Boosting the localization precision of dSTORM by biocompatible metal-dielectric coated glass coverslips

Super-resolution techniques such as direct Stochastic Optical Reconstruction Microscopy (dSTORM) have become versatile and well-established tools for biological imaging over the last century. Here, we theoretically and experimentally show that clever combination of different fluorescence modalities allows further improvements. We found that the interaction of fluorophores with plasmonic surfaces boost super-resolution performance in dSTORM approaches as it allows for tailoring the excitation and emission properties. The strength of the approach is that no further specialized microscope setup is required as the described enhancement solely rely on metal-dielectric coated glass coverslips that are straightforward to fabricate. Such biocompatible plasmonic nanolayers enhance the signal-to-noise ratio of dSTORM, and thus sharpens the localization precision by a factor of two.


Super-resolution techniques such as direct Stochastic Optical Reconstruction Microscopy
(dSTORM) have become versatile and well-established tools for biological imaging over the last century. Here, we theoretically and experimentally show that clever combination of different fluorescence modalities allows further improvements. We found that the interaction of fluorophores with plasmonic surfaces boost super-resolution performance in dSTORM approaches as it allows for tailoring the excitation and emission properties. The strength of the approach is that no further specialized microscope setup is required as the described enhancement solely rely on metal-dielectric coated glass coverslips that are straightforward to fabricate. Such biocompatible plasmonic nanolayers enhance the signal-to-noise ratio of dSTORM, and thus sharpens the localization precision by a factor of two.
To date, advanced fluorescence microscopy methods for resolution enhancement are based on either on-off fluorophores (for PALM [1], STORM [2], dSTORM [3], [4] or STED [5]), Structured Illumination Microscopy (SIM) [6], 4Pi-microscopy [7] and deconvolution routines [8]. Despite their great success, all of these techniques have their advantages and challenges. Intrinsically, localization techniques usually only provide limited temporal resolution, but excellent localization precision in an affordable, straightforward setup. SIM together with deconvolution routines only provides a limited spatial resolution improvement and the need of expensive hard-and software, however lately provides quite impressive acquisition speeds ( [9], [10]) for imaging of tick samples.
The STED or 4Pi-microscope setup and alignment requires relatively high light doses and sophisticated optical engineering, however providing excellent super-resolution in 2D or 3D. One technique that is widely used due to its advantageous cost-benefit ratio is dSTORM. The conceptual approach of dSTORM is analogous to the approach of STORM [2], however simplified without the need of an activator molecule. In dSTORM, a super-resolved image is obtained by subsequent imaging of well-separated subsets of blinking fluorophores. The localization of a single fluorophores can be determined with a precision exceeding the Abbe limit by far, so that dSTORM pushes the resolution to typically 30 nm without further tweaks and tricks.

2/16
For localization based methods like dSTORM the localization precision is the crucial limit for the accessible resolution which can be estimated based on the image and signal parameters ([11]- [13]) and mainly depends on signal brightness and contrast.
There have been attempts to increase super-resolution performance by maximizing the photon yield with optimized dyes [14], additives [15] or by cooling the sample down to very low temperatures [16] or doubling detection efficiency with a 4Pi-Setup [17]. Still all this approaches lack the property of being universally applicable and partly have issues regarding biocompatibility.
They have therefore not reached widespread application.
Combining optical with plasmonic approaches opens exciting perspectives: So called surface plasmons in specially designed nanostructures can generate extremely high photon densities in a nanoscopic volume that is much lower than the Abbe criteria usually allows [18]. This phenomenon of metal-fluorophore interaction has been already theoretically described in 1978 [19] and a first experimental example was given by Drexhage [20]. The interaction of fluorophores with plasmonic surfaces enables amplified fluorescence, increased photostability [21] and distance dependent dynamical [22] and spectral emission shifts [23].
Several applications in the field of plasmon enhanced fluorescence imaging and spectroscopy based on metal films ( [24], [25]), hyperbolic metamaterials [26] and nanoantenna structures ([27]- [31]) have already been presented. Many of the plasmonic structures mentioned here rely on a very complex nanostructure design and yield limited biocompatibility, limited applicability and sophisticated implementation.
Here we show how to use relatively simple biocompatible nanolayers to be applied in superresolution bioimaging. A first attempt to combine the effects of metal-fluorophore interaction with super-resolution microscopy has been made by Berndt et al. [32]. Berndt and co-workers were able to determine the axial position of immobilized microtubules with nanometer precision based on the distance-dependent fluorescence lifetime modulation near a metal-coated surface. In a next step this method was also used to resolve the height profile of the membrane of a living cell with nanometer accuracy ( [22], [33]). As stand-alone techniques, both approaches are only able to provide super-resolution capabilities in the axial dimension. The strength of the concept presented here lies in the straightforward implementation.
Here, we present how simple, biocompatible metal-dielectric nanocoatings can enhance the lateral resolution performance of dSTORM in a standard epifluorescence dSTORM setup. First, we characterized the technique by using the well-known geometry of fluorescently labeled microtubules. To demonstrate the potential of our optoplasmonic approach we resolved the eight-3/16 fold symmetry of the GP210-ring associated with the nuclear pore complex and show the enhanced localization precision. The nuclear pore complex is a membrane protein which is of interest to many biological questions as it plays a key role in the regulation of molecular traffic between the cytoplasm and the nucleus [34]. The eightfold symmetry of the GP210 protein associated with anchoring the nuclear pore complex in the nuclear membrane [35] can be resolved by dSTORM or STED as shown previously ( [36]- [38]), however, it remains inaccessible by diffraction limited fluorescence imaging methods.

