Characterization and development of photoactivatable fluorescent proteins for single-molecule–based superresolution imaging

Significance

Photoactivatable fluorescent proteins (PAFPs) are important probes for superresolution fluorescence microscopy, which allows the spatial organization of proteins in living cells to be probed with sub–diffraction-limit resolution. Here, we compare four properties of PAFPs that are critical for superresolution imaging and report two new PAFPs that exhibit excellent performance in all four properties.

Abstract

Photoactivatable fluorescent proteins (PAFPs) have been widely used for superresolution imaging based on the switching and localization of single molecules. Several properties of PAFPs strongly influence the quality of the superresolution images. These properties include (i) the number of photons emitted per switching cycle, which affects the localization precision of individual molecules; (ii) the ratio of the on- and off-switching rate constants, which limits the achievable localization density; (iii) the dimerization tendency, which could cause undesired aggregation of target proteins; and (iv) the signaling efficiency, which determines the fraction of target–PAFP fusion proteins that is detectable in a cell. Here, we evaluated these properties for 12 commonly used PAFPs fused to both bacterial target proteins, H-NS, HU, and Tar, and mammalian target proteins, Zyxin and Vimentin. Notably, none of the existing PAFPs provided optimal performance in all four criteria, particularly in the signaling efficiency and dimerization tendency. The PAFPs with low dimerization tendencies exhibited low signaling efficiencies, whereas mMaple showed the highest signaling efficiency but also a high dimerization tendency. To address this limitation, we engineered two new PAFPs based on mMaple, which we termed mMaple2 and mMaple3. These proteins exhibited substantially reduced or undetectable dimerization tendencies compared with mMaple but maintained the high signaling efficiency of mMaple. In the meantime, these proteins provided photon numbers and on–off switching rate ratios that are comparable to the best achieved values among PAFPs.

Photoactivated localization microscopy, stochastic optical reconstruction microscopy, and related imaging methods take advantage of photoswitching and imaging of single molecules to circumvent the diffraction limit of spatial resolution in light microscopy (1–3). In these methods, only a subset of the fluorescent labels in the sample is switched on at any given time such that the positions of individual fluorophores can be localized from their images with high precision. Iteration of this process allows numerous fluorescent labels to be localized and an image with sub–diffraction-limit resolution to be reconstructed from the fluorophore localizations. Fluorescent proteins that can be activated from dark to fluorescent or converted from one color to another are widely used for such imaging approaches (4, 5). Although photoactivatable fluorescent proteins (PAFPs) are generally dimmer than photoswitchable dyes (6, 7) and hence give lower image resolution, the ease and high specificity of labeling protein targets in living cells with fluorescent proteins makes PAFPs highly appealing probes for imaging the dynamics of cellular structures (4, 8).

For single-molecule–based superresolution imaging methods, several properties of PAFPs are particularly important for the image quality. Here, we focus on four such key properties. (i) The first property is the photon budget, defined as the average number of photons emitted in each switching event. Given that the error of localizing an individual fluorophore approximately scales with the inverse square root of the number of detected photons, a higher photon budget leads to higher localization precision and hence higher image resolution (7, 9). (ii) The second property is the on–off switching rate ratio (on–off ratio), defined as the ratio between the on-switching (activation) and off-switching or photobleaching rates under the illumination of the imaging light only (7). Even in the absence of activation light, the imaging light itself can also switch on the PAFPs, albeit at a low rate. Thus, the ratio between the on-switching and off-switching rates under this condition determines the lower bound of the fraction of PAFP molecules in the on-state at any given time. The presence of activation light would increase this fraction. When the product of this fraction and the density of fluorescent labels reaches approximately one fluorophore per diffraction-limited volume, it becomes difficult to resolve and precisely localize the activated fluorophores. Hence the on–off ratio limits the density of fluorescent labels that can be localized, which in turn affects the effective image resolution based on the Nyquist sampling theorem (10). (iii) The third property is the dimerization tendency. Many PAFPs have a weak tendency to form dimers; this could even be true for the PAFPs that are reported as being monomeric. When these proteins are fused to target proteins that also tend to polymerize, they may cause undesired aggregation of the target proteins and distort the native distribution of the protein of interest. (iv) The fourth property is the signaling efficiency, defined as the ratio between the number of detectable PAFP-fusion molecules per cell and the expression level of the fusion protein. Fluorescent proteins do not necessarily fold with 100% efficiency. Among the folded molecules, not all of them will become mature at the time of imaging. Among the matured PAFP molecules, only a subset can be photoactivated and imaged. Because of these deficiencies, the number of fusion molecules detected could be substantially lower than the expression level of the fusion protein. PAFPs with higher signaling efficiencies will lead to higher localization densities for a given target protein, which will in turn increase the effective image resolution.

