A highly photostable and bright green fluorescent protein

The low photostability of fluorescent proteins is a limiting factor in many applications of fluorescence microscopy. Here we present StayGold, a green fluorescent protein (GFP) derived from the jellyfish Cytaeis uchidae. StayGold is over one order of magnitude more photostable than any currently available fluorescent protein and has a cellular brightness similar to mNeonGreen. We used StayGold to image the dynamics of the endoplasmic reticulum (ER) with high spatiotemporal resolution over several minutes using structured illumination microscopy (SIM) and observed substantially less photobleaching than with a GFP variant optimized for stability in the ER. Using StayGold fusions and SIM, we also imaged the dynamics of mitochondrial fusion and fission and mapped the viral spike proteins in fixed cells infected with severe acute respiratory syndrome coronavirus 2. As StayGold is a dimer, we created a tandem dimer version that allowed us to observe the dynamics of microtubules and the excitatory post-synaptic density in neurons. StayGold will substantially reduce the limitations imposed by photobleaching, especially in live cell or volumetric imaging.

O ver the past few decades, owing to improvements in gene transfer and protein-targeting techniques, researchers have succeeded in labeling subcellular components, such as organelles and the cytoskeleton, with fluorescent proteins (FPs), enabling efficient observation of these structures in living cells and organisms 1,2 . For most applications, as low as possible expression levels of an FP-tagged protein are preferable to minimize the perturbation of the system under observation 3 . However, obtaining high signal-to-noise-ratio images from samples with low-level expression requires strong excitation light, which often causes severe photobleaching. Similarly, obtaining a high spatiotemporal resolution to study the fast dynamics of fine subcellular structures requires continuous acquisition of images on a time scale from seconds to minutes 4,5 , which inevitably leads to noticeable FP photobleaching. Highly photostable FPs are, therefore, needed to enable fast and long super-resolution imaging.
To date, the development of photostable FPs has nearly always been accompanied by a decrease in brightness 6,7 . Exceptions could be observed in the development of stable and bright blue-emitting FPs, such as Azurite 8 and EBFP2 (ref. 9 ), and a recently published photostable yellow-emitting FP, mGold 10 . The main complication arises from molecular oxygen (O 2 ), which is required for maturation of chromophores of FPs, and high accessibility to O 2 contributes to the increase in maturation speed and maximal brightness in cells 11 . However, decomposition of FP chromophores also involves O 2 , which reacts in a photochemical reaction with the chromophores that remain in their singlet or triplet excited states 12 . In this regard, enhanced O 2 accessibility decreases photostability. Accordingly, there is a tradeoff between brightness and photostability in FP performance.
A typical example of this tradeoff can be seen in the evolution of two monomeric orange-emitting FPs, mOrange 13 and mKus-abiraOrange (mKO) 14 , which were developed independently by two research groups. mOrange matures faster but photobleaches more easily than mKO. Subsequently, a variant of mOrange (mOrange2) was developed that achieved photostability at the expense of maturation speed 6 . On the other hand, a fast-maturing variant of mKO (mKO2) was developed at the expense of photostability 15 . In addition, although a large number of bright green-emitting FPs have been developed, most of them are less photostable than enhanced GFP (EGFP). Because of the apparent inverse relationship between photostability and brightness in the current generation of FPs, few substantial gains in designing bright and photostable products have been made.
Here we report the cloning of an FP with exceptional photostability from the jellyfish C. uchidae. In contrast to most previous engineered FPs, introduction of a single point mutation substantially increased the brightness of the FP without compromising the photostability. The resulting protein, called StayGold, can be localized to subcellular components with appropriate tags. To fully benefit from the rich photon budget, we combined StayGold with super-resolution SIM. In live cell imaging experiments, we visualized the ER and mitochondria with improved spatiotemporal resolution and a considerable extension of the observation period. In coefficient 19 (Supplementary Fig. 4a) was 159,000 M −1 cm −1 at neutral pH. CU17S/V168A had the same excitation and emission spectra as CU17S (Fig. 1g), and the fluorescence quantum yield (QY f ) was 0.93. These values indicate that the molecular brightness of CU17S/V168A surpasses the most popular Aequorea GFP variant EGFP or other bright green-emitting FPs (Table 1). Furthermore, we noted that CU17S/V168A possesses outstanding photostability, which we thought justified its being called 'StayGold' .
Outstanding photostability of StayGold. We compared the photostability of StayGold with that of four green-emitting FPs: EGFP (ref. 11 ), SiriusGFP (ref. 20 ), mNeonGreen (ref. 21 ) and mClover3 (ref. 22 ). These five FPs were treated under the same conditions in parallel. First, the proteins were expressed in E. coli and purified. Based on the protein concentrations determined with a Bradford Assay Kit, we prepared 1 μM FP solutions in polyacrylamide gel. The solution/gel mixtures, sandwiched between two coverslips, were subjected to photobleaching and imaging experiments with continuous unattenuated arc-lamp illumination (5.6 W cm −2 ). This simple comparison revealed that StayGold was brighter and strikingly more photostable than any of the other FPs (Fig. 1h, top). We adopted the currently accepted standard method for evaluating FP photostability 6 . First, the extinction coefficients ( Supplementary  Fig. 4a) at the center wavelength of the illumination (488 nm) and the fluorescence quantum yields of the other FPs were determined (Table 1). Then, consideration of the irradiance (5.6 W cm −2 ) and the determinants of brightness of the five FPs yielded the normalized curves for their photobleaching (Fig. 1h, bottom). For each FP, the time for photobleaching from an initial emission rate of 1,000 photons/s/molecule down to 500 (t 1/2 ) was calculated (Table 1). Of the four previously reported FPs, EGFP had the highest t 1/2 value at 701 seconds. However, the t 1/2 of StayGold far exceeded this value, surpassing 10,000 seconds. These results indicate that StayGold can emit over ten-fold more photons before photobleaching than the other green-emitting FPs.
