A liver-on-chip model of fatty liver injury under controlled flow conditions using HepG2 cells and AI-based image analysis

Authors: Verónica Peña León^1*, Miguel del Valle Fuentes¹, Laura Aguado Cabanillas¹, Angélica Corral², Diego Velasco² and Alberto E Quintero^1,2

1 B5tec Solutions SL, Madrid, Spain
2 Universidad Carlos III de Madrid, Madrid, Spain

Executive Summary

Metabolic dysfunction-associated fatty liver disease (MAFLD), formerly referred to as non-alcoholic fatty liver disease (NAFLD), is one of the most prevalent chronic liver disorders worldwide. The development of predictive and human-relevant in vitro liver models is essential to better understand disease mechanisms and support the preclinical evaluation of emerging therapeutic strategies.

This technical whitepaper presents the validation of the B5tec Liver-on-a-Chip (LoC) platform, a microphysiological system combining controlled recirculating perfusion, automated flow regulation, and AI-assisted image analysis. The platform integrates a proprietary wireless flow controller that enables stable medium recirculation without direct tissue manipulation, reducing operator-dependent variability and improving experimental reproducibility.

Using HepG2 hepatocytes cultured under controlled perfusion conditions (30 µL/min), we demonstrate that dynamic flow maintains cell viability and supports continued proliferation throughout the experimental period. Following exposure to free fatty acids (FFAs), hepatocytes developed intracellular lipid accumulation and biochemical signs of liver injury. Importantly, controlled recirculating flow promoted a more gradual and physiologically relevant injury response than conventional static culture systems, supporting the development of progressive fatty liver injury phenotypes under dynamic conditions.

To complement biological characterization, we developed a customized deep learning-based image analysis workflow incorporating a Human-in-the-Loop (HITL) strategy. This approach substantially improved cell segmentation performance and enabled automated, unbiased, and reproducible quantification of lipid accumulation and viability-related phenotypes.

Together, these results demonstrate the potential of combining organ-on-chip technology, controlled perfusion, and AI-driven image analysis to establish scalable and reproducible in vitro models of fatty liver injury suitable for mechanistic studies and future preclinical drug evaluation.

Introduction and Clinical Challenge

Metabolic dysfunction-associated fatty liver disease (MAFLD), formerly referred to as non-alcoholic fatty liver disease (NAFLD), is one of the most prevalent chronic liver disorders worldwide. Characterised initially by hepatic lipid accumulation (steatosis), the disease may progress toward inflammation, fibrosis, and ultimately metabolic dysfunction-associated steatohepatitis (MASH).

Despite its increasing clinical relevance, the development of predictive in vitro models capable of reproducing key aspects of disease progression remains challenging. Conventional experimental approaches present important limitations:

• Static Cell Culture Models (2D): Traditional multiwell cultures do not reproduce the dynamic microenvironment of the liver, including physiological fluid flow, nutrient exchange, and mechanical stimulation. Under lipid overload conditions, these systems often exhibit rapid cellular stress and injury responses that may limit the study of progressive disease-related phenotypes over extended culture periods.
• Animal Models: Although widely used in preclinical research, animal models present species-specific metabolic, genetic, and immunological differences that may affect translational predictability. In parallel, increasing regulatory and ethical pressure is driving the adoption of New Approach Methodologies (NAMs) that reduce reliance on animal experimentation.

Organ-on-Chip (OoC) technology offers a promising alternative by recreating physiologically relevant microenvironments through controlled perfusion and microfluidic culture. In liver applications, dynamic flow can support long-term cell viability, improve tissue homeostasis, and provide a more representative platform for studying lipid-associated liver injury and evaluating potential therapeutic interventions.

The B5tec Liver-on-a-Chip Platform

The B5tec Liver-on-a-Chip platform combines microfluidic engineering, automated flow control, and advanced image analysis to provide a robust and reproducible environment for liver cell culture under dynamic conditions. The system is based on three core technological pillars:

• Precision Flow Regulation: a proprietary wireless flow controller integrating custom hardware and dedicated software enables stable, programmable, and reproducible flow conditions. Continuous monitoring and regulation ensure consistent perfusion throughout the experiment.
• Closed-Loop Recirculation: the platform incorporates automated medium recirculation, allowing continuous nutrient supply and waste removal while minimizing manual intervention. This approach reduces the risk of bubble formation and mechanical disturbances commonly associated with repeated medium exchanges.
• Experimental Standardization: automated perfusion protocols reduce operator-dependent variability and improve experimental reproducibility. Such standardization is essential for generating reliable datasets suitable for disease modelling and future drug screening applications.

Experimental Methodology

To evaluate the performance of the platform and its suitability for modelling fatty acid-induced liver injury, a comparative study was performed using four experimental conditions.

Experimental Groups

Experimental Timeline

The study was conducted over a total period of 120 hours.

