Researchers from the Korbel Group at EMBL Heidelberg have developed an innovative AI-driven tool that enhances the investigation of chromosomal abnormalities linked to cancer. This technology aims to uncover the conditions that lead to these errors, potentially offering deeper insights into how cancer initiates.
"Chromosomal abnormalities are a significant factor in aggressive cancers, closely associated with patient mortality, metastasis, recurrence, chemotherapy resistance, and rapid tumor development," explained Jan Korbel, a senior scientist at EMBL and lead author of a recent study published in the journal Nature. "We aimed to determine what influences the likelihood of chromosomal alterations occurring during cell division."
A Century-Old Cancer Theory
The link between chromosomal abnormalities and cancer has been theorized for over a century. Theodor Boveri, a German scientist, first proposed this concept in the early 1900s after observing cells through a microscope, suggesting that abnormal chromosomal structures could contribute to cancer development.
Despite the longstanding theory, researching these abnormalities has proven challenging. Only a limited number of cells exhibit chromosomal defects at any time, and many of these cells perish through natural selection. Consequently, scientists have traditionally relied on manual searches under microscopes, isolating only a few cells for further examination.
Marco Cosenza, a Research Scientist in the Korbel Group, sought a solution by collaborating with other EMBL teams facing similar challenges. Together, they created an automated platform that merges microscopy, single-cell sequencing, and artificial intelligence, named machine learning-assisted genomics and imaging convergence (MAGIC).
AI-Powered "Laser Tag" for Cells
MAGIC operates similarly to an automated laser tag system, scanning cells to identify those with distinctive visible features. In this study, the focus was on a structure known as a 'micronucleus'.
Micronuclei are small compartments within cells that harbor DNA fragments separate from the main genome. Cells with micronuclei are more prone to developing additional chromosomal abnormalities, thereby increasing their cancer risk.
Upon detecting a cell with a micronucleus, the system marks it with a laser, utilizing a photoconvertible dye that changes color when exposed to light.
"This project united many of my interests," remarked Cosenza. "It encompasses genomics, microscopic imaging, and robotic automation. During the COVID-19 lockdown in 2020, I dedicated time to learning and applying AI computer vision technologies to biological image data we had previously collected. We then designed experiments to validate and advance our findings."
How MAGIC Operates
The MAGIC system executes a series of automated steps. Initially, an automated microscope captures a vast array of images from a cell sample. A machine learning algorithm, trained on manually labeled examples of cells containing micronuclei, then analyzes these images.
If the algorithm identifies a cell with a micronucleus, it relays the location to the microscope, which subsequently directs a light beam to permanently tag the cell with the dye. Researchers can later isolate these tagged cells from living populations using flow cytometry techniques, enabling detailed genomic analysis.
By streamlining the traditionally labor-intensive process of searching for micronuclei, MAGIC can analyze nearly 100,000 cells in less than a day.
Unveiling Chromosomal Error Rates
The research team utilized MAGIC to investigate chromosomal abnormalities in cultured cells derived from normal human cells, discovering that over 10% of cell divisions result in spontaneous chromosomal errors. This rate nearly doubles when the p53 tumor suppressor gene is mutated.
Additionally, the researchers explored factors influencing chromosomal abnormalities, including the presence of double-stranded DNA breaks.
Broad Implications for Biological Research
The study involved collaborations within and outside EMBL, including contributions from the Advanced Light Microscopy Facility (ALMF) and the German Cancer Research Centre (DKFZ). MAGIC is designed for flexibility; while trained to detect micronuclei in this study, the AI could be adapted to identify various other cellular features.
"As long as a feature can be visually distinguished from a regular cell, AI can be trained to detect it," Korbel noted. "Our system has the potential to propel future discoveries across numerous biological fields."