Clinical characteristics of the GBS cohort

Baseline demographics and clinical characteristics are presented in Table 1. Ninety-six GBS patients, 34 women, were included in the study. The clinical GBS subtypes according to the Wakerly criteria were classic (n = 80), paraparetic (n = 3), PCB variant (n = 3), BWDP (n = 1) and MFS (n = 9). According to the neurophysiological examination, cases were classified as AIDP (n = 46), AMAN/AMSAN (n = 10), equivocal (n = 9), and normal (n = 18). NCS was missing for 11 patients. The albumin ratio of GBS subjects was elevated (n = 70), normal (n = 26), and not done (n = 1). Preceding infection was reported for 75 patients with respiratory tract infections (n = 49), gastrointestinal tract infections (n = 13), and others (n = 2). Treatment consisted of a five days course of intravenous immunoglobulin (IVIG 0.4 g/L/day, n = 42), plasmapheresis (n = 24), a second or more IVIG five days course (n = 7), or IVIG and plasmapheresis in combination (n = 3). Nine patients did not receive treatment, and information was missing for 11. The median GBSDS was 2 (IQR 2–4) at diagnosis, 4 at nadir (IQR 2–4), and 1 (IQR 0–2) at the last follow-up 1 (median 340 days, IQR 88–738).

Eighty-nine GBS patients had serum (n = 21), CSF (n = 26), or both (n = 42) collected within 30 days from clinical onset. Additionally, eight patients had only samples collected later in the disease course: serum (n = 6), CSF (n = 1), or both (n = 1). Serial serum samples (3–11 samples) were available for 19 patients, and CSF (2–3 samples) were obtained from 18 patients.

Comparison between GBS patients and healthy controls

GBS patients had higher sNfL levels than HC (median [IQR] 50.2 ng/L [16.7–209 ng/L] vs. 12.3 ng/L [6.1–19.8 ng/L], age-adjusted p < 0.0001) and CSF NfL levels (972 ng/L [510–972 ng/L] vs 499 ng/L [264–983 ng/L], age-adjusted p < 0.0001) compared with HC. The sNfL Z-scores were significantly higher in GBS patients (3.1 [1.9–3.7]) compared with HC (0.7 [− 0.2 to 1.5], p < 0.0001), and the NfL ratio was significantly lower (31 [14.1–68] vs 42.4 [33.3–55.5], p = 0.02).

The association between NfLvalues, baseline clinical characteristics and GBS severity

The association between sNfL levels, sNfL Z-score, NfL ratio, NfL index, Qalb and demographics, clinical characteristics, and GBS severity was evaluated (supplementary data: Table 1). NfL levels in serum and CSF and sNfL Z-scores increased with increasing GBSDS at diagnosis, p = 0.004, < 0.0001 and 0.0002 respectively. Patients with AMAN/AMSAN had higher sNfL (Med [IQR] 234 ng/L [9.4–341 ng/L]) and sNfL Z-scores (3.5 [1.0–3.9]) than those with normal neurophysiological subtype (sNfL 6.9 ng/L [5.3–20.4 ng/L], sNfL Z-score 0.69 [− 0.3 to 2.9]) or AIDP (sNfL 37.3 [23.4–111], sNfL Z-score 3.0 [2.2–3.5]), p = 0.002 and 0.02 respectively (Fig. 1). Patients with MFS had higher sNfL levels (187 ng/L [51.2–563 ng/L]) and sNfL Z-scores (3.7 [3.2–3.9]) than patients with classic GBS (sNfL 40.5 ng/L [18.8–197 ng/L], sNfL Z-score 3.0 [2.1–3.6]), PCB (6.0 ng/L [5.2–6.9], sNfL Z-score − 0.23 [− 0.2 to 0.7]) and paraparetic subtype (sNfL 66.6 ng/L [4.2–129 ng/L], sNfL Z-score 1.4 [− 0.36 to 3.2]), p = 0.03 and 0.002 respectively. The NfL index was higher in patients with normal neurophysiological subtype (6.9 [1.9–10.8]) compared with patients with AIDP (3.2 [1.4–6.1]), AMAN/AMSAN (2.1 [0.3–9.0]) and equivocal (0.6 [0.3–2.1]), p = 0.04. Multiple comparisons of sNfL and sNfL Z-score in clinical and neurophysiological suptypes are shown in supplementary data, Table 2–5. NfL ratio, NfL index and Qalb at baseline was otherwise not significantly influenced by gender, GBSDS, clinical GBS subtype, type of preceding infection, or neurophysiological subtype.