Results
To find the ideal design of the metal-dielectric coating for our application we performed simulations of the distance dependent fluorescence enhancement for different substrate geometries considering the specific fluorophore excitation and emission properties. To test the super-resolution performance, we carried out dSTORM experiments both on the especially designed substrate and on a bare glass coverslip under otherwise equal conditions. Both image stacks were analyzed with the same settings for localization and image reconstruction.
We detected a clear fluorescence signal enhancement of a factor 1.6 that was also translated into a localization precision enhancement of a factor 1.25 (see figure 3).

Discussion
In conclusion, we have shown that by an anticipatory design of silver-silicon nitride substrates the localization precision of dSTORM is enhanced when imaging microtubules or the nuclear pore complex. The metal-dielectric coated substrates are a versatile tool: The simple tree-ply design of our metal-dielectric coatings grants a straightforward one-step fabrication and allows tailoring the shape of the resulting enhancement field to the sample geometry while taking into account the excitation and emission properties of the fluorescent label at hand. Besides this, the silicon nitridecoating prevents fluorescence quenching and assures the biocompatibility of the substrate allowing the cultivation of living cells. Here we fabricated the described nanostructures under clean-room conditions. However, those could be even fabricated by non-specialists and without clean-room access using suitable tabletop thin film deposition systems, which will be implemented by us in future, work.
The handling of the prefabricated metal-dielectric substrates is user-friendly, as it does not require any further training or caution. Thus, the technique provides a high degree of versatility concerning application and implementation in any (dSTORM) microscopic setup where only the coverslip is modified. As other optoplasmonic approaches our method is surface bound, which only takes effect up to a height of about 160 nm above the substrate interface. Here, this is beneficial as allows for selectively enhancing the fluorescence signal in the plane right in vicinity above the surface while suppressing background signal from regions further away. This is increasing the contrast like in an evanescent filed illumination approach as for total internal reflection fluorescence (TIRF). We understand that this could also be a limitation in other applications resulting in limited or insufficient penetration depth. In principle, it is possible to extend the axial range of the enhancement field when exciting long range plasmons. But as the confinement of the field decreases with growing surface distance the beneficial enhancement effects would be diminished when spread over a larger operating range.
As we have demonstrated, this approach is especially interesting for super-resolution imaging, but it could also be advantageous in contrast enhanced single molecule tracking or single molecule detection in high throughput screening assays.
The simulations even predict a signal enhancement of up to a factor four, which would translate into a precision enhancement of a factor two. The maximum values were not yet reached experimentally, most likely due to underestimation of the sample's distance or photophysical 6/16 effects we are not (yet) aware of. We believe, even with the current enhancement, this technique can tip the scales in favor of a clearly resolved structure with further experimental developments of such an optoplasmonic approach becoming more than promising.     Image reconstruction and analysis: dSTORM images were reconstructed with the ImageJ plugin ThunderSTORM [41]. For each experiment, the same localization parameters were used.
For the microtubule experiments, localizations with less than four neighbors in a 60 nm radius were discarded. In case of the nuclear pore complex, reconstruction localization duplicates were filtered by merging localization events that repeatedly occurred within a radius of 15 nm with a maximum of 50 off-frames in between. Only localizations with an intensity of more than 500 and less than 5000 photons were considered and in order to discard events from unspecific background a density filter of a minimum of 5 events in a radius of 50 nm was applied.

13/16
Simulation: Simulations of the enhancement effect were performed with the commercial software Comsol Multiphysics™ 4.4. For the excitation enhancement, a plane wave excitation encountering the metal-dielectric substrate of the experimentally used geometry was compared with the excitation field in case of a plane glass substrate. In order to theoretically predict the emission enhancement, the distance dependent radiative rate of a dipole located near a metal-dielectric substrate of a glass coverslip was compared. Based on the retrieved height dependent excitation and radiative enhancement rates an efficient enhancement rate was calculated. In case of the microtubules the height distribution of the fluorophores is very well defined by the sample geometry and was also considered. Simulations of the fluorescence images of the nuclear pore complex were calculated with MatLab (Mathworks Inc., Natick, MA).