Glial Fibrilar Acidic Protein как знак болезни в повторном многоразовом склерозе: постспециальный анализ испытаний этапа 3 Ozanimod.

Introduction

Glial fibrillary acidic protein (GFAP) is a Type III intermediate filament that provides structural stability to astrocytes. Astrocytes have many cellular functions in a healthy central nervous system (CNS), including maintenance of the blood–brain barrier, synapse development and plasticity, and support of neurological function [ 1 , 2 , 3 ]. In response to CNS damage or disease, astrocytes undergo biochemical and structural changes, and GFAP protein expression is upregulated to support the formation and maintenance of process elongation and thickening in these activated (reactive) astrocytes [ 3 , 4 , 5 ]. Elevated concentrations of GFAP in the circulation are therefore associated with astrocytic damage and activation, and preclinical evidence suggests that GFAP accumulation in neurologic disorders may exacerbate disease pathology through the formation of positive feedback loops that impact its synthesis and degradation [ 3 , 4 , 5 , 6 , 7 ]. Underlying pathological processes affect the dynamics of GFAP, where the half‐life of GFAP for acute neurologic events (e.g., traumatic brain injury) ranges between 24–72 h [ 8 ], but in chronic neuroinflammatory diseases, the counterbalance between GFAP accumulation and clearance has not been well defined [ 2 ]. GFAP is being investigated as a biomarker for CNS disorders, including various neurodegenerative diseases, traumatic brain/spinal cord injury, stroke, brain tumors, neuromyelitis optica, and multiple sclerosis (MS) [ 2 , 9 ].

In participants with MS, higher GFAP concentrations were observed in plasma and cerebrospinal fluid compared with healthy individuals and were correlated with disability and disease progression [ 10 , 11 , 12 , 13 , 14 , 15 , 16 ]. A meta‐analysis found that participants with progressive MS were more likely to have higher GFAP concentrations in cerebrospinal fluid than those with relapsing–remitting MS [ 17 ]. In addition, increased serum GFAP was associated with white matter damage in participants with MS, supporting the use of GFAP as a biomarker for MS pathology [ 18 ]. Overall, GFAP may be a useful marker for various measures of disease progression in MS; however, additional studies are needed to validate clinical use of this biomarker [ 17 ].

Here, the potential for baseline plasma GFAP as a biomarker for relapsing MS (RMS) was assessed using pooled data from two Phase 3, randomized, active‐comparator trials of ozanimod [ 19 , 20 ], an oral sphingosine 1‐phosphate receptor modulator that reduces the migration of lymphocytes into the CNS [ 21 ]. This post hoc, exploratory analysis evaluated the relationship between baseline demographic, clinical, radiologic, and laboratory parameters and baseline GFAP concentration, as well as the relationship between baseline GFAP concentration and on‐treatment clinical and radiologic outcomes in participants with RMS.

Methods

Phase 3 Ozanimod Trials

SUNBEAM and RADIANCE are previously published, randomized, Phase 3, multicenter, double‐blind, double‐dummy studies comparing daily oral ozanimod 0.92 mg and 0.46 mg with once weekly intramuscular interferon (IFN) β‐1a 30 μg [ 19 , 20 ]. The primary publications for these randomized trials followed the Consolidated Standards of Reporting Trials (CONSORT) guidelines and each contain a CONSORT flow diagram [ 19 , 20 ]. SUNBEAM was conducted from December 3, 2014 to December 22, 2016, and continued until the last participant completed 12 months of treatment [ 19 ]. RADIANCE was conducted from December 3, 2013 to April 13, 2017, and participants were treated for 24 months [ 20 ]. Both trials enrolled adults 18–55 years of age with RMS who had brain magnetic resonance imaging (MRI) lesions consistent with MS, at least one relapse within 12 months before screening or at least 1 relapse within 24 months before screening plus ≥ 1 gadolinium‐enhancing (GdE) lesion within 12 months before randomization, and an Expanded Disability Status Scale (EDSS) score of 0–5 [ 19 , 20 ]. In both studies, ozanimod was initiated via a 1‐week dose escalation regimen consisting of 0.23 mg on Days 1–4, 0.46 mg on Days 5–7, and then the assigned dose of either 0.46 mg or 0.92 mg beginning on Day 8.