Fig. 1

sNfL Z-score (a), NfL ratio (b), and NfL index (c) in different neurophysiologocial subtypes. Line and whiskers represent median and interquartile range, dots individual values. *p < 0.05, stastitically significant results from multiple comparisons analyses are shown. Abbreviations: NfL Neurofilament light chain, AIDP acute inflammatory demyelinating polyneuropathy, AMAN/AMSAN acute motor/sensorimotor axonal neuropathy

Evolvement of NfL levels over time

To map the evolution of NfL levels over time, we calculated the median values and IQR of sNfL, CSF NfL and the NfL ratio depending on which week after clinical onset the samples were collected and analyzed sNfL and CSF NfL in available serial samples.

Between baseline and week two, the median sNfL level increased from 16.5 ng/L (IQR 5.5–46.2 ng/L) to 89.5 ng/L (IQR 18.3–291 ng/L), p = 0.003. The increase in CSF NfL was more modest, median 617 ng/L (IQR 389–1058 ng/L) to median 828 ng/L (IQR 340–3808 ng/L, p = 0.27). While the sNfL concentration peaked in week five, the CSF NfL level peaked in week four.

The median NfL ratio decreased in the first two weeks, from 56.6 (IQR 23.9–73.7) to 32.9 (9.9–27.3), p = 0.09, while the Qalb rose from 7.5 (IQR 6.4–15.8) to 13.4 (IQR 17.3–7.6), p = 0.23 (Fig. 2).

Fig. 2

Evolvement of NfL ratio and Qalb from week one to week 5 after symptom onset expressed as median and interquartile range. NfL neurofilament light chain; Qalb albumin quotient

Nineteen patients had three or more serum samples collected between day 3 and 335 from symptom onset (mean days 49) and the average duration from clinical onset to sNfL peak levels was 22.8 days (Fig. 3a).

Fig. 3

a Serial sNfL levels in individual patients by days from clinical GBS onset. b CSF NfL levels from clinical onset (week one) and at follow-up (week three). Abbreviations: sNfL serum neurofilament light chain and CSF-NfL cerebrospinal fluid neurofilament light chain, Med median, IQR interquartile range

Two CSF samples were available from 18 patients. The average duration between samples was 20.8 days (range 3–174). Median CSF-NfL increased from 709.5 ng/L (IQR 254–1362 ng/L) to 1402 ng/L (IQR 774–8170 ng/L) (p < 0.001) (Fig. 3b).

NfL and GBSDS outcome

We dichotomized patients into severe GBS (GBSDS > 2, i.e., inability to walk) and those with normal or less disability at three and 12 months. Patients with MFS and PBC variants were excluded from this analysis, except for the need of respiratory support, since the ability to walk is usually not impaired in these subtypes. Results are shown in Table 2. NfL levels in serum and CSF and the sNfL Z-scores were significantly higher in those with severe GBS and those requiring respiratory support compared with less disabled GBS patients. No association was shown between the NfL ratio/NfL index and GBS severity. The Qalb was significantly higher in those needing respiratory support and with GBSDS > 2 at three months but not at 12 months.

