Linda van den Berk

The occurrence of serious side effects is one of the main reasons that a drug does not reach the market or is withdrawn after approval. The liver, kidneys, and heart are often affected by side effects, and the damage caused is often irreversible. Most side effects arise from using a drug at too high a dose and are therefore reasonably predictable. However, in a very small portion of patients, toxicity also occurs during normal use; this is called idiosyncratic toxicity. There are multiple causes of idiosyncratic toxicity, including interactions between the drug and patient-dependent factors such as an individual's genetic makeup, the presence of underlying diseases, and inflammatory reactions. Drugs that cause serious idiosyncratic toxicity are prescribed less frequently and receive a severe warning in the package leaflet. It is important to research the mechanisms leading to these side effects to predict at an early stage whether a patient is sensitive to this toxicity, with the aim of preventing eventual organ damage.

Most medications activate certain processes, known as adaptive stress responses, to protect organs from damage. If the damage remains within limits, a cell can recover. However, when the damage reaches a threshold, it will switch to specific cell death programs. Both stress, cell death, and an underlying infection can cause a certain type of immune cells, called macrophages, to produce the pro-inflammatory protein TNF-α. In a healthy or minimally stressed cell, TNF-α can promote survival via the activation of the transcription factor NF-κB. In the case of a cell highly stressed by drug use, the presence of TNF-α can disturb this balance and lead to cell death.

To determine early in drug development whether a drug causes certain stress, preclinical experiments are needed. Various types of preclinical tissue culture models are available to test medicines. However, for specifically studying TNF-α-induced stress, a preclinical model is needed that responds well to this inflammatory protein to maximize the chance of picking up this type of side effect in the laboratory.

A relatively new category of models for studying organ development, disease, and toxicity is centered around the use of pluripotent stem cells. These stem cells are reprogrammed from a somatic cell, for example, from skin, urine, or blood, and have the capacity to subsequently differentiate into any cell of the body. By mimicking embryonic development outside the body, hepatocytes (liver), cardiomyocytes (heart), and the proximal tubule (kidney) can be created and then exposed to drugs to study drug safety.

In this thesis, we compared various preclinical models and used them to measure (TNF-α-induced) stress to determine which model is most useful for us in the lab.

In Chapter 2, we looked at whether an existing model using HepG2 liver carcinoma cells could be made more relevant. Under 2D culture conditions, these cancer cells have low expression of proteins essential for drug conversion. By culturing HepG2 cells in 3D in Matrigel, the expression of such proteins important for drug metabolism is increased. However, Matrigel originates from mouse tumors and exhibits significant variability between different batches, meaning reproducibility cannot be guaranteed. By replacing Matrigel with a self-synthesized gel containing arginylglycylaspartic acid (RGD) peptides, we were able to generate 3D spheroids with better metabolism than those made in Matrigel, based on the expression of various enzymes, drug transporters, and other essential markers relevant in the liver. This new gel seems to be a good alternative for creating HepG2 spheroids, but future research must determine whether this new model is also better at predicting toxicity after drug exposure.

To predict immune-mediated toxicity, a liver model is needed that responds well to pro-inflammatory proteins. Therefore, in Chapter 3, we assessed how five different liver models react to different immune-related cytokines, such as TNF-α, IL-1β, IL-6, IL-10, IL-12, and IL-18. The various liver models consisted of primary human hepatocytes, considered the gold standard in research into drug side effects in the liver. We also cultured HepG2 cancer cells in two different culture media and differentiated pluripotent stem cells into hepatocytes. The final model consisted of spheroids composed of multiple cell types (hepatocytes, stellate cells, and Kupffer cells) present in the liver to better mimic the liver. Based on TempO-seq analysis, a targeted RNA sequencing technique, most cytokines, including IL-6, IL-10, IL-12, and IL-18, had a minimal effect on all these liver models. The pro-inflammatory cytokine IL-1β caused a response in all five liver models, but we could not observe a difference in the degree of reaction. However, when we studied the effect of TNF-α, we could notice differences in how a liver model reacts. Primary human hepatocytes reacted much less strongly to TNF-α compared to the other models, leading us to conclude that these primary hepatocytes are not the most suitable model for investigating the interaction between a drug and TNF-α. In contrast, hepatocytes made from pluripotent stem cells showed a similar but many times stronger response to TNF-α compared to primary hepatocytes. Therefore, we subsequently further investigated in the stem cell model whether non-steroidal anti-inflammatory drugs (NSAIDs) and TNF-α interact and induce immune-related stress. Ultimately, we concluded that this interaction and the accompanying cell death were as expected with several NSAIDs already known to cause immune-induced liver toxicity.

