Intrathecal kappa free light chain synthesis is associated with worse prognosis in relapsing–remitting multiple sclerosis

The study population was first dichotomized according to whether patients exhibited evidence of disease activity (EDA) according to the No Evidence of Disease Activity (NEDA)-3 criteria [22]. EDA-3 was defined as the occurrence of either clinical relapses, and/or confirmed disability worsening (CDW) that was sustained for at least 6 months, and new T1 gadolinium-enhanced lesions/new/newly enlarging T2W lesions. A clinical relapse was defined as neurological signs and symptoms lasting at least 24 h and that could not be explained by another cause [18]. Patients were thus dichotomized to those who maintained NEDA-3 status during the whole follow-up period and those who demonstrated EDA-3.

Thereafter, we chose to focus on those individuals in the EDA-3 group who experienced CDW unrelated to preceding relapses (i.e., had PIRA). The main endpoint of interest in the study was hence a first PIRA event, which was defined according to previous publications [13, 14, 16]. A first PIRA event was defined as experiencing CDW in the EDSS at 6 months during a period not preceded by a relapse (at least 3 months before the PIRA event or 6 months after the first demyelinating event). The first EDSS score obtained between 6 and 12 months after the first demyelinating attack or alternatively 3 months after any other attack was referred to as the BL EDSS score and RBL EDSS score, respectively. As previously described [13], no RBL EDSS score could be lower than the first recorded (BL) EDSS score. CDW was defined as an increase in the EDSS score of 1.5, 1.0, or 0.5 if the BL/RBL EDSS score was 0, 1.0 to 5.0, or greater than 5.0, respectively. The date of PIRA was the date of CDW confirmation. According to this definition, individuals included in the study were stratified into two groups, non-PIRA and PIRA.

Paired CSF and serum samples were acquired and consecutively analysed during the routine diagnostic work-up. Concentrations of KFLC in serum and CSF were measured using the N Latex FLC kappa kit, on an Atellica NEPH 630 instrument (Siemens), following the instructions by the manufacturer. The KFLC index was calculated using the equation [(CSF KFLC/serum KFLC)/(CSF albumin/serum albumin)]. CSF and serum albumin concentrations were measured using the ALBT2 Reagent cassette on a cobas c module instrument (Roche). The CSF/serum albumin ratio was calculated as [CSF albumin (mg/L)/serum albumin (g/L)]. CSF and serum IgG concentrations were measured using the IGG-2 reagent cassette on a cobas c module instrument (Roche). The IgG index was calculated as [CSF IgG (mg/L)/serum IgG (g/L)/CSF albumin (mg/L)/serum albumin (g/L)].

CSF-specific IgG oligoclonal bands (OCBs) were determined using an in-house isoelectric focusing (IEF) method on 7.7% polyacrylamide gels with subsequent silver staining. Matched patient serum and CSF samples were run on adjacent lanes, and CSF-specific IgG OCBs were defined as extra bands in the gamma-zone, which were not present in the corresponding serum sample. For quality control, a positive CSF sample with known CSF-specific OCBs was run on each gel. A cut-off value of IgG OCB ≥ 2 was considered positive. Board-certified laboratory technicians, who were blinded to the clinical status, using strict procedures for quality control and run-approval, performed the analyses. All analyses were performed at the Sahlgrenska Neurochemistry Laboratory in Mölndal, Sweden.

As part of the diagnostic routine for MS investigations we also measured neurofilament light (NfL, n = 131), glial fibrillary acidic protein (GFAP) (n = 131), and tau (n = 125) concentrations in CSF. CSF NfL concentration was measured using a sensitive sandwich enzyme-linked immunosorbent assay (ELISA; NF-light® ELISA kit; UmanDiagnostics AB, Umeå, Sweden; Catalog # 10–7001 CE) as previously described [23], or with an in-house ELISA method as described previously in detail [24]. Comparison of the in-house ELISA and the UmanDiagnostics ELISAs showed CSF NfL concentrations in the same range and a strong linear correlation between CSF NfL values [24]. CSF GFAP concentration was measured by ELISA, as previously described [25]. CSF tau concentration (INNOTEST® hTAU Ag; Product # 81,572) was measured by ELISA, as previously described [26].

