Polarization as a new dimension in super-resolution microscopy

The physical phenomenon of fluorescence has a number of fundamental dimensions, e.g., intensity, wavelength, time, and polarization. In particular, the fluorescence polarization effect—first discovered in 1926¹—arises from the transverse nature of light waves (i.e., from dipole orientations). Various fluorescence anistropy techniques have previously been developed to study the dipole orientation of fluorophores. For example, fluorescence polarization microscopy (FPM) is used extensively for biological imaging applications. In this technique, the angle of a fluorophore is measured so that the orientation and structural details of a targeted protein can be resolved. Conventional FPM methods, however, are limited because of the presence of many molecules within the diffraction-limited volume. This means that the fluorescence polarization information is collected from dipoles with many different orientations.

The idea of using super-resolution to improve imaging resolving power was first proposed in 1995.² This idea has since been realized, i.e., with a photobleaching-photoactivation process used to separate molecules (with a resolution of about 20nm) in the time domain.³ Previously developed super-resolution microscopy approaches, which extend vision beyond the diffraction limit, are mostly based on the intensity, wavelength, and temporal dimensions of fluorescence. Although the fourth dimension of fluorescence, i.e., polarization, can also be used to modulate fluorescence (without restriction to specific fluorophores), this mode of super-resolution microscopy has only recently been investigated. Indeed, a new technique—sparse deconvolution of polarization-modulated fluorescent images (SPoD)—was first developed in 2014 (with which a resolution of 5nm was demonstrated at 1 frame/second).⁴ Although super-resolution can be achieved with this technique, the dipole orientation information is lost during the SPoD reconstruction and an interesting debate—whether or not fluorescent polarization can in fact be used to yield additional super-resolution information—has since ensued.^{5, 6}

Instead of looking at this problem from a conventional fluorescence-intensity point of view, we take both the fluorescence intensity and the fluorescence anisotropy (i.e., polarization) into account. In our work,⁷ we thus introduce a new technique—known as super-resolution dipole orientation mapping (SDOM)—in which we use the dipole angle superimposed on a super-resolution image. In this way, we have demonstrated that polarization can provide additional structural information to that of a super-resolution image and have resolved the ongoing controversy.^{7, 8}

The principle behind our SDOM algorithm is illustrated in Figure 1(a). In this example, two neighboring fluorophores (100nm apart and with different dipole orientations) are shown in red and green. If the polarization states of the two fluorophores are perpendicular to each other, only one can be excited under a specific polarization condition and the two points can thus be distinguished from each other. In other words, by rotating the polarization of excitation, the emission ratio between the two molecules is modulated and they are separated in the polarization domain. We use our sparsity-enhanced deconvolution to estimate the effective dipole intensity under polarization modulation. Furthermore, to fully exploit all the polarization modulation information, we apply a least-squares estimation to extract the dipole orientation. We can either superimpose the dipole orientation (polarization information) on top of a super-resolution image or it can be used in a 3D coordinate system, as illustrated in Figure 1(b) and (c) for a dipole orientation difference of 90°, respectively. In the SDOM orientation mapping image (resolution of about 130nm) the two molecules always cross each other and cannot be resolved. In contrast, when the dipole orientation information is used to create the 3D image, they can be completely separated. Our SDOM technique can therefore be used both to resolve and to provide information on the arrangement of fluorescent molecules.

Figure 1. Illustration of the super-resolution dipole orientation mapping (SDOM) technique. (a) Two fluorophores, 100nm apart, with different dipole orientations are shown in red and green. When excited by rotating polarized light they emit periodic signals. The emission ratio between the two molecules can be modulated accordingly and used to separate them in the polarization domain. The SDOM procedure provides a super-resolution image of the effective dipole intensities (compared with an unresolved wide-field image). Arrows indicate the different dipole orientations. Scale bar denotes 200nm. (b) SDOM result for two fluorophores, superimposed on top of a super-resolution image, where the two molecules cannot be separated. (c) The same data shown in a 3D coordinate system (XY is the plane of the super-resolved intensity image and θis the dipole orientation). In this perspective, the two molecules can be completely resolved.⁷

In our study we have compared equivalent results from both the SPoD and our own SDOM methods. As shown in Figure 2, no information is lost in our method (because of over-deconvolution) compared with the results of the SPoD approach. Moreover, our method provides the substantial advantage of achieving super-resolution imaging of fluorophore orientations. In particular, our imaging reveals that different borders of a dendritic spine neck in hippocampal neurons exhibit a heterogeneous distribution of dipole orientations.

Figure 2. Comparison of results from (a) conventional sparse deconvolution of polarization-modulated fluorescent images (SPoD) and (b) SDOM for the same neuron data set.⁴Close-ups of the dipole orientation mapping from the boxes are shown in (c) and (d), where the SPoD and SDOM images are shown in the top and bottom panels, respectively. Arrows illustrate the dipole orientations.⁷

The very fast imaging speed (up to 5 frames/second) of our SDOM technology, coupled with its very low (milliwatt level) excitation-light power requirements, means that our approach is ideal for observations of living cells. We have thus used SDOM to image fluorophore-labeled septin (a protein) in living yeast (Saccharomyces cerevisiae) cells. A cross-sectional and a top-view image of the samples are shown in Figure 3, as both wide-field and reconstructed SDOM images. We observe the typical double-ring structure of the septin in the SDOM image, but it cannot be resolved in the wide-field image.

Figure 3. Top-view (a) and cross-sectional view (b) of a fluorophore-labeled protein in live yeast cells. A wide-field (top left) and SDOM (bottom right) image for both views is shown. Inset in (a) shows the fluoresence along the marked (dotted line) cross section. Scale bars denote 500nm.⁷

In summary, we have developed a new super-resolution imaging technique known as SDOM in which we use the polarization modulation of the excitation laser and demodulation of intensity and polarization to achieve improved spatial resolution (i.e., beyond the diffraction limit) and accurate dipole orientation detection. We have demonstrated that SDOM provides favorable results compared with the conventional SPoD technique, and we have used it to reveal several interesting observations of biological samples. In addition, the very fast imaging speeds and low power requirements associated with our technique mean that it is particularly suited to imaging living cells. Although SDOM can be used to successfully obtain super-resolution information with orientation mapping, it still has some limitations. As part of our future work, we will therefore establish additional algorithms to solve orientation and orientation-distribution issues at the super-resolution level. Despite recent disputes concerning the role of polarization in super-resolution imaging, our work is part of a continuing surge of interest in this topic.^9–11

This work is part of a collaboration with the groups of Dayong Jin (University of Technology Sydney, Australia), and Jun-Tao Gao (Tsinghua University, China). To facilitate further development of this work, the developed code has been deposited on GitHub.