Neurol Ther (2023) 12:303–317 https://doi.org/10.1007/s40120-022-00427-8 ORIGINAL RESEARCH Serum Neurofilament Light Trajectories and Their Relation to Subclinical Radiological Disease Activity in Relapsing Multiple Sclerosis Patients in the APLIOS Trial Amit Bar-Or . Xavier Montalban . Xixi Hu . Harald Kropshofer . Petra Kukkaro . Neva Coello . Inga Ludwig . Roman Willi . Martin Zalesak . Krishnan Ramanathan . Bernd C. Kieseier . Dieter A. Häring . Morten Bagger . Edward Fox Received: October 4, 2022 / Accepted: November 17, 2022 / Published online: December 19, 2022 Ó The Author(s) 2022 ABSTRACT 20 mg. Frequent sNfL sampling (14 time points over 12 weeks) and monthly MRI scans enabled Introduction: Several studies have described key analyses including assessment of the group- prognostic value of serum neurofilament light level temporal relationship of sNfL levels with chain (sNfL) at the group level in relapsing on-study subclinical development of gadolin- multiple sclerosis (RMS) patients. Here, we ium-enhancing (Gd ?)T1 lesions. Prognostic aimed to explore the temporal association value of baseline sNfL (‘‘high’’ vs. ‘‘low’’) level for between sNfL and development of subclinical subsequent on-study clinical relapse or Gd ? T1 disease activity as assessed by magnetic reso- activity was assessed. Individual patient-level nance imaging (MRI) at the group level and development of on-study Gd ? T1 lesions was evaluate the potential of sNfL as a biomarker for compared across three predictors: baseline capturing subclinical disease activity in indi- Gd ? T1 lesion number, baseline sNfL (‘‘high’’ vs. vidual RMS patients. ‘‘low’’), and time-matched sNfL. Methods: In the 12-week APLIOS study, patients Results: In patients developing Gd ? T1 lesions (N = 284) received subcutaneous ofatumumab at week 4 (absent at baseline), sNfL levels increased during the month preceding the week-4 Supplementary Information The online version MRI scan and then gradually decreased back to contains supplementary material available at https:// baseline. High versus low baseline sNfL conferred doi.org/10.1007/s40120-022-00427-8. A. Bar-Or (&) H. Kropshofer  P. Kukkaro  N. Coello  I. Ludwig  Center for Neuroinflammation and Experimental R. Willi  M. Zalesak  K. Ramanathan  Therapeutics and Department of Neurology, B. C. Kieseier  D. A. Häring  M. Bagger Perelman School of Medicine, University of Novartis Pharma AG, Basel, Switzerland Pennsylvania, Philadelphia, PA, USA B. C. Kieseier e-mail: amitbar@pennmedicine.upenn.edu Department of Neurology, Medical Faculty, X. Montalban Heinrich-Heine University, Dusseldorf, Germany Department of Neurology-Neuroimmunology and E. Fox Centre d’Esclerosi Múltiple de Catalunya (Cemcat), Central Texas Neurology Consultants and Dell Hospital Universitari Vall d’Hebron, Barcelona, Medical School, The University of Texas at Austin, Spain Round Rock, Austin, TX, USA X. Hu Novartis Pharmaceuticals Corporation, East Hanover, NJ, USA 304 Neurol Ther (2023) 12:303–317 increased risk of subsequent on-study clinical The present study explored both the relapse or Gd ? T1 activity (HR, 2.81; p \ 0.0001) temporal association between sNfL and in the overall population and, notably, also in the development of subclinical disease patients without baseline Gd ? T1 lesions (HR, activity as assessed by MRI at the group 2.48; p = 0.0213). Individual patient trajectories level and evaluated the potential of sNfL revealed a marked difference in Gd ? T1 lesions as a biomarker for capturing newly between patients with the ten highest vs. lowest developing subclinical disease activity in baseline sNfL levels (119 vs. 19 lesions). Prognos- individual relapsing MS (RMS) patients tic value of baseline or time-matched sNfL for on- study Gd ? T1 lesions was comparable to that of What was learned from the study? the number of baseline MRI Gd ? T1 lesions. Our findings indicate that sNfL may have Conclusions: sNfL measurement may have utility in monitoring newly developing utility in capturing and monitoring subclinical subclinical disease activity in RMS disease activity in RMS patients. sNfL assess- patients as shown by its temporal ments could complement regular MRI scans and association with new gadolinium- may provide an alternative when MRI assess- enhancing (Gd ?) T1 lesion activity, its ment is not feasible. overall comparable prognostic value to ClinicalTrials.gov: NCT03560739. baseline MRI, and its added value over Classification of Evidence: This study provides MRI in prognostic value of on-study class I evidence that serum neurofilament light Gd ? T1 lesion formation in patients free may be used as a biomarker for monitoring of brain MRI disease activity at baseline subclinical disease activity in relapsing multiple sclerosis patients, as shown by its elevation in Overall, our results suggest that when the weeks preceding the development of new frequent MRI scans are not feasible, serial gadolinium-enhancing T1 lesion activity. sNfL measurements may provide an attractive alternative to MRI monitoring if Keywords: Biomarker; Ofatumumab; Relapsing a standardized test becomes available to multiple sclerosis; Serum neurofilament light physicians chain; Subclinical disease activity Key Summary Points INTRODUCTION Why carry out this study? Multiple sclerosis (MS) is a chronic inflamma- Multiple studies have demonstrated that tory, demyelinating, autoimmune disorder of serum neurofilament light chain (sNfL) the central nervous system, leading to accu- levels are on average abnormally increased mulation of disability. The clinical disease in untreated multiple sclerosis (MS) course of MS is heterogeneous, and much of the patients compared to controls, mainly disease is subclinical [1]. The monitoring and driven by focal inflammatory central detection of suboptimal response to treatment, nervous system activity, and are reduced prognosis of clinical outcome, and therapeutic by effective therapies. However, insights decision-making in individual patients are key are lacking into the kinetics of sNfL