T2 and T1 (with and without gadolinium) brain MRIs were performed during screening and at Months 6 (SUNBEAM only), 12 (SUNBEAM and RADIANCE), and 24 (RADIANCE only). Whole brain volume (WBV) was measured using SienaX [ 19 , 20 ].

Study Procedures

GFAP concentrations were measured by Quanterix Corporation (Lexington, MA, USA) in duplicate, heparinized plasma samples obtained at SUNBEAM and RADIANCE baseline. A single molecule array (Simoa) GFAP discovery kit and HD‐X analyzer were used to determine GFAP concentrations (Quanterix, Billerica, MA, USA) [ 22 ]. The limit of detection and lower limit of quantification of the assay were previously determined to be 0.211 pg/mL and 0.686 pg/mL, respectively [ 22 ]. The average within‐run coefficient of variation across five experiments of three replicates each ranged from 6.3% to 10.9% [ 22 ].

Outcomes

In this study, there were two main analyses for GFAP: (1) baseline relationships assessing GFAP as a biomarker of concurrent status and (2) baseline GFAP as a predictor of outcomes. All outcomes were analyzed in participants with GFAP measurements from SUNBEAM and RADIANCE.

For Part 1, individual baseline demographic (age, sex, body mass index [BMI]), clinical (number of relapses in the previous 12 months and EDSS scores), radiologic (number of GdE lesions, number of T2 lesions, and WBV), and laboratory parameters (neurofilament light chain [NfL] concentration) and relationships with baseline GFAP concentration as a dependent variable were assessed.

For Part 2, on‐treatment outcomes, number of relapses through Months 12 (SUNBEAM only) and 24 (RADIANCE only), number of GdE lesions at Month 12 (SUNBEAM and RADIANCE), number of new/enlarging T2 lesions through Month 12 (SUNBEAM and RADIANCE), WBV at Month 12 (SUNBEAM and RADIANCE), EDSS score at Month 12 (SUNBEAM and RADIANCE), proportion of participants with no evidence of disease activity (NEDA‐3) at Month 12 (SUNBEAM and RADIANCE), and relationships with baseline GFAP concentration as a predictor were assessed. NEDA‐3 was defined as no relapses, no EDSS progression (confirmed after 3 months), no new/enlarging T2 lesions, and no new GdE lesions [ 23 ].

Statistical Analyses

Due to the hierarchical testing used in the primary trials and the exploratory nature of this study, all statistical comparisons were hypothesis generating and there was no adjustment for multiplicity of testing, thus p values are nominal. For each relationship, only those with paired data available for GFAP and each individual outcome were included in the analysis.

Part 1: Individual baseline demographic, clinical, radiologic, and laboratory parameters and relationships with baseline log(GFAP) concentration were determined via robust linear regression analysis using MM estimation with bisquare weight functions and 85% Gaussian efficiency [ 24 ]. Log1p(radiologic parameter) = log(radiologic parameter +1) was used to avoid issues with 0 in the data set.

Part 2: On‐treatment clinical outcomes were assessed using Poisson models, negative binomial models, MM‐estimator robust linear models, Gaussian linear models, or binomial logistic regression models with baseline log(GFAP) concentration as the predictor. The deviance, or variance, explained in the models by GFAP levels is presented for all on‐treatment outcomes. The study was included as a fixed effect in the model for all outcomes, with the exception of the number of relapses, which was analyzed separately for each study. Scan count was included as an offset in the model for the number of GdE and new/enlarging T2 lesions. Duration was included as an offset in the model for the number of relapses. Treatment arm was included as a term for all on‐treatment outcomes. Terms for the baseline value of the parameter were included for WBV and EDSS scores.

Penalized multivariable lasso models were used to identify independent predictors of baseline GFAP concentration (factors: age, sex, BMI, baseline absolute lymphocyte count [ALC], baseline NfL concentration, number of relapses in the previous 12 months, months since relapse, baseline number of GdE lesions, baseline number of T2 lesions, baseline WBV, and baseline EDSS score) and on‐treatment response variables (outcomes: number of relapses, GdE lesions, and new/enlarging T2 lesions; WBV; EDSS score; and the proportion of participants with NEDA‐3). For all assessments, model selection was determined via lasso with 10‐fold cross validation and the one‐standard error rule [ 25 ]. Where necessary, Poisson or Gaussian assumptions were used in place of negative binomial or robust linear assumptions.

