Monocyte NOTCH2 expression predicts IFN-β immunogenicity in multiple sclerosis patients

Conflict of interest: BH served on scientific advisory boards for F. Hoffmann-La Roche Ltd., Novartis, Bayer AG, and Genentech; he has served as DMSC member for AllergyCare; he or his institution have received speaker honoraria from Biogen Idec, Teva Neuroscience, Merck Serono, Medimmune, Novartis, Desitin, and F. Hoffmann-La Roche Ltd.; his institution has received research support from Chugai Pharmaceuticals and Hoffmann-La-Roche; he holds part of 2 patents — 1 for the detection of antibodies and T cells against KIR4.1 in a subpopulation of MS patients and 1 for genetic determinants of neutralizing antibodies to interferon β. EKH has received honoraria/research support from Actelion, Biogen, Czech Ministry of Education project PROGRES Q27/LF1, Sanofi Genzyme, Merck Serono, Novartis, Roche, and Teva.


Introduction
Multiple sclerosis (MS) is a multifocal demyelinating disease of the CNS affecting about 2.5 million people worldwide. Although the disease course is highly variable, the majority of patients develop irreversible disability, and MS remains a major cause of neurological disability in young adults (1). Disease etiology remains uncertain but involves a combination of genetic and environmental cues (2) associated with myelin-specific Multiple sclerosis (MS) is an autoimmune disease characterized by CNS inflammation leading to demyelination and axonal damage. IFN-β is an established treatment for MS; however, up to 30% of IFN-β-treated MS patients develop neutralizing antidrug antibodies (nADA), leading to reduced drug bioactivity and efficacy. Mechanisms driving antidrug immunogenicity remain uncertain, and reliable biomarkers to predict immunogenicity development are lacking. Using highthroughput flow cytometry, NOTCH2 expression on CD14 + monocytes and increased frequency of proinflammatory monocyte subsets were identified as baseline predictors of nADA development in MS patients treated with IFN-β. The association of this monocyte profile with nADA development was validated in 2 independent cross-sectional MS patient cohorts and a prospective cohort followed before and after IFN-β administration. Reduced monocyte NOTCH2 expression in nADA + MS patients was associated with NOTCH2 activation measured by increased expression of Notch-responsive genes, polarization of monocytes toward a nonclassical phenotype, and increased proinflammatory IL-6 production. NOTCH2 activation was T cell dependent and was only triggered in the presence of serum from nADA + patients. Thus, nADA development was driven by a proinflammatory environment that triggered activation of the NOTCH2 signaling pathway prior to first IFN-β administration.
autoreactive CD4 + T cells primed in the periphery and mediating a complex cascade of events leading to progressive neurodegeneration. Recent evidence also implicates wider immune cell involvement from both the adaptive and innate arms of the immune response, including B cells and monocytes (3)(4)(5)(6)(7).
In the last 25 years, many disease-modifying drugs (DMD) have become available for patients with MS that aim to prevent, rather than repair, tissue injury (8). IFN-β (either IFN-β-1a or IFN-β-1b) was one of the first biopharmaceuticals approved for clinical use and has been a first-line DMD in patients with relapsing remitting (RR) MS since 1993 (9)(10)(11)(12).
However, all protein biopharmaceuticals, including IFN-β, can induce an immune response (immunogenicity) associated with the production of antidrug antibodies (ADA). The consequences of immunogenicity range from transient ADA to persistant high-titer ADA with clinical effects (13). Circulating ADA can alter the bioavailability of the drug, either shortening or increasing the elimination half-life of the molecule. Importantly, some ADA have neutralizing activity (neutralizing ADA [nADA]) -leading to reduced drug efficacy (14) -and, in some cases, can induce autoimmunity to endogenous molecules (15). Between 5% and 30% of patients with MS treated with IFN-β develop nADA within the first 9-18 months of therapy. In these patients, nADA have been reported to reduce the cellular response to IFN-β, impacting the therapeutic outcomes of the treatment (16). The formation of ADA can depend on both properties of the product, such as impurity and contamination 332 markers on monocytes, B cells, and CD4 + T cells from healthy donors (HCs, n = 10), treatment naive (n = 10), and IFN-βtreated (n = 15) patients with multiple sclerosis (discovery cohort, Table 1).
