Dynamic full-field optical coherence tomography
Nature
Abstract
Optical coherence tomography offers astounding opportunities to image the complex structure of living tissue but lacks functional information. We present dynamic full-field optical coherence tomography as a technique to noninvasively image living human induced pluripotent stem cell-derived retinal organoids. Coloured images with an endogenous contrast linked to organelle motility are generated, with submicrometre spatial resolution and millisecond temporal resolution, creating a way to identify specific cell types in living tissue via their function.
The comprehension of the human body and its mechanisms at the subcellular scale is still an open area of research. During the seventeenth century, the first examinations of life under the microscope were conducted directly on humans, animals and bacteria1. Then, at the end of the nineteenth century, cell culture studies began to replace in vivo studies, as this allows the creation of in vitro models beneficial for the comprehension of biological phenomena in different environments2,3. Because of the two-dimensional nature of early cell cultures, the possibilities of understanding tissues and organs as a whole were limited. Recently, three-dimensional (3D) cultures have been developed from stem cells to generate organoids that mimic a variety of tissues and serve as models of human development4 and disease studies5,6,7. Organoids could also serve as sources of human tissues for transplantation and as platforms for drug screening8,9,10. These self-organizing structures develop cellular compositions and architectures similar to in vivo tissues, thereby replicating biologically relevant intercellular phenomena in vitro11,12.
For each biological trend, optical imaging devices have been developed and optimized to image tissues, cell cultures and, recently, organoids, which are one of the most fundamental tools in biology, clinical pathology and medical diagnosis13. There are many challenges in imaging 3D structures: due to their relatively transparent nature, it is difficult to obtain contrast on specific structures without staining. Moreover, 3D samples require optical sectioning to discriminate the layer in focus from out-of-focus layers. In this study, we present dynamic full-field optical coherence tomography (D-FFOCT) as a technique to image retinal organoids derived from human induced pluripotent stem cells (hiPSCs)14, which are a major breakthrough in the study of the retina and retinal diseases15,16,17. These hiPSC-derived retinal organoids are routinely imaged with various techniques (see Supplementary Table 1). However, each of the existing methods presents major drawbacks, such as the need for fixation or mechanical sectioning, rendering the study of dynamic phenomena impossible; the need for labelling, requiring cumbersome and costly preparation; or a lack of functional contrast, indicating only cell presence and not cell health or behaviour18,19,20. Optical coherence tomography (OCT) is commonly used in biology and medicine to obtain 3D images of microstructures in tissue. OCT contrast arises from the local endogenous optical backscattering level21. The main drawback of traditional OCT is the trade-off between imaging depth and resolution. To increase the lateral resolution, the numerical aperture of the system must be increased. As a consequence, the depth of field decreases, and only a small layer of the sample can be imaged. Current OCT systems therefore have a lateral resolution on the order of 10 µm, which is insufficient to resolve cell structures laterally. Using an incoherent light source and a camera, time domain full-field OCT (FFOCT) is an en face variant of OCT with a higher spatial and temporal resolution than OCT in the en face plane22. As FFOCT acquires an entire en face plane rather than a line in the depth dimension, the numerical aperture can be arbitrarily increased without any imaging depth trade-off. Using the FFOCT experimental setup shown in Fig. 1(a) and detailed in the Methods section, a novel contrast mechanism has recently been exploited by measuring temporal fluctuations of back-scattered light in a technique called dynamic FFOCT (D-FFOCT)23. By revealing sub-cellular structures that have very weak back-scattering, these dynamic measurements provide contrast based on local intra-cellular motility24,25 with sub-micrometre resolution, and can achieve millisecond temporal resolution to study fast phenomena.
Acquisition of images a, Full-field OCT setup combined with wide-field fluorescence microscopy (side view). PZT piezoelectric translation, TS translation stage, BPF bandpass filter, HPF high pass filter. As fluorescence is recorded in the same spectral range as that used for D-FFOCT, D-FFOCT and fluorescence images could not be recorded at the same time; for this purpose, a flip mirror was added to switch from one modality to another. b 3D cube of data (x, y, t) acquired before processing. Each time evolution of a pixel is processed independently. c An intensity trace is plotted for a pixel inside a living retinal organoid. Post-processing steps d–f. Dynamic images are computed in the HSV colour space. d Hue is computed with the mean frequency, from blue (low temporal frequencies) to red (high temporal frequencies). e Saturation is computed as the inverse of the frequency bandwidth; as a consequence, a signal with a broader bandwidth (e.g., white noise) appears dull, whereas a signal with narrow bandwidth appears vivid. f The value is computed as the running standard deviation23. Bottom row is a D29 retinal organoid. g Computation of the mean frequency (hue); h, frequency bandwidth (saturation); and i, dynamic (value) before j, reconstruction. Scale bar: 50 µm
In FFOCT, light coming back from the sample slice of interest interferes with light coming back from the reference mirror and is projected onto the camera (Fig. 1(a)). To compute a D-FFOCT image, a movie (typically 512 frames) of the interferogram pattern is recorded and processed (see Methods) to extract local fluctuations and render them coloured (Fig. 1(b–j)). Using the hue-saturation-value (HSV) colour space, image brightness is linked to fluctuation amplitude, whereas colour is linked to fluctuation speed, from blue (slow) to red (fast) through green (in between). By translating the sample in the axial direction to acquire a stack of planes, see Supplementary Vid. 1, a 3D volume can be reconstructed (see Methods). Alternatively, a series of dynamic images may be acquired in the same plane to follow the evolution of activity over several hours in a time-lapse fashion with a temporal resolution of up to 20 ms (see “Methods”).
