Comparative Analysis of the Euroimmun CXCL13 Enzyme-Linked Immunosorbent Assay and the ReaScan Lateral Flow Immunoassay for Diagnosis of Lyme Neuroborreliosis

DISCUSSION

CXCL13 CSF levels rise early in the course of LNB and decline during treatment (1). Borreliae that enter the CNS are recognized by monocytes, macrophages, or dendritic cells, which produce proinflammatory cytokines; in addition, several chemokines are released to attract other immune cells (21). CXCL13 is produced by antigen-presenting cells (22). It belongs to the CXC chemokine family and is a selective chemoattractant for B lymphocytes and follicular B helper T cells (also known as follicular T helper cells) via the chemokine receptor CXCR5 (23, 24). CXCL13 is involved in the development and maintenance of secondary lymphoid organs and is also produced during inflammation in nonlymphoid tissues, such as the CNS (25).

In this study, we showed that CXCL13 ELISA values are significantly elevated in patients with definite LNB (median, 1,409 pg/ml; range, 91.4 to 39,370 pg/ml) compared to control patients (median, 20.7 pg/ml; range, 4.5 to 14,645 pg/ml). With an optimal cutoff value of 78.6 pg/ml, the CXCL13 ELISA had a sensitivity of 100% and a specificity of 84.8%. Our data confirm previous studies that also reported highly elevated CXCL13 levels in LNB. In 2018, Rupprecht et al. performed a meta-analysis on 18 studies involving 618 patients with acute LNB and 2,326 control patients with other neurological diseases. The pooled sensitivity and specificity for CSF CXCL13 with an optimal cutoff value of 162 pg/ml were 89% and 96%, respectively (7). In this meta-analysis, the cutoff values determined by the included studies ranged from 18 to 1,229 pg/ml. One possible explanation for this strong heterogeneity could be the utilization of different CXCL13 assays. However, only two assays were used, namely, the R&D Systems ELISA (R&D Systems Europe, Abingdon, United Kingdom) in 15 studies and the Euroimmun ELISA (Euroimmun AG, Lübeck, Germany) in 3 studies. The cutoff values determined by the three studies using the same assay we did (Euroimmun ELISA) were 55 pg/ml, 93.8 pg/ml, and 200 pg/ml (18, 26, 27). The first study by Remy et al. retrospectively examined 53 children with LNB and 91 controls and found an optimal cutoff level of 55 pg/ml with a sensitivity of 96.7% and a specificity of 98.1%. The median CXCL13 concentration in the CSF of children with definite LNB was 774.7 pg/ml, and values ranged from 58.9 to 13,487 pg/ml (26). Wagner et al. reported a slightly higher optimal cutoff value of 93.8 pg/ml, leading to a sensitivity of 95% and a specificity of 97% in an unselected cohort of 459 patients, 20 of them having LNB. The median CXCL13 concentration in the CSF of LNB patients was 900 pg/ml, and values ranged from 10 to 6,500 pg/ml (18). Finally, Waiß et al. analyzed 10 patients with LNB and 53 controls and determined a sensitivity of 100% and a specificity of 92.4% using a literature-based cutoff value of 200 pg/ml. The median CXCL13 concentration in the CSF of LNB patients was 525.5 pg/ml, and values ranged from 254 pg/ml to >900 pg/ml (27). A direct comparison of these results to ours is difficult because there are significant differences in terms of study population and design. Remy et al., for example, examined only children and adolescents, and the median age was 8 years (26). In contrast, the median age of our patients was 55 years. An analysis of the literature on CXCL13 and the corresponding cutoff levels as a function of age was inconclusive. Therefore, it is currently unclear whether age accounts for the differences between the studies.

In contrast to our case-control design, Wagner et al. performed a cross-sectional study (18). According to the above-mentioned meta-analysis, the cutoff levels of cross-sectional studies were lower than those of case-control studies; thus, the different design cannot explain the lower cutoff value found by us (7).

Another difference between the three studies mentioned above and ours is the lower specificity of 84.8% found by us. Responsible for this are six false-positive patients with VZV encephalitis (n = 3; 30% of all VZV encephalitis patients) and bacterial meningitis (n = 3; 37.5% of all patients with bacterial meningitis). While the former showed moderately elevated CXCL13 levels (mean, 288 pg/ml), the CXCL13 levels of the latter were strongly increased (mean, 8,373 pg/ml). The causative bacterial pathogens were C. fetus and H. influenzae in two patients with purulent meningitis and S. intermedius in a patient with bacterial brain abscess caused by poor dental care. Previous studies also reported elevated CXCL13 CSF levels in VZV encephalitis (25, 26). From five of these samples, CSF cell counts were reported; all the patients had CSF pleocytosis. Consequently, CSF pleocytosis cannot aid as an additional discriminator between LNB and VZV encephalitis or bacterial meningitis. To judge the impact of these positive results on the CXCL13-assisted diagnosis of LNB, one has to take into consideration the most common CNS pathogens, e.g., S. pneumoniae, tick-borne encephalitis virus, enterovirus, VZV, herpes simplex virus, and Neisseria meningitidis (28). Only VZV seems to pose a frequent problem. However, VZV is efficiently detected in CSF using nucleic acid amplification techniques. Therefore, VZV encephalitis in patients with elevated CSF CXCL13 can easily be ruled out. Other, less common CNS disorders are more difficult to diagnose. Among these, neurosyphilis is probably the most challenging disease entity, because cross-reactivity can lead to weakly positive results in anti-Borrelia ELISAs and neurosyphilis is known to cause elevated CXCL13 levels (29, 30). Therefore, it is advisable to perform an additional serologic test to rule out Treponema pallidum infection if LNB is suspected and CXCL13 levels are elevated.

