Bleaching/blinking assisted localization microscopy for superresolution imaging using standard fluorescent molecules

Abstract

Superresolution imaging techniques based on the precise localization of single molecules, such as photoactivated localization microscopy (PALM) and stochastic optical reconstruction microscopy (STORM), achieve high resolution by fitting images of single fluorescent molecules with a theoretical Gaussian to localize them with a precision on the order of tens of nanometers. PALM/STORM rely on photoactivated proteins or photoswitching dyes, respectively, which makes them technically challenging. We present a simple and practical way of producing point localization-based superresolution images that does not require photoactivatable or photoswitching probes. Called bleaching/blinking assisted localization microscopy (BaLM), the technique relies on the intrinsic bleaching and blinking behaviors characteristic of all commonly used fluorescent probes. To detect single fluorophores, we simply acquire a stream of fluorescence images. Fluorophore bleach or blink-off events are detected by subtracting from each image of the series the subsequent image. Similarly, blink-on events are detected by subtracting from each frame the previous one. After image subtractions, fluorescence emission signals from single fluorophores are identified and the localizations are determined by fitting the fluorescence intensity distribution with a theoretical Gaussian. We also show that BaLM works with a spectrum of fluorescent molecules in the same sample. Thus, BaLM extends single molecule-based superresolution localization to samples labeled with multiple conventional fluorescent probes.

An image created from a fluorescently labeled biological sample is composed of the point-spread functions (PSF) of many fluorescent molecules. The resolution of a diffraction limited fluorescent microscope is ≈300 nm, precluding the fluorescent molecules overlapping within this radius from being separated from each other (i.e., resolved). Many biological processes involve changes that occur on spatial scales below this resolution limit, and, indeed, many cellular organelles are themselves entirely diffraction limited. Therefore, the development of microcopy techniques that produce images that circumvent the diffraction limit has been the topic of active research in recent years (1, 2). The result has been the development of several, so-called superresolution fluorescence techniques, including stimulated emission depletion microscopy (STED) (3, 4), saturated structured illumination microscopy (SSIM) (5–7), and the family of single molecule localization microscopy techniques, founded by photoactivated localization microscopy (PALM) and stochastic optical reconstruction microscopy (STORM) (8–15).

Of the superresolution techniques, the single molecule localization techniques achieve the highest spatial resolution (1). The position of a single fluorescent molecule can be determined within tens of nanometers by fitting its PSF to a theoretical 2D Gaussian. This type of single molecule fitting cannot be implemented when PSFs of neighboring molecules overlap (8). PALM and STORM overcome this problem by using stochastic activation of photoactivatable or photoswitchable probes, respectively, in which only a few molecules in a sample are emitting at any one time (8, 9). This method allows for a cycle in which a sparse field of single fluorescent molecules is imaged, then subsequently turned off, and followed by the switching on of a sparse set of different molecules. The localization precision of each molecule in each image can then be determined and a superresolution image can be created by combining the localizations of all of the molecules (8, 9).

The PALM/STORM concept has been expanded to localize up to two different proteins using PALM (14) and three different synthetic dyes using STORM (10). Furthermore, an increased knowledge of the photoswitching characteristics of fluorescent proteins and synthetic dyes has led to the development of some simplifications of the PALM and/or STORM methods. For example, direct STORM (dSTORM) takes advantage of the ability to switch Cy5 or Alexa Fluor 647 with a cycle of green and red laser light (11). It was subsequently shown that it is possible to switch most Alexa dyes by imaging them in the presence of a reducing agent (β-mercaptoethanol or DTT) with only excitation laser light (16). Also, intense excitation laser light can be used to turn off many molecules, and then individual molecules can be imaged as a subset of the molecules slowly turn back on over time (ground state depletion followed by individual molecule return, GSDIM) (17). In another approach, genetically expressed fluorescent proteins are prebleached to a virtual dark field of view where only sparse single molecules blink on/off (18, 19). However, the number of blinking fluorescent proteins left after the prebleaching is low relative to the initial pool.

After imaging, the above methods use a similar fitting algorithm to precisely localize the single molecules and reconstruct a superresolution image. Although the imaging techniques outlined above allow for the creation of superresolution images, they are technically challenging and require a large investment in both time and money to incorporate them into a laboratory's tool box. Additionally, neither PALM nor STORM can be performed by simply imaging of standard fluorescent proteins and/or synthetic dyes.

