Non-invasive label-free imaging analysis pipeline for in situ characterization of 3D brain organoids

Abstract

Brain organoids provide a unique opportunity to model organ development in a system similar to human organogenesis in vivo. Brain organoids thus hold great promise for drug screening and disease modeling. Conventional approaches to organoid characterization predominantly rely on molecular analysis methods, which are expensive, time-consuming, labor-intensive, and involve the destruction of the valuable three-dimensional (3D) architecture of the organoids. This reliance on end-point assays makes it challenging to assess cellular and subcellular events occurring during organoid development in their 3D context. As a result, the long developmental processes are not monitored nor assessed. The ability to perform non-invasive assays is critical for longitudinally assessing features of organoid development during culture. In this paper, we demonstrate a label-free high-content imaging approach for observing changes in organoid morphology and structural changes occurring at the cellular and subcellular level. Enabled by microfluidic-based culture of 3D cell systems and a novel 3D quantitative phase imaging method, we demonstrate the ability to perform non-destructive high-resolution quantitative image analysis of the organoid. The highlighted results demonstrated in this paper provide a new approach to performing live, non-destructive monitoring of organoid systems during culture.

Introduction

Developing treatments for brain diseases requires understanding the human brain’s anatomy, connectivity, and function, necessitating suitable pre-clinical models. Three-dimensional (3D) stem cell cultures (termed human brain organoids) have been developed to address the lack of robust pre-clinical models that accurately recapitulate human brain development^1–5. Although human organoids offer a unique opportunity for modeling organ development that recapitulate human organogensis, major challenges still exist in growing and studying organoids in a precise manner. The first main challenge is that traditional cell culture systems fail to emulate the complex micro-environments of human organ formation. As such, organoids are often heterogeneous and not grown in a reproducible manner^6–8. In an effort to grow organoids reliably, organoids have been grown in custom meso- and micro-fluidic devices (termed “organ-on-a-chip”) where the conditions can be precisely regulated to achieve desired outcomes with quality by design^9–12. The use of these devices to culture organoids has demonstrated great promise in increasing their cellular diversity, reducing their necrotic core formation, and producing disease models that better recapitulate human conditions^13–16.

The second main challenge in growing and studying organoids is the inability to perform detailed non-invasive in situ imaging with access to cellular and subcellular structures over long time scales. Typically, to obtain access to the cellular/subcellular structure (as well as molecular composition), fluorescence-based imaging modalities^17,18 and/or immunohistochemistry methods have been employed^{1,5,13,19–27}. These imaging platforms, however, rely on (1) labeling agents, which are not amenable for imaging organoids over long periods of time, unless cells are derived from genetically altered animal models^28,29, and/or they rely on (2) end-point assays, which are destructive. Both of these alternative are also expensive, time-consuming, and labor-intensive. As a result, it remains extremely challenging to monitor/assess organoids (particularly human derived organoids) regularly, in a non-invasive manner over their long (often several months) developmental processes.

To gain some insight into the overall health of the organoid during culture in a non-invasive manner, brightfield imaging is commonly employed. This technology is simple, ubiquitously available, and provides meso-scale structures of the organoid such as volume and circularity, estimated from two-dimensional (2D) projection images^13,14,30,31; however, it lacks cellular and sub-cellular detail. More recently, optical coherence tomography (OCT) has also been applied to monitor organoids non-invasively^32,33, but OCT is generally also unable to resolve cellular or subcellular structures.

To address this challenge, we apply an emerging label-free imaging technology called quantitative oblique back-illumination microscopy (qOBM) to non-destructively image brain organoids in situ^34–36. qOBM enables 3D quantitative phase imaging (QPI) with epi-illumination; and like QPI^37–41 qOBM provides clear, quantitative contrast of cellular/subcellular structures, but with the significant and unique advantage that qOBM can do so in 3D, in thick scattering samples with epi-illumination^42–44. With inherent refractive index (RI) contrast, qOBM yields unprecedented access to subcellular structural detail of organoids in situ without exogenous labels, making it an optimal tool for imaging brain organoids and monitoring their development during culture.

In this work, we develop a non-invasive, label-free image analysis pipeline for the live, non-destructive, longitudinal monitoring of organoids using fluidic technology, qOBM, and brightfield microscopy. We leverage a mesofluidic bioreactor to enable the automated culture in a reliably and reproducible manner, while enabling longitudinal monitoring of the organoids. We first fine tune the fluidic device design parameters for long-term culture with proper nutrient exchange while permitting monitoring with brightfield and qOBM imaging. We then apply this integrated system to image healthy and disease organoid models. We demonstrate high-content imaging that can identify cellular phenotypes and subtle morphological changes of the two groups longitudinally. Our combined brightfield-qOBM-mesofluidics system has the potential to non-invasively differentiate between disease and non-disease states in organoid models, which can provide insights for downstream molecular analysis. Together, the proposed pipeline comprises a new approach for live, non-destructive monitoring of organoid systems.

Results

Non-destructive imaging pipeline of brain organoids

Our combined imaging approach for label-free imaging of 3D organoids involved performing brightfield imaging and qOBM imaging in an optimized fluidic platform. Utilizing both imaging modalities enables high-content non-destructive analysis of organoids which can then be used to guide further downstream analysis (Fig. 1).

Fig. 1.

Conventional and Proposed Organoid Analysis Pipeline: (A) schematic of organoid growth over time, including an increase in the size of the organoid, which can be monitored with brightfield and changes in cell content, morphology, and texture –which can be observed with qOBM. Not drawn to scale. (B) conventional endpoint analyses including Haematoxylin & Eosin (H &E) staining and immunohistochemistry (IHC) staining. IHC stains are blue: DAPI (nuclei), green: Ki67 (proliferation marker), red: SOX2 (neural stem cell marker), and purple: MAP2 (neuronal marker). The images shown are for illustrative purposes only. (C) schematic of brightfield imaging of the organoids in the custom mesofluidic devices with a representative image. (D) schematic of the qOBM imaging system used for the organoids in the mesofluidic device. Bottom contains a 20$\times$ image. Right insets contain 40$\times$ images of the indicated regions in the 20$\times$ image. Scale bars are 200 μm.