originating from the manufacturing process (17), and patient-related features, such as specific HLA alleles (18). In addition, differences in immune cell phenotype and activation status could also play an important role during ADA induction. However, no predictive biomarkers are established to improve treatment choices for patients.  (61). Three cohorts were recruited: a cross-sectional discovery cohort, which was either IFN-β-treatment naive (MS-N) or treated with IFN-β for at least 12 months (MS-T); a cross sectional validation cohort, which included MS-T patients on IFN-β for at least 12 months; and a prospective cohort, which included MS patients untreated at sampling and follow-up for 12 months before neutralizing antidrug antibody (nADA) determination. Cohorts recruited as part of the ABIRISK consortium (www.abirisk.eu/), with samples collected from patients before their first treatment with IFN-β and after 12 months of treatment. In the discovery cohort, a statistical difference was detected between Expanded Disability Status Scale (EDSS) scores from MS-N and MS-T patients (2-tailed t test; P ≤ 0.05). No statistically significant differences were observed in all other clinical/demographic characteristics between the patient groups at baseline (2-tailed t test and χ 2 test; P >0.05). A Scores on the EDSS range from 0-10, with higher score indicating worse disability. B Patients received either IFN-β-Ia or IFN-β-Ib preparations. IQR, interquartile range; N/A, not available.
In the present study, high-throughput flow cytometry technology (LegendScreen) and a systems immunology approach was applied to identify potential predictive biomarkers of nADA development in MS patients treated with IFN-β. Using a systematic analysis framework accounting for the influence of disease progression and IFN-β activity, a phenotypic signature associated with nADA development was identified. Importantly, reduced neurogenic locus notch homolog protein-2 (NOTCH2) expression on monocytes was associated with the presence of nADA even before the initiation of IFN-β therapy and was instrumental in predicting the eventual development of nADA in a prospective cohort of MS patients. Moreover, reduced NOTCH2 expression was associated with increased activation of NOTCH2 signaling and polarization of monocytes toward a proinflammatory phenotype. Thus, we highlight a potentially new tool to assess the response of patients to IFN-β therapy and identify potential immunological mechanisms involved in the development of immunogenicity.

Results
LegendScreen high-throughput flow cytometry identified distinct peripheral blood immune cell signatures. The Leg-endScreen platform (BioLegend) was validated as a tool to assess the surface signature of immune cell subsets in peripheral blood mononuclear cells (PBMCs) from human healthy control (HC) donors. PBMCs prelabeled with lineage-specific fluorescently conjugated antibodies identifying CD4 + T cells, B cells, and monocytes were analyzed in tandem with the 332 cell surface markers of the LegendScreen array ( Figure  1A and Supplemental Figure 1A; supplemental material available online with this article; https://doi. org/10.1172/jci.insight.99274DS1 for experimental and gating strategy). LegendScreen inter-and intra-assay reproducibility was verified by assessing expression of the 332 markers on PBMCs from the same HC analyzed on 3 different days. Both the expression of the markers on each cell subset (Supplemental Figure 1B) and the mean fluorescence emissions of plates 1-4 (P1-P4 of the LegendScreen array) from each run (Supplemental Figure 1C) were highly correlated. Also, using known lineage-specific markers, the specificity of the platform for each cell subset (monocytes, B cells, and CD4 + T cells) was corroborated (Supplemental Figure 1, D-F). Unsupervised hierarchical cluster analysis showed that expression levels of the 332 markers making up the LegendScreen array could discriminate between monocytes, B cells, and CD4 + T cells ( Figure 1B). However, many of the 332 markers were negative in all subsets tested. This was confirmed when LegendScreen signatures from HCs and patients with MS were combined ( Figure 1C). Therefore, to increase the statistical power of our analyses, the markers consistently negative in all samples tested were excluded (Supplemental Figure 2, A and B, and Methods). Using this approach, 211 monocyte, 170 B cell, and 125 CD4 + T cell markers were retained and used for all future analyses (Supplemental Table  1). Thus, the LegendScreen platform was robust and could be used to stratify human PBMCs into biologically relevant groups and identify proteins differentially associated with specific cell types.
Immunogenicity development in IFN-β-treated MS patients was associated with a specific monocyte signature. We used the validated LegendScreen platform to investigate immune signatures associated with nADA in 28 MS patients treated with IFN-β for at least 12 months (MS-T) who had (nADA + ) or had not (nADA -) developed immunogenicity to the drug ( Table 1, cross-sectional discovery cohort). A stepwise approach was used. Firstly, differentially expressed markers (DEMs) between nADA + and nADA -MS-T patients were identified using 2-tailed unpaired t test analysis and FDR <0.2 ( Figure 2A and Supplemental Table 2). The majority of DEMs were expressed on monocytes (n = 23), with only 3 B cell markers and no CD4 + T cell markers identified.