When acquiring images of a population of standard fluorophores, the overall fluorescence intensity decays because of fluorescence photobleaching. Two techniques published in 2004 demonstrated the superresolution localization of two to five single molecules in a diffraction limited spot using loss of signal as individual molecules photobleached (20, 21). The first of the two papers reported a technique called SHRImP (single-molecule high-resolution imaging with photobleaching) (20). The authors conjugated two ends of a DNA molecule each with a single fluorescent dye. They were able to locate the position of the two dyes separated by 10–20 nm with 5-nm localization precision (20). The second paper reported NALMS (nanometer-localized multiple single-molecule) fluorescence microscopy where five molecules in a diffraction limited spot were individually detected by using a similar fitting method (21). Despite this ability, the sole use of photobleaching and photoblinking for mapping out the distribution of single molecules at high density, as in PALM or STORM, has not been established.

The major technical difficultly in using bleaching and/or blinking of fluorescent molecules to produce superresolution images is the requirement for imaging all of the fluorescing molecules in a sample without saturating the camera, while, at the same time, being able to detect single molecules above camera noise. We present here a practical experimental technique for producing image series in which single molecules can be revealed by high magnification streaming acquisition of fluorescent samples followed by image subtraction. This technique, which we call bleaching/blinking assisted localization microscopy (BaLM), is based on the detection of fluorescent molecules that are either lost to irreversible photobleaching or lost/gained by blinking. BaLM does not require prebleaching before single molecules are detected, distinguishing it from prior approaches that require prebleaching to detect single molecules (18, 19). BaLM works with all commonly used synthetic fluorescent dyes and genetically expressed fluorescent proteins. Thus, BaLM can be readily used in multicolor superresolution experiments, mapping out the molecular distribution of up to four different proteins in a sample, as shown here. Furthermore, BaLM allows for molecules to be localized by similar algorithms already used for the analysis of PALM data. As such, BaLM offers a practical and versatile approach for obtaining superresolution images that can stand alone or be used in conjunction with other superresolution approaches.

Results

Fluorescent molecules in a fixed cell can either turn off (i.e., irreversibly bleach or reversibly blink-off) or turn on (i.e., blink-on) from a fluorescent population over time with continuous excitation. In theory, turn-off (i.e., bleach or blink-off) events should be detectable simply by subtracting the subsequent image from each image of a stream acquisition. Likewise, blink-on events should be detectable by subtracting from each image the previous one. Each subtracted image should then contain the PSF of individual fluorophores that either turned off or on during acquisition of the two frames. The position of each fluorophore can then be determined by fitting each molecule to a theoretical Gaussian function, similar to how molecules are localized in PALM/STORM (8, 9). BaLM takes advantage of these characteristics of fluorescent molecules. That is, it creates a superresolution image from single diffraction limited spots generated by subtraction of consecutive images from a stream acquisition of a bleaching fluorescently labeled sample. The single diffraction limited spots being quantified in BaLM are a result of the disappearance of molecules that are irreversibly photobleaching or blinking off, and the appearance of molecules that are blinking on.

It has been analytically shown that maximum precision of single molecules imaged with a low background is obtained with a pixel size equal in dimension to the PSF of the imaging system (22). This calculation is why PALM, which has an inherently low background (8), uses pixel dimensions close to the PSF of the system (typically ≈150 nm diameter-sized pixel) in acquiring data. In BaLM, where single molecules are identified from difference images, there is much higher background. The higher background comes from a variety of sources, including fluctuations in ensemble fluorescent intensities from frame to frame and from bleaching/blinking of nearby molecules. If one subtracts images acquired with a pixel size similar to the PSF of the peak intensity under these conditions, single maxima get averaged out because of the higher background. It thus becomes difficult to detect single peaks by using the 150-nm pixel size used in PALM, as schematically illustrated in Fig. S1A. To overcome this problem, BaLM uses higher magnification to obtain smaller pixel size (≈60 nm). This magnification allows much better detection of single maxima in the difference images arising from bleaching/blinking of molecules (see illustration in Fig. S1B). Additionally, by spreading out the PSF of the molecules over more pixels, it is possible to image more molecules in a diffraction limited spot before saturating each individual pixel.

