Recent years have witnessed rapid development of super-resolution fluorescence imaging methods which substantially surpass the diffraction limit1,2
. These include methods based on patterned illumination, such as stimulated emission depletion (STED and related RESOLFT) microscopy1,3
and saturated structured illumination microscopy (SSIM)4
, as well as methods based on single-molecule switching and localization, such as stochastic optical reconstruction microscopy (STORM)5
and (fluorescence) photoactivated localization microscopy ((F)PALM)6,7
. Resolutions of 10–100 nm have been achieved when imaging biological samples using these methods, allowing otherwise hidden details of subcellular organelles or structures to be revealed. However, given that most protein molecules are only a few nanometers in size, considerably higher resolutions will be required to directly resolve molecular interactions in cells.
STORM and (F)PALM achieve sub-diffraction-limit resolution by sequentially switching on and localizing individual fluorophores in order to build up a high-resolution map of the probes that decorate an imaging target. An important requirement of the method is that the probes can be prepared in a dark state with only a controllably small fraction switched on at any time. Since the localization precision scales approximately as the inverse square root of the number of photons detected from the activated probe, the image resolution depends critically on the number of photons detected per switching event for the probe. A variety of photoswitchable/photoactivatable fluorescent probes have been previously used, including organic dyes8–10
, fluorescent proteins11
, and quantum dots12
. These probes typically give on the order of 103
detected photons per switching event. Although the best fluorescent dyes can emit ~106
detected photons contiguously and provide a localization precision of ~1 nm13,14
, this potential has not yet been realized in super-resolution imaging due in part to the inability to prepare these fluorophores in an activatable dark state.
Here, we report a procedure to chemically convert fluorophores to a stable dark state that can be subsequently photoactivated to the fluorescent state in a buffer optimized for high photon emission. The procedure entails treatment of fluorescent dyes with a reducing agent (sodium borohydride, NaBH4
), which converts the dyes to a long-lived reduced, or caged, form (). Conveniently, the procedure may be carried out in situ, after staining the sample. Because the absorption spectra of the reduced dyes are strongly blue-shifted, the dyes are effectively converted to a dark state. Reduced cyanine dyes have been previously used as sensors for detecting electron beams15
or reactive oxygen16
. Here, we demonstrate that reduced dyes are excellent photoactivatable fluorophores that yield orders of magnitude more photons than currently existing ones and can be used as probes for super-resolution imaging. The reduced dyes may be considered as “caged” because they contain a new covalent bond formed by addition of a hydride anion to the fluorophores (unpublished data) and because they may be photoactivated to return the dyes to their bright, fluorescent form.
Figure 1 Bright photoactivatable dyes created by reductive caging. (a) Left panel, initial fluorescence signal of a fixed cell immunolabeled for microtubules with Cy3B. Middle panel, fluorescence signal after reduction with NaBH4. Right panel, fluorescence signal (more ...)
Photoactivation of the reduced dyes can be facilitated by illumination with ultraviolet (UV) or violet light. To demonstrate the effectiveness of the NaBH4-induced caging and subsequent photo-uncaging processes, we immunolabeled microtubules in fixed cells with a bright, fluorescent dye, Cy3B. Incubation with NaBH4 quenched the initial fluorescence (under 561 nm excitation; , left) to a nearly undetectable level (, middle). After illumination with UV or violet light, the molecules became fluorescent again when excited with 561 nm light (, right). The recovery yield was 40%, i.e. 40% of the reduced dye molecules were photoactivatable. The incomplete recovery could be due to photobleaching of the dye molecules by the activation light or the formation of some non-recoverable reduced products.
We identified several commercially available dyes across the visible spectrum which can be reduced and subsequently photoactivated in this manner. The recovery fraction was 66% for the blue dye Atto 488, 35–40% for the yellow dyes Cy3 and Cy3B, and 12–17% for the red dyes Alexa 647 and Cy5.5 (Supplementary Fig. 1
). The relatively low recovery yield of the red dyes can be compensated for in super-resolution imaging by labeling individual proteins of interest with multiple dye molecules, although this type of labeling may not be achievable for all proteins.
