Morphokinetic analysis of live-cell imaging data

Morphokinetic analysis of live-cell imaging data Belyaev I. 1,3, Marolda A.2, A. Medyukhina1, K. Huenniger2, O. Kurzai2,4, M. T. Figge1,3 1Research Group Applied Systems Biology,, Leibniz Institute for Natural Product Research and Infection Biology – Hans Knöll Institute, Jena, Germany 2 Research Group Fungal Septomics, Leibniz Institute for Natural Product Research and Infection Biology – Hans Knöll Institute, Jena, Germany 3Institute of Microbiology, Faculty of Biological Science, Friedrich-Schiller-University Jena, Germany 4University of Würzburg, Würzburg, Germany 6. Single cell characterisation An example of different cell morphotypes (left) and density distributions plots for whole populations for feature descriptors (right). We used cell footprint area as descriptor of size and gradient-based features as a roughness measure. Bimodality is found in characteristics of infected populations. 7. Non-spreading cells modelling The ratio of correct/erroneous classification for each movie over all iterations for C. albicans infected (red) and C. glabrata infected (green) blood samples. Classification was done using SVM with leave-one –out sampling. 8. Analysis of static images 9. Analysis of semi-kinetic data We used Data-driven SIMCA (Pomerantsev, Rodionova, 2014) for static cell classification. This method utilised PCA to approximate any regular behavior within the class. The score distance h and orthogonal distance v are used for model interpretation and new objects classification. Box diagrams of the estimated fraction of non-typical cells for each movie (left) and the pulled data sets for the three infection scenarios (right). Bootstrapping was performed with 103 iterations. Contact: [email protected], Research group Applied Systems Biology, HKI, Jena, Germany Reference: 1. Pomerantsev A. L., Rodionova O. Ye. Concept and role of extreme objects in PCA/SIMCA, Journal of Chemometrics, vol. 28, 2014. 2. Hong C.-W. Current Understanding in Neutrophil Differentiation and Heterogeneity, Immune Netw., 2017, 17(5):298-306. Review article.. Automated characterisation of cells based on interpretable features, in order to establish a dynamic hemogram from whole blood infection assays that goes beyond standard blood count examination by integration of information on migration and interaction of blood cells, especially neutrophils. 2. From blood sample to diagnose: the workflow 3. Ex vivo human whole-blood infection assay and neutrophil isolation Imaging Cell detection Descriptor extraction Tracking Semi-supervised classification Population analysis Ex vivo assay Human whole blood from healthy volunteers was either infected with C. glabrata or C. albicans and compared to mock-infected control samples. Polymorphonuclear neutrophils (PMNs) are isolated from whole blood for each condition. Non-target cells are removed by immuno-magnetic depletion using MACSxpress Beads, yielding untouched target cells of high purity. 4. Live-cell imaging of primary human neutrophils Examples of single frames for mock-infected (left) and infected (right) samples. It is possible to see two types of morphological appearances: spreading (S) and non-spreading (N) PMNs. For the majority of PMNs in the C. albicans infection scenario the typical duration of cell spreading episodes is shorter than in the C. glabrata infection case. Medium: RPMI1640 with 5% heat-inactivated human serum. Vital dye: propidium iodide (PI), 2.5 ng/ml. Environment conditions: 37°C, 5% CO2. Imaging: Zeiss LSM 780, DIC, time-lapse, 7 sec, ~0.2 μm/px. 5. Hypothesis A cell population in any specimen can be described as a mixed distribution: 𝑀𝑆𝑆 = 1 − 𝜇 𝑁 +𝜇𝜇, where 0≤ 𝜇 < 1 is the fraction of S-cells. It was observed that S-cells were rarely present in mock-infected samples, which allows creating a soft model of N-cells . 1. Project aim 10. Summary A visual inspection of misclassified sets revealed similar PMNs behavior in different infection scenarios. It indicates, that our approach works properly, but it also means that the hypothesis about infection-specific time pattern of spreading regime was not confirmed. There are at least two contributing factors, which we can speculate about: the donor specificity and PMNs population microheterogenity (Hong, 2017).