Fast interrogation of intestinal organ-on-chip devices through inverted single-plane illumination microscopy (iSPIM) using an electro-tunable lens

Abstract

We have developed an inverted single-plane illumination microscope adapted for the interrogation of intestinal epithelium organ-on-chip devices. In this kind of system, samples are normally scanned in a slow fashion through motorized stages. Here, we have engineered a faster sample remote focusing scan mode by the use of an electro-tunable lens synchronized with Galvo mirrors, overcoming some limitations of previous systems. In addition, we have also engineered and optimized scaffolds recapitulating the 3D topography of the intestinal epithelium. Finally, we evaluated the performance of the system using fixed and in-vitro intestinal epithelium organ-on-chip devices, demonstrating the possibility of live imaging for several hours without distorting sample evolution.

1. Introduction

The intestinal epithelium is one of the most important tissue barriers and the major absorption site in the body. The small intestinal epithelium is one of the fastest proliferating tissues in the body with a turnover rate of 4-6 days in human [1]. This is fueled by the proliferation of stem cells present in the invaginations, called crypts. Their progeny migrates up along the finger-like protrusions, called villi, while differentiating to form the mature epithelium. This phenomenon is tightly controlled by gradients of biochemical factors present along the crypt-villus axis provided by the cellular microenvironment that leads to the compartmentalization of proliferative cells in the crypts, and differentiated cells in the villi. Imbalance between proliferation and differentiation can lead to hyperproliferation and ultimately tumor growth [2]. Indeed, colorectal cancer has now become the most prevalent cancer in men and the second most common in women [3,4].

Functional in vitro models of intestinal epithelium have been pursued for a long time. They are key elements in basic research, disease modelling, drug discovery, and tissue engineering and have become prime models for adult stem cell research [5]. The identification of Lgr5 + intestinal stem cells (ISC) and advances in their culturing methods have made it possible to create in vitro near-physiological tissues called ‘organoids’ [6]. Organoids are 3D structures formed by ISCs, which recapitulate some of the tissue functionality in vivo such as distinct proliferative/differentiated cell compartments and a central lumen region. Despite the huge leap, organoids have represented in ex vivo models of intestinal epithelium, organoids are 3D closed spheres, thus providing difficult access to the lumen compartment and therefore hampering their potential applications in drug screening or personalized medicine. On top of this, because they are stochastically developing tissues, they exhibit great morphological variability between them [7]. This inherent heterogeneity scales up by the fact that they require to be embedded in Matrigel, an animal-derived matrix with high batch-to-batch variability.

Synthetic extracellular matrix (ECM) analogues, on the other hand, pose as good alternatives to their native counterparts, as they provide well-defined, controlled microenvironments [8–11]. Therefore, there is a need for engineering culture platforms that provide a physiologically relevant environment, while preserving the stem and differentiated cells present. The current engineered culture platforms of the intestinal epithelium are limited as they fail to combine all the key features of intestinal epithelium, such as distinct stem/proliferative and differentiated cell types, 3D architectural features, and gradients of ISC niche biochemical factors. From the bioengineering point of view, we have developed scaffolds with finger-like protrusions that mimic the villus structures found in the in-vivo gut tissue. The microfluidic device was composed by two chambers separated by a membrane and the 3D scaffold. Biomolecules identified as stemness factors, were delivered at the basal (lower) compartment and diffused to the apical (upper) compartment, therefore establishing in vivo-like biochemical gradients through the 3D scaffold. These gradients were employed to foster cell positioning along the vertical axis of the villus structures, aiming the confinement of stem and proliferative cells at the bases (crypts), and the positioning of differentiated cells at the tip of villi.

The analysis of these big and complex 3D structure, highly differentiated and dense samples require advanced imaging techniques in order to extract relevant information. In the last decade, Light Sheet Fluorescence Microscopy (LSFM) (also referred as Selective Plane Illumination Microscopy – SPIM [12]) has shown its unique capabilities overcoming many of the limitations of confocal microscopy. Among them, high resolution and sensitivity, wide field of view, fast acquisition and minimal photo-damage and photo-bleaching, make it ideal for 3D long-term multicolor live imaging of biological processes [13]. Nowadays, there exist a huge variety of LSFM configurations, each of them with their pros and cons [14].

