Peripheral blood neurofilament light chain levels: the neurologist’s C-reactive protein?

This scientific commentary refers to ‘Serum neurofilament as a predictor of disease worsening and brain and spinal cord atrophy in multiple sclerosis’, by Barro et al.. (doi:10.1093/brain/awy154). Neurofilament proteins are the major scaffolding proteins of axons. Neurofilaments are heteropolymers composed of light (NfL), medium (NfM) and heavy (NfH) chain subunits (Giovannoni, 2006). Pathological processes that cause axonal damage result in the release of neurofilament proteins into the extracellular space. These then diffuse into the CSF and are transported into the peripheral blood. Levels of neurofilament proteins may thus be a good surrogate for quantifying axonal damage. Numerous studies have reported on the potential utility of CSF and peripheral blood neurofilament levels as a biomarker in several diseases characterized by axonal loss, including stroke, small vessel disease, HIV infection, head injury, amyotrophic lateral sclerosis, Alzheimer’s disease, Huntington’s disease, acute spinal cord injury, neuromyelitis optica and multiple sclerosis. Until recently one had to rely on measurement of neurofilament levels in CSF, which has limited its adoption. However, the emergence of the single molecule arrays (Simoa), an ultra-sensitive enzyme-linked immunosorbent assay (ELISA) technique, for measuring peripheral blood NfL levels has led to a renaissance of NfL as a biomarker in diseases including multiple sclerosis (Disanto et al., 2016; Kuhle et al., 2017; Barro et al., 2018; Siller et al., 2018). Although neurofilaments have good face validity as a biomarker, a critical question is how soon can peripheral blood NfL levels be adopted as a surrogate endpoint in the context of multiple sclerosis clinical trials? The well-established Prentice criteria for defining the necessary attributes of a surrogate endpoint require that (i) the baseline measurement of NfL is predictive of outcome; (ii) changes in NfL over time are predictive of outcome; and (iii) changes in NfL levels in response to a therapy are also predictive of outcome (Prentice, 1989). In this issue of Brain, Barro and co-workers confirm that serum NfL levels are higher in patients with multiple sclerosis compared to healthy controls (Barro et al., 2018); their cohort included patients with clinically isolated syndromes and relapsing-remitting, secondary progressive and primary progressive multiple sclerosis. Serum NfL levels above the 90th percentile of healthy control values were an independent predictor of Expanded Disability Status Scale (EDSS) worsening the following year (Barro et al., 2018), thereby fulfilling the first Prentice criterion. The probability of EDSS worsening increased with higher serum NfL percentile category, supporting it as a quantitative biomarker. Not surprising was the finding that increased serum NfL levels were associated with disease activity as defined by the presence of contrast enhancing and new or enlarging T2 lesions on MRI. What about end-organ damage? The higher the serum NfL levels, the more pronounced was future brain and cervical spinal cord volume loss on MRI. What this study does not fulfil, however, are the second and third of Prentice’s criteria above. These results confirm the utility of serum NfL levels as a baseline predictor in relation to converting from a clinically isolated syndrome to clinically definite multiple sclerosis (Disanto et al., 2016). What about NfH? In an earlier prospective study of acute optic neuritis, Petzold and colleagues showed that 50% of subjects presenting with acute optic neuritis had detectable plasma NfH levels (Lycke et al., 1998; Petzold et al., 2004b). Plasma NfH correlated inversely with visual acuity at presentation and levels were higher in subjects with poor recovery of vision compared to those with good recovery (Petzold et al., 2004b). In an observational study of 30 patients with relapsing multiple sclerosis being treated with interferon beta, median plasma NfH levels were significantly higher than in healthy controls (Petzold et al., 2004a); 58% of responders and 72% of interferon non-responders had detectable blood levels (Petzold et al., 2004a). However, plasma levels of NfH tend to be low and are not informative in a significant proportion of patients with multiple sclerosis, which limits the wider use of plasma neurofilament levels in this disorder. In a double-blind placebo-controlled trial of lamotrigine