The qOBM system consists of a conventional brightfield microscope with a modified illumination module^34–36,44. Rather than the classic transmission-based illumination used in brightfield microscopy, qOBM illuminates samples using four LED light sources (720 nm) deployed through optical multimode fibers arranged around the objective, 90$_{}^{\circ }$ from each other, as seen in Fig. 1D. With this configuration, the light effectively illuminates the focal imaging plane at a net oblique angle which provides phase contrast. Using a deconvolution algorithm, we obtain quantitative phase contrast images which encode the refractive index properties of the samples^34–36,44. The images provide 3D cellular and sub-cellular contrast up to 190 μm into the organoid and enable us to track differences in organoid growth over time in a non-invasive manner using a simple, low-cost, and easy-to-use system capable of high-throughput imaging.

The organoids in the fluidic device are first imaged using brightfield microscopy to capture their gross macro morphology (without cellular detail), as seen in Fig. 1C. Next, they are transferred to the qOBM setup to characterize organoid cell morphology, as seen in Fig. 1D.

Mesofluidic platform for non-invasive qOBM imaging of organoids

qOBM imaging can be conducted in standard well plates with modifications to ensure sufficient light scattering. These modifications are especially relevant for organoids at early time points when organoids are small and not able to internally scatter enough photons to provide the oblique back scatter illumination for qOBM. As such, a scattering layer (a simple piece of paper works well) is needed on top of the imaging chamber to increase the number of photons reaching the objective. In standard well plates, the height of the lid of the culture plate can be limiting in that the paper scattering layer sits too far away from the organoid. While modifications could be made to the standard well plates, we designed a fluidic device with a shorter overall height to bring the scattering layer closer to the organoid. Using the device also enables automated culturing in a closed system which lowers the risk of contamination during long-term culture and provides a more robust and reliable platform for the organoids.

Depending on the application, the device can be used as a culturing and imaging platform or just an imaging platform, and has been described extensively elsewhere⁴⁵. Briefly, the device consists of culturing chambers that are connected together via bifurcated 600 μm × 600 μm (wxh) perfusion channels located at the top and bottom of the chambers. The device is fabricated by bonding two PDMS layers together. These PDMS layers are bonded to a glass slide to facilitate high-content imaging. The bottom PDMS layer determines the height of the culture chamber, while the top PDMS layer provides the additional channel or flow path needed for creating crossflow in the device wells. The cell culture medium is perfused from the inlet channel at the bottom of the chamber and exits the outlet channel at the top. We found that flowing media from bottom (inlet) to top (outlet) led to fewer leaks from the silicon tape used to seal the device. The device is connected to a peristaltic pump to ensure automated and continuous media exchange during culture⁴⁵. Previous work demonstrated successful organoid culture, longitudinal brightfield imaging and sample quality control using these devices. The organoids cultured in these cross-flow-based perfusion devices showed comparable cell types and structures to conventional culture methods with minimal signs of cell death⁴⁵.

When utilized as a culture platform and a platform for qOBM imaging, the device geometry has to be jointly optimized to maintain the overall organoid health and enable live imaging of organoids (Table S1). To ensure proper growth of organoids in the mesofluidic device, we evaluated two criteria: the shear stress experienced by the organoids and the robustness of the devices while undergoing perfusion. We evaluated the effect of the device configuration on the shear stress experienced by the organoids in the device wells since shear stress could affect cell behavior and phenotypic expression⁴⁶. The shear stress was calculated using the equation $T=\frac{5\mu Q}{w{h}^2}$⁴⁷, where T is fluid shear stress (in dyn/cm$_{}^{-2}$), μ is viscosity of water (in cP), Q is the volumetric flow rate(in ml/hr), h is the main channel height (in mm), and w is the main channel width (in mm). Since the shear stresses calculated for the different device configurations (range: 0.004–0.03 dyn/cm$_{}^{-2}$) were less than 1 dyn/cm$_{}^{-2}$ (upper bound determined from⁴⁷), we assumed that shear stress did not have a considerable effect on organoids grown in our system and would not change drastically with the device configurations that were tested.

The second criterion for optimizing the device configurations involved evaluating the robustness of the device to enable perfusion flow. We performed flow-visualization studies to observe fluid flow in the devices (data not shown) and also performed dry-runs with cell culture media in an incubator to mimic organoid culturing over multiple days (8–10 days). Organoids were then cultured in device configurations that passed the flow visualization test and the dry-run experiments; only devices with no leaks from both experiments and devices that showed complete reagent exchange during flow-visualization experiments were used in further studies. From these experiments, we discovered that a 1 mm device height would not operate robustly under perfusion due to the instability of the luer fittings at the inlet and outlet ports of the device. We observed that this device height (and shorter heights) led to leaks that could result in cell death due to contamination or non-robust media exchange. Additionally, we observed that devices with low volumes would result in faster nutrient depletion during imaging sessions as the devices are disconnected from the pump set up for imaging. Hence, we determined that we could reduce the well height (bottom PDMS layer) to $\sim$2.5 mm while maintaining the top layer’s height at $\sim$4 mm to stabilize the luer fittings at the inlet and outlet ports of the device. Finally, we designed the well diameter to be 8 mm to ensure sufficient media was available during imaging (Figure S1A). Based on this optimization, we proceeded with devices $\sim$7.5 mm in total height. Next, we evaluated how well organoids could grow in the modified fluidic device. Prior studies have demonstrated similar levels of cell death between organoids grown in the mesofluidic devices and organoids grown in conventional methods such as Spin Omega Bioreactors^45,48. To ensure the organoids studied here were viable, we used the organoid size (as indicated by area) obtained from brightfield imaging to approximate organoid growth (Figure S1B). We observed that the modified platform could maintain organoids in culture for at least two weeks, as indicated by the increase in the organoid area during this time (Fig. 2). As we show in Fig. 1, this device geometry is well suited for qOBM and brightfield imaging.

Fig. 2.

Characterization of organoid growth via brightfield imaging. Characterization of the area of organoids cultured in the mesofluidic platform with device height of 7.5 mm. Significance was calculated using a two-tailed unpaired t-test with Welch-correction for two groups. ** indicates a p-value < 0.01.

