Table 1.
Key features of the selected MS biomarkers.
Table 2.
Demographic data of the patients with MS, CIS and controls.
Table 3.
Case breakdown of the model building and model testing data sets for all individuals included.
Table 4.
Case breakdown of the model building and model testing data sets within MS cohort included.
Table 5.
Correlation matrix for variables included in the modeling.
Figure 1.
PCA broadly distinguishes between inflammatory and non-inflammatory conditions.
PCA scores plot for the model building dataset (Set 1). Most non-inflammatory cases (blue) are associated with positive scores, located in the upper part of the plot. There is more variation in the distribution of the inflammatory cases (yellow) but many associate with negative scores (A). The PCA loadings plot shows that all the measured variables except age underlie the observed scores distribution seen in A (B). Scores of the test set (Set 2) were predicted using the model derived for Set 1. Non-inflammatory conditions, again, cluster mostly in the upper part of the plot (C). Extreme outliers (≥3 SD) in both scores plots belonged to cases of herpes encephalitis and neuroborreliosis.
Figure 2.
The inflammatory signature for MS is associated with patient age.
Scores and loadings respectively for the PCA model of all MS patients with complete ELISA data (Set 3). The loadings show that higher age is associated with lower levels of all inflammatory and axonal injury markers in the CSF. EDSS makes no significant contribution to the model (A and B). The model derived for Set 3 was used to predict the scores for the test set (Set 4). The scores plot shows a similar distribution to that in 2A (C). PCA modelling was repeated for Set 3 samples excluding the age variable (D and E); SUS-style plots plot the scores (D) and loadings (E) from the two models against each other. The scores and loadings from the 2 models are highly correlated signifying that the age variable in itself is not artificially driving the scores distribution seen in 2A. Scores are coloured according to age group: ≥54 years (blue circles), ≤53 years (yellow circles). Loadings are coloured green.
Figure 3.
PCA of relapsing-remitting and progressive MS subgroups.
Scores and loadings respectively for the PCA model of RRMS patients only with complete ELISA data (Set 3 RRMS patients) (A and B). Scores and loadings respectively for the PCA model of progressive patients only with complete ELISA data (Set 3 PMS patients) (C and D). Scores are coloured according to age group: ≥54 years (blue circles), ≤53 years (yellow circles).
Figure 4.
OPLS analysis separates out the variability associated with individual y response variables.
SUS plots compare the loadings (p(corr)) from four different OPLS models for all MS patients with complete ELISA data (Set 3) where each of the ELISAs was treated as the y response variable in turn (A–D). SUS plot methodology was also used to compare the corresponding scores (tcv[1]). The strongest correlation between the scores was seen for MMP9 and CXCL13 (E; R2 = 0.7771). Individuals older than 54 years expressed low levels of both markers whereas more variable expression was evident in the younger age group (E–H). Weaker correlations are seen between the scores for CXCL13 and OPN, CXCL13 and NFL, and NFL and OPN (F–H). Scores are coloured according to age group: ≥54 years (blue circles), ≤53 years (yellow circles). Loadings are coloured green.