In this work, we measured the above properties of 12 commonly used PAFPs, including PAGFP (11), Dendra2 (12, 13), mEos2 (14), mEos3.2 (15), tdEos (16), mKikGR (17), PAmCherry (18), PAtagRFP (19), mMaple (20), PSCFP2 (13, 21), Dronpa (22), and mGeosM (23). From this screen, we found that none of these PAFPs was simultaneously optimal in all four criteria described above. For example, PAtagRFP and mEos3.2 exhibited the highest photon budgets among PAFPs, excellent on–off ratios, and undetectable dimerization tendencies, but showed poor signaling efficiencies. Alternatively, mMaple provided excellent signaling efficiency and on–off ratio with a photon budget nearly equal to those of PAtagRFP and mEos3.2, but had a substantial dimerization tendency. To address this limitation, we developed two new PAFPs based on mMaple that exhibited substantially reduced or undetectable dimerization tendencies while maintaining the high signaling efficiency, high photon budget, and low on–off ratio of mMaple. These PAFPs will substantially facilitate superresolution imaging of cellular structures.

Results

Photon Budget of Photoactivatable Fluorescent Proteins.

To evaluate the properties of PAFPs under conditions similar to typical superresolution imaging experiments, we fused each PAFP to various target proteins and expressed these fusion proteins in either mammalian cells or bacteria. To measure the photon budget, we fused the PAFPs to the mammalian focal adhesion protein Zyxin, transiently transfected BS-C-1 cells with the fusion constructs, and imaged the cells using the superresolution imaging mode in which individual activated proteins were imaged. The distributions of the photon numbers detected per activation event were determined for all 12 PAFPs. Four example distributions, for mEos3.2, mMaple, PSCFP2, and PAGFP, are shown in Fig. 1. The mean photon numbers determined from these distributions are listed in Table 1.

Fig. 1.

Photon number measurements of PAFPs. The histograms are example photon number distributions of mEos3.2 (A), mMaple (B), PSCFP2 (C), and PAGFP (D) measured by imaging individual Zyxin-PAFP fusion proteins in live BS-C-1 cells. The mean photon numbers are indicated.

Table 1.

Properties of PAFPs

PAFP	Preactivation/postactivation emission wavelength, nm*	Photon no.	On–off switching rate ratio	ClpP clustering†	No. of localizations per cell‡	Maturation time, min§
Dendra2	507/573	686	4.2 × 10−6	−	1,810	38
mEos2	519/584	745	2.9 × 10−6	+	1,290	340
mEos3.2	516/580	809	2.6 × 10−6	−	1,950	330
tdEos	516/581	774	3.2 × 10−6	−	1,800	330
mKikGR	515/591	599	4.1 × 10−6	+	3,800	31
PAmCherry	—/595	706	7.8 × 10−6	+	4,200	61
PAtagRFP	—/595	906	5.7 × 10−6	−	760	200
mMaple	505/583	798	1.9 × 10−6	+	24,000	48
mMaple2	506/582	783	1.0 × 10−6	+	21,000	62
mMaple3	506/583	675	6.2 × 10−7	−	12,300	49
PAGFP	—/517	313	1.3 × 10−3	−		<10
PSCFP2	468/511	223	8.1 × 10−6	+
Dronpa	—/517	262	5.8 × 10−4	−		25
mGeosM	—/514	248	4.9 × 10−4	+		<10

A 405-nm laser was used for photoactivation. The photon number and on–off switching rate ratio were measured in live BS-C-1 cells. The ClpP clustering, number of localizations per cell, and maturation time were measured in live E. coli cells.

