RESEARCH ARTICLE Associations of sNfL with clinico-radiological measures in a large MS population Elias S. Sotirchos1,a , Kathryn C. Fitzgerald1,a , Carol M. Singh2, Matthew D. Smith1, Maria Reyes-Mantilla1, Carrie M. Hersh3, Megan H. Hyland4, Ryan Canissario4, Sarah B. Simmons5, Georgina Arrambide6, Xavier Montalban6, Manuel Comabella6 , Robert T. Naismith7, Min Qiao7, Lauren B. Krupp8 , Jacqueline A. Nicholas9, Katja Akgu € n10, Tjalf Ziemssen10 , Richard Rudick11, Elizabeth Fisher , Robert A. Bermel , Ellen M. Mowry & Peter A. Calabresi1 2 5 1 1 Department of Neurology, Johns Hopkins University School of Medicine, Baltimore, Maryland, USA 2 Biogen, Cambridge, Massachusetts, USA 3 Lou Ruvo Center for Brain Health, Cleveland Clinic, Las Vegas, Nevada, USA 4 Department of Neurology, University of Rochester Medical Center, Rochester, New York, USA 5 Mellen Center, Neurological Institute, Cleveland Clinic, Cleveland, Ohio, USA 6 Department of Neurology and Centre d’Esclerosi Mu noma de Barcelona, ltiple de Catalunya, Vall d’Hebron Hospital Universitari, Universitat Auto Barcelona, Spain 7 Department of Neurology, Washington University in St. Louis, St. Louis, Missouri, USA 8 Department of Neurology, New York University, New York City, New York, USA 9 OhioHealth Multiple Sclerosis Center, Riverside Methodist Hospital, Columbus, Ohio, USA 10 Center of Clinical Neuroscience, Department of Neurology, University Clinic Carl-Gustav Carus, Dresden, Germany 11 (formerly) Biogen, Cambridge, Massachusetts, USA Correspondence Abstract Elias S. Sotirchos and Peter A. Calabresi, Department of Neurology, Johns Hopkins Objective: Evaluation of serum neurofilament light chain (sNfL), measured Hospital, 600 N. Wolfe St., Pathology 627, using high-throughput assays on widely accessible platforms in large, real-world Baltimore, MD 21287, USA. Tel: +1 (410) MS populations, is a critical step for sNfL to be utilized in clinical practice. 614-1522; Fax: +1 410-502-6736; E-mail: Methods: Multiple Sclerosis Partners Advancing Technology and Health Solu- ess@jhmi.edu and calabresi@jhmi.edu tions (MS PATHS) is a network of healthcare institutions in the United States and Europe collecting standardized clinical/imaging data and biospecimens dur- Received: 19 September 2022; Revised: 3 November 2022; Accepted: 8 November ing routine clinic visits. sNfL was measured in 6974 MS and 201 healthy con- 2022 trol (HC) participants, using a high-throughput, scalable immunoassay. Results: Elevated sNfL levels for age (sNfL-E) were found in 1238 MS partici- Annals of Clinical and Translational pants (17.8%). Factors associated with sNfL-E included male sex, younger age, Neurology 2023; 10(1): 84–97 progressive disease subtype, diabetes mellitus, impaired renal function, and active smoking. Higher body mass index (BMI) was associated with lower odds doi: 10.1002/acn3.51704 of elevated sNfL. Active treatment with disease-modifying therapy was associ- a Joint first authors. ated with lower odds of sNfL-E. MS participants with sNfL-E exhibited worse neurological function (patient-reported disability, walking speed, manual dex- terity, and cognitive processing speed), lower brain parenchymal fraction, and higher T2 lesion volume. Longitudinal analyses revealed accelerated short-term rates of whole brain atrophy in sNfL-E participants and higher odds of new T2 lesion development, although both MS participants with or without sNfL-E exhibited faster rates of whole brain atrophy compared to HC. Findings were consistent in analyses examining age-normative sNfL Z-scores as a continuous variable. Interpretation: Elevated sNfL is associated with clinical disability, inflammatory disease activity, and whole brain atrophy in MS, but interpreta- tion needs to account for comorbidities including impaired renal function, diabetes, and smoking. 84 ª 2022 The Authors. Annals of Clinical and Translational Neurology published by Wiley Periodicals LLC on behalf of American Neurological Association. This is an open access article under the terms of the Creative Commons Attribution License, which permits use, distribution and reproduction in any medium, provided the original work is properly cited. E. S. Sotirchos et al. sNfL in MS PATHS with demographics, comorbid conditions, MS clinical char- Introduction acteristics and clinico-radiological outcomes, in a large, Multiple sclerosis (MS) is an immune-mediated, inflamma- international, real-world, multi-center population of PwMS. tory disorder of the central nervous system. Despite being classically considered a demyelinating disorder, neuro- Methods axonal injury occurs early in the disease course and repre- sents the pathologic substrate for permanent neurological Study participants disability in people with MS (PwMS).1 In clinical practice, disease monitoring in PwMS is performed by clinical evalu- Data for this study were collected as part of the Multiple ation and use of conventional magnetic resonance imaging Sclerosis Partners Advancing Technology and Health (MRI) measures, including new T2 lesions and/or presence Solutions (MS PATHS) network between November 2016 of T1 post-gadolinium (Gd)-enhancing lesions. Notably, and May 2021. The design of MS PATHS has been previ- these conventional MRI measures assess for the presence of ously described in detail.15 Briefly, MS PATHS is an inflammatory disease activity rather than neuro-axonal ongoing initiative conducted in 10 MS centers (seven in injury, and biomarkers identifying the latter are an impor- the United States, three in Europe), that have standard- tant unmet need in MS and other neurological conditions. ized elements of their clinical practice and collaborated Neurofilaments are neuron-specific cytoskeletal proteins with Biogen to implement a centralized health informa- that are released into the extracellular space following tion exchange architecture for research purposes. MS neuro-axonal damage, and have thus been proposed as PATHS was designed around the concept of a learning putative biomarkers of neuro-axonal injury in multiple health system, merging research with ongoing patient care neurological diseases, including MS.2 Neurofilament light by collecting standardized data during routine clinic vis- chain (NfL) has particularly been shown to be a promis- its. Enrollment in MS PATHS is open to individuals with ing biomarker because of its high solubility. The a physician-confirmed diagnosis of MS (including clini- application of newer immunoassays has enabled the mea- cally isolated syndrome [CIS]) seeking care at participat- surement of the low concentrations of NfL in blood (usu- ing institutions. Additionally, healthy controls (HC) have ally serum) with high accuracy and reproducibility, and been recruited at participating healthcare institutions to increased NfL levels have been found in blood in several be demographically representative of the MS cohort with neurological disorders with underlying neuro-axonal respect to age, sex, and race. Pertinent exclusion criteria degeneration, including MS.3,4 for HC included: (1) pregnancy, (2) diagnosis of migraine In MS, there is evidence that serum NfL (sNfL) levels requiring medication or any other subject-reported diag- correlate with CSF NfL, are associated with clinico- nosis of neurological disease or condition, and (3) any radiological inflammatory disease activity, disability subject-reported diagnosis of an autoimmune disorder. progression, and brain atrophy, and are modulated by All participants provided written informed consent to col- disease-modifying therapies (DMTs).3,5–9 