Data are presented as mean ± SD or as median and interquartile range (IQR), as appropriate. Data distribution was assessed with the Shapiro-Wilks test. The Mann–Whitney U test, unpaired T test, χ² test, and Fisher’s exact test were used for group comparisons as appropriate. The Mann–Whitney U test was used for all comparisons of CSF biomarkers in EDA-3 vs NEDA-3 and non-PIRA vs PIRA. EDSS values at BL/RBL and follow-up stratified by non-PIRA/PIRA were compared with the Wilcoxon matched-pairs signed rank test, and the p value threshold for multiple comparisons was set with the Holm-Šídák method. Next, based on the results of CSF biomarker comparisons, we select CSF KFLC, KFLC index and CSF tau for further analysis in univariable and multivariable cox proportional hazards regression models. As described above, two endpoints were used as time-dependent variables, EDA-3, and PIRA. Cox proportional hazards regression models were performed and the adjusted hazard ratios (aHR) along with corresponding 95% confidence intervals (CI) were calculated. Time to EDA-3 and PIRA were investigated with Kaplan–Meier survival analyses with corresponding logrank tests. For visualization in Kaplan–Meier curves with KFLC index as a predictor variable and based on previous work on the prognostic value of KFLC index, we stratified the cohort according to the cut-off value KFLC index > 100 [9, 12]. Based on previous investigations on prognostic factors, we chose to adjust the models for the following potential confounding covariates: age at the time of CSF sampling, sex, total follow-up time, time from sampling to DMT start, disease course at onset (CIS/RRMS), onset EDSS in the case of EDA-3 (i.e. the first EDSS score recorded at the time of sampling), or BL/RBL EDSS as defined above in the case of PIRA, and brain MRI characteristics (T2W lesions and BL CEL). In the model with EDA-3 as a dependent variable, we adjusted for exposure to high-efficacy DMT from onset. In the model with PIRA as a dependent variable, we adjusted for treatment strategy during follow-up (three categories: 1. patients who received first-line therapy during the whole follow-up; 2. patients who escalated therapy during the follow-up; and 3. patients who initiated high-efficacy DMT at baseline). Natalizumab, anti-CD20 therapies (rituximab or ocrelizumab), alemtuzumab, and autologous hematopoietic stem cell transplantation (AHCST) were classified as high-efficacy DMT, whereas sphingosine-1-phosphate receptor modulators (fingolimod, ponesimod, siponimod or ozanimod), cladribine tablets, teriflunomide, dimethyl fumarate, platform therapies, or no DMT, were grouped as low/moderate efficacy (first-line) DMT. We tested KFLC index > 100 as a binary categorical variable separately. Results of the other confounding covariates are presented only for the model including KFLC index as a continuous variable, as they did not substantially differ in the other models. Due to the exploratory nature of the study, no correction for multiple comparisons in the survival analyses was made. Statistical significance was assumed at p < 0.05. All statistical analyses and figures were performed/created with IBM SPSS version 28.0.1.0 (Armonk, NY: IBM Corp. 2011) and GraphPad prism version 9.1.0.

All patients included in this study had given informed consent to be registered in the SMSreg. All individual data from the different sources were made anonymous to the authors by the replacement of the personal identity numbers by unique number codes for use in the present study. The study has been approved by the Swedish Ethical Review Agency (Dnr: 2020-06851).

Anonymized data, not published in the article, will be shared on reasonable request from a qualified investigator.

Patients who presented with an EDA-3 event during follow-up had higher KFLC index at BL compared with patients who remained stable (median KFLC index 121.4 [IQR 56.7 – 191.4] vs. 66.3 [25.3–177.2], p = 0.015) (Table 1, Fig. 2A). Likewise, patients who presented with a PIRA event during follow-up had higher median KFLC index at BL (148.5 [106.9–253.5] vs. 78.3 [28.9–186.5], p = 0.009) (Fig. 2B). Furthermore, CSF levels of KFLC were significantly higher in EDA-3 vs. NEDA-3 and PIRA vs. non-PIRA (Fig. 2C, D). The IgG index was also noted to be significantly higher in PIRA vs. non-PIRA but not in EDA-3 vs. NEDA-3 (Fig. 2E, F). The correlation between KFLC index and IgG index was very high (Spearman’s rho = 0.85, p < 0.001), and we therefore opted to focus on KFLC index and not the IgG index hereafter. Patients in the EDA-3 group had slightly but significantly higher CSF tau levels compared with the NEDA-3 group (Fig. 2G). No differences in CSF NfL and CSF GFAP concentrations between any of the groups were observed (Figs. 2I–L).

