Machine learning and cell imaging combine to predict the effectiveness of multiple sclerosis drugs

Researchers from Brazil have worked with French institutions to develop a tool that can predict how patients will respond to natalizumab, one of the most commonly used drugs to treat multiple sclerosis.

Despite its effectiveness, approximately 35% of users are not fully responding to treatment and experience a return of symptoms within two years of initiating treatment. Additionally, it can reduce the frequency and severity of occurrence and slow the progression of the disease, but can cause side effects such as serious infection (progressive multifocal leukocytosis), headaches, muscle and stomach pain, fatigue, and increased risk of depression.

Using innovative methodologies, a group of scientists have achieved important advances in precision medicine. This advancement will allow patients to receive targeted treatments that improve their future quality of life, have fewer side effects and have positive outcomes in a shorter period. It also reduces costs for public health systems. Natalizumab, a monoclonal antibody, is offered in Brazil as a national public health network (known as SUS), a “Sistemaúnicode Saúde” which costs an average of 10,000 brl per patient per month.

This drug acts as an antibody and blocks the binding of the immune system protein VLA-4 to a molecule called VCAM-1. This prevents immune cells from entering the brain and causing inflammation. After treatment, immune system cells such as CD8⁺ T cells become more rounded. This change is primarily related to the remodeling of actin, a protein that promotes cellular support, but also plays a role in cell movement, shape and interaction.

Using high content imaging (HCI), scientists found that inadequate outcomes from natalizumab treatment are associated with different actin remodeling responses in CD8.⁺ T cells and their ability to extend even when under the influence of drugs. The cells form and become more vertical. These findings were published in the journal Natural Communication.

“The results are important as they can avoid improving the quality of life of patients, unnecessary side effects and delayed treatment, and cost optimization, as in the case of SUS in Brazil. Chaves is currently a researcher at the Institute for Infectious Inflammation in Toulouse (Infinity), France, but has been studying multiple sclerosis for many years in the local unit of the Oswald Ceará, the state of Ceará, Brazil.

Multiple sclerosis (MS) is an autoimmune, inflammatory, and degenerative neurological disease that affects the central nervous system, leading to exercise, cognitive, and psychiatric disorders. Symptoms range from loss of muscle strength to difficulty walking, memory loss, difficulty in paying attention, and mood swings. It is estimated that 2.8 million people worldwide have MS, including around 40,000 in Brazil. Most diagnosis occurs in young adults between the ages of 20 and 50, with women being affected two to three times more than men.

Solve the system

High Content Cell Imaging (HCI) combines advanced microscopy techniques with automated image analysis to extract multiple cell-by-cell information, including shape and size, organelles distribution, protein localization, drug responses, and genetic disorders. HCI is mostly used in cancer research.

Employing this type of analysis has taken a step further compared to other individualistic medical research that generally uses cytometry (analysis of the physical and chemical properties of cells), serology, or transcriptomics (an evaluation of the methods in which DNA information is transcribed into RNA and is used to produce proteins and other molecules).

In this study, the researchers began by applying it in vitro in vitro to blood cells containing T cells, including T cells, from patients with MS who have not yet been treated with drugs. Cells were stimulated via VLA-4 and plated into VCAM-1-coated plates. The sample was from individuals linked to French institutions.

Characteristics of over 400 morphological profiles, including area, width/length ratio, and actin tissue, were extracted. Of these characteristics, 130 presented research-related information. Using machine learning, researchers have created over a million combinations.

This study achieved 92% accuracy in the discovery cohort and 88% in the validation cohort when predicting clinical response to natalizumab treatment. CD8⁺ T cells have been proven to be an associated subpopulation of this prediction. Non-responder patients showed a more resistant actin remodeling profile, characterized by reduced loss of polarity and increased mobility. This suggests that the CD8 maintains its moving state⁺ T cells can impair therapeutic efficacy.

“The project is very interesting and innovative. A great insight was to take images, convert them into numbers and use this table in machine learning. I'm sure we can replicate this type of approach for other illnesses and treatments.” He is also the author of the article.

Nakayama and Juan Carlo Santos E. Silva, Faculty of Pharmaceutical Sciences at the University of Sao Paulo (FCF-USP), worked on modeling and developing machine learning.

What to expect

Chaves and Nakaya told Agência Fapesp that they intend to continue using the methodology and verify the results with a larger sample.

“Now we are looking for ways to test with more patients, including patients from other countries and regions. Another pathway is to make morphological markers more accessible with the possibility of using simpler and cheaper equipment. We also have the possibility to apply methodology to other diseases.