R version 4.2.2 Patched (2022‐11‐10 r83330) was used for all analyses [ 26 ]. The tidyverse suite of packages was used for various aspects of data restructuring and for the production of graphs [ 27 ]. The mgcv package was used for fitting the negative binomial models [ 28 ]. The robustbase package was used for robust regression [ 29 ], and RobStatTM was used for robust analysis of variance [ 30 ]. The glmnet package was used for lasso modeling [ 31 ].

Standard Protocol Approvals, Registrations, and Patient Consents

The institutional review board or ethics committee at each site approved the study protocol Table  S1 ) and all participants provided written informed consent. These trials conformed with the World Medical Association Declaration of Helsinki and are registered at ClinicalTrials.gov (SUNBEAM: NCT02294058 ; RADIANCE: NCT02047734 ) and EudraCT (SUNBEAM: 2014‐002320‐27; RADIANCE: 2012‐002714‐40).

Results

Disposition and Baseline Demographics and Disease Characteristics

Baseline GFAP assessments were available for 1117 of 1346 (83.0%) participants in SUNBEAM and for 939 of 1313 (71.5%) participants in RADIANCE. A total of 4231 samples were analyzed; 57 samples were excluded from the analysis (20 had insufficient volume, 36 had failing controls or high coefficients of variation, and 1 was below the limit of detection). Demographics and baseline disease characteristics among participants for whom baseline GFAP assessments were available and the total population from SUNBEAM and RADIANCE are listed in Table  1 and Tables  S2 and S3 , respectively.

TABLE 1 Demographics and baseline disease characteristics among participants in SUNBEAM and RADIANCE with baseline GFAP assessments. Parameter SUNBEAM ( n  = 1117) RADIANCE ( n  = 939) IFN β‐1a 30 μg ( n  = 373) Ozanimod 0.46 mg/d ( n  = 382) Ozanimod 0.92 mg/d ( n  = 362) IFN β‐1a 30 μg ( n  = 317) Ozanimod 0.46 mg/d ( n  = 318) Ozanimod 0.92 mg/d ( n  = 304) Age, mean (SD), years 35.9 (9.0) 35.9 (9.4) 35.1 (9.4) 35.2 (9.0) 35.6 (8.6) 36.1 (8.6) Sex Male 122 (32.7) 120 (31.4) 123 (34.0) 89 (28.1) 97 (30.5) 97 (31.9) Female 251 (67.3) 262 (68.6) 239 (66.0) 228 (71.9) 221 (69.5) 207 (68.1) BMI, mean (SD), kg/m 2
24.3 (4.7) 24.0 (4.6) 24.0 (4.5) 24.0 (4.5) 24.0 (4.4) 24.3 (5.1) GFAP concentration, pg/mL Mean (SD) 128.2 (73.7) 124.5 (59.9) 127.2 (63.1) 127.7 (64.6) 131.3 (64.2) 132.9 (68.6) Median (Q1−Q3) 111.4 (84.6–149.1) 115.0 (83.3–149.9) 113.0 (87.0–152.1) 113.7 (87.0–152.9) 116.2 (91.5–158.6) 120.6 (88.2–157.4) NfL concentration, pg/mL Mean (SD) 20.5 (17.9) 18.3 (13.7) 19.4 (16.5) 17.9 (15.5) 17.7 (16.0) 20.2 (25.8) Median (Q1−Q3) 14.9 (9.8–24.5) 14.0 (9.5–22.3) 14.9 (10.9–22.1) 13.5 (9.6–20.2) 13.1 (9.2–20.1) 13.9 (9.3–21.5) Number of relapses in previous 12 months, mean (SD) 1.3 (0.5) 1.3 (0.6) 1.3 (0.6) 1.3 (0.6) 1.4 (0.7) 1.3 (0.6) Participants with GdE lesions, n (%) 0 196 (52.5) 208 (54.5) 195 (53.9) 177 (55.8) 181 (56.9) 171 (56.2) 1–2 104 (27.9) 100 (26.2) 92 (25.4) 70 (22.1) 66 (20.8) 70 (23.0) 3–5 32 (8.6) 44 (11.5) 40 (11.0) 40 (12.6) 37 (11.6) 39 (12.8) > 5 41 (11.0) 30 (7.9) 35 (9.7) 30 (9.5) 34 (10.7) 24 (7.9) Participants with T2 lesions, n (%) 0 0 (0) 2 (0.6) 1 (0.3) 0 (0.0) 2 (0.6) 0 (0.0) 1–20 69 (20.3) 52 (14.6) 61 (18.4) 55 (17.4) 60 (18.9) 55 (18.1) 21–50 127 (37.4) 146 (40.9) 132 (39.9) 138 (43.5) 145 (45.6) 124 (40.8) 51–75 70 (20.6) 71 (19.9) 56 (16.9) 63 (19.9) 53 (16.7) 66 (21.7) > 75 74 (21.8) 86 (24.1) 81 (24.5) 61 (19.2) 58 (18.2) 59 (19.4) WBV, mean (SD), cm 3
1442.6 (78.3) 1446.2 (80.5) 1454.9 (78.6) 1449.0 (78.8) 1451.1 (69.7) 1442.5 (78.0)
Note: Percentages may not add to 100 due to rounding. Data are for participants who had baseline GFAP values and observed values for the other variables. Abbreviations: BMI, body mass index; GdE, gadolinium‐enhancing; GFAP, glial fibrillary acidic protein; IFN, interferon; NfL, neurofilament light chain; SD, standard deviation; WBV, whole brain volume.

