Photoregulated fluxional fluorophores for live-cell super-resolution microscopy with no apparent photobleaching

Photoswitchable molecules have multiple applications in the physical and life sciences because their properties can be modulated with light. Fluxional molecules, which undergo rapid degenerate rearrangements in the electronic ground state, also exhibit switching behavior. The stochastic nature of fluxional switching, however, has hampered its application in the development of functional molecules and materials. Here we combine photoswitching and fluxionality to develop a fluorophore that enables very long (>30 min) time-lapse single-molecule localization microscopy in living cells with minimal phototoxicity and no apparent photobleaching. These long time-lapse experiments allow us to track intracellular organelles with unprecedented spatiotemporal resolution, revealing new information of the three-dimensional compartmentalization of synaptic vesicle trafficking in live human neurons.


Introduction
Substances that isomerize upon photoirradiation are useful as molecular devices because their properties can be switched at will with light 1-6 . Fluxional molecules, which undergo rapid degenerate rearrangements in the electronic ground state, exhibit stochastic switching behavior. Despite the great interest in this phenomenon in organic 7,8 , coordination 9 , main group 10 , organometallic 11,12 , and theoretical chemistry 13, 14 , only a few applications of fluxional molecules are known 15,16 . We envision that the juxtaposition of photoswitching and fluxionality could bring about new functional molecules with practical applications. Specifically, we hypothesize that such a molecule could be useful for single-molecule localization microscopy (SMLM) in living cells because it could ameliorate both phototoxicity and photobleaching, which are two critical limitations of this technique.
These fluorophores exhibit a ground-state equilibrium between a fluorescent and a dark species, providing sparse distribution of fluorescent molecules without photoactivation, which greatly decreases phototoxicity and photobleaching (Fig. 1b).
Despite the great potential of spontaneously blinking dyes, their performance in terms of resolution achieved, image quality, and apparent photobleaching depends on the fraction of molecules that are fluorescent at equilibrium. This fraction is strongly determined by the pH and polarity of the medium. Although these dyes have been used successfully to image specific molecular targets for short periods of time 28,29 , long time-lapse imaging has only been realized in the low-polarity environment of membranes ( Fig. 1b) 27 . In contrast, we argue that a combination of photoactivation and fluxionality would provide a way to control the fraction of fluorescent molecules independently from the properties of the medium. Starting from a dark, non-fluxional isomer, photoactivation would convert a fraction of the total molecules to a fluxional form (Fig. 1c). In this population of fluxional molecules,  Here we report the design, synthesis, validation, and application of fluorescent molecules that become fluxional upon photoactivation. These processes are characterized by single-molecule imaging, demonstrating that the compound indeed becomes fluxional upon photoisomerization. Employing this probe, we are able to perform very long (>30 min) live-cell time-lapse SMLM, in 2D and 3D, with minimal toxicity and no apparent photobleaching. Finally, we apply this fluorophore to study the dynamics and three-dimensional compartmentalization of synaptic vesicle trafficking in live human neurons.

Results
Design, synthesis, and mechanistic studies. To create photoregulated fluxional fluorophore PFF-1 (Fig. 2a), we grafted an acylhydrazone photoswitchable unit 30,31 onto a rhodamine B scaffold. Probe PFF-1 was synthesized in two steps and high yields from rhodamine B (Fig. 2b). As a mechanistic control compound, we also synthesized probe 2, which could be obtained from intermediate 3 (Fig. 2b). We hypothesized that prior to photoactivation, PFF-1 should exist predominantly as the E isomer of the acylhydrazone and a dark, non-fluxional, spirocyclized derivative ( Fig. 2a). This structure was confirmed by X-ray crystallography (Supplementary Fig.   1) and electronic absorption spectroscopy ( Supplementary Fig. 2). Upon photoisomerization with light of 410 nm, the fluorescence of a solution of PFF-1 in aqueous buffer (pH = 7) increases by 12-fold, whereas photoirradiation of compound 2 gives only a 1.6-fold increase under the same conditions (Fig. 2c). Furthermore, photoactivation of PFF-1 at pH = 5 led to an even larger increase in fluorescence of 22-fold, but for compound 2 the increase was only 3-fold. 1 Both the spirocyclic and open rhodamine forms absorb light at 254 nm with essentially the same extinction coefficient ( Supplementary Fig. 2). In contrast, the open rhodamine form absorbs strongly at 560 nm, whereas the spirocyclic form does not absorb in this region at all. We reasoned that we could use the HPL chromatograms measured at these wavelengths to determine the fraction of molecules in both the E and Z isomers that are present as the open rhodamine form.
