Imagine that a promising new drug for chronic pain is stopped during clinical trials. Not because it is ineffective, but because it unexpectedly causes liver damage.
This scenario of hepatotoxicity is often a bitter reality in the pharmaceutical industry, as the liver plays a crucial role in drug metabolism. Hepatotoxicity is a common cause of drug candidate failure.¹ Therefore, understanding liver health and drug-induced hepatotoxicity is not only a scientific endeavor; it can also be a matter of life or death for patients who urgently need a drug that is not approved because it causes liver damage.
Innovative technologies such as AI (artificial intelligence), driving image analysis, are becoming an important tool in ensuring the drug safety and efficacy. Researchers are trying to bridge the gap between in vitro models and human responses in drug development, toxicology, and disease modelling. One such researcher is Simon Plummer.

Simon Plummer, managing director of MicroMatrices Associates Ltd. in Dundee, UK, leads a team focused on researching toxicology through the molecular and morphological analysis of microtissues. Their work involves integrating molecular genomic data with image analysis techniques to better understand how tissues respond to drugs and agrochemicals.

By studying hepatotoxicity, MicroMatrices seeks to improve knowledge of liver health and its critical role in drug metabolism and safety. This research aims to contribute to the development of safer medicines and help customers make informed decisions during the drug development process. Through its work, MicroMatrices aims to translate laboratory findings into practical healthcare applications.

In the pre-AI era, Simon Plummer and his team have used threshold-based image analysis to address issues with multiplex immunofluorescence (mIF) stained images. Threshold-based analysis is more challenging to apply to image analysis of morphological characteristics in hematoxylin and eosin (H&E) stained images due to the non-specific nature of the stain.

One of the challenges they faced in their liver hypertrophy project was to define hepatocyte cell membranes in H&E stained images of liver microtissue sections across species (human and rat). This proved easy in human hepatocytes because there was good contrast between the cytoplasm and the cell membrane in these images.

However, in rat liver microtissue sections, it was difficult to visually identify the cell membrane against a more intensely stained cytoplasm. To create an AI algorithm that accurately performs this task in rat tissue, they Immunohistochemistry (IHC) stained the rat liver microtissues in parallel sections, using a rat hepatocyte marker.

Overall, the MicroMatrices team has significantly reduced the mentioned challenges through AI-driven image analysis.

AI resolves issues associated with spatial/morphological measurements because it can efficiently automate making spatial distance measurements. Historically, histopathological analysis of this kind has been performed semi-quantitatively. AI provides a quantitative solution with greater speed alleviating subjective bias.

This AI-driven image analysis approach is particularly advantageous for leveraging AI in the analysis of the extensive datasets obtained from microtissue microarrays (microTMAs).

The microTMA platform was developed by MicroMatrices in order to efficiently compare treatment groups from an entire microtissue plate-based experiment on a single microscope slide. Furthermore, the planar geometry of the microTMA allows for the generation of up to 20 parallel slides for multiplexed investigations. As this platform enables the simultaneous examination of multiple tissue samples, experimental efficiency is significantly enhanced.

By integrating AI, it is possible to process and analyze these large volumes of data more efficiently, in this case in drug development and toxicology studies, yielding accurate quantitative results with greater speed and precision.

Manually annotating and measuring all the individual hepatocyte cytoplasmic areas would have been very inefficient, however using arivis Cloud streamlined this process.

The process of liver microtissue analysis begins with fixing the liver microtissue, followed by the preparation of microTMA. Once the microarrays are prepared, the next step is to section the microtissues, followed by H&E staining to visualize the tissue structures.

After staining, the samples are scanned using a slide scanner, in conjunction with ZEN microscopy software. This allows high-resolution imaging of the stained sections.

Finally, the images are annotated and analyzed using ZEISS arivis Cloud, which allows for training of AI models with no need to code, to facilitate advanced image processing and data interpretation.

Using microscopy and AI analysis, Simon Plummer and his team were able to measure hypertrophy and demonstrate that this response could be recapitulated in both rat and human liver microtissues. The results showed a significant increase in the cytoplasmic area of the hepatocytes, and the induction of phase 1 and phase 2 enzymes was confirmed by proteomics analysis. Interestingly, the sensitivity of the AI algorithm in measuring this response differed between species.

It was observed that the specificity of the AI algorithm was lower in images of rat liver microtissues, compared to those of human liver microtissues. This discrepancy was attributed to the cell membrane being more clearly defined in human samples due to qualitative differences in the H&E staining.

Thus, the MicroMatrices team addressed questions related to the recapitulation of a key event – liver hepatocyte hypertrophy – in the mechanism of liver carcinogenesis in rats. They used 3D microtissues to model this response and assess, whether it also occurs in human liver microtissues.

Simon Plummer and his team summarized their findings in a paper: published in Frontiers in Drug Discovery, it illustrates their approach to developing technologies that contribute to addressing the challenges of translating drug responses across species. The team's research focused on analyzing microTMAs constructed with drug-treated microtissues to facilitate early pipeline decision making for their biotech and pharmaceutical industry clients.

In our future research, we aim to apply AI to analyze immunohistochemical and fluorescent images.

Early exploration of AI-driven image analysis has revealed several advantages for its application in microscopy. Simon Plummer’s team has found that AI-driven analysis is efficient for measuring distance and morphological parameters, providing accurate and reliable data.

For identifying variations in diffuse staining intensity within images, threshold-based image analysis is a suitable method, while AI-driven approaches may serve different analytical purposes. This highlights the importance of selecting the appropriate technique based on specific research needs, as different methods can be tailored to address diverse analytical challenges.

The next steps in the research are to use AI to identify morphological changes in other microtissues, including those from the brain, heart, and kidney.

MicroMatrices, founded in 2011, is a contract research organization (CRO), dedicated to understanding the molecular mechanisms of drug action in tissues.

The company’s name is derived from the idea that tissues are complex, interconntected matrices of molecules and structures that contain information relevant to biological function.

Its motivation is to develop technologies and provide services to quantify this molecular and morphological data to better understand mechanisms of drug action and safety.

How the microTMA technology SpheroMatrices for 3D cell culture high-throughput histology for drug development and chemical safety testing works