June 18, 2024 | Researchers in Israel who pioneered the discipline of “mechanomics” are now using the concept to fill the chasm between the expanding volume of ‘omics data and its translation into real-world clinical scenarios. Their aim is to help physicians predict the aggressiveness of cancer and a tumor’s potential resistance to treatment and, ultimately, improve patient outcomes, according to Ofra Benny, a professor at the school of pharmacy and doctoral student Yoel Goldstein, in the faculty of medicine at Hebrew University of Jerusalem.
The diagnostic approach uses machine learning to measure the nanoparticle uptake pattern of cancer cells. Artificial intelligence (AI) has in this way “revolutionized” the converging fields of biology and materials science, says Benny.
The methodology, described in a recently published article in Science Advances (DOI: 10.1126/sciadv.adj4370), uses the connection between cell biomechanics and cancer cell functions to classify cells through mechanical measurements. Three pairs of human cancer cell subpopulations—lung, with varying levels of cisplatin resistance; prostate, with different metastatic potential; and melanoma, which vary in their invasiveness—were studied.
Notably, researchers were able to accurately predict invasiveness 95% of the time, based solely on patterns of particle uptake. “This is very promising because it means that we can conclude the behavior of cancer cells... with a relatively simple tool,” Benny says.
Discerning the behavior of cells is complicated because it is the “sum of huge number of parameters,” she adds. The business of measuring the mechanical properties of cells involves complicated methods such as squeezing cells under a microscope (atomic force microscopy) to determine whether they are soft or stiff, which itself results from a combination of multiple molecular pathways.
It is difficult work because mechanical properties can adapt to match their surroundings. And doing the analysis on a big-data scale was, up until now, an impossibility, says Benny.
To acquire particle uptake patterns, cells were exposed to nanoparticles of different sizes, each associated with a unique color, or fluorescent signature, she explains. Machine learning was used to precisely quantify the number of particles consumed by each cell, with the uptake patterns being analyzed by flow cytometry. The training phase was done on cells with known functions, thereby allowing to associate between particle uptake patterns and specific biological functions.
Particle uptake patterns, reflecting multifactor phenotypic data, became the means to classify cancer cell subtypes. The next step of advancing this technology would be to move from the cancer cell lines used in the study to clinical samples with more inherent noise and complexity and, longer term, test the platform’s ability to predict the fate of patients on different treatments.
It has proven difficult to predict how a tumor will behave because cancer is a complex set of diseases. Even patients with cancer in the same tissue can have strikingly different experiences and outcomes based on the subpopulation of tumor cells, Benny says.
“We know that no matter what treatment patients get, eventually, in most cases, [they] either develop resistance to the specific drug or have recurrence or metastasis,” she continues, noting that invasiveness is the main reason for cancer morbidity and mortality. Although many great tools exist to help make predictions—genomics, metabolomics, proteomic—the information they produce is often useless to oncologists because it doesn’t tell them which drug to give a patient or if a certain tumor will become invasive or stay local.
Based on prior research demonstrating that cancer cells can efficiently engulf nanoparticles and microparticles (Science Advances, DOI: 10.1126/sciadv.aax2861), Benny and her colleagues thus began to explore how to look at tumor cells in a more holistic way. The earlier study found a three-way correlation between cell deformability, phagocytic capacity (maximum number of particles that can be taken up by one macrophage), and cancer aggressiveness, pointing to the possibility of a mechanical surrogate marker of malignancy.
Tumor cells, unlike normal ones, are “mostly soft and can move around and twist their shape,” she says, as has been shown previously by the group and numerous other studies. The important new discovery was that these difficult-to-measure differences are linked to how cancer cells absorb nanoparticles, which in turn reflect their function. The research team therefore decided in the latest study to use nanoparticles as a kind of biosensor.
They began with a subpopulation of cancer cells whose resistance to certain drugs was tested and known, to train their machine learning algorithm. The model was then used to look at the different particle uptake pattern of the lung cancer cell line pair to classify those that were and weren’t treatment-resistant, which might also be learned via existing laboratory tests.
To take it one step further, the research team went through the same exercise for another cell function—namely invasiveness, which was likewise linked with biomechanical changes—this time using the prostate and melanoma cell lines. “We took cells that physically looked similar but behaved differently,” Benny says, as signaled by their particle uptake pattern.
Given any cancer and any drug, the AI-powered platform could provide answers to any clinical question related to the mechanical properties of cells, says Benny. The algorithm, once trained and given enough data, could start making predictions.
Benny and her colleagues have a patent on the cancer assessment methodology and kit, she says. The multi-sized particle set used in the study will not change because it is part of the translational process that also includes clinically relevant instrument protocols.
Everything happens outside of the body, she notes, so the nanoparticles needn’t be biodegradable at the same level as if they needed to be injected. They are made of solid plastic (synthetic polymer) particles that get externally added to cell line cultures.
Last year, her lab started a spinoff company, called Pre-Cure, with a mission to develop new solutions in precision therapy for cancer patients. These include a 3D-printed tumor-on-a-chip that can be used to screen different drugs to predict the best ones to go after an individual patient’s tumor.
The plan is to combine the new mechanomics methodology with the tumor model to enable personalized medicine, says Benny. “Then we may be able to predict whether a specific patient would need more monitoring... or maybe another course of treatment.”
The cancer analysis platform is expected to be a laboratory developed test, but over the longer term could be used for personalized drug delivery using nanoparticles to target tumors, she says. In terms of diagnostics, an approach combining imaging scans and tissue biopsies with mechanomics probably makes the most sense since cancer is such a complicated disease.