^*The mMaple2 and mMaple3 emission wavelengths were measured in this work with purified proteins. The other wavelengths are cited from refs. 4 and 16.

^†The “+” indicates that ClpP-PAFP exhibits clustered distributions in at least a subset of cells, whereas “−” indicates that ClpP-PAFP does not exhibit clustered distributions in any cells. The results on Dendra2, Dronpa, and mEos2 are consistent with a previous report (26).

^‡The number of HU-PAFP localizations per E. coli cell.

^§Maturation time is defined as the half-life of the immature state.

Fewer photons were detected from the green PAFPs (PAGFP, PSCFP2, Dronpa, and mGeosM) than from the red ones (Dendra2, mEos2, mEos3.2, tdEos, mKikGR, PAmCherry, PAtagRFP, and mMaple). However, within the same color group, the difference in photon budget was less than twofold. In addition to the above-listed fluorophores, we also imaged rsFastLime (24) and rsEGFP (25). Both of these proteins gave relatively low photon budget (<60 photons per switching event), which would lead to relatively poor localization precision. We thus did not further characterize these proteins. It is, however, worth noting that these proteins are excellent choices for a different mode of superresolution imaging [reversible saturable optical fluorescence transitions (RESOLFT)] due to the large number of switching cycles that they exhibit before photobleaching (24, 25).

On–Off Ratio of Photoactivatable Fluorescent Proteins.

To determine the on–off ratio, we measured the rates for switching on and switching off (or photobleaching) the PAFPs in the presence of imaging light only. By definition, the on-switching rate is the increment in probability of the on-switching events per unit time. To measure this quantity, we imaged the Zyxin-PAFP–expressing cells in the superresolution mode for a short period without any activation light (with imaging light only). The samples were then imaged to completion with an additional activation light at 405 nm. The ratio of the total number of activation events accumulated by a certain time during the period without activation light over the total number of activation events by the end of the imaging process was determined. The slope of this cumulative activation probability against time then gave the on-switching rate (Fig. 2 A and B, and Table S1). The off-switching rate was determined from the inverse of the mean lifetime of the on-state of each PAFP in the presence of the imaging light (Fig. 2 C and D, and Table S1). The ratios between the on- and off-switching rates were determined for all 12 PAFPs (Table 1). Because both on- and off-switching rates scale linearly with the illumination intensity, the on–off ratio should be independent of the imaging light intensity. The on–off ratios were generally very small (10⁻⁵ to 10⁻⁶) except for PAGFP, Dronpa, and mGeosM, which were ∼10⁻³.

Fig. 2.

On–off switching rate ratio measurements of PAFPs. Sample data are presented for PSCPF2 (A, C, and E) and PAGFP (B, D, and F). (A and B) Cumulative on-switching probability as a function of time without activation light. The slope of the line gives the on-switching rate. (C and D) Distribution of the on-state lifetime. The blue bars represent measured data, which were fitted with a binned exponential function (magenta dots) $P_n\mathrm{=}{\displaystyle {\int }_{n\mathrm{-}1}^n\left(1/m\right)e^{\mathrm{-}\frac{x}{m}}dx}$, to take into account that lifetimes are rounded to the next larger integer. m is the mean lifetime, the inverse of which gives the off-switching rate. The on–off switching rate ratio is defined as the on-rate divided by the off-rate. Each frame corresponds to 16 ms. (E and F) Representative superresolution images of Zyxin-PSCFP2 and Zyxin-PAGFP in live BS-C-1 cells.