Demographics and baseline disease characteristics among participants in SUNBEAM and RADIANCE with baseline GFAP assessments.

Note: Percentages may not add to 100 due to rounding. Data are for participants who had baseline GFAP values and observed values for the other variables.

Abbreviations: BMI, body mass index; GdE, gadolinium‐enhancing; GFAP, glial fibrillary acidic protein; IFN, interferon; NfL, neurofilament light chain; SD, standard deviation; WBV, whole brain volume.

Part 1: Relationship Between Baseline Demographics and Disease Characteristics and Baseline GFAP Concentration

Based on robust linear regression models, there was evidence of a relationship between baseline GFAP concentration and sex (nominal p  < 0.0001); an inverse relationship between baseline GFAP concentration and baseline BMI and WBV (nominal p  < 0.0001); and a positive relationship between baseline GFAP concentration and baseline age, NfL concentration, number of GdE and T2 lesions, and EDSS score (nominal p  < 0.0001) (Table  2 ). There was no clear relationship between baseline GFAP concentration and the number of relapses in the previous 12 months (nominal p  = 0.2415) (Table  2 ). When included in a multivariable lasso model, baseline GFAP concentration remained independently predicted by sex and baseline BMI, NfL concentration, number of T2 lesions, and EDSS score but not baseline age, WBV, or number of GdE lesions Figure  S1 . The lasso model also showed no relationship between baseline GFAP concentration and the number of relapses in the previous 12 months Figure  S1 .

TABLE 2 Relationships between individual baseline demographic, clinical, radiologic, and laboratory parameters and baseline log(GFAP) concentrations. SUNBEAM and RADIANCE ( n  = 2056) Baseline parameter Estimate (SE) Nominal p value Robust R
2
Age 0.0056 (0.0011) < 0.0001 1.6% Sex, male −0.1417 (0.0219) < 0.0001 — Log(BMI) −0.4851 (0.0559) < 0.0001 4.4% Log(NfL concentration) 0.2986 (0.0149) < 0.0001 18.9% Number of relapses in previous 12 months 0.0207 (0.0177) 0.2415 0.3% Log1p(number of GdE lesions) 0.1561 (0.0125) < 0.0001 8.4% Log1p(number of T2 lesions) 0.1776 (0.0127) < 0.0001 10.7% EDSS score 0.0957 (0.0087) < 0.0001 6.7% WBV −0.0012 (0.0001) < 0.0001 5.3%
Note: Log‐transformed baseline GFAP concentration was modeled separately as a linear function of each of the transformed baseline parameters above (independent of each other) using robust regression (MM‐estimation with 85% Gaussian efficiency and bisquare weight functions). Study was included as a fixed effect in the model. All statistical comparisons were hypothesis generating and there was no adjustment for multiplicity of testing; p values are therefore nominal. Abbreviations: BMI, body mass index; EDSS, Expanded Disability Status Scale; GdE, gadolinium‐enhancing; GFAP, glial fibrillary acidic protein; NfL, neurofilament light chain; SE, standard error; WBV, whole brain volume.