Next, a systematic analysis framework was implemented to discriminate between DEMs associated with the presence of nADA alone rather than disease progression and/or the effect of IFN-β therapy (Figure 2B). Five DEMs expressed on monocytes (mesenchymal stem cells and neural progenitor cells antigen [MSC-NPC], CD138, TIM3, CD254, and C-type lectin domain family 9-A [CLEC9A]) correlated significantly with MRI evidence of MS disease activity in the brain (measured as change in quantitative MRI [QMRI] T1 and T2 lesion volume in the previous 12 months) (19) and were removed from further analysis ( Figure 2, C and D). For all other DEMs, no significant correlations with T1 and T2 lesion volume were observed (Supplemental Table 3).
DEMs associated with IFN-β activity were identified by comparing their expression in nADA + , nADA -, and IFN-β-treatment naive (MS-N) MS patients ( Table 1, discovery cohort) using 1-way ANOVA. DEMs whose expression was significantly different when comparing nADAand MS-N patients, but not between nADA + and MS-N ( Figure 2E, Table 2, and Supplemental Figure 3), were considered to be associated  Table 1, discovery cohort) and who were either nADA -(n = 12) or nADA + (n = 16) were analyzed using LegendScreen. Volcano plots comparing fold change (Log 2 ) in marker expression against the significance of marker expression between nADAand nADA + MS patients in monocytes, B cells, and T cells (Log 10 P value). Only markers listed in Supplemental Table 1 were included in the analysis. Horizontal dashed line represents P value threshold (P = 0.05), vertical dashed lines represent fold change in marker expression (dashed fold change, 0; dotted blue lines fold change, ±1). Markers significantly different between nADA -/nADA + after 0.20 FDR and with a Log 2 fold change ≥ ±0.4 were selected for further analysis (Supplemental Table 2). (B) Summary flow chart of the screening strategy used to identify the nADA-associated phenotypic signature. (C and D) Correlation of differentially expressed markers (DEMs) selected in A and Supplemental Table 2 with disease progression assessed by MRI measurements T1 (C) and T2 (D). Changes in brain lesions detected by comparing MRI scanning at the time of PBMC isolation to MRI scans performed 12 months before sampling. Spearman's correlation coefficient (r) (see with the effect of IFN-β and removed from further analysis. Thus using high-throughput flow cytometry, an immune signature associated with nADA development in a cohort of IFN-β-treated MS patients and exclusively expressed on CD14 + monocytes was pinpointed (Supplemental Table 4).
To evaluate whether altered surface expression of the nADA-associated markers was related to changes in gene expression, quantitative PCR (qPCR) analysis was performed on RNA isolated from FACS-sorted CD14 + monocytes from nADAand nADA + MS patients ( Figure 2F). No difference in gene expression was found in the NOTCH2, CD200R, CD284, CD262, CD317, and CD245 between nADAand nADA + MS-T patients, suggesting that changes in cell surface expression may be mediated by posttranscriptional regulation mechanisms. However, MxA and CD169/SIGLEC1 (genes known to respond to refs. 14,20) were significantly upregulated in monocytes from nADAcompared with nADA + MS patients, showing that the presence of nADA blocks the effect of IFN-β therapy in these cells. Thus, the results suggest that changes in cell surface expression did not correlate with gene transcription and RNA detection and may be mediated by posttranscriptional regulation mechanisms (21,22).
Reduced NOTCH2 surface expression on monocytes was associated with nADA development in MS patients treated with IFN-β. To validate the identified nADA-associated signature, a second independent cross-sectional cohort comprising 23 relapsing-remitting MS (RRMS)/clinically isolated syndrome (CIS) patients treated for 12 months with IFN-β and tested for the presence of nADA at sampling (Table 1, validation cohort; Figure 3A) was analyzed for the expression of the markers listed in Supplemental Table 4. This analysis confirmed that NOTCH2, CD245, CD169, CD262, and CD317 were associated with nADA development; however, CD200R and CD284 were not found to be differentially expressed in this cohort and were excluded from the signature. Therefore, 5 markers expressed on monocytes were confirmed to be associated with the presence of nADAs in MS patients treated for over 12 months with IFN-β (Table 3).
