Principle and practical implementation

The proposed approaches are illustrated on Fig.1: compared to a wide-field RIC microscope, a digital micromirror device (DMD) is inserted in the optical path and is optically conjugated with the focal plane. By applying a black-and-white amplitude mask on the device, the amplitude of the incoming light is modulated at will at frequencies up to 20 kHz and a spatial sampling in the focal plane of 216 nm (for a 60× objective, as set by the camera pixel size and DMD-to-camera magnification factor). The resulting pattern is imaged on a camera and provides optical sectioning through a matching detection pattern or through post processing of the acquired image: in LC imaging (Fig.1, left), a line of controlled width is scanned across the field of view and the signal on the camera is acquired using a rolling shutter (i.e., signal acquisition starts one line after the other, instead of simultaneously for the whole camera) within a matching moving line of fixed width [19, 20]. In this case, similarly to laser scanning confocal microscopy (LSCM), optical sectioning is obtained through rejection of the defocussed reflections of the illumination line that fall mostly out of the detection line for a sufficiently large defocus [12, 16].

Open in new tabFig 1: RIC microscope with optical sectioning.

Center, a schematic view of a RIC microscope is presented, along with the illumination and detection schemes used for line confocal (LC, left) and structured illumination (SI, right) sectioning. The illumination path consists of an incoherent white light source illuminating a digital micromirror device (DMD). Its illumination cone must be large enough to fully neglect the subsequent diffraction and related chromaticism induced by the DMD. The DMD then is imaged onto the sample plane of the objective after spatial filtering through an aperture diaphragm (setting the illumination numerical aperture (INA) and hence the coherence volume within which RIC is observed [2]) and a field diaphragm (limiting the field of view). The field diaphragm is not strictly necessary as a virtual diaphragm can be set by the DMD by limiting the patterned area, but is useful for DMD alignment. The illumination beam is further filtered in polarisation using a polarising beamsplitter cube and a quarter wave plate (QWP) to separate illuminating from reflected light. The gray arrows show the polarisation of the beam at various positions in the beam path. Incoming and refected beams are drawn with an offset from the optical axis for the sake of clarity only, but are in reality incoming at the center of the objective back aperture. The reflected light is then imaged onto a sCMOS camera, either directly (not shown) or after passing through a slit and a beam splitting system to obtain simultaneously two RIC images at different wavelengths (the case of two colours is shown here as an example). Multicamera imaging is also possible to increase the number of simultaneously recorded images.

Alternatively in SIM, a series of illumination structures such as grids with varying orientations and phases are projected onto the focal plane (Fig.1, right) [21, 22], and the signal is reconstructed by calculating the standard deviation of the resulting image stack [10]. More complex algorithms can be used to improve the signal-to-noise ratio and the resolution of the reconstructions, but they all rely on the combination of at least two [23] and up to several tens of images [24], thereby reducing by the same factor the frame rate of the sectioned image acquisition [25]. The most common implementation, used in this paper, uses 3 images of a 1D sinusoidal pattern shifted between each image by one third of a period (I₁, I₂ and I₃). While more efficient algorithms may be used to minimise the image noise and improve its resolution [26], we have restricted ourselves here to the canonical standard deviation calculation approach that has the benefit of simplicity and permits real time video-rate image reconstruction on a recent computer: $I_{SIM}=\sqrt{{\left(I_1-I_2\right)}^2+{\left(I_2-I_3\right)}^2+{\left(I_3-I_1\right)}^2}$(1)

Using the same three images, a widefield image can also be reconstructed using the formula: $I_{WF}=I_1+I_2+I_3$(2)

LC and SIM are complementary and can be implemented using the same experimental setup, and we detail their relative advantages and limitations in the Discussion section. Complete experimental details on their practical implementation can be found in the Methods section and S3 Fig, and should allow the reader to set up an equivalent sectioning module on their microscope. Briefly, hardware TTL signals are used to synchronize the display of preloaded black and white patterns on the DMD, the acquisition of images on one or several sCMOS cameras, and if needed the motorised displacement of the sample (to acquire several fields of view or adjusting the focus, for example). For SIM, a full image is acquired on the camera with simultaneous exposure of all pixels, while LC uses progressive scan on the camera at a speed synchronised with that of the illumination line movement on the DMD. Conversely, LC images are directly displayed and recorded, while SIM images require further postprocessing to cancel the illumination patterns and obtain a sectioned image (e.g. equ. 1). A home-written LabView software is described in S4 Fig that permits real-time visualisation of SIM sectioned images.