From the superresolution images of Zyxin, it is evident that the PAFP with a smaller on–off ratio gave images with a higher localization density and hence higher image quality (Fig. 2 E and F). This correlation can be understood based on the Nyquist sampling theorem, which suggests that the final resolution of an image is at least twice the average distance between the localized probes (10).

Dimerization Tendency of Photoactivatable Fluorescent Proteins.

Dimerization or oligomerization of fluorescent proteins may cause undesired aggregation of the target proteins. Here, instead of measuring the dimerization affinity of PAFPs in vitro, we directly probed whether the PAFPs could cause aggregation of a target protein using a previously reported method (26). In this approach, the fluorescent proteins are fused to the Escherichia coli protease ClpP, which itself oligomerizes to form a tetradecameric complex. It has been suggested that ClpP proteins tend to aggregate and form a single visible punctum in E. coli when fused to a fluorescent protein with a substantial dimerization tendency, whereas fusion to a fluorescent protein with a weak or no dimerization tendency tends to display a diffuse localization pattern (26). We thus fused E. coli codon-optimized PAFP sequences to the chromosomal copy of clpP, and report whether cells exhibit the single-punctum phenotype (Table 1 and Fig. 3A).

Fig. 3.

Dimerization tendency of PAFPs and its effect on the distributions of target proteins. (A) Phase contrast (Left) and conventional fluorescent images (Right) of live E. coli cells expressing ClpP-PAFP fusions. PAFPs with substantial dimerization tendencies (mEos2, mMaple) result in the formation of ClpP puncta. PAFPs with little to no dimerization tendencies (mEos3.2, PAtagRFP) produce a diffusive ClpP distribution. (B) Overlaid superresolution (magenta) and phase-contrast (gray) images of live E. coli cells expressing H-NS-PAFP fusions. (C) Overlaid superresolution and phase-contrast images of fixed E. coli cells expressing Tar-PAFP fusions. (D) Superresolution images of fixed Cos-7 cells expressing Vimentin-PAFP fusions or fixed untransfected Cos-7 cells with immunofluorescent labeling of Vimentin (IM). (Scale bars: 1,000 nm.)

Fusion with mKikGR, mGeosM, mMaple, PAmCherry, PSCFP2, or mEos2 all led ClpP to form a single punctum in at least a subset of cells, suggesting that these PAFPs exhibit appreciable dimerization tendency, whereas Dendra2, mEos3.2, tdEos, PAtagRFP, PAGFP, and Dronpa did not show high enough dimerization tendency to drive detectable ClpP aggregation (Table 1 and Fig. 3A). As a cautionary note, because different target proteins have different expression levels and intrinsic oligomerization or polymerization tendencies, PAFPs that cause appreciable aggregation of ClpP may not necessarily cause aggregation of other target proteins, and vice versa.