challenges in the management of MS and are abnormalities in relation to newly currently assessed only with clinical and mag- developing disease activity, the potential netic resonance imaging (MRI) measures. use of sNfL to capture sub-clinical disease Neurofilament light chain (NfL) is a activity, and how sNfL measurements may cytoskeletal protein exclusively expressed by complement magnetic resonance imaging neurons and their axons, and its release into the (MRI) scans in clinical practice in order to cerebrospinal fluid (CSF) and blood is a highly monitor disease activity specific sign of neuroaxonal injury [2–4]. A Neurol Ther (2023) 12:303–317 305 strong correlation has been demonstrated syringe and SensoReadyÒ auto-injector pen. In between CSF and serum/plasma levels of NfL as total, 284 patients received ofatumumab 20 mg a blood biomarker for monitoring neuronal (0.4 ml) subcutaneous injections on days 1, 7, injury and treatment response in MS [2, 3]. At and 14 (initial doses) and thereafter every the group level, elevated NfL concentrations at 4 weeks from week 4 onwards (subsequent baseline are associated with subsequent on- doses) [8]. study relapses, MRI lesions, brain volume loss Eligibility criteria at screening included an (BVL), spinal cord atrophy, and disability age of 18–55 years; a diagnosis of RMS; an worsening [2, 4]. While such studies have pro- Expanded Disability Status Scale score of vided evidence supporting serum NfL (sNfL) as a 0–5.5; C 1 relapse in the year before screening potential biomarker of both current and future or C 2 relapses in the 2 years before screening or disease activity in patients with relapsing MS a positive gadolinium-enhancing (Gd ?) scan (RMS) [5], investigations of the temporal asso- during the year before randomization; and ciation between sNfL and the development of neurologically stable disease in the month prior subclinical disease activity as assessed by MRI to randomization. Further details of the study and evaluation of the potential of sNfL as a design and patient population are reported biomarker for capturing subclinical disease elsewhere [7]. activity in individual RMS patients are needed. Ofatumumab, a fully human anti-CD20 sNfL Sampling and Assay Characteristics monoclonal antibody, with a monthly 20 mg subcutaneous (SC) dosing regimen, demon- Blood sampling for sNfL measurement was done strated superior efficacy versus teriflunomide in the full study population following the 14 mg oral once daily and a favorable safety sampling scheme for the evaluation of phar- profile in patients with RMS in the phase 3 macokinetics for the bioequivalence assessment ASCLEPIOS I and II trials. Ofatumumab signifi- at 14 time points over 12 weeks, i.e., at baseline/ cantly lowered sNfL levels versus teriflunomide pre-treatment period and on study days 1, 4, 7, from the first post-baseline assessment at 14, 28, 42, 56, 57, 59, 63, 70, 77, and 84 (Fig. 1) month 3 through 24 [6]. [8]. The 12-week, open-label, phase 2 APLIOS Samples were analyzed using the Siemens study in RMS [7] incorporated frequent blood sNfL assay, on ADVIA CentaurÒ, a novel, vali- sampling (14 time points) and monthly MRI dated assay. Briefly, the ADVIA Centaur NfL scans. This design offered a unique opportunity assay is an automated two-site sandwich to study sNfL trajectories, explore the temporal immunoassay using direct chemiluminometric association between sNfL levels and subclinical technology. The assay uses two mouse mono- disease activity, and evaluate the potential of clonal antibodies (mAb) that bind to the NfL, sNfL as a biomarker for capturing and moni- each recognizing a unique epitope. The anti- toring subclinical disease activity in individual bodies are licensed from Quanterix and thus patients with RMS. have the same specificity as antibodies in the Simoa NF-lightTM assay. The solid phase reagent MATERIALS AND METHODS contains a biotinylated anti-NfL mAb, coupled to a paramagnetic particle conjugated with APLIOS Study Design and Patient streptavidin. The Lite reagent contains a mon- Population oclonal anti-NfL antibody conjugated with acridinium ester for chemiluminescent detec- tion. The accumulated light signal is directly The APLIOS study was a 12-week, randomized, related to the sample NfL concentration. This open-label, parallel-group, multicenter, phase 2 assay uses a full 7-point standard curve for cal- study in RMS (NCT03560739) that demon- ibration. The analytical sensitivity of LoB was strated the bioequivalence of two formulations: determined to be 0.89 pg/ml, LoD was 1.49 pg/ a subcutaneous injection (20 mg) pre-filled 306 Neurol Ther (2023) 12:303–317 Fig. 1 APLIOS study design with sNfL sampling time while a patient was on steroid therapy for relapse and points. aNine months or until the B cells returned to their within 14 days upon termination of steroid therap. eOn baseline value or to LLN; brandomization was stratified by non-injection visits the PK (Plasma) and neurofilament body weight (\ 60 kg, 60–90 kg, and [ 90 kg); cday 7 to samples were drawn at any time during the visit. AI day -1; din case of relapse, if an MRI for efficacy was to be autoinjector; BL baseline; HCP healthcare professional; scheduled within 14 days of the initiation of steroid LLN lower limit normal; MRI magnetic resonance treatment, the MRI was performed before steroid treat- imaging; PFS prefilled syringe; PK pharmacokinetics; sNfL ment was initiated. No MRI for efficacy was performed serum neurofilament light chain ml, and LLoQ was 1.85 pg/ml. The linearity was patients with Gd ? T1 lesions on week 4 and 8 established from 1 to 646 pg/ml in the serum. scans. For each case, an additive mixed model The linear regression result was R2 = 0.996 was fitted to allow a qualitative assessment of (p \ 0.001) using JMP (version 14.0, SAS group-level sNfL over time. The model was Institute). adjusted for age and log of baseline sNfL levels and included a nonparametric smoother of MRI Assessments and Analyses their time, random intercept, and random slope. MRI