To assess the sectioning ability achieved with either of these methods, we have recorded the signal reflected by a test sample composed of a glass coverslip onto which a nanometric layer of gold has been deposited that provides the optical contrast, and which is covered with immersion oil of refractive index matching that of the glass. This permits neglecting all aberrations introduced upon changes in refractive index at the interface. Here, it should be noted that quantitative image analysis in RICM usually relies on the use of small illumination numerical apertures (INA) so that illumination can be considered quasi-parallel. INA is defined as n sin(α_ill), with n the refractive index of the sample and a_ill the maximum angle of the incoming illumination. A low INA also permits optimising the visibility of interferences within the focal volume, and allows visualizing objects up to ≈ 10 μm above the reference surface (together with the appropriate tuning of the temporal coherence) [18]. As a result, the width of the patterns p that can be projected onto the focal volume for optical sectioning is larger than the diffraction limit of the imaging objective (typically $p_{min}\approx \frac{\lambda }{2\text{INA}}>\frac{\lambda }{2\text{NA}}$, with NA the full numerical aperture of the objective). This, in turn, limits the performance of optical sectioning thickness to a minimum of $\delta \approx \frac{n{p}_{min}}{\text{INA}}\ge \frac{n\lambda }{{\text{INA}}^2}$. Note tat or a given value of p, axial sectioning does not depend on the wavelength, facilitating multicolour SI-RICM. A full description of the image formation process in optically sectioned RICM is beyond the scope of this paper and is the topic of another manuscript in preparation by the authors. Considering that INA in RICM is often limited to 0.3-0.5, one gets a shortest axial sectioning in the range 2-10 μm.

While this is far from the sectioning obtained in TIRF microscopy (a widely-used option for imaging samples close to the coverslip surface), the proposed approaches are more easily compatible with quantitative interference imaging, are easily implemented in the conventional microscope and provide a larger field of view (typically ≈ 10× larger when using a 20× objective, at the expense of a lateral resolution ≈ 1.5 — 2x larger). To demonstrate their usefulness, we provide below three case studies where optical sectioning permits the quantitative analysis of complex samples.

Application example 1 : cell imaging in a complex environment

RICM has widely been used for marker-free quantification of cell adhesion or membrane fluctuations [27–29]. While a powerful tool, in its classical configuration its lack of sectioning ability has hampered its use in complex environments such as in the presence of several layers of cells in which quantitative interpretation of images is challenging. In such cases, incoherent background subtraction cannot remove the spurious spatially-heterogenous, time-dependent incoherent signals arising from outside of the coherence volume.

As a proof-of-principle demonstration of optical sectioning with LC, RICM images of red blood cells in PBS sedimented on a glass slide, with and without LC sectioning, are shown on Fig.3. The lower membranes of red blood cells produce well-contrasted fringes that can be used to reconstruct their height profile. As expected, the widefield image exhibits a much higher background intensity due to incoherent parasite reflections, such as stemming from the liquid/air interface at the top of the sample. However, even when subtracting the constant offset from the images, incoherent reflections from area of cells outside the focal volume are visible in the image. In some cases, a second cell resting on a sedimented one modifies the contrast of the fringes, rendering the quantitative analysis of the latter challenging (arrows on Fig.3b). This is representative of typical physiological conditions, where the density of cells within a microvessel can reach 30-40% volume fraction [30]. Studying cell-wall interactions at such high density would thus induce massive out-of-focus heterogeneous reflections from blood cells inside the channel.

Open in new tabFig 3: Red blood cell imaging in a crowded environment

RICM images of red blood cells sedimenting on a glass surface without (a-b) and with (c) LC sectioning. (a) Raw wide field image. (b) Same image with background removed. Outlines of cells stemming from incoherent reflections are still visible in the images (red arrowhead) and can superimpose with the interference pattern (yellow arrowhead). This results in a variable intensity of the interference fringes, as demonstrated in (d) where the profile along the green line is plotted with and without LC sectioning. Scale bar, 10 μm.

In contrast, the same image acquired using LC-RICM yields a much clearer image of the cell structures in the vicinity of the coverslip that provide interferential contrast. Note that the images with and without LC sectioning were acquired sequentially, so that a slight sedimentation and differences in membrane fluctuations are visible in between the two images.

Application example 2 : quantifying the topology of complex surfaces

RICM can also be used as a non-invasive probe of the topology of surfaces decorated with organic material, including wet samples that cannot be dried without altering their structure. Compared to its derived counterpart biolayer interferometry (BLI) [31], it can provide sub-micron resolution in the imaging plane that permits assessing not only the average layer thickness but also the thickness distribution. This is usually performed using multicolour RICM, providing unambiguous information on the layer thickness up to about 1 μm [3, 4]. Quantitative analysis of the RICM signal is however often hampered by the spurious reflections from other interfaces. In particular, it is often convenient to use air objectives for intermediate characterisation steps, so that further functionalization can be performed without cleaning of the lower face of the coverslip, and without sample contamination with immersion oil. This leads to a strong reflection from the non-functionalized side of the coverslip which for samples immersed in water is typically one order of magnitude larger than the signal from the interface of interest itself. In this context, optical sectioning greatly improves the quantification of the surface topology.