Relationships between individual baseline demographic, clinical, radiologic, and laboratory parameters and baseline log(GFAP) concentrations.

Note: Log‐transformed baseline GFAP concentration was modeled separately as a linear function of each of the transformed baseline parameters above (independent of each other) using robust regression (MM‐estimation with 85% Gaussian efficiency and bisquare weight functions). Study was included as a fixed effect in the model. All statistical comparisons were hypothesis generating and there was no adjustment for multiplicity of testing; p values are therefore nominal.

Abbreviations: BMI, body mass index; EDSS, Expanded Disability Status Scale; GdE, gadolinium‐enhancing; GFAP, glial fibrillary acidic protein; NfL, neurofilament light chain; SE, standard error; WBV, whole brain volume.

Part 2: Relationship Between Baseline GFAP Concentration and On‐Treatment Clinical and Radiologic Outcomes

Relapses

Based on Poisson models, baseline GFAP concentration had a positive relationship with the number of relapses through Months 12 (estimate [standard error (SE)]: 0.6544 [0.1143]; nominal p  < 0.0001) and 24 (estimate [SE]: 0.3991 [0.1090]; nominal p  = 0.0003); the model, which included terms for treatment arm and baseline GFAP concentration as predictors and duration as an offset, explained 5.66% and 2.78% of the deviance for the number of relapses through Months 12 and 24, respectively. The relationship between baseline GFAP concentration and the number of relapses through Month 12 of SUNBEAM and Month 24 of RADIANCE is displayed in Figure  1A,B , respectively. When included in a multivariable lasso model, baseline GFAP concentration remained an independent predictor of the number of relapses through Month 12 but not through Month 24 Figure  S2 .

Regression analysis of the number of relapses through Months 12 (SUNBEAM) and 24 (RADIANCE) and baseline log(GFAP) concentration. Relationships were assessed with Poisson models using baseline log(GFAP) concentration as a predictor. Duration was included as an offset in the model, which also included treatment arm as a factor. In SUNBEAM, relapses were assessed during the treatment period through Month 12. In RADIANCE, relapses were assessed during the treatment period through Month 24; thus, these data were not pooled. The number of relapses through Months 12 (estimate [SE]: ozanimod 0.46 mg, −0.4577 [0.1262]; ozanimod 0.92 mg, −0.6808 [0.1374]) and 24 (estimate [SE]: Ozanimod 0.46 mg, −0.3250 [0.1159]; ozanimod 0.92 mg, −0.5270 [0.1249]) was lower in the ozanimod arms compared with IFN β‐1a (nominal p  < 0.01 for all comparisons). All statistical comparisons were hypothesis generating, and there was no adjustment for multiplicity of testing; p values are therefore nominal. GFAP, glial fibrillary acidic protein; IFN, interferon; SE, standard error.

Lesion Counts

Based on negative binomial models, baseline GFAP concentration had a positive relationship with the number of GdE lesions at Month 12 (estimate [SE]: 0.9835 [0.1448], nominal p  < 0.0001) and the number of new/enlarging T2 lesions over 12 months (estimate [SE]: 0.5688 [0.0823], nominal p  < 0.0001); the model, which included terms for study, treatment arm, and baseline GFAP concentration as predictors and scan count as an offset, explained 11.93% and 4.30% of the deviance for the number of GdE lesions at Month 12 and the number of new/enlarging T2 lesions over 12 months, respectively. The relationships between baseline GFAP concentration and the number of GdE lesions at Month 12 and the number of new/enlarging T2 lesions over 12 months are shown in Figure  2A,B , respectively. When included in a multivariable lasso model, baseline GFAP concentration was not an independent predictor of the number of GdE lesions at Month 12 or the number of new/enlarging T2 lesions over 12 months Figure  S3 .