Since we showed in Chapter 3 that hepatocytes made from pluripotent stem cells respond well to immunomodulators, we decided this was the best model to further investigate the interaction of a drug and TNF-α on larger groups of medicines. This is usually an expensive and time-consuming process. Ideally, the potential toxicity of these candidate drugs should be determined quickly and cheaply based on the activation of stress response proteins, as is the case with inflammation. To determine if inflammation occurs, we can use pluripotent stem cells where certain proteins are linked to a fluorescent protein, differentiate these into hepatocytes, and then expose them to a candidate drug and TNF-α. Under a confocal microscope, these cells then light up green, and we can determine if inflammation has occurred. In Chapter 4 of this thesis, we describe a method to systematically identify genes induced during inflammation in the liver. A20, LCN2, RND1, IRF1, VCAM1, and ICAM1 were eventually selected as specific inflammatory biomarkers, and the fluorescent protein eGFP was linked to these proteins to make the cell light up when inflammation occurs. For A20, IRF1, VCAM1, and ICAM1, we successfully demonstrated that this method of linking a fluorescent protein to a biomarker works in pluripotent stem cells. Additionally, these cells lit up when differentiated into hepatocytes. Finally, we described that these biomarkers not only unravel inflammation in the liver but that these fluorescent pluripotent stem cells could also be differentiated into cardiomyocytes to demonstrate the activation of inflammatory defense mechanisms.

We further demonstrated the utility of these fluorescent stem cells in Chapter 5. Here, we differentiated stem cells where the inflammatory protein ICAM1 is linked to eGFP into hepatocytes, cardiomyocytes, and proximal tubule cells of the kidney. We then treated these cells with a relatively new group of candidate drugs called kinase inhibitors (760 candidates) and TNF-α. We were among the first to show that screening such a large set of medicines is possible with a stem cell model. This screen showed significant overlap between drugs that block ICAM1 expression in the different cell types. However, when looking at candidates that increase ICAM1 expression, and thus inflammation, we see almost no similarity between these candidate drugs and the different cell types. Subsequently, we delved deeper into candidates that increase ICAM1 expression in the proximal tubule of the kidney, as this is associated with attracting and activating immune cells that can cause further damage to cells and organs. We validated that kinase inhibitors which cause increased kidney inflammation in clinical trials were also identified in our screen. Additionally, we showed at a molecular level that this is likely due to the increased secretion of certain pro-inflammatory chemokines responsible for attracting immune cells that stimulate further inflammation and possible damage. Finally, we describe that these kinase inhibitors often do not inhibit a kinase protein specifically but often have off-target effects and therefore potentially cause more side effects. This highlights the importance of being able to predict in early stages of drug development whether a candidate will cause side effects.

In this thesis, we have described how we can identify the most relevant preclinical model for immune-related toxicities in the liver, kidneys, and heart. The use of fluorescent reporters for inflammation in pluripotent stem cells and their differentiated cell types contributes to this. Furthermore, we can test large groups of new medicines in a relatively simple way with these preclinical experiments. Based on these outcomes, other follow-up tests can be performed that contribute to a broader insight into the molecular mechanisms behind these toxicities. This contributes to the cheap and rapid detection of potential side effects in early stages of drug development, and in this way, new medicines will hopefully appear on the market faster in the future.