To demonstrate that BaLM works using a standard EMCCD camera, we used a secondary magnification element to create an effective pixel size of 61 nm and imaged microtubules in COS-7 cells labeled with a primary antibody to α-tubulin and an Alexa Fluor-488 labeled secondary. Microtubules are biopolymers with a 25-nm diameter tubular structure, making them useful test objects for superresolution techniques (10). The diffraction-limited image from the first frame of a time-lapse acquisition of microtubules from the periphery of a fixed cell is shown in Fig. 1 A and B. Frames from this series were subsequently subtracted to detect signal intensities that disappeared/appeared. Three representative frames from a subtraction series is shown in Fig. 1C. Single local maxima were then identified in subtracted images as having a higher signal than the background (Fig. 1C Middle). These local maxima were then defined as single molecules if, (i) their radius was <2.5× the SD of the effective PSF, and, (ii) they were emitting between 2,000 and 12,000 photons over a 1-s integration time (Materials and Methods). This range was determined based on previously reported measurements of photon counts of Alexa dyes (23), as well as examination of the photon counts of single molecules at the end of bleaching acquisitions in our experiments (Materials and Methods). The distribution of the photon counts of molecules satisfying the criteria described above is shown in Fig. 1D. Single molecules were then fit with a theoretical Gaussian to obtain localization precisions (Fig. 1 E and F; σ = 48.0 ± 26.1 nm, see Materials and Methods). The rendered superresolution localizations of the molecules in Fig. 1C Middle are shown in Fig. 1E.

Fig. 1.

Detection of molecules and construction of BaLM image. (A) Diffraction limited TIRF image of microtubules localized with an Alexa Fluor 488 secondary antibody. Arrowheads denote fluorescent beads used for image alignment. (B) Magnified view of the yellow box in A. (C) Three representative consecutive frames of a subtraction image series showing before and after two molecules bleached/blinked off. (D) The distribution of photon counts of single molecules (n = 15 microtubules). (E) The localization of the two molecules in C Middle. (F) The distribution of localization precisions of BaLM localized molecules (n = 15 microtubules). (G) Schematic demonstrating how localized molecules from different frames are summed to create a BaLM image reconstruction. (H) A BaLM image reconstruction of the microtubules shown in B. (I) Image intensity linescans (normalized to 1) through the dotted line from the diffraction limited image in B (purple line in graph) and the dotted line from the BaLM reconstruction in H (blue line in graph). (Scale bars: A, 5 μm; B, 1 μm; C and E, 0.6 μm.)

To create the BaLM reconstruction, all superresolution localizations were combined into one single probability map (see schematic in Fig. 1G and reconstruction in Fig. 1H). Note that the microtubule-based structures in the BaLM reconstruction appear to be thinner than in the diffraction-limited image (Fig. 1H compared with Fig. 1B). To confirm this difference in width, we examined line scans through the same region in the diffraction limited data and the BaLM reconstruction (Fig. 1I). Note that the two microtubules in this region are resolved in the BaLM data, but not in the diffraction-limited data (Fig. 1I).

When doing immunofluorescence labeling of proteins, there is always background fluorescence from nonspecific interactions with either the growth substrate or cell (Fig. S1). These fluorophores will show up in the BaLM analysis and in any other point localization-based superresolution technique (Fig. S2). Although spatial filtering of the data could eliminate these molecules, as has been recently shown (24), we present nonfiltered images to emphasize the quality and sensitivity of BaLM imaging.

It can take a few hours to perform BaLM analysis with our custom Matlab-based software to get an estimate of the localization precision of each molecule within a typical dataset of 1,000–3,000 images. However, identifying molecules and creating a BaLM reconstruction from the center of mass of these molecules can be done quickly with a freely available ImageJ plugin software package, QuickPALM (Fig. 2). This plugin also makes it possible to process BaLM data exclusively by using ImageJ plugins in <5 min (Materials and Methods). QuickPALM defines single molecules based on the size, signal/noise ratio, and shape of their PSF but does not calculate a precision. However, once the user knows the average precision of single molecules for a certain set of imaging conditions, then QuickPALM becomes a quick and useful tool that can be used to produce BaLM images. The diffraction limited image of microtubules labeled with Alexa Fluor 488 from a bleaching data set and the subsequent BaLM image reconstructed with QuickPALM are shown in Fig. 2 A–C. Line scans drawn across an area of overlapping microtubules (Fig. 2C Inset) reveal no separation of the microtubules in the diffraction limited image but clear separation in the reconstructed BaLM image.

Fig. 2.