However, the lect and share pseudo-anonymized data; the study was vast majority of studies in which sNfL has been studied in approved by the institutional review board of each partic- MS have been single-center studies or post hoc analyses of ipating institution. clinical trial cohorts, which may lack generalizability. Furthermore, while it is clear that age is a very important MS performance test factor that needs to be accounted for when interpreting sNfL levels, the effect of comorbidity or varying MS charac- A full list of data elements collected as part of MS PATHS teristics on sNfL in MS has not been rigorously character- has previously been reported.15 Briefly, MS PATHS incor- ized at scale. Importantly, data support that factors porates a self-administered iPad-based patient assessment including body mass index (BMI), diabetes mellitus, and tool, the MS Performance Test (MSPT), to collect a struc- impaired renal function may influence sNfL levels.9–13 tured history, patient-reported outcomes, and an assess- Additionally, most existing studies have been performed ment of neuroperformance, using an adaptation of the using single molecule array (Simoa), which is run on a MS Functional Composite (MSFC) to assess walking research platform. In order for sNfL testing to be more speed, manual dexterity, and cognition.16 Specific clinical readily implemented in the clinical arena, evaluation of disability outcomes of interest considered in this study high-throughput assays on routine clinical laboratory plat- include: (1) Patient Determined Disease Steps (PDDS), forms is necessary.14 (2) a 25-foot walking speed test, similar to the Timed 25- In this study, we sought to evaluate associations of sNfL, Foot Walk, (3) a manual dexterity test, similar to the 9- measured using a high-throughput scalable immunoassay, Hole Peg Test, and (4) a processing speed test, similar to ª 2022 The Authors. Annals of Clinical and Translational Neurology published by Wiley Periodicals LLC on behalf of American Neurological Association. 85 sNfL in MS PATHS E. S. Sotirchos et al. the Symbol Digit Modalities Test. The PDDS is a patient- and stored at 80°C prior to thawing and subsequent anal- reported outcome, adapted from the clinician-assessed ysis. sNfL values obtained by the two methods exhibited Disease Steps, which has previously been shown to exhibit good correlation and agreement (r = 0.85, ICC 0.87; Fig. 1). excellent reproducibility, to be sensitive to disability pro- gression, and to correlate well both cross-sectionally and Magnetic resonance imaging longitudinally with the Expanded Disability Status Scale (EDSS).17 The electronic adaptation of the MSFC has Standardized MRI protocols are utilized in MS PATHS to similarly been validated against the analogous technician- acquire brain imaging on Siemens 3 T Scanners, as previ- administered tests and has been shown to exhibit excel- ously described.15 Brain MRIs are acquired as clinically lent correlations in PwMS.16,18,19 Additional data ele- indicated, generally occurring annually for monitoring ments, including medical history, medication lists, and purposes. Radiology report data are extracted from the physical/laboratory assessments are automatically EMR, including the number of T1 post-Gd-enhancing extracted from the electronic medical record (EMR). lesions (if a post-contrast scan was acquired) and the number of new T2 lesions since the prior MRI (if a com- parable scan was available), as read by the local radiolo- Biospecimen collection and sNfL gists at each institution. Quantitative MRI metrics are measurement also derived from brain MRIs as part of MS PATHS, All participants in MS PATHS are also offered the option including brain parenchymal fraction (BPF) and T2 lesion at each clinic visit to provide blood samples for a bio- volume. These values are automatically calculated utilizing banking sub-study. Blood samples for the MS PATHS a software prototype developed jointly by Biogen and Sie- biobanking sub-study are collected at each study site, mens, MSPie (MS PATHS Image Evaluation).23 BPF is a according to a standardized MS PATHS protocol.15 normalized measure of whole brain volume, calculated Briefly, for serum collection, blood is collected by using a combination approach based on segmentation of venipuncture in serum separator tubes (SST), mixed by the brain parenchyma and total intracranial volume in inversion, and allowed to clot for ~30 min at room tem- the 3D FLAIR and 3D T1 images. Whole brain atrophy is perature in an upright position. Within 1 h of collection, calculated as the difference in BPF values from two time SST are centrifuged for 15 min at ~1500 g. Following points. In analyses of a scan–rescan substudy of MS- centrifugation, the serum is aliquoted and frozen immedi- PATHS including three MS-PATHS sites at which 30 MS ately at 70°C (10°C). If a 70°C freezer is not avail- patients underwent four MRIs on two different Siemens able, samples may be frozen and stored at 20°C until 3 T MRI scanners within 1 week, MSPie BPF estimated shipped. Samples are shipped on dry ice weekly to central by MSPie exhibited excellent reproducibility, with a mean laboratories for long-term storage at 70°C. coefficient of variation of 0.18%. T2 lesions are seg- Serum samples for all MS PATHS biobanking sub-study mented automatically in MSPie based on 3D FLAIR and participants from the first available timepoint were 3D T1 images using the LeMan-PV algorithm.24 retrieved for sNfL measurements. Samples were shipped to Siemens Healthcare Laboratory, LLC (Berkeley, CA, USA) Statistical methods on dry ice and upon receipt, samples were stored at 80°C. Prior to analysis, samples were thawed at room tempera- Statistical analyses were performed with R Version 4.1.2 ture and measurements were performed on the Atellica (https://www.r-project.org/). We used sNfL measurements Solution platform, using a high-throughput scalable from the MS PATHS HC cohort to derive age-normative acridinium-ester immunoassay, which has a range of 3.9 to Z-scores using Generalized Additive Models for Location, 500 pg/mL and excellent repeatability.14,20,21 All runs Scale, and Shape (GAMLSS).5,25,26 We selected the Box- included low (~7 pg/mL), medium (~40 pg/mL), and high Cox Cole and Green distribution family as this minimized (340 pg/mL) quality control samples to ensure assay stabil- the Generalized Akaike information criterion. Analyses ity over time, with acceptable precision (coefficient of vari- were performed using the age-normative sNfL Z-score as a ation 10.2%, 5.4%, and 7.3%, respectively) and no continuous variable (truncated at the 99.9th percentile), as significant change over time. Furthermore, sNfL measure- well as using a pre-defined Z-score cut-off of 1.96 (97.5th ment in different serum aliquots from the same blood centile curve) to define “elevated” (sNfL-E) versus “nor- draws was performed previously on a Simoa HD-1 Ana- mal” sNfL (sNfL-N), as this is the typical approach used to lyzer for a subset of participants (n = 2143), using the define reference ranges for clinical laboratory tests.27 sNfL Simoa NfL Advantage Assay Kit.22 For the Simoa sNfL values below the lower limit of quantification (LLoQ; measurement, the samples were similarly shipped on dry 3.9 pg/mL) were replaced by the LLoQ (n = 18 MS samples ice to a central laboratory (Quanterix; Billerica, MA, USA) < LLoQ; none of the HC samples). 