Abstract: In conclusion, we believe NL-SIM to be a powerful approach in the exciting new field of superresolution light microscopy. All superresolution techniques excel in certain aspects and fail in others—the best technique will be determined by the demands of the application. We believe NL-SIM to be the best for those applications that benefit from low light intensity and few exposures, at a resolution of approximately 40 nm in two dimensions with currently available photoswitchable fluorescent probes. In the future, NL-SIM should be a capable technique for addressing a wide range of fundamental biological questions. The major obstacle our technique faces for the realization of this goal is the limited set of photoswitchable fluorescent probes available. Although many have been developed, few are suitable for our purposes. Our method requires reversibly switchable probes, and the resolution of our technique is directly related to the number of switching cycles that can be obtained before the fluorophores photobleach. With the addition of toxic, antifading chemicals and long exposure times, we obtained an acceptable number of cycles from Dronpa—63 for 40- to 50-nm imaging in 2D. However, if a better, more light-stable probe were developed, we believe that NL-SIM could be implemented in three dimensions and to living biological samples, as has been done for linear SIM. We next verified that our technique was useful for biological imaging. To that end, we turned to structures within mammalian cells. The first structure we visualized was the nuclear pore—a complex structure in eukaryotic cells that is a gateway between the nucleus and the rest of the cell. We genetically fused Dronpa to Nup98 and POM121, proteins known to be associated with the nuclear pore, and expressed both fusion proteins in human HEK293 cells. Imaging the nuclei with NL-SIM, we observed strikingly different localization patterns between the two proteins, even though they looked nearly identical under conventional resolution. Nup98, a protein known to be on both the nucleoplasmic and cytoplasmic side of the pore, formed small, uniform punctae, with an average full width at half-maximum of approximately 55 nm. On the other hand, POM121, a protein located in the membrane, formed a variety of different structures, many of them resembling rings with a diameter of 40–70 nm. We next imaged cytoskeleton made of a protein called actin in mammalian cells. We labeled actin with Lifeact, a small peptide that was recently developed to be selective for actin. Again, we genetically fused Dronpa to the Lifeact peptide and expressed the fusion protein in mammalian CHO cells. The actin network is varied and complex; imaging it with our NL-SIM at a resolution of 60–70 nm provided the clearest picture of this complexity ( Fig. P1 ). Fig. P1. We used fluorescent photoswitching with structured-illumination light to increase the resolution of a wide-field microscope. The off state of the photoswitchable protein Dronpa was driven with a pattern of light (blue), so that only molecules in the minima of the pattern remained fluorescent (green). As the off state was saturated, the region of fluorescent molecules became smaller than diffraction (dashed green). This can be visualized in k-space where high-resolution information exists further away from the origin. Our method (green) resolves information beyond the conventional microscopy (pink) or even structured illumination using linear fluorescence (blue). To demonstrate our technique, we looked at the nuclear pore and the actin cytoskeleton. To test our system, we first decided to visualize an already known structure: Microtubules are a well-studied component of the eukaryotic (nonbacterial) cytoskeleton, a network of cellular filaments that, among other functions, provide mechanical support to the cell. We polymerized microtubules in vitro and attached Dronpa to them. Imaging the fluorescent microtubules with our NL-SIM microscope, we resolved them to approximately 40 nm, four times the resolution of a conventional microscope. We first built a custom total internal reflection fluorescence (TIRF)-SIM microscope similar to those described in earlier studies. The illumination pattern was produced by a diffraction grating that could be translated and rotated to produce all the pattern orientations we needed for resolution enhancement. We focused the +1 and -1 order diffracted beams from the grating at the edges of a 1.46-N.A. TIRF objective, so that the illumination pattern was generated in a small (100–200 nm) region above the coverslip. This setup allowed us to image large 2D fields of view. Low light levels from two lasers were used to switch Dronpa between its dark and fluorescent states. We collected the fluorescence by the same objective and imaged it onto a 1,024 × 1,024 CCD camera. Localization-based technologies exploit the precision to which light from a single molecule can be mathematically determined ( 1 ). The goal of these techniques is to localize each individual fluorescent