Near-Membrane Refractometry Using Supercritical Angle Fluorescence

Abstract

Total internal reflection fluorescence (TIRF) microscopy and its variants are key technologies for visualizing the dynamics of single molecules or organelles in live cells. Yet truly quantitative TIRF remains problematic. One unknown hampering the interpretation of evanescent-wave excited fluorescence intensities is the undetermined cell refractive index (RI). Here, we use a combination of TIRF excitation and supercritical angle fluorescence emission detection to directly measure the average RI in the “footprint” region of the cell during image acquisition. Our RI measurement is based on the determination on a back-focal plane image of the critical angle separating evanescent and far-field fluorescence emission components. We validate our method by imaging mouse embryonic fibroblasts and BON cells. By targeting various dyes and fluorescent-protein chimeras to vesicles, the plasma membrane, as well as mitochondria and the endoplasmic reticulum, we demonstrate local RI measurements with subcellular resolution on a standard TIRF microscope, with a removable Bertrand lens as the only modification. Our technique has important applications for imaging axial vesicle dynamics and the mitochondrial energy state or detecting metabolically more active cancer cells.

Introduction

Total internal reflection fluorescence (TIRF) has evolved from an expert technique to a routine contrast mode used for single-molecule and single-organelle tracking at or near the basal plasma membrane of cells adhered to a glass substrate (1). Photoactivated localization microscopy and stochastic optical reconstruction microscopy are localization-based superresolution techniques that have further expanded the range of TIRF applications. The emission counterpart of TIRF, supercritical angle fluorescence (SAF) (2), is increasingly being used for surface microscopies (3, 4, 5). For TIRF, the presence of a refractive-index (RI) boundary between the glass substrate (of index n₂) and the aqueous sample (n₁) results in the emergence of an evanescent wave (EW) that skims the interface layer and provides excitation-light confinement, restricting fluorescence to a thin optical section near the basal plasma membrane. In SAF microscopies, the otherwise undetected evanescent emission component of surface-proximal fluorophores couples to the interface, where it becomes propagative and detectable in the far field, provided the objective has a sufficiently large numerical aperture (NA) for the SAF emission to be captured, i.e., NA = n₂ sin θ_NA with θ_NA > θ_c.

Besides its importance as a parameter characterizing protein density, structure, and function of cells, the sample RI is a fundamental optical variable for surface microscopies; the critical angles for total reflection at the excitation or emission wavelength, λ, θ_c = asin(n₁(λ)/n₂(λ)), the axial decay length of the EW, δ = λ/(4π(n₂²·sin²θ − n₁²)^1/2), and the dipole radiation pattern are all modified by the local sample RI, n₁(x, y). However, this sample RI is generally undetermined during imaging. Even knowledge of the average near-membrane RI for the very cell under study would greatly enhance our ability to choose appropriate incidence and detection angles, to better understand and eliminate image imperfections (e.g., those resulting from non-totally reflected, and hence propagating, excitation light emerging from high-RI regions), and to interpret TIRF and SAF in quantitative terms, e.g., for axial profilometry, size or concentration measurements, or axial single-vesicle or single-molecule tracking.

Hilbert-phase microscopy (6), digital confocal microscopy (7, 8) and full-field optical coherence tomography (9) all allow RI measurements with subcellular resolution, but none of these techniques selectively probes the near-membrane space. On the other hand, digital confocal microscopy in a TIR geometry (10) and surface-plasmon-based RI sensing (11) both measure the RI near the basal plasma membrane, but they either need a reference beam (and hence modifications to the TIRF illuminator) or require metal-coated substrates. Alternatively, near-field scanning-probe techniques (12) are generally too invasive for studying biological phenomena in live cells. An interesting but surprisingly rarely used technique for probing the near-membrane RIs and substrate-cell distance is reflectance interference contrast (RIC) microscopy. See Discussion for details.

In this work, we present and validate a simple scheme for near-membrane refractometry that requires a Bertrand lens (BL) as the only extra component. We measured the RI of mitochondria, the endoplasmic reticulum (ER), and secretory vesicles/lysosomes in the very region probed by TIRF and SAF surface microscopies in live cells. We found significant RI differences among organelles that are in line with RI values obtained from cell fractionation and organelle purification experiments. We also demonstrate that our technique is capable of resolving cell-type-specific RI differences by comparing organelle RIs from mouse embryonic fibroblasts with that of an optically denser human cell line, BON cells (a model system for the study of carcinoid tumors), and we discuss the limits and future developments of our approach.

