Advancing preclinical drug evaluation through automated 3D imaging for high-throughput screening with kidney organoids

Abstract

High-throughput drug screening is crucial for advancing healthcare through drug discovery. However, a significant limitation arises from available in vitro models using conventional 2D cell culture, which lack the proper phenotypes and architectures observed in three-dimensional (3D) tissues. Recent advancements in stem cell biology have facilitated the generation of organoids—3D tissue constructs that mimic human organs in vitro. Kidney organoids, derived from human pluripotent stem cells, represent a significant breakthrough in disease representation. They encompass major kidney cell types organized within distinct nephron segments, surrounded by stroma and endothelial cells. This tissue allows for the assessment of structural alterations such as nephron loss, a characteristic of chronic kidney disease. Despite these advantages, the complexity of 3D structures has hindered the use of organoids for large-scale drug screening, and the drug screening pipelines utilizing these complex in vitro models remain to be established for high-throughput screening. In this study, we address the technical limitations of kidney organoids through fully automated 3D imaging, aided by a machine-learning approach for automatic profiling of nephron segment-specific epithelial morphometry. Kidney organoids were exposed to the nephrotoxic agent cisplatin to model severe acute kidney injury. An U.S. Food and Drug Administration (FDA)-approved drug library was tested for therapeutic and nephrotoxicity screening. The fully automated pipeline of 3D image acquisition and analysis identified nephrotoxic or therapeutic drugs during cisplatin chemotherapy. The nephrotoxic potential of these drugs aligned with previous in vivo and human reports. Additionally, Imatinib, a tyrosine kinase inhibitor used in hematological malignancies, was identified as a potential preventive therapy for cisplatin-induced kidney injury. Our proof-of-concept report demonstrates that the automated screening process, using 3D morphometric assays with kidney organoids, enables high-throughput screening for nephrotoxicity and therapeutic assessment in 3D tissue constructs.

1. Introduction

The field of drug discovery plays a pivotal role in advancing healthcare, expanding our scientific knowledge, and ultimately enhancing the quality of life. In this domain, two primary approaches are employed to unearth potential drug candidates in the early stages of the drug discovery process [1]. The first approach, known as target-based screening, centers its efforts on specific proteins. This strategy is typically employed when a comprehensive understanding of the molecular underpinnings of a disease is available. It seeks to pinpoint compounds that can interact with the target molecule, providing a precise and mechanistic route to therapeutic development. In contrast, phenotypic screening represents the second approach, concentrating on observable changes in cells. This method aims to identify compounds capable of producing desired therapeutic effects or modifying disease phenotypes, even in cases where the specific target remains enigmatic. Phenotypic screening proves especially versatile when dealing with diseases of intricate, multifactorial origins. Both target-based and phenotypic screening have significantly contributed to the field of drug discovery. It is worth noting that many first-in-class drugs have their roots in phenotypic screening, underlining its enduring significance and potential [2].

In the context of phenotypic screening, the selection of reliable models that faithfully replicate disease phenotypes in experimental settings is imperative. In other words, the choice of an appropriate in vitro model is critical for the success of phenotypic screening [3]. This emphasis on model quality is further substantiated by the evolving landscape of high throughput screening (HTS) technologies [4]. Over the years, the focus of HTS has shifted from quantity to the prioritization of screening test quality [5]. Traditionally, cell lines have served as the gold standard in HTS, owing to their ease of use [6]. However, recent advancements in bioengineering and stem cell biology have ushered in a new era, allowing for the creation of organoids. These three-dimensional (3D) cellular constructs provide a more faithful representation of organ pathophysiology and enhance our capacity to uncover innovative therapeutic interventions.

For instance, kidney organoids derived from human pluripotent stem cells mimic the intricate complexity of the kidney comprising various cell types, each possessing unique functions and characteristics [7–9]. The nephron, which serves as the functional unit of the kidney, assembles into segmented structures composed of podocytes and tubular epithelia, supported by interstitial stromal cells and endothelia [10, 11]. These diverse cell types within the kidney interact with each other in complex ways, playing pivotal roles in various pathophysiological processes [12]. Consequently, the use of kidney organoids presents a new opportunity to advance our comprehension of kidney disease mechanisms and accelerate drug discovery efforts [13]. Recent studies provide compelling evidence of the superiority of kidney organoids over traditional 2D cell culture models, with the expression of SLC22 transporter family (OAT1, OAT3, OCT2) in organoid proximal tubules [10, 12, 14–18]. This is of particular significance as the expression of such drug transporters in experimental models is indispensable for studies related to renal toxicity of drugs, a primary contributor to acute kidney injury among hospitalized patients [19, 20].

While kidney organoids exhibit great potential for therapeutic development through drug screening, substantial challenges must be addressed to make them practical models for drug discovery. Unlike 2D cell culture models, organoids form 3D tissues, necessitating the development of innovative methods for phenotypic screening during drug evaluation. Particularly for HTS, it is imperative to establish a practical, automated, and unbiased screening process with minimal manual intervention. In this study, we present a pioneering high-throughput approach that leverages cutting-edge technologies of kidney organoids, tissue clearing techniques, and machine-learning-based 3D imaging and morphometric characterization. We employed an FDA-approved drug screening compound library and implemented automated segment classification and analysis as a reliable and reproducible assessment tool. Given the ability to reconstruct the 3D architecture of renal tubules throughout the entire organoid, we hypothesized that quantitative values of these structures can serve as readouts for drug screening. This novel method enables the identification of potential nephrotoxic and nephroprotective agents in 3D kidney organoid screening, providing more intricate phenotypic information than current standards relying on traditional cell lines.

2. Materials and methods

2.1. Cell culture, kidney organoid differentiation and maintenance

BJFF.6 human induced pluripotent stem cells (hiPSCs) were provided by Sanjay Jain at Washington University. hiPSCs were cultured and maintained in StemFit Basic04 complete medium (Ajinomoto Co., Inc.) in hESC-qualified Geltrex (Thermo Fisher Scientific) coated 6-well plates, with initial supplementation after cell passage with a ROCK inhibitor (Y-27632, Tocris). The cells were passaged weekly, using Accutase (StemCell Technologies). Kidney organoids were generated following a previously published method [7]. Briefly, after passage, cells were seeded at a density of 6000 cells per well in Geltrex-coated 24-well plates. After 3 d of culture in StemFit Basic04 supplemented with Y-27 632, stem cells were differentiated into nephron progenitor cells (NPCs) in 2D culture. This was performed first by the addition of CHIR 99021 (5.5 μM, Fisher Scientific) and dorsomorphin (200 nM, Fisher Scientific) to the culture medium (advanced RPMI supplemented with 1% GlutaMAX, Thermo Fisher) from day 0 to day 4, then by the addition of Activin A (10 ng ml⁻¹, R&D) from day 4 to day 7, then by the addition of recombinant human FGF9 (10 ng ml⁻¹, Fisher Scientific) from day 7 to day 8. At day 8 of differentiation, the NPCs were detached using Accutase, and the suspension was transferred in ultra-low attachment 96-well plates at the concentration of 50 000 cells per well to form 3D spheroids. These were further differentiated into nephron organoids by adding CHIR (2.5 μM) and FGF9 (10 ng ml⁻¹) from day 8 to day 10, then FGF9 alone (10 ng ml⁻¹) from day 10 to day 14. Kidney organoids were thereafter cultured in advanced RPMI supplemented with 1% GlutaMAX with medium change thrice a week. For the drug screening procedure, kidney organoids were cultured within ultra-low attachment Akura 96 Spheroid Microplates (InSphero).

2.2. Supplemented factor exposure

Cisplatin.

Day 49 kidney organoids were used as they display a more mature phenotype and expression profile, particularly in relation to tubular drug transporters (supplementary figure S1(a)–(k) and S6(c)) [12, 14, 21]. The size of organoids increases with longer duration of culture, not only due to an accumulation of stromal cells [22], but also to an increase in structure volume (supplementary figure S1l). Cisplatin (Sigma-Aldrich) was added in the basal medium at a concentration of 5 μM from day 49 of differentiation of kidney organoids, for one week, with medium change thrice a week [12]. Corresponding controls underwent the same medium change.

Additional compounds.

Each compound present in the Tocriscreen FDA-approved drugs library was added to the medium of one organoid (n = 1) at the concentration of 10 μM from day 49 to day 56 of differentiation, with the same frequency of medium change as the cisplatin-treated organoids and the control organoids, i.e. thrice a week. After the initial screening, compounds were selected to test nephrotoxicity and compared to cisplatin-treated organoids and control organoids (n = 3 organoids per condition). Regarding the assessment of nephroprotection against cisplatin-mediated injury, organoids were co-exposed to cisplatin and the candidate drugs at the abovementioned concentrations for the same duration, and were compared to cisplatin-only treated organoids and control organoids (n = 3 organoids per condition, reproduced 2 times, hence a total of 3 independent experiments [23]).

2.3. Immunofluorescence staining and tissue clearing

Whole mount 3D staining and Tissue clearing.

The entire staining and imaging processes were performed without transferring the organoids, which remained in the Akura 96-well plates for spheroid culture. After fixation with 4% paraformaldehyde (PFA) for 1 h, organoids were blocked with a buffer used for blocking aspecific antigen sites and diluting antibodies (BB/ADB, 1 wt% bovine serum albumin (BSA), 5 vol% donkey serum, and 1 vol% Triton X-100 in phosphate buffered saline (PBS-T)) supplemented with streptavidin/biotin blocking kit (Vector lab) for 1 h. After washing with 0.1 vol% PBS-T three times, organoids were incubated with primary antibodies (Rat anti-CDH1, 1:500, ab11512, Abcam, Goat anti-PODXL, 1:500, AF1658, R&D, and Biotinylated LTL, 1:200, B-1325, Vector Lab) diluted in the BB/ADB buffer overnight at 4 °C. After washing with 0.1 vol% PBS-T, organoids were incubated with secondary antibodies conjugated with dyes that are excited at the longest wavelengths, optimal for confocal imaging of thick tissues [24] (Alexa Fluor^™ 647-conjugated Donkey anti-Rat IgG, 1:500, Life Technologies, Alexa Fluor^™ 555-conjugated Donkey anti-Goat IgG, 1:500, ThermoFisher, Alexa Fluor^™ 750-conjugated Streptavidin, 1:500, ThermoFisher) and nuclear staining (SYTOX^™ blue, 1:1000, Invitrogen) diluted in the BB/ADB buffer overnight at 4 °C. After washing with 0.1 vol% PBS-T3 times, organoids were fixed with 4% PFA for 30 min, and subsequently dehydrated by incubating them at 4 °C with Ethanol solutions at increasing concentrations (50% half day, 70% overnight, 100% ethanol for 2 d). Tissue clearing was performed using Ethyl Cinnamate (Sigma-Aldrich), 30 min before the start of image acquisition.

Immunohistochemistry for Frozen Sections.

Organoids were collected and fixed with 4% PFA in PBS for 1 h, washed three times with PBS, incubated with 30% sucrose (w/w) overnight at 4 °C and embedded in O.C.T. compound (Fisher Scientific) to create frozen blocks. 10 μm sections were washed with PBS three times, incubated with blocking buffer (BB; 5vol% normal donkey serum and 0.3vol% Triton X-100 in PBS) for 1 h. For LTL-biotin staining, a streptavidin/biotin blocking kit (Vector Laboratories, SP-2002) was used according to the manufacturer’s protocol. After washing with PBS three times, the sections were incubated with primary antibodies in antibody diluting buffer (ADB; 1wt% BSA and 0.3vol% Triton X-100 in PBS) overnight at 4 °C. Samples were washed with PBS three times, incubated with Alexa Fluor secondary antibodies and nuclear staining by SYTOX blue (Invitrogen) in ADB for 1 h at room temperature, and washed with PBS three times. Finally, samples were mounted with the VECTASHIELD mounting medium (Vector Laboratories) and sealed using a cover slip. Primary antibodies are listed in supplementary table S1.

2.4. Image acquisition and analysis

3D fluorescence images of the cleared kidney organoids were taken within the Akura flat-bottom 96-well plate, using the STELLARIS 8 confocal microscope (Leica Microsystems). To enable automated imaging, a predefined plate layout was created, allowing the imager to automatically scan the selected regions corresponding to each well. The images were acquired using the sequential method in Line mode, at a resolution of 1024 × 1024 pixels. z-stacks of the whole organoids were taken, using a 5 μm interval, allowing us to produce 3D images of organoids.

The acquired images were analyzed using the IMARIS 3D image analysis software, connected to Image J-Fiji with the LABKIT (machine-learning) plugin [25]. This plugin allowed pixel classification after training the software to identify the regions of interest. The created protocol performed image segmentation according to the selected laser channel, corresponding to each segment of the nephron (LTL+ segments, proximal tubular epithelium; CDH1+ segments, distal tubular epithelium; PODXL+ segments, glomerulus-like podocyte clusters). The Batch pipeline of the IMARIS software was then used to automatically process the images corresponding to all the organoids.

2.5. Quantitative reverse transcription polymerase chain reaction (PCR)

Kidney organoid samples underwent total RNA isolation with TRIzol (Invitrogen). Subsequently, cDNA synthesis was carried out using 400 ng of total RNA and the high-capacity cDNA reverse transcription kit (Applied Biosystems) in accordance with the manufacturer’s instructions. Real-time PCR was executed utilizing iTaq SYBR Green Supermix (Bio-Rad) on the QuantStudio3 Real-time PCR systems. GAPDH served as the housekeeping gene, and values were computed using the delta delta CT method. Primer sequences are provided in supplementary table S2.

2.6. Data processing and statistical analyses

The machine learning-based structure identification yielded comprehensive quantitative datasets including morphometry data (supplementary table S3). The volume of each structure, the total volume of each specific segment, and the number of structures were analyzed as the most relevant morphometric analyses, consistent with in vivo assessment [26, 27]. Regarding the drug screening initial evaluation of potential hits, total LTL volume served as the initial evaluation, confirmed by Z-score transformation applied using the formula:

$$z=\frac{x-\mu }{\sigma }$$

where $x$ is the raw score, $\mu$ is the population mean, and $\sigma$ is the population standard deviation.

The compounds associated with the most extreme Z-scores were further evaluated to confirm the phenotype.

Data were presented as scatter plots and mean. One-way ANOVA and multiple comparisons using the Holm-Sidak test were performed to evaluate the statistical significance of differences between groups. A p-value < 0.05 was considered significant. Graphs and statistical analyses were performed using Graphpad Prism (9.0).

3. Results

3.1. Optimizing kidney organoid production and imaging for drug screening

For reliable phenotypic screening in drug discovery, addressing key challenges recognized in organoid research is crucial in kidney organoid production. First and foremost, the issue of kidney organoid immaturity and their reported tendency to degrade with fibrosis has been acknowledged by several research groups [28–31]. To tackle this concern, we have adopted a differentiation approach that mimics metanephric kidney development, involving the specification of posterior intermediate mesoderm and metanephric mesenchyme, which includes SIX2-positive NPCs [7, 9, 32]. This directed differentiation process initially induces SIX2+ NPCs through a 3-step growth factor and small molecule treatment (figure 1(a)). The process begins with treatment involving a GSK-3β inhibitor, CHIR99021 (CHIR), and a BMP inhibitor (Dorsomorphin) for 4 d, followed by treatment with activin for 3 d and FGF9 administration for an additional day. The induction of NPCs is evaluated on day 8 of differentiation through SIX2 immunostaining (figure 1(b)).

Figure 1.

Kidney organoid generation and z-stack 3D images. (a) Schematic depicting the protocol used for directed differentiation of human pluripotent stem cells to human kidney organoids in vitro. (b) Immunostaining of nephron progenitor cells in 2D culture, showing the expression of SIX2. Scale bar = 500 μm. (c) Representative image of a kidney organoid at day 14 of differentiation under bright field microscopy, showing the development of tubular structures (white arrow). Scale bar = 200 μm. (d) Immunostaining demonstrating the presence of nephron structures at an early stage after differentiation (day 21, left panel) and at a later stage (day 49, right panel). Scale bar = 200 μm. (e) Representative 3D images of a kidney organoid at day 21 of differentiation. Scale bar = 100 μm. NPCs, nephron progenitor cells, Dorso, dorsomorphin, PODXL, podocalyxin, LTL, Lotus Tetragonolobus Lectin.

Following dissociation into a single-cell suspension, NPCs were reseeded into a 96-well ultra-low-attachment plate. To minimize exposure to undefined components, organoids were cultured in suspension without any extracellular matrix, such as the animal-derived Matrigel commonly used in various organoid cultures. To induce mesenchymal-to-epithelial transition, simulating in vivo processes prompted by ureteric bud stimulation [33, 34], NPC spheroids were subjected to a combination of FGF9 and a low-dose CHIR treatment for 2 d, followed by FGF9 without CHIR for 4 d. Cells accumulated and formed 3D spheroidal structures, with certain structures in organoids emerging by day 14 (figure 1(c)). Subsequently, nephron organoids were cultured in a basic medium without any inducing reagents. By day 21, the organoids displayed segmented nephron components and interstitial cells, including proximal tubules, distal tubules, glomeruli, and endothelial cells (figure 1(d) and supplementary figure S2(a)). To address batch variations, another significant challenge in organoid research [35], quality controls were applied to all batches of organoid differentiation. This included SIX2 immunocytochemistry at the NPC stage and whole-mount staining for the three nephron segments using anti-podocalyxin (PODXL) antibody, biotinylated LTL, and anti-CDH1 antibody. A high SIX2 staining intensity at day 8 of differentiation was correlated with satisfying kidney organoid differentiation (supplementary figure S2(b)). Of note, in-batch control samples were systematically used to correct for batch-to-batch variation. Viable nephron structures were observed in the totality of the organoids, including in the center (Z-stack images in supplementary video S1 and S2). Additional staining and transcript expression were checked for these segments, namely WT1, NPHS1, LRP2 and ATP1a1 (supplementary figures S2(c) and (d)). With these strategies to enhance metanephric specification and minimize batch variations, the generated kidney organoids proved to be stable for long-term culture, consistent with findings in previous studies (figure 1(d)) [12].

The kidney organoid generated through this approach had a thickness of approximately ~1000 μm in diameter. Imaging such thick tissues presents challenges for drug screening because conventional fluorescence microscopy techniques, typically used for thinner samples, are inevitably constrained by light scattering issues. While tissue sectioning of frozen or paraffin-embedded samples is feasible and commonly employed, such manual processes are not only impractical for large-scale drug screening, but also hinder the possibility of grasping the 3D architecture. Therefore, we have adopted a tissue-clearing method that involves a dehydration process, followed by organoid clearing with ethyl cinnamate [36]. This preparation enabled us to capture z-stack whole 3D images from the top to the bottom using a standard confocal microscopy setup with a single laser. The acquired z-stack images were reconstructed into 3D renderings, showcasing multiple glomerular and tubular structures within each organoid (figure 1(e)).

3.2. Improving efficiency and precision of 3D imaging for drug screening with organoids

Image-based drug screening is a time-consuming process, and the requirement of z-stack imaging for organoids substantially increases the imaging time, making it impractical. Additionally, kidney organoids are randomly located within each well, necessitating manual intervention to determine the X and Y axis values for each organoid sample. To address these inherent challenges, we strategically employed 96-well ultra-low-attachment plates with flat and narrow bottoms. This plate allowed us to capture entire organoids at the predefined center position of the standard 96-well plate format, eliminating the need for manual selection of X–Y positions. Their compatibility with direct imaging by confocal microscopy, with the flat bottom, additionally enables high-resolution imaging, thereby circumventing the complications associated with organoid thickness when combined with tissue clearing. This preparation allowed us to capture nephron structures via z-stack whole 3D images directly within the confines of the 96-well plates using confocal microscopy in an automated fashion (figure 2(a)).

Figure 2.

Machine-learning-based object analysis in 3D images of kidney organoids. (a) Representative confocal microscope image of a well containing a kidney organoid stained with nuclear staining (SYTOX blue), a proximal tubule marker (LTL), a marker for loops of Henle and distal nephrons (CDH1), a podocyte marker (PODXL), the merged image, and the bright field image (right panel). (b) Image analysis using the machine learning Labkit pipeline to identify the objects corresponding to each nephron structure (top panels) compared to intensity-based image segmentation (bottom panels) Scale bar = 50 μm. (c) LTL+ segments within a kidney organoid transformed into 3D objects using machine learning. Scale bar = 50 μm. (d) Representative images of immunofluorescent staining of control and cisplatin-treated organoids, showing a decrease in LTL+ structures corresponding to proximal tubules, with no overt change in podocyte clusters expressing PODXL. Scale bar = 200 μm. (e) Quantification of total volume and number of LTL+ and PODXL+ structures for control and cisplatin-treated organoids. PODXL, podocalyxin, LTL, Lotus Tetragonolobus Lectin. *, p < 0.05.

To quantitatively dissect the 3D information, z-stack images need to be converted into 3D objects. Typically, threshold-based approaches using the signal intensities are taken for this step; however, background noises tend to be higher in such thick tissues like organoids even with tissue clearing. To minimize the background signals that can create false objects, we employed a machine-learning-based approach to create the 3D objects using Labkit [25]. This method showed the more precise classification of LTL+ proximal tubules, when compared to a conventional Intensity threshold approach (figure 2(b)). This comprehensive 3D analysis by the IMARIS software provided us with several parameter results comprising, for instance, the number of classified objects, the area, volume of each classified object, among others (figure 2(c) and supplementary table S3).

To validate whether this 3D image analysis can be used to evaluate tubular toxicity, we subjected kidney organoids to cisplatin. This cisplatin dose was formulated based on previously published studies reporting dose-dependent toxicity, established at 5 μM to induce moderate injury [37]. The protocol was simplified, based on previous studies that performed repetitive cisplatin treatment to induce severe tubular injury [12]. A time course experiment was initially conducted from day 0 to day 7 of cisplatin treatment. This experiment revealed a significant loss of LTL-positive tubular structures by day 7 (41% average loss, p = 0.019) (supplementary figure S3(a)), establishing day 7 as the duration for the experimental treatment. The z-stack images of these and control organoids visualized the LTL+ proximal tubular structures (figure 2(d)). The quantification of the total volume of LTL+ objects created by the machine-learning method revealed a significant reduction in the total volume of proximal tubular structures in organoids treated with cisplatin, while the number of LTL+ objects per organoid did not statistically change (figure 2(e)). Importantly, cisplatin treatment did not affect the PODXL+ podocytes, as confirmed by the assessment of the total volume, justifying the use of cisplatin at this 5 μM concentration for tubular toxicity studies. This result supports our hypothesis that drug screening is feasible with kidney organoids by using 3D image analyses.

3.3. Drug screening using FDA-approved drugs in kidney organoids

To demonstrate a proof-of-concept for drug screening in kidney organoids, we used an FDA-approved drug library in a cisplatin injury model (figure 3(a)). In this model, kidney organoids at day 49 of differentiation were exposed to 5 μM cisplatin for one week. Each drug, at a concentration of 10 μM, was added to individual wells containing kidney organoids cultured in 96-well plates. Throughout the one-week treatment, media were replaced every two to three days to ensure adequate nutrient and drug supplementation. Subsequently, all organoids were fixed with 4% PFA in the cell culture plates and underwent immunostaining for nephron segments, using PODXL, LTL, and CDH1 markers. Following tissue clearing, we performed automated acquisition of z-stack images using a confocal microscope (figure 3(b)). All z-stack images underwent 3D object analysis through the machine learning method and the batch pipeline of the IMARIS software, which enables automated image processing. This automated imaging pipeline of organoids revealed that some drugs preserved the total volume of LTL-positive structures, while others further decreased the LTL-positive total volume as shown by the Z-score values (figure 3(c) and supplementary figure S3(b)). To validate these findings from the initial screening, we selected representative drugs according to the Z score, that were either high or low compared to the other samples, suggesting potential protective or toxic effects, respectively.

Figure 3.

FDA-approved drug screening using kidney organoids and volumetric analysis. (a) A schematic illustration of the screening protocol. (b) Bright field and fluorescence confocal microscopy of each well containing one kidney organoid. (c) Distribution of Z scores of LTL+ structures total volume of the organoids, exposed to one drug of the library each, during the first screening phase. Blue dots: drugs selected for their potential nephroprotective effect. Red dots: drugs selected for their potential nephrotoxic effect.

3.4. Toxicity evaluation

The initial screening showed that FTY 720 (Fingolimod), Ciclopirox, and CPT 11 (Irinotecan) exhibited a considerable reduction in the LTL total volume per organoid compared to the control with cisplatin alone. Each of them was subjected to a second round of rigorous testing for one week without cisplatin treatment to assess their nephrotoxicity in kidney organoids (figure 4(a)). These validation assessments revealed that Ciclopirox, in particular, demonstrated loss of most of the nephron structures, indicative of severe nephrotoxicity. CPT 11, also exhibited a severe level of injury surpassing that caused by cisplatin. FTY 720 showed a decreasing trend in the LTL total volume (figures 4(b), (c) and supplementary figure S4). To evaluate the cell-type specificity, we assessed the toxicity to podocytes by measuring the total volume of PODXL+ glomerular structures. Notably, none of the three drugs decreased the PODXL+ total volume per organoid (figure 4(d)). This second phase of drug evaluation confirmed the initial screening. It also suggests that the toxicity of FTY 720, Ciclopirox, and CPT 11 is primarily mediated by drug transporters that are predominantly expressed in proximal tubules [14].

Figure 4.

Evaluation of nephrotoxic drugs. (a) A schematic diagram of the toxicity assay protocol. (b) Representative images of immunofluorescent staining of kidney organoids subjected to the tested compounds. Scale bar = 200 μm, for the top panel, scale bar = 50 μm, for the bottom panel. (c) Quantification of LTL+ total volume in each tested organoid of the toxicity evaluation phase. (d) Quantification of PODXL+ total volume in each tested organoid of the toxicity evaluation phase. ^**, p < 0.01, ^***, p < 0.001, ^****, p < 0.0001. PODXL, podocalyxin, LTL, Lotus Tetragonolobus Lectin, FTY, Fingolimod, CPT, Irinotecan.

3.5. Potential drugs that ameliorate cisplatin nephrotoxicity

Cisplatin is a commonly used chemotherapy drug in clinical practice, in the treatment of various tumor types, including lung and breast cancer [38]. Nephrotoxicity is a well-recognized side effect of cisplatin [39]. Kidney damage resulting fromcisplatin treatment can have serious consequences, including the discontinuation of chemotherapy and the potential need for renal replacement therapy. However, there are currently no established drugs known to effectively mitigate cisplatin-induced kidney injury. To address the challenge of cisplatin-induced nephrotoxicity, we embarked on an exploration of potential FDA-approved drugs that might alleviate this condition. In this endeavor, we re-evaluated the top three FDA-approved drugs that showed improvement in the volume of LTL-positive tubules during the initial screening. These included Imatinib (anti-cancer), Tizanidine (muscle relaxant), and Olopatadine (antihistamine). Additionally, we tested Olmesartan, an antihypertensive drug that has previously been reported for its potential renal protective properties [40, 41].

Kidney organoids were exposed to each of these drugs at a concentration of 10 μM in the presence of cisplatin at 5 μM for one week (figure 5(a)). Analysis of 3D objects from z-stack images revealed a significant reduction in the total volume of LTL+ proximal tubules when cisplatin was used alone. However, there was an increased tendency for volume preservation when Imatinib and Olmesartan were introduced (figures 5(b), (c), (d) and supplementary figure S5). Notably, we observed structural improvements in each proximal tubular structure with the addition of Imatinib. Subsequently, we assessed various parameters to evaluate the proximal tubules and found that the LTL+ volume within each structure, as opposed to the total volume within each organoid, showed a statistically significant improvement with Imatinib (figure 5(e)).

Figure 5.

Nephroprotection assay. (a) A schematic diagram of the Nephroprotection assay protocol. (b) Representative images of immunofluorescent staining of kidney organoids in control conditions, or subjected to cisplatin ± the potential nephroprotective compounds. Scale bar = 200 μm. (c) Representative images of object classification using the machine learning process showing the change in LTL+ structures within organoids, in the control condition, or subjected to cisplatin ± the potential nephroprotective compounds. Scale bar = 100 μm. (d) Quantification of LTL+ total volume in each tested organoid. *, p < 0.05, **, p < 0.01. (e) Quantification of the volume of each LTL+ structure in each condition. **, p < 0.01. PODXL, podocalyxin, LTL, Lotus Tetragonolobus Lectin.

Proximal tubular epithelial cell injury was further evidenced in cisplatin-treated organoids by the de novo expression of KIM1, consistent with a 5.6 mean fold change of the KIM1 transcript HAVCR1, of the DNA damage marker phosphor-histone H2AX, and of the epithelial to mesenchymal transition marker Vimentine (figure 6). Tubular injury was partly prevented by imatinib treatment as observed by the decrease in DNA damage with LTL-positive epithelial cells. Under basal condition, a certain degree of cell proliferation was observed, which significantly increased after cisplatin treatment, with no statistical difference in the cisplatin + imatinib group (supplementary figure S6).

Figure 6.

Cell states in organoids treated with cisplatin and Imatinib. (a) Representative images of KIM1 and γH2X immunofluorescent staining of kidney organoids in control conditions, or subjected to 5 μM cisplatin ±10 μM Imatinib. Scale bar = 40 μm. (b) Quantification of KIM1 positivity proximal tubular cells and γH2AX positivity in proximal tubular cell nuclei. *, p < 0.05, **, p < 0.01 ***, p < 0.001. (c) qPCR analysis of kidney organoids in control conditions, or subjected to cisplatin ± Imatinib for HAVCR1 (KIM1). Values are normalized against the control. ***, p < 0.001, ****, p < 0.0001. (d) Representative images of vimentine (VIM) immunofluorescent staining of kidney organoids in control conditions, or subjected to 5 μM cisplatin ±10 μM Imatinib. Scale bar = 40 μm.

Although further studies are required to confirm the efficacy and mechanisms of Imatinib in preventing cisplatin-induced nephrotoxicity, these results underscore the utility and feasibility of kidney organoid screening through 3D imaging. This phenotypic screening pipeline can be adapted for larger-scale screenings in future studies.

4. Discussion

In this study, we present a machine learning-based approach for the automatic profiling of nephron segment-specific epithelial morphometry in kidney organoids. This approach allowed the identification of potential nephrotoxic and nephroprotective agents when compared or associated with a well-characterized nephrotoxic agent, cisplatin. The whole process, from image acquisition to image analysis was performed automatically, demonstrating a potential for translation to high-throughput screening.

The rich vascularization of kidneys allows efficient electrolyte and acid-base homeostasis. An inevitable backlash of this high efficiency is the susceptibility of kidneys to drug-induced toxicity [42]. As a matter of fact, nephrotoxicity is a major cause of acute kidney injury [43], which is associated with adverse clinical outcomes including increased risk of mortality and morbidity [44]. Nephrotoxicity also represents a major hurdle in drug development as it accounts for 19% of drug attrition during phase 3 clinical trials [42, 45], demonstrating the frequent underestimation of nephrotoxicity in the early stages of drug discovery. Predicting nephrotoxicity in humans at the non-clinical stage is thus a crucial need that remains to be addressed. Indeed, in vitro assays for nephrotoxicity have limited predictive capabilities regarding toxicity in humans [19], as they rely on indirect injury biomarkers or cell viability analysis [46]. Moreover, rescue therapies are more difficult to apprehend by the use of these biomarkers. In vivo injury responses are, instead, evaluated by morphometry analyses based on the established correlation between structure and function [47]. This structure-based analysis is one of the key advantages of kidney organoid use in drug screening. The morphometry approach was validated by the use of cisplatin, which represents a well-characterized injury control model [12], showing a decrease in proximal tubular epithelium volume and the number of structures following exposure to cisplatin. Kidney organoids also serve as a human-derived model system, conversely to animal models that often fail to predict human response due to interspecies variability [48]. This is even more so when studying nephrotoxicity, as murine models are known to be less prone to kidney damage and fibrosis than humans [49]. Therefore, kidney organoids are anticipated to contribute to drug discovery, toxicity evaluation, and our understanding of underlying mechanisms.

The 3D morphometric approach successfully identified both nephrotoxic compounds and potential drugs for attenuating cisplatin-induced injury. Ciclopirox, a widely used topical medication, is effective in treating fungal infections of the skin and nails [50]. Additionally, as a pan-histone demethylase inhibitor, it affects gene expression by modulating histone modifications, making it a potential candidate for anticancer therapy [51, 52]. While efficacy in reducing cyst expansion in a polycystic kidney disease model has been demonstrated at lower concentrations, our results are consistent with previous findings, showing that cytotoxicity occurs at concentrations exceeding 2 μM [53].

CPT11, or Irinotecan, is categorized as a topoisomerase 1 inhibitor, a class of chemotherapeutic agents used in cancer treatment. Its hepatic metabolism would predict a low risk of kidney toxicity, although there are several reports of acute kidney injury on pre-existing chronic kidney disease following CPT11 administration [54]. Finally, FTY720 acts as a sphingosine 1-phosphate (S1P) receptor agonist and is used in the treatment of multiple sclerosis as an immunomodulatory agent. It has been reported that FTY720 concentrations exceeding 2 μM induced significant cytotoxicity in mouse bone marrow cells [55], consistent with the mild toxicity observed in the tubules at a concentration of 10 μM in our results. Overall, upon examination of our nephrotoxicity assays following the initial drug screening in the context of previous literature, our results suggest that kidney organoids can serve as a relevant preclinical model for the investigation of nephrotoxicity.

Regarding the potential nephroprotective agents in cisplatin-induced injury, our screening method allowed the identification of a new candidate. Imatinib belongs to the class of tyrosine kinase inhibitors. It revolutionized the field of onco-hematology in the 2000s by dramatically improving the survival outcomes of patients with chronic myeloid leukemia and Philadelphia chromosome-positive acute lymphoblastic leukemia [56]. Our results suggest a protective effect of imatinib when co-administered with cisplatin. Interestingly, imatinib is known to inhibit creatinine tubular secretion [57], suggesting an inhibition of the proximal tubular creatinine transporters, which are also cisplatin transporters [58]. In addition, cisplatin elicits kidney toxicity by penetrating the proximal tubular epithelial cells, subsequently accumulating within these cells and inducing DNA damage [11, 12]. Interestingly, Imatinib promotes DNA repair in irradiation-induced cellular damage [59], consistently with our findings, strengthening the view of direct cellular protection by this drug. Our results are consistent with a previous report demonstrating the nephroprotective effect of imatinib in a rat model of cisplatin-induced acute tubular injury [60]. The consistency observed in our results thus encourages us to validate the robustness of our kidney organoid assessment system. This validation extends beyond nephrotoxicity screening and covers the exploration of potential therapeutic interventions to prevent or treat nephrotoxicity and other etiologies of kidney disease.

We have developed an automated screening methodology using kidney organoids, which provides a new perspective on high-throughput screening. The automated segmentation and classification yielded comprehensive quantitative characterization, including volume, and number of structures, but also fluorescence intensity, density, or circularity. All these features might represent new predictive tools in pre-clinical studies, in the same line as recent innovations in big data processing in the field of medical imaging [61]. This methodology can also be extended to other cell types within kidney organoids, such as podocytes, interstitial cells, and endothelial cells. This opens up the potential to investigate segment-specific toxicity and injury, and to identify new targets in kidney disease drug development, ranging from genetic to immune-mediated conditions. Future steps to the scalability of this methodology will include high throughput size organoid production as reported by Tran et al [62], and automated organoid production and culture, along with cell cultures [63]. Hit expansion studies are also warranted, as well as in vivo testing to provide a comprehensive view of the applications of kidney organoids in drug development.

There were two major limitations in our screening study. Firstly, the acquisition of images using our confocal microscope was time-consuming. It took approximately 6 h to capture z-stack images of one entire plate of 96-wells. This prolonged duration indicates that the technological improvement in 3D image acquisition is necessary to efficiently test larger numbers of drug candidates. Secondly, regarding automated 3D image quantification using batch analysis, while it is a powerful tool for object classification for multiple samples, it is not flawless. In one 3D image during the second phase of therapeutic testing, reclassification of LTL+ objects with an additional machine-learning process was necessary to accurately capture the 3D objects. This technical discrepancy in the current software could potentially introduce bias into the results of the initial large screening, necessitating subsequent validation experiments using selected drugs with manual intervention for more accurate image analyses. Therefore, for those seeking highly accurate data, individual analysis and/or further improvements of the algorithms in the Imaris software are necessary.

In conclusion, we developed a new automated 3D morphometry profiling to facilitate the identification of potential drugs targeting specific segments of the kidney. We anticipate that our strategy will make a transformative contribution to drug discovery by mitigating the safety risk and accelerating access to effective and safe medications. This transformative contribution holds the promise of reshaping the landscape of pharmaceutical research and development, fostering a more efficient and secure path to improved healthcare solutions

Data availability statement

All data that support the findings of this study are included within the article (and any supplementary files).