Introduction

Multiple sclerosis (MS) is an immune-mediated, chronic neurological disease and a leading cause of disability in young patients1. The clinical course of MS is heterogenous and difficult to predict, with a wide range of symptomatic manifestations between individuals2. Patient phenotypes include clinically isolated, radiologically isolated, relapsing–remitting, primary or secondary progressive, or progressive-relapsing MS, with a broad spectrum of disease severity within this framework1. Other phenotypic classifications have been proposed including clinically isolated syndrome versus relapsing–remitting disease or progressive disease further stratified by active or inactive disease within the past 12 months3.

While clinicians are guided by the McDonald criteria in establishing a diagnosis, several common syndromes often masquerade as MS and misdiagnoses do occur. In one 2016 study by Solomon et al. following 110 patients with an incorrect MS diagnosis, 33% held this misdiagnosis for 10 years or longer, and 31% experienced “unnecessary morbidity as a direct result” of their misdiagnosis according to neurologist assessment, in part due to the consequences of immunotherapy or corticosteroid use4. Although the McDonald criteria were revised in 2017, interrater variability in patient assessment persists, suggesting further opportunities to improve the reliability of MS diagnoses5.

MRI is currently the most important paraclinical tool for MS diagnosis and observation6. Cerebrospinal fluid (CSF) immunoglobulin G analysis and clinical outcome measures (expanded disability status scale, nine-hole peg test, symbol digit modality test, and timed 25-foot walk) are several avenues, among others, available to establish a diagnosis and quantify disease severity7,8. Additionally, the presence of oligoclonal bands and low levels of vitamin D have been shown to portend conversion from clinically isolated syndrome to clinically definite MS9,10. Despite these tools, there are currently no standard criteria to comprehensively measure disease progression or predict clinical phenotype. Identification of additional biomarkers could allow us to better classify and treat newly diagnosed MS patients11. This approach fits within the framework of personalized medicine, a concept postulated to represent the future of neurologic treatment as a less expensive and more effective approach to MS patient care.

Since it bathes all CNS structures, cerebrospinal fluid offers a more direct approximation of cellular activity within the central nervous system than does peripheral blood12. Here, we performed an exploratory analysis of CSF immune protein expression in a cohort of Disease Modifying Therapy (DMT)-naïve MS patients. We chose a commercially available high throughput panel of 92 immunologic analytes encompassing cytokines and markers of cells with known roles in MS pathogenesis, including IL-12, IFNγ, TNF, CD40, and CD8, as well as exploratory proteins without an established role in MS (FGF-19). We aimed to determine relationships between these intrathecal immune proteins and a variety of clinical and radiographic measures relevant to MS phenotypic expression and disease severity. We chose a hypothesis-generating approach in an effort to better understand the underlying immune expression pathways associated with MS pathophysiology.

Results

Cohort description

Fifteen relapsing remitting MS (RRMS) and 5 progressive MS (PMS) patients were included (see patient data in Table 1). Mean age was 42.4 ± 13.7 (SD). Fifteen patients (75%) were female. Mean EDSS (available for 14 patients) was 2.6 ± 1.4, with a median IQR of 2.25 (2.0, 3.0). Of these, one RRMS patient was classified as having “inactive” disease demonstrated by meeting No Evidence of Disease Activity- 3 (NEDA-3) criteria in the past 12 months from LP, defined by (1) no clinical relapses, (2) no progression of EDSS within 6 months, and (3) no radiographic progression on brain MRI (Lu et al. 2018). The remaining 19 patients were considered to have “active” disease, not meeting NEDA-3 within 12 months prior to LP. Additionally, 2 RRMS patients with active disease were diagnosed with Tumefactive features and were subsequently designated in the following analyses as having Tumefactive MS.

Table 1 Patient characteristics and clinical information.
Full size table

Intrathecal immune correlates of clinical disability

CSF, brain MRI, and clinical metrics were collected and analyzed (Fig. 1A). Several intrathecal immune proteins correlated with higher clinical disability (Fig. 1B). Positive correlations were observed between intrathecal IL-4 and Expanded Disability Status Scale (EDSS) or 25-foot walk time (25-FWT), as well as between CD8A and 9-hole peg test (9HPT) (Fig. 1C–E). A significant negative6 correlation existed between Single Digit Modalities Test (SDMT) score and each of the following analytes: CD244, OPG, SCF, Flt3L and 4E-BP1 (Fig. 1F–J).