Various imaging methods have been used to image organ-on-chip (OOC) platforms, including brightfield, phase-contrast, confocal microcopy, and light-sheet fluorescence microscopy, among others [15,16]. While several LSFM systems have been demonstrated for single-cell imaging flow cytometry microfluidic devices [17,18], relatively few LSFM imaging systems have been presented for cell culture OOC platforms. Some examples include a specific system tailored for microfluidic chips to study vasculature [19] or LSFM systems for the characterization of intestinal epithelial scaffolds [20,21].

We have recently adapted our light-sheet-based imaging platform (called Flexi-SPIM [22]) for quantifying the temporal evolution of biochemical gradients of the ISC niche in intestinal epithelium hydrogel-based scaffolds [23]. As we will show, this system has serious limitations in terms of chip area that can be scanned, the OOC size, as well as the ease of interchangeability between different chips.

To properly interrogate such OOC devices, we have designed and built a novel LSFM system based on the inverted selective plane illumination microscopy (iSPIM) setup [24]. From the optical point of view, to offer faster scanning speeds, we have engineered a LSFM system that uses a galvanometric mirror in the illumination path, which moves the light sheet perpendicularly to the direction of propagation, synchronized with an electrically tunable lens (ETL) in the detection path, modifying the focus plane of the detection objective [25]. In conjunction with XY stages this fast-scanning mode allows fast inspection of big organ-on-chip areas (5 mm x 5 mm) with cellular resolution in just 3 minutes per channel. Here we present a fully automated platform for fast interrogation of complex organ-on-chip devices, providing multicolor 3D live imaging of the samples over several (4 to 5) hours. We have also explored and compared two different iSPIM scanning modalities: traditional slow-scanning mode (sample translation across a static light-sheet using XY stages) and our fast-scanning mode (light-sheet scanning across a static sample using galvo mirrors and an ETL). In this paper we present the development of core technologies to assess the in vivo-like functional properties of engineered intestinal epithelial tissues, combining light-sheet fluorescence microscopy (LSFM) imaging with a custom-made microfluidic device to visualize organoid-derived monolayers on 3D synthetic scaffolds.

2. Materials and methods

2.1. General description of the set-up

We have built our iSPIM set up on top of a Nikon Eclipse body, as shown in Fig. 1(a), where we have modified some parts in order to attach the objective holding block, a breadboard containing the illumination and detection objectives, the filter wheel and the tube lens. On the illumination path (Fig. 1(b)), we use an ArKr diode laser (Innova 70C Spectrum, Coherent) as light source, providing 488, 514, 561, 638 nm lines. Wavelength selection and laser power control is performed by an AOTF (AOTFnC-400.650-TN, AAOptoelectronics). We expand the beam through a Galilean telescope, made of two achromatic lenses (L₁ = AC254-050-A-ML and L₂ = AAC254-150-A-ML, Thorlabs), and direct it towards a 75 mm focal distance cylindrical lens (LJ1703RM-A, Thorlabs). This section of the set-up is mounted on an elevated breadboard, in order to progressively raise the beam up the entrance port of the objective holding block. Then, the beam is focused by this cylindrical lens onto a galvanometric mirror (GVS002, Thorlabs), which in turn is projected through a second Galilean telescope (L₃ = AC254-150-A-ML and L₄ = AC254-200-A-ML, Thorlabs) onto the back focal plane of the illumination objective. In that way, the cylindrical lens will generate the light-sheet plane that will be translated across the sample by the galvanometric mirror. For illumination we use a 10x water immersion objective (0.30 NA, 3.5 mm WD, Nikon CFI Plan Fluorite). An adjustable mechanical slit (VA100C, Thorlabs), inserted after the first telescope, allows reducing the illumination numerical aperture, thus providing thicker, but more uniform light-sheet illumination, needed for large fields of view (see Supplement 1 ^{(2.1MB, pdf)} Fig.1).

Fig. 1.