Finally, the expression of the remaining markers was examined in PBMCs collected from a prospective cohort of 18 newly diagnosed RRMS/CIS patients before the first injection of IFN-β and followed up for 12 months (Table 1, prospective cohort). Marker expression was evaluated blindly and stratified according to the presence of nADA assessed after 12 months of IFN-β treatment (nADA prospective + [nADA p + ] or nADA p -). Strikingly, only NOTCH2 was significantly reduced on the surface of CD14 + monocytes from untreated MS patients that went on to develop nADA after 12 months of IFN-β therapy ( Figure 3B). IFN-β treatment did not influence the expression of NOTCH2 on monocytes overtime, nor did the type of treatment or the nADA induction and titre (Supplemental Figure 4, A-D, and Supplemental Table 6). Therefore, these data suggest that reduced detection of NOTCH2 on CD14 + monocytes could be a potential biomarker to predict nADA development in MS-N patients.
Combining NOTCH2 surface expression and monocyte subset frequency predicted nADA development in IFN-βtreated MS patients. To gain insight into the pathophysiological significance of monocyte NOTCH2 expression associated with nADA induction, we analyzed the expression of NOTCH2 in functional monocyte subsets. Human monocytes can be subdivided into 3 subsets based on the expression of CD14 and CD16 in peripheral blood; namely, classical (CD14 + CD16 − ), intermediate (CD14 + CD16 + ), and nonclassical (CD14 lo CD16 + ) monocytes (23,24). NOTCH2 signaling is known to regulate monocyte cell fate by driving monocyte differentiation from the mainly phagocytic classical subset toward more proinflammatory, intermediate, and nonclassical phenotypes (25). Therefore, we assessed subset frequency in both the validation and prospective MS patient cohorts (26,27) (Table 1 and Figure 4A for gating strategy). MS-T patients with established nADA (nADA + ) had significantly elevated numbers of proinflammatory nonclassical monocytes (CD14 lo CD16 + ) and reduced classical monocytes (CD14 + CD16 -) compared with nADA -MS-T patients (Figure 4, B and C). The increase in the nonclassical monocyte subset in nADA + patients was not associated with disease activity (Supplemental Figure 5). Patients from the prospective cohort recruited before IFN-β administration and classified as nADA + after 12 months of IFN-β therapy (nADA p + ) had significantly elevated intermediate also Supplemental Table 3). (E) DEM expression in treatment-naive (MS-N; n = 15), nADA -(n = 9), and nADA + (n = 11) IFN-β-treated (MS-T) patients. Box and whisker plots showing expression of NOTCH2, CD200R, CD169, CD284, CD245, CD262, and CD317 on monocytes that passed inclusion criteria in Table  2. One-way ANOVA with Tukey's correction for multiple comparisons, *P ≤ 0.05, **P ≤ 0.01, ***P ≤ 0.001, ****P ≤ 0.0001. See Supplemental Figure 3 for analysis of excluded markers. (F) Transcriptional regulation of the selected nADA-associated markers (Supplemental Table 4). mRNA was isolated from FACS-sorted CD14 + monocytes from nADA -(n ≥ 3) and nADA + (n ≥ 3) MS-T patients and analyzed for CD169, NOTCH2, CD200R, CD284, CD262, CD317, and CD245 and for IFN-β-responsive gene MxA as a control analyzed by qPCR relative to cyclophilin or GAPDH. Fold change compared with HCs shown. Twotailed t tests, **P = 0.01. Box plots show the median and 25th and 75th percentiles; whiskers represent minimum and maximum values.
rather than nonclassical monocytes compared with nADA p patients ( Figure 4, D and E). Correspondingly, NOTCH2 expression was reduced in all monocyte subsets in nADA + and nADA p + patients compared with nADAand nADA p patients, respectively (Figure 4, F and G). This suggested that reduced NOTCH2 expression and altered monocyte subset frequencies could be a feature in both patients who had already developed nADA and in those patients that would go on to develop nADA after IFN-β treatment. Scatterplot 3-dimensional (3-D) and unsupervised hierarchical cluster analysis using intermediate and nonclassical monocyte frequency and total monocyte NOTCH2 expression levels clustered MS-N patients from the prospective cohort into 2 distinct groups; 1 group that went on to develop nADA (nADA p + ) and 1 group that did not (nADA p -) ( Figure 4, H and I), suggesting that these parameters could be used to stratify patients before they receive treatment to identify those at greatest risk of nADA development. In support of this, receiver operating characteristic (ROC) curves constructed using a predictor built using NOTCH2 expression and nonclassical monocyte subset frequency in the discovery cohort and applied to predict ADA status in the validation cohort show an AUC value of 0.88, indicating a higher predictive power than the NOTCH2 expression alone (NOTCH2 mean fluroescence intensisty [MFI] and NOTCH2 percentage AUC values 0.85 and 0.70, respectively) ( Figure 4J). Moreover, repeating the same approach using patients from the prospective cohort to build the predictor model (a logistic regression model) and applying the predictor model to treatment-naive MS patients whose baseline samples were analyzed using the LegendScreen platform (Table 1), we identified that NOTCH2 MFI could discriminate nADA p and nADA p + patients before initiation of therapy (AUC = 0.75) (Supplemental Figure 6).
