Artificial neural network modeling of crosslinked elastin -like polypeptide formulations parameters effects on physical properties and cellular interactions for cartilage tissue engineering
Tissue engineering and regenerative medicine efforts for articular cartilage have investigated multiple materials to serve as scaffolding for guiding cartilage regeneration. Materials have included those that are prefabricated or those that may be injected and have been formed from a variety of natural and synthetic materials. Several studies have demonstrated that material parameters affect mechanical properties of these scaffolds as well as their ability to support chondrogenesis by encapsulated or seeded cells, but none have simultaneously investigated the effects of perturbations of multiple formulation parameters on competing outcome measures, including physical and biological outcomes. Therefore, this dissertation sought to investigate the effects of formulation parameters of one novel material, crosslinked elastin-like polypeptides (ELPs), on multiple and diverse outcomes measures and evaluate the ability of self-organizing maps to recognize patterns in outcomes data in order to define relationships between them and ELP formulation parameters.
Quantitative measures of cell encapsulation and viability, matrix accumulation, and metabolite consumption and evolution were used to assess the effects of ELP formulation parameters on biosynthesis for encapsulated cells. ELPs of high molecular weight and high crosslink densities were found to support the highest levels of cell encapsulation, while all but three formulations experienced a decline in cell number during culture. In contrast, ELPs with low concentrations and moderate crosslink densities were found to support higher levels of matrix synthesis and ideal ratios of metabolite consumption and evolution.
The use of early measurements of media metabolites was evaluated as a means of predicting longer-term matrix accumulation in a non-destructive manner. It was found that lactate concentration at day 7 is strong predictors of both sulfated glycosaminoglycans (sGAG) and hydroxyproline (OHP). Further evidence for using this technique was evidenced by formulations demonstrating ideal ratios of metabolite consumption and evolution also resulting in long-term accumulation of matrix components with ratios near those found in native tissue.
A semi-quantitative histological grading scheme was used to evaluate matrix and cell distribution visually. In addition, principal components analysis was used to determine the relative contribution of each scoring category to the overall histological outcome. This analysis revealed that scores for safranin-o and matrix (collagen accumulation) contributed more variability to the data than scores describing cellularity and distribution, which suggested that this type of grading scheme should be subjected to a multivariate method such as principal components in order to "rank" scoring categories, rather than simply summing scores to arrive at a cumulative score.
The effect of ELP formulation parameters on the mechanical properties of crosslinked gels was also evaluated. ELPs with very high crosslink densities and high concentrations resulted in gels with higher mechanical properties that exhibited more elastic behavior compared to other formulations.
In general relationships between ELP formulation parameters, mechanical, and biological outcomes could not be determined using ANOVA with post-hoc testing beyond broad effects, which suggested the need for more sophisticated mathematical tools for defining these relationships. To that end, experimental results from all outcomes measures were used to investigate the utility of a self-organizing map (SOM) network for this purpose. SOM demonstrated the ability to recognize patterns in experimental data and provided a visual representation of relationships between ELP formulation parameters and measured outcomes. Mapping showed that mechanical properties resulted in the best mean separation amongst neurons and showed that crosslink density was the strongest predictor of most outcomes, followed by starting ELP concentration. The map also grouped formulations together that resulted in the highest values for matrix production, greatest changes in metabolite concentrations, and highest histological score, indicating that the map was able to recognize patterns in experimental data amongst diverse variables. In addition, mapping parameters were defined for this dataset that resulted in the map with the highest overall quality, statistically separating means amongst neurons. These results point toward the utility of this mathematical tool for defining relationships amongst formulation parameters and diverse outcomes measures.