Table 2 NfL parameters and albumin quotient compared between groups based on outcomesNfL Z-score to determine GBS prognosis

We evaluated the potential of the sNfL Z-score to predict the risk of severe residual disability in GBS. ROC analysis was carried out for respiratory support and GBSDS > 2 at three months (Fig. 4). A Z-score of > 3.2 had a sensitivity of 82% and specificity of 71% for GBSDS > 2 at three months and a sensitivity of 79%, and a specificity of 65% for requiring respiratory support. Given that 25% of patients will need respiratory support, a sNfL Z-score of > 3.2 will have a positive predictive value (PPV) of 43% and a negative predictive value (NPV) of 91%.

Fig. 4

ROC curves of the sNfL Z-score for respiratory support and GBSDS > 2 at three months. Abbreviations: GBSDS Guillain–Barré syndrome disability scale, AUC area under the curve

Comparison of albumin ratio, Nfl ratio and NfL index as diagnostic biomarkers for GBS

We excluded MFS and PCB subtypes from the comparison analysis as these rare GBS subtypes usually have low impact on peripheral nerves compared with classic GBS and therefore may be considered outliers among GBS subtypes.

The Qalb and the sNfL concentration correlated significantly in GBS patients (r = 0.4, p = 0.02). No correlation was found between Qalb and sNfL in the MS or ALS populations. Qalb was not available for HCs (Supplementary data, Table 6).

The NfL ratio was lower in GBS patients (median [IQR] 30.8 [15.6–65]) compared with HC (42.4 [ 33.3–55.5]), active MS (49.2 [ 27.9–96.9]), non-active MS (38.9 [23.1–71.3]) and ALS (69.4 [52.2–113]), p < 0.0001 (Fig. 5a). After applying multiple comparison tests, the difference between NfL ratios was statistically significant between GBS and active MS (p = 0.048) and between GBS and ALS (p < 0.0001) (Supplementary data, Tables 7 and 8).

Fig. 5

a, b Comparison of NfL ratio and NfL index between study populations and c ROC curves for NfL ratio, NfL index and Qalb. Dots represent individual values, line and whiskers median and interquartile range. a NfL ratio in HC, GBS, MS and ALS. b NfL index in GBS, MS an ALS. *p < 0.05, **** p < 0.0001. c ROC for NfL index and Qalb in GBS versus MS and ALS and for NfL ratio in GBS versus MS, ALS and HC. NfL neurofilament light chain, Qalb albumin quotient, HC healthy controls, GBS Guillain–Barré syndrome, MS multiple sclerosis, ALS amyotrophic lateral sclerosis, ROC receiver operator charachteristics curve

To estimate the influence of an impaired BCSFB on the sNfL concentration, we calculated the NfL index. The NfL index was statistically significantly lower in GBS (median [IQR] 2.6 [1.2–6.1]) compared with active MS (9.8 [7.3–24.3]) non-active MS (7.7 [4.3–13.9]), and ALS (13.6 [8.8–20.2]), p < 0.0001 (Fig. 5b). The multiple comparison test showed statistically significant differences between GBS and the other study populations (Supplementary data, Tables 7 and 9).

The Qalb, NfL ratio and NfL index were determined in GBS, MS and ALS patients and their diagnostic value for GBS was determined in ROC analyses.

The AUC for NfL ratio was 0.66 (95% CI 0.55–0.78, p = 0.0018), and at an NfL ratio < 37.5, the sensitivity was 64% (95% CI 48–75%) and the specificity 69% (95% CI 62–75%).

The ROC analysis for NfL index showed that the AUC was 0.86 (95% CI 0.78–0.93, p < 0.0001), and at an NfL index of < 6.9, the sensitivity was 81% (95% CI 65–90%) and the specificity 70% (95% CI 60–78%).

The AUC for Qalb was 0.88 (95% CI 0.81–0.95, p < 0.0001), and at an Qalb of > 7.1, the sensitivity was 83% (95% CI 68–92%) and the specificity 80% (95% CI 71–87%) (Fig. 5c).