Optimized two-color super resolution imaging of Drp1 during mitochondrial fission with a slow-switching Dronpa variant

When we initially attempted to examine suborganelle structures with two-color PALM, we encountered the photoactivation/emission of dense populations of Dronpa molecules per frame (50 ms) even under low or zero activation laser power (Movie S1). Unlike other red PA-FPs, such as the Eos-FP family and PAmCherry1, in which photoactivation is tightly controlled by 405-nm light (21), this spurious “basal” or “spontaneous” photoactivation rate of Dronpa prevents the identification of single-molecule events and causes a high background fluorescence level even in the absence of 405-nm laser, particularly in densely labeled samples. This effect fundamentally limits the discrimination required for single-molecule identification and localization achievable with Dronpa. When we attempted to inactivate Dronpa molecules with strong illumination by 488-nm laser power before imaging, as recommended in the published method (3), the molecules could no longer be reactivated or excited and were presumed to be photobleached. Although these problems likely have prevented Dronpa from wider use in super resolution microscopy, systematic studies on these issues have been lacking. We hypothesized that the excitation light (488 nm) was also capable of photoactivation and the source of the practical problems with Dronpa (Fig. 1A). To validate this idea, we examined in vitro the photo-physics of individual Dronpa molecules in Tris buffer by a total internal reflection microscope (SI Materials and Methods) (22). Similar to previous work, the emission trace of a single Dronpa molecule, under 488-nm excitation only, features several bursts separated by a few seconds (Fig. 1B) (4, 6). However, in contrast to the previously published observations, we found that the interval between bursts becomes significantly shortened with increasing 488-nm powers. For the complete characterization of Dronpa at the single-molecule level, three quantities were measured and statistically analyzed for various 488-nm and 405-nm power: the number of times single molecules photo-switch (N_Blink), the time spent in the fluorescent state (T_ON), and the time spent in the dark state (T_OFF). N_Blink and T_ON are related to the kinetics of leaving the fluorescent state “F,” whereas T_OFF is related to the kinetics of entering the fluorescent state (Fig. 1A).

N_Blink and T_ON are decreased by the increase of 488-nm excitation laser power, although not affected by 405-nm activation laser power (Fig. S1 A and B). Thus, the Fluorescent (F) → Dark (D) and Fluorescent (F) → Bleached (B) kinetics depend only on 488-nm light (Fig. 1A). In contrast, T_OFF is altered by 405 nm and 488 nm laser power (Fig. S1C). As revealed by the cumulative probability distribution function (CDF) of T_OFF (Fig. 1C), T_OFF increases as 405-nm laser power decreases, confirming that the photoactivation kinetics (D → F) is proportional to the 405-nm activation laser power. Noticeably, even at zero 405-nm laser power, 50% of the Dronpa molecules are photoactivated within 3.4 s (Fig. 1C). This basal photoactivation rate increases as the 488-nm excitation laser power increases, indicating that the photoactivation kinetics (D → F) is also proportional to the 488-nm activation laser power (Fig. 1D). At high 488-nm laser power the basal photoactivation rate dominates, such that the 405-nm laser power contribution becomes almost negligible (Fig. 1D). Therefore, a more realistic model of Dronpa includes 488-nm, as well as 405-nm, light in the photoactivation kinetics (Fig. 1A). To mitigate the issue, we developed a Dronpa variant with an approximately twofold lower basal 488 nm-induced photoactivation.

The V157G mutation of rsFastLime significantly decreases the T_OFF by fourfold in bulk measurements compared with Dronpa, likely through the reduction of steric hindrance to the chromophore’s cis/trans isomerizations (8, 23). When we attempted to use rsFastLime at a 20-Hz frame rate and simultaneous illumination with 405-nm and 488-nm light, fluorescent events occurred with such rapidity that no individual events could be distinguished per frame. Therefore, we hypothesized that increasing the steric hindrance at V157 would inhibit the cis/trans isomerization of the Dronpa chromophore, slowing the 488 nm-induced basal photoactivation and improving temporal separation between single molecules (Fig. 2A). We chose to replace Dronpa V157 with the nonpolar amino acid leucine or phenylalanine. Both mutants and wild type were expressed in Escherichia coli and tested for fluorescence and photo-switchability. DronpaV157L maintained fluorescence and UV induced photo-switching. DronpaV157F lacked visible fluorescence and was discarded for the remainder of the studies.

