Advanced spinning disk‐TIRF microscopy for faster imaging of the cell interior and the plasma membrane

Understanding the cellular processes that occur between the cytosol and the plasma membrane is an important task for biological research. Till now, however, it was not possible to combine fast and high‐resolution imaging of both the isolated plasma membrane and the surrounding intracellular volume. Here, we demonstrate the combination of fast high‐resolution spinning disk (SD) and total internal reflection fluorescence (TIRF) microscopy for specific imaging of the plasma membrane. A customised SD‐TIRF microscope was used with specific design of the light paths that allowed, for the first time, live SD‐TIRF experiments at high acquisition rates. A series of experiments is shown to demonstrate the feasibility and performance of our setup.


Introduction
In the last 15 years, light microscopy underwent a profound revolution. Imaging with subwavelength-sized resolution of both fixed and living samples is now possible due to the flourish of different novel super resolution techniques, as are, for example, STED (stimulated emission depletion), PALM (photoactivation localisation microscopy), STORM (stochastical optical reconstruction microscopy) and SIM (structured illumination microscopy) (Hell & Wichmann, 1994;Gustafsson, 2000;Betzig et al., 2006;Hess et al., 2006;Rust et al., 2006;van de Linde et al., 2011). Moreover, high-resolution live imaging of macroscopic specimen became accessible after the implementation of novel approaches, that is SPIM (selective plane illumination microscopy) (Huisken et al., 2004;Keller et al. 2008). These imaging techniques benefit from technological developments that have strongly improved the sensitivity of detectors as well as the speed and capacity of data storage devices. All these progresses are extremely beneficial for the biological and biomedical field (Toomre & Bewersdorf, 2010;Eggeling et al., 2015;Sydor et al., 2015). Technological a.failla@uke.de development, however, is still ongoing and will further permit relevant improvements to the speed, performance and resolution of imaging systems. For example, fast super resolution imaging of living animals with subcellular resolution is a target that might be achieved in the near future.
Light microscopy progresses result to be also beneficial for the research focused on studying membrane-mediated interaction between the cell and its environment. This is extremely important since a wide variety of processes are regulated via the plasma membrane, for example endo-/exocytosis, adhesion, migration or signalling. The understanding of those phenomena could provide relevant progress in many biological and biomedical branches such as neurobiology, developmental biology, tumour biology and microbiology (Axelrod, 2008;Mattheyses et al., 2010).
However, membrane-mediated processes are still not easily accessible by light microscopy. There are two prerequisites for imaging such events. The first: it is essential to isolate and localise the cell membrane. The second: it is necessary to track the fast movement of subcellular organelles in the volume enclosed by the cell membrane. Despite most of the available super resolution methods allow three-dimensional (3D) imaging, only TIRF microscopy, due to the exponential decay of its signal, can specifically localise the ventral plasma membrane of adherent cells, by comparing the time variation of the signal arising from subcellular structures, for example vesicles approaching or leaving it, with the exponential decay curves shown in Figures S1(C), (D). However, the localisation of the plasma membrane should be combined with the tracking of subcellular organelles, like vesicles, moving between the plasma membrane and the cytoplasm. In other words, TIRF microscopy needs to be combined with fast 3D imaging, a condition still not achievable by super resolution microscopy techniques.
At the present state of the art, fast highly resolved 3D stacks can be provided by three different approaches. The first is wide field microscopy supported by deconvolution algorithms. It is simple, fast and cheap, can be easily combined with TIRF microscopy but, as all the deconvolution-based imaging, is not fully reliable for processing low signal-to-noise ratio acquisitions. The second is 3D super-resolution microscopy such as 3D SIM or lattice light sheet microscopy (Shao et al., 2011;Chen et al., 2014). Acquisition speed and resolution are high and thus might also be combined with TIRF in the future. This approach will prospectively provide the better signal-to-noise ratio without losing resolution. However, the implementation of this configuration appears to be complex. The third approach is spinning disk microscopy. This method provides high-speed, high-resolution and good signal-to-noise ratio, especially if compared with wide field microscopy.
The idea of integrating spinning disk and TIRF microscopy has already been published (Trache & Lim, 2009;Stehbens & Wittmann, 2014). However, none of these examples can be considered as a proof of principle for a setup thought to fulfil the demands illustrated in the previous paragraphs. In fact, in these microscopes, the spinning disk and the TIRF unit were placed onto two different detection paths, requiring long time for switching between them. The switching time is indeed the critical factor to prevent an efficient spinning SD-TIRF acquisition in a fast live imaging experiment.
In this work, we will show for the first time, according to our knowledge, a custom-designed microscope combining both dual channel spinning disk and TIRF microscopy. This imaging system is able to isolate the plasma membrane and, at the same time, to track events in the cell volume enclosing it. The key factor to achieve this goal was to use the same detectors for both SD and TIRF microscopy. This allowed us to perform fast sequential imaging switching between spinning disk and TIRF. As a proof of principle, we will show a series of examples of live cell movies, such as exocytosis, vesicle trafficking and cytoskeletal dynamics. Please note that the purpose of the experiments shown in this work is to demonstrate the capabilities of the setup without performing any quantitative analysis on the images. Thus, the description of imaging data is only qualitative.