(a) General design of the iSPIM system with ETL refocusing. (b) Orthogonal views of the illumination path, designed to ensure a thin light sheet and a large field of view. (c) Design for the detection path. When the light sheet aligns with the image plane, the ETL has no effect (top). However, when the light sheet does not align with the image plane, it causes a displacement “d” in the image formation, and the ETL compensates for this displacement (bottom). In both cases, the images are generated in focus in the camera.

The detection path is composed of two parts (Fig. 1(c)). The first one, on the objective holding block, contains the detection water immersion objective (10x, 20x or 40x), a motorized filter wheel (FW103 H, Thorlabs), and a 200 mm tube lens (L₅ = TTL-200-A, Thorlabs), attached using different cage system elements (Thorlabs). The second one consists on a periscope system containing two high quality achromatic lenses (L₆ = AC508-200-A-ML and L₇ = AC508-100-A-ML, Thorlabs) and the electrically tunable lens (ETL) (EL-16-40-TC-VIS-20D, Optotune), resulting a 0.5x demagnification to reduce the number of pixels per field of view, in order to speed up the acquisitions, normally limited by the sCMOS’ camera frame rate (OrcaFlash4 v.3, Hamamatsu). All the optical elements are positioned correctly to ensure proper conjugation. It’s important to notice that the actual configurations of the setups using 10x 0.3 NA water immersion objective for detection, provides 5x magnification with 2.6 µm resolution. The use of the ETL, a device that modifies the curvature of its surfaces as a function of the applied voltage, allows us to modify electronically its focal distance from -10 to 10 diopters. In our system, the ETL provides fast and precise refocusing of the image, following the scanned light-sheet plane and generating a clear (i.e., focused) image in the plane of the camera.

To quickly record the large amount of generated data without any data loss, we use PCIe solid-state disks. This allows recording at a speed higher than 196 MB/s. Moreover, the computer is equipped with a fast data acquisition card (PCIe-6363, National Instruments) to generate the drive voltages for both the ETL and the galvanometric mirror, as well the synchronization TTL train for the camera. We have developed custom-made software tools to automate image acquisition (LabView, National Instruments). The image processing was performed using Fiji software [26], especially the TransformJ plugin [27] and for data representation we used ClearVolume plugin [28].

2.2. Microfluidic device and sample mounting system including incubation chamber

We have microfabricated poly(ethylene) glycol diacrylate (PEGDA) based scaffolds recapitulating the 3D topography of the intestinal epithelium (Fig. 2(a)) and able to support the formation of primary intestinal monolayers (Fig. 2(b-d)). This hydrogel material has tissue-like stiffness and proper porous size to allow the diffusion of key biochemical factors for intestinal stem cell proliferation and differentiation. The scaffolds were functionalized with collagen type I to promote cell adhesion and was combined with a microfluidic device, which, through a buried channel, allowed the delivery of such biochemical factors in a controlled way, establishing gradients of signaling along the vertical direction (Fig. 2(b)) [29]. Alternatively, we have also used simplified setup that would permit facile testing of the complex biological system, the standard Transwell setup, which separated the culture well into two compartments: lower (source) and upper (sink) compartment. The microstuctured hydrogels were fabricated on porous membranes were functionalized with collagen type I and mounted onto the Transwell inserts. The gradients were generated by adding the ISC niche biochemical factors only to the lower (source) compartment. Culture medium without the factors was added to the upper compartment that acted as the sink. The main goal was achieved by culturing intestinal organoid-derived cells (enriched for ISCs with a special medium composition) on the scaffolds bearing the gradients.

Fig. 2.

(a) Crypt-villus structure of the small intestinal tissue where cell compartmentalization is guided by gradients of biochemical factors (Wnt2b, Noggin, R-spondin). (b, c) Scheme of the microfluidic device that contains a mini-cell culture well to allocate the fitting of the hydrogel scaffold. A porous membrane and a buried microchannel to permit the diffusion-mediated delivery of biochemical species in a gradient fashion. (d) Picture of the fabricated chips. (e) Scheme of the designed incubation chamber for iSPIM imaging, including the XY motor, Z motor and heating plate. (f) Detail picture of the sample mounting. (g) Evolution of the room (blue) and sample (red) temperature.