We determined the bulk spectroscopic properties of purified in vitro DronpaV157L and Dronpa. The excitation and emission maxima of DronpaV157L did not differ significantly from Dronpa; however, DronpaV157L has ∼25% reduced brightness (Fig. S2 A and B). Comparison of the UV-induced photo-switching properties of Dronpa and DronpaV157L were determined by an in vivo photo-switching assay of the fluorescent proteins expressed in E. coli (Fig. S2 C and D). DronpaV157L showed a slightly slower photo-decay time (∼0.7 s) than Dronpa (∼0.5 s), but the fluorescence growth rate was not affected (Fig. S2 E and F). In addition, DronpaV157L photobleached more slowly (2.5 vs. 1.7 min) than Dronpa (Fig. S2 C and D). We renamed DronpaV157L “rsKame.” Dronpa was named for “dron,” a Japanese ninja word for vanishing, and PA for photoactivation (24). rsKame is a combination of reversibly switchable (rs) and “kame,” the Japanese word for turtle. Although the slower photo-switching of rsKame could be inferred by the ensemble measurement, we chose to more accurately understand its kinetics at the single-molecule level.

To better understand the effects of increased steric hindrance proximal to the Dronpa chromophore, we examined rsKame as single molecules using the same photo-physical single-molecule characterization as was described for the characterization of Dronpa (SI Materials and Methods) (22).

For moderate 488-nm laser power (0.5 W/mm²), the N_Blink of rsKame (2.4) was slightly less than that of Dronpa (3.1), whereas T_ON of rsKame (44 vs. 42 ms) was slightly longer (Fig. S3 A and B). As expected, these are consistent with slowed 488-nm deactivation [F (cis) to D (trans)] kinetics by the increased steric hindrance posed by the leucine side chain of rsKame at position 157. However, the effect is minor and becomes almost negligible when the photobleaching becomes dominant at strong 488-nm laser power (5.9 W/mm²) (Fig. S3 A and B).

The T_OFF of rsKame in the presence of low 405 nm and 488 nm is ∼2 times as long as Dronpa (Fig. S3 C and D), demonstrating that the increased steric hindrance affects the photoactivation [trans (D) to cis (F)] kinetics more than the deactivation kinetics. However, the difference in T_OFF between rsKame and Dronpa is reduced and eventually becomes negligible as 405-nm laser power increases. Data sets from a third laser power (0.2 W/mm²), which demonstrated similar photo-physical trends, further clarified the effect of 488-nm light on the photo-switching properties of rsKame and Dronpa (Fig. S3).

For the super resolution imaging application, the suppression of basal level 488 nm-induced activation at zero 405-nm power is especially important. The CDF of T_OFF shows that the basal activation of rsKame is reduced by twofold compared with Dronpa (Fig. 2B). Fully analyzed super resolution images of the OMM labeled with rsKame in EpH4 cells or the IMM in HeLa cells, with only 488-nm illumination (Movie S1), show clearly delineated membranes with increased resolution compared with those labeled with Dronpa (Fig. 2 C and D and Fig. S3 E and F). Moreover, the number of rsKame photoactivation events in the presence of 488 nm alone decays 1.6 times slower than for Dronpa, further demonstrating a slower basal photoactivation rate in vivo for rsKame (Fig. 2E). Because fewer molecules are photoactivated at any given time, single-molecule events can be better separated in time for rsKame. In addition, a reduction in the background fluorescence from molecules out of focus improves the contrast of molecules in focus and thus their localization accuracy, although rsKame and Dronpa have similar photon budgets (Fig. 2F, Fig. S3 G and H, and Movie S1). Therefore, the increased steric hindrance of leucine at position 157 in rsKame mitigates the 488 nm-induced photoactivation of the molecule and renders it more suitable for PALM imaging than Dronpa.

We also compared rsKame with another green PS-FP, mGeos-M, that has similar photo-physical properties (25). However, mGeos-M exhibits more significant basal photoactivation by 488 nm than rsKame (Fig. S4 and Movie S2).

We chose to pair rsKame with PAmCherry1, a dark to red PA-FP. Because 405-nm light activates both rsKame and PAmCherry1, to develop our sequential two-color PALM imaging protocol, we measured the photoactivation rate of PAmCherry1 for varying 405-nm laser powers, using the previously described single-molecule photo-physical characterization method. The photoactivation rate of PAmCherry1 was more than 1,000-fold slower than rsKame for any given 405-nm laser power (Fig. S5A). We estimated that over a 5-min experiment (the average time for imaging rsKame), less than 2.5% of PAmCherry1 molecules would be activated. Therefore, sequential two-color PALM by imaging first rsKame and then PAmCherry1 is possible (Fig. S5B; Materials and Methods).

To demonstrate the efficacy of our two-color PALM method, we chose to visualize two major substructures of the mitochondria, the repeatedly and deeply invaginated IMM and the smooth OMM. Nuclear encoded mitochondrial proteins are directed to mitochondria by localization sequences (MLS), often found at the N or C terminus (26). These labels and additional transmembrane or localization sequences direct mitochondrial proteins to specific compartments and/or membranes (26). We chose two proteins whose MLS and specific membrane integration had been characterized by truncations fused to reporter proteins.