Figure 1
figure 1

CSF protein correlates of clinical status in MS patients. (A) Study overview. This image was created with BioRender.com (B) Pearson correlation matrix of clinical disability scores and normalized protein expression values from Olink Target 96 Inflammation panel. (C) IL4 correlates with EDSS (note: EDSS within 8 months of lumbar puncture available for n = 14 patients). (D) IL4 correlates with 25-foot walk. (E) CD8A correlates with 9-hole peg test. (F) CD244, (G) OPG, (H) SCF, (I) Flt3L and (J) 4E-BP1 correlate with poorer SDMT performance (n = 20 DMT-naïve MS patients).

Full size image

Intrathecal immune correlates of age

The following immune proteins were positively correlated with age at the time of lumbar puncture: CDCP1, IL7, TWEAK, CASP8, LIF, LIFR, LAP TGF-beta-1, HGF, PD-L1, and OPG (Fig. 2A–J). In contrast, IL-13 was negatively correlated with patient age (Fig. 2K).

Figure 2
figure 2

Intrathecal cytokines associated with age of MS patients at time of Lumbar Puncture. (A) CDCP1, (B) IL7, (C) TWEAK, (D) CASP8, (E) LIF, (F) LIFR, (G) LAP TGF-beta-1, (H) HGF, (I) PD-L1, and (J) OPG increased in greater age. (K) IL-13 was lower in older patients. (n = 20 MS patients).

Full size image

Immune analytes differ by MS subtype and patient sex

Next, intrathecal analytes were compared across disease subtype and biological sex. Male sex was associated with higher FGF19 (Fig. 3A–C,G). Amongst patients with RRMS, female patients had higher IL-10RA and IL-20 than male patients (Fig. 3B,H,I). These differences persisted when accounting for age (supplemental Tables 4a–4c, supplemental Fig. 3). No differences were identified between relapsing remitting and progressive MS subtypes (Fig. 3D). Tumefactive MS exhibited higher CCL4 than their non-Tumefactive counterparts. (Fig. 3E,J). No inflammatory analytes were specific to RRMS (Fig. 3F).

Figure 3
figure 3

Intrathecal cytokines differ by clinical subtype and male/female sex. (A–C) Volcano plots comparing male versus female sex across disease subtypes for (A) all patients, (B) only amongst patients with RRMS, and (C) only patients with PMS. Adjusted p values are as follows: (A) Full cohort: FGF-19 p = 0.0003 (B) RRMS: FGF-19 p = 0.02; IL-10RA p = 0.02; IL-20 p = 0.02 (C) PMS: FGF-19 p = 0.07. (D–F) Volcano plots comparing cytokines across disease subtype. Adjusted p values are as follows: (D) RRMS vs PMS: SLAMF1 p = 0.97 (E) Tumefactive MS vs Other: CCL4 p = 0.02. (F) RRMS vs Other was ns. Panels A–F were calculated using Olink’s volcano plot software. A red data point reflects significance (p < 0.05) after adjusting for multiple testing via Benjamini-Hochberg method. Dotted line indicates significance (p < 0.05) prior to adjusting for multiple testing. (G–J) Analysis of protein biomarkers identified in volcano plots by sex and disease subtypes. (G) FGF-19, (H) IL-10RA and (I) IL-20 expression in males versus females grouped as all patients, RRMS or PMS as indicated. (J) CCL4 expression in tumefactive versus other subtypes. Panels G–J were calculated using unpaired 2-tailed t tests as indicated (*p < 0.05, **p < 0.01, ***p < 0.001). Full cohort: n = 20; RRMS: n = 13; PMS: n = 5; Tumefactive: n = 2. Please refer to Supplemental Tables 1 and 2 for a comprehensive list of analytes and p values.

Full size image

Intrathecal immune correlates of brain segmental volumes by MRI

Immune biomarkers were then compared against volumetric MRI features (Fig. 4). IL-33 was negatively correlated with cortical gray matter (p = 0.02), total gray matter (p = 0.02), thalamic volume (p = 0.049), and caudate volume (p = 0.01) (Fig. 4A–E). IL-33 was positively correlated with total lesion volume (p = 0.03) and ventricular size (p = 0.03) (Fig. 4C & F). Intrathecal CD8a negatively correlated with thalamic (p = 0.03) and caudate volumes (p = 0.04) (Fig. 4G). Older age correlated with decreased gray matter (p = 0.03) and increased ventricular size (vCSF) (p = 0.008). Finally, greater T2 FLAIR lesion volume predicted increased ventricular size (p = 0.002) (supplemental Fig. 1). A comprehensive list of significant associations between protein analytes and MRI brain volumes can be found in supplemental Table 5.