Materials and Methods

Reagents

Dulbecco’s modified Eagle’s cell culture medium (DMEM; Gibco, Billings, MT), fetal calf serum, penicillin/streptomycin, and 100-nm diameter yellow-green-emitting (505/515 nm) InSpeck beads were from Invitrogen (Cergy Pontoise, France). Fluorescein isothiocyanate (FITC) and FITC-dextran (FD-10S) were from Sigma (Deisenhofen, Germany). Beads were resuspended in ethanol (EtOH) and deposited as a sparse monolayer on the coverslip surface by solvent evaporation of a small (μL) droplet. FM2-10 and FM1-43 were bath applied (1 μg/mL for 20 min) and the cells were thoroughly washed two times before imaging. FITC-dextran was added at 1 mg/mL for 2 h and the cells were imaged after changing the extracellular solution, leaving only endo-/pinocytosed FITC in intracellular vesicles.

Cell culture, plasmids and transfections

For cell culture, coverslips (BK-7, 25-mm diameter, Schott-Desag D263M, Thermo Fisher Menzel, Braunschweig, Germany) were sterilized by passing them through 70% EtOH baths, and kept individually in EtOH in a tightly sealed six-well plate. Before use, the EtOH was washed and the coverslips were rinsed with sterile water to avoid bacterial infections and covered with culture medium (see below). In some experiments, coverslips were prebleached by intense overnight 337-nm irradiation with 260-μJ pulses emitted from a pulsed nitrogen laser (VSL337ND-S, SpectraPhysics, Santa Paula, CA).

Mouse embryonic fibroblasts (MEFs) were cultured at 37°C in a 5% CO₂ atmosphere in a medium of 86% DMEM, 9.6% fetal bovine serum, 1% L-glutamine (200 mM, PAA Laboratories, Pasching, Austria), penicillin/streptomycin (50 μg/mL), HEPES (pH 7.3, 1 M), pyruvate (100 mM), and 0.01% β-mercaptoethanol (50 mM, Gibco).

BON cells (derived from a metastatic human carcinoid tumor of the pancreas (13, 14)) were originally provided by C. M. Townsend (Galveston, TX) and cultured as described (15). Cells were maintained at 37°C in a humidified 5% CO₂ atmosphere in DMEM F12 with L-glutamine, supplemented with 10% fetal calf serum and penicillin/streptomycin (100 U/mL and 100 μg/mL, respectively). On the day before transfection, cells were replated on collagen-coated coverslips at 50,000 cells/cm². Isolated round adherent cells were imaged 24–48 h after transfection with Jet Prime (Polyplus, Illkirch, France) using standard protocols (see Table S1 in the Supporting Material for details).

For imaging, irrespective of cell type, coverslips were shaken in cell culture medium and rinsed to remove detached and faintly adherent dead cells, tightly fixed in a custom-made Teflon coverslip holder, and covered with imaging medium (the same as the culture medium, but with DMEM containing no phenol red). All experiments were performed in a static bath at room temperature (20–22°C) on a home-built microscope.

Combined TIRF-SAF microscopy

The beam of an Ar⁺-ion gas laser (Reliant 150 Select, DZ Laser Service, Centerville, UT) was filtered to remove the 458- and 514-nm lines, shuttered (LS3, Uniblitz, Vincent Associates, Rochester, NY), and delivered to the optical table via a mono-mode polarization-maintaining optical fiber (Qioptiq PointSource, Hamble, UK). The vertical linear polarization was adjusted with a zero-order half-wave plate (λ/2, WPH05M-488, Thorlabs, Newton, NJ) to maximize the throughput of the (1,1)-order of our acousto-optical deflector (AOD) beam scanner (see below).