86 ª 2022 The Authors. Annals of Clinical and Translational Neurology published by Wiley Periodicals LLC on behalf of American Neurological Association. E. S. Sotirchos et al. sNfL in MS PATHS Figure 1. sNfL measured on the Siemens Atellica Platform versus the Quanterix Simoa Platform. Plot of sNfL levels (pg/mL) measured on the Siemens Atellica (y-axis) and Quanterix Simoa platforms (x-axis) in n = 2143 participants. The overlaid red line corresponds to the ordinary least squares regression line fit. The corresponding marginal densities for each variable are shown in the plot margins. ICC, intra-class correlation coeffi- cient; CI, confidence intervals. Furthermore, we also derived age-normative sNfL Z- <18.5 (“underweight”), 18.5–24.0 (“reference”), 25.0–29.9 scores using a similar approach, using sNfL measured in (“overweight”), 30.0–39.9 (“obese”), and ≥ 40 (“morbidly the National Health and Nutrition Examination Survey obese”). DMT use was categorized as “IFN-b/GA” (NHANES), as reported in detail previously.13 Given that (interferon-beta and glatiramer acetate), oral (terifluno- the previously reported NHANES derived model include mide, sphingosine-1-phosphate inhibitors, and fumaric age, creatinine, and glycosylated hemoglobin (HbA1c), acid esters), “infusion or immune reconstitution therapy” with creatinine available for a subset of our cohort and (anti-CD20 agents, natalizumab, alemtuzumab, and HbA1c not available for the vast majority of our cohort, cladribine), “other or not listed,” and “untreated.” Miss- derived age-normative curves assume a creatinine of ing indicator levels were used in multivariable analyses 0.79 mg/dL (median value of MS PATHS participants for variables available only for a subset of the cohort with available creatinine measurements [n = 4667] with- (listed in Table 1). While we considered alternative out evidence of chronic kidney disease [i.e., estimated approaches for missing data including imputation or glomerular filtration rate > 60 mL/min/1.73 m2]) and complete case analysis, some key variables were unlikely HbA1c of 5.22% (median value of non-diabetics in to be reliably imputed using available predictors (e.g., NHANES).13 eGFR without a creatinine measurement), and complete Age and disease duration were categorized each into case analysis generally reduces the effective sample size. five groups based on quintile splits. Estimated glomerular Furthermore, given that race is largely a social construct filtration rate (eGFR) was calculated using the creatinine and MS PATHS includes sites across the United States value (if available) closest to the date of blood sampling and Europe, race in this study may be a surrogate of dif- for sNfL measurement, using the CKD-EPI 2021 equa- fering societal dynamics. Therefore, we report results tion incorporating age and sex, omitting race, and was without the inclusion of race as an independent variable. categorized as >90 (“normal”), 60–90 (“mildly impaired), Secondary analyses considered race with similar results and <60 mL/min/1.73 m2 (“moderately to severely observed. Using the PDDS, self-reported disability was impaired”).28 Body mass index (BMI) was categorized as categorized as “mild” (no gait impairment), “moderate” ª 2022 The Authors. Annals of Clinical and Translational Neurology published by Wiley Periodicals LLC on behalf of American Neurological Association. 87 sNfL in MS PATHS E. S. Sotirchos et al. Table 1. Baseline demographics and clinical characteristics of the MS participants. Overall (N = 6974) sNfL-N, (N = 5736) sNfL-E (N = 1238) Age (years), mean (SD) 46.9 (12.0) 47.2 (11.5) 45.5 (14.2) Female, n (%) 5017 (72%) 4149 (72%) 868 (70%) Race, n (%) White 5751 (82%) 4804 (84%) 947 (76%) Black 501 (7.2%) 360 (6.3%) 141 (11%) Other 332 (4.8%) 265 (4.6%) 67 (5.4%) sNfL (pg/mL), median (IQR) 11.1 (8.4, 14.8) 10.1 (8.0, 12.8) 21.6 (16.6, 28.9) Brain parenchymal fraction, mean (SD) 0.851 (0.025) 0.851 (0.025) 0.848 (0.028) T2 lesion Volume (mL), median (IQR) 6.9 (3.2, 14.9) 6.3 (3.0, 13.7) 10.4 (4.9, 22.3) Disease duration (years), median (IQR) 12.2 (5.9, 20.5) 12.4 (6.1, 20.4) 11.4 (4.7, 21.1) MS subtype, n (%) CIS/RRMS 4558 (65%) 3866 (67%) 692 (56%) Progressive MS 2059 (30%) 1597 (28%) 462 (37%) Self-reported disability status, n (%) Mild 3860 (59%) 3308 (61%) 552 (48%) Moderate 1295 (20%) 1052 (19%) 243 (21%) Severe 1390 (21%) 1037 (19%) 353 (31%) DMT class None 940 (13%) 710 (12%) 230 (19%) IFN-b/GA1 1373 (20%) 1153 (20%) 220 (18%) Oral1 2213 (32%) 1921 (33%) 292 (24%) Infusion/IRT1 1729 (25%) 1373 (24%) 356 (29%) Other2 or Unknown 719 (10%) 579 (10%) 140 (11%) eGFR (mL/min/1.73 m2), n (%) >90 3342 (48%) 2743 (48%) 599 (48%) 60–89 1226 (18%) 1003 (17%) 223 (18%) <60 99 (1.4%) 50 (0.9%) 49 (4.0%) Body mass index, n (%) <18.5 105 (1.5%) 72 (1.3%) 33 (2.7%) 18.5–24.9 2213 (32%) 1768 (31%) 445 (36%) 25–29.9 1876 (27%) 1574 (27%) 302 (24%) 30–39.9 1830 (26%) 1521 (27%) 309 (25%) >40 427 (6.1%) 352 (6.1%) 75 (6.1%) Diabetes mellitus, n (%) 434 (6.2%) 321 (5.6%) 113 (9.1%) Smoking status, n (%) Non-smoker 5676 (81%) 4685 (82%) 991 (80%) Current smoker 1207 (17%) 975 (17%) 232 (19%) Characteristics available for a subset of the cohort included: race (n = 6584), disease subtype (n = 6627), disease duration (n = 6355), PDDS (n = 6545), DMT class (n = 6255), eGFR (n = 4667), body mass index (n = 6451), diabetes status (5606), smoking status (n = 6883), and brain parenchymal fraction/T2 lesion volume (within 365 days of blood sampling; n = 3514). sNfL-E, elevated sNfL; sNfL-N, “normal” sNfL; SD, standard deviation; IQR, interquartile range; CIS, clinically isolated syndrome; RRMS, relapsing– remitting MS; PDDS, Patient Determined Disease Steps; DMT, disease-modifying therapy; IFN-beta, interferon-beta; GA, glatiramer acetate; IRT, immune reconstitution therapy; eGFR, estimated glomerular filtration rate. 1 IFN-b/GA (IFN-b n = 611; GA n = 762), Oral (Fumarate n = 1054; S1P inhibitor n = 874; Teriflunomide n = 285), Infusion/IRT (anti-CD20 n = 943; Natalizumab n = 684; Alemtuzumab n = 89; Daclizumab n = 10; Cladribine n = 3). 2 A small proportion of participants (n = 47) reported use of oral immunosuppressive medications (including mycophenolate, methotrexate, cyclos- porine, or cyclophosphamide) or intravenous immune globulin (n = 7). (early gait impairment), and “severe” (assistive device for on the distribution in the MS participants) in order to ambulation or non-ambulatory).17 Neuroperformance facilitate interpretation of the results. domains were transformed to Z-scores using regression- Assessments of the associations of sNfL-E with the based equations derived from a study of adult healthy various patient demographics, comorbidity, and clinical volunteers (n = 517).29 T2 lesion volume (log- characteristics were performed using logistic regression; transformed) and BPF were converted to Z-scores (based similar models considered age-normative sNfL Z-scores 88 ª 2022 The Authors. Annals of Clinical and Translational Neurology published by Wiley Periodicals LLC on behalf of American Neurological Association. E. S. Sotirchos et al. sNfL in MS PATHS Figure 2. Serum neurofilament chain by age in the MS participants. Distribution of sNfL in the MS participants by age. The entire distribution is shown in the inset plot in the upper-right corner, with the larger plot demonstrating the distribution in the range up to 100 pg/mL (red box in the inset plot). The overlaid line corresponds to the age-normative 97.5th percentile curve derived from the healthy controls, with sNfL levels below the line (purple) and above the line (green) being categorized as “normal” (sNfL-N) and “elevated” (sNfL-E), respectively. using linear regression. Neuroperformance measures and Results