Imaging human brain organoids

To demonstrate the utility of our imaging approach, we compared healthy and disease-model organoids using the fluidic device and qOBM setup. Tuberous Sclerosis Complex is a developmental disorder that affects multiple systems causing nonmalignant hamartomas that can affect the skin, heart, kidney, lung, and brain⁴⁹. In the brain, TSC is associated with epilepsy, autism & intellectual disability and is characterized by loss-of-function mutations in the TSC1 and/or TSC2 genes^2,50. These proteins regulate the mammalian target of Rapamycin (mTOR) pathway⁵¹. Loss of TSC1/TSC2 results in overactivation of the mTOR pathway, which is implicated in numerous biological processes related to cell growth, proliferation, metabolism, and protein synthesis². Previous work in developing in vitro neuronal stem cell models of TSC have increased our understanding of the disease expression. 2D and 3D in vitro cultures revealed minor differences in phenotypic expression between TSC and healthy controls during early neuronal differentiation^2,52–54. These differences included increased neural rosette sizes, increased neural progenitor (NPC) proliferation, increased cell size, and increased cell death. However, during later stages of neuronal differentiation, noticeable differences in neuronal and astroglial differentiation were observed, with a preference for the astro-glial fate over neuronal fate observed in cells with TSC mutations.

We hypothesized that longitudinal non-invasive imaging could help identify subtle morphological changes in organoid phenotypes. The protocol for TSC organoid culture was performed according to previously described methods for cortical spheroid formation (Fig. 3A)^5,20,55. In this study, we performed multiple rounds of imaging where we imaged organoids for 6 weeks of the culture process during the neural precursor expansion and neuronal differentiation processes. Using our high-content label-free imaging approach, we characterized organoid development via various metrics ranging from the whole organoid morphology to changes in neural rosette size and shape to more subtle changes in cell content over the course of the differentiation. These metrics were obtained from patient-derived organoids and used to identify differences between organoids with TSC mutations and healthy controls. Organoids with TSC mutations are considered our experimental group and were obtained from 3 different cell lines (TSC01, TSC10, TSC10E2) and our healthy controls were obtained from 2 different cell lines (PGP1, C1-2).

Fig. 3.

Whole-organoid morphology brightfield analysis of organoids with TSC mutations and healthy controls. (A) Schematic of protocol for TSC organoid culture. (B–D) Quantification of brightfield metrics related to organoid shape and size of the control organoids (left) and the experimental organoids (right); diameter (B), circularity (C), solidity (D), and aspect ratio (E). Images were taken using the mesofluidic device. N = 15 organoids (control) and N= 17 organoids (experimental) for Week 2, N = 21 organoids (control), N = 27 organoids (experimental) for Week 3, N = 19 organoids (control), N = 15 organoids (experimental) for Week 4, N = 8 organoids (control), N = 11 organoids (experimental) for Week 5 , N= 7 organoids (control), N = 11 organoids (experimental) for Week 6. Data is representative of 4 independent experiments with 6-8 organoids from each experiment group per experiment. Due to limited sample availability, data were pooled from both mesofluidic and conventional cultures. Week 2 - Week 4 data: combination of mesofluidic and conventional culture. Week 4- Week 5 data: mesofluidic culture only. Organoids with low quality brightfield images were discarded from the analysis. Error bars represent the standard deviation. Significance was calculated using a two-tailed unpaired t-test with Welch-correction for two groups. * indicates a p-value < 0.05. ** indicates a p-value < 0.01.

Analysis of whole organoid morphology

Using metrics obtained from brightfield imaging and qOBM imaging, we measured morphological changes that occur during organoid development on the whole-organoid. We obtained various brightfield metrics describing organoid size and shape (Fig. 3B–E). Using the diameter as a measure of organoid growth, we observed that although the initial population of TSC-experimental group organoids appears smaller than the healthy controls, the diameters of organoids from both groups become comparable over time. We also note that the organoid diameters increase over time, as expected. Next, we measured the circularity, solidity, and aspect ratio to characterize the organoid shape. While none of these variables demonstrate consistent statistically significant differences in the mean between the control and experimental organoids, we see the early onset of trends indicative of the differences in organoid formation seen in previously studied day 40+ organoids with TSC mutations^2,52,53. For instance, the circularity of the TSC-experimental group organoids appears to slightly decrease over time. In contrast, the circularity of the control organoids appears more constant throughout the experiment. Similarly, the solidity of the TSC-experimental organoids also decreased over time, while the solidity of the control organoids appeared constant. Additionally, organoids with TSC mutations had fairly consistent aspect ratios over development time; however they demonstrated increased variation compared to the control organoids with significant f-tests in week 4 (F_(18,14) = 0.34, p = 0.015), week 5 (F_(7,10) = 0.09, p = 0.003), and week 6 (F_(6,10) = 0.16, p = 0.027). This variation indicates the formation of more complex structures in the TSC. Interestingly, at day 25 (midway through week 4 of culture), the neural medium supplied to the organoids is supplemented with media to promote the differentiation of neural progenitors. Thus, the results suggest that the changes observed may be due to differences in neuronal progenitor cell proliferation or potential differentiation.

To get a detailed analysis of the trends observed with brightfield imaging, we performed a similar analysis with qOBM to observe structural changes occurring on the organoid surface, up to 190 $\mathrm{\mu }$m into the organoid. Low magnification imaging (20$\times$) with qOBM revealed that the TSC-experimental group organoids had more folds and complex structures forming on their surface when compared to the control (Fig. 4A) with markedly improved distinction between the control and experimental groups at later time points compared to the brightfield metrics. Specifically, the qOBM images revealed internal organoid fissure lines where an organoid appeared to have folded up on itself. An example of the fissure lines can be seen in Fig. 4A. Both control and TSC-experimental group organoids exhibit these folding structures; however, the number of organoids with observable folds were lower for the control group, even at early time points, and decreased rapidly over time. In contrast, a larger percentage of TSC-experimental group organoids exhibited fissures and folds, and the percentage did not decrease over time, as seen in Fig. 4B. At the fissure lines, the cells along the edge appeared as directional, elongated, well-aligned cells not seen elsewhere in the organoid. Areas with (green box) and without (yellow box) directional cells can be seen in Fig. 4D &E.

