Human bone marrow mesenchymal stem cells (hBMSCs) are widely used cell source for clinical bone regeneration. Achieving the greatest therapeutic effect is dependent on the osteogenic differentiation potential of the stem cells to be implanted. However, there are still no practical methods to characterize such potential non-invasively or previously. Monitoring cellular morphology is a practical and non-invasive approach for evaluating osteogenic potential. Unfortunately, such image-based approaches had been historically qualitative and requiring experienced interpretation. By combining the non-invasive attributes of microscopy with the latest technology allowing higher throughput and quantitative imaging metrics, we studied the applicability of morphometric features to quantitatively predict cellular osteogenic potential. We applied computational machine learning, combining cell morphology features with their corresponding biochemical osteogenic assay results, to develop prediction model of osteogenic differentiation. Using a dataset of 9,990 images automatically acquired by BioStation CT during osteogenic differentiation culture of hBMSCs, 666 morphometric features were extracted as parameters. Two commonly used osteogenic markers, alkaline phosphatase (ALP) activity and calcium deposition were measured experimentally, and used as the true biological differentiation status to validate the prediction accuracy. Using time-course morphological features throughout differentiation culture, the prediction results highly correlated with the experimentally defined differentiation marker values (R>0.89 for both marker predictions). The clinical applicability of our morphology-based prediction was further examined with two scenarios: one using only historical cell images and the other using both historical images together with the patient's own cell images to predict a new patient's cellular potential. The prediction accuracy was found to be greatly enhanced by incorporation of patients' own cell features in the modeling, indicating the practical strategy for clinical usage. Consequently, our results provide strong evidence for the feasibility of using a quantitative time series of phase-contrast cellular morphology for non-invasive cell quality prediction in regenerative medicine.

Bone augmentation via tissue engineering has generated significant interest. We hypothesized that periosteum-derived cells could be used in place of bone marrow stromal cells (which are widely used) in bone engineering, but the differences in osteogenic potential between these 2 cell types are unclear. Here, we compared the osteogenic potential of these cells, and investigated the optimal osteoinductive conditions for periosteum-derived cells. Both cell types were induced, via bFGF and BMP-2, to differentiate into osteoblasts. Periosteal cells proliferated faster than marrow stromal cells, and osteogenic markers indicated that bone marrow stromal cells were more osteogenic than periosteal cells. However, pre-treatment with bFGF made periosteal cells more sensitive to BMP-2 and more osteogenic. Transplants of periosteal cells treated with BMP-2 after pre-treatment with bFGF formed more new bone than did marrow stromal cells. Analysis of these data suggests that combined treatment with bFGF and BMP-2 can make periosteum a highly useful source of bone regeneration.