Evaluation of these chromatograms revealed that the Z isomer has a much higher absorbance at 560 nm compared to the E isomer. This result indicates that formation of the fluorescent rhodamine form is preferred in the Z but not E isomer. Based on these absorbance measurements ( Supplementary Fig. 8), we determined that the percentage of molecules that is in the fluorescent state is approximately 39% for the Z isomer, but only 0.004% for the E isomer at pH = 5 ( Supplementary Fig. 8).
Because the population of the Z isomer before irradiation is only 0.18% ( Supplementary Fig. 8), the total percentage of fluorescent molecules is ~0.07%.
After photoactivation (100 s, ~8.5 mW cm -2 ), the population of the Z isomer increases to 0.64%, which corresponds to ~0.25% of fluorescent molecules. These experiments confirm that even after photoactivation, the total fraction of fluorescent molecules is very small, which is a desirable feature for SMLM. Unfortunately, it was not possible to separate the two isomers at pH = 7.4 ( Supplementary Fig. 5), but the overall fraction of fluorescent molecules could be estimated before and after photoactivation as 0.003% and 0.012%, respectively.
These experiments support the mechanism depicted in Fig. 2a, in which formation of the Z isomer of the acylhydrazone in PFF-1 brings the protonated 2-pyridyl substituent in close proximity to the carbonyl group, facilitating proton transfer-induced ring opening. This mechanism is further supported by density functional theory (DFT) modeling (Fig. 2d). DFT modeling also suggests that the barrier of interconversion between the open fluorescent rhodamine and the dark spirocyclic form is energetically low (6.6 kcal mol -1 ) for the Z isomer (Fig. 2d). To demonstrate that the Z isomer interconverts rapidly between the dark spirocyclic and fluorescent rhodamine forms (i. e. it is fluxional), we embedded PFF-1 in a polyvinylalcohol (PVA) film on a coverslip, irradiated this film with a photoactivation pulse (405 nm, 2.6 W cm -2 , 20 ms), and imaged single molecules with a 561 nm (0.25 kW cm -2 ) laser in total-internal reflection mode. This method has been used before to evaluate the blinking properties of photoactivatable small molecules 32 . We validated the method by employing the non-fluxional compounds 2 and rhodamine B as negative controls and a spontaneously blinking molecule (HMSiR) as a positive control ( Fig. 3 and Supplementary Fig. 9). This experiment revealed that single molecules of PFF-1 transitioned several times between their dark and emissive forms, confirming the fluxionality of the probe (Fig. 3a,b). The switching behavior of PFF-1 was robust, providing numerous switching cycles (221±17, N = 30) and a large number of emitted photons per switching cycle (640±93, N = 30). These parameters are comparable to those of commonly used SMLM dyes 32 , but unlike these previously reported compounds, PFF-1 does not require continuous photoactivation to switch.
Moreover, single molecules of PFF-1 could be localized with an average precision of 22±9 nm in films, demonstrating that the fluxionality of PFF-1 could be used for single-molecule localization. Time-lapse SMLM in living cells for more than 30 minutes. The toxicity of PFF-1 was assessed in HeLa cells ( Supplementary Fig. 10). No loss of viability was detected at concentrations up to 100 µM (limit of solubility of PFF-1 in growth medium) after 48 h. Spinning-disk confocal microscopy was used to prove that PFF-1 diffused freely across the plasma membrane of mammalian cells and could be photoactivated with light of 405 nm ( Supplementary Fig. 11). PFF-1 displayed strong fluorescence after photoactivation in the acidic medium of lysosomes ( Supplementary Fig. 12). Under identical photoactivation conditions, the fraction of fluorescent molecules of PFF-1 is larger at low pH than in neutral solution because more molecules of PFF-1 are protonated, which enables fluxionality. The total fraction of fluorescent molecules can be modulated by photoactivation and, unlike the spontaneously blinking dye HMSiR 28 , PFF-1 can be employed at either neutral or low pH, simply by tuning the photoactivation step.
