2.1.

Lateral Shearing Interferometry

Lateral shearing interferometry is a technique to measure the phase gradient of a given light wavefront in one direction, from its replication into two identical but tilted copies. This offers the key advantage that it works without any use of a reference light beam. More precisely, the incident wavefront is replicated, thanks to a sinusoidal amplitude grating. After a few millimeters propagation, a mutual interference pattern is recorded with a video camera. In the case of a flat incident wavefront, the interference pattern gives regular fringes of period $p$ and hence produces two well-defined spots corresponding to the fringe frequency $1/p$ in the Fourier plane. In the case of an arbitrary wavefront, the interference gives a non-regular interferogram that presents local changes in the fringe spatial frequency with respect to the corresponding phase gradient of the incident beam.

However, reconstruction methods require the wavefront derivatives along two orthogonal directions to avoid error propagation and then to fully recover the actual two-dimensional phase distribution of the incident field. Multiwave interferometry solved the problem by offering a single-shot acquisition of at least two wavefront derivatives.¹⁸ In the particular case of quadriwave lateral shearing interferometry (QWLSI), four replicas are created by a specific two-dimensional diffraction grating, allowing two gradients along two orthogonal directions to be measured and then integrated to determine both the intensity and phase of the incident field.¹⁹ As for one-dimensional lateral shearing interferometry, one could use a sinusoidal two-dimensional amplitude grating to generate the four required replicas. In practice, the so-called Modified Hartmann Mask is preferably used because of its much simpler fabrication process: it is an only 3-level amplitude component that has been optimized to diffract more than 90% of the light energy into the selected replicated wavefronts, by properly choosing the ratio between the size of the square holes and the pitch of the mask.²⁰ Phase gradients are recovered in the Fourier space, by means of a deconvolution around the nominal interferogram fringe frequency $1/p$. Unlike classical interferometers, where a coherent reference arm is mandatory, lateral shearing interferometry is self-referenced, making it particularly insensitive to environmental vibrations, and achromatic.⁸ The wavefront sensor we used was a QWLSI commercially available product that has been specifically optimized for biological applications (SID4Bio, Phasics S.A., France). Such a model gives $300\times 400$ phase and intensity measurement points with a lateral pitch of 29.6 µm in the image plane.

2.2.

Experimental Setup

Combination of phase and epifluorescence imaging is generally interesting because of the complementary information the two modalities can offer. Optical path differences (OPD) provide structural information, while fluorescence reveals specific functional parameters about the biological system. Some works combining phase and fluorescence imaging have already been reported.^21,22 Here we used fluorescence in order to specifically identify sub-cellular compartments.

Figure 1 shows the experimental setup we used. The wavefront sensor substituted for a conventional camera plugged into a conventional inverted microscope equipped with a standard broadband halogen lamp as its light source. The microscope (Eclipse TiU, Nikon, Japan) also included an expanded space stratum structure providing an additional back port combined with a second fluorescence filter turret. That is how simultaneous quantitative transmission phase imaging and epifluorescence imaging was performed. A $750\pm 20\mathrm{nm}$ band-pass filter was added on the transmission illumination path before the sample in order to illuminate it in wide field with near infrared (N-IR) light. The upper filter turret carried a high-pass dichroic filter, reflecting light from 725 nm to around 1 µm and allowing light transmission from 350 to 725 nm. The second filter turret carried conventional fluorescence filter cubes to allow multi-marker fluorescence imaging. The QWLS interferometer was plugged onto the microscope back exit port to measure the exit wavefront on the image plane in the N-IR spectral band. A cooled intensity camera (DX2, Kappa opto-electronics GmbH, Gleichen, Germany) was plugged into one of the lateral microscope video port to record fluorescence coming from the sample. A 720 nm short-pass filter was added just before this camera to filter N-IR light that could remain. The two sensors were trigged to measure the fluorescence and the phase signal at the same time. In order to merge phase and fluorescence information and thus to obtain composite phase/fluorescence images, a calibration relative to residual tilt, lateral misplacement and differential magnification between the two sensors was required. An algorithm was then applied to extract the exact position, angle, and dilation of the phase image compared to the fluorescence one. In order to be fully diffraction-limited on the QWLSI sensor, we replaced the back tube lens (400 mm instead of 200). Consequently, the final magnification was 200 in phase for a $100\times$ objective, and still 100 in fluorescence.

Fig. 1

Schematic layout of the setup.

The typical irradiance of transmitted light at the sample plane was less than $5\mathrm{nW}/{\mathrm{\mu m}}^2$ for all the measurements presented in the manuscript. This level of exposure allowed for non-invasive continuous live-cell imaging over extended periods of time in the hour-level range. Fluorescence excitation was automatically cut between acquisitions to avoid photobleaching.

2.3.