Compared with the classical LSFM architecture, the iSPIM approach offers more flexibility for sample loading, allowing to insert the samples mounted on traditional MatTeck dishes. We have also implemented on our iSPIM system a new mounting protocol adapted for live imaging (Fig. 2(e)), providing cell viability conditions (i.e. temperature and CO₂ control). We have designed a Z axis motorized (NewFocus) 3D printed platform, attached on top a motorized XY stage (NSA12, Newport), to control sample positioning. Moreover, 3D printed holders, compatible with MatTeck plates, were designed in order to properly hold and immerse the generated microfluidic devices (Fig. 2(f)) and water dipping objectives. To maintain the microfluidic device at 37°C, we designed a proportional-integral-derivative (PID) temperature controller connected to a thermoresistor to heat an aluminum plate on the base of the imaging platform. Maintaining the heating plate at 43°C resulted in a sample medium temperature of 37°C at the sample position after 10 minutes, and it remained stable over several hours (Fig. 2(g)). To perform long-term imaging of living cells, proper CO₂ incubation is needed, which also provides sample isolation to avoid contaminations and medium evaporation. We designed a flexible rubber enclosure chamber, able to adapt the 45° orientation of the iSPIM objectives. We 3D printed the masters and used dental mold polymers (Twinsil soft, PicoDent), which only polymerizes after the mixing of two chemicals, to create this chamber. A flow of 5% CO₂ is introduced into this enclosure chamber maintained the medium pH almost constant along the 8 hours of experiments.

3. Results

3.1. Comparison between modalities: slow XY stage scanning vs. fast ETL scanning

Imaging complex and large 3D structures under a classical LSFM configuration is always challenging due to the physical constrains imposed by the two orthogonal objectives architecture and its respective working distances. As a first approach, we tried to image these chips using our Flexi-SPIM [22] multimodal microscope (Fig. 3(a)), where illumination is performed horizontally to the optical table by a 4x air objective and detection in an up-right configuration with water dipping 10 x objectives. To avoid that the first rows of pillars of the sample create an occlusion of the excitation laser beam, preventing to image the following rows, we needed to tilt the sample with respect to the laser beam. For that reason, we created a 3D printed incubation chamber with a pedestal tilted 20° from the horizontal plane. Although good results were obtained for small fields of view (see Visualization 1 ^{(13.9MB, avi)} ), this sample-mounting configuration was not amenable for regular visualization of the whole organ-on-chip device as well as for chip interchange.

Fig. 3.

Fluorescence images of fixed cell samples after 5 days in culture (maintaining the gradients by periodical replenishment of the culture media every 24 h) displaying proliferative cells (Ki67) (green), F-actin (red) and nucleus (DAPI staining). Images acquired with (a,b) Flexi-SPIM; (c,d) iSPIM system with slow scanning using XYstage; and (e,f) iSPIM system with fast scanning using synchronized electric tunable lens and galvo. Scale bar: 250 µm. For a 3D reconstruction of the samples, see Visualization 1 ^{(13.9MB, avi)} , Visualization 2 ^{(6.8MB, avi)} , Visualization 3 ^{(10MB, avi)} .

For that reason, we developed an iSPIM set-up displayed in Fig. 1(a). The iSPIM configuration solves the problem since both objectives face the sample at 45° from the top, and samples can be mounted easily in a MatTeck or Petri dish (Fig. 2(f)). As explained before, in our iSPIM system samples can be scanned in slower fashion through a motorized XY stage (Fig. 3(c)) or, alternatively, through fast scanning by remote focusing techniques using an ETL synchronized with the galvanometric mirror (Fig. 3(e)). To validate the performance of the system we imaged the same fixed sample under different available light-sheet microscopes on the lab (Flexi-SPIM (Fig. 3(b), iSPIM with motor scanning (Fig. 3(d)) and iSPIM with ETL scanning (Fig. 3(f) (also see Visualization 1 ^{(13.9MB, avi)} , Visualization 2 ^{(6.8MB, avi)} , Visualization 3 ^{(10MB, avi)} )). Although both, iSPIM slow and fast scanning modes provide high quality images, it is worth to point out some details regarding the light-sheet waist position along scanning, the useful FoV and the subsequent image post-processing needed.