radiological outcomes were compared between sNfL groups and their associations with age-normative sNfL Study population Z-scores were assessed in generalized linear models adjusted for age, sex, BMI, disease duration, disease sub- Serum samples were available for 7246 PwMS and 201 type, DMT class, eGFR category, diabetes mellitus, and HC. Samples from 272 MS participants were excluded smoking status (covariates selected based on the findings due to insufficient quantity of serum in the aliquots in the baseline, cross-sectional analyses). Self-reported (n = 204) or excessive hemolysis (n = 68), leaving 6974 disability status was compared between sNfL groups MS participants with samples available for sNfL measure- using multinomial regression and longitudinal analyses ment. Characteristics of the MS participants and HC are of BPF (dependent variable) were performed using shown in Table 1 and Table S1, respectively. Participant mixed-effects models including time, sNfL group, and recruitment by site is shown in Table S2. the cross-product of time and group, as well as random intercepts and random slopes in time to account for Associations of sNfL with demographics, within patient correlation, and were similarly adjusted, clinical characteristics, and comorbidities as previously described.30 Analyses were also similarly performed including the As expected, sNfL levels increased with age in the HC age-normative sNfL Z-score as a continuous variable, using (Fig. S1). Elevated sNfL levels above the age-normative restricted cubic splines in order to assess for non-linear 97.5th percentile (derived from the HC) were present at relationships; restricted cubic spline models were fit with baseline in 1238 MS participants (17.8%; Fig. 2 and four knots, placed at the 5th, 35th, 65th, and 95th percentile Table 1). Interestingly, while the NHANES-derived age- of the age-normative sNfL Z-score distribution.31 Linear normative sNfL Z-scores for the MS participants demon- versus spline models were formally compared using likeli- strated excellent correlations with those derived using the hood ratio tests; spline models are reported if deviation MS PATHS HC data (Fig. 3A), their distribution was from linearity was detected (P < 0.05).31,32 similar to the NHANES reference population (Fig. 3B) ª 2022 The Authors. Annals of Clinical and Translational Neurology published by Wiley Periodicals LLC on behalf of American Neurological Association. 89 sNfL in MS PATHS E. S. Sotirchos et al. Figure 3. Associations between age-normative Z-scores derived using the MS PATHS HC vs NHANES. Panel A depicts the correlation of the MS participant age-normative sNfL Z-scores derived using the MS PATHS HC sNfL data (x-axis) versus age-normative Z-scores derived using the NHANES reference population (y-axis). Panel B depicts the distribution of the age-normative sNfL Z-scores in the MS population, normalized for age using MS PATHS HC (purple) or NHANES (green). Notably, while it is clear that the distribution of the MS PATHS HC-derived Z-scores deviate compared to the reference population (standardized to have a mean of zero and standard deviation of one in the MS PATHS HC), the distribution of the NHANES-derived Z-scores is similar to that of the reference population (i.e., the MS participants in MS PATHS do not appear to have abnor- mally elevated sNfL Z-scores compared to the NHANES reference population). and only 199 MS participants (2.9%) had sNfL levels Baseline associations of sNfL with clinical above the NHANES-derived age-normative 97.5th per- disability and radiological measures centile. Initial analyses assessed the association of demographic Cross-sectional associations of sNfL group or MSPATHS characteristics, clinical characteristics, and comorbidities HC-derived age-normative sNfL Z-scores with neuroper- with the presence of elevated sNfL in MS (Table 2). Fac- formance measures and MRI volumetrics are shown in tors associated with increased odds of elevated sNfL Table 3. Clinical disability was worse in those with sNfL- included male sex (adjusted OR: 1.20, 95% CI: 1.04 to E compared to those with sNfL-N, as evidenced by higher 1.39, P = 0.012), progressive disease subtype (adjusted self-reported disability (adjusted OR—moderate vs mild OR: 1.63, 1.41 to 1.88, P < 0.001), diabetes mellitus (ad- disability: 1.39, 95% CI: 1.15 to 1.67, P < 0.001; severe vs justed OR: 1.67, 95% CI: 1.31 to 2.12, P < 0.001), current mild disability: 2.26, 95% CI: 1.85 to 2.75, P < 0.001) smoking (adjusted OR: 1.21, 95% CI: 1.02 to 1.43, and worse neuroperformance (adjusted difference in Z- P = 0.029), and moderately to severely decreased eGFR scores—walking speed: 0.54, 95% CI: 0.80 to 0.28; (adjusted OR 5.21; 95% CI: 3.39 to 8.0, P < 0.001). Over- manual dexterity: 0.45, 95% CI 0.58 to 0.33; pro- all, higher age was associated with lower odds of elevated cessing speed: 0.30, 95% CI: 0.38 to 0.22; P < 0.001 sNfL compared to younger patients (reference age group: for all). 18–38 years). Individuals with short (<5 years) or long Furthermore, in those with available brain MRI within (>23 years) disease duration had higher odds of elevated 1 year of blood sampling (n = 3514; median absolute sNfL compared to other disease duration categories. time between blood sampling and MRI: 96 days, IQR: 23 Increasing BMI was associated with lower odds of ele- to 183 days), sNfL-E participants had lower BPF and vated sNfL, and this association was consistent across higher T2 lesion volume compared to sNfL-N (adjusted BMI categories. Use of an MS DMT was associated with differences in Z-scores—BPF: 0.20, 95% CI: 0.28 to lower odds of elevated sNfL, although 70% of patients 0.12; T2 lesion volume: 0.42, 95% CI: 0.33 to 0.51; with elevated sNfL were on an MS DMT. Overall, these P < 0.001 for both). Findings were similar for the contin- findings were consistent in analyses of the continuous uous MS PATHS HC-derived age-normative sNfL Z- age-normative sNfL Z-scores, derived either from the MS scores. Prior MRIs obtained at least 6 months before but PATHS HC (Table S3) or NHANES (Table S4). within 2 years (median time from prior MRI: 1.0 years, 90 ª 2022 The Authors. Annals of Clinical and Translational Neurology published by Wiley Periodicals LLC on behalf of American Neurological Association. E. S. Sotirchos et al. sNfL in MS PATHS Table 2. Associations of demographic and clinical characteristics with elevated sNfL. Univariable analysis Multivariable analysis Characteristics Odds ratio 95% CI P-value Odds ratio 95% CI P-value Age, years Q1: 18–36 — — — — Q2: 37–42 0.47 0.39, 0.57 <0.001 0.45 0.37, 0.55 <0.001 Q3: 43–50 0.38 0.31, 0.46 <0.001 0.34 0.27, 0.42 <0.001 Q4: 51–58 0.44 0.36, 0.53 <0.001 0.34 0.27, 0.43 <0.001 Q5: >58 0.66 0.55, 0.78 <0.001 0.39 0.31, 0.49 <0.001 Sex Female — — — — Male 1.11 0.97, 1.27 0.12 1.20 1.04, 1.39 0.012 Body mass index, (kg/m2) 18.5–24.9 — — — — <18.5 1.82 1.18, 2.76 0.006 1.84 1.16, 2.87 0.008 25–29.9 0.76 0.65, 0.90 <0.001 0.68 0.58, 0.81 <0.001 30–39.9 0.81 0.69, 0.95 0.009 0.65 0.55, 0.77 <0.001 >40 0.85 0.64, 1.10 0.2 0.61 0.45, 0.81 <0.001 Disease duration, (years) Q1: < 5 — — — — Q2: 5–9 0.67 0.55, 0.82 <0.001 0.76 0.62, 0.94 0.012 Q3: 10–15 0.63 0.52, 0.78 <0.001 0.77 0.62, 0.96 0.020 Q4: 16–23 0.59 0.48, 0.72 <0.001 0.79 0.63, 0.99 0.045 Q5 > 23 0.82 0.68, 1.00 0.045 1.02 0.80, 1.28 0.9 MS subtype CIS/RRMS — — — — Progressive MS 1.62 1.42, 1.84 <0.001 1.63 1.41, 1.88 <0.001 DMT class None — — — — IFN-b/GA 0.59 0.48, 0.72 <0.001 0.63 0.50, 0.78 <0.001 Infusion/IRT 0.80 0.66, 0.97 0.021 0.63 0.51, 0.77 <0.001 