed in inclusion bodies. 36 Thus, one can use this method to rapidly determine whether an FP has good folding and maturation properties.

Labeling density
An appropriate labeling density, neither too high nor too low, is the key factor for the higher resolution of both stochastic and patterned subdi®raction excitation SR techniques. A high labeling density provides a high resolution; however, the activating laser power (generally 405 nm) must be optimized to activate separated single molecules at a time for PALM/STORM imaging because more molecules in the o® state produce a higher background, and a higher illumination intensity or a longer time is needed to saturate the molecules to the o® state for saturation-based SR techniques. Furthermore, a higher labeling density decreases the dynamic°uctuation in SOFI imaging.

Optimal Photomodulatable FPs for Di®erent SR Techniques
In addition to the properties mentioned above, other properties of PMFPs, including photons per switching event, on-o® duty cycle, contrast ratio and photon conversion e±ciency largely dictate the quality of SR images. Di®erent SR techniques require the consideration of di®erent photophysical and biochemical properties. In this review, we discuss the properties of PMFPs that a®ect their performance in the stochastic or patterned subdi®raction excitation-based SR techniques. Mainly based on our recently developed PCFP and RSFP, we provide important and practical suggestions for choosing PMFPs for most promising live-cell SR techniques, speci¯cally the SOFI and NL-SIM techniques.

PMFPs for PALM/FPALM
In PALM/FPALM imaging, the resolution depends on the localization precision and molecular density. The former describes how well the center of a molecule can be determined, whereas the latter indicates how many molecules can be determined per area unit. Although molecular density has been neglected for a time, it is essential for good PALM imaging. According to the Nyquist criterion, the average distance between two neighboring molecules should be no more than half of the achievable resolution, so a certain labeling density should be achieved. Moreover, the formula of the localization precision below 37 shows that the photon number emitted by a single FP and the contrast ratio of the FP are also important: where s is the standard deviation of the point spread function, a is the pixel size of the imaging detector, N is the photon number and B is the background noise. All three types of PMFPs can be applied to PALM imaging. We refer the readers to the recent excellent review of PMFPs for PALM/STORM techniques 8 and Xiaowei Zhuang's research paper. 24 Here, we highlight the important properties of recent PMFPs and present our considerations on their selection for PALM/STORM techniques.
Many PMFPs have been developed for PALM/ STORM imaging 8 ( Table 1). As mentioned above, the monomeric property is our¯rst consideration for choosing a PMFP. mEos2 is one of the most widely used PCFPs for PALM/STORM microscopy. However, it was found to form dimers and higher-order oligomers at high concentrations. 31,38 We have shown that mEos2 causes incorrect intracellular aggregates when fused to membrane proteins, including G protein-coupled receptor GRM4 and glucose transporter 4 (GLUT4) 23 (Fig. 1). Siyuan Wanga et al. also reported that mEos2 exhibits an arti¯cial visible punctum or clusters in Escherichia coli when fused to the protease ClpP, nucleoid-associated protein H-NS and Tar protein and causes Vimentin¯laments to cluster into thick bundles in mammalian cells. 24 Other PMFPs that have dimerization tendencies and fusion problems include mKikGR, mGeosM, mMaple, PAmCherry, PSCFP2 and tdEos 24 (Table 1). Notably, the authors stated that for those proteins that oligomerize or cluster even without fusion to FPs, even a weak dimerization tendency of FPs may amplify the clustering e®ect of the target protein and cause ar-ti¯cial clusters. 24 Therefore, because it is not known before the experiment whether the target protein has intrinsic aggregation, PMFPs that are less dimeric or not dimeric, including mEos3.2, Dronpa, PAGFP, mMaple3 and PAtagRFP, as well as our recently developed Skylan-S 25 and Skylan-NS, 22 should be considered¯rst (Table 1).
Next, we consider the photon budget and on/o® contrast ratio of PMSFs (Table 1). A higher photon budget leads to higher localization precision and, hence, higher image resolution. 39 Among the monomeric PMFPs mentioned above, PAtagRFP and mEos3.2 exhibited the highest photon budgets (800-900) among PAFPs 23,24 (Table 1). The green PAGFP has two-fold fewer photons (200-300) than mEos3.2 and PAtagRFP 24 (Table 1). Slowly switched RSFPs often have higher number of photons per switching event and are thus better suit for PALM. The green RSFPs rsFastLime, 40 rsEGFP 41 and updated version rsEGFP2 42 gave relatively low photon budgets (< 60 photons per switching event), which would lead to relatively poor localization precision 24 for PALM/STORM imaging. However, it is worth noting that rsEGFP2 is an excellent alternative choice for RESOLFT due to the large number of switching cycles before photobleaching. 42 Skylan-S is our recently developed RSFP speci¯c for SOFI imaging. However, it can also be used for the PALM/STORM technique at a higher illumination intensity because the photon number before bleaching is high enough. 25 The on/o® contrast ratio is another key property to consider for PALM/STORM imaging (Table 1). For example, Skylan-NS, 22 an RSFP we recently developed speci¯cally for NL-SIM and RESOLFT techniques, has a high single-molecule photon number but is suboptimal for PALM/STORM imaging due to its low single-molecule contrast ratio. The bulk°uorescent signal is much higher than that of other green RSFPs, but it is hard to detect single molecules, possibly because Skylan-NS maturates very fast (unpublished) and can be easily activated to the on state by 488 nm illumination. For PALM/ STORM imaging, all the PMFPs are illuminated during imaging, but only one or a few single molecules within the di®raction-limited zone are in the on state and excited to emit photons. However, the abundant other molecules within the di®ractionlimited spot and in the o® state will produce a high background that a®ects the detection of the single molecule. The on/o® contrast ratio is de¯ned as the ratio of°uorescence between the on state and o® state under the illumination of the imaging light only. 43 It is worth noting that the on/o® contrast depends highly on the illumination intensity and exposure time, which should be optimized for each PMFP to obtain the optimal single-molecule detection. Because the emission spectrum of activated single molecules is red shifted (compared with that of the inactivated molecules) and detected in a di®erent color channel, PCFPs have much higher contrast ratios than RSFPs and PAFPs and hence better image quality (resolution). Therefore, PCFPs such as mEos3.2 and mMaple3 are the¯rst consideration for PALM/STORM imaging of only one target protein. mEos3.2 has a very high photon budget and can be used for live-cell PALM microscopy using sCMOS camera-speci¯c single-molecule localization algorithms. 44 Compared with mEos3.2, mMaple3 was reported to have higher signal e±ciency, which is de¯ned as the ratio between the number of detectable PMFP-fusion molecules per cell and the expression level of the fusion protein, 24 and thus a higher labeling density and resolution in theory; however, the lower photon budget of single molecules and lower on-o® ratio of mMaple3 in the red channel may decrease the spatial resolution in practical use. For RSFPs, the on/ o® contrast ratio is highly dependent on the labeling density and the activating laser and excitation laser intensities. rsKame is an excellent candidate as a green marker for dual color PALM/STORM. 45 The ideal red partner still needs to be developed, as PAmCherry mentioned above has the problem of dimerization, whereas PAtagRFP has a very low signal e±ciency. 24 The properties discussed above are also suitable for the updated versions of PALM/STORM that image several overlapping single molecules simultaneously.