scans, including assessment of Gd ? T1 lesions, were performed at baseline and weeks 4, sNfL over Time Versus Disease Activity 8, and 12, with timing of the baseline scans and Individual Trajectories ranging from 57 days to 1 day prior to treat- Baseline sNfL was defined as the last assessment ment initiation. Gd ? T1 lesions from scans on or before day 1, and the median baseline collected within 14 days after termination of sNfL level across all patients was 9.1 pg/ml. The steroid therapy were excluded from the sNfL concentrations were analyzed using spa- analyses. ghetti plots, and patients were grouped by (1) sNfL during the study showing values con- Statistical Analyses stantly above baseline sNfL median (N = 60), (2) sNfL during the study showing both values above and below (‘‘crossed baseline median’’) Temporal Association Between Serial sNfL (N = 146), and (3) sNfL during the study show- Levels and Gd 1 T1 Lesions ing values constantly below the baseline sNfL The temporal association of serial sNfL levels median (N = 78). For the above-defined longi- and MRI activity was analyzed in patients with tudinal sNfL groups, group-level summaries of no Gd ? T1 lesions at baseline and with no confirmed clinical relapses (according to pre- missing scans and separately for patients with specified criteria [6]), Gd ? T1 lesions, and the Gd ? T1 lesions on week 4 scan only and Neurol Ther (2023) 12:303–317 307 proportion of patients with no evidence of dis- (1) model ‘‘baseline Gd ? T1’’ used the number ease activity (NEDA-3) were assessed. NEDA-3 of Gd ? T1 lesions on the baseline scan as the was defined as no confirmed clinical relapses, predictor, (2) model ‘‘baseline sNfL’’ used the no Gd ? T1 lesions on any post-baseline scans baseline sNfL measure ([ 9.1 pg/ml and no new/enlarging T2 lesions (relative to vs. B 9.1 pg/ml) as the predictor, and (3) model baseline) on any post-baseline scans, and no ‘‘time-matched sNfL’’ used the sNfL level on the 3-month confirmed disability worsening. The MRI scan day (or if unavailable, the last sNfL individual sNfL trajectories and disease activi- measurement before the scan day; [ 9.1 pg/ml ties of patients with the ten lowest and ten vs. B 9.1 pg/ml) as the predictor. All models highest baseline sNfL levels were also presented. were adjusted for age. Their total numbers of confirmed clinical AUCs for the development of week 4, 8, and relapses and Gd ? T1 lesions were summarized. 12 Gd ? T1 lesions were calculated from ten- fold cross-validation. Confidence intervals (CIs) Prognostic Value of Baseline sNfL Levels of AUCs were obtained through 10,000 boot- for On-Study Disease Activity strap resampling iterations. We dichotomized cohorts as baseline sNfL level above median: ‘‘high’’ ([ 9.1 pg/ml) or below Standard Protocol Approvals, median: ‘‘low’’ (B 9.1 pg/ml) to analyze the risk Registrations, and Patient Consents of on-study disease activity (confirmed clinical relapses or Gd ? T1 lesions) using a Cox The study was conducted in accordance with regression model of time since treatment start the International Conference on Harmonisa- to the first confirmed clinical relapse or first tion guidelines for Good Clinical Practice [9] observation of on-study Gd ? T1 lesions, and the principles of the Declaration of Helsinki whichever was earlier. The analysis was per- [10]. The protocol was approved by an inde- formed in all patients and in the subgroup of pendent ethics committee or institutional patients free of Gd ? T1 lesions at baseline. For review board for each study site, and all patients the analysis of all patients, the model included provided written informed consent before baseline sNfL category as factor (‘‘high’’ vs. commencing trial-related procedures. This arti- ‘‘low’’) and age, baseline number of Gd ? T1 cle is based on previously conducted studies and lesions, baseline volume of T2 lesions, baseline does not contain any new studies with human EDSS, and number of relapses in previous year participants or animals performed by any of the as continuous covariates. For the analysis authors. including only patients free of Gd ? T1 lesions at baseline, the model included baseline sNfL Data Availability Statement category as factor and age, baseline volume of T2 lesions, baseline EDSS, and number of The datasets generated during and/or analyzed relapses in previous year as continuous during the current study are available from the covariates. corresponding author on reasonable request. Patient-level Identification of On-Study Gd 1 T1 Lesion Activity RESULTS Using area under the receiver-operating char- acteristics curve (ROC AUC), we assessed the Patient Demographics and Baseline accuracy of prediction of whether individual Characteristics patients would have on-study Gd ? T1 lesions on each of week 4, 8, and 12 scans, using In total, 284 patients were included in the logistic regression models. For each of the three study. At baseline, patients in the high baseline post-baseline scans, the development of Gd ? sNfL ([ 9.1 pg/ml) group versus those in the low T1 lesions was compared across three models: baseline sNfL (B 9.1 pg/ml) group were found to 308 Neurol Ther (2023) 12:303–317 Table 1 Patient demographics and baseline disease characteristics by baseline sNfL low (B 9.1 pg/ml) vs. high ([ 9.1 pg/ ml) Characteristics Low sNfL High sNfL All patients (N = 146) (N = 138) (N = 284) [7] Age, years 36.1 ± 8.12 38.6 ± 9.56 37.3 ± 8.92 Sex, female, n (%) 102 (69.9) 97 (70.3) 199 (70.1) Race, white, n (%) 145 (99.3) 130 (94.2) 275 (96.8) Weight, kg 76.55 ± 17.81 70.69 ± 18.55 73.70 ± 18.38 BMI, kg/m2 26.42 ± 5.65 24.58 ± 6.48 25.53 ± 6.13 MS duration since first symptom, years 8.40 ± 7.15 10.19 ± 8.28 9.27 ± 7.75 No. of relapses in the year before the study 1.3 ± 0.70 1.4 ± 0.73 1.3 ± 0.72 No. of relapses in the month 12 to month 24 before 0.9 ± 1.03 1.1 ± 2.01 1.0 ± 1.58 the study EDSS score 2.82 ± 1.32 3.08 ± 1.26 2.95 ± 1.30 No. of Gd ? T1 lesions 0.5 ± 2.55 2.6 ± 6.47 1.5 ± 4.97 Patients with Gd ? T1 lesions, n (%) 31 (21.2) 76 (55.1) 107 (37.7) Patients with previous DMTs, n (%) 103 (70.5) 91 (65.9) 194 (68.3) Data are presented as mean ± SD, unless otherwise specified BMI body mass index; DMTs disease-modifying therapies; EDSS Expanded Disability Status Scale; Gd ? gadolinium enhancing; MS multiple sclerosis; sNfL serum neurofilament light chain be slightly older (38.6 vs. 36.1 years) and had a lesions at baseline and were observed to develop lower BMI (24.58 vs. 26.42 kg/m2), a longer Gd ? T1 lesions at both week 4 and 8 scans disease history (10.19 vs. 8.40 years), and a (most likely lesions persisting for [ 4 weeks higher number of Gd ? T1 lesions (2.6 vs. 0.5) after they were first observed) (Fig. 2B), except (Table 1). that in this group, sNfL did not decrease to baseline levels by week 12. Analysis that Temporal Association of sNfL Levels accounts further for sex was also considered and and Gd 1 T1 Lesions the findings remain the same. The unique approach of frequent blood sam- sNfL Concentrations over Time pling and imaging in APLIOS enabled us to and Disease Activity assess the temporal relationship between sNfL and the development of subclinical new disease At the group level, ofatumumab treatment was activity, assessed as new Gd ? T1 lesions. We associated with a consistent decline in sNfL observed that in patients without Gd ? T1 levels from baseline up to 12 weeks of study lesions at baseline who developed one or more treatment (Supplementary Material; data on Gd ? T1 lesion/s at week 4, the group-level file), similar to the trend observed in the pivotal average sNfL levels increased during the month phase 3 studies [6]. before the week 4 MRI scan, reached maximum We first considered whether sNfL levels over levels over a few weeks, and then gradually time were associated with measures of disease decreased to baseline levels (Fig. 2A). This was activity, as previously reported [2, 4], by divid- also the case in patients who had no Gd ? T1 ing patients into three groups based on their Neurol Ther (2023) 12:303–317 309 Fig. 2 Temporal association of sNfL levels and Gd ? T1 Additive mixed model with a nonparametric smoother of lesions in patients free of Gd ? T1 lesions at baseline: time adjusted for age and log of baseline sNfL. aPatients A patients (N = 13) Who only had Gd ? T1 lesions on were free of Gd ? T1 lesions at baseline. Gd ? gadolin- the week 4 scan and B patients (N = 7) who had ium enhancing; sNfL serum neurofilament light chain Gd ? T1 lesions on both week 4 and week 8 scans. longitudinal sNfL levels. sNfLs over time by Over time, the proportion of patients with longitudinal sNfL group are presented as spa- Gd ? T1 lesions was also higher in the group ghetti plots in Fig. 3. We found that the pro- with sNfL levels above the baseline median portion of patients with confirmed clinical versus those with sNfL levels that crossed the relapses was higher in the group of patients baseline median or below the baseline median, with sNfL levels constantly above the baseline with 64.7% vs. 31.1% and 16.2% at week 4, median (9/60; 15%) versus groups with sNfL 25.9% vs. 10.8% and 9.0% at week 8, and 11.9% levels that crossed the baseline median (4/146; vs. 4.4% and 3.9% at week 12, respectively. 2.7%) or constantly below the baseline median The mean number of Gd ? T1 lesions by (2/78; 2.6%). longitudinal sNfL group over 12 weeks is Fig. 3 sNfL over time versus disease activity and individ- median; the group ‘below baseline median’ includes ual trajectories. Dotted line: baseline median (9.1 pg/ml). patients whose serial sNfL values were all B baseline sNfL, serum neurofilament light chain. Day 1 represents median; the group ‘crossed baseline median’ includes the baseline. The group ‘above baseline median’ includes rest of the patients patients whose serial sNfL values were all [ baseline 310 Neurol Ther (2023) 12:303–317 presented in Table 2. As shown, the mean patients with the ten lowest baseline sNfL number of Gd ? T1 lesions over time was levels. higher in the group of patients with sNfL levels above the baseline median versus those with Prognostic Value of Baseline sNfL Levels sNfL levels that crossed or were below the for On-Study Disease Activity baseline median. The proportion of patients with NEDA-3 In both the overall patient population and the assessed by longitudinal sNfL subgroup was subset of patients who were free of Gd ? T1 higher in patients with below baseline median lesions at baseline, we confirmed that baseline sNfL (51/78; 65.4%) compared to those with sNfL level had a prognostic value for on-study above (13/60; 21.7%) or crossing the baseline disease activity (confirmed clinical relapses or sNfL median levels (73/146; 50.0%; Fig. 4). Gd ? T1 lesions). Specifically, high baseline To illustrate individual patient sNfL levels sNfL levels conferred an increased risk of on- over the course of the study and the associated study confirmed clinical relapses or Gd ? T1 relapses and Gd ? T1 lesions, sNfL trajectories lesions versus low baseline sNfL levels (hazard in patients with the ten lowest and ten highest ratio [HR], 2.81 [95% CI: 1.78, 4.42]; p \ 0.0001; baseline sNfL levels are presented in Fig. 5A and Fig. 6A). Among the patients free of Gd ? T1 5B, respectively. lesions at baseline (N = 176), the high baseline Gd ? T1 lesions and clinical relapses were sNfL levels also predicted an increased risk of more frequently observed over the course of the on-study confirmed clinical relapses or Gd ? T1 study in patients with the highest baseline sNfL lesions compared with low baseline sNfL levels levels than in those with the lowest baseline (HR, 2.48 [95% CI: 1.15, 5.39]; p = 0.0213) sNfL levels. Over time (including baseline), in (Fig. 6B). patients with the ten highest versus ten lowest Analysis that accounts further for sex was baseline sNfL levels, seven versus four patients also considered, and the findings remain the reported at least one Gd ? T1 lesion and seven same. vs. one patients had a scan with at least two An important question relates to the ability lesions. The total number