High-Precision Mapping of Membrane Proteins on Synaptic Vesicles using Spectrally Encoded Super-Resolution Imaging

Abstract

The spatial resolution of single-molecule localization microscopy is limited by the photon number of a single switching event because of the difficulty of correlating switching events dispersed in time. Here we overcome this limitation by developing a new class of photoswitching semiconducting polymer dots (Pdots) with structured and highly dispersed single-particle spectra. We imaged the Pdots at the first and the second vibronic emission peaks and used the ratio of peak intensities as a spectral coding. By correlating switching events using the spectral coding and performing 4–9 frame binning, we achieved a 2–3 fold experimental resolution improvement versus conventional superresolution imaging. We applied this method to count and map SV2 and proton ATPase proteins on synaptic vesicles (SVs). The results reveal that these proteins are trafficked and organized with high precision, showing unprecedented level of detail about the composition and structure of SVs.

Graphical Abstract

We used the intensity ratio of the first and second vibronic peaks of semiconducting polymer dots (Pdots) to correlate photoswitching events. Multi-fold localization precision enhancement was achieved by averaging over switching events from the same Pdot. This method was used to count and map two membrane proteins on synaptic vesicles.

Synaptic vesicles (SVs) play a central role in neurotransmission. The functions of SVs, which include uptake, storage, and stimulus-dependent release of neurotransmitters via endocytosis and exocytosis, are mediated by membrane proteins on their surface.^[1] Therefore, it is important to develop a detailed and quantitative understanding of the membrane proteins on SVs. However, SVs (~40-nm diameter) are well below the resolution limit of conventional fluorescence microscopy. Without using complex instrumentation or significantly slowing the acquisition rate, conventional single-molecule localization methods such as stochastic optical reconstruction microscopy (STORM) and photoactivated localization microscopy (PALM) also have difficulty mapping out the spatial distribution of proteins on single SVs.^[2] In PALM, fluorescent probes are sequentially photoactivated, localized, and photobleached. In STORM, the fluorescent probes are stochastically switched “on” and “off”. While a probe can be switched “on” many times, it is difficult to correlate switching events dispersed in time. Therefore, the spatial resolution of these methods is limited by the photon number of a single localization event and cannot be further enhanced by changing imaging settings or post-acquisition processing.^[3]

Here we overcome this limitation by developing a new class of ultra-bright photoswitching semiconducting polymer dots (Pdots) which exhibit structured and highly dispersed single-particle emission spectra. We imaged the Pdots at the first and second vibronic emission peaks and used the ratio of peak intensities as spectral coding to correlate switching events. The high brightness of the Pdots offered high encoding capacity where dozens of spectra could be resolved, providing an additional way to separate Pdots that could not be differentiated by direct localization. This method allows averaging over many switching events from the same probe, which improves the theoretical localization precision $\sqrt{N}$ fold (N is the number of frames averaged), making it uniquely well-suited for characterization of membrane proteins on vesicles such as SVs. In this work, we demonstrated 4–9 frame binning and achieved a 2- to 3-fold experimental resolution improvement versus conventional superresolution probes.^[2a] We applied this method to determine the copy numbers of synaptic vesicle protein 2 (SV2, 5.4±0.8 per SV) and proton ATPase (1.1±0.4 per SV). High precision mapping revealed that SV2 proteins are located closer to each other than to proton ATPases, suggesting spatial organization of SV2s on SVs. The results provide an unprecedented level of detail about the composition and structure of SVs. The methods developed here can also be applied to other biological vesicles such as extracellular vesicles, which could provide insight into their biological function and advance related diagnostic applications.