BaLM reconstructions created with the ImageJ plugin, QuickPALM. (A–C) Diffraction limited image (A), BaLM image (B), and overlay (C) of a microtubule labeled with Alexa Fluor 488. Graph in C is a line scan through the yellow line in C. (D) Overlay of the diffraction limited image (purple) and BaLM image (light blue) of myosin IIC-GFP. (Scale bars: 2 μm.)

BaLM reconstructions also can be acquired from samples expressing GFP-tagged proteins. To demonstrate this possibility, we exogenously expressed the molecular motor, myosin IIC fused to EGFP, in COS-7 cells (Fig. 2D). Myosin II has a stereotypical periodic localization along actin filament-based stress fibers that can be detected with diffraction limited imaging (Fig. 2D, see periodic purple signal from myosin IIC-GFP that distributes along elongated actin filaments). BaLM imaging of this sample provides significantly higher resolution of myosin IIC-EGFP, showing new substructure associated with each diffraction-limited puncta (Fig. 2D, see BaLM localizations in blue). The mean localization precision (σ) of individual EGFP molecules imaged with BaLM was 57.2 ± 22.9 nm.

To compare BaLM's performance with another superresolution technique, we compared microtubule structures localized with BaLM to those localized with PALM (Fig. 3). PALM imaging and processing was performed as reported (8) on COS-7 cells expressing α-tubulin fused to photoactivatable GFP (PA-GFP) (Fig. 3 A and B). Our localization precision was found to be ≈20–30 nm, which is similar to previous reports of PALM using PA-GFP (8). Examples of diffraction-limited images and PALM or BaLM reconstructions are shown in Fig. 3 A–D. To quantify the improvement in the resolution obtained with PALM and BaLM to that of the diffraction-limited images, we created line scans of the intensity distribution across the width of microtubules and measured the width of the distribution at half of the maximum signal (Fig. 3 E and F, dotted lines). As expected for the diffraction-limited image, the half-width was ≈300 nm (301.8 ± 45.7 nm). We found that the half-width of microtubules in both PALM and BaLM images was significantly narrower than those from the diffraction-limited images (PALM, 46.6 ± 6.3 nm; BaLM, 64.5 ± 11.7 nm; Fig. 3G). PALM localization of microtubules is slightly more precise because of the better precision obtained from PALM data compared with BaLM. This difference is primarily due to the lower background signal in PALM data.

Fig. 3.

Structural resolution of BaLM compared with PALM. (A and B) Total raw diffraction limited image (A) and PALM image (B) from a cell expressing α-tubulin-PAGFP. (C and D) Diffraction limited image (C) and BaLM image (D) of microtubules labeled with Alexa Fluor 488. (E) Linescans (lines in A and B) through a single microtubule in both the diffraction limited image (purple) and the PALM image (blue) showing the decreased width of the microtubule as visualized with PALM. (F) Linescans (lines in C and D) through a single microtubule in both the diffraction limited image (purple) and the PALM image (blue) showing the decreased width of the microtubule as visualized with BaLM. (G) Quantification of the widths of microtubules measured as the width of signal distribution from diffraction limited images (301.8 ± 45.7 nm, n = 10 microtubules, 5 linescans per microtubule), PALM images (46.6 ± 6.3 nm, n = 10 microtubules, 5 linescans per microtubule), and BaLM images (64.9 ± 11.7 nm, n = 10 microtubules, 5 linescans per microtubule). Error bars denote SD. *P value from Student's t test <0.001 when distribution are compared with diffraction limited microtubule widths. (Scale bars: 1 μm.)

To further validate the performance of BaLM, we next compared BaLM images of microtubules to that obtained with conventional immunogold electron microscopy (Fig. 4). The localization resolution of the immunogold technique is mainly limited by the size of the immunogold particle and the antibodies. Therefore, we performed electron microscopy on rotary shadowed cells that were fixed by using the same procedure as for the BaLM samples. Microtubules were localized with a secondary antibody conjugated to 15-nm gold particles (Fig. 4C). To compare the distribution of BaLM-localized fluorophores along microtubules with the immunogold particles, we measured the distance from the molecule or gold particle's centroid to the middle of the microtubule (see schematic in Fig. 4D). The average distance from the center of microtubules for the immunogold particle was 17.0 ± 13.7 nm, whereas the average distance for BaLM-localized molecules was 32.7 ± 22.8 nm. The better performance of electron microscopy stems from the subnanometer localization precision with which the gold particles can be localized. However, BaLM produces a higher density of labeling and can be more practical for the simultaneous localization of different proteins.

Fig. 4.