Microscope setup
In Figure 1, the custom setup of the integrated spinning disk-TIRF (SD-TIRF) microscope from Visitron systems (Puchheim, Germany) is illustrated. The setup is composed of an inverted Nikon Ti microscope (Tokyo, Japan) connected to the scan units via two independent ports. The left port is linked to a spinning disk unit (CSU, Yokogawa CSU W1, Tokyo, Japan) in two camera configuration. The back port is connected to an iLAS2 scanner unit (TIRF, Roper Scientific, Martinsried, Germany) designed for selective TIRF illumination and photobleaching/conversion experiments. As excitation sources, six laser lines, namely 405/445/488/515/561/640 nm, are coupled into the system via multiple beam-paths emerging from a laser combiner unit. For excitation and detection, a Nikon Plan-Apochromat TIRF 100×/NA 1.45 oil objective is used. In the detection path, the emitted light is split inside the CSU scan head by one of two switchable dichroics, that is 561 nm long-pass and 514 nm long-pass. After passing through emission filters (band-pass centred at 525 nm, width 50 nm, and band-pass centred at 609 nm, width 54 nm, for green and red fluorescence, respectively) placed in front of each detector, the light is finally directed to two EM-CCD cameras (Photometrix Evolve, Tucson, AZ, USA, 16 µm sensor pixel size, 16 bit pixel depth, and image format 512 × 512 pixel). The final pixel size in the image is 85 nm. A fast piezo focus drive (Ludl, Hawthorn, NY, USA) was used for 3D image acquisition and to overlap the SD and TIRF datasets. The VisiView software was controlling the setup and run in streaming mode to ensure highest frame rates. The data acquired can be imported into ImageJ by the bioformats plugin, however, to generate combined SD-TIRF hyperstacks, an additional ImageJ macro was used. Images and movies in this manuscript were generated either by FIJI (Schindelin et al., 2012) (Figs. 2-4 and Movies S1-S3) in combination with the integrated plugins 'Orthogonal Views' and '3D Project . . . ', or by Volocity (PerkinElmer, Rodgau, Germany) ( Fig. 5 and Movies S4A-C). Vesicle tracking was performed with the 'Track objects manually' function in Volocity.
This microscope allows for fast simultaneous dual-channel spinning disk imaging followed by simultaneous dual-channel TIRF imaging. The fluorescence light emitted from the sample is either passing the dual spinning disks (SD mode, Fig. 1A) or bypassing them (TIRF mode, Fig. 1B). This configuration gives rise to three main advantages: first, the collected TIRF signal is maximised, since it is not reduced by the SD pinholes. Second, the system uses for the first time, according to our knowledge, the same detectors for SD and TIRF imaging, providing a perfect overlap of both TIRF and SD datasets. Thus, manual or software-based image alignment issues and distortions, which could be experienced in another custom built setup similar to (Trache & Lim, 2009;Stehbens & Wittmann, 2014), are avoided. Third, the acquisition speed is enhanced, because only two components are moved for rapid switching between SD and TIRF mode: the fluorescence turret and the SD unit. In more detail, the switching time of the turret between the empty position for SD imaging and the adjacent position housing a quad-band filter cube for TIRF imaging is 150 ms (Nikon, 2016). To move the SD in or out requires about 500 ms. No additional time is consumed by rapidly changing the illumination (<10 ms) within the laser combiner. Altogether this results in a possible minimum cycle time for single plane, dual-channel SD-TIRF image datasets of 1286 ms with 100 ms exposure for SD and TIRF channels, respectively. In comparison, the same dataset acquired in single camera mode will take 1906 ms. Thus, the SD-TIRF switching time is faster than changing the microscope detection path as reported in other setups (Stehbens & Wittmann, 2014).