To further illustrate the potential effect of PAFP dimerization on target proteins, we imaged E. coli nucleoid-associated protein H-NS, E. coli chemotactic receptor Tar, and mammalian intermediate filament protein Vimentin fused to different PAFPs. In an earlier paper (27), we have reported that H-NS appears as a few large and discrete clusters in cells when fused to mEos2, a previously reported monomeric version of the Eos fluorescent protein (14, 16). We also observed similar results when H-NS was fused to PAmCherry (27), another previously reported monomeric PAFP. Fusion with these PAFPs did not appear to perturb the functional activity of H-NS (27). Given the residual dimerization tendency of mEos2 and PAmCherry detected by the ClpP assay (Table 1 and Fig. 3A), we extended the study of H-NS using other PAFPs here (Fig. 3B). Notably, H-NS generally appeared as large and discrete clusters in cells when fused to PAFPs that exhibited appreciable dimerization tendency in the ClpP assay, such as mEos2, PAmCherry, and mMaple. H-NS also appeared as large clusters when fused to tdEos, even though tdEos did not show sufficient dimerization to drive ClpP aggregation. However, fusion of H-NS to Dendra2, mEos3.2, and PAtagRFP, which did not exhibit detectable dimerization tendency by the ClpP assay, appeared much more dispersed in the nucleoids. We also observed the more dispersed phenotype with immunofluorescence imaging of H-NS, but did not present this comparison here because fixation of the bacterial nucleoid is in general discouraged. It is known that H-NS oligomerizes/clusters even without fusion to fluorescent proteins (28, 29), and this intrinsic H-NS clustering effect is important for its in vivo function as a transcriptional silencer (29) and brings gene loci in different chromosome regions into spatial proximity in wild-type cells (27). However, fusion with mEos2, PAmCherry, mMaple, and tdEos may have exaggerated the H-NS clustering effect. Without prior knowledge of the residual dimerization tendency of these “monomeric” PAFPs, we previously interpreted the large clusters of H-NS-mEos2 and H-NS-PAmCherry earlier as a property of H-NS (27). Results here suggest that this interpretation is likely inaccurate because the clustering of H-NS is likely enhanced by the underappreciated dimerization effect of mEos2 and PAmCherry. Although it is formally possible that the more monomeric PAFPs could disrupt intrinsic H-NS clustering by discouraging PAFP self-interactions, we consider this a less likely possibility.

As another example, Tar preferentially clustered at cell poles when fused to mEos2 and mKikGR (Fig. 3C) (30), but spread out more evenly along the envelope of E. coli when fused to the more monomeric mEos3.2 (Fig. 3C), suggesting that the preferential cell pole distribution might have been exaggerated by the dimerization of mEos2 and mKikGR. Such aggregation effect could also apply to eukaryotic systems. For example, fusion to mEos2 and mKikGR caused Vimentin filaments to cluster into thick bundles in mammalian cells, whereas Vimentin-mEos3.2 appeared as thin filaments that were similar to immunofluorescence images of Vimentin in untransfected cells (Fig. 3D).

Signaling Efficiency of Photoactivatable Fluorescent Proteins.

Even when the PAFP is fused to the target protein at its endogenous chromosomal locus, the number of detectable fluorescent proteins does not necessarily reflect the expression level of the fusion protein. This deficiency not only prevents quantitative analysis of the target protein, but may also reduce the resolution of the image by effectively decreasing the labeling density. Here, we compared the relative signaling efficiency of different PAFPs by determining the number of single-molecule localizations per cell of the PAFPs fused to a common target protein under endogenous expression. We fused the PAFPs to the E. coli gene hupA, which encodes a subunit of the nucleoid-associated protein HU, at its endogenous chromosomal locus and determined the total number of HU-PAFP localizations per cell (Fig. 4 and Table 1). The four green PAFPs were too dim to be detected as single molecules in E. coli, whereas the eight red PAFPs showed similar photon budget to those measured in mammalian cells (data not shown). Among the red PAFPs, mMaple provided by far the largest number of HU localizations, which was 6- to 32-fold higher than those of other HU-PAFP fusion proteins (Fig. 4).

Fig. 4.

Signaling efficiency comparison among PAFPs. (A) Superresolution images of live E. coli cells expressing HU-PAFP fusions (magenta) overlaid with phase-contrast images (gray). Superresolution images were acquired until all of the PAFP molecules in the field of view were bleached. (B) Numbers of observed localizations per cell for different HU-PAFP fusions. (Error bars represent SEMs.)

The difference in the number of HU-PAFP localizations for different PAFPs may be attributed to multiple reasons. One of the possible reasons could be the difference in the number of blinking (switching) events per fluorophore. We measured the average number of blinking events in fixed HU-PAFP–expressing cells (Table S2). The average number of blinking events spanned only a small range from 1.7 to 3.3, which are similar to the results derived from purified PAFP in vitro (Fig. S1). The localization numbers divided by the average numbers of blinking events yielded the numbers of imaged molecules per cell (Table S2), which is still much bigger in the case of HU-mMaple–expressing cells (4- to 36-fold higher than in other HU-PAFP–expressing cells).