Quantitative Phase Measurement Procedure

Let us define first the optical path difference (OPD) as

Once the quantitative phase information is obtained as described above, it is possible to post-process the actual phase image with the help of different types of numerical filters in order to enhance the phase contrast of target details. Typically, high-pass filtering in the spatial frequency domain is an interesting filter as it removes low-frequency information (global shape of the cell, nucleus) and increases the relative dynamics of much smaller details (organelles and membrane ruffles for example). High-pass filtering is tunable and does not present any decrease in the lateral resolution, but it does, of course, modify the global value of the local phase distribution and introduces artifacts. As a consequence, it must be banished in the case of quantitative measurements.

2.4.

Biological Samples and Models

In this paper, we mostly used COS-7 cells, adherent African green monkey kidney fibroblast-like cells. They are particularly interesting for studying membrane dynamics and intracellular movements.²³ They are also easy to transduce. They were seeded in a 2-well Lab-Tek (Nunc, Roskilde, Denmark) and transduced with CellLight reagents (Life Technologies, Saint Aubin, France) the day before experiment in order to specifically label given organelles with fluorescent fusion proteins. Imaging and cell cultures were done in Dulbecco's Modified Eagle Medium, supplemented with 5% fetal calf serum and sodium pyruvate. For long imaging experiments, cells were placed in a 37 °C box over the microscope, with a small incubation chamber providing humidified air with 5% carbon dioxide.

We also used a periodical microstructured etched fused silica glass as a model solid sample. Surface topography was determined by atomic force microscopy [Fig. 2(a)], showing an etching depth of 30 nm. The refractive index of the fused silica used (ACM, France) was 1.453 around the 770 nm central wavelength of the measurement beam.

Fig. 2

Measurements on a calibrated etched fused silica sample. (a) Topographic measurement by atomic force microscopy; (b) quantitative OPD map of the sample; (c) histogram of OPD values on each pixel of image (b); and (d) comparative results between theoretical and experimental OPD of the sample in several solutions with different refractive indices. Scale bar: 10 µm.

3.1.

Quantitative Phase Imaging on a Calibrated Model

In order to characterize resolution and reliability of our system, we first tested it on solid calibrated structures. They were embedded in liquid media with known but varying refractive indices and could work as a model of biological features of interest. More precisely, we successively measured the OPD image of our silica model sample when embedded in different liquid media (pure water, water with varying glucose concentration, and immersion oil), with known refractive indices determined, thanks to an Abbe refractometer (2WAJ, HuiXia Supply, China). Figure 2(d) shows the highly coherent results that we obtained when comparing the theoretical OPD given by Eq. (1) with the experimental data. Such an experiment also enables us to determine the noise level: Fig. 2(b) gives the OPD map as measured, and Fig. 2(c) gives its OPD histogram. The sharpness of this histogram leads us to define the error bars of Figure 2(d) as $\pm 0.5\mathrm{nm}$. Full width half maxima of the two histogram peaks also give $\pm 0.5\mathrm{nm}$ as the typical noise level in the OPD images.

3.2.

Optical Detection of Cellular Features

Fixed human neurons have been imaged (Fig. 3) in order to show that QWLSI is a well-adapted tool to detect and observe thin membrane filaments, by exhibiting a contrast that is comparable to that obtained with fluorescence imaging. The advantage of the phase technique is that it does not require fluorescence labeling that can be saved for functional imaging.

Fig. 3

Fixed human neurons images. (a) Green fluorescent protein (GFP) fluorescence; (b) raw QWLSI phase contrast image; and (c) high-pass filtered sub-image with lookup table saturation to provide enhanced useful contrast. ($40\times$, N.A. 1.3.) Scale bar: 10 µm.

Another interesting point of phase contrast microscopy is that one can detect subcellular features as soon as they present a refractive index that slightly differs from that of the cytosol. In that case, using high-pass filtering in the spatial frequency domain of phase images increases the resulting contrast and therefore enables the clear detection of the target features. As an example, Fig. 4 shows an 18-min long sequence of filtered QWLSI phase images of a HeLa cell mitosis, revealing different phases of the cell-division cycle. In Fig. 4(b), the cell is in metaphase, where spindle fibers align chromosomes in the middle of the cell, along the metaphase plate. Figure 4(c) shows anaphase, where chromatids separate and move to opposite sides of the cell. Figure 4(d) corresponds to telophase, where chromatids arrive at the opposite poles and new membranes form around the daughter nuclei. Finally, cytokinesis appears [Fig. 4(e)], where two daughter cells are now individualized.

Fig. 4

Sequence of filtered QWLSI phase images recorded during HeLa cells mitosis. (a) Full field image, and (b) to (e) zoom on the cell of interest. (b) Metaphase, (c) anaphase, (d) telophase, and (e) cytokinesis. ($40\times$, N.A. 1.3.) Scale bars: 20 µm.