As a demonstration, we have imaged a multilayer sample of alternating cellulose nanocrystals (CNCs) and xyloglucan (XG) particles adsorbed on a polyethileneimine (PEI) layer. XGs are the major hemicelluloses in the primary cell wall of dicotyledonous plants, while CNCs in plant cell walls combine with other biomolecules to form supramolecular assemblies reinforcing its structure. The strong interaction between CNCs and XGs permits building biomimetic nano composite films reminiscent of plant cell walls, whose thickness is controlled by the number of (CNC/XG) layers [32]. While such layers can be characterised by AFM at small scale (typical a few μm²), or by neutron reflectivity (with lateral averaging over typically several cm²) [32], multicolour RICM with optical sectioning provides a fast, sub-micrometric, large field-of-view characterization of the film height that can be performed as a quality control at different stages of film deposition. We demonstrate how with a 20×, air objective, line confocal RICM permits monitoring the thickness of a PEI(CNC/XG)₅ CNC film over a 200 ×500 μm² field of view in ≈ 1s with a ≈ 500nm lateral resolution using 3 visible wavelengths (453, 532 and 632 nm) that allow reconstructing pseudo-colour images (Fig.4a). Comparison between widefield and optically sectioned images reveals a striking improvement in contrast and an increased visibility of small structures, as well as a lower sensitivity to out-of-focus impurities (Fig.4a, red arrowheads). Even after FFT filtering to remove the slowly varying background, the visibility of small structures is still hampered by noise and residual background in wide field RICM images compared to line confocal images (Fig.4b, yellow arrowheads).

Open in new tabFig 4: Topology of a biomimetic multilayer film.

(a) A PEI(CNC/XG)₅ CNC film immersed in water obtained by layer-by-layer deposition is visualised without (top) and with (bottom) line confocal sectioning in 3-colour RICM, yielding a false colour RGB image that informs on the topology of the film. In the absence of sectioning, a large incoherent background reduces the contrast of the signal of interest. Scale bar, 50 μm. (b) Zoom on the green rectangle in (a), with a Fourier filtering the wide field image (top). This preserves the RICM contrast only if assuming a zero low frequency component in the signal of interest (an assumption that fails in the case of samples thinner than about 100 nm). Even with this filtering, the contrast is still poor compared to the corresponding sectioned image (bottom). Scale bar, 10 μm. (c) Computed hue corresponding to a given height assuming a layer refractive index of 1.4. (d,e) Reconstruction of the height profile based on the 3-colour RICM signal, on the full image (d) and on the green inset (e). The two images share the same colour scale. (f) Histogram of heights measured on each pixel from the full RICM image.

Furthermore, sectioned images can be used to quantitatively reconstruct the topology of the film using simple assumptions for its optical properties: assuming a homogeneous, isotropic refractive index (a reasonable assumption for a disordered film of polysaccharides) ranging from 1.47 (fully dehydrated) to 1.4 (partly hydrated, polysaccharide content 50%) we can reconstruct the variations of hue expected as a function of film thickness (Fig.4c and methods). Interestingly, the colour scale is only weakly affected by the large change in refractive index that we have considered, yielding only a ≈ 5% systematic change in the estimated thickness between these two strongly different hydration assumptions. Based on this hue scale, machine learning is used to map the 3-colour RICM image onto a series of height probability images using 20-nm steps [33]. These images are subsequently combined to obtain a quantitative thickness image over the whole field of view (Fig.4d and 4e, see Methods). These maps provide a wealth of information such as the average film thickness (here ≈ 121 ± 5nm), roughness (here ≈ 24 ± 2nm, see methods for the definition of roughness) and characteristic lateral size of height variations (here ≈ 0.65 μm, close to the resolution limit ≈ 0.5 μm hinting that smaller features are present but less visible or poorly resolved, in agreement with AFM images [32]), as well as a full height histogram (Fig.4f). Previous AFM characterizations have shown that the thickness of hydrated PEI(CNC/XG)₅ CNC films was 124 ± 21nm [34], a value with which our height measurement through confocal RICM is in excellent agreement, demonstrating the robustness of our optical measurement experimental setup and processing pipeline.