We used our combined spinning TIRF (4, 5) and SAF (16) (spTIR-SAF) microscope described earlier. Briefly, a pair of AODs (AA.Opto, St-Rémy-en-Chevreuse, France) scanned the expanded beam that was focused by a six-element scan lens (Rodagon, Rodenstock, Feldkirchen, Germany) to a tight spot in the objective back focal plane (BFP). All other lenses were achromatic doublets (Linos, Göttingen, Germany). The αPlan-Apochromat 100×/NA 1.46 oil objective (Zeiss, Oberkochen, Germany) was piezo positioned for accurate focusing at the near-membrane layer (PIFOC, Physik Instrumente, Karlsruhe, Germany). Fluorescence was detected through the same objective that was used for TIRF excitation, extracted with a 2-mm-thick dichroic mirror (zt491 RDCXT, AHF, Tübingen, Germany), and filtered with two stacked holographic notch filters (to remove the totally reflected 488-nm beam). The fluorescence was either imaged directly or the objective BFP was imaged, using a BL (f = 100 mm), onto an electron-multiplying charge-coupled device camera (EMCCD, QuantEM512C, Photometrics, Tucson, AZ).

Compared to the earlier published version (16), four modifications were made to our microscope: 1) a λ/4 plate (WPQ05M-488, Thorlabs) was inserted right after the AODs to produce a circularly polarized 488-nm beam and excite all fluorophores independent of their dipole orientation; 2) the second lens of the beam-compressing, i.e., scan-angle-increasing, telescope (f₁ = 100 mm, f₂ = 50 mm) was mounted on a z-adjustment positioner (G061165000, Linos) and the axial lens position was fixed by minimizing the beam divergence, measured as the spot diameter of an on-axis beam emerging from the microscope objective when projected at the ceiling; 3) similar to (17), we mounted an inexpensive CMOS camera (DCC 1545M, Thorlabs) in an equivalent BFP of the microscope excitation optical path for better alignment and monitoring of the excitation light distribution; and 4) a white-light HighLED (40 mW; G06 5150 000, Linos, and LXHL-MW1C, Luxeon Star, Lethbridge, Canada) provided bright-field illumination to select fields of view that contained a single isolated cell that responded to morphological criteria.

RI calibration

RIs $\left(n_{\text{D}}^{\left(20°\right)}\right)$ of precalibrated commercial sucrose “Brix” standards having 10, 12.5, 20, 25, 30, 35, 45, 50, 55, and 60°Bx (%w/w sucrose in water, from Reagecon, Shannon, Ireland) were first verified at room temperature with an Abbe refractometer (WYA, Shanghai, China) using the sodium-D line of a low-pressure sodium-vapor lamp (589 nm). They showed a systematic 2% offset compared to the specified values (see Table S2). Fluorescent beads stuck to a coverslip surface and covered with these sucrose media or water, or not covered (i.e., exposed to air), were used, and the resulting sample images (fluorophore distribution) and BFP images (fluorescence radiation pattern) were collected.

Calibration of the pixel size and RI measurements

Sample-plane images were calibrated with a microscope slide having an etched 10-μm grating (Linos). For aperture-plane pixel calibration, we placed a miniature transparent glass microruler (Linos) in the BFP 8.5 mm above the objective shoulder and measured the line spacing on BFP images. The total magnification was 90.6 nm/pixel for sample-plane images and 11.5 μm/pixel for aperture-plane images. RI values of the medium surrounding the fluorescent sample and the objective NA were established considering that r_c = $f\times \langle n_1\rangle$ and r_max = f × NA, and r_c and r_max were measured from the BFP images (Fig. 1 A, inset). Using these equations plus the BFP pixel size and f = 1.65 (the nominal value for a Zeiss 100× objective), we found RI values for air (n_air = 1.003 ± 0.003) and water (n_water = 1.339 ± 0.003) that were in excellent agreement with their nominal values. The objective NA found from r_max (NA = 1.465 ± 0.009) also coincided with the specified one (NA = 1.46).

Figure 1.

SAF-based sample RI measurement. (A) Simplified optical layout. Obj, objective; TL, tube lens; FL, focusing lens; S, sample; DC, dichroic mirror; M, mirror. The excitation wavelength was 488 nm. The inset shows a BFP image of a thin layer of fluorescent (λ_ex/λ_em = 488/515 nm) 100-nm-diameter beads in water. r_c identifies the critical angle for a given RI. r_max corresponds to the objective maximum angle of collection. (B) RIs measured by Abbe refractometry from a series of calibrated sucrose standards versus the corresponding r_c. The red line is a linear fit to the data, with slope = 1.6528 ± 0.0003 and R = 1. See Materials and Methods for details. To see this figure in color, go online.

Determination of r_c and r_max

To find the inner (r_c) and outer radius (r_max) from BFP images, a custom routine was written in MATLAB (The MathWorks, Natick, MA) that first took the derivative along both the x- and y-directions of the image and then fitted a circle with the maxima of the derivative image (see Fig. S4). Due to fitting, the accuracy of the radius determination is subpixel (in the same manner that a Gaussian fit can localize a single fluorescent molecule with subpixel accuracy).