Sample preparation
All cells were cultured in a 37°C, 5% CO 2 and humidified incubator. The following growth media and transfection methods were used: HEK293 and HeLa cells were grown in DMEM/10% foetal calf serum/penicillin and transfected with Turbofect (ThermoFisher Scientific, Waltham, MA, USA). N2a cells were grown in DMEM with 10% exosome-depleted foetal calf serum (Atlas Biologicals, Fort Collins, CO, USA) and the exosomes were purified as described previously (Falker et al., 2016). Exosomes were PKH26-labelled (Sigma Aldrich, #MINI26-1KT, St. Louis, MO, USA) and measured to be 132 nm in diameter. Peripheral human macrophages were isolated from buffy coats, cultured as described previously (Linder et al., 1999) and transfected with the Neon R electroporator (Invitrogen, Carlsbad, CA, USA). Transient cell transfection was performed 1 day (N2a, HeLa, macrophages) or 3 days (HEK293), respectively, before the experiment. The following plasmids were transfected: GFP-Snx27, RFP-β1AR, GFP-CLN3, KIF9-GFP  and alpha-tubulin-mCherry. As outlined in the experimental procedures, Lysotracker R Red DND-99 (Molecular Probes, Eugene, OR, USA) was added at 50 nM concentration 30 min before and 1 µM isoproterenol was added directly before starting the acquisitions. Cells were seeded on poly-L-lysine coated no. 1.5 glass bottom dishes (MatTek, Rochester, MN, USA). Imaging was performed in growth medium on a stage top incubator (Okolabs, Pozzuoli, Italy) at 37°C and 5% CO 2 .

Single-colour spinning disk and TIRF imaging can be combined in a single dataset
In a first experiment, the interaction of fluorescently labelled exosomes with neuronal cells was investigated. As transmitting agents, exosomes seem to play a major role in the spread-ing of neurodegenerative diseases (Kalani et al., 2014). A day before the experiment, exosomes were purified from the neuronal cell line N2a and then marked with the red fluorescent dye PKH26. Before starting imaging, N2a cells were allowed to endocytose labelled exosomes for 90 min at 37°C at a ratio of circa 50 particles per cell. The dynamic redistribution in the cell was then recorded with SD-TIRF microscopy. A z-stack covering a 14 µm thick volume (35 z-planes with 0.4 µm spacing) was necessary to visualise the entire distribution of exosomes in the cell ( Fig. 2A). However, only by means of TIRF microscopy it was possible to see when and where there was a close contact of vesicles with the ventral plasma membrane. The time lapse sequence in Figure 2(C) (for a movie see Movie S1) clearly showed the contact of a single intracellular exosome with the plasma membrane (arrowhead, exosome was also visible in the cyan-coloured TIRF channel) for more than 360 s, when  it suddenly detached and lifted off. At t = 380 s, no signal was present in the bottom plane, neither TIRF nor SD, but the XZ projection showed the detached exosome. Several other exosomes in the cell frequently came into contact with the plasma membrane. This was proved by the detection of their signal in the TIRF channel. In fact, SD alone would not have been enough to precisely locate the exosome in the axial direction, as shown exemplary for two vesicles in Figure 2(B). Indeed, the evanescent field produced by the TIRF illumination penetrates only about 190/100 nm (experimental overestimate/theoretical estimate) into the cell, while the depth of field of a single SD-plane is circa 780/500 nm (experimental overestimate/theoretical estimate) (for more details regarding SD-TIRF experimental estimates of the point spread functions, please refer to Fig. S1 in the supplementary online material).

