Potential role of CSF cytokine profiles in discriminating infectious from non-infectious CNS disorders

Abstract

Current laboratory testing of cerebrospinal fluid (CSF) does not consistently discriminate between different central nervous system (CNS) disease states. Rapidly distinguishing CNS infections from other brain and spinal cord disorders that share a similar clinical presentation is critical. New approaches focusing on aspects of disease biology, such as immune response profiles that can have stimulus-specific attributes, may be helpful. We undertook this preliminary proof-of-concept study using multiplex ELISA to measure CSF cytokine levels in various CNS disorders (infections, autoimmune/demyelinating diseases, lymphomas, and gliomas) to determine the potential utility of cytokine patterns in differentiating CNS infections from other CNS diseases. Both agglomerative hierarchical clustering and mixture discriminant analyses revealed grouping of CNS disease types based on cytokine expression. To further investigate the ability of CSF cytokine levels to distinguish various CNS disease states, non-parametric statistical analysis was performed. Mann-Whitney test analysis demonstrated that CNS infections are characterized by significantly higher CSF lP-10/CXCL10 levels than the pooled non-infectious CNS disorders (p = 0.0001). Within the infection group, elevated levels of MDC/CCL22 distinguished non-viral from viral infections (p = 0.0048). Each disease group of the non-infectious CNS disorders independently showed IP-10/CXCL10 levels that are significantly lower than the infection group [(autoimmune /demyelinating disorders (p = 0.0005), lymphomas (p = 0.0487), gliomas (p = 0.0294), and controls (p = 0.0001)]. Additionally, of the non-infectious diseases, gliomas can be distinguished from lymphomas by higher levels of GRO/CXCL1 (p = 0.0476), IL-7 (p = 0.0119), and IL-8 (p = 0.0460). Gliomas can also be distinguished from autoimmune/demyelinating disorders by higher levels of GRO/CXCL1 (p = 0.0044), IL-7 (p = 0.0035) and IL-8 (p = 0.0176). Elevated CSF levels of PDGF-AA distinguish lymphomas from autoimmune/demyelinating cases (p = 0.0130). Interrogation of the above comparisons using receiver operator characteristic analysis demonstrated area under the curve (AUC) values (ranging from 0.8636-1.0) that signify good to excellent utility as potential diagnostic discriminators. In conclusion, our work indicates that upon formal validation, measurement of CSF cytokine levels may have clinical utility in both identifying a CNS disorder as infectious in etiology and, furthermore, in distinguishing viral from non-viral CNS infections.

Fig 1. Heat map and dendrogram based on agglomerative hierarchical clustering (AHC).

(A) The heat map visually depicts the cytokine levels using colors (red = low, green = high) shown on the right. The heat map was created in conjunction with agglomerative hierarchical clustering (AHC). The three classes shown in the AHC are reflected here. Vertical, yellow lines have been drawn to aid in this separation of classes. C = control; L = CNS B-cell lymphoma; IF = infection, T = tumor (high grade gliomas), A = autoimmune; MS = multiple sclerosis. (B) The AHC based dendrogram includes all 43 cases and shows separation of CNS diseases into three major classes demonstrating a trend toward clustering of disease types based on overall cytokine profiles. See supplementary table (S5 Table) for a key to the abbreviations of the disease type/class used in the heat map and dendrogram.

Fig 2. Discriminant analysis observation plot.

According to discriminant analysis, 100% of our 43 cases were assigned appropriately to their respective disease groups (infection, autoimmune, tumor, DM, control).

Fig 3

(A-I): Results from mann-whitney analyses for specific cytokines (IP-10/CXCL10, MDC/CCL22, IL-7, IL-8, GRO/CXCL1, VEGF, and PDGF-AA). Each graph shows the cytokine level distribution for the respective disease groups. In each graph, horizontal lines with an asterisk (*) indicate the presence of statistically significant differences between groups for the given cytokine. The number of asterisks corresponds to the calculated p-value (* = p < 0.05, ** = p < 0.01, *** = p < 0.005, **** = p < 0.001).

Fig 4

(A-F): ROC curves for IP-10/CXCL10 (A), MDC/CCL22 (B), IL-7 (C), IL-8 (D), GRO/CXCL1 (E), and PDGF-AA (F). The title of each graph includes the groups that were compared, as well as the respective AUC.

Fig 5. Proposed algorithm for diagnosis of CNS diseases using a selective cytokine panel.

The cytokines included in the algorithm include IP-10/CXCL10, MDC/CCL22, IL-8, GRO/CXCL1, and PDGF-AA. Potential “cut-off” values for interpretation are presented here. Sensitivity, specificity, and likelihood ratios (if available) corresponding to the cut-off values have been generated from the ROC analyses and are also featured.