When in the slow-scanning mode, an organ-on-chip device is scanned across the static light sheet using a XY stage (Fig. 3(c)), the LS waist illuminates always the same height of the pillars, providing a more homogeneous illumination across the entire chip. For every X stage position, the projected image on the camera is fixed on the FoV center (see Visualization 2 ^{(6.8MB, avi)} ), allowing to crop the image and to reduce the overall amount of data recorded. However, in order to recover the real sample volume a highly cost computationally image processing is needed, including image deskewing (applying a shear transformation with ImageJ’s TransformJ plugin [27]) and a 45° rotation to visualize the 3D structure in an orthogonal fashion, i.e. visualizing the sample from the top of the pillars. These image processing steps increase substantially the amount of memory needed. To image the whole 5 mm x 5 mm area, 4 to 5 long regions of the scaffold are sequentially scanned with the Y stage (see Fig. 5(a)).

Fig. 5.

Evaluation of the whole chip (day3) using the iSPIM set up with motor scanning (a) green channel, showing GFP and autofluorescence of the pilar structures. (b) Maximum projection red channel, showing tdTomato expressing cells (c) Transversal view of the two channels showing the vertical distribution of cells. The red area was interrogated over 5 hours, every 10 minutes, in different chips at different culturing times without any appreciable side effect of the laser, and achieving single cell resolution. Transversal views of chips at day 1,2 and 3: (d) single plane and (e) maximum projection. (f-h) Zoom in of three different subregions of the chip indicated in blue. See Visualization 5 ^{(15.1MB, avi)} .

For the fast-scanning mode with the ETL-galvo approach (Fig. 3(e)), the beam is quickly scanned across a static sample. For that reason, while at one edge the waist will be sitting on the top of the pillars, on the other edge this waist will be sitting on the pillar base, introducing some out of focus light across the FoV. We reduced this effect by closing the illumination slit, thus reducing the NA on one axis (see Fig. 1(b)) and generating a thicker but more homogenous LS along the whole FoV ( Supplement 1 ^{(2.1MB, pdf)} Fig. 1). On the camera plane, due to the demagnification, we can also observe the back focal aperture disk, which limits the overall FoV to 300 × 300 µm² (Fig. 3(f)).

Beside those limitations the fast-scanning mode using ETL presents two main advantages. While scanning, the image formed crosses the FoV, so no deskewing is needed. Only a 45° rotation of the image stack, which is much less time and resource consuming than deskewing, is performed to visualize the pillars from top to bottom (see Visualization 3 ^{(10MB, avi)} ). The main advantage of the proposed set-up, is related to the imaging speed. The use of the ETL allows us to record a full FoV volume (with 300 frames) in just two seconds for each channel. To cover the whole chip surface, we move the sample in a grid fashion using the XY motors, creating a mosaic after stitching (see Visualization 4 ^{(10.9MB, avi)} ). Then, it takes only around 7 minutes to scan the whole chip (considering 36 FoV volumes and two colors). To image the same sample using the XY stage scanning (slow mode) it would take up to 20 min.

3.2. High-throughput imaging of micro engineered intestine

Thanks to that speed we were able to evaluate in a fast manner different cell culturing conditions. The villus-like PEGDA-based hydrogels were fabricated on PET membranes, mounted on Transwell inserts and functionalized with collagen type I. Then, the proteins were delivered in the Transwell insert, i.e., from the basolateral side, to create the spatial gradients. For this first step, we employed as source medium primary fibroblasts conditioned medium (CM)/ENRCV and we corrected the concentrations of the ISC niche factors for the protein accumulation taking place at the membrane as predicted by the in-silico simulations [23]. To account for the cellular effects of the biomolecular gradients formed, four different culture conditions were established: (i) asymmetric, where primary fibroblast (CM)/ENRCV medium was delivered basolaterally and basal medium was delivered apically; (ii) uniform, with the basolateral and apical delivery of primary fibroblast (CM)/ENRCV medium; (iii) asymmetric 2.0, where the concentrations of Noggin and R-spondin were doubled compared to asymmetric while keeping basal medium at the apical chamber, and (iv) asymmetric Wnt2b, where the primary fibroblast CM was replaced by the recombinant protein Wnt2b, so ENRCV medium plus Wnt2b were delivered in the basolateral chamber and basal medium at the apical side.