A 3D reconstruction of a 28-day-old (D28) retinal organoid is depicted in Fig. 2(a), corresponding to an optic vesicle stage during retinogenesis9,10,15,16, along with a sub-volume in Fig. 2(b), highlighting the layered internal retinal progenitor cell organization. A cross-section is shown in Fig. 2(c), in which the elongated shape of cells is seen. A time-lapse video at 50 µm depth was acquired on the same organoid to study its temporal evolution over three hours (see Fig. 2(d) and Supplementary Video 2 for the full recording). In these acquisitions, different dynamic profiles of cells can be observed: surface cells exhibit faster dynamics than those inside the sample volume. This could be explained by the fact that at the surface of the organoid, the cells are in contact with the external environment, making them more vulnerable to change and often leading to their death. In Fig. 2(d), cells in the centre of the organoid exhibit fast and intense activity until their disappearance, possibly indicating that they are undergoing apoptosis. Evolution of cell dynamics near a clear boundary between two distinct types of cells is also visible. On one side of the boundary, cells exhibit faster and stronger dynamics, suggesting a differentiation process towards specific retinal lineages15, and on the other side, small rounded progenitor cells have slower activity. These two cell types are therefore distinguishable by their dynamic profile alone. Generated D-FFOCT images present a consistent colormap in which each frequency is continuously represented by the same colours; therefore, similar results are obtained for different retinal organoids at the same developmental stage. Supplementary Video 3 shows a time-lapse movie of the D28 retinal organoid shown in Fig. 2(a–c) alongside a D29 retinal organoid. The same clear boundary between distinct types of cells is present for both. By processing the data contemporaneously on the GPU using a modified version of Holovibes software26, an enhanced temporal resolution of 20 ms was achieved, which represents a 500-fold improvement without the need to store the raw data (up to 4 Go.s−1). The price paid for this improvement is the use of an alternative version of the dynamic computation, which is noise-sensitive and non-quantitative (see “Methods”). Figure 2(f–h) shows high temporal resolution (20 ms) images of a D147 retinal organoid. The typical rosette organization of retinal cells, previously described14,15, is visible in the centre, i.e., photoreceptors (seen from the side) in the inner part of the rosette centre with emerging outer segments (indicated by white lines on Fig. 2(h)), and other surrounding retinal cells are evident. Photoreceptor nuclei exhibit different dynamic profiles (Fig. 2(g))—either compact and uniform, inflated, or absent—which may correspond to the nuclear G0/G1, dying and M states, respectively. The gain in temporal resolution allows the study of fast biological processes such as organelles moving inside the cytoplasm (see Supplementary Video 4). A series of retinal organoids imaged by D-FFOCT at consecutive steps of development showed the gradual differentiation of retinal cell progenitors into neural cells and photoreceptors (Supplementary Fig. 2), as validated by comparison with a similar organoid series imaged with immunofluorescence on a confocal microscope14.
a 3D reconstruction of the spherical D28 retinal organoid, composed of cells ~5 µm in diameter. Red arrows highlight surface cells exhibiting fast dynamics. b Image represents a sub-volume of a (blue square). c Image represents a cross-section in (a) (green dashed line) in which one can see the organization of the layers inside the retinal organoid. d High-magnification images of two different areas of the organoid during a 3h time-lapse acquisition: magnified images in the top row show the change in dynamic profile that could reflect a differentiation process (the boundary between the two types of cells is represented by a red dotted line); images in the bottom row show a very active zone composed of cells exhibiting fast and high dynamics, possibly undergoing apoptosis, in the centre of the organoid. e Colour bar of the D-FFOCT images for the 3D and time-lapse acquisitions with a consistent colormap for (a–d). High-temporal-resolution imaging performed on a D147 retinal organoid. f Part of the retinal organoid revealing fusiform structures corresponding to emerging photoreceptor outer segments in the centre of the rosette. g Magnified view of nuclei in three different states around the rosette: (i) a nucleus in a normal state with a compact, uniform shape and is very bright (i.e., exhibiting a high activity); (ii) an seemingly dying, inflated nucleus, exhibiting almost no activity; and (iii) a nucleus undergoing division with no defined nuclear membrane in the cytoplasm, and two distinct parts (white arrows) of the content of a nucleus (suggesting mitosis of the nucleus with chromosomes already divided, with the same subcellular activity level as the “normal” nucleus). h Magnified image of the photoreceptor outer segment-like structures imaged side-on; three of them are marked with a white line. Scale bar: 20 µm.