Accuracy and precision of RI measurement estimation

The accuracy of the method was measured by taking BFP images of a monolayer of subresolution fluorescent beads (100-nm diameter) covered with water or one of the glucose solutions mentioned in Table S2 for three independent alignments (along the optical axis) of the BL. The mean of the standard deviation of the RI found for each solution was 0.0026 RI units. This error takes into account not only the alignment precision but also the accuracy of the fitting process. This value was considered as the stereotypic error of all RI measurements.

Statistics

All experiments are reported as the mean ± SD of at least three independent experiments. For RI measurements in cells, at least six cells for each labeling and cell type were measured.

Results

SAF-based RI measurements reproduce Abbe refractometry

Our technique is based on the realization that for fluorophores located near a dielectric interface (n₁/n₂) and embedded in a medium of index n₁, the RI is encoded in a fluorophore radiation pattern. The radiation pattern is accessible by imaging the aperture plane (or BFP) instead of the sample plane. Provided that the NA of the objective is large enough (θ_NA > θ_c), the RI can then be measured from the boundary separating under- and supercritical fluorescence emission components (18, 19) (Fig. S1). The limit between them is the critical angle, θ_c,fluo, at the wavelength of fluorescence emission. As a consequence of the Abbe sine condition, angles are converted to radii in the BFP, and the sample RI and objective NA correspond to r_c = f × RI (Fig. 1 A, inset, yellow line) and r_max = f × n₂·sin θ_NA = f × NA (red line), respectively. Here, f is the objective focal length, f = f_TL/M (where f_TL is the focal length of the tube lens and M the lateral magnification). On any standard microscope, this critical-angle boundary can be imaged by the simple insertion of a BL in the microscope emission path (Fig. 1 A, red). RI measurements require the calibration of BFP radii in terms of RI values. For that, one can either use media of known RI (as done below) or else use the outer radius, r_max, limiting the BFP image. r_max depends on the effective NA of the microscope objective (18), and with 100-nm-diameter fluorescent beads in air on top of the coverslip, we found NA_eff = 1.465 ± 0.009, consistent with the nominal NA of this objective and an earlier measurement of the same lens (16).

We validated our SAF-based RI measurement by first measuring a series of commercial sucrose standards used in the food industry. RI values of the sucrose solutions were independently checked by Abbe refractometry, and the measured RIs were plotted against the r_c values obtained from BFP images of a thin layer of fluorescent beads topped with the sucrose solutions (Fig. 1 B). As expected from the r_c = f × RI relation, a line fit of the sucrose RIs with the measured critical radii gave a slope with the effective focal length f_eff = 1.6528 ± 0.0003 mm. This value is in excellent agreement with the nominal focal length of the Zeiss αPlanApo 100×/1.46 NA objective (f = 1.65). We conclude that SAF-based refractometry faithfully reproduces average RIs measured independently with an Abbe refractometer.

SAF-based measurements of organelle RI

As our technique probes the local RI of the medium in which SAF-emitting fluorophores are embedded, we reasoned that RI measurements with subcellular resolution should be feasible by targeting the fluorophores to specific subcellular compartments. To this end, we used different organelle-specific stains, labeling either the plasma membrane, endo/lysosomal vesicles, the ER, or mitochondria. Using azimuthal beam-spinning TIRF (17) excitation at θ = 68°, we confirmed the specificity of the staining from the respective fluorophore distributions on sample-plane images (Fig. 2 A). At the same time, by imaging the aperture plane (BFP), we measured the RIs of different fluorophore-labeled organelles in MEFs from their corresponding radiation patterns (Fig. 2 B).

Figure 2.

Sample plane and corresponding BFP images for different organelle labelings. Images in TIRF (A) for the different subcellular stainings in MEF cells and the corresponding BFP images (B). Red and yellow circles identify r_c and r_max, respectively. Note that, except for membrane/vesicular staining, r_max, related to the NA of the objective, cannot be seen easily due to the low SAF/UAF ratio. Scale bars, 10 μm (A) and 1 mm (B). To see this figure in color, go online.