Fig. 4.

Low magnification (20$\times$) qOBM imaging of the two experimental groups reveals differences in cell morphologies on the surface of the organoid. (A) A representation of how the folds present in week 5 organoids. Right image shows a zoomed-in region with directional cells around the folding lines. (B) shows the presence of fissures over time, with the control organoids showing a decrease in fissures throughout development. Experimental group organoids do not show the same level of decrease. Error here is represented as $\frac{1}{\sqrt{n}}$ as given by Poisson counting statistics. (C) represents the values of the fractal features for the 7 $\mathrm{\mu }$m and 20 $\mathrm{\mu }$m patterns corresponding to the elongated cells along the fissures. These values follow the same trend in the controls and experimental organoids as exhibited in (B). Significant differences exist between the control and experimental group organoids post- exposure to neuronal differentiation media. (D, E) show the distribution of fractal values before and after culture in neuronal differentiation media to demonstrate the decrease of directional cells among the control group. They also contain images to show how different fractal values appear within the distribution. The selected images are border regions that contain both directional and circular cells. The fractal values of the boxed regions in the image are plotted with their respective symbols on the histogram. Significance was calculated using a two-tailed unpaired t-test with Welch-correction for two groups. Scale bars are 100 $\mathrm{\mu }$m. * indicates a p-value < 0.05. *** indicates a p-value < 0.001. **** indicates a p-value < 0.0001.

Analysis of cellular morphologies

Automated feature analysis was conducted to analyze differences in image texture and shape of organoids that presented fissure lines on their surface. We found trends in fractal dimension that correlated with the areas of the organoid containing directional cells. In particular, curved patterns corresponding to the cellular membrane and boundary were identified. Specifically, fractal dimension at scales of 5–9 $\mathrm{\mu }$m and 17–23 $\mathrm{\mu }$m were identified as corresponding to the oblong shape of directional cells - we selected the middle fractal values from those ranges (7 $\mathrm{\mu }$m and 20 $\mathrm{\mu }$m) for comparative analysis. We tracked the presence of features over time (Fig. 4D,E), comparing the TSC-experimental group organoids to the controls, and found that the prevalence of these features displays a similar pattern as seen in the presence of fissures in the organoids (Fig. 4C) where the fractal dimension values decreased over time in the controls but remained constant in the TSC-experimental organoids with significant differences (p < = 0.01) after Week 4, which is post-exposure to neuronal differentiation medium. The fractal values plotted indicate that fewer of the elongated, directional cells surrounding the folding and edges of the organoids are present in the control group post-exposure to neuronal differentiation medium. We further analyze those fractal values distribution in images pre-and-post-neuronal differentiation medium (histograms in Fig. 4D,E). Pre-exposure to neuronal differentiation medium (Week 1–3), the distributions of the TSC-experimental and control organoids overlap and display no significant differences. This suggests insignificant differences in the number of elongated, directional cells imaged. Post-exposure to neuronal differentiation medium (Week 5–6), the control distribution shifts left, i.e., displaying fewer elongated, directional cells. The TSC-experimental distribution post-exposure to neuronal differentiation medium also exhibits a skew towards fewer directional cells, albeit more subtle. However, more importantly, these organoids may maintain more cells containing directionality, as seen in Fig. 4D,E. This distribution suggests that both the control and TSC-experimental organoids start with a similar number of areas that exhibit these elongated, directional cell phenotypes, but post-exposure to neuronal differentiation medium, the controls do not show this phenotype. At the same time, it persists significantly in the TSC-experimental organoids.

Analysis of rosette morphology

Next, we studied the differences in the number of rosettes and their morphology in control and TSC-experimental organoids. The rosettes of the organoids are lumen-filled structures with a cellular arrangement of elongated processes that radiate outward from a central core. These neural tube-like structures are the sites of neurogenesis in brain organoid models⁵⁶. Prior studies have not revealed differences in the number of rosettes at early time points of TSC-experimental organoid culture^52,57. However, when scanning through the surface layers of the organoids, significantly more rosettes were observed in the TSC-experimental organoids versus the controls. qOBM imaging consists of surface-level imaging of approximately 190 $\mathrm{\mu }$m into the organoid, so while we cannot make claims regarding rosette count or development in the entire organoids, we observe that the TSC-experimental organoids contain a greater number of rosettes growing close to the organoid surface and outer layers. One such representation of the increased number of surface-level rosettes can be seen in Fig. 5A, where Week 4 organoids are compared: the control organoid(left) shows no rosettes, while the TSC- experimental (right) exhibits a large number of rosettes on a single plane. Figure 5B compares rosette count between TSC-experimental and control organoids. We note that the TSC-experimental organoids exhibit relatively constant surface-level rosettes over time. In contrast, the control organoids show decreased rosettes growing near the surface.

Fig. 5.

A demonstration of rosettes in the organoid. (A) (left) shows a control organoid with no surface-level rosettes and (right) shows a TSC organoid with 10 rosettes visible in the field of view. (B) shows that the TSC-experimental organoids demonstrated a statistically significantly higher number of rosettes on the surface of the organoid at all time points after Week 3. Error here is represented as $\frac{1}{\sqrt{n}}$ as given by Poisson counting statistics. (C) shows the segmentation of 4 different rosettes. From left to right, they show a pair of rounded rosettes with the lumen centered, an irregularly shaped rosette with a centered lumen, a rounded rosette with a lumen not centered, and an irregularly shaped rosette with an uncentered lumen. (D) shows the circularity of rosettes over time. Note the larger variance in the TSC rosette circularity compared to the controls. Error bars represent the standard deviation. (E) shows the distance between the center of the rosettes and the center of the lumen. Note how the lumen is less centered in the TSC-experimental rosettes than in the controls and how those differences increase over time. Error bars represent the standard deviation. Significance in (D) and (E) was calculated using a two-tailed unpaired t-test with Welch-correction for two groups. All scale bars are 100 $\mathrm{\mu }$m. * indicates a p-value < 0.05. ** indicates a p-value < 0.01. **** indicates a p-value < 0.0001.

