A transparent waveguide chip for versatile TIRF-based microscopy and nanoscopy

Total internal reflection fluorescence microscopy (TIRF) has enabled low-background, live-cell friendly imaging of cell surfaces and other thin samples thanks to the shallow penetration of the evanescent light field into the sample. The implementation of TIRF on optical waveguide chips (c-TIRF) has overcome historical limitations on the magnification and field of view (FOV) compared to lens-based TIRF, and further allows the light to be guided in complicated patterns that can be used for advanced imaging techniques or selective stimulation of the sample. However, the opacity of the chips themselves has thus far precluded their use on inverted microscopes and complicated sample preparation and handling. In this work, we introduce a new platform for c-TIRF imaging based on a transparent substrate, which is fully compatible with sample handling and imaging procedures commonly used with a standard #1.5 glass coverslip, and is fabricated using standard complementary metal-oxide-semiconductor (CMOS) techniques, which can easily be scaled up for mass production. We demonstrate its performance on synthetic and biological samples using both upright and inverted microscopes, and show how it can be extended to super-resolution applications, achieving a resolution of 116 nm using super resolution radial fluctuations (SRRF). These new chips retain the scalable FOV of opaque chip-based TIRF and the high axial resolution of TIRF, and have the versatility to be used with many different objective lenses, microscopy methods, and handling techniques. We thus see c-TIRF as a technology primed for widespread adoption, increasing both TIRF's accessibility to users and the range of applications that can benefit from it.


Introduction
A common trade-off in microscopy is between resolution, speed, and photodamage or photobleaching. Total internal reflection fluorescence (TIRF) microscopy, however, is a rare example of a method which improves all three simultaneously, compared to confocal microscopy. In TIRF, only the bottom ~100 nm of the sample is excited 1-4 , which improves the axial resolution, eliminates outof-focus light, and protects the bulk of the sample from photodamage. Additionally, because TIRF is a widefield technique, acquisition of a full image can occur within milliseconds. TIRF's ability to focus exclusively on a thin layer at the surface of the cell have made it an excellent tool for studying, among others, the dynamics of focal adhesions 5 , the inner workings of endocytosis 6 , the kinetics of cell surface receptors 7 , and docking of synaptic vesicles with neurons 8 .
The enabling mechanism of TIRF is the generation of a thin, exponentially decaying layer of light at a surface, called the evanescent field. When light is directed into an interface between media with a high index of refraction contrast (HIC) at a sufficiently high angle, the light is totally reflected within that interface; while the light itself does not escape from the high index material, an evanescent field is generated along the surface that the light travels 2,9 . The angle necessary for TIRF has traditionally been achieved using a high numerical aperture (NA) objective lens 1,10,11 , through a prism 1,4,11 , through the use of grating couplers 12-15 and more recently by coupling into the side facet of optical waveguide chips [16][17][18][19][20][21] . These photonic chips are fabricated using technology similar to computer chips, and thus have the potential to be mass produced at low cost. While objective lens-based TIRF is restricted to using a high NA lens, thus limiting its field of view (FOV), the evanescent field generated by waveguides is independent from the imaging pathway, enabling them to be used with any imaging objective on a standard microscope 12,18,20 . Furthermore, waveguide chips have been extended to super-resolution modalities, including single molecule localization microscopy (SMLM) 18,21 , entropy based superresolution imaging (ESI) 19 , and structured illumination microscopy (SIM) 21 . In addition to achieving a sub-diffraction localization precision of 72 nm, chip-based SMLM was able to do this with an unprecedented FOV of 0.5x0.5 mm 2 , approximately 100 times larger than objective lens based techniques 18 . Similarly, chip-based TIRF-SIM has surpassed the 2X resolution enhancement of conventional SIM, achieving a 2.4X resolution enhancement due additional benefits from the fringe pattern being generated in a high index material instead of free space 21 .
Thus far, most chip-based microscopy has been performed on waveguides fabricated on top of opaque substrates 12,16-18,20,21 . With the sample sitting on top of the waveguide, the chips can therefore only be used with an upright microscope (Fig 1a) because the emitted light cannot pass through the opaque substrate to the objective of an inverted microscope. On an upright microscope, however, the light must travel from the bottom of the sample where it is emitted, through the rest of the sample (which may be complex and highly scattering) and its surrounding media, and typically through a glass coverslip to the objective. This translates to aberrations in the image, not because of inferior optics but because of the scattering of light from the sample itself and from index mismatches from the medium in which cells are kept (Fig. 1a). This problem is even more pronounced for super-resolution techniques, for example in SMLM, where aberrations limit both the localization precision of the single molecule and the resulting resolution enhancements, and in SIM where these aberrations can result in additional reconstruction artefacts.
Additionally, opaque chips have several practical handling difficulties, particularly with their use on an upright microscope 11, 22 . Typically, a thin PDMS frame is placed on the chip around the specimen to contain the media, and a glass coverslip seals the chamber from the top during imaging; while this limits evaporation from the sample, it precludes the addition of reagents or the use of micromanipulation tools during imaging, the low volume can be problematic for live cell imaging, and the chamber often induces tilt in the coverslip, leading to additional image aberrations. Unless the objective is designed for water, the media in the chamber induces yet more aberrations from index mismatch. Furthermore, the opacity of the chip makes it impossible to check on the sample using a basic brightfield or phase contrast microscope, so the confluency and health of cells cannot be checked prior to labelling and imaging.