We examined the brightness-adjusted photostability of fluorescence from living cultured HeLa cells that stably expressed each of the five FPs ( Fig. 1i and Supplementary Fig. 5). The conspicuous photostability of StayGold was again apparent. For example, whereas the t 1/2 of StayGold was greater than 10,000 seconds, that of EGFP was below 500 seconds (Table 1). In one attempt, we mixed StayGold-expressing cells and mNeonGreen-expressing cells for direct comparison of photobleaching in a single field of view and verified the large difference in their photobleaching rates ( Fig. 1j and Supplementary Video 1, first half). It should be noted that the two cell populations were indistinguishable at t = 0 because they had the same initial fluorescence intensities. We also performed a one-to-one comparison between StayGold-expressing and EGFP-expressing cell lines and found that the former was brighter and more photostable Fluorescence images of an isolated polyp (b) and a female medusa (c) superimposed on differential interference contrast images. Scale bars, 0.5 mm. d, Normalized excitation (dotted line) and emission (solid line) spectra of a tissue homogenate prepared from C. uchidae medusae. e, Amino acid sequence alignments of CU17S, StayGold, EGFP and DsRed. Residues whose side chains form the interior of the β-barrel are shaded (EGFP and DsRed). Asterisks: residues responsible for chromophore synthesis. The V168A mutation is indicated in red. f, Absorption spectra of CU17S (dotted line) and StayGold (solid line), normalized against the peak at 280 nm. g, Normalized excitation (dotted line) and emission (solid line) spectra of CU17S and StayGold. h, i, Photostability of green-emitting FPs under continuous WF illumination (5.6 W cm −2 ). Plotted as measured intensity versus time (top) or as intensity versus normalized total exposure time with an initial emission rate of 1,000 photons/s/molecule (bottom). h, Purified protein (1 μM) in polyacrylamide gel. The data are shared with Fig. 2b. i, Expressed in human (HeLa) cells in HBSS. j, k, Fluorescence images of StayGold-expressing and mNeonGreen-expressing (j) or EGFP-expressing (k) cells (stable HeLa cell transformants) at the indicated times (minutes:seconds). Illumination intensity: 16  than the latter ( Fig. 1k and Supplementary Video 1, 2nd half). The superior photostability of StayGold relative to EGFP was observed across the full range of light intensities of the arc-lamp illumination ( Fig. 1l). To examine the involvement of O 2 in the photobleaching of the five green-emitting FPs, we exposed FP-expressing living cells to continuous arc-lamp illumination under anoxic and hyperoxic  conditions, in addition to normoxic conditions, and found that StayGold was O 2 -sensitive, as were the other FPs ( Supplementary  Fig. 6). We then performed time-lapse imaging experiments after cDNA transfection into HeLa cells. Although StayGold showed stronger fluorescence than mNeonGreen half a day or 3 days after transfection ( Supplementary Fig. 7a), their maturation rates proved to be the same (Fig. 1m). Quantification of the cellular brightness of the five green-emitting FPs verified the most efficient maturation of both StayGold and mNeonGreen (Supplementary Fig. 7b and Table 1). Compared to StayGold, CU17S is poorly folded in both bacterial and mammalian expression systems ( Supplementary Fig. 4b). It is noted, however, that folded CU17S is as photostable as StayGold ( Fig. 2a and Supplementary Table 1a).
Over the past 2 decades, photostability has been improved for red-, orange-and cyan-emitting FPs to develop several robust FPs, such as TagRFP-T (ref. 6 ), mOrange2 (ref. 6 ), mScarlet-H (ref. 7 ) and mTFP1 (ref. 23 ). Although a direct comparison of FPs of different colors is impossible, it is possible to quantitively assess their photostability on the basis of t 1/2 calculations ( Supplementary Fig. 4c). Figure 2b and Supplementary Table 1b show the normalized photobleaching data for 16 FPs that we investigated; the findings demonstrate the superiority of StayGold's glowing capability within the realm of FPs.
Labeling ER lumen with StayGold-based markers. We expressed StayGold and other green-emitting FPs as soluble markers of the ER lumen by fusing the calreticulin signal sequence and the ER-retrieval motif to their N-and C-termini, respectively (er-FPs). In our efforts to direct StayGold to the ER lumen, however, we faced two technical challenges.
First, cysteine residues in an FP prevent correct folding of the β-barrel due to the formation of unnecessary disulfide bonds in the oxidizing ER environment. To date, cysteine-less variants of FPs (oxFPs), including oxGFP, have been successfully engineered and used to efficiently label the ER from the inside 1 . StayGold has five cysteine residues. After site-directed random mutagenesis, we generated oxStayGold, in which Cys174 and Cys208 are replaced with Ile. Photobleaching experiments using purified proteins showed that these two mutations did not affect the outstanding photostability of StayGold ( Fig. 2a and Supplementary Table 1a).
Second, StayGold's head and tail are relatively short (Fig. 1e), which makes this FP sensitive to C-and N-terminal fusions. We borrowed termini from other proven FPs and found that nine amino acids from the N-terminal region of EGFP (n1 or n2) and ten amino acids at the C-terminus of dfGFP (ref. 24 ) (c4) could be fused to StayGold (Supplementary Fig. 8) to improve its targeting function to some subcellular components ( Supplementary Fig. 9).

Analysis of ER network rearrangements.