To coat the OMM with PAmCherry1, we fused the C-terminal MLS and transmembrane region of murine B-cell lymphoma-extra large (BclXl) to the C terminus of PAmCherry1 with a 25-amino-acid linker (PAmCherry1-Lk-BclXl^201-233). BclXl is a protein involved in apoptosis and anchored to the OMM with the main body of the protein in the cytoplasm (27, 28). To coat the IMM, we fused the N-terminal MLS and transmembrane region of human ubiquinol-cytochrome c reductase synthesis-like (BCS1L) to the N terminus of rsKame (BCS1L^1-160-Lk-rsKame). BCS1L is a mitochondrial translocase protein and anchored to the IMM with the main body of the protein in the matrix (29, 30).

EpH4 cells, cotransfected with PAmCherry1-Lk-BclXl^201-233 and BCS1L^1-160-Lk-rsKame, were imaged by our two-color PALM method (Materials and Methods). Two-color PALM images allowed us to elucidate cells containing predominantly elongated mitochondrial networks (Fig. 3 A and B) from cells containing predominantly punctate mitochondria (Fig. 3 G and H). From the clear OMM boundary in PALM images, we could measure the width of elongated mitochondria (200-250 nm) and punctate round mitochondria (300-350 nm) (Fig. 3 C and I). In contrast, the width of IMM was shorter than OMM, appearing completely enclosed by the OMM (Fig. 3 D, E, J and K). The distinction is further demonstrated in intensity profiles made by drawing a line transversely across an individual mitochondrion and measuring PALM localization intensity along the line (Fig. 3 F and L). In elongated IMM, we could observe structures composed of a series of transverse and somewhat linearly oriented fluorescent molecules, presumably outlining the highly folded structure of the cristae (Figs. 3D and 4A and Fig. S3I).

Two-color PALM images of the IMM and OMM revealed diverse mitochondrial morphologies (Fig. 4). We observed “donut”-shaped mitochondria, a recently described morphology with pathophysiological significance under stress conditions (Fig. 4 A and B) (31, 32). In previous studies the typical diameter was ∼1.3 μm by conventional fluorescence microscopy. However, we measured donuts as small as ∼500 nm (Fig. 4 A and B). Previously observed donut mitochondria with small diameters were likely misclassified as a separate “blob” morphology because of diffraction limited resolution. Some observed IMMs only partially filled the OMM compartments (Fig. 4 C and D). Such heterogeneous IMM structures may represent mitochondria undergoing IMM remodeling (33, 34).

We applied our two-color PALM method to examine the structural characteristics of mammalian Drp1 oligomerization during mitochondrial fission in situ. Drp1 helical rings were imaged by labeling the OMM with PAmCherry1-Lk-BclXl^201-233 and fusing Drp1 to rsKame. Previous studies with Drp1 subunits fused N-terminally to fluorescent proteins incorporated into fission rings during assembly showed normal fission progression (15, 18, 35, 36). We fused rsKame to the N terminus of human Drp1 separated by a 14-amino-acid linker (rsKame-Lk-Drp1). HeLa cells were cotransfected with PAmCherry1-Lk-BclXl^201-233 and rsKame-Lk-Drp1 and imaged by PALM as previously described (Materials and Methods). We determined the ratio of rsKame-Lk-Drp1 to endogenous Drp1 to be 0.62 to 0.38, which, assuming that a low percentage of rsKame will misfold, indicates at least half of the Drp1 molecules in the helical rings are labeled (Fig. S6A).