Design of transparent waveguide chip
To ensure the highest compatibility with a wide variety of microscopes, the waveguide chips should closely match the properties of a conventional glass coverslip, including the transparency, thickness, and index of refraction (RI). Additionally, the material must be compatible with the waveguide fabrication processes, have low surface roughness, and ideally be relatively inexpensive and readily available. The borosilicate glass used for traditional coverslips is unfortunately incompatible with the higher-temperature fabrication steps (such as annealing), and has a somewhat high propagation loss 26 . While sapphire is both transparent and compatible with high-temperature processes, it is expensive and has an RI significantly different from that of coverslip glass (nsapphire = Fabrication of these waveguides is particularly tricky due to the fragility of the thin (180 µm) fused silica wafer. Handling procedures must be even more stringent because both sides of the substrate should be clean and free from scratches, as the top side hosts the optical waveguides and the emitted light will pass through the bottom side on an inverted microscope. Standard photolithography processes and ion beam etching were used to fabricate tantalum pentoxide strip waveguides on fourinch wafers, and the steps are shown in Fig. S1a and b with additional details provided in the methods section. The wafers were diced into individual chips during the back end processing, which is done in a non-cleanroom environment, and were temporarily bonded with a thick (1 mm) glass wafer (Fig. S2) to protect them from breaking during this process. This provided mechanical stability from the pressure during wafer dicing and polishing, and could be safely removed without damaging the waveguides afterwards, without adding significant cost to the process. Finished chips on both opaque silicon and transparent fused silica substrates are shown in Fig. 2a and b. For the transparent substrate, waveguides of varying widths were examined by scanning electron microscopy (top view and cross section shown in Fig. 2c and d) and showed no visible fabrication defects.
As the roughness of the substrate and the deposited Ta2O5 has both a strong impact on the waveguide propagation losses and can influence the aberrations when imaging on an inverted microscope, we used an atomic force microscope (AFM) to examine the roughness both before and after deposition (Fig. S3). While the transparent silica substrate was rougher (rms = 0.460 nm) than a standard oxidized wafer (rms = 0.178 nm), both were significantly less rough than a glass coverslip typically used for microscopy (rms=1.5 -2 nm) 34 . After sputtered deposition of 250 nm of Ta2O5 onto the substrates, we measured the new rms surface roughness values to be 1.061 nm for the transparent substrate and 0.632 nm for the standard oxidized substrate; these measurements suggest that the roughness of the base substrate will impact the roughness after subsequent deposition steps, and ultimately will affect the final waveguide losses.
The difference between the RI of the waveguide and the surrounding medium dictate the strength of the evanescent field, thus it was important to measure any changes that occurred in the material during the fabrication process. The RI of the Ta2O5 film deposited on the fused silica substrate was measured in the visible spectrum using a spectroscopic ellipsometer and compared with the same film deposited on an opaque substrate ( The final quality control check before the waveguides could be used for imaging was to measure the propagation losses using scattering analysis 35 . The propagation loss for narrow (5 µm wide) Ta2O5 waveguides on the transparent substrate was found to be 6.6 dB/cm at 660 nm, which is notably higher than for similar waveguides on opaque substrates (3.7 dB/cm). This higher loss is mainly related to the Ta2O5 film quality, which depends on the initial substrate surface quality, and in the future can be reduced by improving the substrate roughness. For wider waveguides (>50 µm width), the propagation losses were on the order of 1 dB/cm; these lower losses are because the modes of the propagating light have less overlap with the roughness of the waveguide sidewalls. As a wider waveguide translates to a larger field of view, all of the c-TIRF imaging reported here used these wider, lower loss waveguides (200 -500 µm), which are still narrow enough to ensure sufficiently high power density for imaging.