24 hours after the formation of the gradients, we seeded mouse intestinal organoid derived single cells. Using an initial cell seeding density of approximately 5·10⁵ cells/sample, we observed that three days after seeding, epithelial monolayers had formed and were fully covering the surface of the hydrogels (see Fig. 4). Next, we wondered if the process to form a monolayer was dependent on the presence of ISC niche biomolecular gradients. For this, we seeded organoid-derived single cells on top of scaffolds bearing gradients and on scaffolds with no gradients, in other words, with uniform concentration of the factors. Every 24 hours we renewed the media of both apical and basolateral compartments to maintain the gradients stable, as predicted by the in-silico simulations. The high-throughput achieved by our microscope allowed us to evaluate in two imaging sessions 10 full chips, and 9 half chips. These include five different gradient strategies (Uniform, Asymmetric, Asymmetric Wnt2b, Asymmetric 2.0, Asymmetric Ki67) after 24 and 72 hours after cell seeding (see Supplement 1 ^{(2.1MB, pdf)} Fig. 2). One day after cell seeding, cells were starting to form a monolayer, covering approximately 30% of the surface under both conditions (Fig. 4(a,c)). 72 hours after seeding, the monolayer coverage reached to 70% of the surface in the Asymmetric condition (Fig. 4(d)) while in Uniform condition (Fig. 4(b)), the surface was almost fully covered.

Fig. 4.

High-throughput evaluation of the whole chip with fixed cell using the ETL/galvo scanning. Different gradient conditions were tested displaying proliferative cells (Ki67) (blue) and F-actin (green). Here we show two of them: Uniform at (a) day 1 and (b) day3; Asymmetric2.0 at (c) day 1 and (d) day 3; (e-h) Zoomed areas of (a-d) respectively. Scale bar: 300 µm. Also, see Supplement 1 ^{(2.1MB, pdf)} and Visualization 4 ^{(10.9MB, avi)} .

These results suggested that organoid-derived cells were able to form complete monolayers on the surface of the hydrogels irrespective of the presence or absence of such gradients. This implies that the type of ISC niche factor gradients formed in our system were not necessary for epithelial cells to be able to form intact monolayers; provided that the minimum concentration of factors required for cell growth were supplied. Still, it remained to be deciphered whether gradients of ISC factors determine cell compartmentalization in vitro.

3.3. Live imaging of the stem cell compartmentalization

To elucidate if gradients of ISC factors guide cell compartmentalization in vitro, we analyzed both the distribution of cells within the villus-like and crypt-like regions. Live imaging of the organoid-derived cells dividing and migrating onto the 3D villus-like scaffolds upon the generation of biochemical gradients was achieved (Fig. 5). The scaffolds were functionalized with collagen type I and the gradients of ISC niche biochemical factors were generated prior to seeding. Intestinal organoids were obtained from Lgr5-EGFP-ires-creERT2/ Rosa2btdTomato mouse. Intestinal organoids were previously induced with tamoxifen and the derived single cells were seeded onto the scaffolds. After incubating the samples overnight, we performed live imaging using the iSPIM setup that allowed culture medium temperature to be maintained at 37°C. We performed live cell migration imaging over 5 hours on a small region (red square in Fig. 5(a)) at three different time points (24, 48, and 72 hours after seeding) using the slow motor scanning iSPIM mode. In these first experiments, the culture medium used during acquisition was supplemented with 25 mM HEPES buffer, to compensate for the lack of CO₂ control. In order that the lack of CO₂ control does not affect the results, after 48 and 72 hours we imaged two different chips, maintained on fully controlled conditions in an incubator.

During the time of acquisition, the cells were maintained alive and endogenous signals of GFP (marking Lgr5+ ISCs) and tdTomato (marking the progeny of Lgr5+ ISCs) could be detected (see Visualization 5 ^{(15.1MB, avi)} ). Moreover, the distribution of the cells as a function of culture time was assessed and found that at day 1 the cells were primarily located at the hydrogel base. At increased culture times, the cells were found distributed along the pillars (Fig. 5(c)), indicating that the cells were growing to cover the microstructured hydrogel surface.