We optimized the imaging conditions to record long time-lapse SMLM experiments in live HeLa cells using PFF-1. Prior to imaging, cells were incubated with 500 nM PFF-1 for 10 min and then subjected to a photoactivation pulse (405 nm, 2.6 W cm -2 , 10 ms). After this initial photoactivation step, cells were either imaged using only the fluxionality of PFF-1 (single photoactivation, Fig. 4a,c) or by repeating the photoactivation pulse every 10 min to replenish the population of fluxional molecules (sequential photoactivation, Fig. 4b,c). In either case, single molecules could be localized with an average precision of 33 nm over an acquisition time of 30 min. In the case of single photoactivation, however, substantial photobleaching occurred after 15 min (Fig. 4a,c). Sequential photoactivation (every 10 min), on the other hand, maintained the number of detected molecules essentially constant during the whole acquisition (Fig. 4b,c and Supplementary Movie 1). This experiment confirms that PFF-1 enables extremely long time-lapse SMLM at irradiation intensities that are much lower than those used in typical experiments (Supplementary Table 1). Furthermore, we evaluated the morphology of the cells, disruption of the plasma membrane, and activation of caspase-3 as signs of phototoxicity. Even after this very long time-lapse acquisition, the irradiated cells displayed only modest signs of phototoxicity ( Supplementary Fig. 13), highlighting the advantages of using a combination of very mild photoactivation and fluxionality for single-molecule localization.
Under these imaging conditions, we were able to distinguish individual vesicles within diffraction-limited areas ( Supplementary Fig. 14), confirming that cellular structures that would have been blurred by diffraction became discernible using our imaging approach. Moreover, the overall high labeling density and fast switching kinetics of PFF-1 allowed us to localize approximately 1,100 molecules per µm 2 within vesicles using only 23 camera frames. Using these signals, we were able to obtain reconstructed snapshots with a Nyquist-limited resolution of nearly 60 nm every 0.5 s. This excellent spatiotemporal resolution allowed us to track in detail the fast motion of single lysosomes. The example displayed in Fig. 4d depicts a vesicle displaying both fast directional motion and slow Brownian-like diffusion, in agreement with previous reports of lysosome dynamics 33 .
We also performed time-lapse 3D SMLM using astigmatism and adaptive optics 34 .
These images revealed the 3D distribution of lysosomes within a 1 µm-thick imaging volume in a living cell (Fig. 4e,f). The adaptive optics implementation allowed us to increase the localization precision to 11±5 nm and 140±60 nm in the lateral and axial direction, respectively. A single lysosome moving primarily along the axial direction could be tracked in this experiment (Fig. 4g), illustrating the level of spatiotemporal detail with which these organelles could be visualized employing probe PFF-1.
Besides being useful to image and track lysosomes, probe PFF-1 could also be used to label other cellular components at higher pH. We synthesized probe MitoPFF-1, which is a derivative of PFF-1 functionalized with a mitochondria-targeting triphenylphosphonium group 35 . Although mitochondria display neutral or slightly basic pH and therefore would decrease the fluorescent fraction of MitoPFF-1 in its fluxional equilibrium, we were able to image mitochondria with super-resolution using this probe simply by tuning the photoactivation conditions ( Supplementary Fig. 15).
Additionally, a taxol conjugate of PFF-1, termed TaxoPFF-1, was prepared to test These tracks, however, have connections to these hotspots (Fig. 5f). The diffusion of a vesicle between H3 and H4 confirmed that vesicles are able to jump out of the hotspot, join the track, move rapidly through it, and fall into the next hotspot (Fig. 5g).

Discussion
The combination of photoactivation and fluxionality in the same fluorophore allowed us to control its emission in a manner that alleviates photobleaching and phototoxicity in live-cell SMLM. This feature facilitates very long (>30 min) time-lapse imaging of single-molecules within living cells, with a temporal resolution of seconds, even in small and densely labeled sub-cellular compartments. To the best of our knowledge, our time-lapse SMLM experiments are the longest ever reported, and even after such long acquisition times, no apparent photobleaching was observed.
The performance of these dyes was not affected by the medium, allowing for superresolved imaging of both neutral and acidic intracellular organelles, and thereby providing a generalizable mechanism for the development of small-molecule dyes for long-term SMLM.