Pdots have been widely used in biological imaging applications due to their exceptional brightness and photostability.^[4] Here we developed a new class of ultra-bright photoswitching Pdots that exhibit structured and highly dispersed single-particle emission spectra, which make them well suited for spectrally-encoded superresolution microscopy. For the backbones of the light-emitting polymers in the Pdots, we selected two semiconducting polymers that exhibit emission spectra with pronounced vibronic progression, poly(9,9-dioctylfluorene) (PFO) and poly(9,9-dioctylfluorene-co-bithiophene) (PF2T). The fullerene derivative phenyl-C61-butyric acid (PCBA) was grafted onto the side chains of the light-emitting polymer backbones as electron acceptor units to facilitate photo-induced formation of hole polarons (Figure 1a). Hole polarons are efficient fluorescence quenchers in semiconducting polymer systems,^[5] and stochastic fluctuations of the photogenerated hole population in Pdots leads to spontaneous fluorescence switching (see Supporting Information for discussion of the photoswitching mechanism, Figure S1).^[6] It is noteworthy that the semiconducting polymers above cannot be synthesized under typical Suzuki polymerization conditions. Instead, we adopted a unique microwave-assisted polymerization scheme, which involves using tetrahydrofuran as the solvent and 150 W microwave to accelerate the reaction. The organic base tetraethylammonium hydroxide (NEt₄OH) was added to avoid CO₂ gas generation during polymerization (Figure S2, Figure S3). After optimization, both polymers were synthesized with high yields (87% for PFO and 84% for PF2T) and high molecular weights (3.6×10⁴ Da and 4.2×10⁴ Da). We tested various PCBA grafting ratios and found that 2% PCBA offered optimal photoswitching behavior for our application. Previous studies have reported that PFO and PF2T exhibit differently shaped emission spectra in different host materials, depending on molecular packing and inter/intra-chain interactions.^[7] We synthesized an inert matrix polymer (DDA-PIAM, Figure 1a) and incorporated it into the Pdots via coprecipitation. We found that doping with DDA-PIAM can result in a variety of emission spectra at the single-particle level. DDA-PIAM can also provide handles for bio-conjugation. However, due the long side chains, it has higher tendency to be trapped deep inside the Pdots, with fewer units on the surface. As a result, it effectively perturbed the semiconducting polymer chain-packing inside the Pdots, but it was not very efficient in providing surface functional groups. Greatly increasing DDA-PIAM doping ratio solved the problem of bio-conjugation, but it also reduced the brightness of the Pdots. It was previously reported that blending with poly(styrene-co-maleic anhydride) (PSMA) can effectively produce surface functional groups on Pdots for bio-conjugation.^[8] Considering these facts, we tested various DDA-PIAM and PSMA doping ratios and chose Pdots doped with 20% DDA-PIAM and 10% PSMA, as this combination delivered the optimal performance, including distinct single-particle spectral variation, facile bio-conjugation, and high single-particle brightness. Two Pdots were synthesized: PFO-PCBA/20%DDA-PIAM/10%PSMA (PFO-PCBA Pdots) and PF2T-PCBA/20%DDA-PIAM/10%PSMA (PF2T-PCBA Pdots). The Pdots were ~14 nm in diameter based on dynamic light scattering (Figure 1b). Figure 1c shows their bulk emission spectra, and indicates the first and second windows used for imaging (colored bands).

Figure 1.

a) Chemical structures of PFO-PCBA, PF2T-PCBA, DDA-PIAM, and PSMA. b) Number-weighted particle size distributions of PFO-PCBA Pdots (blue, 14.1±6.8 nm diameter) and PF2T-PCBA Pdots (green, 14.6±5.6 nm). c) Emission spectra of PFO-PCBA (blue) and PF2T-PCBA (green) Pdots. The colored bands indicate the first and second windows used for imaging. d) Schematic of instrumentation for spectrally-encoded superresolution imaging. e) Schematic of the data analysis process.

Spectrally-encoded superresolution imaging was performed using a custom-built wide-field imaging system based on an inverted microscope (Figure 1d). The Pdots were excited at 405 nm (typical excitation power density, ~2 kW/cm²) and were imaged at a frame rate of 50 Hz. Emission from the first and second vibronic peaks was split and focused onto different sensor areas on a sCMOS camera. A cylindrical lens was added to the detection optics to introduce astigmatism so that the axial position could be obtained from the aspect ratio of a point spread function (PSF) (Figure S4).^[9] Localization analysis was performed separately with images acquired at the first and second wavelengths, which involved fitting the PSF to an elliptical Gaussian function (Figure 1e). The localized positions in the two images were correlated and corrected using a homography matrix generated by imaging multi-color beads. The precise location of a Pdot was obtained by intensity-weighted averaging of localized positions in the two images. In addition to locations, the intensity ratio (R) at the first and second wavelengths was recorded as spectral coding to identify different Pdots. After localization, a superresolution image was generated based on X, Y, and Z values for preliminary evaluation of the data. Then, the localized points were regrouped based on their R values. Since the density of the localization clusters was significantly reduced after spectra separation, the centers of individual localization clusters can be determined. For each cluster, binning was performed with localization points within an uncertainty radius. Finally, all the localization points after binning were combined together to reconstruct the resolution-enhanced image. (Figure 1e; see Supporting Information for detailed procedures, Figure S5).