Fluorescence imaging using waveguides on a transparent platform
As our transparent platform provides imaging capability with both upright and inverted configurations, we have developed an experimental setup that integrates both these configurations To allow for a large illuminated FOV we chose to use very wide waveguides, e.g. 200 -500 µm wide, which will simultaneously guide several modes inside. The interference of these modes creates a wavy, uneven illumination pattern, which can generally be evened out by translating the coupling objective lens with respect to the input facet of the waveguide over time, but the lack of certain modes can lead to stripe-like artefacts as seen in some of the images. However, this uneven illumination pattern can be exploited for fluctuation-based super-resolution imaging, and because of the high RI of the waveguide material the illumination patterns can contain higher spatial frequencies than with conventional free-space optics 2 , potentially leading to even higher resolution.
While TIRF platform is biocompatibility. While the growth of the HeLa cells on these chips already confirms the biocompatibility shown previously for Ta2O5 waveguides 18 , live HeLa cells labelled with SiR-Actin were additionally imaged using the inverted setup as a demonstration (Fig. S6). In the future, however, a stage-top incubation system will be necessary to ensure long-term viability.
Next, we have demonstrated the implementation of a high-NA oil immersion objective lens (60X 1.49 NA) with c-TIRF in the inverted configuration; due to its short working distance it was not possible to image in the upright configuration using this objective. We used immersion oil with an RI of 1.46 (refractive index liquid, Cargille Labs) to match the RI of the fused silica substrate and reduce spherical aberration, with further fine-tuning using the correction collar of the objective lens. Beads imaged using this 1.49 NA oil objective lens (Fig. 4c) are better resolved than with the 1.2 N.A. water immersion objective lens (Fig. 4b), and the smaller depth of focus helps to reject the light purple) shows two clearly resolved actin fibers, the c-TIRF profile (Fig. 6e, green) shows only a slight separation between the same two fibers. For the second actin bundle (Fig. 6f), c-TIRF shows only one large fiber while SRRF has the resolution to distinctly resolve two separate fibers. The overall resolution enhancement is further confirmed with Fourier ring correlation (FRC) 36 , which measured the resolution in the SRRF image to be 116 nm (Fig. S8). To the best of our knowledge, this is the first demonstration of super-resolution waveguide-based TIRF on a transparent substrate.

Discussion and Conclusions
Chip-based TIRF provides significantly improved versatility compared to traditional TIRF techniques by unlocking the restriction of the objective lens and decoupling of the excitation and emission pathways.
Additionally, the inherent variation in the illumination pattern in broad waveguides makes the c-TIRF platform a natural partner for fluctuation-based super-resolution techniques, and the high RI of the waveguide enables even smaller patterns to be generated for fluctuations than is possible with freespace optics. As a proof-of-concept, we have shown 116 nm resolution using SRRF on a transparent waveguide, but this can easily be extended to additional super-resolution techniques such as ESI, direct stochastic optical reconstruction microscopy (dSTORM), or superresolution optical fluctuation imaging (SOFI). These techniques can also be applied to images taken with a lower magnification objective; while the resulting resolution may be above the typical ~250 nm 'super-resolution' limit, it will still be significantly improved compared to both widefield imaging with the same objective or reconstructions of those widefield images using these same fluctuation-based algorithms because the evanescent field from c-TIRF prevents the blurring caused by out-of-focus light. The use of CMOS-compatible fabrication for these chips means production can be scaled up for eventual mass production at a low cost, allowing them to be used in many more labs and environments than traditional TIRF. Because of the decoupling of the emission pathway from the TIRF excitation, these chips could be combined with more durable cameras and collection optics, enabling them to be directly used in harsher environments, such as inside a humid incubator, under water, or at extremely high temperatures. Alternatively, because the fabrication is CMOS-compatible, chips could conceivably be designed to contain the waveguide, integrated filters, camera, and an LED light source as a single, integrated CMOS-based device. Additionally, as waveguide technology allows light to be sent along an innumerable variety of complex paths not possible in free-space, we see natural extensions for future designs to facilitate e.g. FRAP, photoactivation, or phototaxis along specific patterns. Our aim was to make TIRF microscopy significantly more user-friendly and accessible, and as this promising technology is increasingly adopted we anticipate seeing even more applications in fields where TIRF was previously impractical.