We also observed that neural rosettes in TSC-experimental organoids tended to be more densely packed and irregularly shaped than those in the control organoids. Hence, we analyzed the rosettes’ size, shape, morphology, and image texture. We calculated the circularity of the rosette and lumen and the centeredness of the lumen in the rosette. Examples of centered, non-centered, rounded, and irregularly shaped rosettes can be seen in Fig. 5C. We found that the lumens of the rosettes were significantly less centered in the rosettes of the TSC-experimental organoids, as seen in Fig. 5E. We also noted a disparity in the rosette circularity of the TSC-experimental groups versus the control organoids post-exposure to neuronal differentiation medium, as seen in Fig. 5D. We found significant statistical test scores (f and t) between the control and TSC-experimental organoids post-neuronal differentiation, indicating a greater variation in the TSC-experimental rosette population and their mean. These findings agree with previous in vitro studies where the TSC-experimental group were observed to have altered neural rosette morphologies, possibly due to mTOR overactivation in neural progenitor cells^54,58.

Analysis of cellular content

During imaging we noted that more high-refractive index, bright spots appeared in the TSC-experimental organoids than in the control organoids. In qOBM imaging, a brighter region corresponds to an area with a higher refractive index (RI), implying the presence of subcellular structures including lipid droplets, cellular membranes, and nucleic material^35,60,61. With a lateral resolution of 0.6 μm, we can track the structure of high RI content with subcellular resolution. The high RI material was segmented from the image using a $\mathrm{\Delta }n$ > = 1.46 (Fig. 6A). Figure 6B shows that as the cultures grew, the control and TSC-experimental organoids increased in high RI content. However, the TSC-experimental organoids increased at a greater rate. By the final week of the study, the difference between the TSC-experimental and control organoids was statistically significant (p = 0.02).

Fig. 6.

Analysis of cell content using refractive index information. (A) shows qOBM segmentation with pink as the segmented high RI material. (B) shows the high refractive index data shows the percentage of the organoid composed of high RI. We note the growth over time with significant differences between the control and TSC organoids in Week 6 of organoid culture. (C) shows the distribution of fractal values with higher values representing repeated pattern values. The heat maps show the distribution of the linear fractal pattern (x-axis) and the 2D circular fractal pattern (y-axis) pre-and-post-exposure to neuronal differentiation medium. The image on the right exhibits sample regions and the corresponding fractal values in an experimental organoid growing in neuronal differentiation medium. Significance was calculated using a two-tailed unpaired t-test with Welch-correction for two groups. All scale bars are 50 $\mathrm{\mu }$m. * indicates a p-value < 0.05. Heatmap was generated in Matlab version R2023b⁵⁹.

Leveraging the quantitative nature of qOBM imaging, we used automated feature analysis to compare the amount of high RI spots between the organoid types. We identified the fractal values corresponding with the high RI droplets, including a curved linear fractal dimensions with a curvature diameter between 4 and 8 $\mathrm{\mu }$m and a two-dimensional circle fractal pattern corresponding to circular shapes of 34–38 $\mathrm{\mu }$m$_{}^2$. These fractal patterns show significant differences (p < = 0.01) between the TSC-experimental group and control organoids post-exposure to neuronal differentiation medium indicating differences in the presence of the high-RI material, as seen in Fig. 6C. In further exploring the distributions of the fractal values, we see that prior to neuronal differentiation, the control and TSC-experimental group organoids share a similar distribution, with many of the values overlapping (as seen in Fig. 6C). Post-exposure to neuronal differentiation medium, a shift occurs in which the control and TSC-experimental group organoids exhibit higher overall fractal values (indicating greater fractal structures than pre-exposure to neuronal differentiation medium and greater high-refractive index content). Post-exposure to neuronal differentiation medium, we also observe that the TSC-experimental organoids exhibit larger fractal values than the control organoids with a larger distribution, indicating higher heterogeneity of cellular structures. As indicated by the stars overlaid on the heat maps, even a tiny subsection of the organoid can have vastly different fractal values based on the structures in the sections analyzed. We note that the control organoids have a smaller, more homogeneous distribution of values post-exposure to neuronal differentiation medium; meanwhile, the TSC-experimental organoids display a larger overall distribution with higher value nodes, as seen in Fig. 6C. These results suggest that organoids display areas of varying lipid concentration, with some areas possessing a higher percentage than others. We hypothesize that these spatial differences observed might be due to the distribution of different cell types in these areas.

Lipid staining and histology verification

To validate some of the findings from our study we performed Oil Red O (ORO) and Hematoxylin staining on organoid sections Fig. 7A). Hematoxylin staining was performed to detect the cell nucleic content. Using the results from the Hematoxylin staining, we qualitatively assessed neural rosette morphology, distribution, and cell density in the organoid samples. Additionally, we performed ORO staining to identify possible reasons for differences in refractive index content between the control and TSC-experimental groups since it enables the identification of neutral lipids and lipid droplets in tissue sections⁶². We qualitatively observed that the TSC-experimental organoids, on average, tend to have more nucleic content than the healthy controls, as indicated by the deeper expression of the hematoxylin stain in the TSC-experimental group organoid samples. These results are consistent with previous findings on increased cell size and cell division in TSC-experimental samples compared to controls. The neural rosette structures in the TSC-experimental group organoids were also more densely packed than those in the control organoids.

Fig. 7.

Histological assessment reveals similar cell morphologies and lipid content differences in experimental and control organoids to non-invasive imaging. (A) Top: Representative Oil Red O images of organoid sections showing lipid droplets (red, indicted by yellow asterisks) at day 24 for experimental (1st column) and control (2nd column) samples. Bottom: Representative Oil Red O images of organoid sections showing lipid droplets (red, indicated by yellow asterisks) at day 42 for experimental (1st column) and control (2nd column) samples. Slices were counterstained with Hematoxylin. The brightness and contrast of images were adjusted for visualization. Scale bar:40 $\mathrm{\mu }$m. (B) Schematic showing the image processing pipeline for quantifying ORO particles in the organoid sections (C) Quantification of ORO-positive particles in organoid sections of experimental group organoids (left) and control organoids (right) at two different time points: day 24 and day 42. Error bars represent the standard deviation. All scale bars are 100 $\mathrm{\mu }$m. Significance was calculated using a two-tailed unpaired t-test with Welch-correction for two groups. D24: 112 images from 17 sections (control) and 75 images from 20 sections (experimental). D42: 81 images from 25 sections (control) and 71 from 22 sections (experimental). *** indicates a p-value < 0.001.