The ER is organized into an interconnected reticular network of branching tubules that extends throughout the entire cytosol. Although previous studies mostly imaged only a part of the ER network (the peripheral ER), such local imaging is not appropriate for the assessment of photobleaching in the ER because diffusional exchange occurs between bleached and unbleached FPs. Therefore, we always placed the entire cell within the field of view. For fast and cell-wide super-resolution imaging of the ER network, we employed three-dimensional structured illumination microscopy (3D-SIM), a WF microscopy technique that demands great photostability of dyes because it acquires and combines 15 (five phases × three angles) images per frame 25 . Unlike multiphoton and light sheet microscopy, 3D-SIM excites fluorophores both above and below the focal plane while enabling optical sectioning ( Supplementary Fig. 10). Accordingly, while a z-series is collected for volumetric visualization of a specimen, each fluorophore is almost continuously exposed to excitation light. We performed comparative volumetric 3D-SIM imaging experiments and observed substantial photobleaching of er-oxGFP, but not er-(n2)oxStayGold(c4), after acquisition of a z-series stack (Fig. 3c).
Two recent studies elegantly visualized the peripheral ER network using grazing incidence (GI)-SIM, a new two-dimensional (2D)-SIM technique that creates an appropriately thick illumination field over a cover glass ( Supplementary Fig. 10). This technique can image the entire volume of the basal cell cortex for more comprehensive analysis of the ER than total internal reflection fluorescence structured illumination microscopy (TIRF-SIM). Owing to the higher fluorescence intensity, GI-SIM achieves higher temporal resolution than TIRF-SIM. Using GI-SIM, Nixon-Abell et al. and Guo et al. imaged the ER near the basal plasma membrane at 40 and 266 frames per second, respectively, at a ~100-nm resolution, to discover that the so-called 'peripheral ER sheets' are dynamic, dense ER tubular arrays that oscillate very rapidly (4-10.1 Hz) 4,5 . We confirmed these findings by employing an advanced 3D-SIM system with a cell-wide field of view that covered the perinuclear as well as peripheral regions. The entire tubular network was visualized at a temporal resolution of 134.47 frames per second (Extended Data Fig. 1 and Supplementary Video 2).
To thoroughly investigate multiple mechanisms of tubular ER formation, Guo et al. also used GI-SIM to observe ER network Absorbance maximum. b Emission maximum. c Absolute extinction coefficient at λab (left) and 488 nm (right). The measurement was based on the fact that, after alkali denaturation of these FPs, the chromophore, containing a dehydrotyrosine residue conjugated to the imidazolone group, absorbs light maximally at 447 nm with a molar extinction coefficient of 44,000 M −1 cm −1 (ref. 19 ). See Supplementary Fig. 4a. d Fluorescence quantum yield measured using an absolute photoluminescence quantum yield spectrometer. e Product of ε (λab) and Qy f . This value reflects the molecular brightness of an FP. f Cellular brightness calculated from data shown in Supplementary Fig. 7b. The fluorescence from each green-emitting FP (with excitation at 488 nm) was corrected by mCherry fluorescence and then normalized to that of EGFP (ref. 43 ). Equimolar co-expression of a green-emitting FP and mCherry using the bicistronic expression system 44 . g Time in seconds to reduce emission rate from 1,000 to 500 photons/s/molecule under WF illumination. h Amount of purified protein from 1 L of bacterial culture. All values were measured in this study. SiriusGFP is a variant of EGFP that exhibits a two-fold increase in photostability relative to EGFP but a three-fold decrease in brightness 20 .
rearrangements with a temporal resolution of 0.5 Hz 5 . By contrast, we acquired single-layer 3D-SIM images of three neighboring HeLa cells expressing er-(n2)oxStayGold(c4) (Fig. 4a,b) at 2.6 Hz continuously over 6 minutes, during which the cells were challenged with a histamine and an anti-histamine reagent sequentially to initiate Ca 2+ mobilization and then shut it down (Supplementary Video 3). As evidenced by fast Fourier transform (FFT) spectral data 26 (Fig. 4c), the super-resolution quality of the 3D-SIM images was well preserved throughout the experiment (Extended Data Fig. 2a). To comprehensively characterize ER network dynamics, we performed fully automatic image processing (Extended Data Fig. 2b). After extraction of fluorescent ER tubules, a 2K image was divided into sub-blocks, each of which consisted of 16 × 16 pixels. We used a custom-made algorithm to quantify the displacement of ER tubules in each sub-block. As shown by the heat maps (Fig. 4d, top), the ER network was most mobile near the plasma membrane at the free edge. Interestingly, the overall ER network was rather immobile (Fig. 4d, bottom) while inositol 1,4,5-trisphosphate-induced Ca 2+ release took place (Fig. 4e). After conducting separate experiments using er-(n2)oxStayGold(c4) and er-oxGFP and analyzing the data comparatively, we confirmed that high photostability is essential to this type of ER imaging by 3D-SIM (Supplementary Fig. 11 and Supplementary Discussion 1). We also generated another construct, er-(n2)oxStayGold (Fig. 4f,g), which appeared to label ER more moderately than er-(n2) oxStayGold(c4). The reversible, Ca 2+ mobilization-dependent stabilization of ER network dynamics was verified in a sustainable 3D-SIM experiment that visualized er-(n2)oxStayGold-accumulating ER tubules at 1.1 Hz (Fig. 4h,i and Supplementary Video 4).

Labeling mitochondrial matrix with StayGold-based markers.