We characterized the number of photons per switching event and the localization precision of individual dye molecules after activation. An example fluorescence time trace of a single Cy3B molecule before and after photoactivation is shown in . On average, we detected 270,000 photons over 10 s from each Cy3B molecule after photoactivation and prior to photobleaching or switching off using a low intensity of 0.11 kW/cm2
561 nm excitation (). By tracking the center positions of the images of individual Cy3B molecules over multiple frames, we measured the localization uncertainty for each molecule per switching event (Supplementary Fig. 2
). The average localization precision was determined to be 1.6 nm (), which corresponds to a potentially obtainable resolution of 3.7 nm (= 2.35 × localization precision assuming a Gaussian distribution). Similarly, another yellow dye, Cy3, gave 220,000 detected photons over 10 s at 0.11 kW/cm2
561-nm excitation, with a 1.8 nm localization precision and 4.4 nm potentially obtainable resolution ( and Supplementary Fig. 2
). The two red dyes, Alexa 647 and Cy5.5, emitted more photons per activation event, providing 1.7 and 1.1 million detected photons over 120 s at 0.08 kW/cm2
657-nm excitation, respectively ( and Supplementary Fig. 2
). The measured localization precisions of these two red dyes, 0.7 nm and 0.8 nm, correspond to potentially obtainable resolutions of 1.7 nm and 2 nm, respectively ( and Supplementary Fig. 2
). The blue dye Atto 488 emitted fewer photons (11,000 photons detected) over 0.8 s and gave a poorer localization precision (4.7 nm) and obtainable resolution (11 nm) at 0.5 kW/cm2
488-nm excitation ( and Supplementary Fig. 2
). Nonetheless, it still emitted substantially more photons than previously reported blue photoswitchable/activatable dyes10
. Notably, although the fluorophores could be switched off more rapidly with higher excitation intensities, the photon yields of these dyes decreases with increasing excitation intensities (Supplementary Fig. 3
). Thus, there is a trade-off between image resolution and acquisition speed. Additional properties relevant to super-resolution imaging, including the off-switching rate and on-off duty cycle (defined as the fraction of time that the molecule spends in the on state10
) at different excitation powers, are provided in Supplementary Fig. 3
We note that a different method to utilize reduced dyes, namely mixing suitable concentrations of NaBH4
and an oxidant to obtain on-off equilibrium switching of cyanine dyes, has been proposed for super-resolution imaging17
To demonstrate that the reductive caging and photoactivation procedure indeed allows for super-resolution imaging, we performed STORM imaging of cells immunostained for tubulin using antibodies labeled with multiple dye molecules to help compensate for low dye recovery yields. The immunostained samples were reduced in situ using NaBH4. Weak 405 nm light was then used to activate only a small fraction of the dyes at a time, allowing for high-precision localization of individual molecules. The activation, imaging, and localization procedure was iterated over 40–80 min until a sufficiently high density of localizations was acquired to allow the reconstruction of a super-resolution image. shows the STORM images obtained for the representative blue, yellow, and red dyes, Atto 488, Cy3B, and Cy5.5. Cy3 and Alexa 647 gave images of similar quality to that obtained with Cy3B and Cy5.5, respectively (data not shown).
Figure 2 STORM images of microtubules in cells stained by indirect immunofluorescence with (a) Atto 488, (b) Cy3B, and (c) Cy5.5. Samples were reduced in situ with NaBH4, washed, and then imaged with 488 nm, 561 nm, and 647 nm excitation light in a buffer designed (more ...)
To avoid antibody-induced broadening effects and to illustrate higher image resolution obtainable with these bright probes, we prepared samples of known structures that were directly labeled with dye. shows an image of microtubules which were polymerized in vitro and labeled directly with Cy3B. Cy3B was chosen here since it offered the best tradeoff between photon yield, recovery fraction, and imaging speed. Highly uniform labeling was achieved, with an average distance of 5 nm between neighboring localizations. To image these densely labeled samples, we used higher illumination intensity (0.6–1.2 kW/cm2
) to increase imaging speed and improve the on-off contrast ratio, and hence fewer photons (on average ~100,000) were detected per activation event (Supplementary Fig. 3
). Accordingly, the obtainable resolution was decreased from 3.7 nm to 6 nm and the total imaging time was 80 min. Brightfield fiducial markers were used to correct for drift, resulting in residual drift errors of <1 nm from frame to frame, and <2 nm over 1–2 hours.
Figure 3 STORM image of microtubules polymerized and labeled in vitro with Cy3B. (a) A STORM image of microtubules (green) with several magnified zoom-in images shown in insets. A portion of the corresponding conventional fluorescence image (magenta) is overlaid (more ...)