Oral 0.47 0.39, 0.57 <0.001 0.45 0.37, 0.55 <0.001 Other or unknown 0.75 0.59, 0.94 0.016 0.78 0.61, 1.01 0.060 eGFR, (mL/min/1.73 m2) >90 — — — — 60–89 1.02 0.86, 1.21 0.8 1.20 1.00, 1.44 0.053 <60 4.49 2.99, 6.73 <0.001 5.21 3.39, 8.00 <0.001 Diabetes mellitus 1.46 1.16, 1.83 <0.001 1.67 1.31, 2.12 <0.001 Current smoking 1.12 0.96, 1.32 0.15 1.21 1.02, 1.43 0.029 Odds ratios were derived from logistic regression models assessing sNfL-E versus sNfL-N as the dependent variable. The multivariable model included all the characteristics listed in the table as independent variables. Statistically significant results (P < 0.05) are bolded. sNfL-E, elevated sNfL; sNfL-N, “normal” sNfL; SD, standard deviation; IQR, interquartile range; CIS, clinically isolated syndrome; RRMS, relapsing– remitting MS; PDDS, Patient Determined Disease Steps; DMT, disease-modifying therapy; IFN-beta: interferon-beta; GA, glatiramer acetate; IRT, immune reconstitution therapy; eGFR, estimated glomerular filtration rate. IQR 0.8 to 1.3 years) were available for 1651 participants Contrast-enhanced MRI was available within 60 days (n = 285 sNfL-E, 17.3%) allowing assessment for interim prior to blood sampling for 976 participants (n = 163 radiological inflammatory disease activity (i.e., new T2 sNfL-E, 16.7%). One or more gadolinium-enhancing lesions). One or more new T2 lesions were found in lesions was detected in 17.2% of sNfL-E vs 6.4% of sNfL- 26.3% of sNfL-E vs 10.9% of sNfL-N participants (ad- N participants (adjusted OR: 3.68, 95% CI: 1.97 to 6.79, justed OR: 2.66, 95% CI: 1.86 to 3.77, P < 0.001; P < 0.001; Fig. 4B), and similar results were found when Fig. 4A), and consistent findings were found when assess- assessing the MS PATHS HC-derived sNfL Z-score as a ing the MS PATHS HC-derived sNfL Z-score as a contin- continuous variable (adjusted OR: 1.50 per one-unit uous variable (adjusted OR per one-unit increase: 1.36; increase, 95% CI: 1.25 to 1.79, P < 0.001). Of those with 95% CI: 1.23 to 1.50, P < 0.001). one or more gadolinium-enhancing lesions (n = 80), 35% ª 2022 The Authors. Annals of Clinical and Translational Neurology published by Wiley Periodicals LLC on behalf of American Neurological Association. 91 sNfL in MS PATHS E. S. Sotirchos et al. Table 3. Baseline associations of sNfL with neuroperformance measures and MRI volumetrics. sNfL-E versus sNfL-N sNfL age normative Z-score Adjusted mean Adjusted beta difference1 95% CI P-value coefficient1,3 95% CI P-value Walking speed 2 0.54 0.80 to 0.28 <0.001 0.23 0.30 to 0.16 <0.001 Manual dexterity2 0.45 0.58 to 0.33 <0.001 0.14 0.17 to 0.11 <0.001 Processing speed2 0.30 0.38 to 0.22 <0.001 0.10 0.12 to 0.08 <0.001 Brain parenchymal fraction2,4 0.20 0.28 to 0.12 <0.001 0.06 0.08 to 0.04 <0.001 T2 lesion volume2,4 0.42 0.33 to 0.51 <0.001 0.12 0.10 to 0.14 <0.001 1 Results derived from linear regression models adjusted for age, sex, BMI, disease duration, disease subtype, DMT class, eGFR, diabetes mellitus, and smoking status. 2 Neuroperformance domains were transformed to Z-scores using regression-based equations derived from a study of adult healthy volunteers. T2 lesion volume (normalized to total brain volume and log-transformed) and BPF were converted to Z-scores based on the distribution in the MS population. 3 Adjusted beta coefficients correspond to the change in the Z-score for a given measure for a one-unit increment in the sNfL age-normative Z-score (derived from the MS PATHS healthy control cohort). There was no evidence for deviation from linearity for any of the examined measures (likelihood ratio test for comparison between natural cubic spline model and linear model, P-value >0.05 for all). 4 Brain parenchymal fraction and T2 lesion volume available for n = 3514 with MRI within 1 year of blood sampling (median [IQR] absolute time between blood sampling and MRI was 96 days [23 to 183 days]). Figure 4. Serum neurofilament light chain group and associations with recent inflammatory disease activity. (A) New T2 lesion count compared to previous available brain MRI in sNfL-N (“normal” sNfL) and sNfL-E (“elevated” sNfL) MS participants. (B) Gadolinium (Gd)-enhancing lesion count in sNfL-N (“normal” sNfL) and sNfL-E (“elevated” sNfL) MS participants with available contrast-enhanced brain MRI within 60 days prior to sNfL measurement. were sNfL-E, whereas of those without gadolinium- follow-up period of up to 2 years, in 2251 MS participants enhancing lesions (n = 896) 6.4% were sNfL-E. (n = 400 sNfL-E, 17.8%) with at least 6 months of available Baseline associations of sNfL with clinical disability and MRI follow-up (median follow-up: 1.7 years; IQR 1.1 to radiological measures were also assessed utilizing 1.9 years). Longitudinal MRI was also available for 148 HC NHANES-derived age-normative Z-scores and exhibited participants (median follow-up: 1.2 years, IQR: 1.0 to similar relationships (Table S5). 2.1 years). sNfL-E participants exhibited 63% faster whole brain atro- phy compared to sNfL-N participants (annualized percent Longitudinal analysis of brain atrophy and change in BPF: 0.26%/year vs 0.16%/year; adjusted differ- new T2 lesions ence: 0.10%/year, 95% CI: 0.14% to 0.06%, P < 0.001; We examined associations between sNfL levels, new T2 Fig. 5A). Both groups exhibited faster rates of brain atrophy lesion development, and whole brain atrophy over a compared to HC (annualized percent change in BPF in HC 92 ª 2022 The Authors. Annals of Clinical and Translational Neurology published by Wiley Periodicals LLC on behalf of American Neurological Association. E. S. Sotirchos et al. sNfL in MS PATHS Figure 5. Associations of serum neurofilament light chain with prospective brain atrophy and T2 lesion development. (A) Comparison of annualized percent change (mean and 95% confidence intervals) in brain parenchymal fraction between HC (healthy controls), sNfL-N (“normal” sNfL), and sNfL-E (“elevated” sNfL) MS participants. (B) Relationship of annualized percent change in brain parenchymal fraction (BPF) with age- normative sNfL Z-scores (derived from MS PATHS HC) in the MS participants. The estimates were derived from a mixed effects regression model using restricted cubic splines to model the sNfL Z-score (likelihood ratio test compared to linear model P = 0.002). The solid line corresponds to the estimated mean annualized percent change in BPF, and the bounds of the shaded area to the 95% confidence intervals. The vertical dashed line corresponds to the 97.5th percentile cut-off used to define sNfL-E versus sNfL-N (i.e., age-adjusted sNfL Z-score of 1.96). (C) New T2 lesion count during follow-up in in sNfL-N (“normal” sNfL) and sNfL-E (“elevated” sNfL) MS participants. 