A second possible reason for the difference in localization numbers could be the difference in HU-PAFP expression levels. However, quantitative Western blot experiments showed that most HU-PAFP constructs are expressed at similar levels (Table S2), except for tdEos. We divided the number of imaged molecules per cell by the number of expressed fusion proteins per cell to calculate the percentage of PAFP imaged (Table S2). The percentage for mMaple imaged was still 5- to 22-fold higher than the other PAFPs. Therefore, neither the number of blinking events nor the expression level is a major contributing factor to the difference in the observed localization numbers.

A third potential reason is the difference in the fluorescent protein maturation time. Because E. coli has a relatively short doubling time under our growth conditions, a substantial fraction of the PAFP molecules may not have sufficient time to mature and become fluorescent. To test this idea, we measured the in vivo maturation time of several PAFPs by using kanamycin to block protein synthesis in E. coli cells expressing HU-PAFP, and then recording the increase in cellular fluorescence due to PAFP maturation (Fig. S2 and Table 1). Indeed, compared with some of the PAFPs with low signaling efficiencies, mMaple has a substantially faster maturation time. However, the difference in maturation time was substantially smaller than and hence not sufficient to account for the observed difference in the localization numbers (SI Note and Fig. S3).

Other potential contributing factors include the fraction of properly folded PAFPs at equilibrium, the fraction of folded PAFPs that can ultimately mature, the fraction of folded and mature PAFPs that can be photoactivated, and the fraction of activated PAFPs that are sufficiently bright to image. Given that different red PAFPs tend to have similar photoactivation efficiencies (31) and photon budgets (Table 1), the last two factors are less likely explanations for the observed range of signaling efficiencies. More experiments are required to test these possibilities.

Finally, we measured the signaling efficiency of the PAFPs in mammalian cells by transient transfection of A549 cells with the zyxin::PAFP constructs. For fast data collection, conventional fluorescent microscopy was used and cells were imaged until the PAFPs were fully bleached. The number of imaged molecules in each cell was estimated by dividing the total photon count from the cell by the product of the average photon number per switching event (Table 1) and the average number of blinking events for each PAFP (Table S3). The results indicate that mMaple again offers much higher signaling efficiency (7- to 70-fold higher) than the other red PAFPs, even in mammalian cells. Among the green PAFPs, PSCFP2 offers a comparable signaling efficiency to that of mMaple (Table S3).

New PAFPs with High Signaling Efficiency and Low Dimerization Tendency.

The above results show that mMaple has a much higher signaling efficiency than all of the other tested red PAFPs. However, the dimerization tendency of mMaple could lead to aggregation effects on the target proteins. Dendra2, mEos3.2, and PAtagRFP exhibit undetectable dimerization tendency, but have low signaling efficiency. It is thus desirable to develop a new PAFP that has both high signaling efficiency and low dimerization tendency. To this end, we engineered two new PAFPs by introducing point mutations into mMaple designed to destabilize the dimerization of this protein.

We first took inspiration from the two mutations that make mEos3.2 more monomeric than mEos2: I102N and Y189A (15). Based on a sequence alignment between mEos2 and mMaple, we made the comparable mutations, I111N and Y198A, in mMaple. We then tested the properties of this protein, which we termed mMaple2, by fusing it to Zyxin, ClpP, and HupA and performing measurements on the photon budget, on–off ratio, dimerization tendency, and signaling efficiency as described above. mMaple2 exhibited a similar photon budget, on–off ratio, and signaling efficiency as those of mMaple (Fig. 5 A–C, Table 1, and Tables S2 and S3). ClpP-mMaple2 proteins still formed puncta in some cells. However, the percentage of cells showing single punctum was significantly reduced in comparison with ClpP-mMaple, suggesting a lower dimerization tendency of mMaple2 (Fig. 5D).