Using the fast ETL/galvo scanning mode, we compared the behavior of these cells under three different culturing conditions with three gradient distributions (Uniform (Fig. 6(a)), Asymmetric (Fig. 6(b)) and Asymmetric Wnt2b (Fig. 6(c))) (see Visualization 6 ^{(2.5MB, avi)} ). On those experiments, we included 5% CO₂ perfusion into the specially designed incubation chamber adapted for the iSPIM configuration. The use of this mode allowed us to visualize the whole chip, with two colors over 4 to 5 hours with 15 minutes time resolution, although given the scanning speed of the system this time resolution could be reduced to 7 minutes. This result suggested that in the Uniform condition, almost all proliferative cells are stem cells, implying that the transit-amplifying population is almost absent, and that there are almost no quiescent cells within the stem cell pool. Conversely, in Asymmetric condition, there is a fraction of proliferative cells that corresponds to cells that are not stem cells, probably transit-amplifying cells. Therefore, the Asymmetric condition seems to be favoring a more homeostatic-like tissue.

Fig. 6.

Live imaging using the ETL/galvo scanning of the whole chip. Maximum projection mosaic after 45° rotation for different gradient conditions: (a) Uniform (b) Asymmetric (c) Asymmetric Wnt2b. (d-f) Maximum projection and transverse projections at specific positions (14, 10 and 19, respectively) for the different gradient conditions at time point zero. Scale bar: 300 µm. Videos could be found in Visualization 6 ^{(2.5MB, avi)} .

4. Discussion and conclusions

In vitro assays are recently shifting from the use of 2D cell monolayers to the use of organotypic, 3D cell culture, as it is becoming increasingly evident that the “flat biology” approach is not predictive of in vivo tissue responses. Despite its undoubtedness utility, organoid technology has intrinsic limitations. Organoid 3D closed geometry impedes an easy access to the lumen, thus hindering drug screening experiments mimicking the in vivo route through the apical cell side. Live imaging is also challenging, as organoids are located in multiple focal planes and embedded in a 3D matrix. Most importantly, this 3D matrix is typically Matrigel, an animal-derived ECM with high batch-to-batch variability and ill-defined composition. As it has been shown that the tumor microenvironment is a key factor in cancer progression and metastasis, an accurate control of the biochemical environment might be beneficial to interrogate cancer stem cell behavior.

We have developed a new LSFM instrument, based on the iSPIM configuration, adapted for the interrogation of organ-on-chip devices. In this system, the sample can be scanned in slower fashion through a motorized XY stage. Alternatively, the fast scan of the samples through remote focusing techniques is possible using an ETL synchronized with a galvo mirror. Both modalities are fully controlled using a custom-made Labview code, which provides full automation of the system and allows multiple acquisition modes. We have also engineered and optimized the sample mounting protocols for both systems, including proper sample access and temperature control. On the other hand, we have designed and fabricated microfluidic chips allocating hydrogel scaffolds that permitted both, the generation of biomolecular gradients and their visualization with LSFM. We have previously measured [23,30] the biomolecular gradients formed on different hydrogel samples and corroborated the data using in silico models obtained through FEM simulations.

As a proof-of-principle of the versatility and the high-throughput achieved by our microscope we imaged different intestinal epithelium organ-on-chip devices, both fixed and live, including different gradient strategies (Uniform, Asymmetric, Asymmetric Wnt2b, Asymmetric 2.0, Asymmetric Ki67) after 24 and 72 hours after cell seeding. Moreover, we evaluated the performance of the system using fixed and alive samples and demonstrated the possibility to perform imaging session of several hours without distorting sample evolution. Finally, we were able to track the evolution of the cell cultures over three days in three different chips (see Fig. 5).