0.09%/year; adjusted differences vs HC—sNfL-E: 0.16%/ laboratory platform, in a large, multi-center, real-world year, 95% CI: 0.20% to 0.12%, P < 0.001; sNfL-N: population of PwMS. We found that a significant propor- 0.06%/year, 95% CI: 0.09% to 0.04%, P = 0.017). tion of PwMS had elevated sNfL relative to age-adjusted Higher MS PATHS HC-derived age-normative sNfL Z-scores reference ranges derived from HC. Factors associated with were associated with faster brain atrophy (adjusted change in elevated sNfL for age in MS included demographic factors annualized percent brain atrophy: 0.026%/year per one- (male sex, younger age), clinical characteristics (progres- unit increase, 95% CI: 0.03% to 0.02%, P < 0.001), but sive disease subtype, lack of DMT use, short or long dis- this relationship appeared non-linear (likelihood ratio test ease duration), and comorbidities (impaired renal P = 0.003), with relatively stable brain atrophy rates for sNfL function, diabetes mellitus, and smoking). Notably, higher Z-scores less than two (i.e., corresponding to the 97.5th per- BMI was associated with lower sNfL. Furthermore, sNfL centile [z = 1.96] cut-off used to identify sNfL-E) but pro- elevations were associated with more severe physical and gressively increasing rates of brain volume loss for those with cognitive disability, brain atrophy, and radiological higher sNfL Z-scores (Fig. 5B). One or more new T2 lesions inflammatory disease activity. was detected during follow-up in 20.5% of sNfL-E vs 12.1% Previous studies have established a strong association of sNfL-N participants (adjusted OR: 1.94, 95% CI: 1.42 to between age and sNfL levels.3 Notably, we found that, 2.62, P < 0.001). Findings were consistent when assessing the when accounting for the age-adjusted distribution of sNfL MS PATHS HC-derived age-normative sNfL Z-score as a in the MS PATHS HC, younger PwMS were more likely continuous variable (adjusted OR: 1.19 per one-unit increase, to have elevated sNfL levels for age. This is consistent 95% CI: 1.10 to 1.29, P < 0.001). with the observation that overt inflammatory disease These analyses were also performed utilizing NHANES- activity (and consequent neuro-axonal injury) is more derived age-normative Z-scores with similar findings for frequent in younger PwMS and decreases with aging.33 brain atrophy (adjusted change in annualized percent Furthermore, shorter disease duration (<5 years) was also brain atrophy: 0.041%/year per one-unit increase, 95% independently associated with elevated sNfL, consistent CI: 0.050% to 0.033%, P < 0.001; Fig. S2) and new with this phenomenon, although those with longer disease T2 lesion development (adjusted OR: 1.34 per one-unit duration (> 3 years) also exhibited similarly elevated increase, 95% CI: 1.16 to 1.54, P < 0.001). sNfL, which could relate to increasing contribution of neurodegenerative processes. Furthermore, male sex was found to be associated with higher age-adjusted sNfL. Discussion Natural history studies have suggested that among people In summary, we assessed sNfL, measured using a novel with MS, males are at a higher risk of disability and pro- high-throughput immunoassay on an existing clinical gressive disease, and it is likely that this finding is ª 2022 The Authors. Annals of Clinical and Translational Neurology published by Wiley Periodicals LLC on behalf of American Neurological Association. 93 sNfL in MS PATHS E. S. Sotirchos et al. reflective of this association. However, while an individual reference populations. This observation is unlikely to be participant-level meta-analysis of CSF NfL also found related to a technical issue due to different collection pro- higher levels in males with MS, this was also observed in cedures and/or processing, given that sNfL levels have healthy controls and other non-inflammatory neurological been previously reported to remain stable in unprocessed diseases.34 This observation needs to be explored further blood samples stored up to 7 days at room temperature, in large control populations in order to assess the contri- to withstand repeat freeze–thaw cycles, and to be similar bution of biological sex to blood NfL levels independent across different serum collection tube types.20,39,40 Given of neurological disease. the above, we speculate that this difference may relate to Additionally, our findings support the importance of selection bias of individuals recruited from tertiary refer- accounting for comorbidities when interpreting sNfL. Our ral centers (with the MS PATHS HC sampled from a sim- observation of a relationship between reduced eGFR and ilar population), whereas NHANES is designed to be a higher sNfL is consistent with recent reports of large ref- representative sample of the non-institutionalized US erence populations (including NHANES), as well as smal- population.41 This highlights the importance of consider- ler prior reports of older adults without neurological ing a reference population that is representative of the disease.9,10,12,13 The precise mechanism by which renal population under consideration when interpreting sNfL function is associated with sNfL is not clear but could be levels at an individual level. Important strengths of our related to possible renal clearance of blood NfL and/or study include the large sample size with standardized neurological complications of renal disease.35 This is espe- MRIs available for a large subset of the cohort, multi- cially relevant in people with progressive MS, who may center data from across two continents, the incorporation have lower than predicted eGFR.36 Diabetes mellitus and of standardized collection of measures assessing ambula- smoking were also associated with higher sNfL, which tory disability, manual dexterity and cognitive disability, could be related to complications such as diabetic neu- and the availability of data related to comorbidities. Fur- ropathy and cerebrovascular disease, but these factors thermore, we employed a high-throughput, scalable may also be associated with a pro-inflammatory state and immunoassay for sNfL measurement that can be run on exacerbate underlying MS disease processes.37,38 BMI, an existing routine clinical laboratory platform, with mea- however, exhibited an inverse association with sNfL, con- surements demonstrating good correlation with Simoa, sistent with prior reports, possibly related to a dilutional and consistent associations with clinical and radiological effect of blood volume which is highly correlated with measures. BMI.9,11–13 Our study bears several limitations that warrant discus- We also confirmed that elevated sNfL was associated sion. sNfL was measured only at a single timepoint for cross-sectionally with both physical and cognitive disabil- this analysis. Given the dynamic nature of inflammation ity in MS, as well as lower BPF and higher T2 lesion vol- and consequent neuro-axonal injury in MS, it is likely ume. These associations were generally modest, in line that serial measurements of sNfL will be more informa- with prior smaller studies. This is consistent with the per- tive.7 Serial blood sampling is being performed as part of ception of sNfL as a dynamic biomarker reflecting recent MS PATHS and future planned analyses will investigate and ongoing neuro-axonal injury, whereas established the utility of serial sNfL measurements. Furthermore, we clinical disability and brain atrophy in MS represent the were unable to perform a detailed analysis of the impact cumulative effects of preceding neuro-axonal injury.5,6 of inflammatory disease activity at the time of sampling Furthermore, elevated sNfL was associated with markedly on sNfL and its relationship with outcomes, given that accelerated short-term brain atrophy, and significantly exact timing of relapses and contrast-enhanced MRI was higher odds of prior, ongoing and future inflammatory not available routinely for the studied population within disease activity. Importantly though, rates of whole brain close proximity to blood sampling. Additionally, the atrophy were also faster compared to HC for those with duration of follow-up at present in MS PATHS was sNfL within the age-normative reference range. insufficient to assess for clinical disability progression in a Interestingly, when using NHANES as the reference sufficiently large sample of participants. Longitudinal population, age-normative sNfL levels in the MS partici- follow-up of MS PATHS study participants is ongoing, pants were