Serum factors drive increased NOTCH2 activation in nADA + and nADA p + MS patients. Reduced detection of surface NOTCH2 in monocytes (Figure 4, B-G) could be due to increased binding to NOTCH2 ligands, triggering the expression of specific NOTCH2 target genes -including HES1 (hairy and enhancer of split-1) and DTX1 (Deltex1) (28)(29)(30). Indeed, NOTCH2-responsive genes HES1 and DTX1 were upregulated ( Figure 5, A and B), and the mRNA and protein expression of proinflammatory IL-6 were increased in monocytes from nADA + and nADA p + MS patients compared with nADAand nADA p patients, respectively (Figure 5, C-G). Thus, in MS patients who had developed or who went on to develop nADA, NOTCH2 appeared to be preferentially activated, potentially driving proinflammatory cytokine production and monocyte differentiation. Confirming this, when HC monocytes were cultured in the presence of serum from nADA + /nADA -MS patients, only nADA + serum induced increased HES1 and DTX1 expression, an effect that was blocked by inhibiting NOTCH2 activation using a specific inhibitor of γ-secretase (N-[N-(3, 5-difluorophenacetyl)-l-alanyl]-S-phenylglycine t-butyl ester; DAPT) ( Figure 5, H-J). Moreover, the induction of NOTCH2 target genes by nADA + patient serum was associated with reduced transcription  (Table 1, discovery cohort) using one-way ANOVA (See Figure 2E, and Supplemental Figure 3) of CD163, a marker of classical monocytes, and increased expression of genes associated with inflammation and nonclassical monocyte subsets including STAT1 ( Figure 5K and Supplemental Figure 7) (31,32).
Finally, the mechanism of monocyte NOTCH2 activation was assessed in a coculture experiment (Figure 5L). Induction of HES1 by nADA + serum was only observed in monocytes cocultured with T cells -not when monocytes were cocultured with B cells or in monocyte-only cultures ( Figure 5L). Therefore, the data suggest that patients with nADA had serum factors that triggered the activation of NOTCH2 on monocytes in a non-cell intrinsic manner and that NOTCH2 activation could initiate the differentiation of proinflammatory monocyte subsets and provide an environment favoring ADA development ( Figure 6). Thus, patients who had this proinflammatory signature prior to their first treatment with IFN-β could be predisposed to develop nADA.

Discussion
Systems biology approaches are used increasingly to identify molecular signatures that orchestrate complex immunological processes. Such studies support the concept that multiplex measurements, including gene expression, high-density analysis of cell populations, and cellular and serological responses, can predict the type and magnitude of human immune responses (33). Here, high-throughput flow cytometry was used to identify a unique phenotypic signature associated with the development of neutralizing antibodies to IFN-β in patients with MS. Notably, reduced expression of NOTCH2 on monocytes was associated with nADA development in patients with MS treated with IFN-β and was also identified as a potential predictive marker of nADA development in patients who were drug naive using a prospective patient cohort. Serum factors in patients with nADA triggered NOTCH2 activation and subsequently reduced  Table 1) were analyzed for the expression of NOTCH2, CD200R, CD284, CD169, CD245, CD262, and CD317 on CD14 + monocytes using flow cytometry. Two-tailed unpaired t tests, *P ≤ 0.05, **P ≤ 0.01, ***P ≤ 0.001, ****P ≤ 0.0001. (B) PBMCs from treatment-naive (MS-N) patients (n = 18) taken before the first injection of IFN-β (prediction cohort, Table 1) were analyzed for the expression of NOTCH2, CD169, CD245, CD262, and CD317 on CD14 + monocytes using flow cytometry. The expression of nADA-associated markers was stratified according to the presence of nADA in the same patients after they had been treated for 12 months with IFN-β; nADA prospective-positive (nADA p + ) (n = 11) and nADA p -(n = 7). Mann Whitney U test, *P < 0.05. Box plots show the median and 25th and 75th percentiles; whiskers represent minimum and maximum values.
detection of surface NOTCH2 in monocytes. This process drove the differentiation of proinflammatory monocyte subsets, an effect that required T cells but not B cells. Thus, this work has identified a potentially novel mechanism influencing the development of immunogenicity to biopharmaceuticals and a potential biomarker to predict antidrug responses. The role of NOTCH2 and monocyte subsets in the development of nADA in the wider context of other biopharmaceutical drugs and in different diseases is, at present, under investigation (34)(35)(36).