• 0 h – Cell Seeding: HepG2 hepatocytes were seeded into the microfluidic culture chamber.
• 0–24 h – Stabilization Phase: Cells were maintained under static conditions to promote attachment and monolayer formation.
• 24 h – Flow Initiation: Continuous recirculating perfusion was initiated in the dynamic groups at 30 µL/min.
• 48 h – Fatty Acid Challenge: A free fatty acid (FFA) cocktail was introduced to induce lipid-associated liver injury.
• 120 h – Endpoint Analysis: Cell viability, proliferation, lipid accumulation, and biochemical injury markers (ALT, AST, and LDH) were evaluated. Fluorescence imaging was performed using Hoechst, BODIPY, and Live/Dead staining protocols.

Key Results

A. Controlled Flow Promotes a More Progressive and Physiologically Relevant Injury Response

Continuous recirculating perfusion at 30 µL/min maintained culture viability and supported sustained cellular growth throughout the experimental period. Following exposure to free fatty acids (FFAs), clear differences were observed between static and dynamic culture conditions.

• Fluorescence images obtained using BODIPY staining revealed intracellular lipid accumulation in both static and flow conditions.
• However, static cultures exhibited a more pronounced increase in the release of liver injury biomarkers, including ALT, AST, and LDH, indicating greater cellular stress and damage.
• In contrast, cultures maintained under controlled flow remained viable and metabolically active following FFA exposure.
• Although biochemical markers of injury were also detected under dynamic conditions, their increase was more moderate, suggesting a more gradual cellular response to lipid overload.

BODIPY staining confirmed progressive intracellular lipid accumulation in perfused cultures, demonstrating the development of steatotic phenotypes while maintaining a viable cell population throughout the experimental period. This combination of sustained viability and gradual lipid accumulation provided a quantifiable experimental window for monitoring disease-related phenotypes over time.

Together, these findings suggest that controlled recirculating perfusion supports a more physiologically relevant progression of fatty acid-induced liver injury compared with conventional static culture systems. By reducing the abrupt injury responses frequently observed under static conditions, the platform enables the study of progressive steatosis and associated cellular alterations in a dynamic microphysiological environment.

B. AI-Assisted Quantification Using a Human-in-the-Loop Strategy

Accurate image analysis represents a significant challenge in dense hepatic cultures, where closely packed cells and progressive lipid accumulation can reduce the performance of conventional segmentation approaches. To address this limitation, B5tec developed an AI-assisted image analysis workflow based on the Cellpose deep learning framework combined with a Human-in-the-Loop (HITL) optimization strategy.

• The baseline Cellpose cyto3 model identified 1008 cells in representative HepG2 culture images.
• Following expert-guided refinement and retraining, the optimized model detected 1724 cells within the same dataset.

This improvement substantially enhanced segmentation performance and enabled more accurate quantification of lipid accumulation and viability-related parameters. The resulting workflow provides automated, reproducible, and observer-independent image analysis suitable for high-content organ-on-chip studies. This approach is particularly valuable for monitoring subtle morphological changes associated with progressive fatty liver injury and supports robust quantitative analysis in high-content organ-on-chip studies.

Discussion and Future Perspectives

The results presented here demonstrate both the technical robustness and biological relevance of the B5tec Liver-on-a-Chip platform. Controlled recirculating flow maintained stable hepatic cultures, supported long-term viability, and enabled the development of fatty acid-induced injury phenotypes under dynamic culture conditions. Combined with AI-assisted image analysis, the platform provides a scalable framework for quantitative studies of liver injury and metabolic disease mechanisms.

Potential applications include:

• Preclinical Drug Evaluation: The platform may support the assessment of compounds targeting lipid accumulation and early liver injury by providing a reproducible and physiologically relevant in vitro testing environment.
• Advanced Human-Relevant Liver Models: The modular architecture of the system facilitates future integration of additional liver-relevant cell types, including endothelial cells, Kupffer cells, and hepatic stellate cells, enabling the development of increasingly complex and physiologically representative liver microphysiological systems.

As regulatory agencies and industry continue to adopt New Approach Methodologies (NAMs), scalable organ-on-chip platforms combined with automated data analysis are expected to play an increasingly important role in biomedical research and preclinical development.

About B5tec Solutions

B5tec Solutions S.L. is a biotechnology and engineering company based in Madrid, Spain, dedicated to the development of advanced organ-on-chip technologies, wireless flow control systems, and automated microphysiological platforms. By combining microfluidics, electronics, software, and artificial intelligence, B5tec develops innovative tools that support human-relevant research models, accelerate biomedical innovation, and contribute to the transition toward animal-free testing strategies.

References

• European Society for Organ-on-Chip (EUROoCS) roadmap documents & technical guidelines, 2025.
• Smith, A. et al. “Modelling NAFLD/MAFLD in dynamic microphysiological systems: the role of shear stress in lipid metabolism.” Journal of Hepatology in vitro , vol. 12, no. 3, pp. 245-258, 2024.