Statistical analysis

To assess the effect of treatment (sham vs. OA) and time (day) on galectin concentrations in synovial fluid, galectin ELISA data were first tested for normality using a Shapiro-Wilk W test and were found to be right skewed. Log transformation was performed to achieve normality. In order to account for the hierarchical nature of the data in the experimental models, a mixed linear model was employed because each horse was repeatedly measured on each limb and each limb repeatedly over days. The fixed effects in the model included treatment (sham vs. OA), day and a treatment^*day interaction term, and random effects included horse and individual limb nested within horse to account for the non-independence of the observations. Predefined post hoc comparisons of specific contrasts for each time point were performed to assess differences between sham and OA joints, with a Bonferroni correction applied based on the number of multiple comparisons to correct for the false discovery rate. Model diagnostics were performed and showed normality and homoscedasticity of residuals. Spearman correlation analysis was performed to determine associations between synovial fluid galectin-1 and galectin-3 levels and previously published lubricin ELISA data for the first experimental cohort and for the naturally occurring OA cases (32). Raw RT-qPCR and ELISA data from naturally occurring samples were log-transformed to achieve normality of the data, and data were analyzed using a one-way ANOVA with Dunnett's post hoc tests for multiple comparison correction, designating healthy carpal joints (OA severity = 0) as the control group. Significance was set at α = 0.05.

Immunohistochemistry scores (0 to 3 scale) were treated as ordinal categorical outcomes, and weighted kappa statistics were calculated for inter-observer agreement. Immunostaining results were assessed using Wilcoxon matched-pairs signed rank tests due to the small sample size and non-normal distribution of data. For the percentage of galectin-1 positive chondrocytes throughout the entire cartilage section, scores were treated as continuous outcomes and analyzed using Wilcoxon matched-pairs signed rank tests. All modeling and parametric analyses were performed using JMP Pro 13 software (SAS; Cary, NC), and non-parametric test statistics, kappa statistics and graphs were generated using Prism 7 (GraphPad; La Jolla, CA).