Flat-top TIRF illumination boosts DNA-PAINT imaging and quantification

Super-resolution (SR) techniques have extended the optical resolution down to a few nanometers. However, quantitative treatment of SR data remains challenging due to its complex dependence on a manifold of experimental parameters. Among the different SR variants, DNA-PAINT is relatively straightforward to implement, since it achieves the necessary ‘blinking’ without the use of rather complex optical or chemical activation schemes. However, it still suffers from image and quantification artifacts caused by inhomogeneous optical excitation. Here we demonstrate that several experimental challenges can be alleviated by introducing a segment-wise analysis approach and ultimately overcome by implementing a flat-top illumination profile for TIRF microscopy using a commercially-available beam-shaping device. The improvements with regards to homogeneous spatial resolution and precise kinetic information over the whole field-of-view were quantitatively assayed using DNA origami and cell samples. Our findings open the door to high-throughput DNA-PAINT studies with thus far unprecedented accuracy for quantitative data interpretation.

1. The simple addition of a beam shaping device to homogenize the excitation in TIRF-based single molecule localization microscopy is not significant enough per se and the effects of the homogenized excitation of fluorescence as detailed by the authors are largely expected. As the authors have pointed out themselves, a few groups have already used (albeit) different methods for homogenizing the excitation field for flat-fielded excitation in single molecule fluorescence microscopy. The simple addition of another, commercially available, flat-fielding device is, in my opinion, not significant enough, plus other, previous attempts of flat-field, e.g. by Axelrod (probably the "father" of TIRF microscopy) by beam-scanning, as well as the well-known ringTIRF illumination scheme by the group of Derek Toomre were ignored. Furthermore, the reasoning for the need for flat-fielding is flawed. The authors appear to imply that coherent illumination is somehow required for TIRF illumination (e.g. on page 3), when in fact it is not. Coherent excitation might improve the efficiency of TIRF excitation, but the TIRF condition can be achieved with any form of light, even fully incoherent light, and, indeed, many commercial adapters for TIRF illumination in fluorescence microscopy don't necessarily require laser excitation.
2. In the second sentence on page 2, it is argued that due to the "non-fluorogenic" nature of imagers, some form of selective plane illumination is required?? Obviously, this should say "fluorogenic nature"? The main reason why selective plane, and in particular TIRF illumination, is required is that there are many imager strands freely diffusing in the background, which need to somehow be excluded from the excitation.
3. In my opinon, the only major new result of this manuscript is the observation that flat-fielding improves the photon statistics to the point where it can be used to effectively and efficiently excluded multiple localizations (see Fig. 3f) without the need for sophisticated multi-peak fitting algorithms. If one considers how much time and effort is still spent in trying to utilize data from multiple localizations of densely labeled samples by developing even more sophisticated methods, see e.g. the recent "HAWK" procedure by the group of Susan Cox in Nature Methods, then anything that helps improve this situation by simpler means is worth reporting. So, if the authors could refocus their paper on this application and elaborate on it, then I could still be convinced that the paper could be worthy of publication in Nature Communications, but, in its current form I would have to recommend that the paper be rejected and published in another, lower impact journal.
Reviewer #2 (Remarks to the Author): In their manuscript "Flat-top TIRF illumination boosts DNA-PAINT imaging and quantification" Stehr et al present the application of a recently first presented refractive approach to provide uniform illumination for TIRF to single molecule localisation microscopy, specifically DNA-PAINT.
The MS makes a number of interesting points. 1) In comparing the current widely used TIRF approaches that have an essentially Gaussian profile, the authors introduce an interesting segmented analysis for the Gaussian case.
It appears to this reviewer that several of the advantages that the MS puts forward as a result of the uniform illumination intensity, can be achieved by applying the proposed criteria in segments, as already done in some of the comparisons, i.e. (1) effective removal of double binding events via a simple photon number criterion, (2) binding time based distinction of docking strands. The criteria would have to be chosen by segment but this is clearly not computationally expensive nor complex. In this regard it seems to me that some of the advantages appear overstated and it should rather positively be stated that a segment based analysis can provide a fairly straightforward improvement of Gaussian illumination as well.
Clearly, the uniform illumination makes this more convenient and a larger field of view is available with high signal-to-noise ratio data.
2) A couple of consequences of the insertion of the beam-shaping device should be more clearly elaborated on: -how much does the insertion of the device reduce the peak intensities than can be reached? this could be based on comparison to the segmented analysis, by stating which segment has equivalent illumination intensity for same laser power and the intensity ratios for the other segments.
-does the slightly lower peak intensity that can be reached still allow maximal harvesting of photons from DNA-PAINT imagers in the scenarios tested?
-how does the device affect the HILO mode which is also of high interest, particularly for biological samples. The supplementary has images for this configuration but the main text lacks any comment on the uniformity/non-uniformity of the resulting distribution.
3) The binding time based analysis has a number of features that should be clarified (material relating to Supp Fig. 10 and corresponding text areas): -I found the terminology not particularly well chosen and hard to follow until I had unpacked what terms such as 'mean frame time' and 'pick' meant. Both seem to me non-standard and not very helpful without clearer definition.
-assuming a pick refers to a small, and often sparse ROI in the data, possibly selected in software via a feature of the resulting rendered super-resolution image. Then the analysis appears to relate to the events associated with this ROI or 'pick' and their distribution in time is analysed. The 'mean frame time' is then the mean of the time distribution of events, where the natural time unit of SMLM data is the frame number in which an event was detected. The width of this time distribution, measured by its standard deviation, is then the second measure used. I would find it useful if a clear plain word definition of the terms used (without reference to specific software etc) can be provided, as I have attempted here. -can such analysis be used for non-sparse regions in biological samples, e.g. within the areas covered by microtubules; it seems that the analysis suggested has a sparsity requirement that is well fulfilled with e.g. origami, but probably not with many biological samples -I briefly looked at the software linked from the text to do the pick-based analysis (https://github.com/DerGoldeneReiter/qPAINT). It was not clear which scripts where used in the set of files and directories in the distribution. Could an example with a test data set be provided for clarification?
The authors have significantly modified their original document and satisfied most of my original concerns. I have not objections to the publication of this manuscript, but would like to suggest a few more additional (minor) modifications: 1) As the authors pointed out in their response to my original comments, the approaches taken by Axelrod and Toomre and colleagues do not provide flat-fielding. Instead they are used to "even out" and "homogenize" TIRF illumination. To credit Toomre for this development, I would suggest adding the following citation to the references: Yang Q, Karpikov A, Toomre D, Duncan JS. 3-D reconstruction of microtubules from multi-angle total internal reflection fluorescence microscopy using Bayesian framework. IEEE Trans Image Process. 2011;20(8):2248-2259 2) I think it would be wise to prominently discuss the fact that the improved segmented analysis does not apply to the majority of biological samples, in order to avoid that some researchers who might have just casually read the manuscript begin to pursue this in their research. A statement similar to the one provided in response to reviewer 2: "These types of segmented analysis would work well e.g. in the case of compact, separable protein structures such as nuclear pore complexes, however will most likely fail to produce satisfactory results in the case of e.g. cytoskeleton or organelle structures." would be beneficial -either in the introduction or in the conclusions to the paper.