Structural resolution of BaLM compared with electron microscopy. (A and B) Diffraction limited image (A) and locations of the center of molecules localized with BaLM (B) from a single microtubule. (C) Electron micrograph of a rotary-shadowed microtubule labeled with gold particles. (D) Schematic shows how the centers of the molecules or gold beads were measured from the center line of a microtubule and graph shows the averages and SDs of these distances. BaLM molecules deviated an average of 32.7 ± 22.8 nm (n = 258 molecules from three microtubules) from the center of microtubules, whereas immunogold particles deviated 17.0 ± 13.7 nm (n = 139 gold particles from six microtubules). ^#P < 0.05. (Scale bars: 100 nm.)

The number of proteins that can be localized with BaLM should only be limited by the ability to separate the emission of different fluorophores while still being able to detect single molecules above camera noise. To explore this possibility, we investigated the feasibility of doing BaLM analysis on multiple types of fluorescent dyes in a single sample. We first labeled three different proteins in cells using immunofluorescence. Myosin IIA, clathrin light chain, and microtubules were labeled with Alexa Fluor 647, Alexa Fluor 561, and Alexa Fluor 488, respectively (Fig. 5A). A dataset in which all molecules were irreversibly bleached was acquired for each channel in the following order: Alexa Fluor 647, followed by Alexa Fluor 488 and then Alexa Fluor 561. BaLM localization of these three proteins revealed discrete nonoverlapping structures, indicating that there was insignificant bleed-through between channels (Fig. 5A). Additionally, we labeled microtubules with a mixture of competing fluorescently conjugated secondary antibodies: Alexa Fluor 405, Alexa Fluor 488, Alexa Fluor 561, and Alexa Fluor 647. Although the secondary antibodies showed different labeling efficiencies (Alexa Fluor 647 showed particularly low labeling affinity when mixed with competing antibodies), BaLM worked in all four channels. The diffraction limited and BaLM images of a single microtubule from the four separate channels is shown in Fig. 5B. This difference demonstrates that one can use BaLM to produce similar structural information in multiple channels. Finally, it should be possible to expand the use of BaLM to more than four channels.

Fig. 5.

Multicolor BaLM. (A) BaLM reconstruction showing the localization of microtubules (purple), clathrin pits (yellow), and myosin IIA filaments (blue) in the same cell. Immunofluorescent labeling of myosin IIA, clathrin light chain, and microtubules was performed with Alexa Fluor 647, Alexa Fluor 488, and Alexa Fluor 561, respectively. The channels were imaged in the following order: Alexa Fluor 647, Alexa Fluor 488, and Alexa Fluor 561. (B) The diffraction limited images (Left) and the BaLM images (Right) of a single microtubule that has been labeled with a mixture of Alexa Fluor 405 (purple), Alexa Fluor 488 (green), Alexa Fluor 561 (orange), and Alexa Fluor 647 (red) secondary antibodies is shown. The channels were imaged in the following order: Alexa Fluor 647, Alexa Fluor 405, Alexa Fluor 488, and Alexa Fluor 561. (Scale bars: 2 μm.)

Discussion

In this article, we describe a single molecule superresolution imaging technique, BaLM, which relies on the intrinsic bleaching and blinking behavior of fluorescent molecules. BaLM is a practical procedure for generating superresolution images of samples containing any commonly used fluorophore. BaLM does not require the use of engineered photoactivatable proteins or photo-switchable synthetic dye pairs, although it should work with these as well.

A common need in fluorescence microscopy is the ability to image multiple fluorophores and to compare localization of multiple proteins. Currently, it is technically challenging to localize more than two different proteins by using PALM because of the limited variety of photoactivatable probes that can be used in combination. By contrast, BaLM can use the same probes used in conventional fluorescence microcopy. An example of this versatility is illustrated in Fig. 5, where a microtubule labeled with secondary antibodies having one of four different Alexa dyes was imaged and processed with BaLM analysis. Previous multicolor superresolution imaging typically uses two (25) or three fluorophores (10). With BaLM, the number of different fluorophores that can be combined in the same sample is limited only by the user's ability to separate their fluorescence emission. Using BaLM, researchers can design experiments using several different imaging channels with a combination of genetically expressed or synthetic fluorescent molecules.

