Neurofilament light predicts neurological outcome after subarachnoid haemorrhage

Abstract

To improve outcome prediction following subarachnoid haemorrhage (SAH), we sought a biomarker integrating early brain injury and multiple secondary pathological processes in a prospective study of 42 non-traumatic SAH patients and 19 control individuals. Neurofilament light (NF-L) was elevated in CSF and serum following SAH. CSF and serum NF-L on Days 1-3 post-SAH strongly predicted modified Rankin score at 6 months, independent of World Federation of Neurosurgical Societies (WFNS) score. NF-L from Day 4 onwards also had a profound impact on outcome. To link NF-L to a SAH-specific pathological process, we investigated NF-L's relationship with extracellular haemoglobin. Most CSF haemoglobin was not complexed with haptoglobin, yet was able to be bound by exogenous haptoglobin i.e. haemoglobin was scavengeable. CSF scavengeable haemoglobin was strongly predictive of subsequent CSF NF-L. Next, we investigated NF-L efflux from the brain after SAH. Serum and CSF NF-L correlated positively. The serum/CSF NF-L ratio was lower in SAH versus control subjects, in keeping with glymphatic efflux dysfunction after SAH. CSF/serum albumin ratio was increased following SAH versus controls. The serum/CSF NF-L ratio correlated negatively with the CSF/serum albumin ratio, indicating that transfer of the two proteins across the blood-brain interface is dissociated. In summary, NF-L is a strong predictive marker for SAH clinical outcome, adding value to the WFNS score, and is a promising surrogate end point in clinical trials.

Keywords: glymphatic system; haemoglobin; haptoglobin; neurofilament light; subarachnoid haemorrhage.

Figure 1

CSF and serum NF-L predict long-term neurological outcome after SAH. (A) Serial CSF and serum NF-L over a 2-week period following SAH [median ± IQR (shaded zone)]. (B) In both CSF and serum, the maximum NF-L observed for each patient over the 2-week sampling period was significantly greater than the control population (mean ± 95% CI; unpaired t-test, P < 0.0001. CSF: control = 19, SAH = 42. Serum: control = 17, SAH = 41). (C) Ordinal logistical regression shows high CSF NF-L between Days 1–3 predicts poor outcome at 6 months on the mRS (odds ratio 4.3, 95% CI 1.1–16.2; P = 0.031, n = 24 and 12 for high and normal CSF NF-L, respectively). (D) Ordinal logistic regression shows high serum NF-L between Days 1 and 3 predicts poor outcome at 6 months on the mRS (odds ratio 2.3, 95% CI 0.6–6.6; P = 0.04, n = 7 and 27 for high and normal serum NF-L, respectively). High CSF and serum NF-L were defined as higher than the upper limit of the respective control population-based reference ranges. The NF-L reference range was defined as the mean ± 1.96 SD of the control population log-normal distribution.

Figure 2

Scavengeable haemoglobin is elevated in the CSF of SAH patients. (A) Pie charts representing the relative proportions of bound, scavengeable and unscavengeable haemoglobin over the 2-week sampling period. Increasing diameter reflects increasing total haemoglobin. Days and day groupings indicated in the panel. (B) Size-exclusion chromatography to identify haemoglobin (Hb) species. Neat or haptoglobin (Hp)-saturated samples reveal haem-containing haemoglobin peaks in the 415 nm Soret band.

Figure 3

Scavengeable haemoglobin is correlated with the brain injury marker NF-L. Multiple linear regression was used to predict CSF NF-L levels using the highest preceding scavengeable haemoglobin (Hb), while controlling for WFNS, age and sex (R² = 0.503, scavengeable haemoglobin P = 0.0003; n = 36); scavengeable haemaglobin plotted against model fitted (i.e. predicted) CSF NF-L values.

Figure 4

Brain-to-blood NF-L efflux. (A) Scatter plot of CSF NF-L versus serum NF-L reveals a significant Pearson r correlation (r = 0.62, P < 0.0001, n = 41). (B) Qalb is elevated from Day 4 until the end of the 2-week sampling period. The red line is the mean value for the control population. (C) The maximum Qalb for each patient over the 2-week sampling period is significantly greater than the control population (mean ± SEM; unpaired t-test, P < 0.0001, n = 20 and 34 for controls and SAH, respectively). (D) Scatter plot of Qalb versus the serum/CSF NF-L ratio reveals a significant negative Pearson r correlation (r = −0.58, P = 0.0008, n = 42).