PFO-PCBA and PF2T-PCBA Pdots both exhibit emission spectra with clear vibronic structure. In single-particle imaging studies, R varied from particle to particle (Figure 2a, b). PFO and PF2T are sensitive to their environment and exhibit differently shaped emission spectra in different host materials.^[7] Previous research has shown that when intra-chain coupling dominates, the first vibronic transition is strongly allowed, resulting in a large R; when inter-chain coupling dominates, the first transition has low oscillator strength, resulting in a lower R.^[10] In our Pdots, it is likely that the relatively low doping level of DDA-PIAM matrix polymer (20%) induced variation of chain-chain coupling between particles, resulting in a variety of single-particle emission spectra. The variation in R can be used as spectral coding to correlate blinking events. By performing single-particle imaging simultaneously in two channels, two blinking Pdots located within a diffraction-limited spot were differentiated simply by intensity analysis (Figure 2c). This is particularly useful when the distances between Pdots are smaller than the localization uncertainties. Two adjacent Pdots that were not directly resolved by single-particle localization are shown in Figure 2d. By analyzing the intensity ratios of the two channels and performing 4-point binning of frames that share similar R, we recovered the precise locations of the Pdots. Analysis of the localization histograms indicated that the experimental localization precision increased 1.98 fold after the spectrally-encoded 4-point binning, consistent with the theoretical expectation that N point binning should result in $\sqrt{N}$ fold improvement in the localization precision.

Figure 2.

a, b) Representative Single-particle emission spectra of PFO-PCBA (a) and PF2T-PCBA (b) Pdots. c) Fluorescence intensity trajectories of two adjacent PF2T-PCBA Pdots, acquired from the first (red) and second (blue) imaging channels. Different colored bands indicate blinking events from different Pdots. d) Histograms of localized positions show that the two Pdots are differentiated by spectrally-encoded binning. e) Distributions of R values for the PFO-PCBA Pdots (blue) and PF2T-PCBA Pdots (green).

As a single-molecule technique, this method depends critically on photon budget. For the PFO-PCBA Pdots, the average numbers of photons detected per frame from the two images (N₁, N₂) were 3.2×10³ (N₁) and 1.0×10³ (N₂); for the PF2T-PCBA Pdots, the average numbers of photons were 2.0×10³ and 2.4×10³ (Figure S6). Based on these photon numbers and the data analysis method described above, the theoretical per frame localization uncertainties of PFO-PCBA and PF2T-PCBA Pdots were 4.2 and 4.0 nm, respectively (Supporting Information).^[11] The experimental localization error, however, was typically ~10 nm for both Pdots (see Supporting Information for resolution measurement, Figures S7 and S8). We also evaluated the encoding capacity of the Pdots, which is dictated by the measurement uncertainty as well as by the range of R values. Based on Poisson statistics and the error propagation rule, the relative R measurement uncertainty σ_R is given by $\sqrt{N_1+N_2}/\sqrt{N_1{N}_2}$. For the PFO-PCBA Pdots, R ranged from 1.0 to 5.1; for the PF2T-PCBA Pdots, R ranged from 0.4 to 1.5, with typical σ_R values of 0.036 and 0.030, respectively (Figure 2e). Using these values, we determined the encoding capacity C for PFO-PCBA and PF2T-PCBA Pdots to be 25.0 and 23.6, respectively, which means that theoretically over 20 Pdots can be resolved within a localization resolution-limited spot (see Supporting Information for determination details). It should be noted that the spectral de-mixing was performed after superresolution localization, which means that we only have to de-mix spectra of Pdots within a localization-resolution-limited area, not a large diffraction-limited-area (Figure S5). Considering the experimental localization standard deviation of ~10 nm and Pdot size of 14 nm, this encoding capacity is sufficient for our application. By analyzing the biological samples of interests, we found that there were typically 2–3 Pdots in a localization resolution-limited spot. Based on the number of Pdots and the probability distribution of R, the probability for spectrally-encoded binning to work for a particular localization resolution-limited spot was over 91% for both Pdots, which means that the resolution of over 91% of the imaging areas can be enhanced (Supporting Information). The remaining small percentage of the area was characterized by larger clusters of several unresolved Pdots; these clusters were excluded when performing quantitative analysis of biological structures. To ensure enough localization data points, we typically performed 4–9 frame binning (see Supporting Information for comparison of different binning results, Figure S10). The spatial resolution achieved was 2- to 3-fold better the resolution obtained from the conventional STORM probes, like Alexa Fluor-647.^[2a]

