A network analysis approach using transcriptomic and phenotypic properties to identify the effects of phosphate deficiency on fracture healing
Deng, Zi Jun
MetadataShow full item record
Approximately 6.3 million fractures occur annually in the U.S. and almost 10% of these fractures fail to heal normally. These non-union fractures adversely affect the patients’ quality of life and are an economic burden, due to both treatment costs and lost or reduced employment. Much of the biological and molecular basis of fracture healing and non-unions remains poorly understood. Previous studies have shown that hypophosphatemia produces a rachitic state and diminishes the endochondral ossification of long bone regeneration. Our studies used phosphate deficiency to produce a rachitic animal model in which to study delayed fracture healing. The goal of this study is to investigate the effects hypophosphatemia have on the healing bone’s temporal mRNA expression profiles, its correlations with bone phenotypes, and the effects diet and genetic strain have on significant bone-healing genes. Three strains — A/J, C57BL/6J, C3H/HeJ (AJ, B6, and C3 respectively) — of skeletally mature male mice had stabilized fractures produced in the right femur. Hypophosphatemia was produced by feeding a group (Pi) of mice with a low phosphate diet starting two days before surgery until 14 days after surgery when the regular diet is reintroduced. The control group (Ctrl) was fed the regular diet throughout. At harvest time points (post-operative days; POD) 3, 5, 7, 10, 14, 18, 21, 28, and 35, RNA was extracted from the fracture callus and quantified via microarray analysis. From a different set of mice, the calluses were extracted on POD 14, 21, and 35, for phenotype measurements. Diet and strain significant genes were identified by ANOVA with 11,037 from a total of 21,187 genes. These 11,037 genes were evaluated by Weighted Gene Co-expression Network Analysis (WGCNA) to correlate transcriptomic data to phenotypic properties of the healing bone. Additionally, the genes were also analyzed using a custom polynomial clustering method in order to cluster genes together based on similar temporal expression profiles. WGCNA results showed that, out of the list of 11,037 genes, 10,620 genes (from Ctrl group) and 10,351 genes (from Pi group) respectively clustered either into a group positively correlated with or a group negatively correlated with bone structural properties. In terms of biological functions present in the two gene groups, the positively correlated group consists of immune-related functions while the negatively correlated group consists of bone, cartilage, and vasculature functions. There were a greater number of bone-healing functions in the Pi groups relative to the Ctrl groups; this finding is consistent with known accelerated bone healing in the Pi group after phosphate has been returned to the diet at POD 14. Polynomial clustering results showed specific temporal gene expression differences between strains. Some genes, such as HIF-1α, had the same temporal gene expression regardless of diet or strain and are likely to be unaffected by phosphate deficiency while also being conserved across genetic strains. Other genes, such as IHH, had AJ temporal gene expressions, distinct from the B6 and C3, displaying a diminished peak with no compensation after phosphate recovery. Lastly, few genes, such as BMP2 and RUNX2, showed no Pi diet effects in AJ and C3 but were affected by the Pi diet in B6. Genes such as these may be attributable to previous findings where B6 bones are structurally less mineralized compared to AJ and C3 at POD 35 and beyond. These studies provided a highly detailed understanding of the temporal changes in the transcriptome in relation to both bone healing and the underlying changes at the organ level (bone phenotype). Through the analysis of specific genes’ temporal expressions, our findings further defined the role of phosphate deficiency in impaired bone repair. Future directions are to find the central hub gene, genes that are interconnected with the greatest number of other genes, for each different temporal genetic expression motif. Once identified, those temporally-clustered motifs and their central hub genes will be correlated directly with bone phenotypes in order to understand how these temporal transcriptomic profiles are associated with biomechanical and structural bone properties in the healing bone.