Every third Friday of the month, researchers affiliated with Cancer Center Amsterdam (CCA) gather for the ‘CCAll’ lectures where a selected principal investigator (PI) and research team present an overview of their scientific interests and experimental data. At the coming CCAll meeting on September 15, Dr. Bart Westerman and his team will discuss their work on crafting more effective therapeutic approaches against glioblastoma brain tumors.

The challenge of treating glioblastoma

Glioblastoma is the most common primary brain cancer. It is a devasting disease because it infiltrates the brain, is highly resistant to therapy, and is protected by the blood brain barrier that limits the delivery of toxic substances to the brain including most drugs.

The diffuse growth of tumor cells in the brain reduces the effectiveness of surgery and chemoradiotherapy. The disease is also prone to relapse in areas that were not initially affected or treated. Attempts to improve therapy have shown that there is a translational gap between laboratory findings and clinical practice. While this is true for many tumor types, the gap for glioblastoma is especially wide given its unique and protected location.

We can point to few reasons explaining this gap and the limited progress in bridging it:

(1) only few drugs can cross the blood brain barrier,

(2) there are only a few targetable genetic modifications in glioblastoma and these are associated with pathway cross-talk and redundancy,

(3) there is limited immune system involvement - we have found that only 4% of patients have a substantial level of T-cell infiltration.

Given the failure of the traditional single-target or single-modality therapies in the clinic, our group focuses on using multi-target therapeutic approaches.

Our group aims to:

1. Apply machine learning to understand the molecular drug structures contributing to blood brain barrier penetration

2. Use machine learning coupled to wet-lab experiments to identify effective multi-target therapies and prioritize them based on their expected adverse side effect profiles

3. Match these multi-target therapies to mechanisms that trigger immune an immune reaction against cancer cells

Use of cheminformatics to identify compounds that enter the brain

We have developed a machine learning method to understand which molecular structures of small molecule compounds contribute to blood brain barrier (BBB) penetration. Our results show that drug features that predicted passive diffusion over membranes overlap with features that explain endothelial permeation of approved central nervous system-active drugs for a large part. We have also identified which physical properties and molecular substructures contribute to active influx and efflux to and from the brain. These findings provide guidance towards identifying brain permeable compounds by optimally matching physicochemical and molecular properties to different transport mechanisms.

Machine learning to identify effective multi-target therapies and limited toxicity

Glioblastoma is genetically driven by kinases such as EGFR, CDK4/6 and PDGFRA. We have previously found that the simultaneous inhibition of multiple kinases (called poly-pharmacology) can lead to more durable effects, especially when combined with multi-target drugs that induce apoptosis and/or endoplasmic reticulum (ER) stress.

To match multi-target drugs to their respective tumor targets, we have generated a target predictor for kinase inhibitors. Based on the availability of drug-kinase structures, we developed a convolutional neural network model to predict the target profiles of these kinase inhibitors using 270,000 compound ligand measurements. We are currently performing a similar approach for drugs that induce apoptosis and/or ER stress.

A concern when working with multiple targets is that this could lead to unexpectedly strong side effects. Therefore, we have also developed a back-translational approach to assess whether multi-target drugs are likely to cause unexpected side effects in patients.

Characterize immune cell infiltration and its relation to therapy

We have found that only 4% of patients have a substantial level of T-cell infiltration, whereas the majority of tumors contain up to 50% macrophages. We have also found that genetic loss of the interferon alpha/beta locus correlates to the absence of T-cell infiltration into malignant brain tumors. We will investigate which environmental cues affect CD8+ T cell infiltration in relation to the presence of macrophages leading to immune response dampening.

For more information, contact Dr. Bart Westerman.

Team members:

·Ammarina Beumer Chuwonpad (postdoctoral fellow)

·Olivier J. M. Béquignon (postdoctoral fellow)

·George Kanev (postdoctoral fellow)

·Leon van Hout (PhD student)

·Megan Houweling (PhD student)

·Asli Kucukosmanoglu (PhD student)

·Fleur Cornelissen (PhD student)

·Silvia Scoarta (PhD student)

·Yoran Broersma (PhD student)

·Xiangming Cai (PhD student)

·Anna Giczewska (PhD student)

Bart Westerman (PI)

Collaborating companies



Financial support

Brain Tumour Charity








Text by Dr. Bart Westerman

This article was created for Cancer Center Amsterdam.