To measure the lipid content of the organoid samples, we quantified the amount of ORO particles in the organoid sections (Fig. 7B). Our results revealed that TSC-experimental group organoids had a higher ORO+ particle content than healthy controls both before (day 24) and after exposure to neuronal differentiation medium (day 42) (Fig. 7C). These differences in lipid distribution between TSC-experimental group organoids and healthy controls could explain disease expression in these organoids, as over-accumulation of lipids in neural stem cells has been associated with deregulating their ability to proliferate and differentiate⁶³. Overall, our results highlight the utility of our non-invasive approach in identifying subtle differences in cell content and metabolism, which can be investigated further with downstream molecular assays.

Discussion

Organoids have been shown to demonstrate superior fidelity to human development compared to animal models in both typical and pathological scenarios. Nevertheless, effectively tracking these 3D systems in a non-invasive, in situ manner remains a challenge. This research employs a label-free, non-invasive imaging technique to continuously observe diseased organoids and provides cellular and subcellular distinctions continuously over time. The platform is also capable of capturing developmental differences beginning at previously unstudied early time points. These capabilities enable the detection of variations in neural rosette morphology, structural development, and lipid metabolism between healthy and disease model organoids.

Our neural rosette morphology results were consistent with previous findings in 2D cultures. Some morphological changes we observed indicated changes in structure formation and lipid metabolism. For example, we observed increased folds and complex structure formation in TSC-experimental organoid models in early-stage organoids. However, we have yet to identify a potential cause for this observation. Previous brain organoid studies have however highlighted the role of ASD-related genes such as PTEN on cortical folding²². It was discovered that the down-regulation of the PTEN gene in neural progenitor cells (NPCs) led to increased cortical expansion and folding in cerebral organoids²². The down-regulation of PTEN could provide one explanation for the more complex structures observed in the TSC-experimental organoids due to its inhibitory role on PI3K, a known antagonist of TSC1/TSC2⁶⁴. However, further investigation is needed to confirm this.

We also observed increased high refractive index content, which maybe attributed to lipids, in the TSC-experimental organoids compared to their healthy controls. Our results indicate possible dysregulation of lipid metabolism in the TSC-experimental organoids via quantification of the qOBM images and ORO staining. One limitation of the ORO staining is that it is a non-specific stain⁶². As a result, it is difficult to determine what lipid species may be affected by the TSC mutation. Additionally, ORO staining helps detect hydrophobic and neutral lipids and not polar lipids (e.g., sphingolipids), which play critical roles in cognitive development^65–67; hence, our findings only provide an estimate of lipid presence in the organoid sections. Further, ORO staining is an endpoint measurement. As such, the tissue used for ORO staining can not be grown further or used for additional analysis. As such, a limited sample size was used. Additionally, the qOBM segmentation procedure described in this paper does not isolate only lipids, it may also include nucleic material and other structures with high RI. Nevertheless, the qOBM image segmentation can be used as a proxy to estimate lipid accumulation, with reasonable agreement to ORO staining (although not perfect one-to-one agreement). Results show a general agreement in that the TSC-experimental organoids contain greater lipid or high RI content at later days. Further lipidomics studies are necessary to uncover the underlying mechanisms behind the observations made in this paper and identify what cell types may be contributing to the lipid count differences. Attempts were made to conduct whole mount immunostaining on the same organoids imaged with qOBM; but given the high failure rate of the procedure, these experiments were not successful (label-free methods, such as qOBM, do not suffer from this significant disadvantage). We leave additional analyses that may provide direct comparisons of qOBM and molecular imaging for future work to characterize the organoids including immunohistochemistry staining and hematoxylin & eosin staining.

Indeed other optical methods have been used to image spheroids and organoids, including other forms of quantitative phase imaging (QPI) and dynamic full field (DFF)-OCT, however these methods have important limitations. Previously implemented phase imaging methods have mostly been restricted to thin samples (< 100 $\mathrm{\mu }$m) as they operate in transmission mode, or they depend on polarization and have limited subcellular detail^68–70. Alternatively, DFF-OCT can provide similar structural information as qOBM with similar penetration depth⁷¹, but is more complex, expensive and data-intensive as many acquisitions are necessary to render the structural detail⁷¹. On the other hand, qOBM is fast, low-cost, simple, and can even render similar dynamic/metabolic activity information⁴⁴. Thus, qOBM offers both high throughput and high-subcellular detail in an easy-to-use and low-cost embodiment.

In this work, we demonstrated an effective tool to non-invasively monitor human brain organoids over cell culture, providing access to morphological changes occurring over time without disrupting the cellular environment. Using this fluidics-enabled imaging method, we performed a high-content analysis which revealed differences in cell morphology and content between TSC-experimental organoids and healthy controls. The ability of qOBM to provide cellular and sub-cellular detail in 3D up to 190 $\mathrm{\mu }$m into the organoid is key to enabling unique quantitative features in the organoids. Differences in the folding of the organoids, shape of the organoids, distribution of lipids and differences that occur within the rosette structures of the organoid were highlighted in this study.

Utilizing the fluidic platform enabled the tracking of culture and monitoring of these live organoids while minimizing the risk of contamination due to the ability to provide automated cell feeds. The device was also specifically design to enable qOBM and brightfield imaging, which jointly enable the tracking of individual organoids over time to compare intra-organoid heterogeneity, a feature that will be used in future studies. Further, qOBM has demonstrated the ability to perform this imaging non-invasively within an incubator^72,73. With this, future studies may be conducted where organoids are monitored continuously during cell culture, and without needing to be disconnected from the fluidic devices for imaging.

Enabled by fluidic technology, we demonstrate that our imaging approach provides detailed high-content information about organoid development during the culture process. We envision this combined qOBM-fluidics technology being used to quantify important structural properties of organoids to help improve our understanding of endogenous developmental processes and better guide organogenesis and disease modeling. With a growing field of studying organoids grown in microfluidic devices (so called “organs-on-a-chip”)⁶, we envision the proposed qOBM-fluidic approach as a suitable platform for studying the organogenesis of organoids developed from a wide variety of organs and diseases.