Spinning TIRF provides a more even illumination across the field of view than does unidirectional EW excitation (17, 20, 21, 22) and hence should permit a more reliable measurement of the average near-membrane RI. Table 1 compiles the RI values that we obtained from BFP images of near-membrane endo-/lysosomes (labeled with FITC-dextran, FM2-10, or FM1-43, respectively), mitochondria (labeled with MitoTracker-Green or Mito-enhanced green fluorescent protein (EGFP)), and the ER (labeled with ER-EGFP). We found cellular RI values ranging from 1.344 to 1.38, indicating a dominantly aqueous, i.e., cytosolic, local dye environment (n_water = 1.335 at 510 nm (17)). Cellular SAF-based RI measurements were consistent for different regions of interest, and across cells, lending further support to the robustness of our approach (Fig. S2).

Table 1.

SAF-Based RI Measurements of Different Organelles in MEF Cells

Labeling	RI	SD
FITC	1.344	0.003
FM1-43	1.353	0.005
FM2-10	1.347	0.003
Mito Tracker	1.349	0.002
Mito-EGFP	1.378	0.002
ER-EGFP	1.38	0.02

No axial dependence of the RI within the volume probed by TIRF

Does the dense protein packaging at cell adhesion sites and the high density of membrane proteins affect the RI of the near-membrane layer compared to deeper cytoplasmic layers? To address this question, we measured RIs in MEFs for different polar angles, θ = 68° and 73°, corresponding to nominal EW penetration depths, δ(θ), of the order of 91 ± 14 nm to 69 ± 6 nm (considering n₁ = 1.37 ± 0.2, the exact value will depend on the unknown n₁). To our surprise, we observed no systematic trend, and this irrespective of the precise fluorescent labeling used (Fig. S3). However, upon consideration of the detailed ring structure on BFP images, the lack of a clear z-dependence can be understood (Fig. 3). In fact, in addition to the RI of the embedding medium, the axial fluorophore distance from the reflecting interface equally impacts the radial intensity distribution (3).

Figure 3.

Influence of axial fluorophore localization on BFP images. (A) Top: Sketches of vesicular and plasma membrane labeling (left), as well as reticular labeling in MEFs (right). Note the difference in fluorophore distance, z, from the interface. Bottom: Zoom on the corresponding BFP images showing the fitted r_max (red line) and r_c (yellow line). Scale bar, 1 mm. (B), intensity profiles along the purple and green dotted lines on the BFP images in (A), and the fitted r_max and r_c values indicated by red and yellow arrowheads, respectively. Note the relative dominance of UAF (green trace) and SAF (purple), respectively. To see this figure in color, go online.

For example, for MEFs labeled with FM dyes, we observed a well-contrasted, bright SAF ring, as the fluorophores were located in the basal plasma membrane and in near-membrane vesicles, close to the reflecting interface (Fig. 3, left). In contrast, for mitochondrial or ER labeling, fluorophores were on average located farther from the interface, and the emission of these organelles is dominated by under-critical-angle fluorescence (UAF) (right) (see (23)). Thus, depending on the precise fluorophore localization, the critical-angle boundary separating under- and supercritical radiation components will be visible either as a bright-to-dark transition (UAF dominating) or a dark-to-bright transition (SAF dominating), but neither transition will affect our capacity to identify the boundary and measure r_c and, hence, n₁.

As a corollary, we conclude from this observation that, despite having chosen a polar excitation beam angle θ > θ_c,ex to be in TIRF condition, organelles that were located deeper in the cytoplasm were excited and capable of contributing UAF to the fluorescence emission. Thus, the excitation light was not perfectly evanescent but must have presented long-range (scattered) light components, too (16, 20, 24). Altogether, for a given cell, a tiny modification of the beam angle (and the nominal penetration depth) did not grossly modify the excited fluorophore population.

SAF resolves cell-type-specific RI differences

We next measured the RIs of organelles in BON cells using the same method and labeling protocols as for MEFs, as described above.

BON cells are a neurosecretory tumor cell line with properties close to those of chromaffin cells of the adrenal gland, i.e., they are compact, round cells packed with thousands of large, dense-core granules, for which we would expect a higher RI compared to MEFs. The RIs for lysosomal, mitochondrial, and ER labeling are shown in Fig. 4.

Figure 4.

Subcellular RI measurements in different cell types. SAF-based RI measurements in live BON cells and MEFs for different organelle stainings. Error bars correspond to the mean ± SD of RI values from 6–8 different cells for each condition. To see this figure in color, go online.