RSFPs for SOFI
SOFI is a purely calculation-based imaging approach that can produce background-free, contrastenhanced SR images based on the temporal correlation analysis of°uorescence°uctuation/blinking over hundreds of raw images. 18 In contrast to PALM/STORM imaging, in which only one or several single molecules are stochastically excited, SOFI uses light to stochastically activate small subsets of RSFPs within a densely labeled structure to produce a°uorescence°uctuation. The RSFP properties critical for SOFI imaging include (1) the averaged°uorescence intensity in the°uctuation state, (2) the on/o® contrast ratio, (3) the photostability, and (4) the oligomerization tendency. Thē rst three properties determine the°uctuation range of the imaged pixels and the SOFI signal, which are essential to the spatial resolution, and the last may lead to arti¯cial aggregation of the target proteins. As in PALM/STORM imaging, the on/o® contrast ratio in SOFI imaging is highly dependent on the labeling density, laser power and exposure time. There are only a few RSFPs reported for SOFI imaging. Dronpa and rsTagRFP could be used for SOFI imaging. However, both have low averaged°u orescence intensity in the°uctuation state, photostability, and on/o® contrast ratios, which produce low SOFI signals. 25 Recently, we developed the novel monomeric green RSFP Skylan-S. Compared with Dronpa, Skylan-S has a much higher on/o® contrast ratio and photostability 25 (Fig. 2). Notably, Skylan-S exhibits a 4-fold improvement in the°uctuation range of the imaged pixels and an order-of-magnitude increase in averaged°uorescence intensity in the°uctuation state. The higher°uctuation and averaged°uorescence intensity in the°uctuation state of Skylan-S produces a higher second-order cumulant value than that provided by Dronpa at each time point. With Skylan-S, a high resolution SOFI image of live cells can be obtained.