We performed spectrally-encoded superresolution imaging of two biological nanostructures: microtubules (MTs) and SVs. Prior to imaging, MTs were incubated with biotin-conjugated anti-α-tubulin antibodies, then with streptavidin-conjugated PFO-PCBA Pdots. Figure 3a shows a 3D superresolution image of MTs. Although the MT network is already resolved at sub-diffraction-limit resolution via direct localization (upper-right image in Figure 3a), spectrally-encoded binning further improves the resolution and reveals additional details (lower-left image). The resolution enhancement is evident by examining lateral slices of MTs (Figure 3b, c), in which the hollow MT structure is only partially resolved before binning and becomes clearly resolved after binning (Figure 3d).^[2a] For SV labeling, PFO-PCBA and PF2T-PCBA Pdots were conjugated to anti-SV2 and anti-proton ATPase antibodies, respectively. Molar ratios of Pdot to antibody were controlled to produce single antibody-conjugated Pdots. Initial localization revealed a single cluster of proton ATPase and a continues distribution of SV2 (Figure 3e), consistent with previous reports that SV2s are more abundant than proton ATPase on SVs.^[12] Since SVs are only ~40 nm in diameter, additional resolution enhancement is required to map the precise locations of SV2s. After spectral-encoded binning, the continues distribution of SV2 separates into several clusters in the 3D projection image (Figure 3f). By counting the number of clusters, we determined protein copy numbers of 5.4±0.8 per SV for SV2 and 1.1±0.4 per SV for proton ATPase (Figure S11, Supporting Information)—values that are in the same range as copy numbers determined previously by other techniques.^[12] While the Pdots are relatively large compared to dyes, they exhibit a comparable size with Immunoglobulin G antibody binding arm and typically reside at the outer layer of the labelled structure, which should have low spatial hindrance effect. The superresolution and protein counting results confirmed that Pdot-conjugated-antibodies exhibited similar labelling densities on MT and SVs, as compare to dye-conjugated-antibodies (Supporting Information, Figures S8 and S9). Based on 3D superresolution images, we calculated average interprotein distances and distance distributions (Figure 3g, Figure S11, Supporting Information). It should be noted that the measured protein distance is a convolution between the actual protein distance and the measurement uncertainty. The measurement uncertainty broadens the actual protein distance distribution. However, since the broadening is likely symmetric and is not protein-dependent in this case, the average protein distance can still be measured with high precision. The results showed that the SV2-SV2 distance (37.7 nm) was smaller than the SV2-Proton ATPase distance (46.5 nm), suggesting the different spatial organizations of the two proteins on SVs, which is likely related to the SVs function and requires future studies to elucidate in full detail.

Figure 3.

a) 3D superresolution image of a microtubule network labeled with PFO-PCBA Pdots. The upper-right and lower-left halves of the image are before and after spectrally-encoded binning, respectively. Scale bar, 1 μm. b, c) Lateral slice images of MTs before b) and after c) 9-point spectral-encoded binning. Scale bar, 100 nm. d) Intensity versus position plot of microtubule cross-sections from slices indicated in panels b and c, showing greatly enhanced resolution of the hollow microtubule structure following spectrally-encoded binning (green). e, f) 3D projection superresolution image of SVs before (e) and after (f) spectrally-encoded binning, showing SV2 (green) and proton ATPases (red). Scale bar, 10 nm. g) Interprotein distance distributions for SV2–SV2 (green) and SV2–proton ATPase (blue) on SVs. Wilcoxon rank sum test of the two distributions yielded P value of 3.7 × 10⁻²¹, indicating significant difference in the distribution medium.

Here we demonstrated spectrally-encoded superresolution microscopy based on a new class of photoswitching Pdots that exhibit structured and highly dispersed single particle emission spectra. Images were acquired simultaneously over two wavelength ranges that cover the first and second vibronic peaks of Pdot emission, and the ratio of intensities of the first and second vibronic emissions was used as spectral coding to correlate switch “on” events. The high brightness of the Pdots offered high encoding capacity where dozens of spectra can be resolved, providing an additional way to differentiate Pdots that are not separated by direct localization. This method allows averaging over many switch “on” events that originate from a single probe to further improve the resolution. In this work, we demonstrated 4–9 frame binning and achieved a 2- to 3-fold experimental resolution improvement versus conventional superresolution imaging. We applied this method to counting and mapping two membrane proteins on synaptic vesicles—SV2 and proton ATPase—providing an unprecedented level of detail about the composition and structure of synaptic vesicles. In the future, development of photoswitching, spectrally-encoded Pdots with different emission colors can enable multiplex superresolution imaging of nanoscale bio-structures, such as exosomes and synaptic vesicles. Simultaneous, high-precision localization of multiple bio-markers on vesicle surface can be used to study interactions of bio-molecules and identify sub-populations of vesicles, which can provide important insight into their biological functions and inform their use in disease diagnostic applications.