NOTCH2 signaling is a highly conserved pathway involved in cell fate decisions, proliferation, and survival (37). Mammalian cells have 4 NOTCH receptors (NOTCH-1 to -4) and 5 NOTCH ligands (Delta-like-1, -3, and -4 and Jagged-1 and -2). NOTCH/NOTCH-ligand binding induces γ-secretase-mediated cleavage and shedding of the NOTCH extracellular domain, activation of the NOTCH signaling pathway, and induction of NOTCH target genes, including HES1 and DTX1 (28,30,38). Our data revealed that NOTCH2 gene expression was similar between nADAand nADA + MS patients, while membrane expression was reduced in the latter, most likely indicating an association between NOTCH pathway activation and nADA induction. Alternatively, posttranslational modification of the NOTCH extracellular domain, which has been reported to control the strength of NOTCH-pathway signaling events and the selective interaction between NOTCH and NOTCH ligands, could also account for the reduced detection of membrane NOTCH2 in nADA + MS patients (39,40). However, our finding that CD14 + monocytes isolated from nADA + MS patients had significantly increased expression of both HES1 and DTX1 genes compared with both nADA -MS patients and HCs supported the hypothesis that the reduced expression of membrane NOTCH2 was due to increased receptor activation. NOTCH target gene transcription can also be stimulated by TLR-2 and TLR-4 activation with LPS and the fungal analog zymosan, respectively, independently of NOTCH receptor activation (41,42). However, our observation that nADA + , but not nADA -, patient serum could upregulate HES1 and DTX1 transcription in HC monocytes and that this could be inhibited by pretreating PBMCs with the NOTCH-specific γ-secretase inhibitor DAPT demonstrated the direct association between NOTCH receptor, NOTCH target gene activation, and the loss of tolerance to IFN-β in MS. The fact that the surface expression of all 4 mammalian NOTCH receptors on CD14 + monocytes was measured in the discovery cohort and only the surface expression of NOTCH2 was found to be significantly different between the nADA + and nADA -MS patients groups suggested that NOTCH2 activation in CD14 + monocytes was associated with the generation of ADA against IFN-β in MS.
Human monocytes can be subdivided into 3 subsets based on the expression of CD14 and CD16 and characterized by a specific gene and microRNA (miRNA) signature (23,31). The classical (CD14 + CD16 − ) monocyte subset constitutes 85%-90% of the circulating monocyte pool, whereas the remaining 10%-15% consist of intermediate (CD14 + CD16 + ) and nonclassical (CD14 lo CD16 + ) monocytes (31,43). In humans, monocyte precursors differentiate into classical monocytes in the BM and further polarize toward intermediate and nonclassical monocytes in the peripheral blood (24). While the function of monocyte subsets is still debated, recent evidence suggests that classical monocytes are mainly phagocytic and not inflammatory, while nonclassical monocytes produce inflammatory cytokines and exhibit proprieties of antigen presentation, and the intermediate subset may be transitional with both inflammatory and phagocytic functions (44). Indeed, NOTCH2 activation regulates the in vivo differentiation of Ly6C lo monocytes, the murine equivalent of the human nonclassical monocytes, from peripheral LyC hi (murine classical monocytes) monocytes (25). This suggests that the observed upregulation of NOTCH2 signaling in nADA + patients could be responsible for the induction of nonclassical monocytes in this patient subset. We show  Monocytes from nADA + and nADA p + patients were also characterized by increased expression of the proinflammatory cytokine IL-6 compared with nADAand nADA p -, respectively. Taken together, these data support a link between NOTCH2 activation and the differentiation of monocytes from classical (mainly phagocytic) to proinflammatory intermediate and nonclassical subsets (24,25).