Indeed, we found a systematically higher RI for BON cells compared to MEFs for all organelles except the ER, for which the difference was not significant. We conclude that SAF-based refractometry distinguishes different cell types, holding the potential to separate, e.g., healthy cells from tumor cells.

Discussion

We devised and validated a simple technique for measuring the average RI at or near the basal plasma membrane of live cells cultured on a glass substrate. Our technique requires fluorescently labeled cells, as it builds on analyzing the radiation pattern of near-interface fluorophores. The RI measurement is based on extracting the critical radius (i.e., the critical angle on the emission site) separating super- and under-critical fluorescence emission components, which varies as a function of the local RI of the embedding medium (Fig. 1). Our technique requires an objective capturing n angles larger than the critical angle at the wavelength of fluorescence, which is the case for all lenses compatible with objective-type TIRF. Conceptually simple, our technique permitted us to measure the RIs of different subcellular organelles by specific dye targeting (Fig. 2; Table 1) and to resolve RI differences among different cell types (Fig. 4). Although the exact radiation pattern depends on both the sample RI and axial fluorophore localization (Figs. 3 and S3), we found no clear trend when probing surface-proximal or distant cell layers using variable-angle TIRF microscopy (Fig. 3). In the following paragraphs, we discuss these results.

Multilayer surfaces

The cavity effects of multiple dielectric layers can produce complex resonances in the radiation pattern and multiring systems in the BFP image (2). For cells cultured on a glass substrate, such layers result from the polymer coating on the coverslip, which is often used to favor cell adhesion, or they can be deposited by the cells themselves as a consequence of the secretion of cell-adhesion molecules or extracellular matrix proteins.

We tested whether such multilayer effects had to be taken into account for the interpretation of our BFP images. We took images of fluorescent microspheres on bare or poly-ornithine-coated coverslips, as well as on coated coverslips on which MEFs were grown during 1 week. However, we did not observe any significant differences in the emission pattern (data not shown).

Near-critical emission of surface-distant fluorophores

Following the equations in (25) and (26), we can calculate the angular emission pattern (radiation pattern) for a fluorophore located at an axial distance z from the water/glass interface. Even at z = 10 λ_em (λ_em being the emission wavelength) the radiation pattern still shows an intensity peak near the critical angle, θ_c,fluo. However, a bright intensity ring is observed for radii corresponding to subcritical polar angles, i.e., θ < θ_c,fluo. The difference with the radiation patterns of fluorophores located much closer to the surface is that no supercritical emission (θ > θ_c,fluo) is possible for distant emitters. This is because the surface-distant fluorophores cannot emit far-field light into these “forbidden angles,” as they are too distant from the surface for their near field to couple to the interface and produce SAF. Nevertheless, they have an intensity peak at θ_c,fluo, not at supercritical but at very high under-critical angles (θ $\mathrm{\lesssim }$ θ_c,fluo). As a consequence, even for surface-distant fluorophores, r_c,fluo—and hence θ_c,fluo and the RI—can be obtained from the BFP image. The question is whether the ring intensity relative to the image background due to UAF and instrument and cellular autofluorescence is sufficient for precise edge detection.

Influence of the excitation-light distribution on the radiation pattern

Depending on the precise excitation-light confinement, different fluorophore populations are excited: in TIRF, despite the presence of nonevanescent (long-range) components (20, 24), SAF-competent emitters usually dominate, but even these near-surface dipoles always emit a mix of SAF and UAF. On the contrary, for epifluorescence or confocal laser scanning microscopy, both near-interface and distant fluorophores contribute to the BFP image. Surface-distant emitters contribute background (UAF emission) and render the detection of the critical angle, θ_c,fluo, and hence the RI measurement, more difficult. The detectability of the SAF ring will be determined by the quantity, proximity and molecular brightness (εϕ) of near-interface fluorophores. Therefore, although SAF detection is, in principle, compatible with wide-field excitation, SAF-based refractometry will always benefit from the additional excitation-light confinement of TIRF. The analogous effect was already observed when using SAF imaging (see (20) and Fig. S1). Near-membrane organelles are better visible, because SAF detection improves the axial optical sectioning from the excitation confinement of TIRF by adding an emission spatial filtering (Fourier-plane filtering). Thus, although both EW phenomena operate on the same length scale (of the order of 100 nm), SAF improves the axial sectioning and in part compensates for the imperfections of TIRF excitation.

