Simultaneous 3D super-resolution fluorescence microscopy and atomic force microscopy: combined SIM and AFM platform for cell imaging

Correlating data from different microscopy techniques holds the potential to discover new facets of signaling events in cellular biology. Here we report for the first time a hardware set-up capable of achieving simultaneous imaging of spatially correlated super-resolution fluorescence microscopy and atomic force microscopy, a feat only obtained until now by fluorescence microscopy set-ups with spatial resolution restricted to the Abbe resolution limit. We hereby remove the need to perform independent measurement and subsequent data averaging required to eliminate cell-to-cell variation in observed signals. We detail system integration, demonstrate system performance and report imaging of sub-resolution fluorescent beads and genome-engineered human bone osteosarcoma epithelial cells.


Introduction
Visualization of biological specimens can be accomplished through several available microscopy techniques, each of them presenting inherent benets and constraints. The combination of dierent imaging platforms can further extend their applications [1,2]. The major signicance of correlative imaging relies on its capacity to provide complementary information such as morphological, chemical and biophysical details from the same sample during a single experiment.
Atomic Force Microscope (AFM) was introduced in 1986 by Binning et al. [3], being a high-resolution surface probe technique that allows topographical and nano-mechanical characterisation. AFM has become a powerful tool to team up with optical microscopes, especially regarding the biosciences eld.
On the other hand, super-resolution uorescence optical microscopy methods have been a major breakthrough for biological samples examination, having being actively developed in the latter years to surpass the optical diraction barrier [4]. Nonetheless, they are limited in some aspects such as for example nanomechanical sample characterization. Therefore, a correlative microscopy approach is benecial as it allows to pair up the high spatial resolution and mechanical characterization potential of an AFM, with the additional molecular information available, through various uorescent markers or immunostains [5,6].
Despite the great potential on simultaneous operation of AFM and superresolution uorescence microscopy, no one to the best of our knowledge has reported it. Current AFM techniques combining optical uorescence imaging normally perform successive measurements and then superimpose the images obtained from the separate acquisitions. Main reasons for this are: i) imaging in a simultaneous mode causes perturbation on the AFM cantilever operation due to uorescence excitation light, and ii) noise transfer from the optical microscope can disturb AFM measurements [7,8].
The integration of AFM with uorescence super-resolution schemes faces a variety of problems impeding simultaneous data acquisition. For instance, high-powered light sources interact strongly with gold coated AFM cantilevers, causing excessive heating or even cantilever coating deterioration.
High powered light sources are commonplace for the depletion beam of a STED set-up as well as accelerators for a fast read-out in STORM microscopy.
Furthermore, STORM microscopy requires uorophores exhibiting blinking behaviour which is typically promoted by immersing the sample in a buer containing an enzymatic oxygen scavenger. The latter precludes simultaneous operation of the AFM and uorescent measurements, as the ingredients of the buer stick to the AFM cantilever, so correlative imaging is normally performed by acquiring rst the AFM image and then adding the buer for super-resolution microscopy [16,19]. Recent research has proposed a new dye to circumvent this problem that allows correlative AFM and STORM imaging without the need to change the buer [18].
Here, we report on a new super-resolution microscopy platform that allows to perform simultaneously uorescence microscopy and atomic force microscopy. In this context, we combine AFM with Structured Illumination Microscopy (SIM), a super-resolution technique that allows to surpass the diraction barrier by a factor of two in every spatial direction [20]. Contrary to other super-resolution modalities, SIM relies on a uorescent excitation light pattern favourable to avoid AFM cantilever disruption during simultaneous operation, as we reported previously on a novel system for synchronic uorescence optical sectioning microscopy through aperture correlation microscopy using a Dierential Spinning Disk (DSD) and AFM [8].
The principle of SIM relies on the projection of a grating onto the sample and the subsequent reconstruction of the high-resolution image from dierent grating imaging positions. This technique has been used for biological imaging, in particular for xed samples, but it has also proven to be easily applicable to live cell imaging [21,22]. SIM is a very promising imaging modality to reveal dynamic processes in live biological samples in 3D and presents some advantages over other super-resolution techniques as it does not require the use of specialized uorophores or sample preparations, in-duces less phototoxicity and can reach higher acquisition rates for longer periods of time as compared to PALM/STORM and STED [23]. Because SIM requires a relatively low illumination power, its integration with AFM is very appropriate for non-disruptive simultaneous operation. Aside, as SIM is subject to suer from artifacts during image reconstruction process [24,25], a combination with AFM can help to validate super-resolution data.
To demonstrate system performance, we rst image a sample consisting of sub-resolution uorescent beads. SIM-AFM simultaneous operation is then explored using CRISPR/Cas9 genome-edited human cells expressing a uorescently tagged plasma membrane transporter from their native genomic loci. This model is particularly relevant for super-resolution microscopy techniques as it avoids the typical overexpression-induced artefacts that often aects protein localization and dynamics [23]. The AFM was operated in an advanced force-spectroscopy based mode, referred to as Quantitative Imaging (QI), which allows nanomechanical characterisation and simultaneous imaging [26]. For static measurements on beads we used force modulation cantilevers (FM, NanoWorld, Switzerland), with a nominal resonance frequency of 75 kHz in air, spring constant of 2.8 N m −1 , reective detector gold coating, and monolithic silicon pyramidal tips with radius of curvature (ROC) of 8 nm. For measurements on cells in liquid, we used qp-BioAC-CI-CB1 cantilevers (NanoSensors, Switzerland), with a nominal resonance frequency of 90 kHz (in air), spring constant of 0.3 N m −1 , partial gold coating on the detector side, and quartz-like circular symmetric hyperbolic (double-concaved) tips with ROC of 30 nm. Figure 1: Simplied schematic set-up of the SIM-AFM system. In this case, structured illumination is generated using a grating (G). Collimated and expanded laser light illuminates the grating, resulting in the diraction of multiple orders. Only 0 th and ±1 st orders are allowed into the illumination path and focused on the back focal plane of the objective and the created stripped illumination pattern excites the sample. Collection of the uorescence signal is achieved by a high aperture microscope objective, a dichromatic mirror (DM) and a tube lens (TL). The AFM laser is reected by a mirror (M) to a highdynamic range camera coupled to a microscope port for cantilever laser spot alignment. At the same time, nanomechanical mapping of the sample is performed by a probe consisting of a exible cantilever and a tip. After simultaneous SIM and AFM images acquisition, they are combined into a single image.   fuorescence 72 h after electroporation. Clonal populations were isolated and characterized, as previously described [29]. Sequencing revealed a one nucleotide sequence variation, which does not alter the protein or intron sequences.