Fig. 5.

mMaple2 and mMaple3 exhibit both high signaling efficiency and low dimerization tendency. (A–D) Photon budget (A), on–off switching rate ratio (B), signaling efficiency (C), and dimerization tendency (D) of mMaple2 and mMaple3 in comparison with mMaple. (Insets in D) Sample fluorescent images of ClpP-mMaple2– and ClpP-mMaple3–expressing E. coli. Error bars are SEMs and are too small to be visualized in A. (E) H-NS appears more spread out in E. coli cells when fused to mMaple3 in comparison with the mMaple and mMaple2 fusion proteins. (F) Tar-mMaple2 and Tar-mMaple3 appear more evenly distributed along the cell envelope and less concentrated at the polar caps than Tar-mMaple. (G) Vimentin-mMaple2 and Vimentin-mMaple3 are less bundled than Vimentin-mMaple. (Scale bars: 1,000 nm.)

To further reduce the residual dimerization tendency of mMaple2, we screened a series of mutations to residues that are predicted to be solvent-exposed near residues 111 and 198. We focused on charged residues and switched them to the opposite charge. One of the derivatives, with mutations E82R, D83K, and D197K, in addition to I111N and Y198A, exhibited undetectable dimerization tendency when fused to ClpP (Fig. 5D and Table 1). This derivative, which we termed mMaple3, again had a similar photon budget to that of mMaple and even a lower on–off ratio compared with mMaple (Fig. 5 A and B, and Table 1). Its signaling efficiency was only moderately reduced from those of mMaple or mMaple2 (Fig. 5C, Table 1, and Tables S2 and S3).

We further tested the aggregation effects of mMaple2 and mMaple3 on H-NS, Tar, and Vimentin. H-NS-mMaple3 showed substantially more spread out H-NS distributions than H-NS-mMaple, although H-NS-mMaple2 still appeared as a few discrete clusters in each cell (Fig. 5E). Tar-mMaple2 and Tar-mMaple3 were substantially less concentrated at the cell poles than Tar-mMaple (Fig. 5F). Vimentin-mMaple2 and Vimentin-mMaple3 filaments were much less bundled than Vimentin-mMaple filaments (Fig. 5G). All of these observations are consistent with the reduced dimerization tendency of mMaple2 and the undetectable dimerization tendency of mMaple3.

The excitation and emission spectra of mMaple2 and mMaple3 were similar to those of mMaple (Table 1 and Fig. S4).

Discussion

In this work, we characterized four properties of PAFPs that are important for superresolution imaging based on single-molecule switching and localization. First, the photon budgets are not substantially different for different PAFPs within the same color group, but the red PAFPs provide substantially more photons (600–900 photons per activation event) than the green PAFPs (200–300 photons). Even the red PAFPs’ photon budgets are substantially lower than those of the bright photoswitchable/photoactivatable dyes (several thousand to one million photons) (7, 32). Given that the localization precision scales approximately linearly with the inverse square root of the photon numbers, the PAFPs thus give substantially lower localization precision than photoswitchable dyes. Second, the on–off ratios are also not substantially different for different PAFPs within the same color group, but the red PAFPs tend to show much lower on–off ratio (10⁻⁵ to 10⁻⁶) than the green PAFPs (∼10⁻³). PSCFP2 is a noticeable exception with an on–off ratio similar to the red PAFPs. The on–off ratios of the red PAFPs are substantially lower than those of the popularly used photoswitchable dyes (10⁻³ to 10⁻⁴) (7), and hence PAFPs can provide substantially higher localization density. Third, natural fluorescent proteins tend to dimerize or tetramerize. Although mutant fluorescent proteins have been made to reduce dimerization, many of the so-called monomeric fluorescent proteins still have some residual dimerization tendency and thus can cause undesired aggregation and mislocalization of the target proteins to which they are fused. Among the 12 PAFPs tested here, the majority appeared to exhibit a substantial propensity toward aggregation or bundling of target proteins. The exceptions are Dendra2, mEos3.2, PAtagRFP, PAGFP, and Dronpa, which did not show a detectable aggregation effect for the target proteins tested here. It is important to note that these results do not imply that such aggregation will not happen when these PAFPs are fused to other target proteins with strong intrinsic clustering or polymerization tendency. As another cautionary note, it is also possible that monomeric fluorescent proteins that discourage self-interactions may disrupt natural oligomerization of the target proteins. Thus, it is a good practice to verify the spatial organization derived from PAFPs with alternative approaches, such as immunohistochemistry or by using SNAP/CLIP/HALO or short peptide tags to label proteins with dyes (33). The fourth property that we probed is the signaling efficiency, which determines the number of detectable molecules of the PAFP fusion protein for a given target protein. When the signaling efficiency is low, the number of localizations is not sufficient to map out the fine organization of the target protein. Notably, among the eight red PAFPs tested, mMaple has the highest signaling efficiency, which is about one order of magnitude higher than the other PAFPs. The downside of mMaple is, however, its relatively high dimerization tendency.