in secondary progressive multiple sclerosis, serum NfH levels correlated with multiple disability endpoints at 24 months, but again serum NfH was only detectable in a proportion of study subjects. The latter finding and the higher variability of the NfH assay (Petzold et al., 2010) make serum NfH less trial-ready than serum NfL as a potential surrogate outcome measure. In a recent small single centre study, T2 lesion volume increased and brain parenchymal volume decreased more rapidly in patients with higher baseline serum NfL levels (Siller et al., 2018). Treatment with a disease-modifying therapy reduced serum NfL levels and patients with relapse or with radiological activity had significantly higher serum NfL levels than those in remission or without new MRI lesions (Siller et al., 2018). In a study in early multiple sclerosis that investigated the potential treatment effect of riluzole, as an add-on neuroprotective agent to weekly intramuscular interferon-β (IFN-β)–1a, riluzole was found to have no effect on serum NfL levels (Kuhle et al., 2017). However, serum NfL levels decreased over time and were correlated with EDSS change, neuropsychological outcomes and brain volume change at 12 and 24 months. Brain volume decreased more rapidly in subjects with high baseline serum NfL at 12 and 24 months and the relationship was stronger at 24 months. In this study, higher and increasing serum NfL levels predicted a higher number of gadolinium-enhancing MRI lesions, confirming that serum NfL levels are useful as an inflammatory marker in multiple sclerosis (Kuhle et al., 2017). We are potentially close to having an answer to this question. Several valuable phase 3 datasets have recently been presented at international meetings suggesting that this is indeed the case. Unfortunately, these data have yet to be published in peer-reviewed journals. In a clinical cohort of 286 patients with multiple sclerosis, treatment with a disease-modifying therapy reduced serum NfL levels and patients who had a clinical relapse or radiological activity on MRI had significantly higher serum NfL levels compared to patients in ‘remission’ (Novakova et al., 2017). However, whether the lowering of serum NfL levels predicts long term outcome in these patients is unknown (Novakova et al., 2017). CSF neurofilament studies in relation to multiple sclerosis go back two decades to the first report by Lycke and colleagues (1998) who measured CSF NfL in subjects participating in a trial of acyclovir. Since then the evidence-base in support of NfL levels as a prognostic, disease activity and treatment response marker has gradually increased. However, the wide adoption of NfL as a biomarker in multiple sclerosis and related disorders has been limited by the need for a lumbar puncture. Now that NfL can reliably be detected in the peripheral blood, this will change. In summary, NfL is a non-specific axonal damage marker that is released in response to central or peripheral nervous system damage. In multiple sclerosis, serum NfL levels are acutely elevated in relation to disease activity and predict poor outcomes. Early indications are that serum NfL levels respond to effective disease-modifying therapies and that this response will predict a better long-term outcome. We need to establish whether serum NfL could be used to enrich neuroprotective trials by selecting those subjects that are most likely to progress in the next 2–3 years. I envisage short term neuroprotective studies using area under the curve (AUC) analyses of serum NfL levels as an outcome for screening effective neuroprotective drugs for late phase development. A good analogy is the area under the C-reactive protein curve in rheumatoid arthritis (Pencina et al., 2008). Once the single molecule array ultra-sensitive ELISA technique for measuring peripheral blood NfL levels has been validated across laboratories I predict its rapid adoption by the wider multiple sclerosis community. I have little doubt that it will become a surrogate endpoint and will be used in both clinical trials and in general neurology practice for prognostic profiling, as a treatment response marker and as a treatment target. Why wouldn’t one want to minimize the area under the serum NfL curve as a treatment target in multiple sclerosis? Tragically it has taken 20 years for us to get to this point. In future, we need a system and biomarker validation platform to speed-up biomarker discovery in multiple sclerosis and other neurological conditions.