Regression analysis of the number of GdE lesions at Month 12 (SUNBEAM and RADIANCE) and the number of new/enlarging T2 lesions over 12 months (SUNBEAM and RADIANCE) and baseline log (GFAP) concentration. Relationships were assessed with negative binomial models using baseline log(GFAP) concentration as a predictor. Scan count was included as an offset in the model, which also included terms for study and treatment arm. The number of GdE lesions at Month 12 (estimate [SE]: ozanimod 0.46 mg, −0.6728 [0.1498]; ozanimod 0.92 mg, −1.1715 [0.1620]) and the number of new/enlarging T2 lesions over 12 months (estimate [SE]: ozanimod 0.46 mg, −0.3578 [0.0905]; ozanimod 0.92 mg, −0.5050 [0.0923]) were lower in the ozanimod arms compared with IFN β‐1a (nominal p  < 0.0001 for all comparisons). All statistical comparisons were hypothesis generating, and there was no adjustment for multiplicity of testing; p values are therefore nominal. GdE, gadolinium‐enhancing; GFAP, glial fibrillary acidic protein; IFN, interferon; SE, standard error.

Brain Volume

Based on MM‐estimator robust linear models, baseline GFAP concentration had an inverse relationship with WBV at Month 12 (estimate [SE]: −3.6935 [0.4924], nominal p  < 0.0001); the model, which included terms for study, baseline WBV, treatment arm, and baseline GFAP concentration as predictors, explained 98.96% of the variation of WBV at Month 12. The inclusion of baseline WBV was largely the reason for the high amount of variation explained. The relationship between baseline GFAP concentration and WBV at Month 12 is shown in Figure  3 . When included in a multivariable lasso model, baseline GFAP concentration was not an independent predictor of WBV at Month 12 Figure  S4 .

Regression analysis of WBV at Month 12 (SUNBEAM and RADIANCE) and baseline log (GFAP) concentration. Relationships were assessed with an MM‐estimator robust linear model using baseline log(GFAP) concentration as a predictor. Terms for study, baseline WBV, and treatment arm were also included in the model. WBV at Month 12 was higher in the ozanimod arms (estimate [SE]: ozanimod 0.46 mg, 1.4839 [0.5193]; ozanimod 0.92 mg, 2.3590 [0.5328]) compared with IFN β‐1a (nominal p  < 0.01 for all comparisons). All statistical comparisons were hypothesis generating, and there was no adjustment for multiplicity of testing; p values are therefore nominal. GFAP, glial fibrillary acidic protein; IFN, interferon; SE, standard error; WBV, whole brain volume.

Expanded Disability Status Scale Scores

Based on Gaussian linear models, baseline GFAP concentration had a positive relationship with EDSS score at Month 12 (estimate [SE]: 0.1018 [0.0271], nominal p  = 0.0002); the model, which included terms for study, baseline EDSS score, treatment arm, and baseline GFAP concentration as predictors, explained 81.84% of the variation of EDSS score at Month 12. The inclusion of baseline EDSS was largely the reason for the high amount of variation explained. The relationship between baseline GFAP concentration and EDSS score at Month 12 is shown in Figure  4 . When included in a multivariable lasso model, baseline GFAP concentration was not an independent predictor of EDSS score at Month 12 Figure  S5 .

Regression analysis of EDSS score at Month 12 (SUNBEAM and RADIANCE) and baseline log (GFAP) concentration. Relationships were assessed with a Gaussian linear model using baseline log(GFAP) concentration as a predictor. Terms for study, baseline EDSS score, and treatment arm were also included in the model. EDSS score at Month 12 was similar between the ozanimod arms (estimate [SE]: ozanimod 0.46 mg, −0.0434 [0.0305]; ozanimod 0.92 mg, −0.0383 [0.0309]) and the IFN β‐1a arm (nominal p  > 0.05 for all comparisons). All statistical comparisons were hypothesis generating, and there was no adjustment for multiplicity of testing; p values are therefore nominal. EDSS, Expanded Disability Status Scale; GFAP, glial fibrillary acidic protein; IFN, interferon; SE, standard error.