The activation and recruitment of peripheral blood monocytes to the CNS is a hallmark of active disease associated with ongoing demyelination and axonal injury, underlining the role of myeloid cells in the disease pathophysiology (8,45). An association between MS disease activity and circulating myeloid cell activation and proinflammatory cytokine release has been demonstrated previously (46)(47)(48)(49)(50). Higher frequencies of IL-6 and IL-12 secreting monocytes have been reported in MS patients compared with healthy donors (46). Moreover, the frequency of IL-12 releasing monocytes has been shown to correlate with the presence of gadolinium-enhancing MRI in patients (47). However, the evidence surrounding perturbations in circulating monocyte subset frequencies in MS patients has been inconsistent (48,49). Previously, IFN-β treatment has been shown to either induce accumulation of circulating CD16 + intermediate and nonclassical monocytes (49) or to increase the intermediate subset frequency at the expense of nonclassical monocytes in MS (50). Here, we observed increased CD16 expression on CD14 + monocytes from MS-T compared with untreated patients, suggesting that IFN-β directly induced CD16 expression on monocytes. Moreover, the frequency of intermediate monocytes between nADA + and nADA -MS patients was comparable, suggesting that the reduced frequency of nonclassical (CD14 lo CD16 + ) subsets in the nADA + patients was not due to the effect of IFN-β.
It is known that the E. coli-derived nonglycosylated IFN-β-1b is more immunogenic than the fully glycosylated IFN-β-1a produced in mammalian CHO cells (51). The majority of nADA + patients in our cohorts were treated with the more immunogenetic form of IFN-β-1b; this did not allow us to clearly discriminate the effect of the different preparations on the expression of NOTCH2. Furthermore, although NOTCH2 expression on monocytes was a good predictor of immunogenicity against IFN-β, our study was not powered enough to establish a clear cut-point to use in routine clinical practice to identify patients likely to develop nADA. Further studies are warranted to identify such cut-point values.
The stepwise approach used here was designed to identify DEMs associated with nADA, excluding any marker associated with disease progression (e.g., TIM-3) (52) or response to IFN-β therapy (e.g., CD274, CD123, CD129, ect.) (53)(54)(55). IFN-β-neutralizing antibodies only partially explain IFN-β nonresponse in MS (56). Qualitative and quantitative differences in the molecular response to the drug, genetic variants in IFN-β receptors, or signaling components and perturbations of monocytes and the innate immune system are all reported to be associated with reduced response to therapy (57)(58)(59)(60). While -for the purposes of this study -markers associated with disease progression or response to IFN-β therapy were excluded, they could play an important role in MS pathogenesis and/or response to treatment and will be the subject of further investigations.
In summary, we show for the first time to our knowledge that reduced surface expression of NOTCH2 and increased NOTCH intracellular signaling in monocytes was associated with increased frequency of nonclassical (CD14 lo CD16 + ) monocytes and nADA induction in IFN-β-treated MS patients. This signature was identified in both cross-sectional and prospective patient cohorts and supports the hypothesis  HC PBMCs were cultured with 10% serum from MS-T nADA + or nADApatients. Following 18-hour culture, mRNA was isolated from purified monocytes for qPCR analysis. (I) Cumulative expression of NOTCH target genes HES1 and DTX1 relative to cyclophilin A in nADA -(n = 6) or nADA + (n = 5) MS-T patients evaluated by qPCR. Mann Whitney U test, *P < 0.05. (J) PBMCs were precultured for 16 hours ± N-[N- (3,5-difluorophenacetyl-L-alanyl)]-S-phenylglycine t-butyl ester (DAPT) or vehicle (DMSO) before 18-hour culture with 10% serum from nADA + /nADApatients (n = 4). Cumulative expression of HES1 and DTX1 relative to cyclophilin A. Mann Whitney U test, *P < 0.05, **P ≤ 0.01, ***P ≤ 0.001. (K) Cumulative expression of genes previously reported to be associated to classical (CD163) or nonclassical (STAT1) monocyte subsets (32). (L) HES1 expression in sorted CD14 + monocyte cultures and cocultures of CD14 + monocytes and CD19 + B cells or CD14 + monocytes and CD4 + T cells (1:1 ratio) after culture in presence of serum from MS-T nADA + or nADApatients (n = 7). Mann Whitney U test. *P < 0.05. Box plots show the median and 25th and 75th percentiles; whiskers represent minimum and maximum values. that a preexisting variation in the immune profile and cellular make-up of MS patients may contribute to immunogenicity against IFN-β and could be used as a biomarker to predict response to biopharmaceuticals in a clinical setting.

