Study of an Automated High-Throughput Microinjection System for Adherent Cells


Student thesis: Doctoral Thesis

View graph of relations


Related Research Unit(s)


Awarding Institution
Award date12 Jan 2021


Microinjection is a technique that uses a glass micropipette to deliver a minuscule amount of liquid substances into a cell. It has been studied for years and used in biomedical applications, such as clustered regularly interspaced short palindromic repeat (CRISPR) genome editing with the CRISPR-associated protein 9 (Cas9), high-throughput transfection of deoxyribonucleic acid (DNA), and measurement of gap junctional intercellular communication. Robotic microinjection can accurately deliver a large variety of foreign materials to almost every type of cell at the single-cell level and significantly avoid viral- or chemical-mediated cell transfection's immunogenicity, chemical toxicity, and electroporation's high death rate. However, the microinjection technique currently only has a throughput of hundreds to 1000 cells within one production cycle, far less than the number required for clinical use. The throughput cannot be increased by merely conducting additional existing experiments one by one for a longer time to obtain more injected cells because cells outside the incubator only maintain their vitality for a limited time and probably lose their properties' consistency if the overall processing duration is too long. In summary, low throughput remains a problem for successfully implementing robotic microinjection in practical applications. The thesis presents a new automated microinjection system equipped with two micromanipulators and a deep learning-enhanced cell detection and segmentation scheme from the three following aspects.

First, micromanipulators are doubled and integrated onto the same cell processing platform to double the throughput. Two motorized micromanipulators are introduced to construct a unified dual-module system. Such a centralized scheme decreases manual operation, maintenance cost, and malfunction chances than a decentralized scheme. The two micromanipulators insert cells cooperatively in an optimal injection path to save time. Optimization of the injection sequence of a total of n cells with two micropipettes can be formulated to an equality-generalized traveling salesman problem (E-GTSP) of 2n+1 nodes, which can be transformed further into one symmetric traveling salesman problem (TSP) of 4n+2 nodes resolved by the program Concorde solver.

Second, selecting cells for injection is automated to replace manual operations to further improve the system's operational efficiency and increase the throughput in each time unit. Cell detection and segmentation are used to identify living cells in digital images from the microscope camera. Both the traditional and deep learning-based image processing algorithms are used to automate the selection of cells for injection. In the first stage, the living cell nuclei are stained with a fluorochrome Hoechst 33342 to display blue-cyan fluorescence. A series of traditional, data-independent image processing algorithms are used to segment stained cell nuclei. In the second stage, two deep convolutional neural networks are proposed to detect and segment stain-free living cells. The new algorithms avoid monotonous preparation for cell staining, fluorescent dye toxicity, and potential interference with injection samples.

Third, the micromanipulators' working mode is adapted to cooperate with the microscope stage to provide a consistent injection performance to reduce unintended manual intervention, such as changing micropipettes during experiments and increasing throughput in a single experiment. The micropipette tips' penetration depths are adjusted based on the Petri dish bottom's horizontal variation when the motorized stage moves around. One penetration depth throughout the experiment often leads to injection failures. The Petri dish bottom, the dish holder plate, and the motorized stage installation are not strictly horizontal. As such, identifying the moving plane of the dish bottom and adjusting the penetration depth are necessary to avoid frequent re-calibrations and enable nonstop injections. The identification is achieved by moving the stage to three non-collinear locations to collect physical positions and fit them by the moving plane.

Experiments of massive injections on MC3T3-E1 fibroblast-like adherent cells are performed to evaluate cell detection efficiency, injection speed, success rate, and survival rate. Results confirm that one micropipette can inject over 1500 cells, while two micropipettes can inject around 4000 cells in 1 h.

In summary, the proposed microinjection system can achieve high-throughput cell injection with high efficiency. The system can be advantageous when the cells are transfection-challenging, and the injection samples are membrane-impermeable.