The high cytoplasmic population of rsKame-Lk-Drp1 could induce errors in identifying colocalization between Drp1 rings and mitochondria. We reduced this potential error by using a custom built program to analyze Drp1 clusters, filtering out clusters of n < 8 events (Fig. S7). The majority of the large Drp1 clusters were observed to colocalize with the OMM (Fig. 5A and Fig. S7). Distinct OMM morphologies can be used as a reference to determine fission sites; the diameter of the OMM at the site decreases as fission progresses (Fig. S8). Therefore, on the basis of the observed OMM morphology and the incorporation of Drp1 clusters, we identified four different fission states. The fission state in which an elongated or two distinct Drp1 clusters are found on unconstricted or slightly constricted mitochondria was termed Initial (Fig. 5B and Fig. S8 A–D). This is presumed to correspond to the 2D section of a Drp1 ring in its relaxed form before the GTP-hydrolysis-driven conformational change. Constricted is used to describe the fission state in which the Drp1 cluster is localized to a clearly constricted outer membrane (Fig. 5C and Fig. S8 E–H). Presumably this state corresponds to a Drp1 ring that is partially constricted and undergoing GTP hydrolysis. We termed the fission state in which membranes appear to be very tightly constricted beneath Drp1 clusters Scissioned (Fig. 5D and Fig. S8 I–L). The fission state Terminal is used to describe Drp1 clusters observed at one end of mitochondria (Fig. 5E and Fig. S8 M–P). This is likely a partially degraded ring, which is found predominantly at only one of the mitochondrial termini resulting from fission (Fig. 6D). Because it was difficult to determine the precise degree of membrane constriction between the Constricted and Scissioned states of the OMM, we grouped these two states together for analysis under Intermediate (Fig. 5 C and D). We also noted numerous individual Drp1 clusters found at shallow invaginations in the OMM (Fig. 6 A–C). These invaginations and clusters were usually observed only on one side of the mitochondria and may reflect nascent fission sites or early binding of Drp1 to local clusters of Mff (16, 37).

To measure the diameter and the length of individual Drp1 rings, we first drew a rectangular box around each ring (Fig. 5 B and F). The box is reduced to the minimum rectangle that includes all of the Drp1 molecules. The widths and the lengths of these minimal rectangles (Fig. 5 G and H) overestimate the dimensions of the Drp1 ring by twice the mean localization uncertainty of rsKame (22 nm; Fig. S5C) in the PALM images (Fig. 5F). After this correction, the Drp1 ring diameters were as follows: Initial = 139 ± 60 nm, Intermediate = 105 ± 48 nm, and Terminal = 77 ± 39 nm (Fig. 5G). The corresponding ring lengths were as follows: Initial = 68 ± 40 nm, Intermediate = 68 ± 51 nm, and Terminal = 48 ± 38 nm (Fig. 5H). Significantly, although the Drp1 ring diameter decreases by ∼60 nm, accompanying its transition between the Initial and Terminal fission states, no significant change in ring length was observed. We also quantified the diameters of the OMM within the Initial and Intermediate fissions states, which correlated with the Drp1 diameters within error and showed a distinct decrease from the Initial to Intermediate state (Fig. S8 Q and R).

To determine the sensitivity of our analysis method to the number of incorporated labeled Drp1 molecules, we performed simulations on Dnm1 helical rings, based on cryo-EM structural data (20), varying the amounts of fluorescently labeled monomers. The resulting values were consistent at coverage greater than 0.3 (Fig. S6 B and C), significantly less than the experimental coverage (≥0.5) (Fig. S6A). Compared with our measured Drp1 width (W) (Fig. 5G), our values were within the error of the Dnm1 simulations, for both non- and constricted rings; whereas the lengths (L) were much shorter than predicted for a two-turn helix (Fig. S6 B and C).

To first image rsKame, the sample was coilluminated with 488-nm excitation/photoactivation (5.9 W/mm²) and 405-nm activation at 0.6 mW/mm². Fluorescent data were collected at 20 Hz with the 525-nm emission filter until all molecules were photobleached. To image PAmCherry1, the sample was coilluminated with the 561-nm excitation (22.0 W/mm²) and 405-nm photoactivation begun at 0.1 W/mm² and slowly increased to 1.4 W/mm² over time. Fluorescence was collected until all molecules were photobleached. Both PALM datasets were analyzed separately and their super resolution images were overlaid using our custom MATLAB-based analysis software package (22).

rsKame-Lk-Drp1 was placed downstream of a TetOn promoter and cotransfected with PAmCherry1-Lk-BclXl^201-233 into the T-Rex HeLa cells (Invitrogen TetOn system). Expression of rsKame-Lk-Drp1 was induced by doxycycline. Twelve hours before fixation and imaging, cells were fed 40 nm Au fiducial markers coated in FBS (60,000 particles/mL). Four hours before imaging, doxycycline (Invitrogen) and z-vad-fmk (Promega) were added to the total growth media to final concentrations of 0.75 μg/mL and 20 µM, respectively. HeLa cells were then incubated for 110 min at 37 °C. Staurosporine (Cell Signaling Technologies) was added to a final concentration of 1 μM, and cells were incubated for another 100 min at 37 °C. Just before fixation, mitochondria were labeled with MitoTracker Deep Red (Life Technologies/Invitrogen) at a final concentration of 30 nM. Treated cells were fixed in 1% formalin (Sigma-Aldrich) in 1× PHEM (60 mM Pipes, 25 mM Hepes, 10 mM EGTA, and 2.0 mM MgCl₂, pH 7.0) for 10 min at room temperature and imaged in 1× PHEM.