The simplicity and practicality of BaLM can be a critical advantage in relation to other superresolution methods when localization is combined with additional procedures such as array tomography (26, 27) and correlative microscopy (28, 29). Array tomography uses repeated cycles of immunofluorescence labeling, imaging, and stain elution (26, 27). Thus, BaLM could extend the versatility of array tomography by allowing the use of a broad range of readily available fluorescent immunoprobes. BaLM could also be used in correlative electron and fluorescence microscopy for the combined localization of molecules and the ultrastructural visualization of their microenvironment (29–31).

The general method in which molecules are localized with BaLM, PALM, and STORM are very similar and, in fact, these techniques can be used in various combinations. For example, a PALM sample that already contains a photoactivatable protein such as PAGFP can be counterstained with other fluorescence dyes and imaged with both PALM and BaLM. Furthermore, BaLM in principle can be adapted to produce 3D localization by using a similar approach to that developed for 3D STORM (12, 32). This 3D localization can be achieved by using astigmatism in the fluorescence image by simply adding a cylindrical lens in the imaging path to cause fluorescent molecules to appear elliptical (33). The ellipticity of the single molecule image can be used to calculate the molecule's axial position with high precision.

The localization precision in PALM/STORM-type superresolution techniques depends on several factors including camera performance, fluorophore properties, and synchronization of image acquisition to fluorophore activation/emission cycles. Properties of fluorophores such as brightness, quantum yield, and photostability, vary widely (10, 34, 35). The number of photons obtained per activation cycle significantly influences the localization precision (34). Furthermore, the measurement and calculation of the background is essential to the precision calculation. The overall background signal of a BaLM image is increased compared with PALM/STORM because it is produced by the subtraction of two images, thus combining the background noise of the two images. Our measurements for the precision of an Alexa dye with BaLM is ≈50 nm, which is significantly less than the precision reported for PALM/STORM (≈10–20 nm) (8, 9). Although this lower precision may be an inherent disadvantage of BaLM, we feel that its simplicity and accessibility make BaLM a viable alternative to more technically difficult point-localization techniques. Additionally, with the development of fluorescent probes with higher photostability and improvements in the overall performance of EMCCD cameras, the resolution of these two techniques will be more comparable.

Bleaching and blinking behaviors are intrinsic to virtually all fluorescent probes (36). Although bleaching is often regarded as a hindrance, and blinking is often ignored, we show here that both characteristics can be used to produce superresolution images. Fluorescence bleaching has been used to count molecules in a molecular complex (20, 21, 37, 38). BaLM could also be potentially extended to provide information about the actual number of fluorescent molecules in a multilabeled complex sample. During the logarithmic decay of fluorescence, the blinking events are relatively rare in comparison with the bleaching events (37); thus an algorithm could be developed to separately account for the bleaching events and derive approximate quantitative information on the fluorophores being localized, analogous to what has been recently developed for PALM (39).

Materials and Methods

Cell Culture and Sample Preparation.

COS-7 or PtK1 cells were cultured, extracted, and fixed as described (40). An α-tubulin antibody (Sigma) was used to localize microtubules. For expression experiments, cells were transfected with myosin IIC-EGFP or α-tubulin-PAGFP. Replicas of critical point dried samples for electron microscopy were prepared as described (41, 42). See SI Materials and Methods for details.

Image Acquisition.

Four hundred to 3,000 images were acquired by using variable angle TIRF with a Nikon Eclipse Ti microscope with either an Evolve (Photometrics) or an iXon 847 (Andor) EMCCD camera. Total magnification was 40 or 61 nm per pixel. This high magnification allowed the use of more of the dynamic range of the camera while avoiding pixel saturation. See SI Materials and Methods for details.

Practical Image Analyses Using ImageJ.

We used freely available ImageJ plugins (http://rsbweb.nih.gov/ij/plugins) to analyze BaLM datasets. To correct drift in the images, we used the translational setting of StackReg. Bleaching/blinking off events and blinking on events were identified with the delta F down setting or the delta F up setting, respectively, of T functions. BaLM molecules were localized and image reconstructions were rendered by using QuickPALM (43). See SI Materials and Methods for details.

Data Analyses Using BaLM Software.

Custom BaLM programs were written in MatLab (Mathworks). Single molecule localization and estimation of precision of localization were performed in a similar manner as described (8, 22). See SI Materials and Methods for details.

Note Added in Proof.

While this paper was in review, another paper published a similar approach to BaLM using bleaching of a single synthetic dye (44). As shown here, BaLM utilizes both the bleaching and blinking behaviors of synthetic dyes or fluorescent proteins and can be used in multi-spectral imaging.