not elevated compared to the reference distri- and this is an important outcome of interest that will be bution, whereas when using the MS PATHS HC as the assessed. Furthermore, the sample size of the MS PATHS reference population, we were clearly able to detect sNfL HC population was relatively small, but, as we observed, elevations in the MS participants. Notably, NHANES- the use of an unrelated reference population (i.e., derived and MS PATHS HC-derived age-normative sNfL NHANES) may not have been appropriate for our MS Z-scores exhibited excellent correlations and this seemed population, precluding the ability to detect abnormal to represent a systematic difference between the two sNfL elevations, although analyses of continuous 94 ª 2022 The Authors. Annals of Clinical and Translational Neurology published by Wiley Periodicals LLC on behalf of American Neurological Association. E. S. Sotirchos et al. sNfL in MS PATHS age-normative sNfL Z-scores demonstrated consistent Roche, Stendhal and ECTRIMS; speaking honoraria from findings. Finally, since data in MS PATHS are collected as Sanofi, Roche, and Merck. Xavier Montalban reports: part of routine clinical care, measures including MRI and speaking honoraria and travel expenses for scientific meet- creatinine/eGFR were not available for the entire cohort. ings, has been a steering committee member of clinical tri- In conclusion, our study supports that sNfL, measured als or participated in advisory boards of clinical trials in the using a high-throughput and scalable assay, is associated past 3 years with Actelion, Alexion, Biogen, Celgene, EMD with clinical disability, inflammatory disease activity, and Serono, Genzyme, Immunic, Medday, Merck, Mylan, whole brain atrophy in MS, but interpretation needs to Novartis, Roche, Sanofi-Genzyme, and Teva Pharmaceuti- account for body mass index and comorbidities including cal. Manuel Comabella reports: compensation for consult- impaired renal function, diabetes, and smoking. Longitu- ing services and speaking honoraria from Bayer Schering dinal follow-up of MS PATHS study participants is ongo- Pharma, Merk Serono, Biogen-Idec, Teva Pharmaceuticals, ing, and planned future work includes examining Sanofi-Aventis, and Novartis. Robert Naismith reports: sci- associations of sNfL (including serial sNfL measurements) entific advisory board and/or consulting for Abata Thera- with disability progression and long-term radiological peutics, Banner Life Sciences, BeiGene, Biogen, Bristol outcomes. Myers Squibb, Genentech, Genzyme, Janssen, GW Thera- peutics, Horizon Therapeutics, Lundbeck, NervGen, TG Therapeutics. Lauren Krupp reports: scientific advisory Acknowledgments board and/or consulting for Biogen, Novartis, Janssen, Ger- Study funding was provided from the National Institutes son Lehrman, Sanofi, and Biogen; research support from of Health (K23NS117883 to E.S.S.; K01MH121582 to Biogen. Jacqueline Nicholas reports: scientific advisory K.C.F.; U01NS111678 to P.A.C.), National Multiple board and/or consulting for Novartis, Genentech, Green- Sclerosis Society (RG-1904-33834 to E.S.S.; RG-1904- wich Biosciences, Biogen, EMD Serono, and TG Therapeu- 33800 to P.A.C.), and Biogen. tics; compensation for serving on speakers bureaus for EMD Serono, Alexion, Viela Bio, and Bristol Myers Squibb; research support from Novartis, Biogen, and Genentech. Author Contributions Katja Akgün reports: scientific advisory board and/or con- E.S.S., K.C.F., E.M.M., and P.A.C. contributed to the con- sulting for Roche, Sanofi, Alexion, Teva, Biogen, and Cel- ception and design of the study. E.S.S., K.C.F., C.M.M., gene. Tjalf Ziemssen reports: scientific advisory board and/ M.D.S., M.R., C.M.H., M.H.H., R.C., S.B.S., G.A., X.M., or consulting for Biogen, Roche, Novartis, Celgene, and M.C., R.T.N., M.Q., L.B.K., J.A.N., K.A., T.Z., R.R., E.F., Merck; compensation for serving on speakers bureaus for R.A.B., E.M.M., and P.A.C. contributed to the acquisition Roche, Novartis, Merck, Sanofi, Celgene, and Biogen; and/or analysis of the data. E.S.S., K.C.F., E.M.M., and research support from Biogen, Novartis, Merck, and Sanofi. P.A.C. drafted the text and prepared the figures. Richard Rudick is a former Biogen employee and holds stock in the company. Robert Bermel reports: scientific advisory board and/or consulting for Astra Zeneca, Biogen, Conflict of Interest EMD Serono, Genzyme, Genentech, Novartis, and Viela- Elias Sotirchos reports: scientific advisory board and/or Bio, research support from Biogen, Genentech, and Novar- consulting for Alexion, Viela Bio, Horizon Therapeutics, tis, and shared rights to intellectual property underlying the Genentech, and Ad Scientiam; speaking honoraria from Multiple Sclerosis Performance Test, currently licensed to Alexion, Viela Bio, and Biogen. Kathryn Fitzgerald, Mat- Qr8 Health and Biogen. Ellen Mowry reports: research sup- thew Smith, Maria Reyes-Mantilla, Ryan Canissario, Min port from Biogen, Teva, and Genentech, and royalties for Qiao, and Sarah Simmons have nothing to disclose. Carol editorial duties from UpToDate. Peter Calabresi reports: Singh and Elizabeth Fisher are employees of Biogen and consulting fees from Biogen, Nervgen, Idorsia, Avidea hold stock/stock options in the company. Carrie Hersh (now Vaccitech), and Disarm Therapeutics (now Lilly); reports: scientific advisory board and/or consulting for Bio- research support from Principia and Genentech. gen, Novartis, Genentech, Genzyme, EMD Serono, TG Therapeutics, and Bristol-Myers Squibb; compensation for Data Availability Statement serving on speakers bureaus for Genzyme and Biogen; research support from Biogen, Novartis, and Genentech. Requests for individual participant de-identified data Megan Hyland reports: research support from Biogen. would be considered from qualified investigators, based Georgina Arrambide reports: consulting services or partici- on information provided, including the proposed use and pation in advisory boards from Sanofi, Merck and Roche; analysis plan. Requests to access the datasets should be travel expenses for scientific meetings from Novartis, directed to datasharing@biogen.com. Proposals will be ª 2022 The Authors. Annals of Clinical and Translational Neurology published by Wiley Periodicals LLC on behalf of American Neurological Association. 95 sNfL in MS PATHS E. S. Sotirchos et al. reviewed by the MS PATHS data use committee. If the automated immunoassay platform. Front Neurol. proposed use is appropriate, a data sharing agreement will 2022;13:935382. doi:10.3389/fneur.2022.935382 be put in place before a fully de-identified version of the 15. Mowry EM, Bermel RA, Williams JR, et al. Harnessing dataset, including the data dictionary, is made available. real-world data to inform decision-making: multiple sclerosis partners advancing technology and health solutions (MS PATHS). Front Neurol. 2020;11:632. REFERENCES 16. Rhodes JK, Schindler D, Rao SM, et al. Multiple sclerosis 1. Reich DS, Lucchinetti CF, Calabresi PA. Multiple Sclerosis. performance test: technical development and usability. N Engl J Med. 2018;378:169-180. Adv Ther. 2019;36:1741-1755. 2. Petzold A. Neurofilament phosphoforms: surrogate 17. Learmonth YC, Motl RW, Sandroff BM, Pula JH, Cadavid markers for axonal injury, degeneration and loss. J Neurol D. Validation of patient determined disease steps (PDDS) Sci. 2005;233:183-198. scale scores in persons with multiple sclerosis. BMC 3. Khalil M, Teunissen CE, Otto M, et al. Neurofilaments as Neurol. 2013;13:37. biomarkers in neurological disorders. Nat Rev Neurol. 