To overcome this problem, we developed two new PAFPs, mMaple2 and mMaple3, which largely maintained the superb signaling efficiency of mMaple but exhibited substantially reduced dimerization tendency. In particular, mMaple3 exhibited no detectable dimerization tendency. The photon numbers and on–off ratios of both are similar to those of mMaple, which are among the best for PAFPs. Based on the above results, we recommend researchers to label target proteins of interest with mMaple3 when performing single-molecule–based superresolution imaging using fluorescent proteins. mMaple3 provides excellent performance in all four key properties described here, including high signaling efficiency, low dimerization tendency, high photon budget, and low on–off ratio. In the cases that the localization number provided by mMaple3 is suboptimal, one should also consider labeling with mMaple2, which will give substantially higher localization numbers, but it is important to check whether fusion with mMaple2 has led to undesired aggregation effect. The large number of localizations provided by mMaple2 and mMaple3 will not only facilitate mapping out the fine spatial distribution of the target protein, but will also allow the localizations to be divided into more snapshots to facilitate time-lapsed imaging of the dynamics of cellular structures. When localization number is less of a concern, mEos3.2 and PAtagRFP are excellent choices for their high photon budget and undetectable dimerization tendency.

The four properties measured in this work are key properties of PAFPs to consider when imaging the spatial organization of a target protein with single-molecule–based superresolution imaging. However, they are not the only important properties to consider when other experimental requirements need to be taken into account. For example, spectral properties are essential for multicolor imaging. For two-color imaging, green PAFPs are particularly useful (when paired with a red PAFP) even though they generally give lower photon numbers and higher on–off ratios. Dark-to-fluorescent PAFPs are also more favorable than color-changing PAFPs as the former take a smaller spectral space. The number of switching cycles is another important parameter to consider. For quantifying the stoichiometry of the target proteins, it is desirable to have PAFPs that can be switched on only once, although almost all PAFPs tend to blink more than once (31). However, for mapping of protein spatial organization, a larger number of switching cycles per molecule would allow the target structure to be sampled more times, which is advantageous when it comes to time-lapsed imaging for probing dynamics. For a different superresolution imaging method, RESOLFT, reversibly switching fluorescent proteins with a large number of switching cycles (such as rsEGFP) is actually required (25).

Methods

Plasmids were constructed with PCR and isothermal assembly (34). E. coli chromosomal insertions were created by lambda RED recombination (35). A list of the plasmids and strains is provided in Table S4. E. coli cells were grown to exponential phase in M9 minimal media before imaging. Mammalian cell lines were transfected with purified plasmids using nucleofection or lipofection, and incubated for 24–26 h before imaging. Phase-contrast, superresolution, or conventional fluorescence images were collected on an Olympus IX-71 inverted microscope with 405-, 488-, and 561-nm laser lines. Images were analyzed with custom-written software. See SI Text for details.