RSFPs for RESOLFT/Nonlinear SIM (NL-SIM)
RESOLFT and NL-SIM belong to the saturated depletion-based SR techniques, which can be implemented by patterned photoactivation of RSFPs to dramatically lower the illumination intensities and decrease the size of the excitation ensemble (point or line). RSFPs can be switched o® from a long-lived state under dramatically reduced laser power (nearly one million times lower than the intensity used in STED), which helped the invention of reversible saturable optical°uorescence transitions stimulated emission (RESOLFT). Derived from STED, RESOLFT also uses a donutshaped beam to switch o®, but not deplete,°uorescence in the beam-covered region, and only the molecules at the small center of the beam remain°u orescent; by scanning the small center across the sample, RESOLFT can generate a SR image. Similarly, by replacing the donut-shaped beam with other illumination patterns, such as the sinusoidal stripes used in structured illumination microscopy, and by saturating one of the \on" or \o®" states, nonlinear structured illumination microscopy (NL-SIM) increases the imaging speed compared with RESOLFT, and higher resolution than SIM. These two techniques require fewer raw images to reconstruct a¯nal SR structure, making them suitable for live cell imaging.
The following characteristics of RSFPs are most critical for the two SR microscopies:¯rst, the inte-grated°uorescence signal across each switching cycle, which depends on the absorption cross-section, e®ective quantum yield and characteristic switching time from the°uorescent \on" to \o®" state; second, the°uorescence contrast ratio of the \on/o®" states; and third, the photostability under excitation and depletion. As mentioned above, there are not very many truely monomeric RSFPs with high photon budgets suitable for RESOLFT/NL-SIM imaging. The Dronpa and rsEGFP families have been exploited for the saturated SR imaging. However, Dronpa has a limited number of switching cycles, relatively low°uorescence signal and poor contrast ratio under physiological conditions. Using rsEGFP, Grotjohann et al. demonstrated RESOLFT imaging of various living samples at $ 40 nm resolution, including bacteria, mammalian cells, and organotypic tissue slices. 41 Additionally, by developing a fast switching mutant rsEGFP2, they increased the pixel dwell time 25-to 250-fold and revealed changes in the ER structure occurring in < 1 s. 42 rsEGFP2 exhibits extremely high photostability, showing potential utility in timelapse live-cell imaging. However, the low photon numbers restrict its ability to reach the desired resolution at a reasonable SBR. Currently, we developed a truly monomeric RSFP, Skylan-NS (short for Sky lantern for Nonlinear Structured illumination), with excellent RSFP characteristics for SR live imaging. Skylan-NS produces $ 10-fold more photons per switching cycle than rsEGFP2 and a 3.6-or 1.8-fold higher \on/o®" contrast ratio than Dronpa or rsEGFP2, respectively. Notably, it provides $ 700 switching cycles before bleaching to 1/e of the initial°uorescence intensity. cell imaging to achieve $ 60 nm resolution at subsecond rates with only a 100 W/cm 2 illumination intensity for tens of time points across 40 Â 40 m 2 elds-of-view.

Future Directions and Challenges:
Brighter, Higher Contrast Ratio, and more Colorful PCFPs have a higher contrast ratio and spatial resolution than RSFPs. However, they occupy two channels, making dual-color PALM imaging either with another PCFP or RSFP (PAFP) a challenge. Although it has been reported that mEos2 can be used with PSCFP2 for dual-color PALM microscopy by sequential imaging of mEos2 and PSCFP2, 46 it is di±cult to convert all the green mEos2 to red or bleach all the green mEos2 molecules before acquiring the green signal from the photo-converted PSCFP2. This is also the problem for mEos3. Dendra2 has very high conversion e±ciency; however, the single molecule properties are not suitable for PALM imaging (unpublished data). There is an urgent need for PCFPs with high photo-conversion e±ciency and excellent single-molecule photons for dual-color PALM/STORM imaging. Green RSFPs are the best developed ones among RSFPs for both stochastic and patternedexcitation microscopies. In the future, brighter green RSFPs with enhanced contrast ratios and photostability will be required for live-cell SR techniques with higher spatiotemporal resolution and more time points. Additionally, red RSFPs and far-red RSFPs with optimal photochemical characteristics are highly needed for dual color SR imaging.