Competing interests
B.S.A. has applied for two patents for chip-based optical nanoscopy. B.S.A. is a co-founder of the company Chip NanoImaging AS, which commercializes on-chip super-resolution microscopy systems.       image an optical resolution of 116 nm was measured (Fig S7). For averaged c-TIRF and SRRF images, 1500 and 500 frames were used respectively and each of these frames was acquired with 50 ms exposure time. The scale bar in (a -c) is 25 µm and in (c -d) is 6 µm.

Methods
Waveguide fabrication front end process (Fig. S1a) 250 nm thick Ta2O5 film was deposited using RF magnetron sputtering directly onto 4" diameter wafers of 180 µm thick transparent fused silica or of 1 mm thick silicon (with 2.5 µm SiO2). The base pressure of the deposition chamber was kept below 1 x 10 −6 Torr with Ar:O2 flow rates of 20 standard cubic centimetre per minute(sccm) : 5 sccm. A substrate temperature of 200 • C was maintained throughout the deposition time. Ta2O5 films were deposited with a high deposition rate of 2.2 nm/min as determined using a stylus profiler (KLA Tencor P-16+ model). Photolithography was used to create a photoresist mask for further dry etching to fabricate channel waveguides. First, a 1.3 µm thick layer of positive resist (Shipley S1813) was spin-coated on top of the 250 nm Ta2O5 film and then prebaked at 90°C in an oven for 30 minutes. Second, the prebaked photoresist was exposed under a hard chrome mask with waveguide patterns using a mask aligner (MA6, Carl Suss). The time needed to properly expose the photoresist differs depending on the substrate used (based on whether it is reflective, transmissive or absorptive), therefore the exposure time was optimised separately for the transparent substrate (5.8 seconds at 20 mW/cm 2 ) and opaque substrate (5.6 seconds at 20 mW/cm 2 ).
Finally, the exposed patterns were developed using MF-319 developer for 58 seconds.
The dry etching was performed in an ion beam system (Ionfab 300+, Oxford Instruments) using Argon gas with a flow rate of 6 sccm and Trifluoromethane (CHF3) gas with a flow rate of 12 sccm to fully etch the Ta2O5 waveguides. The process beam voltage (500 V), beam current (100 mA), RF power (500 W) and substrate temperature (15°C) were kept constant throughout etching process. In the ion beam milling process, the substrate was placed at an angle of 45° with respect to the incident ion beam to achieve low sidewall roughness. The processed wafer was subjected to plasma-ashing with oxygen gas for 20 minutes to remove the photoresist. After that, the wafers were placed in a 3-zone tube furnace for annealing at 600°C (ramp rate of 3°C/min to 525°C, ramp rate of 2°C/min from 525°C to 600°C) in an oxygen environment (2 liter/min) for 3 hrs to reduce the stress and repair the oxygen deficiency created in Ta2O5 during the fabrication process 37 .
Waveguide fabrication back end end process (Fig. S1b) Photoresist (MICROPOSIT®S1813) was spin-coated on top of the processed wafer on both sides of the substrate to safeguard waveguides from scratching and contamination during the back end process.
Thin fused silica substrate was temporarily bonded with a thick (1 mm) glass wafer with a 4" diameter to ensure that the thin transparent wafer would not break during the back end processing. As shown in Fig. S2a, the thin transparent waveguide wafer was placed on the thick circular glass carrier substrate using quartz wax from Logitech (part no. 0CON-200, Melting point 66 -69°C) and heated to 80°C for its fast melting. Constant pressure was applied after melting the wax to make sure no air bubble remained trapped and that spreading of the wax was uniform throughout the thin transparent wafer as shown in Fig. S2b. After that, the transparent wafer with carrier glass was diced with a diamond blade under water cooling (Isomet 5000, Buehler) into small chips.
Instead of dicing for the opaque substrate, cleaving along the crystal lattice was performed to dice the chips.
The lapping and polishing processes for the transparent and opaque chips were carried out on a Logitech LP50 system. The initial lapping step utilized a 9 µm alumina solution from 30 minutes to one hour, until the waveguide end facets of the diced chips were fully visible and flat. The second lapping step utilized a 3 µm alumina solution and was performed for 30 minutes to reduce the surface roughness to below 3 µm. In both cases, the sample was placed against a rotating metal plate. The alumina lapping solution with 3 µm particles offered sufficient removal rates of 5 µm min -1 . After lapping, the samples were polished on a different LP50 system using a diamond slurry. The surface roughness value for diced waveguide input facets was reduced from a few microns to tens of nanometers after polishing.