To further validate the D-FFOCT signal origin via direct comparison between D-FFOCT and specific fluorescence labelling in the same organoids, a multimodal setup was developed that combines D-FFOCT and fluorescence channels to allow a pixel-to-pixel overlay of D-FFOCT and fluorescence images. A D29 retinal organoid was labelled with a dye targeting the nuclei of dead cells (see “Methods”). D-FFOCT images overlaid with fluorescence wide-field images are shown Fig. 3(a–d). Two fluorescent red spots are clearly visible (Fig. 3(a)) and correspond to very weak dynamic signals in the D-FFOCT image, confirming that dead cells exhibit low activity and that the contrast revealed by D-FFOCT is metabolic. These two areas are magnified in Fig. 3(c, d), and dark zones are encircled by a white dotted line. A retinal organoid was imaged with the combined D-FFOCT-fluorescence system at D126, via which a large number of differentiating photoreceptors can be detected in rosette-like configurations (Fig. 3(e, f)). A red fluorescent zone corresponding to photoreceptors is visible in Fig. 3(e), whereas in the D-FFOCT image (Fig. 3(f)), the different activity level in photoreceptors compared to surrounding cells is sufficient to provide distinction of the cell type through the dynamic signal alone, across a region that is coincident with the fluorescently labelled zone. Underlying biological processes responsible for the dynamic contrast, which has been shown to be linked to cell metabolism23, could include movement of organelles. Preliminary experiments suggest mitochondrial contribution, but other potential contributors, such as Golgi bodies, lysosomes, vesicles, and pigments, have not yet been discarded.
D-FFOCT images are depicted in colour (b–d, f), D-FFOCT images overlaid with wide-field fluorescence images are depicted in greyscale (D-FFOCT) and red (fluorescence) (a, e). a–d D29 retinal organoid labelled with dye targeting the nuclei of dead cells. a One can see two dead cells marked by the red spots, corresponding to the two dark zones on (b), the D-FFOCT image, in which there is no dynamic signal in these zones. c, d Magnified images of the two dark zones (highlighted by a white dashed line, corresponding to the two red spots of fluorescence). e, f D126 retinal organoid derived from a fluorescent cone rod homeobox (CRX) reporter iPSC line exclusively labelling photoreceptors in red (mCherry). e Overlaid image on which the photoreceptor fluorescence matches the blue-green cells of (f), the D-FFOCT image. These areas are highlighted by a white dotted line. These precursors of photoreceptors have their own particular dynamic signature, which allows them to be distinguished from the surrounding cells by D-FFOCT alone. Scale bar: (a–d) 10 µm, (e, f) 50 µm
With D-FFOCT, once installed in the culture lab under controlled environmental conditions, we anticipate being able to follow retinal organoid developmental processes that have been documented by traditional methods14,19, such as the quantification of cells exhibiting the same dynamic profile at multiple time points. In addition, we should also gain additional biological insights owing to the non-invasive nature of our imaging method, as it provides access to the time course of continuous processes such as progenitor cell proliferation and migration, cell type differentiation including evolution of the boundary between neural and non-neural retinal cells (Fig. 2(d)) and evolution of the organoid into layered retina. Interestingly, combining D-FFOCT and a 3D organoid-based model of retinal dystrophies due to patient-specific iPSCs offers the opportunity to understand fundamental subcellular processes leading to diseases in a way that was not previously possible.
Dynamic FFOCT imaging creates a new label-free non-invasive contrast for imaging retinal organoids. As this technique does not damage the samples, it complements and could potentially replace the imaging modalities traditionally used. D-FFOCT allows imaging of different layers at multiple depths while preserving the sample integrity, i.e., using neither exogenous labelling nor destructive methods, and is therefore suitable to follow the evolution of the same organoid at different stages of its development. The high dimensionality of the probed signals (512 interferograms per pixel) is useful for developing statistical approaches such as automated classification and clustering, and we are working towards the identification of cells via their D-FFOCT contrast alone using machine learning algorithms, with the ultimate goal of removing the need for markers entirely. The only missing part, for now, is the lack of ground truth validation data (e.g., segmented cells with labels that could be generated by fluorescence or by annotating experts), which will be a milestone in the further development of this technique.