Altogether, the experiments presented in this paper show the successful fabrication of 3D villus-like PEGDA-based hydrogels with mouse anatomical dimensions, a native tissue-like elasticity and a hydrogel mesh size that allows the diffusion of the factors of the ISC niche. Primary epithelial cells formed a complete monolayer regardless of the presence or absence of gradients. However, only under gradients of the ISC niche (Asymmetric condition), the 3D monolayer exhibited in vivo-like cell compartmentalization, as seen by proliferative and stem cells preferentially located at the bases of the hydrogel and the progeny located at the uppermost regions of the villus-like pillars. Finally, we proved our platform as an ideal tool to screen for different gradients profiles and compositions and read out how epithelial monolayers react, demonstrating its relevance for the study of different types of samples.

The experiments performed at first stage allowed us to detect some problems on the system that needed to be addressed in the near future. Currently, the area of the chip under inspection is scanned sequentially once for each excitation/emission combination. This leads to increased scanning times and inefficient localizations, since cells may move in between different scans, especially if the slow motor scan version of the iSPIM is used. Alternatively, we propose the use of an AOTF and a dual band filter, in order to perform fast two-color imaging of the chip.

The second main technical problem is inherent to light-sheet microscopy and it is the large amount of data collected during an imaging session. As an example, for the interrogation of fixed chips, covering a total area of 5 × 5 mm², recording 30 positions with 3 colors, each dataset weights around 50 Gb. Since we scanned 12 different chips, the total amount of data produced for this experiment was around 0.6 Tb. For live cell tracking experiments, the amount of data is even bigger. These experiments typically lasted 4 hours, obtaining a full scan, dual color, of the lab-on-chip every 15 minutes. Each experiment, with 36 positions per chip (6 × 6 FoV) and 16 timepoints, resulted in around 400 Gb. We have repeated that for 3 different conditions, providing a total amount of 1.2 Tb. It is worth to point out that, since we use the iSPIM configuration, the final amount of data is almost doubled, since raw data needs to be post-processed in order to reorient the dataset to visualize the volume parallel to the chip surface. In any case, this postprocessing is highly simplified in the ETL scanning mode (only rotation of the dataset is needed) compared with the motor scanning mode (deskewing plus rotation), extremely reducing the time and computational cost.

Intestinal epithelial organ-on-chip devices, such as the one presented here, are becoming highly relevant models for high-throughput screening and drug analysis. In the future, our ultimate goal will be to evaluate the in vivo behavior of stem cells cultured in the controlled microenvironment of the designed chips, to visualize the effect of biochemical gradients on stem cell proliferation, migration, and differentiation. This article aims to advance this trend by providing a novel cell culture platform for epithelial tissues that can boost in vitro disease modeling, preclinical drug toxicity screening, and understanding of organ development. In this context, the development of new imaging techniques, such as light-sheet fluorescence microscopy, will contribute to increasing the potential and widespread adoption of these models in biological research and in the preclinical stages of drug development. The development of in vitro human intestinal models that faithfully replicate the in vivo behavior of adult stem cells also opens new possibilities for intestinal tissue regeneration or the personalization of cancer care.

Funding

Ministerio de Economía y Competitividad 10.13039/501100003329 ( CEX2019-000910-S); FUNDACIÓ Privada MIR-PUIG 10.13039/501100021495; Centres de Recerca de Catalunya 10.13039/100015439; Laserlab-Europe 10.13039/100015668 ( 871124); Departament d'Innovació, Universitats i Empresa, Generalitat de Catalunya 10.13039/501100002943 ( 2021 SGR 01495); European Regional Development Fund 10.13039/501100008530 ( RD16/0006/0012); European Regional Development Fund 10.13039/501100008530 ( PID2021-129115OB-I00); NextGenerationEU 10.13039/100031478 ( CPP2022-009979); Ministerio de Ciencia, Innovación y Universidades 10.13039/100014440 ( PID2021-129115OB-I00, CPP2022-009979); Instituto de Salud Carlos III 10.13039/501100004587 ( RD16/0006/0012).

Disclosures

The authors declare no conflicts of interest.

Data availability

Data underlying the results presented in this paper are not publicly available at this time but may be obtained from the authors upon reasonable request.

Supplemental document

See Supplement 1 ^{(2.1MB, pdf)} for supporting content.