Our observation that the measured RI value does not systematically change when the penetration depth of the EW is altered (Fig. S3) can be interpreted in three different ways. 1) The most straightforward interpretation is that there is no measurable difference in the local RI in the range probed by the two EWs. This would mean that on average, the local microenvironment of all SAF-emitting fluorophores was the same in these zones, or at least that the difference, if there is any, was smaller than the error bar of the RI measurement. 2) The subcellular localization of the fluorophores to specific organelles outweighs any z-effects. This would mean that their specific targeting exposes fluorophores to a local chemical microenvironment that merely reflects the organelle composition, and that this environment is little affected by precisely where this organelle is located with respect to the basal plasma membrane. Indeed, the mean RI values over three measurements for the vesicular, membrane, and mitochondrial RIs showed distinct and characteristic values for each type of organelle, but no significant difference at different beam angles. 3) Finally, yet another way of interpreting the lack of z-dependence is that our technique relies on the detectability of the r_c on BFP images, and therefore, by definition, those fluorophores that are located closest to the coverslip surface will dominate the RI measurement. A 20-nm increase in the axial volume probed will change the radiation pattern, i.e., the intensity emitted as SAF and UAF, but it will not change θ_c,fluo (or r_c,fluo).

Comparison with other subcellular RI measurements

Although the importance of the cell RI for cell biology, biomedical imaging, and disease diagnosis is well recognized (see, e.g., (27) for a recent review), local RI measurements in live cells are fairly rare, as are studies that specifically measure the RI in the near-membrane space. Due to their inherent surface selectivity, techniques that use optical effects at the dielectric boundary emerge as a natural choice. In their classical work, Bereiter-Hahn et al. (28) used quantitative RIC microscopy (29) to investigate how cell adhesion to the glass substrate locally modifies the RI. They found higher RI values (1.38–1.40) at points of focal contact, where stress fibers terminate, compared to areas of close contact (1.354−1.368). In areas of the cortical cytoplasm, between focal contacts, not adherent to the glass substrate, RIs between 1.353 and 1.368 were observed, but these numbers are likely to represent mixed values due to the finite volume probed by RIC. Also, in the absence of specific organelle labeling, RIC does not provide RI values for specific subcellular compartments. Although the work of Bereiter-Hahn et al. (28) prompted our variable-angle TIRF experiment at incidence angles of 68° and 73°, our results—slightly different from theirs—might be explained by the distinct cell types used, the smaller penetration depths in our case, or the fact that RIC images are difficult to interpret because the effects of the RI and substrate-membrane spacing both contribute to the local image intensity.

Another way to measure the near-membrane RI is to use the RI dependence of the critical angle for TIR, i.e., on the excitation site. In the past, several authors have noted the emergence of transmitted light propagating at very oblique angles from high-RI sites where TIR is locally disrupted (20, 30, 31, 32). Along the same lines, the steep dependence of the reflected intensity on the RI near the critical angle has been proposed to map the average RI from an angular scan of the reflected-beam intensity (33).

Other techniques that use the information contained in the reflected beam are digital holographic interferome-try which has been implemented in a TIR geometry (10, 34, 35). However, to the best of our knowledge, no cellular RI measurements have been reported. Ash et al. (36) later used this technique to characterize Amoeba proteus cells, as well as SKOV-3 cells (an ovary cancer cell line) and 3T3 fibroblasts.

Similarly, heterodyne interferometry uses the phase difference between s- and p-polarized components of totally reflected light (37) or the RI-dependent phase variation of p-polarized reflected light (38). Related work uses the beam profile (intensity map) of the totally reflected beam of cells cultured on a gold-coated coverslip to measure the surface-plasmon-resonance-generated contrast in reflection/absorption (39). However, an absolute calibration of RI is missing.

On the emission side, Enderlein and co-workers (18) used the BFP ring pattern to measure the effective NA of the microscope objective, an approach that we employed in our earlier work for characterizing the Zeiss objective used here (20) and that directly inspired this study. In this study, the procedure is just the inverse: instead of taking the BFP radii corresponding to the θ_c,fluo of water and air as a reference for calibrating r_max = r_NA,eff, we used r_NA,eff and a series of RI standards to calibrate BFP radii and took the radii measured for different organelle-specific labels for calculating the corresponding RI values. Building on the same idea, the Jaffiol lab proposed an approach for polar-incidence-angle (θ) calibration that is based on the analysis of ring patterns in BFP images and indirectly uses the RI information encoded in the critical angle (19, 40). Similarly, Soubies et al. (41) used different RI standards to obtain a relation between the corresponding θ_c values and BFP radii, which is then used for multiangle TIRF microscope calibration. These two studies are conceptually very close to ours in that they use the angle-radius correspondence, but they did not use the BFP information for cellular refractometry.