No Evidence of Disease Activity

Based on binomial logistic regression models, baseline GFAP concentration had an inverse relationship with the proportion of participants with NEDA‐3 at Month 12 (estimate [SE]: −0.6858 [0.1168], nominal p  < 0.0001); the model, which included terms for study, treatment arm, and baseline GFAP concentration as predictors, explained 2.06% of the deviance of the proportion of participants with NEDA‐3 at Month 12. The relationship between baseline GFAP concentrations and the proportion of participants with NEDA‐3 at Month 12 is shown in Figure  5 . When included in a multivariable lasso model, baseline GFAP concentration was not an independent predictor of the proportion of participants with NEDA‐3 at Month 12 Figure  S6 .

Regression analysis of the proportion of participants with NEDA‐3 at Month 12 (SUNBEAM and RADIANCE) and baseline log (GFAP) concentration. Relationships were assessed with a binomial logistic regression model using baseline log(GFAP) concentration as a predictor. Terms for study and treatment arm were also included in the model, and NEDA‐3 was defined as the absence of all of the following: relapses, GdE lesions, new/enlarging T2 lesions, and disability progression confirmed over 3 months on EDSS at Month 12. The proportion of participants with NEDA‐3 at Month 12 was higher in the ozanimod 0.46 mg arm (estimate [SE]: 0.2891 [0.1258]) compared with IFN β‐1a (nominal p  < 0.05) but was similar between ozanimod 0.92 mg (estimate [SE]: 0.2286 [0.1282]) and IFN β‐1a (nominal p  > 0.05). All statistical comparisons were hypothesis generating, and there was no adjustment for multiplicity of testing; p values are therefore nominal. EDSS, Expanded Disability Status Scale; GdE, gadolinium‐enhancing; GFAP, glial fibrillary acidic protein; IFN, interferon; NEDA, no evidence of disease activity, defined as no relapses, no EDSS progression (confirmed after 3 months), no new/enlarging T2 lesions, and no new GdE lesions; SE, standard error.

Discussion

In this exploratory, post hoc analysis of participants with RMS from two Phase 3 ozanimod clinical trials, baseline plasma GFAP concentration was a relapse‐independent biomarker of baseline disease severity and a predictor of treatment response. These data therefore provide some support for the use of plasma GFAP as a biomarker for disease severity and response to treatment in participants with RMS.

As a well‐established marker of astrocyte injury and activation in CNS diseases [ 2 ], GFAP could be valuable as a biomarker for on‐treatment outcomes in participants with MS. The data presented here suggest that groups of participants with high baseline plasma GFAP concentrations are more likely to experience poor clinical outcomes during treatment with IFN β‐1a or ozanimod in RMS clinical studies. In addition, with two different statistical models, baseline plasma GFAP was predictive of on‐treatment relapses at Month 12. Future studies assessing GFAP concentrations during a clinical relapse, as a marker of relapse severity instead of or in addition to MRI, may also add to the current data regarding the utility of GFAP as a biomarker for patients with RMS.

Given the complexity of MS pathophysiology, assessing a combination of biomarkers that encompasses the various aspects of astrocytic and neurodegenerative pathology in MS may enhance predictive outcomes. A separate analysis of Phase 3 ozanimod RMS trials found plasma NfL to serve as a biomarker for monitoring and predicting disease activity and treatment response [ 32 ]. Confirming the utility of both GFAP and NfL as biomarkers in MS, a real‐world study found higher levels of serum GFAP concentrations to correlate with subsequent disease progression, whereas serum NfL concentrations were more reflective of acute disease activity in participants with MS at high risk of underlying progressive pathology [ 15 ]. A cohort study that acquired data from a prospective, multicenter trial of patients with MS found serum GFAP concentrations to be a prognostic biomarker for future progression independent of relapse activity while the association of serum NfL with progression was less pronounced, further highlighting the potential of serum GFAP as a complementary marker to serum NfL [ 33 ]. Plasma GFAP, however, may be a more stable marker of disease progression than NfL in patients with MS [ 34 , 35 ]. Use of a combination of biomarkers, such as plasma GFAP and NfL, as predictors of on‐treatment outcomes could be valuable in individualizing treatment decisions for participants with MS by observing changes in biomarker levels in response to various MS therapies. Additionally, assessments of plasma samples would be a more convenient and comfortable method of measuring GFAP concentrations as compared to CSF sampling for patients with MS.

