Location matters: highly divergent protein levels in samples from different CNS compartments in a clinical trial of rituximab for progressive MS

The study of in vivo processes and pathophysiological mechanisms in the central nervous system (CNS) of humans poses many challenges. Brain biopsies are not readily available and, where possible, tissue samples processed ex vivo will inevitably suffer from artefacts due to the rapid loss of viability. Imaging studies are limited for translating structure to function, particularly on the cellular or subcellular level. In vitro models still have many challenges to overcome [1].

One of the most common methods to study in vivo processes in the CNS is the biochemical analysis of cerebrospinal fluid (CSF). The concept of the “third circulation” suggesting that CSF flows through the ventricles, cisterns and subarachnoid space and is reabsorbed into the blood at the arachnoid villi, was introduced by Cushing already in 1925 [2]. It has been assumed that CSF samples obtained by lumbar puncture are related to pathophysiological processes in the CNS. This has been utilized for diagnostic purposes as well as for markers of disease progression, e.g. oligoclonal bands in support of the diagnosis of multiple sclerosis and neurofilament-light as an indicator of axonal damage in various neurological diseases [3]. However, to which extent lumbar CSF accurately reflects ongoing processes in the brain is presently unknown.

In an open label, phase 1b study on intraventricular administration of rituximab for progressive multiple sclerosis (PMS), the study design included measurements of protein levels in three different CNS compartments: the interstitial space (ISS) of the brain, the right lateral ventricle and the lumbar CSF space. This enabled us to determine to what degree the composition of proteins was similar between compartments.

In this study we compared the levels of 180 proteins in three different CNS compartments. Our main finding was that there was a strikingly poor correlation between the protein levels of all compartments, for the most part not more than expected by chance. Consequently, a model based on a gradual shift of proteins from the ISS fluid to the ventricles and further on into the lumbar cistern gained no support. On the other hand, we demonstrated a moderate correlation between the protein levels in samples obtained from the same location but at different timepoints, suggesting some degree of individual stability over time in each compartment.

According to mainstream theory, CSF is chiefly formed by the choroid plexus, based upon the historical canine experiments of Dandy [8]. Then, CSF moves through a unidirectional flow through the ventricles, cisterns and subarachnoid space and is finally reabsorbed into the blood at the site of the arachnoid villi [2]. This theory is likely an oversimplification and there are good reasons to believe that CSF circulation is much more complex: a combination of directed bulk flow, pulsatile movements, and continuous bi-directional fluid exchange at the blood brain barrier and the cell membranes at the borders between CSF and ISS spaces probably all contribute to the movement and composition of CSF [9]. Furthermore, the recent discoveries of the glymphatic system and meningeal lymphatic vessels [10‐13] have provided novel insights on the clearance of proteins and metabolites. Our data are more in line with this complex model and highlight the importance of not extrapolating data obtained from one compartment in the CNS to another.

As a first objective we wanted to establish whether the protein composition of the ISS was similar to CSF from the right lateral ventricle. Most of the proteins in our panels could be detected by the PEA, but the Spearman’s rho was centered around zero, with an equal distribution on both sides, suggesting that there was no meaningful correlation between these two compartments. Also, the distribution of the levels of the proteins was fairly equal with 36% of proteins being higher in the ISS and 64% higher in ventricular CSF samples, again suggesting a random effect. However, the results are likely to be influenced by the pore size (100 kDa) of the semipermeable membrane of the MD catheter, causing a reduced recovery of proteins starting at 40–50 kDa in size. Recovery by MD also depends on the individual characteristics of the analyzed proteins, which influence physical passage through the membrane as well as adhesion to the membrane and tubes. Absolute and relative recovery can be enhanced by adding dextran to the MD fluid and the flow rate was chosen in order to optimize the absolute recovery volume in order to facilitate the laboratory analyses [14, 15]. Adhesion of the proteins to the tubes is highest in the first 5–10 min after the system is set up. After that the inner lining of the tubes will be saturated with proteins and recovery will stabilize [15]. Since the MD fluid analyzed in this study was collected during 6 h we believe that the issue of loss of recovery due to adhesion is of minor importance.