of Gd ? T1 lesions of sNfL, measured in an individual patient, to was notably higher in patients with the ten capture the presence of subclinical disease highest vs. ten lowest baseline sNfL levels (116 activity. Using ROC curves, we observed that vs. 19); in patients with the ten lowest baseline the identification of on-study Gd ? T1 lesions sNfL levels, most of the lesions were from one based on sNfL levels in individual patients, patient. One notes that due to the frequent MRI either at baseline or time matched, was com- scans, the total number may include lesions parable in performance to the prediction based observed in more than one scan. Two clinical on the number of Gd ? T1 lesions on the relapses were observed among those ten baseline MRI scan, an observation which was patients with highest baseline sNfL and none in consistent at all time points over 12 weeks Table 2 Number of Gd ? T1 lesions by longitudinal sNfL group Visit Above baseline median Below baseline median Crossed baseline median All patients (mean – SD) (N = 60) (N = 78) (N = 146) (N = 284) Baseline 4.88 ± 9.10 0.24 ± 0.68 0.82 ± 2.76 1.54 ± 4.98 Week 4 2.29 ± 3.60 0.24 ± 0.59 0.61 ± 1.23 0.83 ± 1.98 Week 8 0.63 ± 2.35 0.12 ± 0.43 0.17 ± 0.61 0.25 ± 1.17 Week 12 0.36 ± 1.23 0.07 ± 0.38 0.05 ± 0.25 0.12 ± 0.64 Neurol Ther (2023) 12:303–317 311 Fig. 4 Proportion of patients with NEDA by longitudinal sNfL subgroup Fig. 5 Individual patient trajectories: A Patients with the refers to those visible on brain MRI scans. Additional ten lowest baseline sNfL levels and B patients with the ten lesions may have occurred outside the field of view, e.g., in highest baseline sNfL levels. Dotted line: baseline median the spinal cord. Gd ? gadolinium-enhancing; sNfL serum (9.1 pg/ml). The number of Gd ? T1 lesions annotated neurofilament light chain (Fig. 7). We note that the AUCs are compatible high-sensitivity range, which is of most interest within the estimated confidence intervals, and to identify patients at risk. Analysis that the ROC curves are particularly close in the 312 Neurol Ther (2023) 12:303–317 accounts further for sex was also considered, and the findings remain the same. DISCUSSION APLIOS was an open-label study with all patients initiating treatment with ofatumumab, which is known to substantially limit new focal MS disease activity [6]. In keeping with multiple prior studies reporting longitudinal reductions in sNfL following initiation of most approved MS treatments [4, 6, 11–16], ofatumumab treatment was associated with a consistent decline in sNfL levels from baseline up to 12 weeks [17]. Confirming prior studies [2–5], we observed that elevated sNfL levels could reflect and pre- dict the development of new disease activity at the group level. The high-frequency MRI and high-frequency sNfL sampling in this study provided the unique opportunity to assess the temporal association between sNfL and disease activity confirmed by subclinical newly devel- oping Gd ? T1 lesions. We observed that sNfL levels increased before new lesions were observed on monthly scheduled MRI scans, reached maximal levels over a period of few weeks, and then gradually decreased thereafter Fig. 6 Baseline sNfL level predicts risk of on-study disease to baseline. While one cannot be sure exactly activity: A All patients and B patients free of Gd ? T1 when a new Gd ? T1 lesion formed, our find- lesions at baseline. The stepped lines show Kaplan-Meier estimates of the cumulative proportion of patients with ings indicate that the half-life of sNfL is longer either confirmed clinical relapses or Gd ? T1 lesions by than that of the Gd ? T1 lesions that contribute time (days). In the figure, the steep increase in the to the rise in sNfL. These observations provide estimated proportion of patients with on-study confirmed new insights into the dynamic changes and clinical relapses or Gd ? T1 lesions at day 28 was evolution of neuroaxonal injury and resulting observed because day 28 is the day of the first-post sNfL measures associated with the development baseline MRI scan, uncovering the proportion of patients of new focal inflammatory lesions in patients with radiological disease activity. aThe Cox regression with RMS. They also indicate that sNfL assess- model included the baseline sNfL category as factor, age, ment may capture subclinical disease activity baseline number of Gd ? T1 lesions, baseline volume of that might otherwise be missed without fre- T2 lesions, baseline EDSS, and number of clinical relapses quent MRI scans. in previous year as continuous covariates. bThe Cox The unique frequent assessments of sNfL regression model included baseline sNfL category as factor levels and new disease activity in the relatively and age, baseline volume of T2 lesions, baseline EDSS and large number of patients monitored in this number of clinical relapses in the previous year as study showed that the proportion of patients continuous covariates. Gd ? gadolinium enhancing, HR with NEDA-3 was higher in the group whose hazard ratio, sNfL serum neurofilament light chain Neurol Ther (2023) 12:303–317 313 Fig. 7 ROC curves and AUC for the identification of on- predictor of whether there would be lesions on the study Gd ? T1 lesionsa; 9.1 pg/ml is the median baseline respective scan. All models were adjusted for age. AUCs sNfL concentration for all patients in the safety set whose from week 4, week 8, and week 12 scans were calculated baseline sNfL levels are available. aPrediction of whether a from tenfold cross-validation, and average AUC is the patient would have a Gd ? T1 lesion on the scan of the average across the three scans. For each scan and each specified week using logistic regressions based on the model, 10,000 bootstrap resampling on the predicted following linear predictors: Model ‘ baseline Gd ? T1’’ probabilities and the true labels is performed, and a (green) used the number of Gd ? T1 on the baseline scan residual bootstrap interval is obtained from the boot- as the predictor, model ‘ baseline sNfL’’ (blue) used the strapped