AFM Measurement
Roughness of the substrates and deposited Ta2O5  were recorded versus the tip-sample separation distance, and Sader-Jarvis-Katan formalism was used to transform the cantilever dynamics into the force versus distance curves 38 .

Ellipsometry Measurement
The ellipsometric spectra of the 250 nm thick Ta2O5 film on the transparent fused silica and opaque silicon substrates were collected using a J.A. Woollam VASE ellipsometer in the 400-800 nm spectral range. The measurements were performed at an angle of incidence of 55 o and analysed using the CompleteEASE ellipsometry data analysis program. They were fitted using the Tauc-Lornetz method which is suitable for amorphous semiconductor materials 39 .  Table S1.

Propagation loss of the Ta2O5 waveguides
Propagation losses were analysed using the scattering analysis method 35 . 660 nm laser light was coupled into the waveguide and propagation losses were measured by acquiring images of light scattered from the waveguide surface using a microscope setup previously described 35 . To avoid saturated pixels, the exposure time for the microscope's CCD camera was adjusted. The images were processed with a MATLAB program: an average value for each image was used to find the intensity of propagating light as a function of the waveguide length, and then the result was plotted on a log scale with a linear fit to estimate the propagation loss (dB/cm).

Sample preparation
2 µm diameter, uniformly fluorescent microspheres with an RI of 1.59 and excitation/emission wavelengths of 540/600 nm (FC05F, Bang Laboratories) were used. The stock solution was diluted 1:100 in water and 500 µl of the solution was added into the PDMS well chamber on the transparent waveguide.
HeLa cells were grown in minimum essential medium (MEM) supplemented with 10% fetal bovine serum and 1% penicillin/streptomycin, in a standard humidified incubator at 37 °C with 5% CO2. Cells were seeded into the PDMS well chambers on transparent waveguides (described further in Section 3) 1-2 days before imaging. Cells used for live-cell imaging were incubated with 1 μM SiR-Actin and 10 μM verapamil (Spirochrome) in culture medium for 1 hour before washing. The culture medium was aspirated and replaced with pre-warmed Live Cell Imaging Solution (Invitrogen) before imaging. For phalloidin labeling, HeLa cells were similarly seeded within PDMS chambers, and after 1-2 days of growth fixed for ~20 minutes using 4% paraformaldehyde in PBS. Cells were then washed in PBS followed by incubation with 0.1% Triton X-100 in PBS for 4 minutes, and washed 3 times in PBS for a few minutes each time. Cells were then incubated with Atto-647N phalloidin (1:33 for widefield c-TIRF, 1:100 for SRRF imaging) in PBS for 90 minutes.

Image processing
All microscopy images were processed using FIJI 40,41 . Image sequences with fluctuating modes were first averaged over time, followed by processing steps performed on the averaged image. Background subtraction with a rolling ball radius of 50 pixels was performed for c-TIRF images of cells. Slight differences in rotation and magnification between the cameras for the upright versus inverted camera were compensated for using the Align Image by line ROI plugin for the images of fixed cells; this allowed for direct comparison by merging/overlaying channels. Images shown in this paper have been linearly adjusted for brightness and contrast. Single-channel images use a 'hot' style lookup table for improved contrast.
The super-resolved images were processed with the SRRF plugin for FIJI 25 . Five hundred (500) frames were used to process the SRRF images, each image having 50 ms acquisition time. A ring radius of 0.5 (default) and radiality magnification of 15 were used.      Table 1 of the main manuscript as well. Scale bar is 500 nm. Experimental ellipsometric data were fitted using the Tauc-Lornetz method, which is suitable for amorphous semiconductor materials, to find the RI (n) and extinction coefficient (k). The calculated optical constants (n and k) for the Ta2O5 film on (c) transparent and (d) opaque substrates are shown plotted as a function of wavelength. The RI values extracted from this data at a wavelength of 660 nm are listed in Table 1 of the main manuscript. this is approximately 100X larger than what is possible using traditional objective lens-based TIRF, but without any compromises to the axial resolution inherent to TIRF. Scale bar is 50 µm.   6 was analysed using the FRC plugin in FIJI, and the resulting correlation curve is shown here.

A transparent waveguide chip for versatile TIRF-based microscopy and nanoscopy
Resolution is extracted from the spatial frequency at which the correlation value drops below a standard threshold of 1/7 (dashed line), which here corresponds to a lateral resolution of 116 nm (blue arrow).