3.3.

Quantitative Phase Imaging on Biological Samples

We now present different ways to study living cells from quantitative OPD images obtained with the QWLSI technique.

With the quantitative raw image of Fig. 5(a), we measured OPD in different parts of a COS-7 cell with a high signal to noise ratio [SNR; Fig. 5(b)]. We also combined fluorescence and high-pass filtered phase images, simultaneously, to detect specific organelles in real time, as for peroxysomes [Fig. 5(c) and 5(d)]. Additionally, we simulated DIC images by simply calculating the gradient of the OPD map in a given direction [Fig. 5(e) and 5(f)], thanks to its quantitative artifact-free nature.

Fig. 5

Different ways to use QWSLI images to look at a COS-7 cell. (a) Quantitative phase image with a phase level sectioning (red arrow). (b) Result of this sectioning showing a really good SNR. (c) Fluorescence image of peroxysomes obtained simultaneously with a video camera, combined in (d) with filtered phase image. (e) and (f) simulated post processing DIC images from a single acquisition along two perpendicular axes [(e) 0° and (f) 90°]. ($100\times$, N.A. 1.3.) Scale bar: 10 µm.

Figure 6 shows experimental results obtained with transiently transduced COS-7 cells that expressed fusion proteins specific of mitochondria or actin (CellLight, Invitrogen). Observations were made with a $100\times$, $\mathrm{NA}=1.3$ immersion objective in order to visualize a single cell with a good lateral resolution. Post-processing high-pass filtering was used to increase the phase contrast of target organelles, such as mitochondria, actin stress fibers, or vesicles [Figs. 6(a), 6(d), and 7, respectively]. However, as described in Sec. 2.3, OPD measurements were made on raw images.

Fig. 6

Quantitative phase measurement in COS-7 cells. (a), (b), and (c) Mitochondria tubular network focusing. (d), (e), and (f) Actin stress fiber focusing. (a) and (d) High-pass filtered phase image with phase level sectioning (red lines). (b) and (e) Phase level sectioning along the red lines defined in (b) and (d) but from raw images (not shown). (c) and (f) Simultaneous GFP fluorescence images, needed to localize labeled organelles. ($100\times$, N.A. 1.3.) Scale bar: 10 µm.

Fig. 7

Quantitative phase measurement of lysosomes compared to others vesicles, in COS-7 cells. (a) An example of composite phase/fluorescence images (red: RFP, lysosomes), with a zoomed insert (b). (c) and (d) Phase level sectioning of vesicles along the red and white dotted lines drawn in (b). (e) Relative refractive index ($\mathrm{\Delta }n/\mathrm{diameter}$) distribution of fluorescent lysosomes and other non-fluorescent vesicles (average values and standard deviation are drawn), obtained from 70 measurements on independent vesicles of around 1-µm diameter, showing a significant difference between the two populations.

Mitochondria are not always individualized, but can form a tubular network [Fig. 6(a) and 6(c)]. Identification of mitochondria can be achieved by using morphological criterion on OPD images. Relative OPD of tubular and horizontal mitochondria (compared to the cell background) is quite constant in a given cell [about 6 nm for a diameter of 1.2 µm, see Fig. 6(b)]. Nevertheless, we are unable to quantify OPD of a specific organelle in very dense area, such as the perinuclear zone, where the endoplasmic reticulum and Golgi apparatus are located. We did not detect clear individual mitochondrion in our cells, neither in fluorescence nor in phase.

We then applied the technique to a cytoskeleton study. Actin stress fibers are made of many polymers of actin and are synthesized by the cell in order to keep its structure and to move on the surface [Fig. 6(f)]. Though monomeric actin is not big and dense enough to be seen with our technique, its polymeric form can be detected with the actual setup [Fig. 6(d)]. Figure 6(e) shows an OPD of around 1.5 nm for the biggest fibers, close to the sensitivity limit of the technique ($\mathrm{SNR}=3$).

Figure 7 demonstrates the possibility of determining specific vesicles by only using the quantitative phase-contrast technique. We indeed labeled lysosomes on COS-7 cells with red fluorescent protein fusion proteins [Fig. 7(a)], and measured 32 fluorescent lysosomes and 38 other non-fluorescent vesicles of similar diameters, taken from five different living COS-7 cells [examples in Fig. 7(b) and 7(c)]. The results [Fig. 7(d)] clearly present two significantly different vesicles groups with respect to their relative refractive index $\mathrm{\Delta }n$, where $\mathrm{\Delta }n$ is the difference between the refractive index of the considered vesicle and that of the surrounding medium. Reaching $\mathrm{\Delta }n$ was possible by assuming the vesicles were both homogeneous and spherically shaped, which is often but not always the case.