AUCs. AUC area under the curve; Gd ? baseline sNfL measure ([ 9.1 pg/ml vs. B 9.1 pg/ml) as gadolinium-enhancing; MRI magnetic resonance imaging; predictor, and model ‘ time-matched sNfL’’ (red) used ROC AUC area under the receiver-operating characteristics time-matched sNfL (if not available, the last sNfL value curve; sNfL serum neurofilament light chain before the scan day) ([ 9.1 pg/ml vs. B 9.1 pg/ml) as the sNfL levels were constantly below the baseline population where patients were on highly effi- median throughout the study period compared cacious treatment such as ofatumumab. Time- to both patients with longitudinal sNfL levels matched sNfL (which utilizes the more frequent that crossed the baseline median on one or sampling) was also assessed but did not sub- more occasion, particularly those whose longi- stantially improve the prediction compared to tudinal sNfL levels were constantly above the baseline sNfL over the 12-week study period. baseline median. These results suggest that Although not better than baseline sNfL, the use persistently low longitudinal sNfL levels asso- of on-treatment sNfL did not decrease the per- ciate with NEDA-3. As expected, we also formance, suggesting the value of a more recent observed that baseline sNfL level showed prog- sNfL over a potentially distant ‘‘baseline,’’ e.g., nostic value for on-study disease activity (con- in cases where patients have been on treatment firmed clinical relapses or Gd ? T1 lesions) as for years. those patients who had a high baseline sNfL While at the group level low sNfL was asso- level were at significantly higher risk of devel- ciated with fewer Gd ? T1 lesions, variability oping on-study lesions or confirmed clinical was observed at the individual patient level. relapses than those who had a low baseline sNfL Some individual patient trajectories showed value. We further noted that sNfL measure- high baseline sNfL levels but no baseline or on- ments provided added value over MRI among study Gd ? T1 lesions, which could potentially patients who appeared free of Gd ? T1 lesions be explained by recently resolved lesions that based on MRI scan at baseline. developed relatively close to study entry. New At the individual patient level, the prognos- subclinical lesions and clinical relapses were tic value of baseline sNfL for on-study Gd ? T1 nevertheless more frequent over time in lesions over 12 weeks was comparable to that of patients with the highest compared to those the prognostic value of the number of Gd ? T1 with lowest baseline sNfL levels, indicating lesions on the baseline brain MRI scan, even in a prognostic potential of baseline sNfL even in a 314 Neurol Ther (2023) 12:303–317 population where patients were on highly effi- single-arm trial with a short study duration, cacious treatment. These findings may not be where treatment (ofatumumab) was highly surprising as high baseline sNfL is likely efficacious in suppressing lesion formation and indicative of a higher level of disease activity at reducing sNfL levels. Further findings from a baseline, and during the relatively short study larger independent cohort study could provide duration, patients with a high level of disease a more complete perspective. Currently, the activity at baseline are more likely to manifest research is being conducted with different new disease activity post-baseline. One notes assays and cut-offs and with different endpoints that patients enrolled in this study had to have in mind [22]. The present APLIOS study used documented disease activity sometime within the median sNfL value at baseline (9.1 pg/ml) as 2 years (see ‘‘Methods’’), and the half-life of NfL an analysis threshold to divide the recruited is likely at least 6–8 weeks [18]. High baseline RMS patient population into ‘high’ vs. ‘low’ sNfL levels might also reflect focal inflamma- groups (without the use of external controls). tory MS disease activity in regions not covered The data do not allow claims of optimality by brain MRI (e.g., the spinal cord), damage in regarding this choice of analysis threshold. A the normal-appearing white/gray matter, or threshold optimization should be performed in other sources such as brain trauma [19]. Future the context of an intended use of an in vitro studies may help to further understand the diagnostic device development for NfL to source of elevated NfL levels and their rela- ensure the utility of such a tool in clinical tionship to disease activity. practice. The present analysis, which extends previous studies [2, 6, 13, 20, 21], supports the utility of sNfL in monitoring subclinical disease activity CONCLUSIONS and points to the potential of individual sNfL measures to capture such activity in patients In conclusion, this study suggests that sNfL may with RMS, as determined by MRI. The analysis have utility in monitoring subclinical disease indicates that even in RMS patients without activity in RMS patients as shown by its tem- evidence of MRI or clinical disease activity, poral association with Gd ? T1 lesion activity, presence of high sNfL levels may capture the its overall comparable prognostic value to development of new disease activity as well as baseline MRI, and its added value over MRI in flag patients who may benefit from closer prognostic value of on-study confirmed clinical monitoring and/or reconsideration of treatment relapses or Gd ? T1 lesions in patients free of response. brain MRI disease activity. Therefore, when Together, our findings suggest that, in cer- frequent MRI scans are not feasible, sNfL may tain settings, sNfL measurements may have provide an attractive alternative to MRI if a utility in replacing or complementing standard standardized test becomes available to MRI monitoring of subclinical disease activity physicians. in individual patients. This may be particularly helpful when serial MRIs are prohibitively costly, burdensome, or otherwise inaccessible to ACKNOWLEDGEMENTS patients. The kinetics and evolution of sNfL trajectories that we observed in relation to new The authors