18. Rao SM, Galioto R, Sokolowski M, et al. Multiple sclerosis 2018;14:577-589. performance test: validation of self-administered 4. Bittner S, Oh J, Havrdova EK, Tintore M, Zipp F. The neuroperformance modules. Eur J Neurol. 2020;27:878- potential of serum neurofilament as biomarker for 886. multiple sclerosis. Brain. 2021;144:2954-2963. 19. Rao SM, Losinski G, Mourany L, et al. Processing speed 5. Disanto G, Adiutori R, Dobson R, et al. Serum test: validation of a self-administered, iPad-based tool neurofilament light chain levels are increased in patients for screening cognitive dysfunction in a clinic setting. with a clinically isolated syndrome. J Neurol Neurosurg Mult Scler. 2017;23:1929-1937. Psychiatry. 2016;87:126-129. 20. Sotirchos ES, Fitzgerald KC, Smith MD, et al. Type of 6. Barro C, Benkert P, Disanto G, et al. Serum neurofilament as serum collection tube does not impact neurofilament light a predictor of disease worsening and brain and spinal cord chain levels. Mult Scler Relat Disord. 2022;59:103676. atrophy in multiple sclerosis. Brain. 2018;141:2382-2391. 21. Qiu X, Lee S, Jackson J, et al. Equivalence of serum and 7. Chitnis T, Gonzalez C, Healy BC, et al. Neurofilament plasma neurofilament light chain levels using highly light chain serum levels correlate with 10-year MRI sensitive automated immunoassay. Alzheimers Dement. outcomes in multiple sclerosis. Ann Clin Transl Neurol. 2020;16:e045278. 2018;5:1478-1491. 22. Wilson DH, Rissin DM, Kan CW, et al. The Simoa HD-1 8. Canto E, Barro C, Zhao C, et al. Association between analyzer: a novel fully automated digital immunoassay serum neurofilament light chain levels and long-term analyzer with single-molecule sensitivity and multiplexing. disease course among patients with multiple sclerosis J Lab Autom. 2016;21:533-547. followed up for 12 years. JAMA Neurol. 2019;76:1359- 23. Fisher E, Kober T, Tsang A, et al. Magnetic resonance 1366. imaging (MRI) metrics in routine clinical practice: proof 9. Benkert P, Meier S, Schaedelin S, et al. Serum of concept in MS PATHS (multiple sclerosis partners neurofilament light chain for individual prognostication of advancing Technology for Health Solutions) (1356). disease activity in people with multiple sclerosis: a Neurology. 2020;94:1356 https://n.neurology.org/content/ retrospective modelling and validation study. Lancet 94/15_Supplement/1356 Neurol. 2022;21:246-257. 24. Fartaria MJ, Kober T, Granziera C, Bach CM. 10. Akamine S, Marutani N, Kanayama D, et al. Renal Longitudinal analysis of white matter and cortical lesions function is associated with blood neurofilament light chain in multiple sclerosis. Neuroimage Clin. 2019;23:101938. level in older adults. Sci Rep. 2020;10:20350. 25. Rigby RA, Stasinopoulos DM. Generalized additive models 11. Manouchehrinia A, Piehl F, Hillert J, et al. Confounding for location, scale and shape. J R Stat Soc Ser C Appl Stat. effect of blood volume and body mass index on blood 2005;54:507-554. neurofilament light chain levels. Ann Clin Transl Neurol. 26. Stasinopoulos MD, Rigby RA, Heller GZ, Voudouris V, 2020;7:139-143. Bastiani FD. Flexible Regression and Smoothing: 12. Koini M, Pirpamer L, Hofer E, et al. Factors influencing Using GAMLSS in R. 1st ed. Chapman and Hall/CRC; serum neurofilament light chain levels in normal aging. 2017. Aging (Albany NY). 2021;13:25729-25738. 27. Katayev A, Balciza C, Seccombe DW. Establishing 13. Fitzgerald KC, Sotirchos ES, Smith MD, et al. reference intervals for clinical laboratory test results: is Contributors to serum NfL levels in people without there a better way? Am J Clin Pathol. 2010;133:180-186. neurologic disease. Ann Neurol. 2022;92:688-698. 28. Inker LA, Eneanya ND, Coresh J, et al. New creatinine- 14. Lee S, Plavina T, Singh CM, et al. Development of a and cystatin C–based equations to estimate GFR without highly sensitive neurofilament light chain assay on an race. N Engl J Med. 2021;385:1737-1749. 96 ª 2022 The Authors. Annals of Clinical and Translational Neurology published by Wiley Periodicals LLC on behalf of American Neurological Association. E. S. Sotirchos et al. sNfL in MS PATHS 29. Sokolowski M, Dey T, Strober L, et al. Multiple sclerosis 39. Altmann P, Leutmezer F, Zach H, et al. Serum performance test (MSPT): a demographically stratified neurofilament light chain withstands delayed freezing and normative database (N = 517) of healthy subjects ages 15– repeated thawing. Sci Rep. 2020;10:19982. 89 to aid clinical interpretation of MS patient 40. Altmann P, Ponleitner M, Rommer PS, et al. Seven day neuroperformance. Mult Scler J. 2019;25:78. pre-analytical stability of serum and plasma neurofilament 30. Sotirchos ES, Gonzalez-Caldito N, Dewey BE, et al. light chain. Sci Rep. 2021;11:11034. Effect of disease-modifying therapies on subcortical gray € 41. Kokmen E, Ozsarfati Y, Beard CM, O’Brien PC, Rocca matter atrophy in multiple sclerosis. Mult Scler. WA. Impact of referral bias on clinical and 2020;26:312-321. epidemiological studies of Alzheimer’s disease. J Clin 31. Harrell FE. General aspects of fitting regression models. Epidemiol. 1996;49:79-83. In: Harrell FE Jr, ed. Regression Modeling Strategies: with Applications to Linear Models, Logistic and Ordinal Regression, and Survival Analysis. Springer International Supporting Information Publishing; 2015:13-44. doi:10.1007/978-3-319-19425-7_2 Additional supporting information may be found online 32. Gauthier J, Wu QV, Gooley TA. Cubic splines to model in the Supporting Information section at the end of the relationships between continuous variables and outcomes: article. a guide for clinicians. Bone Marrow Transplant. 2020;55:675-680. Table S1. Characteristics of the healthy controls. 33. Zeydan B, Kantarci OH. Impact of age on multiple Table S2. Distribution of MS participants and healthy sclerosis disease activity and progression. Curr Neurol controls by MS PATHS site. Neurosci Rep. 2020;20:24. Table S3. Associations of demographic and clinical char- 34. Bridel C, van Wieringen WN, Zetterberg H, Tijms BM, acteristics with age-normative sNfL Z-scores (derived Teunissen CE, the NFL Group. Diagnostic value of using the sNfL distribution in the MS PATHS healthy cerebrospinal fluid neurofilament light protein in control cohort). neurology: a systematic review and meta-analysis. JAMA Table S4. Associations of demographic and clinical char- Neurol. 2019;76:1035-1048. acteristics with age-normative sNfL Z-scores (derived 35. Krishnan AV, Kiernan MC. Neurological complications of using the sNfL distribution in the NHANES reference chronic kidney disease. Nat Rev Neurol. 2009;5:542-551. population). 36. Calabresi PA, Austin H, Racke MK, et al. Impaired renal Table S5. Associations of MS participant age-normative function in progressive multiple sclerosis. Neurology. NHANES-derived sNfL Z-scores with neuroperformance 2002;59:1799-1801. measures, MRI volumetrics and new T2 lesion develop- 37. Van Dyken P, Lacoste B. Impact of metabolic syndrome ment. on neuroinflammation and the blood–brain barrier. Front Figure S1. Serum neurofilament light chain levels by age Neurosci. 2018;12:930. doi:10.3389/fnins.2018.00930 in the healthy control cohort. 38. Rosso M, Chitnis T. Association between cigarette Figure S2. Associations of age-normative sNfL Z-scores smoking and multiple sclerosis: a review. JAMA Neurol. derived using NHANES with prospective whole brain 2020;77:245-253. atrophy in the MS participants. ª 2022 The Authors. Annals of Clinical and Translational Neurology published by Wiley Periodicals LLC on behalf of American Neurological Association. 97