Most larger proteins analyzed in ISS will probably have a recovery of < 40%, suggesting that the actual concentrations of proteins in ISS are higher than observed here, which may have led to an overestimation of the difference in protein levels. However, this is less likely to influence the over-representation of proteins from the immunology panel in the ISS, since they have a low molecular weight with a higher expected recovery. Despite the issue of recovery, a number of proteins showed a higher level in ISS compared to ventricular CSF, which strongly suggests a true difference for those particular proteins.

When using dextran, there is always a possibility of dextran leakage from the catheter to the extracellular space, which may influence the local microenvironment around the catheter. For Dextran-70 a histological inflammation reaction has been demonstrated but not for Dextran-500 [16]. Unfortunately, no analysis of the composition of proteins in the extracellular space was done in that study. To our knowledge, no such study has been performed with Dextran-60. Due to the inherent nature of MD, great precaution has to be taken when discussing the absolute levels of proteins obtained by MD. Nevertheless, MD is currently the best method to obtain samples from extracellular space allowing non-targeted mass analysis of protein spectra. Despite the limitations of the technique we believe that our findings lead to the conclusion that it is unlikely that the ISS of the brain directly communicates with the CSF of the ventricles. A more reasonable conclusion is perhaps that these proteins are cleared to the perivenous space and then drained out of brain into the cervical lymphatic system [17, 18].

As a second objective, we wanted to establish whether the protein composition of the CSF from the right lateral ventricle was similar to the CSF in the lumbar subarachnoid space. It has been demonstrated that lumbar injection of radionuclides [19] as well as gadolinium contrast agents [20] do not enter the ventricular system under normal conditions, which supports a caudally directed flow of CSF from the ventricles to the lumbar region. A majority of the proteins in our study displayed higher levels in the lumbar cistern. Furthermore, our data suggested a very poor correlation between the protein levels in the two compartments. These data highlight the complex relationship between different subcompartments of the total CSF compartment including the various ways proteins enter CSF through the blood-brain-barrier (BBB), blood-spinal cord barrier (BSCB) and blood-CSF-barrier (BCB) [21, 22]. For immunological substances, an explanation may be that several of these have their origins in the leptomeninges. Proteins synthesized in the leptomeninges have indeed been shown to be highly enriched in the lumbar compartment compared with the ventricles [23]. For proteins in the neurology panel, the finding of substantially higher concentrations in the lumbar compared with ventricular CSF is more intriguing since the opposite relation has been reported for brain cell-derived structural proteins [23]. One possible explanation to this observation could be that some of the proteins in the neurology panel were more abundantly expressed in the spinal cord, from which the contribution of specific protein content to the lumbar CSF is likely to be greater than from the brain. In addition, the overall rostral—caudal flow of CSF with only around 20% of the total volume enclosed in the ventricles could lead to a slower bulk flow with an accumulation of content in the larger lumbar cistern.

The strongest correlation between compartments in this study was between proteins obtained from the same (ventricular) compartment but at different timepoints. In lumbar CSF, the most common proteins have been extensively studied and rather narrow reference ranges have been established [24, 25], indicating the same type of stability over time within this sub-compartment as well. These observations are further reinforced by a recent study, in which it was shown that the intraindividual variation of the proteome for lumbar CSF was considerably more stable over time than the interindividual variations between different individuals [26]. In addition, previous investigators have demonstrated a small but consistent concentration gradient within the lumbar compartment when CSF was withdrawn in 4 mL portions up to 24 mL CSF [27]. Altogether, these data are consistent with a CSF compartment that can be further divided into at least two sub-compartments (i.e. lateral ventricles and lumbar sac), each with a rather stable composition but a complex and variable exchange of specific proteins.