acknowledge the patients, investi- focal inflammatory disease activity in individ- gators, and staff at participating sites for sup- ual patients over this 3-month study indicate porting the conduct of the study. that sNfL sampling on a quarterly basis may offer adequate monitoring and surveillance of Funding. The study was funded by Novartis subclinical disease activity (at least involving Pharma AG. The study sponsor participated in the brain). the design and conduct of the study, data col- Limitations of this post hoc analysis of data lection, data management, data analysis and from the APLIOS study include that this was a Neurol Ther (2023) 12:303–317 315 interpretation, and preparation, review, and for content, including medical writing for con- approval of the manuscript. The Novartis tent; Major role in the acquisition of data; Study Pharma AG also sponsored the publication of concept or design; Analysis or interpretation of this manuscript along with its associated rapid data. Petra Kukkaro: Drafting/revision of the service fee. manuscript for content, including medical writing for content; Major role in the acquisi- Medical Writing, Editorial, and Other tion of data; Study concept or design; Analysis Assistance. The authors thank Saimithra or interpretation of data. Neva Coello: Drafting/ Thammera and Anuja Shah (Novartis Health- revision of the manuscript for content, includ- care Pvt. Ltd., Hyderabad, India) for providing ing medical writing for content; Major role in medical writing support, which encompassed the acquisition of data; Study concept or design; writing of the manuscript, formatting, refer- Analysis or interpretation of data. Inga Ludwig: encing, preparation of tables and figures as per Drafting/revision of the manuscript for content, journal guidelines, and incorporating the including medical writing for content; Major authors’ revisions and finalizing the draft for role in the acquisition of data; Study concept or submission, all under the direction of the design; Analysis or interpretation of data. authors. All authors edited the manuscript for Roman Willi: Drafting/revision of the manu- intellectual content, provided guidance during script for content, including medical writing for manuscript development and approved the content; Major role in the acquisition of data; final version submitted for publication. Study concept or design; Analysis or interpre- tation of data. Martin Zalesak: Drafting/revision Data Availability. The datasets generated of the manuscript for content, including medi- during and/or analyzed during the current cal writing for content; Major role in the study are available from the corresponding acquisition of data; Study concept or design; author on reasonable request. Analysis or interpretation of data. Krishnan Ramanathan: Drafting/revision of the manu- Authorship. All named authors meet the script for content, including medical writing for International Committee of Medical Journal content; Major role in the acquisition of data; Editors (ICMJE) criteria for authorship for this Study concept or design; Analysis or interpre- manuscript, take responsibility for the integrity tation of data. Bernd Kieseier: Drafting/revision of the work, and have given final approval to of the manuscript for content, including medi- the version to be published. All authors are cal writing for content; Major role in the responsible for intellectual content and data acquisition of data; Study concept or design; accuracy. Analysis or interpretation of data. Dieter A. Häring: Drafting/revision of the manuscript for Author Contributions. Amit Bar-Or: Draft- content, including medical writing for content; ing/revision of the manuscript for content, Major role in the acquisition of data; Study including medical writing for content; Major concept or design; Analysis or interpretation of role in the acquisition of data; Study concept or data. Morten Bagger: Drafting/revision of the design; Analysis or interpretation of data. manuscript for content, including medical Xavier Montalban: Drafting/revision of the writing for content; Major role in the acquisi- manuscript for content, including medical tion of data; Study concept or design; Analysis writing for content; Major role in the acquisi- or interpretation of data. Edward Fox: Drafting/ tion of data; Study concept or design; Analysis revision of the manuscript for content, includ- or interpretation of data. Xixi Hu: Drafting/re- ing medical writing for content; Major role in vision of the manuscript for content, including the acquisition of data; Study concept or design; medical writing for content; Major role in the Analysis or interpretation of data. acquisition of data; Study concept or design; Analysis or interpretation of data. Harald Disclosures. Amit Bar-Or has participated as Kropshofer: Drafting/revision of the manuscript a speaker in meetings sponsored by and 316 Neurol Ther (2023) 12:303–317 received consulting fees and/or grant support written informed consent before commencing from: Janssen/Actelion; Atara Biotherapeutics, trial-related procedures. This article is based on Biogen Idec, Celgene/Receptos, Roche/Genen- previously conducted studies and does not tech, Medimmune, Merck/EMD Serono, Novar- contain any new studies with human partici- tis, Sanofi-Genzyme. Xavier Montalban has pants or animals performed by any of the received speaking honoraria and travel expenses authors. for participation in scientific meetings, has been a steering committee member of clinical trials Open Access. This article is licensed under a or participated in advisory boards of clinical Creative Commons Attribution-NonCommer- trials in the past years with Actelion, Alexion, cial 4.0 International License, which permits Bayer, Biogen, Celgene, EMD Serono, Genzyme, any non-commercial use, sharing, adaptation, Immunic, Medday, Merck, Mylan, Nervgen, distribution and reproduction in any medium Novartis, Roche, Sanofi-Genzyme, Teva Phar- or format, as long as you give appropriate credit maceutical, TG