Dual-colour SD-TIRF imaging reveals spatiotemporal interactions of vesicles at the plasma membrane
The mutual interactions between two fluorescently labelled structures and their correlation with the plasma membrane was studied taking advantage of dual-colour SD-TIRF imaging. For subsequent experiments, two SD and two TIRF acquisition channels were set. The first SD/TIRF channel was configured for detecting GFP (SD-488/TIRF-488), while the second SD/TIRF channel enabled the detection of red fluorophores such as RFP or Lysotracker Red DND-99, respectively (SD-561/TIRF-561). A z-stack of 5.2 µm (13 z-planes with 0.4 µm spacing) was a reasonable compromise to cover most of the fluorescent structures while maintaining a high acquisition speed. In the first example, HEK293 cells were transfected with plasmids coding for GFP-Snx27, known to be involved in the recycling of cell surface receptors, and RFP-β1AR, the β1 adrenergic receptor (Nakagawa & Asahi, 2013). After stimulation with the β1AR-agonist isoproterenol, cells were investigated at the microscope. From a perinuclear pool of RFP-positive vesicles, a single vesicle could be followed while migrating to the cell periphery and touching the plasma membrane (Fig. 3A, arrowhead, SD-561 and TIRF-561 channel of magnified view in Fig. 3B, and Movie S2). The β1AR-vesicle was moving in proximity to the plasma membrane until it colocalised with a GFP-positive Snx27-vesicle (Fig. 3B), which was only visible in the SD-488 but not in the TIRF-488 channel. Upon temporal colocalisation, the β1AR-vesicle left the plasma membrane (no more signal in the TIRF-561 channel), but remained visible for about 20 s in the SD-561 channel until it also dissociated from the Snx27-vesicle and returned to the perinuclear pool.
In the second example, HeLa cells were transfected with GFP-CLN3, an endosomal/lysosomal glycoprotein, and stained for lysosomes with Lysotracker Red DND-99. From the majority of lysotracker-positive CLN3-vesicles, approximately 20% were also detectable in the TIRF-561 channel (see Fig. 4A and Movie S3). Here, the dynamic interactions of CLN3positive lysosomal vesicles have been studied, as depicted in the time lapse sequence in Figure 4(B).
In a zoomed area, two vesicles, marked with a white and a yellow arrowhead, were visible from time point t = 93 s on. One of the two vesicles (yellow arrowhead) was more distant from the plasma membrane, as indicated by the dimmer signal in the SD channels and the absence of the signal in the TIRF channels. We observed temporal colocalisation of several vesicles, when between t = 107 s and t = 116 s the signal increased in all channels and only a single vesicle was detectable. Further and more detailed studies might determine if this signal superposition is the result of a very close interaction, for example fusion. Upon separation or fission, two vesicles remained visible until t = 135 s in short distance to each other, but only one of them in proximity to the plasma membrane (white arrowhead, TIRF channels). Subsequently, both vesicles started to disappear from the plasma membrane. This was evident by watching at the white arrowhead marked vesicle in the XZ reconstruction of time points t = 149 s and t = 154 s in Figure 4(C).