We engineered StayGold and mNeonGreen to localize in the mitochondrial matrix by fusing a tandem repeat of the COX VIII pre-sequence 27 (Supplementary Fig. 12a; mt-StayGold and mt-mNeonGreen). When transfected into HeLa cells, substantial cell-to-cell variation in the expression level was observed for each construct. Among mt-StayGold transfectants, we chose a rather dim cell with modestly labeled mitochondria for continuous 3D-SIM imaging. However, due to the high motility of mitochondria, we had to increase the imaging frame rate and, accordingly, the illumination intensity to achieve sufficient spatial resolution. mt-StayGold provided moderate but well-sustained signals throughout an experiment (Supplementary Video 5), whereas mt-mNeonGreen lost its initial bright signals halfway under the same conditions.
Although we chose cells with moderate FP expression levels for live imaging, it was also interesting to what extent mitochondrial labeling intensity could be increased with the soluble markers. We noticed that mt-(n1)StayGold gave much brighter mitochondrial labeling than mt-StayGold (Supplementary Fig. 12b and Supplementary Discussion 2). We constructed HeLa cells that constitutively expressed mt-(n1)StayGold. The stable transformant proliferated normally ( Supplementary Fig. 5b) while exhibiting strong and uniform labeling of mitochondria, enabling fast, sustained, wide 3D-SIM imaging for efficient detection of their fission and fusion (Extended Data Fig. 3). As these events take place under a variety of metabolic stress conditions 28 , high-resolution imaging of live mitochondria should also be cell-wide and long-term for a comprehensive understanding of their regulations.
StayGold tandem dimer. For protein fusion applications, we attempted to fuse two copies of a StayGold construct that was appended at both the N-and C-termini using a flexible linker (EV linker) 29 to create a tandem dimer ( Supplementary Fig. 13a). The resultant FP construct, tdStayGold, showed the same photostability as StayGold (Fig. 2a, Supplementary Fig. 4b and Supplementary  Table 1a). An OSER assay 3 using CytERM-tdStayGold showed almost no whorl structures, indicating the monovalent fusion of tdStayGold ( Supplementary Fig. 13b).
Microtubule (MT)-associated end-binding protein 3 (EB3), a core component of the MT plus-end protein complex, surfs on the growing tips of MTs. As MT dynamics was found to depend on the monomer/dimer equilibrium of EB3 (ref. 30 ), we tagged this protein with tdStayGold (Extended Data Fig. 4a). We imaged COS cells that expressed EB3 = tdStayGold by WF microscopy continuously. We visualized many fluorescent comets moving along MTs toward the cell periphery with minor background (cytosolic) fluorescence (Supplementary Video 6). No photobleaching was noted during the 30-minute observation. As the EB3 dynamics seemed to be preserved until the end of measurements, it is probable that the illumination mode did not cause any substantial phototoxicity.
PSD-95 is an important excitatory post-synaptic density (PSD) protein that regulates the trafficking and localization of glutamate receptors and signaling molecules to modulate synaptic plasticity 31 . As PSD-95 is involved in multivalent interaction networks in PSD (ref. 32 ), the FP used for tagging this protein should be monomeric. We tagged PSD-95 with a similar tandem dimer, tdoxStayGold. We used spinning disk confocal microscopy to image cultured neurons that expressed PSD-95-tdoxStayGold and observed disc-shaped signal assemblies on dendritic spines and shafts (Extended Data Fig. 4b and Supplementary Discussion 3).

Visualizing SARS-CoV-2 assembly by StayGold/SIM technology.
In 2020, the Coronavirus Disease 2019 (COVID-19) pandemic caused by SARS-CoV-2 substantially affected our study aimed at practical application of StayGold. In hopes of combatting the pandemic with resources at hand, we embarked on an urgent project to develop a technique for visualizing SARS-CoV-2 using FPs. This virus expresses a surface spike (S) glycoprotein that is responsible for viral entry into the host cell; the protein ectodomain is composed of the S1 and S2 subunits. Over the past year, several potent neutralizing antibodies were developed against the S1 protein. We were most interested in nanobodies (Nbs), which are the variable domains of heavy chain-only antibodies and can be genetically engineered to be fluorescent antigen detectors via FP fusions 16,33 . Screening for S1 binding ability in a cDNA library of 10 trillion synthetic Nb sequences yielded a high-affinity (1.4 nM) binder. Due to its high neutralization potency, the Nb, hereinafter referred to as Nb(S1), effectively alleviated symptoms in SARS-CoV-2-infected Syrian hamsters after nasal delivery 34 .
Through structure-guided design, two recent studies successfully engineered multivalent Nb constructs with extremely high neutralization potency in vitro 35,36 . Similarly, we expected that divalent binding by StayGold fusion would enhance target binding by increasing avidity. We inserted the EV linker between Nb(S1) and StayGold to generate Nb(S1) = = StayGold (Fig. 5a). This approach achieved good yield, soluble expression in bacterial culture and  bars in b,d,e,g and i, 10 μm. efficient purification of the recombinant chimeric protein. The full maturation of the StayGold chromophore was verified by absorption measurements (Fig. 5b). Furthermore, the molecular integrity of the fusion construct was verified by SDS-PAGE (Fig. 5c). A surface plasmon resonance experiment revealed that Nb(S1) = = StayGold had somewhat lower affinity toward S1 than Nb(S1) (Fig. 5d).