molecule—or fluorophore—in the sample (typically a single cell or region therein) to a precision well below the diffraction limit. Obtaining the best resolution can require tens of thousands of raw frames, each containing only a few isolated fluorophores. Such high numbers of raw frames ultimately limit the speed at which data can be acquired and, by consequence, the temporal resolution that can be achieved. On the other hand, illumination-based methods treat the sample as a continuous fluorescent object, rather than as a set of single molecules, and use the combination of an illumination pattern and a nonlinear fluorescence response from the sample to obtain theoretically unlimited spatial resolution ( 2 ). A number of nonlinear fluorescence phenomena, including notable ones like stimulated emission ( 3 ) [e.g., stimulated-emission depletion (STED) microscopy] and saturation [e.g., saturated structured-illumination microscopy (SSIM)] ( 2 ), have been exploited for resolution enhancement purposes. Unfortunately, these phenomena typically require extremely high light intensities. Fluorophores are known to rapidly photobleach—or fade—under high light intensities, and importantly, biological samples can be damaged by this energy. Despite these drawbacks, STED microscopy has realized a resolution of 50–70 nm for fluorescent protein-labeled cells ( 4 ). On the other hand, the high light intensity needed for fluorophore saturation has prevented SSIM from demonstrating resolution enhancement for a biological sample. In general, however, structured-illumination microscopy (SIM), the method underlying SSIM, is well suited for biological imaging, because its wide-field or nonscanning nature permits parallel data acquisition over large fields of view. Indeed, SIM, using only linear fluorescence, has permitted rapid live-cell imaging of whole cells ( 5 ), on a variety of biological samples in both two and three dimensions, but the linear fluorescence condition imposes a hard, physical resolution limit of approximately 120 nm. For these reasons, we decided to improve existing nonlinear SIM (NL-SIM) technology, such that it could be applied to biological samples while still realizing ultrahigh resolutions of approximately 50 nm. Instead of saturation or stimulated emission, we chose to exploit photoswitching, which elicits a nonlinear response from the sample at light intensities six to nine orders of magnitude lower than those needed for these other nonlinear phenomena. Using a reversibly photoswitchable fluorescent protein (i.e., a protein that can be switched between two spectrally distinct states using light) called Dronpa, we took 63 raw images with a SIM microscope at a light intensity of 1–10 W/cm 2 and visualized biological structures with a resolution of 40- to 70-nm resolution. Ernst Abbe formulated the diffraction limit of light more than a century ago, and up until the last few years, physicists and biologists have been operating in its shadow. However, in the last decade, a renaissance in fluorescence light microscopy has taken place, and numerous methods are now able to achieve resolution well beyond this “hard” physical boundary. Resolution of a few tens of nanometers is now possible, whereas previously, still subject to diffraction, the resolution was limited to a few hundreds of nanometers. Each of these subdiffraction or superresolution methods is based on one of two ideas: localization precision or patterned illumination light. The light microscope is an almost perfect tool for the life sciences, but its limited resolution means it is often ill suited for investigating life’s smallest and most puzzling mysteries. How is genetic information spatially structured and dynamically orchestrated? How do subcellular structures like organelles and proteins organize and interact with each other? What are the myriad ways in which cells, grow, divide, and communicate on nanometer-length scales? In the last decade, with the help of technological advances, imaging with a light microscope has improved dramatically, and a number of techniques—collectively termed superresolution light microscopy—have succeeded in the endeavor to improve the resolution, such that it is more relevant on the scale of these biological questions. It is unfortunate then that these superresolution methods typically need thousands of raw images, which may not fully capture biology’s rapid and intricate dynamic processes, or extremely high light intensities are likely to damage biological samples. Here, we present a method that achieves a resolution of approximately 40 nm—or four times the resolution of a conventional light microscope—using 10- to 1,000-fold fewer images and a light intensity six to nine orders of magnitude lower than that of other superresolution methods. Thus, our method brings the field of superresolution microscopy closer to its ultimate goal of realizing these ultrahigh resolutions while retaining the qualities that have made and will continue to make the light microscope an invaluable resource for biology.