Analyses of different proteins in body fluids can be done for several purposes where the specific purposes are linked to specific and sometimes different requirements. To establish a biomarker, the protein needs to be statistically associated with a clinical outcome, but knowledge of the actual role of the protein in disease is strictly not necessary. On the other hand, in order to study a pathophysiological disease process, the sampling site needs to some extent mirror the anatomical site where the pathological process occurs, in order to draw valid conclusions from observations. Up to 80% of the proteins in the CSF are believed to originate from the blood, leaving only a fraction of CNS-specific content mirroring disease processes within the CNS [23]. An obstacle to the research of human CNS disorders is the difficulty in obtaining representative tissue samples from living patients with non-malignant disease. Cerebrospinal fluid collected at lumbar puncture has often been regarded as a window to the CNS. In some diseases, CSF findings have contributed to the unfolding of pathological processes in the brain and spinal cord, one example being the dementia field [28]. In those neurodegenerative disorders, the CSF proteins that panned out to be of significance were all structural proteins of the CNS, similar to those in the neurology panel of our study. The most important biomarker of structural damage to the CNS in MS is neurofilament light protein (NFL), which was not a part of the neurology panel used here. NFL is an extraordinarily stable protein [29], which slowly diffuses from the site of injury to the CSF compartment [3]. NFL is believed to be accumulated in the lumbar sac, increasing the sensitivity of this biomarker when obtained at this location [23]. The opposite is to be expected when studying paracrine cytokines or chemokines, which are locally produced in the tissue. These are likely to degrade over time and distance from the site of production. Proteins from the immunology panel in this study were more abundant in the ISS compared with ventricular CSF while the opposite was the case between ventricular and lumbar CSF indicating a complex relationship between the different CNS compartments for these substances (Fig. 5, Additional file 2: Table S1).

Our study had several weaknesses, the major one being the small sample size (n = 10 for paired MD/ventricular-CSF and n = 13 for paired ventricular-CSF/lumbar-CSF samples). The study was planned as an exploratory part of a clinical trial with primary endpoint safety and feasibility, which by its phase 1b nature was limited in terms of patient numbers [4]. The invasive neurosurgical procedures used in the study put a clear limit on patient recruitment. The sampling itself could potentially influence the pressure dynamics in the CSF compartments, which might affect synthesis and/or distribution of some brain-derived proteins. The most common adverse event related to intraventricular injection was a short-lasting vertigo, which we believe was created by a disturbance of pressure equilibrium [4]. However, since collection of samples were done before injection, we believe it is unlikely that this affected the protein concentrations in the samples used for this study.

We decided to adopt a broad approach to the analysis of constituents in the different CNS compartments. The large number of tested proteins in relation to the small number of patients obviously prevents reliable conclusions regarding specific proteins and their role in PMS, although further hypotheses on specific proteins could be formulated and tested in future studies. P-values from a wide explanatory analysis, such as from the present study, cannot be interpreted in the same way as for a confirmatory analysis. Corrections for multiple comparisons to control for family wise error or false discovery rate was deemed not meaningful due to the loss of power it would imply. However, it should be noted that the expected number of type 1-errors when testing 180 proteins is 9, assuming no real differences exist and that we obtained more statistically significant differences in all performed compartment comparisons. Another weakness is that we were not able to make any comparison between protein levels in the ISS and lumbar CSF from samples obtained at the very same timepoint, intuitively perhaps the most interesting. However, the design of the study, with a baseline lumbar puncture performed at a timepoint preceding the surgical procedure with a substantial variability in time, would have made a comparison difficult to interpret. In contrast, the other comparisons between different compartments were done on the samples obtained more or less simultaneously (at a most within a few hours) in the compared compartments. However, the overall lack of correlation between the other investigated compartments in this study makes it unlikely that protein concentrations obtained in lumbar CSF would reliably reflect the composition of proteins pertaining to focal pathology in the brain tissue. This should be confirmed in future studies, by obtaining simultaneous samples from ISS and lumbar CSF and analyze the relationship between key inflammatory substances in these compartments.

The results from this study indicate a substantial heterogeneity regarding the proteins within different compartments of the CNS. Therefore, great care should be taken when analyzing data obtained from analysis of lumbar CSF samples in relation to the underlying pathological processes in the CNS tissue. This seems to be particularly pertinent to the interpretation of levels and patterns of inflammatory substances.