Therapeutics, Excemed, MSIF to the original author(s) and the source, provide and NMSS. Dr. Fox has received compensation a link to the Creative Commons licence, and for research, Consulting, Speakers’ Bureau, and/ indicate if changes were made. The images or or Advisory work from AbbVie, Alexion, Biogen, other third party material in this article are Bristol-Myers Squibb, Chugai, EMD Serono, included in the article’s Creative Commons Genentech Roche, Novartis, Sanofi Genzyme, licence, unless indicated otherwise in a credit and TG Therapeutics. Harald Kropshofer was an line to the material. If material is not included employee of Novartis during the analysis of this in the article’s Creative Commons licence and study until final version of paper development. your intended use is not permitted by statutory But he sadly passed away in May 2022 and regulation or exceeds the permitted use, you authors wish to acknowledge his significant will need to obtain permission directly from the contribution to his work via posthumous copyright holder. To view a copy of this licence, authorship. Krishnan Ramanathan was an visit http://creativecommons.org/licenses/by- employee of Novartis during the analysis of this nc/4.0/. study until final version of paper development. Roman Willi, Petra Kukkaro, Inga Ludwig, Martin Zalesak, Dieter A. Häring, Xixi Hu, Neva Coello, and Morten Bagger are employees of REFERENCES Novartis. 1. Joy JE, Johnston RB Jr, editors. Committee on Prior Presentation. Some results from this multiple sclerosis: current status and strategies for the future, board on neuroscience and behavioral study were summarized in one research abstract health. Washington: National Academies Press and conference posters, presented at the (US); 2001. MSVirtual2020 (the 8th Joint Congress of Americas Committee for Treatment and 2. Kuhle J, Kropshofer H, Haering DA, et al. Blood neurofilament light chain as a biomarker of MS Research in Multiple Sclerosis and the European disease activity and treatment response. Neurology. Committee for Treatment and Research in 2019;92(10):e1007–15. Multiple Sclerosis), 11–13 Sep 2020. 3. Berger T, Stuve O. Neurofilament light chain: an Compliance with Ethics Guidelines. The important step toward a disease biomarker in mul- tiple sclerosis. Neurology. 2019;92:1–2. study was conducted in accordance with the International Conference on Harmonisation 4. Disanto G, Barro C, Benkert P, et al. Serum Neuro- guidelines for Good Clinical Practice [9] and the filament light: a biomarker of neuronal damage in principles of the Declaration of Helsinki [10]. multiple sclerosis. Ann Neurol. 2017;81:857–70. The protocol was approved by an independent 5. Haering DA, Kropshofer H, Kappos L, et al. Long- ethics committee or institutional review board term prognostic value of longitudinal for each study site, and all patients provided Neurol Ther (2023) 12:303–317 317 measurements of blood neurofilament levels. Neu- 14. de Flon P, Laurell K, Sundstrom P, et al. Compar- rol Neuroimmunol Neuroinflamm. 2020;7: e856. ison of plasma and cerebrospinal fluid neurofila- ment light in a multiple sclerosis trial. Acta Neurol 6. Hauser SL, Bar-Or A, Cohen JA, et al. Ofatumumab Scand. 2019;139(5):462–8. versus teriflunomide in multiple sclerosis. N Engl J Med. 2020;383:546–57. 15. Cross A. Ocrelizumab treatment reduced levels of neurofilament light chain and numbers of B cells in 7. Bar-Or A, Wiendl H, Montalban X, et al. Rapid and the cerebrospinal fluid of patients with relapsing sustained B-cell depletion with subcutaneous ofa- multiple sclerosis in the OBOE study S56.008. tumumab in relapsing multiple sclerosis: APLIOS, a Neurology. 2019;92(Supplement 15):52. randomized phase-2 study. Mult Scler. 2021. https://doi.org/10.1177/13524585211044479. 16. Thebault S, Tessier D, Lee H, et al. High serum neurofilament light chain normalises after 8. Jones B, et al. Statistical methodology for highly haematopoietic stem cell transplant for MS. Neurol variable compounds: a novel design approach for Neuroimmunol Neuroinflamm. 2019;6(5):e598. the ofatumumab phase 2 bioequivalence study. Pharm Stat. 2022. https://doi.org/10.1002/pst.2233. 17. Data on file. Ofatumumab (OMB157G). Exploratory analysis. COMB157G2102. Adjusted geometric 9. Dixon JR Jr. The International Conference on Har- mean NfL concentrations over time. Novartis monization Good Clinical Practice guideline. Qual Pharma AG; 2022. Assur. 1998;6(2):65–74. 18. Bergman J, Dring A, Zetterberg H, et al. Neurofila- 10. Declaration of Helsinki: ethical principles for med- ment light in CSF and serum is a sensitive marker ical research involving human subjects. Fortaleza: for axonal white matter injury in MS. Neurol Neu- World Medical Association; 2013. p. 388. roimmunol Neuroinflamm. 2016;3(5): e271. 11. Akgün K, Kretschmann N, Haase R, et al. Profiling 19. McKee AC, Daneshwar DH. The neuropathology of individual clinical responses by high-frequency traumatic brain injury. Handb Clin Neurol. serum neurofilament assessment in MS. Neurol 2015;127:45–66. Neuroimmunol Neuroinflamm. 2019;6(3): e555. 20. Novakova L, Zetterberg H, Sundström P, et al. 12. Sejbaek T, Nielsen HH, Penner N, et al. Dimethyl Monitoring disease activity in multiple sclerosis fumarate decreases neurofilament light chain in using serum neurofilament light protein. Neurol- CSF and blood of treatment naı̈ve relapsing MS ogy. 2017;89(22):2230–7. patients. J Neurol Neurosurg Psychiatry. 2019;90(12):1324–30. 21. Varhaug KN, Barro C, Bjørnevik K, et al. Neurofila- ment light chain predicts disease activity in 13. Jakimovski D, Zivadinov R, Ramanthan M, et al. relapsing-remitting MS. Neurol Neuroimmunol Serum neurofilament light chain level associations NeuroInflamm. 2018;5(1): e422. with clinical and cognitive performance in multiple sclerosis: a longitudinal retrospective 5-year study. 22. Bittner S, Oh J, Havrdová EK, Tintoré M, Zipp F. The Mult Scler J. 2019;26:1670–81. potential of serum neurofilament as biomarker for multiple sclerosis. Brain. 2021. https://doi.org/10. 1093/brain/awab241.