Using the dependency of the fluorescence lifetime on the local chemical microenvironment (42) of genetically targeted GFP fusion proteins, van Manen et al. (43) found RI values of ∼1.38 and ∼1.46 form cytosolic and membranous compartments, respectively. Similar cytoplasmic values (1.36–1.39) were obtained with tomographic phase microscopy, with the RIs corresponding to subcellular organelles (44).

Generally, our RI values seem somewhat smaller than the published values (although a large scatter exists in the literature). This systematic difference could be due to the fact that in our case, most dyes are exposed to mixed environments, i.e., lipid/protein-rich membrane on the organelle face but a more aqueous environment on the cytoplasmic face. Our values are generally closer to the mean cytoplasmic value (1.360 ± 0.004) found using quantitative phase-amplitude microscopy and confocal microscopy (45) than to values obtained using purified organelles. Table S4 compiles published values from endo-/lysosomal and mitochondrial RI measurements. We did not find RI values in the literature for the intact ER, but two older studies reported values ranging from 1.549 to 1.601 for “endoplasmic vesicles” (46) and from ∼1.405 to 1.415 for fractions 1–3 corresponding to the ER in (47) after subcellular fractionation and ultra-centrifugation, respectively.

Accuracy and precision of SAF-based RI measurements

How precise and accurate are our SAF-based RI measurements? Clearly, the quality of our SAF-based RI measurements depends on the accuracy and precision of measuring radii in the objective BFP, which in turn depends on the pixel calibration of the BFP image and the way radii are extracted.

We renormalized the BFP-image pixel size with the known RI values of air and water. Radii were fitted as described earlier (16) (Fig. S4), and the excellent agreement between the apparent and true objective focal lengths (Fig. 1) warrants the validity of our method. Overall, we estimate a 0.2% uncertainty in the precision of our RI measurement. A fundamental limit of our technique is that it reports the average RI of the local environments embedding all (near-surface) fluorophores across the entire field of view. As shown above, the presence of membrane-distant fluorophores that cannot emit SAF, but emit UAF only, will render the detection of the r_c ring more difficult. As such, fluorophore targeting and/or excitation-light confinement improve the spatial resolution of the measurement, as would, e.g., photoswitching and imaging of fluorophore subpopulations or confocal spot detection (23).

Finally, albeit trivial, fluorophore brightness and concentration must exceed the instrument and sample autofluorescence that eventually limits the sensitivity of our method (M.B. and M.O., unpublished data).

Perspectives

Can one go further in the analysis of the subcellular RI distribution? In this study, we used a standard wide-field TIRF microscope with the addition of a BL as the only modification, which has all the advantages and disadvantages of a conventional wide-field measurement. However, as SAF-based refractometry is compatible not only with TIRF—and specifically the now-preferred azimuthal-angle-scanning TIRF (17, 23, 24, 48, 49) used here—but also with confocal-spot-scanning TIRF (23), spatially resolved near-membrane RI measurements could be achieved using this geometry by scanning the sample with a piezo-driven xy-stage relative to the focused EW spot.

On the other hand, at least for thin samples or well-targeted fluorophores, highly inclined laminar optical sheet illumination (17, 50) and epifluorescence will allow very simple RI measurements at high time resolution. By simultaneous imaging of the sample and rear plane, it is possible, by separating the emitted fluorescence with an image splitter (“dual-viewer”) and adding a BL in one of the detection arms, to produce two partial images, one yielding the fluorophore distribution in the sample plane and the other the radiation pattern permitting the RI measurement. These techniques will have important applications in the biomedical sciences. By combining inexpensive LED illumination and SAF-based refractometry in a disposable biosensor, our technique will allow, e.g., low-cost on-chip diagnosis of cellular malignancy (51) or mitochondrial bioenergetics (52).

Author Contributions

M.B., L.R., and M.O. performed experiments and analyzed data. M.O. conceptualized and directed research and wrote the manuscript with contributions from all authors.

Editor: Christopher Yip.