Speaker
Description
Using a machine learning-supported approach for assessing and predicting the susceptibility of articular cartilage to mechanical trauma-induced changes in cellularity
M. Selig1,2, Laura Saager, Klaus Böhme, Bodo Kurz2, and B. Rolauffs1
Presenting Author: Mischa Selig, mischa.selig@uniklinik-freiburg.de
1G.E.R.N. Research Center for Tissue Replacement, Regeneration & Neogenesis, Dept. of Orthopedics and Trauma Surgery; 2Albert-Ludwigs-University of Freiburg, Hugstetter Straße 55, 79106 Freiburg, Germany; 3Institute of Anatomy, University of Kiel, Otto-Hahn-Platz 8, 24118 Kiel, Germany;
INTRODUCTION: The diagnosis and potential prevention of early post-traumatic osteoarthritis (PTOA) remain shot topics in orthopedic research because PTOA is among the leading causes of worldwide disability. In clinical routine macroscopic tissue damage after trauma is being diagnosed using imaging methods such as MRI or x-ray. Besides macroscopic damage, post-traumatic tissue regeneration vs. progressive degeneration to full PTOA depends also on the extent of cell death and survival, which can currently not be assessed clinically. We used here a clinically applicable score, the superficial chondrocyte spatial organization (SCSO) as a surrogate marker for articular cartilage ultrastructure and nanoscale functionality. We asked whether (i) the SCSO determines cell survival after simulated injury and (ii) can be used to predict the extent of cell death and survival. The ability to quantitatively assess a given patient’s cell population after trauma and even predict his / her susceptibility to increased post-traumatic cell death would help improving diagnostic capability.
METHODS: Discs from human OA articular cartilage (AC) explants were fluorescently-labeled with Calcein AM (Thermo Fischer), a cell viability indicator, and Propidium iodide (Sigma- Aldrich), a cell dearth indicator. Each disc’s SCSO was classified as chondrocyte string, double string or cluster organizations. One group of discs was subjected to mechanical injury (50%/0.5 s/0 s). Injured and control discs were stained for cell viability and cellularity, single cell morphology (area, length, width, circularity, roundness, and solidity), and multiple quantitative SCSO (qSCSO) parameters (nearest neighbor cell-cell distance (NNDs), cell intensity, and measures of cell grouping) were calculated for each disc. Both the morphological and the qSCSO data of viable cells were then used for training a random forest regression model (RF, R2: 0.955) to predict the extent of cell death and survival. Statistical analyses were performed with SigmaPlot (α<0.05).
RESULTS: A significantly higher number of chondrocytes organized as strings and double strings survived injury than chondrocytes organized in cell clusters and, conversely, a significantly lower number of chondrocytes organized as strings and double strings died after injury (p<0.001), demonstrating an SCSO-dependent susceptibility to trauma-induced cell death and survival. Chondrocyte morphology and qSCSO parameters were significantly different between SCSO stages and between injured vs. control discs (p<0.001). The RF model excellently predicted the amount of injury-surviving cells after trauma (R2: 0.946).
DISCUSSION & CONCLUSIONS: The extent of chondrocyte cell death after trauma depends on the SCSO, revealing SCSO-dependent trauma susceptibility. Furthermore, qSCSO and single cell morphology can be used as machine learning predictors to predict SCSO-dependent trauma susceptibility with excellent precision.
41883633849
