CytoMEMS Project
- What?
CYTOMEMS aims for the first smart MEMS device performing high content biophysical characterisation of cells for classification by statistical learning. Electrical and mechanical measurements, performed by a MEMS device, are treated in a single system with embedded analog and digital components. Cell identification is computed by statistical learning from a comprehensive set of biophysical parameters. The assembled device computes key cell characteristics to tune the sensor for sorting the cell.
- Why?
Blood carries cells and elements circulating through the body, e.g. immune, red and platelets that are responsible for vital processes. These circulating cells are of great importance considering the number and severity of the associated diseases, e.g. cardiovascular, immunity, hematology, hemostasis and cancer. In diseases of the blood, changes in cell mechanical properties can have profound effects on cells’ ability to flow normally through the vasculature as increased stiffness prevents cells flow through small capillaries. Better knowledge of biophysical changes, e.g. deformability, viscosity and membrane conditions, is crucial to have a better understanding of cell functioning and related diseases.
Single-cell characterization is essential for studies involving circulating cells due to the high heterogeneity, and extremely low concentration of cancer cells. Cell cytometer, providing high throughput using biomarkers, suffers when cell subpopulations lose their properties. For example, circulating tumour cell (CTC) analysis based on EpCAM may miss a potentially important population of cells that have undergone a transition to switch into mesenchymal phenotype while spreading from a primary tumour. Such cells show higher invasiveness and resistance to apoptosis. Also, useful measures to prohibit disease progression requires detecting cancer cells in circulation as early as possible. Thus, a reliable, practical and rapid method is crucial to include cancer cell analysis in routine medical examinations, which demands alternative label- and marker-free methods.
Many biological studies suggest that biophysical properties (i.e. mechanical) of cells can potentially be used to reflect the state of their health. For example, depending on the inflammatory conditions, leukocytes undergo cytoskeleton and mechanical changes, and exposure to pro-inflammatory cytokines or chemotherapeutic drugs strongly decreases cell deformability. Similarly, a reduction in stiffness with increasing metastatic efficiency in human cancer cell lines has been reported.
- How?
The scientific and technological work has been organized in 3 parts.
The MEMS sensor is optimised from our original process. Various electrode configurations are designed to measure either electrical or mechanical cell characteristics under no or controlled deformation. At first multifrequency electrical measurement reveals the cell size, membrane capacitance and subcellular element response. Downstream electrodes are actuated according to size enabling the cell mechanical and electrical characterisations under controlled deformation. The research work on the MEMS aims in finding optimal device design, especially for the moving electrode to reach practicable throughput (50-100 cells/s) with the proper signal generation. Innovative microfluidic development (i) handles the fluid in this open channel configuration, (ii) tunes and controls the transporting flow, (iii) manages the cell passage in the different sensing point and (iv) finally, collects cells.
The signal processing chain pilots the overall operations of the sensor including the challenging real time sensor tuning, the – in flow – data treatment for cell classification and sorting. All these functions are integrated in a single system embedding the required analogue and programmable digital components. For low noise measurements, signal readouts are processed by lock-in amplification. The architecture features challenging time-domain data management to detect and pool asynchronous events relative to a single cell among a constant flow of data generated from new coming cells. Embedded data processing computes in real time the key cell characteristics to tune the sensor configuration (actuated elements) and to classify the cell within the statistical/machine learnt multi parameters landscape.
The statistical/machine learning work package develops the classification and prediction method enabling the statistical cell identification from their biophysical characteristics. These computational methods are trained from large-scale, heterogeneous, continuous and spatial datasets including high numbers of cells’ electrical and mechanical properties. The statistical analysis relies on canonical correlation analyses, classification, regression methods for complex heterogenous, large dimension data. Machine learning methods (SVM, Random Forest, Boosting classification Trees, Knn classification methods) predict the cell class.
- Who?
CytoMEMS brings four complementary skills together:
Team 1: MEMS by Pr. Dominique Collard, LIMMS
Team 2: Statistical learning by Pr. Sophie Dabo, MODAL
Team 3: Microfluidics by Dr. Mehmet Cagatay Tarhan, JUNIA
Team 4: Signal processing chain by Dr. Nicolas Delorme, ASYGN