VeroE6 cells were engineered to constitutively express TMPRSS2, which promotes viral entry by cleaving S protein. VeroE6/TMPRSS2 cells were infected with SARS-CoV-2 and fixed and processed for immunoreaction with Nb(S1) = = StayGold, followed by nuclear counterstaining with Vybrant dye. By conventional WF microscopy, we observed islands of infected cells filled with green fluorescent  (Fig. 5e). Fluorescent puncta were observed in infected cells reacted with Nb(S1) = = StayGold, but not in uninfected cells or non-reacted infected cells, indicating that the signal was specific to SARS-CoV-2 infection and S1 (Fig. 5f). Like many other RNA viruses, SARS-CoV-2 markedly remodels the endomembrane system of the host cell ( Supplementary Fig. 14). The virus-induced, ER-derived membranous structures include convoluted membranes (CMs) and double-membrane vesicles (DMVs). CMs accumulate viral non-structural proteins (nsps) that rearrange host cell membranes to establish DMVs, the major platform for viral replication in double-stranded RNA (dsRNA) species. DMVs and CMs have been well-characterized by a variety of advanced imaging techniques, such as cryo-electron microscopy (cryo-EM) [37][38][39] . By contrast, although viral assembly takes place mainly in the ER-Golgi intermediate compartment (ERGIC), the assembly process involving the S protein has barely been characterized. A recent study, however, visualized SARS-CoV-2 assembly pathways through a unique combination of cryo-EM techniques: 40 newly synthesized S proteins are transported in small (~100 nm) dense vesicles, which are fused with ERGIC single-membrane vesicles (SMVs) measuring a few hundred nanometers in diameter, and the interior of these vesicles is lined with S proteins. For the comprehensive high-spatial-resolution mapping of S proteins, we employed volumetric 3D-SIM to resolve punctate signals (Extended Data Fig. 5). The highly photostable StayGold enabled large-scale super-resolution microscopy observations. With the improved spatial resolution of 3D-SIM (100-120 nm), we were able to observe S signals in the shape of a ring in an x-y image (Fig. 5g). As S proteins on SMV membranes cluster at the assembly site during viral budding, the ring structure appears to be mostly irregular. Subsequently, assembled virions decorated with S proteins are released into the lumen of SMVs; the virions are transported through the secretory pathway and released by the cell. In fact, we often saw green fluorescence signals on the plasma membrane (Fig. 5h).
To further characterize the S signal-containing vesicles, we performed volumetric dual-color 3D-SIM imaging experiments. ERGIC-SMVs, CMs and DMVs were immunolocalized using anti-ERGIC3, anti-nsp8 and anti-dsRNA antibodies, respectively (Fig. 5i-k and Supplementary Fig. 15); these immunolocalizations were labeled with red fluorescence (Alexa Fluor 555). We observed substantial overlap between S and ERGIC3 signals. By contrast, no overlap was observed between S and nsp8 signals or between S and dsRNA signals.
To confirm the reproducibility of Nb(S1) reactivity, we diversified the construct. We fused Nb(S1) to Achilles, a fast-maturing yellow-emitting FP (ref. 41 ), with another linker to generate Nb(S1) = Achilles (Supplementary Fig. 16). In WF microscopy, very similar punctate fluorescence signals were observed in SARS-CoV-2-infected VeroE6/TMPRSS2 cells. Co-immunostaining of nsp8 with red fluorescence (Alexa Fluor 555) revealed considerable variation in the ratio of S and nsp8 signals among infected cells.

Discussion
Live cell imaging has proven to be a powerful method to understand the dynamics of various cellular processes in culture and in vivo. However, because of the limited photostability of the current generation of FPs, maintaining a sufficiently high signal-to-noise ratio for extended periods of time required either a high initial brightness of the sample or a low observation frequency. Both strategies have practical limitations. When cellular proteins are labeled with FPs, a common concern is that bright cells with the corresponding high expression levels of the tagged proteins will show non-physiological behaviors, whereas temporal downsampling may miss transient but important signals. The high photostability and brightness of StayGold allowed us to select moderately bright cells for observation with continuous illumination and to perform experiments in which imaging performance was not limited by photobleaching.
StayGold is endowed with outstanding photostability and excellent brightness (for a discussion of the technical difficulties in determining and comparing photostability, see Supplementary Discussion 4). However, a detailed understanding of how StayGold interacts with O 2 to synthesize its chromophore and why it is so insensitive to photochemical degradation will require determination of the protein structure, which is underway. Also, single-molecule imaging of StayGold, which is also ongoing, will help us to elucidate photophysical properties, including blinking, which is an indicator of inter-system crossing to the triplet state.