Most cases achieved EDA-3 due to MRI signs of disease activity (n = 45, 73.8%). In only 13 patients, EDA-3 was due to a clinical relapse. In univariable cox regression models, both CSF KFLC concentrations and the KFLC index were associated with increased hazard of EDA-3 (HR 1.046, 95% CI 1.012–1.081, p = 0.008 For CSF KFLC; HR 1.001, 95% CI 1.00006–1.003, p = 0.041 for KFLC index) (Table 2). Furthermore, both RRMS disease course and exposure to high-efficacy DMT were protective (HR 0.35, 95% CI 0.19–0.63, p < 0.001 for RRMS; HR 0.54, 95% CI 0.32–0.91, p = 0.02). In contrast, other demographical and clinical factors such as age, sex, BL EDSS and MRI measures were not associated with increased risk of EDA-3. As CSF tau levels were slightly but significantly increased in EDA-3 compared with NEDA-3, we opted to include CSF tau as well in the cox regression model. However, in univariable and multivariable analyses, CSF tau was not predictive of a higher EDA-3 risk.

In a multivariable cox regression model, KFLC index as a continuous variable was an independent predictor of EDA-3 status at follow-up (aHR 1.002, 95% CI 1.001–1.004, p = 0.007) (Table 2). In addition, CSF KFLC also emerged as an independent predictor of EDA-3 status (aHR 1.053, 95% CI 1.016–1.09, p = 0.005). RRMS disease course at onset was independently associated with a protective effect against EDA-3 (aHR 0.235, 95% CI 0.09–0.61, p = 0.003). Likewise, the protective effect of exposure to high-efficacy DMT from disease onset was preserved (aHR 0.521, 95% CI 0.28–0.96, p = 0.036). Similar to the univariable models, other demographic and clinical factors were not predictive of EDA-3, as shown in Table 2.

As a next step, to investigate the time to EDA-3 in a Kaplan–Meier survival analysis, we opted to test KFLC index as a binary categorical variable based on the prognostic cut-off value KFLC index > 100. In univariable and multivariable cox regression models, KFLC index > 100 was associated with increased risk of EDA-3 (multivariable model—aHR 2.1, 95% CI 1.25–3.61, p = 0.005). The median time to EDA-3 for KFLC index > 100 was 23.5 months compared with 84 months in patients with KFLC index ≤ 100 (logrank p = 0.0064) (Fig. 3A).

In patients with EDA-3, 23 patients exhibited CDW, of which five were classified as RAW, and 18 as PIRA (Fig. 1). In the PIRA subgroup, 4/18 (22.2%) had new T2W lesions on MRI during follow-up. In non-PIRA, BL/RBL EDSS were significantly higher compared with last follow-up (p < 0.001), while in PIRA, EDSS at censoring was significantly higher compared with BL/RBL EDSS (p < 0.001).

In a univariable cox regression model, KFLC index was associated with a higher risk of PIRA (HR 1.003, 95% CI 1.001–1.006, p = 0.003 for KFLC index) (Table 2). In addition, a lower BL EDSS was associated with PIRA (HR 0.61, 95% CI 0.4–0.93, p = 0.02). None of the other demographical and clinical factors demonstrated a significant association with PIRA.

In a multivariable cox regression model, KFLC index emerged as an independent risk factor for PIRA at follow-up (aHR 1.005, 95% CI 1.002–1.008, p = 0.002) (Table 2). Likewise, CSF KFLC levels were also independently associated with PIRA at follow-up (aHR 1.08, 95% CI 1.006–1.16, p = 0.033). The association of low BL EDSS diminished in the multivariable model. Notably, exposure to high-efficacy DMT from onset did not show a significant protective effect against the risk of PIRA (aHR 0.6, 95% CI 0.16–2.24, p = 0.45).

Next, similarly to the analysis of time to EDA-3, we opted to test KFLC index > 100 as a binary categorical variable to investigate time to PIRA. In a multivariable cox regression, KFLC index > 100 was an independent predictor of PIRA (aHR 3.7, 95% CI 1.101–12.3, p = 0.03). The median time to PIRA in both non-PIRA and PIRA remained undefined (logrank p = 0.0027) (Fig. 3B).