Methods

Device fabrication for qOBM Imaging

Device fabrication was conducted using previously published protocols⁴⁶ but with modifications. The device design was drawn in SolidWorks, and molds for the devices were made using 3D printing by the company Protolabs. The molds were printed in the material Accura SL 5530. Using the 3D printed molds, mesofluidic devices were fabricated in polydimethylsiloxane (PDMS) (Dow Corning Sylgard 184, Midland, MI) by soft lithography⁷⁴. Briefly, PDMS was mixed in a 10:1 ratio of pre-polymer and crosslinker, degassed to remove air bubbles, poured on the master mold, degassed a second time to remove remaining bubbles, and cured overnight at 80 °C. Following curing, PDMS devices were peeled off the master molds. The molds were not pre-treated prior to use. Additionally, creating the crossflow in the device required two-layer PDMS fabrication. The mold for both layers was identical. For both layers of features, PDMS was poured on the mold to a height of approximately 3 mm to define the height of the culture chamber. Following curing and peeling, cylindrical chambers were made in both feature layers by manually punching holes with an 8 mm biopsy punch (VWR). Inlet and outlet holes were punched with a 2 mm biopsy punch (VWR). The bottom PDMS layer was bonded to a 1 mm thick glass slide. Next, the top and bottom PDMS layers were plasma bonded together and left in an oven at 80 °C overnight to strengthen the bond.

Free-space qOBM system

The qOBM system consists of a conventional brightfield microscope with a modified illumination module^34–36,42. Rather than the classic transmission-based illumination used in brightfield microscopy and QPI, qOBM illuminates samples in epi-mode using four LED light sources (720 nm) deployed through multimode optical fibers arranged around the objective, 90$_{}^{\circ }$ from each other. Through these fibers, the sample is illuminated in epi-mode, where approximately 45 mW are incident on the organoid samples. In the organoids, the photons undergo multiple scattering events causing the photons to change direction, with some being redirected back toward the microscope objective. These redirected photons create an effective virtual light source within the sample with an overall oblique illumination, a process known as oblique back-illumination⁷⁵. Variations in the index of refraction in the sample refract the light either towards or away from the microscopy objective, resulting in intensity fluctuations that encode the RI properties of the sample. This work uses a Nikon S Plan Fluor LWD 99 20$\times$, 0.45 NA, and Nikon S Plan Fluor LWD 40$\times$, 0.6 NA. Light collected by the microscope is detected using an sCMOS camera (pco.edge 4.2 LT). Intensity images collected from two opposing illumination angles are subtracted to produce a differential phase contrast (DPC) image. Two orthogonal DPC images (a total of four acquisitions) are deconvolved with the system’s optical transfer function to finally obtain quantitative phase contrast images. This process has been described in further detail in previous studies^{34–36,42,43}.

Organoid culture

hiPSCs derived from patient donors were used for organoid generation. TSC-experimental organoid formation was performed using pre-established protocols for generating cortical organoids^5,20,55. Briefly, hiPSC colonies were dissociated from a layer of mouse embryonic fibroblast feeders by exposing them to a low concentration of dispase for approximately 30 min. Suspended colonies were transferred into ultra-low-attachment 100 mm plastic plates in hiPSC medium without FGF2. The medium was supplemented with the ROCK inhibitor for the first 24 h (day 0). Dorsomorphin and SB-431542 were added to the medium for the first five days for neural induction. On the sixth day in suspension, the floating spheroids were moved to neural medium (NM) containing Neurobasal, B-27 serum substitute without vitamin A, GlutaMax, 100 U ml⁻¹ penicillin, and 100 $\mathrm{\mu }$l streptomycin. The NM was supplemented with 20 ng ml⁻¹ FGF2 and 20 ng ml⁻¹ EGF for 19 days with medium change every other day. To promote differentiation of the neural progenitors into neurons, FGF2 and EGF were replaced with 20 ng ml BDNF and 20 ng ml^-1 NT3 starting on day 25 till day 42 of the culture. A description of the cell lines used in this study is provided in Table 1.⁻¹

Table 1.

Summary of hiPSC lines used in qOBM study.

Name of cell line	Description
C1-2	Healthy control
426	Healthy control
PGP1	Healthy control
TSC10E2	Isogenic control
TSC10	Patient sample
TSC01	Patient sample

The healthy control organoids were derived from patients with no TSC mutations. In contrast, the TSC01 and TSC10 organoids were derived from patients with mutations in their TSC2 gene. The isogenic control organoid, TSC10E2, was derived from an hIPSC cell line of a healthy patient (PGP1) which was genetically modified by knocking in mutations into the TSC2 gene via CRISPR-Cas 9 gene editing. All studies were approved by Emory University School of Medicine Institutional Review Board (IRB). All methods and experimental protocols were in accordance with institutional guidelines. All subjects or their legal representatives were informed and signed informed consents.

Organoid culture in mesofluidic device

Prior to each experiment,the devices, luer fittings (Nordson Medical), the bubble trap (Cole Parmer), and tubing (1/32” I.D. silicone tubing, 1.6 mm I.D. peristaltic tubing; Cole Parmer) were sterilized by autoclaving. The devices were treated with air plasma the day before organoid loading to render the PDMS hydrophilic. After treatment, the devices were immediately primed with DMEM/F12 (Life Technologies) to maintain the hydrophilicity and then re-sterilized with UV light for 30 mins. The devices were placed in a cell culture incubator overnight after replacing the DMEM/F12 with fresh DMEM/F12. Before organoid loading, the devices were primed with Neural Media supplemented with 20 ng ml⁻¹ FGF2 and 20 ng ml⁻¹ EGF. Organoids were loaded into individual chambers of the device by pipetting with cut 200 $\mathrm{\mu }$L tips. Following the loading of organoids into individual chambers, the devices were sealed with 3M™ Thermally Conductive Adhesive Transfer Tape (Product No: 9882) to seal the culture chambers. Finally, primed tubing and fittings were connected to the device inlet and outlet. The devices were then connected to a peristaltic pump. The entire setup was placed in a humidified incubator (HERAcell 240i, Thermo Scientific) for culture. Organoids were cultured in the mesofluidic devices starting on day 14. This was the earliest timepoint we considered as there would be an increased risk of the organoids getting displaced from the device during perfusion with smaller-sized organoids (< 1 mm) as our channel sizes are 600 $\mathrm{\mu }$m wide. Alternatively, the organoids were cultured in 6-well plates on orbital shakers and transferred to the devices for imaging.

