The molecular profile of synovial fluid changes upon joint distraction and is associated with clinical response in knee osteoarthritis

Summary Objective Surgical knee joint distraction (KJD) leads to clinical improvement in knee osteoarthritis (OA) and also apparent cartilage regeneration by magnetic resonance imaging. We investigated if alteration of the joint's mechanical environment during the 6 week period of KJD was associated with a molecular response in synovial fluid, and if any change was associated with clinical response. Method 20 individuals undergoing KJD for symptomatic radiographic knee OA had SF sampled at baseline, midpoint and endpoint of distraction (6 weeks). SF supernatants were measured by immunoassay for 10 predefined mechanosensitive molecules identified in our previous pre-clinical studies. The composite Knee injury and OA Outcome Score-4 (KOOS4) was collected at baseline, 3, 6 and 12 months. Results 13/20 (65%) were male with mean age 54°±°5yrs. All had Kellgren–Lawrence grade ≥2 knee OA. 6/10 analytes showed statistically significant change in SF over the 6 weeks distraction (activin A; TGFβ-1; MCP-1; IL-6; FGF-2; LTBP2), P < 0.05. Of these, all but activin A increased. Those achieving the minimum clinically important difference of 10 points for KOOS4 over 6 months showed greater increases in FGF-2 and TGFβ-1 than non-responders. An increase in IL-8 during the 6 weeks of KJD was associated with significantly greater improvement in KOOS4 over 12 months. Conclusion Detectable, significant molecular changes are observed in SF following KJD, that are remarkably consistent between individuals. Preliminary findings appear to suggest that increases in some molecules are associated with clinically meaningful responses. Joint distraction may provide a potential opportunity in the future to define regenerative biomarker(s) and identify pathways that drive intrinsic cartilage repair.

1.1.3. We will use the mean value of the two measures collected for each biomarker concentration at baseline. Therefore the unit of analysis will be the knee of the contributing patient. We will analyse the mean baseline value on each patient.
1.1.4. Percentage will be obtained for categorical variables in order to measure the frequency of each category. We will present the mean and its standard deviation (sd) for variables with normal distribution. For non-normal distributions we will utilise the median with the interquartile ranges (IQR) ( Table 1). We will assessed normality with graphic representation (histograms and Q-Q plots). The Q-Q plots helps us to identify if the quantiles of our continuous variable are coincident with the quantiles of the normal distribution. We will complement the analysis of normality using the Shapiro-Wilk test 4 . However, if the graphics look roughly normal we will assume their result over the test.
Analysis 2: we will use non-parametric methods based on ranks because our dataset is small (n=20). Therefore, it is obvious non-normality cannot be corrected by a suitable transformation.
2. Association among markers at baseline, 3 weeks and 6 weeks.
2.1. We will see how correlations among biomarkers are different at baseline, midpoint (3 weeks) and endpoint (6 weeks). (do file "A descriptive JDK.do" lines 261, 294, 298) (Excel file "Descriptive JDK 1.1.xlsx" spreadsheet "Correlations") ( Table 3). We will expect to see different pairwise of biomarkers and also different size of correlation in the different time points (research question 1 -RQ 1). To do so, we will work out the Spearman's rank correlation coefficients 5 . We will apply the following interpretation according to the size of the correlation 6 :   "correlation.xlsx" spreadsheet "Correlations") To do so, we will run the Spearman's rank correlation coefficients. We will exclude from this analysis comparisons on the same biomarker.

Is the absolute value of change related to the absolute value of change of other biomarkers?
We will analyse the correlation in change vs. change, (change= concentration at 6 weeksbaseline concentration) (do file "E1 spearman.do" lines 48-51) (Excel file "correlation.xlsx" spreadsheet "Correlations") To do so, we will run the Spearman's rank correlation coefficients. We will exclude from this analysis comparisons on the same biomarker.
2.5. We will check how much of the change is statistically significant at 6 weeks (endpoint) (RQ 1) ( Table 2). To do so, we will assess the equality of matched pairs of observations (same biomarker collected from the same knee at different time points) using the Wilcoxon matched-pairs signed-ranks test 7 . The null hypothesis is that both distributions are the same.
(do file "E1 spearman.do" lines 62-70, Stata command signrank ) (Excel file "Descriptive JDK 1.1.xlsx" spreadsheet "Wilcoxon Rank at endpoint"). We will use the parametric counterpart paired t-test if the distribution of the biomarker is normal. Note: An experiment run on 40 different patients is not as powerful as a before-and-after comparison using the same 20 patients (see ranksum in Stata).
2.6. We will check how much of the change is statistically significant at 3 weeks (midpoint) (RQ 1) ( Table 2). We like to see early changes in biomarker concentrations. To do so, we will assess the equality of matched pairs of observations (same biomarker collected from the same knee at different time points) using the Wilcoxon matched-pairs signed-ranks test 7 . The null hypothesis is that both distributions are the same. (do file "F signrank midpoint.do") (Excel file "Descriptive JDK 1.1.xlsx" spreadsheet "Wilcoxon Rank at midpoint"). We will use the parametric counterpart paired t-test if the distribution of the biomarker is normal.  (Table 2 already shows this information).
Association of markers at baseline regulated by FGF2 and CTGF, respectively.
3.2. Factor analysis (RQ 1). We will use this approach to find a few combinations of variables, called factors, that adequately explain the overall observed variation, and thus to reduce the complexity of the data. We will test the suitability of the data before the analysis using two test the Kaiser-Meyer-Olkin measure of sampling adequacy 8,9 and the Bartlett's test of Sphericity 10 . We will focus on the rotated factor matrix taken those values over 0.6 of each factor. Biomarkers with >0.6 in the same factor will be part of the same cluster. The factor plot in the rotated factor space will help us to see graphically the clusters. We will generate a   Table   5") (log file: "I KOOS4.log") ( Table 5) Table 5") (" Table 6.pdf").Analysis 6 (suggested in email March 5 2018): Spearman pairwise correlation of change over 12 months in KOOS4 according to baseline and 6 weeks in concentrations of analytes. (RQ 2). (do file "I KOOS4.do") (Excel file "Descriptive JDK 1.2.xlsx" spreadsheet " Table 7") (" Table 7.pdf").Analysis 8 (suggested in email March 5 2018): Simple linear regression between change over 12 months in KOOS4 and categories of change (according to baseline interquartile distribution of concentrations) over 6 weeks in concentrations