Consistent with the regression analyses from this study, other publications on GFAP as a predictor of clinical outcomes in patients with MS found strong associations between GFAP and MRI outcomes that are known predictors of disease progression, such as T2 lesion count and measures of brain volume [ 14 , 36 , 37 ]. In the present study, MRI variables were included in the multivariable lasso model, whereas in some previously reported studies, adjustments were made only for demographic and clinical severity characteristics, such as age, disease duration, and/or EDSS score [ 14 , 15 , 33 ]. It is possible that GFAP did not contribute additional information in the lasso model prediction of some clinical outcomes because this information was already captured by MRI parameters. A lasso regression model was used for variable selection in this analysis because it can efficiently address the potential association between covariates, such as collinearity [ 38 ]. This effect may explain the discrepancies between the initial regressions and the lasso regression results in this study.

This analysis has a number of strengths, including use of combined data from two randomized, Phase 3 clinical trials of ozanimod that enrolled large cohorts of participants with RMS. In addition, the Simoa assay used to quantify plasma GFAP concentrations is highly sensitive [ 39 ].

One key limitation is the exploratory, post hoc nature of this analysis. The results should be interpreted as exploratory and are not intended to be declarative. In addition, p values were not adjusted for multiple comparisons and are thus interpreted as nominal, and log transformation of GFAP levels in the regression models may be challenging to interpret. As these are trial‐level data that may not correspond to patient‐level outcomes, further validation of these findings is needed to assess clinical relevance.

Overall, this exploratory, post hoc analysis of the Phase 3 SUNBEAM and RADIANCE ozanimod trials adds to the current body of knowledge that GFAP is an informative biomarker of relapse‐independent baseline MS disease severity and a predictor of response to treatment in participants with RMS.

Author Contributions

All authors contributed to the study's concept and design, data acquisition, analysis, or interpretation, and manuscript drafting or substantive revision, and all authors approved the submitted version.

Conflicts of Interest

S.H.: employee and shareholder of Bristol Myers Squibb.

J.K.S. and R.M. : former employees of Bristol Myers Squibb.

G.C.: reports compensation for consulting and/or speaking activities from Almirall, Biogen, Celgene, EXCEMED, Forward Pharma, Genzyme, Merck, Novartis, Roche, Sanofi, and Teva.

B.A.C.C.: personal compensation for consulting from Alexion, Atara, Autobahn, Avotres, Biogen, Boston Pharma, EMD Serono, Gossamer Bio, Hexal/Sandoz, Horizon, Immunic AG, Kyverna, Neuron23, Novartis, Sanofi, Siemens, and TG Therapeutics, and research support from Genentech and Kyverna.

D.L.A.: consulting fees from Biogen, Biohaven, Bristol Myers Squibb, Eli Lilly, EMD Serono, Find Therapeutics, Frequency Therapeutics, GlaxoSmithKline, Idorsia Pharmaceuticals, Kiniksa Pharmaceuticals, Merck, Novartis, Race to Erase MS, Roche, Sanofi‐Aventis, Shionogi, and Xfacto Communications; equity interest in NeuroRx.

L.S.: reports consulting for AbbVie, Atreca, Bristol Myers Squibb, EMD Serono, Novartis, Pasithea, Teva, TG Therapeutics, and 180 Life Sciences; research support from Atara and Bristol Myers Squibb.

H.S.: reports compensation from Bristol Myers Squibb, F2G, GlaxoSmithKline, CellCentric, and Gossamer Bio.

L.K.: received no personal compensation. His institutions (University Hospital Basel/Stiftung Neuroimmunology and Neuroscience Basel) received payments for steering committee, advisory, and data safety monitoring board participation, consultancy services, and educational activities from Bayer, Biogen, Bristol Myers Squibb, Celltrion, Clene Nanomedicine, EMD Serono Research and Development, Galapagos, Genentech, Immunic, Janssen, Kiniksa Pharmaceuticals, Laboratoires Juvise Pharmaceuticals, Merck Healthcare, Merck Sharp & Dohme, Minoryx Therapeutics, Neurostatus‐UHB, Novartis, Roche, Sanofi, Shionogi, Wellmera, Zai Lab; research support from Novartis, Roche, and Innosuisse.

J.A.C.: personal compensation for consulting for Astoria, Atara, Biogen, Bristol Myers Squibb, Convelo, and Viatris.

Supporting information

Data S1 : Supplementary Information