We anticipate that er-(n2)oxStayGold(c4) can be effectively combined with methods for rapid super-resolution imaging, such as GI-SIM, to further enhance the maximal duration of imaging experiments. GI-SIM is a specialized 2D-SIM technique that achieves nanoscale resolution on a millisecond time scale. Using GI-SIM, two studies 4,5 discovered the very rapid oscillatory motion of ER tubules, which could not be characterized by conventional spatiotemporal resolution and explains the appearance of continuous sheet-like structures of the ER. Due to its poor z-axis resolution, however, GI-SIM can be applied only to rather loose ER networks in the cell periphery. We employed 3D-SIM instead to image the ER in a cell-wide plane that contained the nucleus. This imaging mode allowed us to detect all changes in the morphology of the ER. Whereas GI-SIM limits the illumination volume to the basal region of a cell, 3D-SIM illuminates the entire cell volume, including portions outside the imaging plane. Thus, our experiments using 3D-SIM required the high photostability of the fluorescent ER marker. Using er-(n2)oxStayGold(c4), we imaged the ER network dynamics of multiple cells simultaneously and continuously for up to 6 minutes, without deterioration of super-resolution image quality ( Fig. 4 and Supplementary Discussion 5).  29 , indicated as a gray bar. b, Absorption spectrum of Nb(S1) = = StayGold. Normalized against the peak at 280 nm. c, Coomassie brilliant blue staining for the visualization of Nb(S1) = = StayGold separated by SDS-PAGE. The migration position is indicated by an arrowhead. d, SPR experiment of S1 binding to immobilized Nb(S1) = = StayGold. Black traces show raw data; red lines show kinetic fit. e, VeroE6/TMPRSS2 cells infected with SARS-CoV-2 at an MOI of 0.1 and fixed at 24 hours post-infection (hpi). Spike protein (green). Nuclei were counterstained with Vybrant (magenta). Representative of n = 9 independent samples. f, SARS-CoV-2-infected (top) and SARS-CoV-2-uninfected (bottom) VeroE6/TMPRSS2 cells. Treated (left) and untreated (right) with Nb(S1) = = StayGold (3.4 μg ml −1 ). Representative of n = 2 independent infections. g, A single-layer 3D-SIM image of SARS-CoV-2 spike protein (green) and Nuclear (blue) extracted from volumetric image data shown in Extended Data Fig. 5. Representative of n = 5 cells over three independent infections. h, A single-layer 3D-SIM image of SARS-CoV-2 spike protein (green), Nuclear (blue) and plasma membrane (red). Representative of n = 2 areas over two independent infections. i-k, Single-layer 3D-SIM images of SARS-CoV-2 spike protein (green) and ERGIC3 (magenta) (i), nsp8 (magenta) (j) or dsRNA (magenta) (k). Nuclear (blue) images are sectional but not SIM-reconstructed. Areas enclosed by white boxes are enlarged below. e,f, WF, wide-field microscopy observation. f-k, VeroE6/TMPRSS2 cells infected with SARS-CoV-2 (MOI = 0.02, 36 hpi). i-k, These single-layer 3D-SIM images are part of volumetric imaging data. Representative of n = 3 imaging experiments from a single sample per staining condition (Supplementary Fig. 15). g,h, Nuclear and plasma membrane images are sectional but not SIM-reconstructed. h,i, Samples are shared. Scale bars in e and f, 20 μm; g, 1 μm; and h-k, 5 μm. All the cells were fixed.
A potential limitation of StayGold is its dimeric nature, and a monomeric version of StayGold is under development. To avoid potential complications that can arise when tagging membrane proteins with dimeric FPs, in our experiments we labeled the ER (Figs. 3 and 4) and the mitochondria (Extended Data Fig. 3 and Supplementary Fig. 12) by expressing StayGold as luminal soluble markers. Although it is possible that the dimeric structure is indispensable for the protein's outstanding photostability, other oligomeric FPs were not exceptionally photostable, and successful monomerization of previous FPs did not, in most cases, affect their photostability substantially (Fig. 2c, Supplementary Fig. 4d and Supplementary Table 1c). Currently, StayGold cannot substitute for popular bright monomeric green-emitting FPs in all applications. mNeonGreen, for example, will still be the first choice for molecular fusion applications in conventional time-lapse imaging experiments where illumination power density can be kept to less than 0.5 W cm −2 .
Substantial efforts have always been required to engineer wild-type FPs for use in fluorescence imaging. Because many wild-type FPs are oligomers, monomerization is one of the most important aims in any engineering effort for an FP (ref. 42 ). It is also noted that most wild-type FPs are sensitive to N-or C-terminal fusions and that the evolution of FPs as useful tools has involved substantial modifications of their N-and C-termini to develop fusion-tolerant FPs, such as the mFruit series 13 . Although successful engineering of truly monomeric forms of StayGold (mStay-Gold) with appropriate appendages at the N-and C-termini may require determination of the crystal structure, we are engaged in their directed evolution via random and rational mutagenesis. mStayGold will then enable the visualization and quantification of tagged proteins at low copy number expressed via genome editing techniques; it will also enable the tracking of such proteins at the single-molecule level in cells over extended periods of time.

Online content
Any methods, additional references, Nature Research reporting summaries, source data, extended data, supplementary information, acknowledgements, peer review information; details of author contributions and competing interests; and statements of data and code availability are available at https://doi.org/10.1038/ s41587-022-01278-2.
Gene construction for mammalian expression (ER targeting). First, (n2) StayGold was constructed by inserting amino acids 4-12 of EGFP between residues 3 and 4 of StayGold. Second, er-(n2)StayGold was generated by extending (n2) StayGold at the N-terminus with the signal peptide from calreticulin and at the C-terminus with an ER retention signal (KDEL). Third, all the five cysteine residues (Cys40, Cys150, Cys165, Cys174 and Cys208) of er-(n2)StayGold were subjected to site-directed random mutagenesis according to a published protocol 52 where multiple degenerative primers were used to mutate amino acid residues randomly at multiple sites simultaneously. Screening for bright fluorescence labeling of the ER of transfected HeLa cells revealed that substituting Ile for both Cys174 and Cys208 was effective, whereas the other cysteines seemed to be indispensable. The resulting construct called er-(n2)oxStayGold was found to have the additional mutation His169Tyr accidentally. The er-(n2)oxStayGold gene was subcloned into the HindIII/EcoRI site of pcDNA3. Finally, er-(n2)oxStayGold(c4) was constructed by inserting amino acids 223-233 of dfGFP (ref. 24 ) into the C-terminal region of er-(n2)oxStayGold.
Protein purification. Recombinant proteins with a polyhistidine tag at the N-terminus were expressed in E. coli (JM109 (DE3)). Transformed E. coli was incubated in a Luria-Bertani medium containing 0.1 mg ml −1 of ampicillin at room temperature with gentle shaking for several days. Protein purification by Ni 2+ affinity chromatography was performed as described previously 53 .

Analytical ultracentrifugation.