Zoom-in images of segments of microtubules and transverse profiles of these segments showed the hollow tubular structure of the microtubule, with higher localization densities near the edges and two peaks separated by 16–18 nm ( insets and ). Analysis of the transverse profiles of ten double-peaked microtubule segments from eight different microtubules gave a mean peak-to-peak separation of 16.4 ± 1.1 nm s.d., which is in quantitative agreement with the known 17 nm inner diameter of microtubules (Supplementary Fig. 4
). In comparison, previous super-resolution images of microtubules recorded with dimmer fluorophores have either failed to resolve the hollow tubular structure or have shown a substantially broadened tubular structure with diameter of ~37 nm due to antibody labeling10,18
. We note that not all microtubules show the hollow tubular structure (). The transverse profiles of these microtubules have a mean width of 26.8 ± 0.8 nm s.d. (ten microtubule segments from eight different microtubules). One possible reason that these microtubules were less well resolved could be due to inadequate fixation or immobilization.
Using the reductively caged Cy3B, we also imaged a thinner object, filamentous bacteriophage M13 virus. Transverse profiles of M13 particles showed ~9 nm widths (Supplementary Fig. 5
), again in good agreement with the known ~7 nm diameter of the virus particle.
The concentrations and incubation times of sodium borohydride used here for STORM imaging did not significantly perturb the structure of either microtubules or M13 bacteriophage as revealed by electron microscopy images, though longer incubations or higher concentrations of NaBH4
can degrade the microtubule samples (Supplementary Figs. 6, 7
It should be noted that at the level of molecular-scale resolution other factors may limit the overall resolution in addition to the photon number. One such factor is the dipole orientation effect if the fluorophore orientation is fixed or substantially constrained during imaging. Such fixed-dipole effects are most prominent for molecules out of focus19
, though small errors also exist for in-focus samples20
. The fluorophores studied here do not exhibit substantially hindered rotation (Supplementary Fig. 8
). The localization errors for molecules within 100 nm of the focal plane were estimated to be <5 nm for the red dyes and <6 nm the orange dyes (Supplementary Fig. 9
). For the well-focused 25 nm wide microtubules and 7 nm wide virus particles studied here, the dipole-induced error is minimal. However, to obtain a molecular-scale (1–2 nm) resolution for thick samples, the fixed-dipole effect can become a substantial limiting factor, and labeling methods to promote freely rotating dyes or analysis approaches to correct for fixed-dipole effects will be required (see also Supplementary Note
A second limiting factor to consider is the labeling strategy and density of fluorescent labels, since the overall obtainable image resolution may also be compromised by an insufficient density of labels. Here, we used direct protein labeling to achieve a relatively high labeling density without considerable broadening of the structure under study. In order to take advantage of the high resolution affordable by these fluorophores for cellular samples, other labeling strategies may be required, such as site-specific labeling using genetically fused peptide tags and enzymes or the use of small drug/ligand molecules or small immunolabels, such as nanobodies18
The high photon yield of the photoactivatable probes developed in this work was achieved at a cost of the long time it takes to switch off or bleach each fluorophore. Accordingly, the acquisition time of the STORM images obtained here was relatively long (tens of minutes), but is compatible with ultrastructural imaging of fixed samples. Because the photon number detected per activation event increases with decreasing illumination intensity, the obtainable resolution could be further increased compared to those observed in , when lower excitation laser intensities are used (Supplementary Fig. 3
), but the imaging speed would be slower. As another tradeoff, the on-off duty cycle also increases when the excitation intensity is decreased, reducing the number of molecules that can be localized in the diffraction-limited area. The high photon yields of these reductively caged probes potentially allow sufficiently high resolution to determine molecular organizations within individual protein complexes. The relatively low number of molecules that would need to be localized in this type of application should both allow substantially faster imaging and alleviate the constraint on localization density imposed by the on-off duty cycle, but may require improvements in the recovery yield of the dyes and the use of high-efficiency labeling strategies.
Beyond our demonstration here, there are also several other potential benefits of the high-photon yields of these photoactivatable probes. First, they should allow improved axial resolution for 3D superresolution imaging in addition to improved transverse resolution. Second, they should allow the use of microscope components that confer advantages to specific experiments but which are typically avoided in super-resolution imaging due to lower collection/detection efficiencies. For example, water-immersion objective lenses, which have lower collection efficiencies and larger point-spread functions than high numerical aperture (NA) oil-immersion lenses, could be used to image deeper into aqueous samples without incurring severe spherical aberrations (as high NA oil-immersion lenses do). The high-photon yield of these fluorophores should allow the use of water-immersion objective lenses while maintaining a very high image resolution. Third, fluorophore emission could be divided amongst several detection planes (multiplane detection) to allow extension of the focal range achievable with superresolution imaging.