Application example 3: probing surface biofunctionalization quantitatively

In addition to being non invasive, spatially-resolved and compatible with wet samples, RICM also permits rapid measurements which allow monitoring of sample formation or changes in situ. Here, we imaged the functionalization of a glass interface with a molecular platform permitting controlled presentation of biomolecules at a solid/liquid interface. The platform is composed of a supported lipid bilayer (SLB) incorporating biotinylated lipids (see Methods) that provide anchor points for tetrameric traptavidin molecules (TAv), a variant of streptavidin (SAv) with a larger binding energy [35]. Because only two to three of the four biotin-binding sites are bound to the SLB [36], this platform provides a way to graft a variety of biotinylated molecules at a controlled orientation and surface density, forming well-controlled biomimetic surfaces widely used for in vitro studies of ligand-receptor, or interactions of cells or viruses with model cell surfaces [37–39]. The method relies on the fact that the biotin-SAv bond is one of the strongest non-covalent bond known in nature, which is even strengthened in the synthetic variant TAv (35 kT/bond). However, while the SLB-SAv platform has been already widely used and characterised, and in particular the ability of mobile SAv molecules to form 2D crystals above a critical surface concentration [40], its synthetic variant TAv has not been characterized so far for this type of application.

Here, we take advantage of the enhanced sensitivity of SI-RICM to monitor the adsorption of TAv to a biotinylated SLB during TAv incubation. Contrary to spectral ellipsometry, which provide a laterally-averaged quantification of the adsorbed amount of protein, SI-RICM does not straightforwardly inform on the adsorbed surface density of proteins in the regime of a fraction of a single monolayer, but provides a direct readout of the homogeneity of the functionalized surfaces. Interestingly, this sub-micron resolution shows that in incubation conditions that provide homogeneous and mobile SAv layers, TAv forms regions of higher densities (appearing darker in RICM) surrounded by lower density areas (Fig.5). Real-time monitoring shows that this process occurs quasi-simultaneously in the whole field of view and is initiated on sub-resolution defects (lateral resolution ≈ 250 nm, using a 60×, 1.35NA oil objective) in the bilayer (see S1 Video). By analogy with SAv, for which literature extensively documents the formation of 2D crystals on lipid layers [40–43], and since when labelled (see below), dense areas display an anisotropic in-plane response and an absence of mobility despite the fluidity of the LSB (confocal imaging and FRAP, not shown), we similarly attribute the dense phase to the formation of 2D crystals.

Open in new tabFig 5: In situ, non-invasive monitoring of TAv adsorption to a biotinylated surface.

(a) 532-nm SI (left) and widefield (right) RICM images of a biotinylated SLB incubated with TAv. Crystals start to form on the surface after ≈ 17 min incubation, with a contrast that remains constant over the whole incubation phase. Scale bars, 20 μm. (b) Relative change in intensity in the crystalline phase compared to the mobile phase with (SIM) and without (WF) SI sectioning, illustrating the gain in contrast thanks to optical sectioning. Error bars are standard deviations on n=24 independent measurements on different crystals of the same sample. (c) Computed concentration of TAv in the mobile phase compared to the crystalline phase using fluorescence confocal microscopy (Fluo), SI and widefield (WF) RICM. SI-RICM is in good agreement with confocal measurements, while widefield RICM underestimates the difference in surface concentration between the two phases. Error bars are standard deviations on n=16 (confocal) or n=24 (RICM) independent measurements on different crystals of the same sample, represented as blue dots. The same sample was measured in RICM and fluorescence to allow direct comparison of the measured densities.

This application highlights the extreme sensitivity of SI-RICM to minute variations of the surface reflectivity which is modulated here by only 0.6%. Importantly, optical sectioning permits analyzing the changes in reflectivity to extract quantitative information about the sample. Here, we observe first that the contrast between the high- and low-density phases is constant over time, indicating that once a sufficient surface density is reached, additional binding of TAv does not increase further the density of the mobile phase, but instead induces growth of the crystalline phase. Furthermore, modelling the optical properties of the SLB-TAv platform and assuming that the TAv crystalline phase density is identical to that of SAv (34 nm²/molecule, or 4.9 pmol/cm², for crystallization conditions similar to ours [41,42]), we can estimate the density of TAv in the disordered phase (see Methods). We find that the mobile phase is about 30-35% less dense than the crystalline phase (corresponding to 3.3±0.1 pmol/cm²), while analyzing widefield images provides a relative difference of only 13% (corresponding to 4.3±0.05 pmol/cm²). This stems from the lesser contrast of the crystals in the presence of an unknown incoherent background.

To validate the value obtained by SI-RICM, we incubated the platform further with biotinylated-FITC and imaged the sample using fluorescence confocal microscopy: comparing the relative fluorescence intensities in the crystalline and mobile regions confirms a relative change of ≈ 30%, highlighting the benefits of optical sectioning for the quantitative analysis of RICM images. Note that while fluorescent tagging provides a more straightforward measure of TAv surface density, it can only be used after the end of incubation, and prevents further functionalization by biotinylated molecules of interest. This highlights the benefits of SI-RICM for investigating surface functionalization.