Sedimentation equilibrium experiments were carried out as described previously 54 . Purified recombinant StayGold at 2.8 μM in 50 mM HEPES-NaOH (pH 7.4) was analyzed on a Beckman Optima XL-A analytical ultracentrifuge at 24 °C, and absorbance was measured at 497 nm ( Supplementary Fig. 2a).

Gene construction for mammalian expression (mitochondrial targeting). (n1)
StayGold was constructed by inserting amino acids 5-13 of EGFP between residues 4 and 5 of StayGold. mt-StayGold and mt-(n1)StayGold were constructed by fusing a tandem repeat of the cytochrome c oxidase subunit VIII pre-sequence 27 to the N-termini of StayGold and (n1)StayGold, respectively.

Gene construction for mammalian expression (PSD targeting).
The PSD-95 gene 58 was amplified using primers containing 5′-XhoI and 3′-EcoRI sites, and the restricted product was cloned into the XhoI/EcoRI sites of pEGFP-N1 (Clontech, Takara Bio) to generate pPSD-95-EGFP-N1. In parallel, the tdoxStayGold gene was amplified using primers containing 5′-BamHI and 3′-NotI sites, and the restricted product was cloned into the BamHI/NotI sites of pBlueScript Coupler 1 (ref. 56 ). From the resultant plasmid, the KpnI/NotI fragment encoding Coupler-tdoxStayGold (=tdoxStayGold) was prepared and cloned into the KpnI/NotI sites of pPSD-95-EGFP-N1. The substitution of the =tdoxStayGold gene for the EGFP gene resulted in the generation of pPSD-95=tdoxStayGold-N1 for transfection of cultured neurons (Extended Data Fig. 4b).
Cell proliferation assay. HeLa/StayGold, HeLa/EGFP, HeLa/mNeonGreen and HeLa/mt-(n1)StayGold cells were tested in comparison with control HeLa cells. In total, 25,000 cells were seeded onto a standard 60-mm dish and maintained in growth medium (DMEM high glucose, supplemented with 10% FBS). Afterwards, cells were detached by trypsin and suspended in 1 ml of growth medium. Then, 25 μl of the cell suspension was mixed with 25 μl of 0.4% trypan blue solution (Sigma-Aldrich, F8154) for manual counting (Supplementary Fig. 5b).

Cellular brightness assay.
HeLa cells were seeded into 96-well glass-bottom plates (Matsunami Glass, GP96000) and maintained in growth medium (DMEM low glucose, supplemented with 10% FBS). On the following day, cells were transfected with cDNAs. Thirty hours after transfection, cells were imaged on an inverted microscope (IX-83, Olympus) equipped with an LED light bulb (X-Cite XYLIS, Excelitas Technologies), an objective lens (Olympus, UPlanXApo ×4/0.16 NA) and a scientific CMOS camera (ORCA-Fusion, Hamamatsu Photonics). Green-emitting FPs were observed using a filter cube (U-FBNA, Olympus). mCherry was observed using a filter cube (U-FMCHE, Olympus). Cells were transfected with 0.2 μg of pCSII-EF/mCherry-T2A-green-emitting FP per well. Green-emitting FP fluorescence was divided by mCherry fluorescence and normalized to the ratio of EGFP/mCherry as described previously 43 (Table 1 and Supplementary Fig. 7b).
The following experimental conditions were used for eight FP groups: The data were analyzed using Excel (2019). The fluorescence intensity at t = 0 was normalized to 1,000 photons/s/molecule, and the time axis was adjusted according to the standard method 6 .
For the examination of reversible photobleaching, the shutter of the illuminator was closed manually for 1-3 minutes during the photobleaching experiment ( Supplementary Fig. 18).

WF photobleaching (living cells in HBSS).
Living cells on 35-mm glass-bottom dishes were incubated in Hanks' Balanced Salt Solution (HBSS) containing 15 mM HEPES-NaOH (pH 7.4) and imaged on an inverted microscope (IX-81, Olympus) equipped with a standard 75-W xenon lamp, a ×40 objective lens (UPlanSApo ×40/0.95 NA) and a cooled CCD camera (ORCA-AG, Hamamatsu Photonics). The data were analyzed using Excel (2019). The fluorescence intensity at t = 0 was normalized to 1,000 photons/s/molecule, and the time axis was adjusted according to the standard method 6 ( Fig. 1i and Table 1).

WF photobleaching (living cells in DMEM).
Living cells on 35-mm glass-bottom dishes were incubated in DMEM (high glucose) containing 10% FBS without L-glutamine and phenol red (Wako, 040-30095) and imaged on an inverted microscope (IX-81, Olympus) equipped with a standard 75-W xenon lamp, a ×40 objective lens (UPlanSApo ×40/0.95 NA) and a cooled CCD camera (ORCA-AG, Hamamatsu Photonics). The data were analyzed using Excel (2019). The fluorescence intensity at t = 0 was normalized to 1,000 photons/s/molecule, and the time axis was adjusted according to the standard method 6 (Table 1 and Supplementary Fig. 6a).
For the generation of anoxic and hyperoxic conditions, a gas cylinder containing 95% N 2 and 5% CO 2 and a gas cylinder containing 95% O 2 and 5% CO 2 , respectively, were purchased from TOMOE SHOKAI Co., Ltd. (Supplementary Fig. 6b).
SIM for live imaging. Super-resolution 3D-SIM images were acquired continuously (without intervals) on an N-SIM S microscope (Nikon Instruments) equipped with an SR HP Plan Apo λS 100×C/1.35 NA silicone immersion objective and a CMOS camera (ORCA FUSION, Hamamatsu Photonics) at room temperature (25-30 °C). When mitochondrial fission/fusion events were observed, the temperature was set at 37 °C. Each frame was constructed from 15 raw images (five phases and three angles). In all cases, 'slice reconstruction' was used to obtain confocal-like optically sectioned images. Image acquisition and analysis were carried out with NIS-Elements AR (version 5.30.00) (Supplementary Videos 3-5).