Brightfield imaging and quantification

Brightfield images of devices were acquired prior to the start of the qOBM imaging session using an EVOS microscope. Organoid size and shape features were quantified from images using FIJI/ImageJ. Multiple images of different regions of the same organoid had to be taken for larger organoids because the organoids were larger than the field of view on the EVOS microscope. The images were then stitched together using a pair-wise stitching plugin on FIJI/ImageJ before quantification⁷⁶.

Imaging organoids with qOBM

A total of 24 control and 24 TSC-experimental organoids were imaged over a four-week time period from day 15 of culture (week 3, pre-neuronal differentiation) to day 42 of culture (week 6, post-neuronal differentiation). The organoids were imaged inside mesofluidic devices 3 times a week with two different magnification objectives: (1) a 20$\times$ (0.45NA, 850 $\mathrm{\mu }$m$_{}^2$ fields of view (FOV)) microscope objective captured images of the organoid up to 190 $\mathrm{\mu }$m in-depth, and (2) a 40$\times$ (0.6NA, 425 m$_{}^2$ FOV) microscope objective captured additional, higher resolution images taken of features of interest, including cell structure, rosette morphology, and unique features that had not previously been well-resolved with the 20$\times$ objective. A subset of organoids were taken for endpoint imaging at day 24, and the remaining subset was processed for endpoint imaging at day 42. An air stream incubator was used during imaging to helped maintain homeostatic temperatures.

Manual feature analysis of qOBM images

Post-imaging, qualitative and quantitative features were extracted from the organoid images, including the presence (enumeration) of observed folds and rosette presence. Additionally, the rosettes and lumen were manually segmented for quantitative analysis. Finally, segmentation was used to separate the content of high-RI particulate structures, corresponding primarily to what we hypothesize to be lipid and nucleic acid material. The relationship between the measured optical phase and the sample’s refractive index is given by,

$$\begin{array}{c}\hfill \mathrm{\Delta }n=\frac{\lambda \mathrm{\Delta }\varphi }{2\pi \mathrm{\Delta }z}\end{array}$$ 1

Where $\mathrm{\Delta }n=n_0-n_m$ represents the refractive index difference between the object ($n_0$) and the surrounding medium ($n_m$), $\lambda$ represents the wavelength, $\mathrm{\Delta }\varphi$ represents the phase difference, and $\mathrm{\Delta }z$ represents the thickness of the object, or in this case, the effective thickness of the slice provided by this 3D imaging method³⁵. Here we take the medium’s refractive index ($n_m$) to be that of a PDMS scattering medium used in imaging to simulate thick brain tissue, 1.3440⁷⁷. For all 20$\times$ images, a RI threshold (n > = 1.46) was used to segment the lipid and nucleic acid material from the rest of the organoid cells in the images^36,78.

Automated feature analysis of qOBM images

Quantitative image features extracted from the qOBM images were analyzed to assess structural differences between the control and TSC-experimental organoids. To accomplish this, images from the 20$\times$, 0.45 NA objective ($720\mathrm{\mu }\text{m}\times 720\mathrm{\mu }\text{m}$) were subdivided into regions of $50\mathrm{\mu }\text{m}\times 50\mathrm{\mu }\text{m}$. With those subdivided regions, features per region were extracted based on texture analysis⁷⁹, fractal analysis^80,81, Fourier space features⁸², and mathematical auto-correlation transformations^83,84. Detailed explanations regarding the computed features can also be found in Ref.⁸². Feature selection ranking was performed using Minimum Redundance and Maximum Relevance, Neighborhood Component Analysis, and the Chi-square tests, as implemented by Matlab’s functions fscmrmr, fscnca, and fscchi2, respectively. The highest 150 ranked features from all three methods that posed the greatest differences between the data sets are discussed in the Results.

Oil red O staining and quantification

The Oil Red O staining protocol was provided by Aqua Asberry, the Laboratory Coordinator for the Parker H. Petit Institute for Bioengineering and Bioscience Research Histology Core at the Georgia Institute of Technology in Atlanta, GA. Briefly, the frozen sections were allowed to thaw and air dry at room temperature for about 10 mins. Next, the sections were submerged in Oil Red O staining reagent for about 20 mins. Following ORO staining, the samples were washed under running water while being careful to avoid direct contact with the water. A counterstain with Hematoxylin was also performed for about 20 secs, and the samples were rewashed under running water until all the excess stains had been eliminated from the washing solution. The samples were then mounted on a coverslip using an aqueous mounting medium. Samples were then imaged using a Zeiss AxioObserver Z1 Fluorescent Microscope. Quantification of total ORO lipid content was performed using Fiji/ImageJ software. Color deconvolution separated the red channel (ORO staining) from the blue channel (Hematoxylin staining) using the built-in matrix: FastRed/FastBlue. Following color deconvolution, the images were thresholded, and the number of pixels was counted to determine the number of ORO particles in the image. This number was normalized by the image area occupied by the pixels.

Author contributions

C.E.S., S.C. and, P.C.C. conceived and conducted the experiments. S.C. designed the fluidic devices. S.C. and W.N. grew the organoids. C.E.S. lead qOBM data analysis with assistance from B.C. S.C. conducted and analyzed ORO experiments. C.E.S. and S.C. interpreted the data. C.E.S. created all figures. C.E.S. and S.C. drafted the manuscript. All authors reviewed the results and approved the final version of the manuscript.

Data availability

The data that support the findings of this study are available from the corresponding author upon reasonable request.

Competing interests

The authors declare no competing interests.

Supplementary Information

The online version contains supplementary material available at 10.1038/s41598-024-72038-2.