Methods
Patient cohorts. Patients were diagnosed with RRMS or CIS according to the revised McDonald criteria (61). Three cohorts were recruited and summarized in Table 1. Briefly, PBMCs and serum were collected from a discovery cohort, either as MS-N or treated with IFN-β for at least 12 months (MS-T). A validation cohort of MS-T patients and a prospective cohort were recruited as part of the Anti-Biopharmaceutical Immunization: prediction and analysis of clinical relevance to minimize the RISK consortium (ABIRISK consortium; www.abirisk.eu/), with samples collected from patients before their first treatment with IFN-β and after 12 months of treatment. Patients included in the study were treated with both E. coli-derived IFN-β-1b (Betaferon/Extavia, n = 33) or mammalian-derived IFN-β-1a (Avonex, n = 8; Rebif, n = 43) products. PBMCs and serum were collected from age-, sex-, ethnicity-, disease duration-, and smoking status-matched HC. In order to reduce drug interference with the expression of immune cell markers, all samples were collected 14 hours after the last administration of IFN-β. PBMCs were isolated using density gradient centrifugation and cryopreserved in liquid nitrogen until use. Serum samples were used to test for the presence of nADA.
MRI acquisition and lesion volumetric assessment. All MRI scans analyzed in the study were performed on the same scanner (1.5 T Gyroscan; Philips Medical Systems) in the Department of Radiodiagnostics at General University Hospital in Prague with the same protocol. The standardized protocol consisted of 2 sequences: fluid-attenuated inversion recovery (FLAIR) and T1-weighted 3-dimensional fast field echo (T1-WI/FFE 3D). Contiguous slices covering the whole brain were acquired with the following parameters.  LegendScreen screening. PBMCs were analyzed using LegendScreen Human Cell Screening Kits (BioLegend; http://www.biolegend.com/legendscreen), according to the manufacturer's protocol with the following modifications. Briefly, 3 × 10 7 (~9 × 10 4 per antibody) PBMCs were labeled with a 1:5 dilution of each PE-labeled mAb targeting 332 surface antigen and 10 isotype controls, making up the LegendScreen panel. After 30 minutes of incubation at 4°C, cells were washed and incubated with CD4-V500 (RPA-T4; BD Biosciences), CD19-APC-Cy7 (SJ25C1; BD Biosciences), and CD14-APC (61D3; eBioscience) to identify CD4 + T cell, B cell, and monocyte subsets for 30 minutes at 4°C. After final washing, cells were fixed before acquisition using the FACSVerse flow cytometer equipped with FACSuite software (Becton Dickinson), and 25,000-40,000 events were acquired per marker. For all samples, aggregates and dead cells were excluded by forward-and side-scatter gating; fluorescence minus one (FMO) controls were used to control for specificity of surface antigen staining. Data were analyzed by manual gating using FlowJo 8.8.7 software (TreeStar Inc.). Percentage of PE-positive cells and PE median MFI were collected for each of the 332 markers screened.
Mixture model for LegendScreen marker selection. The markers expressed by the different cell subpopulations were selected using a 2-component Gaussian mixture model on the log-scale (intensities were highly positively skewed) performed using Rmixmod in R version 3.3.0 and fitted to the data using a stochastic expectation-maximization (EM) algorithm. The mixture modeling was then followed by a 4 class clustering: (a) a unimodal-negative class containing the markers with 2 negative modes; (b) a unimodal-positive class with 2 positive modes; (c) a bimodal class with 1 negative mode and 1 positive mode; and (d) a bimodal class with 2 positive modes. The threshold values were chosen based on several gold-standard negative markers (confirmed in the literature to be negative for each cell subset examined) (64) (Supplemental Table  5 and Supplemental Figure 2, A and B).
Monocyte, T cell, and B cell isolation. Purified cell populations were obtained using negative bead isolation (Stemcell Technologies) according to manufacturer's instructions or FACS sorting.
Statistics. Statistical analysis was performed with the Prism Software (GraphPad), JMP version 12.0.1 software (www.jmp.com), and the R software package Rmixmod (65). The significance of differences among experimental groups was assessed by 2-tailed Student's t test, Mann Whitney U test or 1-way ANO-VA followed by Tukey's post hoc test. Data was presented using box and whisker plots; the box shows the median and 25th and 75th percentiles, and the whiskers represent minimum and maximum values. Correlations were assessed with Spearman's correlation coefficient. P > 0.05 was considered significant; *P < 0.05, **P < 0.01, ***P < 0.001.
In order to assess the capacity of Notch2 and monocyte subset frequency to predict nADA development in MS-N patients, predictors were constructed on the discovery cohort using univariate and multivariate logistic models with NOTCH2 MFI, NOTCH2 percentage-positive monocytes, and frequency of each monocyte subset as explanatory variables. The models with the best fit to the data measured as lowest Akaike Information Criterion (AIC) were applied on the validation cohort to predict the ADA status and build ROC curves, on which the AUC was calculated using the R package ROCR.