Study of the relationship between choroid plexus volume and CSF derived biomarkers in patients with multiple sclerosis

Choroid Plexus (ChP) consists of highly vascularized veil-like organoids located in the brain ventricles and its primary function is the production of the Cerebrospinal Fluid (CSF). ChP structural and functional alterations may contribute to the pathophysiology of neurodegenerative diseases such as Multiple Sclerosis (MS). It can be imaged noninvasively with Magnetic Resonance Imaging (MRI), and recent studies highlighted that ChP is altered and particularly, it is enlarged in MS and other disorders, suggesting its use as a proxy of brain neuroinflammation and raising interest on the ChP volume (ChPV) as potential biomarker of neuroinflammation. The goal of this study is to investigate whether ChP alterations, measured by the ChPV can be related to biomarkers (cytokines and chemokines) obtained from CSF analysis in patients with MS. The final aim is a better understanding of underlying processes of MS and the potential use of ChPV as new biomarker for faster and non-invasive disease monitoring. To reach this objective, the ChP of 167 progressive MS subjects from two acquisition sites has been segmented through ASCHOPLEX, a new approach for ChP automated segmentation. Initially, ASCHOPLEX output was manually validated by a quality check procedure, that highlighted the partial erroneous inclusion of the hippocampus boundaries in the ChP segmented mask in some subjects. This aspect has been deepened through the study on a standardized space (MNI) for comparing the position of ChP segmentations in the brain among the entire cohort. The inclusion of the hippocampus has been corrected by excluding some voxels on the original segmentations through a thresholding operation in MNI space. ChPV was then divided by the Total Intracranial Volume (TIV) to remove any residual brain-size dependance and finally NeuroCombat approach was used to harmonize ChPV by removing unwanted scan variability. An additional dataset of 130 CSF molecules derived from the same subjects has also been provided for this study. The processing of these data consisted in a preliminary statistical study to have an overview of distribution, missing values, and collinearity between molecules. Based on this analysis, the number of collected variables has been reduced, considering both the dataset coverage and the presence of missing values A subset of selected molecules was investigated in detail because of previous evidence of a relationship with MS, and they have been considered a hypothesis-driven selection. NChPV and CSF molecules relationship has been studied using linear regression. To apply it, a previous selection of relevant variables has been performed with Lasso and Stepwise regression, obtaining two sets of data-driven selections. Lasso Regression reaches an adjusted R2=0.23 when using 8 variables derived from the 54 molecules and applied to the entire cohort; Stepwise regression instead, reports an adjusted R2=0.46 when applied on a reduced cohort of subjects and variables without missing values, and R2=0.30 when applied to the entire cohort. Even if the performance of linear models employed in this study are not optimal, some variables show a significant predictive role (i.e. TNF, CX3CL1, IFNg and CCL11). These molecules have been previously highlighted to be related to inflammatory aspect within the CNS of MS patients confirming that the alteration of the CSF biological profile might be related to the alteration of ChP. The main finding of this work is that the molecules derived from data-driven selection give a significant contribution in predicting the NChPV and are consistent with those of the hypothesis-driven selection. Particularly, they are all related to inflammation processes or to immune cells recruiting. To conclude, this study might lay the foundation to a better understanding of the relationship between inflammatory mechanisms in MS and ChPV, showing the promising usefulness of ChPV as a new non-invasive neuroimaging biomarker.