SD-TIRF imaging enhances the spatiotemporal resolution of the dynamic interaction of vesicles with the plasma membrane and the cytoskeleton
It was demonstrated that GFP-KIF9 vesicles move along microtubules, regularly contacting podosomes (Cornfine et al., 2011), actin-rich structures that are involved in matrixdegradation and visible in the ventral site of adherent monocytic cells such as macrophages. To further elucidate the dynamic interplay of KIF9 with microtubules and the plasma membrane, SD-TIRF imaging was used. To do so, GFP-KIF9 was cotransfected with mCherry-alpha-tubulin in primary human macrophages. The GFP-signal was detected in the SD-488 and TIRF-488 channels, while the mCherry signal was displayed only in the SD-561 channel. A finer sampled 2.75-µm z-stack (11 z-planes with 0.25 µm spacing) was necessary and sufficient to resolve most of the peripheral microtubules and KIF9-vesicles albeit maintaining a high temporal resolution. The still image shown in Figure 5 and time lapse sequence (Movie S4A) clearly show temporal overlap between the signal arising from vesicles and microtubules that suggests the colocalisation of vesicles with microtubules. A single vesicle was tracked to reveal its position in all spatial directions over time ( Fig. 5B and Movies S4B and S4C).
The XZ projection in Figure 5(B) clearly demonstrated that the vesicle was very dynamically moving in the axial and lateral directions. Intensity fluctuations visible in the TIRF-488 channel show repetitive contacts of the vesicle with the plasma membrane, possibly at podosomal sites. Please remember that the TIRF-signal of any isolated vesicle is directly correlated with its distance to the plasma membrane (Mattheyses et al., 2010). In details, the brighter is the fluorescence the closest the vesicle is to the plasma membrane (see also Fig. S1B), as confirmed by the axial position of the vesicle in the SD-488 channel in Figure 5(B).

Discussion
In this paper, we showed a greatly improved model of a SD-TIRF microscope and its application in a range of live imaging experiments. Although similar systems have been published before, there were two main disadvantages that might have hampered their application in biological experiments: first, image acquisition was too slow due to the light path design. Second, a perfect overlay between SD and TIRF images was not possible to be achieved, due to the separate detection paths for SD and TIRF modes that cause shifts and distortions. Hence, we decided to design the light path of our microscope to overcome those limitations. The Yokogawa CSU-W1 spinning disk unit has the possibility to move the dual-disk in and out of the detection path; this is a great benefit for the collection of the TIRF signal, which would otherwise be reduced by the pinholes. Additionally, the same cameras could be used for the detection of the SD and TIRF signal. The result was a highprecision overlap of the field of view of the wide field TIRF and laser scanned spinning disk images (checked by imaging fixed samples, see Fig. S2) as well as fast sequential acquisition of the two imaging modes. Running the Roper iLAS² scanner in so-called time-sharing mode allowed us to detect two TIRF channels simultaneously, speeding up the acquisition of complex, multichannel data. Indeed, the speed of the system is currently limited only by the movement of the dual-disk inside the scan head. This setup configuration can be easily improved by taking advantage of technological developments. A faster motor or the possibility to circumvent the disk movement, for example through bypassing the disk with galvanometer mirrors, would further reduce the switching time between SD and TIRF mode.
We demonstrated that, with our microscope, we were able to visualise the temporal overlapping of biological structures and to follow them in live imaging experiments (see . Please note that, in the previous mentioned experiments, we sacrificed imaging quality (high signal-to-noise ratio) to privilege noninvasive imaging conditions, i.e. negligible bleaching and reduced phototoxicity. Moreover, the signal-to-noise ratio can be improved easily by implementing a new generation camera with more sensitivity in the current setup. We were profiting from the fact that SD and TIRF microscopy are two imaging methods that are especially suited for live imaging, due to their low phototoxicity and high acquisition rates. Combining SD and TIRF datasets allowed 3D reconstructions of fast intracellular processes with high resolution of the bottom plane. This is of great importance, if interactions of subcellular organelles with the plasma membrane are investigated. SD imaging alone is too limited in the z-resolution and would not be able to reveal such contacts (see Fig. 2B). TIRF imaging allowed this differentiation. In fact, TIRF allows the specific localisation of a 100-200 nm excitatory zone above the coverslip from the residual volume of the cell (Axelrod, 2008)a compartment known to comprise the plasma membrane. Furthermore, recent approaches with multiangle TIRFM (Fu et al., 2016) have increased the axial resolution down to about 20 nm and could be implemented in our microscope.
Our setup is easy to build up and readily available to the scientific community. The combination of SD and TIRF in a single system designed, as shown in this paper, will encourage researchers to gain insight into cellular mechanisms that could not be unveiled before.

Supporting Information
Additional Supporting information may be found in the online version of this article at the publisher's website: