ReBiA-Robotic Enabled Biological Automation: 3D Epithelial Tissue Production

Abstract

The Food and Drug Administration's recent decision to eliminate mandatory animal testing for drug approval marks a significant shift to alternative methods. Similarly, the European Parliament is advocating for a faster transition, reflecting public preference for animal-free research practices. In vitro tissue models are increasingly recognized as valuable tools for regulatory assessments before clinical trials, in line with the 3R principles (Replace, Reduce, Refine). Despite their potential, barriers such as the need for standardization, availability, and cost hinder their widespread adoption. To address these challenges, the Robotic Enabled Biological Automation (ReBiA) system is developed. This system uses a dual-arm robot capable of standardizing laboratory processes within a closed automated environment, translating manual processes into automated ones. This reduces the need for process-specific developments, making in vitro tissue models more consistent and cost-effective. ReBiA's performance is demonstrated through producing human reconstructed epidermis, human airway epithelial models, and human intestinal organoids. Analyses confirm that these models match the morphology and protein expression of manually prepared and native tissues, with similar cell viability. These successes highlight ReBiA's potential to lower barriers to broader adoption of in vitro tissue models, supporting a shift toward more ethical and advanced research methods.

Keywords: 3R; AI tools; alternatives to animal testing; lab automation; tissue engineering.

Figure 1

Workflows for automated tissue engineering. A) The icons on the left represent the use of automation technologies, from the top an automated microscope, neural networks for data processing, collection of process data in a cloud, and translation into an automated process using a PLC and step sequences. General workflow derived from manual SOP: (1) Manual cell isolation; (2) 2D cell seeding; (3) medium exchange in T‐flask; (4) cell detachment; (5) cell count and resuspension; (6) 3D cell seeding; (7) add proliferation medium; (8) set to ALI; (9) medium exchange in cell culture plate; (10) manual evaluation and further use of the tissues. B) Illustration of the communication interfaces, data flow is clockwise. The thickness of the lines represent an abstraction of the data flow intensity between the respective partners.

Figure 2

Complete robotic system. A) The working area of the robot is designed to ensure that all tailored devices and standard laboratory equipment can be installed within reach of the robot. (1) falcon stack 4 °C; (2) falcon stack room temperature (RT); (3) material port; (4) incubator; (5) centrifuge; (6) laminar flow; (7) pipette rack; (8) robot; (9) pipette separation system; (10) microscope; (11) controlling unit [PLC]. B) Top view of the whole system; (4), (5) access to centrifuge and incubator; (9) pipette separation system; (green framed area) sterile area secured via laminar flow. C) Technician handing over 2D cell culture in T‐flask to the material port. D) Robot taking released culture medium out of RT stack to transfer it to the sterile area. E) Robot using the high‐volume pipetting device to refill culture medium.

Figure 3

Design of devices. Automated tissue model generation required the adaption of the robotic platform and of the laboratory material. To allow maximum flexibility in handling various geometries of equipment, two gripping mechanisms were implemented: parallel and centric gripping. A) For the mobilization of different tools, as well as the opening of screw caps and the transport of centrifuge tubes on the lid, the choice is made for the self‐centering properties of the centric gripper. B) Parallel gripping fingers can handle T‐flasks, centrifuge tubes on the shaft, microtiter plates, and various measuring pipettes. C) The high‐volume pipette (HVP) is developed based on the standardized hardware interface for the centric gripper. D) It is adaptable to use corrugated tips with a nominal volume from 2 mL up to 50 mL. (1) universal grip adapter with pipette centering; (2) headpiece with universal magazine adapter; (3) silicone attachment; (4) connector hull; (5) pneumatic scissors gripper; (6) pipette centering gripper; (7) disposable graduated pipette. E) The combination of parallel and centric gripper allows complex sample manipulation like the dosing of liquid inside of T175 cell culture flask. To ensure the correct positioning of the cell culture transwell inserts or cell crowns, a plate holder as well as an insert holder were integrated. F) The inner structures of the RWP enable the cultivation of 24 tissue models on a shared medium compartment. The positioning mat and plates are adapted for RHE culture using Millicell inserts, and G) cell crowns with a biological matrix for hATM.

Figure 4

Process monitoring. Comparison of data collected during the automated process with data from manual handling steps. A) Dosing accuracy for seeding the cell suspension V  =  200 µL. In comparison rLine(R) 5000 incl. automated tip assembly (n  =  71), single‐channel pipette nominal volume 1000 µL (n  =  72), Multistepper with 10 mL attachment (n  =  71). B) Dosing accuracy at setpoint 35 mL and 20 mL with the robotic system compared to dosing with the measuring pipette (n  =  40). C) Survival rate of hEK after accutase‐mediated detachment. Whiskers indicate minimum and maximum values (A–C). D) Exemplary plot of the online‐acquired measurements of CO₂‐concentration and temperature of a representative 2 h period. The marked areas indicate the unloading and subsequent loading of well plates, showing constant conditions. E) Hands‐on‐time of the robotic system, including the automated preparation of pipettes and transfer of materials into the workspace, loading/unloading of the incubator (n  =  6). Graphs show mean values with SD. F) A schedule displays the time occupancy of the plant to produce and culture 14 batches of RHE over 21 days (14 × 24 RHE  =  336 RHE) within a timeframe of 21 days and 7 h per day representing a workday of manual operators. Seven batches are produced on d1 (dark blue) and d3 (light blue) respectively. The following occupancies represent the culture steps of Airlift and media exchange.

Figure 5

Automated cell classification and counting using a custom trained YOLOv5 based detection pipeline. A) The workflow starts with (1) acquiring multi‐layered images using the microscope; (2) selecting sharp images; (3) employing a sliding window technique to generate cropped images sized to the Neubauer chamber grid; (4) using a YOLOv5‐based detection pipeline to differentiate between live and dead cells; (5) implement non‐maximal suppression to eliminate duplicate detections; (6) finalize the cell count by analyzing labeled instances of live and dead cells. B) The automated microscope comprises (1) motion stages, (2) an objective lens, (3) a light diffuser, and (4) a well plate carrier. C) The Neubauer chamber is positioned within a holder inserted into the microscope loading unit by the robotic parallel gripper. D) Live and dead cells are detected from microscope image acquisition. E) Mean Average Precision (mAP) of 0.98 is achieved for detection of both classes. Moreover, high values of true positive (TP) and true negative (TN) and low values of false positive (FP) and false negative (FN) in the confusion matrix (inset), confirm the efficiency of the classification. F) To validate the automated cell‐counting process, we performed a manual cell‐counting experiment with the collaboration of 15 expert biologists. Graphs show mean values with SD, blue points indicate the prediction of the automated cell count.

Figure 6

Biological evaluation of automated RHE production. A–D) HE‐stained sections of automated RHE (A,B), manually generated RHE (C), and native skin (D). E–G) Immunofluorescent analysis of biological markers CK10 (red) and CK14 (green). Nuclei are stained blue by DAPI. Scale bars = 100 µm (A), Scale bar = 50 µm (B–G). H) Viability analysis by MTT assay comparing the metabolic activity of automatically produced (A n  =  24) and manually generated RHE (M n = 9). I) Analysis of barrier function via impedance spectroscopy. The Bode plot for a representative experiment shows the averaged amplitude and phase response for the automatically cultivated RHE compared to the manual controls (A n  =  12; M n  =  6). Graphs show mean values with SD, H–I.

Figure 7

Skin irritation test adapted by OECD 439. A) HE stained sections of automatically generated RHE, observed 42 h after irritation. Scale bar = 200 µm. B) Quantitative MTT normalized on NC; pictures of RHE after MTT incubation (n = 6; three independent experiments). Graphs show mean values with SD, B.

Figure 8

Comparative analysis of manually and automatically produced hATM. A–D) Histological analyses revealed no obvious morphological differences comparing automatically (A,B) and manually (C,D) produced tissue models. E–J) Immunofluorescent staining confirms cilia‐like structures and differentiated epithelial cells in tissue models, with consistent morphology across E–G) automatically and H–J) manually produced samples, and cell nuclei are stained with DAPI. K. 3D reconstruction of hATM with apical tight junction marker ZO‐1 reveals functional tissue polarization. L–M) Quantitative (L) and qualitative (M) MTT data showed significantly higher cell metabolism in automatically produced tissue models comparing d7 and d21. Graph shows mean values with SD, L (n =  3). *:  p‐value of unpaired t‐test with Welch's correction < 0.05. Scale bar = 100 µm (A,C,E,F,H,I); scale bar = 50 µm (B,D,G,J,K).