At this point, it must be emphasized that CXCL13 is a nonspecific marker for CNS inflammation and is not unique to CNS infection. CXCL13 levels are reported to be elevated in patients with various inflammatory CNS disorders (e.g., clinically isolated syndrome, multiple sclerosis, N-methyl-d-aspartate receptor encephalitis, or Rasmussen encephalitis) (31, 32). Although 60% of these noninfectious inflammatory CNS disorders do not cause CNS pleocytosis, they represent possible differential diagnoses for patients with suspected LNB (32). Consequently, the specificity of CXCL13 for the diagnosis of LNB is strongly dependent on the patient population, i.e., the clinical setting in which the test is used. A high number of patients with noninfectious inflammatory CNS disorders in the case mix would decrease the specificity of the CXCL13 assay, especially in the absence of CSF pleocytosis.

Two of the study patients were particularly interesting because they had positive AIs and lacked CSF pleocytosis. Their CXCL13 results were negative, strongly arguing against LNB. Thus, it may be assumed that antibiotic treatment would have been without effect. Perhaps nonspecific polyclonal activation was the reason for the elevated B. burgdorferi AIs.

One major outcome of our study is the careful comparative assessment of the performances of the CXCL13 ELISA and the CXCL13 LFA. While ELISA-based antigen tests usually show high sensitivity, they suffer from some drawbacks. Factors like a long assay turnaround time of 1 to 2 h; the need for appropriately equipped facilities and trained personnel; and last, but not least, low test frequencies for economic reasons restrict their immediate diagnostic impact. Immunochromatographic assays, like the CXCL13 LFA, may overcome these limitations. They are designed as fast point-of-care tests that can be performed on a single-patient basis without complicated laboratory equipment. In our hands, the CXCL13 LFA proved to be a robust test without invalid test results. The hands-on time and the time to result per test were 1 min and 20 min, respectively. The CXCL13 LFA enables instant testing around the clock and on weekends without the need for experienced personnel. For the LFA, we calculated an optimal cutoff AV of 22.5, leading to a sensitivity of 91.2% and a specificity of 93.5%. The agreement analysis of our CXCL13 ELISA and LFA results showed perfect agreement (100%) for values under 78.6 pg/ml and high agreement (80%) for values above 78.6 pg/ml. Overall, the correlation between the CXCL13 levels determined by the ELISA and the LFA was strong (r = 0.89; P < 0.0001) and in line with the results determined by Pietikäinen et al. (19). So far, the work by Pietikäinen et al. is the only study to compare a CXCL13 LFA and a CXCL13 ELISA. While we both used the same LFA, the CXCL13 ELISA was different (human CXCL13/BLC/BCA-1 Quantikine [R&D Systems, Minneapolis, MN, USA] versus CXCL13 ELISA [Euroimmun AG, Lübeck, Germany]). However, the main difference from the study by Pietikäinen et al. is that we calculated optimal cutoff values by ROC analysis for the CXCL13 ELISA, as well as for the CXCL13 LFA. In particular, the cutoff value for the CXCL13 LFA was extremely helpful, because the manufacturer does not provide one.

Our study has some limitations. First, it was a retrospective case-control study, and we could not retrieve data on antibiotic therapy before lumbar puncture for all patients. Because it is known that CSF CXCL13 decreases rapidly during antibiotic therapy (12), early antibiotic treatment could have diminished the CXCL13 values in our study. Second, the CSF samples were stored at −20°C before performing the CXCL13 assays. Freeze-thaw cycles might have caused degradation of CXCL13, thereby leading to decreased concentrations. However, Hytönen et al. demonstrated that up to five freeze-thaw cycles did not significantly affect the levels of CXCL13 (25). Our samples did not undergo more than three freeze-thaw cycles (see Materials and Methods), so it may be assumed that our samples were of high quality. Third, the numbers of non-LNB patients included in our study were small for several subgroups of diseases (e.g., CNS malignancies).

A question that remains to be answered is how CXCL13 measurement is optimally used in daily routine. Is it necessary, or will antibiotic therapy be started anyway in case of a typical clinical picture and CSF pleocytosis, irrespective of the CXCL13 result? In our experience, which is solely based on European LNB, it is not possible to diagnose LNB from the clinical picture alone. Furthermore, CSF pleocytosis is also present in other CNS infections and inflammatory disorders. With the CXCL13 LFA, a bedside test is now available that provides a result within 20 min. In case of a negative CXCL13 test, the diagnosis of LNB is unlikely, and in combination with a normal CSF cell count, it is virtually ruled out. In addition, the amount of money saved by avoiding ceftriaxone therapy until receipt of the antibody results after 3 days is sufficient to perform approximately 6 CXCL13 LFAs. Thus, the extra financial burden to the health care system of additional CXCL13 testing is limited and serves to spare antibiotics in times in which we face increasing numbers of multidrug-resistant bacteria worldwide. To include the CXCL13 LFA in the workflow of LNB diagnosis, we suggest the following algorithm for patients with a clinical picture compatible with LNB. Initially, CSF samples are analyzed for leukocyte counts and CXCL13 levels. In cases of pleocytosis and strongly increased CXCL13 values, antibiotic therapy should be initiated due to a high probability of LNB. If CXCL13 levels are negative, therapy for LNB may be withheld, irrespective of the CSF cell count, and alternative diagnoses should be considered. In any case, the CSF/serum Borrelia AI should be determined in parallel.