Assessment of SIM image quality. Scores of reconstruction results of 3D-SIM are as follows (from the manual of N-SIM).
8: Reconstruction is executed normally. Every separation component is judged as an effective signal and is reflected on the reconstructed image. 7: The 2nd order diffraction component in one direction is judged as a defective signal; the reconstructed image is generated without the 2nd order diffraction component. 6: The 2nd order diffraction components in two directions are judged as defective signals; the reconstructed image is generated without these components. 5: The 1st order and the 2nd order diffraction components in one direction are judged as defective signals; the reconstructed image is generated without these components.
4: The 2nd order diffraction components in three directions are judged as defective signals; the reconstructed image is generated without these components.
3: The 1st order diffraction component in one direction and 2nd order diffraction components in the other directions are judged as defective signals; the reconstructed image is generated without these components.
2: The 1st order diffraction component in one direction and 2nd order diffraction components in the other two directions are judged as defective signals; the reconstructed image is generated without these components.
1: The 1st order and the 2nd order diffraction components in two directions or more are judged as defective signals; the reconstructed image is generated without these components.
Lattice SIM for live imaging. Super-resolution 3D-SIM images were acquired continuously on ZEISS Elyra 7 equipped with a PlanApo ×40/1.4 NA oil immersion objective at 37 °C. The Burst mode for lattice SIM was used to increase the temporal resolution. Image analysis was carried out with ZEN 2014 (version 9.1) (Extended Data Fig. 1

and Supplementary Video 2).
Spinning disk confocal microscopy. Living cells on 35-mm glass-bottom dishes in HBSS containing 15 mM HEPES-NaOH (pH 7.4) were imaged at 30 °C on an inverted microscope (IX-83, Olympus) equipped with a Nipkow spinning disk unit (CSU X1, Yokogawa) and a ×100 silicone objective lens (UPlanSApo ×100/1.35 NA). Images were collected every 53 ms in the streaming mode using an iXon Ultra 897 EMCCD camera (Andor Technology) with gain ×300. The exposure time was set to 50 ms. Image acquisition and analysis were carried out with Olympus cellSens software (version 2.6) (Fig. 3b).
Purified FP proteins (1 μM FP solutions in polyacrylamide gel) between two coverslips were excited continuously on this microscopy system with a ×40 objective lens (UPlanSApo ×40/0.95 NA). Illumination intensity was 1.7 W cm −2 . Image acquisition was performed every 30 seconds with a short exposure time (250 ms) ( Supplementary Fig. 17).
Automatic quantification of ER network rearrangement. Image processing was performed using a customized program based on C++ and OpenCV 3.4.1 (https://opencv.org). First, regions containing observed cells were selected for image processing, and binarized images were generated from reconstructed 3D-SIM images according to the conventional method 60 . Based on the Otsu method, threshold value was automatically optimized in each image to get the best binarization output; this was crucial for analyzing er-oxGFP images that showed weak fluorescence due to photobleaching ( Supplementary Fig. 11). Second, the regions were divided into 16 × 16-pixel sub-blocks. As the widths of most ER tubules spanned several pixels, each sub-block was considered sufficiently small to efficiently detect alterations of ER network morphology. Third, for each sub-block, frame n -1 and frame n were compared, and the proportion of pixels that underwent conversions between + and -was calculated. The calculation was repeated n -1 times throughout the regions. Finally, the calculated values were normalized into integers ranging from 0 to 255 for the generation of n -1 heat maps. The heat maps shown in Fig. 4 are pseudo-colored images that were generated using the '16 Colors' of ImageJ (Extended Data Fig. 2b).
Analysis of rapid motion of ER tubules. We selected regions of interest (ROIs) that contained ER tubules with relatively stable endpoints. ROIs were subjected to image processing, as follows. First, binarized images were created to mask fluorescent objects, which were mostly tubular structures. Second, we developed a program to crop 16-bit raw images according to masked regions. The program was also used to remove noise components that measured <20 pixel area. This design enabled us to characterize ER tubule structures in great detail (16-bit processing) for kymography. Third, we used ImageJ to manually draw the axis containing two endpoints. Fourth, using another customized program, we traced fluorescence signal distribution along the perpendicular bisector to generate kymographic data. After confirming the single peak characteristics, we plotted the position that gave the maximum intensity over time. Finally, we performed FFT analysis Extended Data Fig. 2 | cell-wide, spatiotemporally high-resolution, sustainable imaging of ER network rearrangement. The three neighboring HeLa cells expressing er-(n2)oxStayGold(c4) (Fig. 4a-d) were analyzed. Scale bars, 5 μm. a, Magnified SIM and binarized images at t = 0 min (First frame) and t = 6 min (Last frame) of a peripheral region of the uppermost cell. Compare with Supplementary Fig. 11i. The preservation of 3D-SIM image quality in the last frame was confirmed in all six-minute experiments that employed er-(n2)oxStayGold(c4). Shown is a representative of n = 9 independent transfections, including experiments #1, #2, #3, and #4 ( Supplementary Fig. 11a-d). b, Automatic quantification of ER network rearrangement. See Methods (Automatic quantification of ER network rearrangement). Magnifications of the boxed region are shown (bottom). Considering that the diameter of an ER tubule spanned several pixels, we reasoned that the sub-block size was optimal for the efficient detection of target movement.