Image registration
The simultaneous operation of AFM with optical microscopy enables the collection of optical sectioning uorescence and nano-mechanical mapping information from a sample. However, one complication of SIM is that, in order to avoid artifacts in the nal image, requires of optimized experimental implementation, consideration of bleaching properties of the sample [30,31] and proper selection of reconstruction parameters [32,33], as SIM has a need of a complex post-processing step. Hence, coupling of AFM and SIM can also be a powerful tool to validate the results obtained with the latter.
The successful combination of AFM with optical images is not a straightforward process. Lenses in microscopes exhibit optical aberrations that distort the image, while AFM registers real-space images using highly linear piezo-electric elements [34]. To circumvent this problem and correctly ovarlay SIM data onto AFM images we use a software module (DirectOverlay, JPK BioAFM, Bruker Nano GmbH, Berlin, Germany) to calibrate the optical image. In this calibration process the cantilever is displaced to a set of predened coordinates in a 3 × 3 or a 5 × 5 grid pattern, registers an optical image at each position and a transform function between the AFM and the optical image is determined [8].

Results and Discussion
At rst, we evaluated the spatial resolution improvement of the structured illumination system (Figure 3). To this end, we used both uorescent beads Once it was veried that it is possible to work below the diraction limit, to represent the AFM images by using a pixel dierence Z-scale lter of some sort. In this particular case, we have treated the AFM height image in Figure   4f with a curvature function [36] that highlights the curvature in the image by calculating the local mean curvature operation A for each pixel: where R 1 and R 2 are the smallest and the largest radii determined at each pixel.
Correlating dierent types of datasets, such as optical and AFM data bodes a dierent set of questions and problems about the eciency of multidimensional data representation. On one side, it is necessary to consider that both SIM and AFM techniques carry by denition very dierent types of information, interpreted by the dierent operation principles and detection methods. This is further complicated by the dierent spatial and axial resolution of both techniques, which can dier in the range of up to 2 (spatial) 300 nm are now possible [37], the Z-scaling (not resolution) in conventional AFM images is typically conned to the overall sample morphology. A cer-tain solution to this problem is the application of force-spectroscopy based imaging (force-controlled AFM), which allows us to interpret the 3D-Force volume data of the acquired AFM dataset, i.e. each pixel has a force curve associated, and analytically subtract dierent quantitative information [38]. In

Conclusion
A 3D SIM super-resolution system has been successfully integrated with a tip-scanning QI TM nanomechanical mechanical mapping AFM for simultaneous SIM-AFM operation. The combined SIM-AFM integration was tested on 100 nm uorescent spheres and xed osteosarcoma cells, demonstrating a good 2D spatial correlation of both signals. The spatial resolution obtained in both cases was approaching the theoretical SIM diraction limit. We also give an example for multidimensional representation of correlated microscopy sets, by introducing section of the 3D AFM force-volume sets. Finally, the The simultaneous operation of AFM and super-resolution uorescence microscopy technique provides a powerful observational tool on the nanoscale, albeit data acquisition is obstructed by a series of integration problems. We believe that the combination of SIM with AFM presents one of the most promising schemes enabling synchronous imaging, allowing the recording of Figure 6: Simultaneous SIM/AFM acquisition. The AFM measurements were carried out on xed USO2 cells in medium/buer with (a) and without N-SIM illumination (b). For convenience and enhanced feature/noise contrast, both AFM topography images in the SIM-AFM overlays are displayed with an edge detection algorithm using a pixel dierence operator in X. The topography images from petri dish surface on three positions (labelled in the gures) were planet (1st order polynomial function) to compensate for tilts in the sample surface, and subjected to surface roughness analysis (c). For comparison reasons, the average roughness (Ra), RMS roughness (Rq) and peak-to-valley roughness (Rt) values are given below the corresponding height proles.
nanomechanical data and cellular dynamics visualization at the same time.

Conict of interest
There are no conicts to declare.

Acknowledgements
Ana I. Gómez Varela wishes to acknowledge support from the Xunta de