Figure 4
figure 4

MRI features correlate with Inflammatory biomarkers. (A–F) Intrathecal IL-33 versus segmental brain volume measurements as follows: (A) cortical gray matter, (B) total gray matter, (C) ventricular CSF volume, (D) thalamic volume, (E) caudate volume, and (F) lesion volume (T2LV). (G) Intrathecal CD8A versus thalamic (blue) and caudate (purple) volumes. X-axis values are normalized protein concentrations (NPX values) per Olink analysis. (n = 17 MS patients for whom MRI data was available).

Full size image

Clinical disability scores correlate with MRI features

Finally, brain volumes on MRI were compared against clinical disability (Fig. 5). SDMT was negatively correlated with total lesion volume and ventricular size (Fig. 5A,B), and positively correlated with thalamic volume (Fig. 5C). There was no significant association between segmental brain volumes and the other clinical metrics of disability included in this study (supplemental Fig. 2).

Figure 5
figure 5

MRI features correlate with Clinical Metrics of Disability. (A–C) SDMT versus segmental brain volume measurements as follows: (A) T2 FLAIR lesion volume, (B) ventricular volume, and (C) thalamic volume. (n = 17 MS patients for whom MRI data was available).

Full size image

Discussion

The goal of this study was to identify CSF immune biomarkers of clinical and MRI phenotypes in Multiple Sclerosis (MS). Our study approach is comparable to a recent investigation by Huang et al. who also employed Olink to examine plasma and CSF biomarkers with the goal of distinguishing healthy from MS for diagnostic purposes13. Our study, however, comprised a cohort of only DMT-naïve MS patients with a focus on stratification of various phenotypic presentations within the disease.

We identified several immune correlates of clinical disability at the time of lumbar puncture, including those reflecting dysregulated T cell immunity.

First, expanded disability status scale (EDSS) scores and 25-foot walk times (25-FWT) were positively correlated with interleukin 4 (IL-4). Both EDSS and 25-foot walk time reflect ability to ambulate: higher values indicate accruing gross motor disability. IL-4 thus may be a marker of decreased motor function, balance, and stability.

Second, 9-hole PEG test (9-HPT) performance was positively (pathogenically) correlated with CD8a. The 9HPT is an assessment of finger dexterity and fine motor coordination, during which a patient uses one hand to place nine pegs into nine associated holes on a board and subsequently removes them as expeditiously as possible. Test output is time to completion. Thus, higher intrathecal CD8a correlated with slower completion of this task. 9-HPT did not significantly correlate with any other CSF analyte, although did exhibit a positive trend with IL-20RA (p = 0.051, R2 = 0.21).

Third, single digit modalities testing (SDMT) was negatively (pathogenically) correlated with each of the following analytes: CD244 (Natural killer cell receptor 2B4), OPG (Osteoprotegerin), SCF (Stem cell factor), Flt3L (Fms-related tyrosine kinase 3 ligand) and 4E-BP1 (Eukaryotic translation initiation factor 4E-binding protein 1). SDMT is a measure of cognitive processing speed in which a patient is asked to pair numbers and figures within an allotted time limit. A higher score indicates a higher number of successful pairings. Therefore, unlike 9-HPT, EDSS, or 25-FWT, a lower SDMT score indicates poorer performance. SDMT was also negatively correlated with total lesion volume and ventricular size, and positively correlated with thalamic volume, linking brain atrophy with impaired cognitive function.

Interestingly, CSF analytes used for the clinical diagnosis of MS – including oligoclonal band count, IgG index, protein, glucose, and white blood cell count – did not correlate with clinical disability at the time of lumbar puncture.

The findings above underscore the known relevance of T cells in multiple sclerosis. Accumulating evidence in the field of MS has pointed to a role for CD8 + T cells in disease pathogenesis14,15. Indeed, these cells are present in abundance in MS lesions and exhibit TCR affinity for CNS antigens16,17,18,19. It is likely that cytotoxic effector cells in the brain mediate damage associated with clinical phenotypes. Contrarily, the exact role of IL-4, a cytokine classically produced by TH2 skewed CD4+ T cells, in MS remains ambiguous. Some studies in the EAE mouse model have indicated that IL-4 may ameliorate disease20,21, though TH2 skewing in other autoimmune diseases can be pathogenic22. IL-4 and IL-4R polymorphisms have been associated with development of MS23,24, though their particular roles remain unclear.

Another proxy for disease progression or severity is compartmental brain atrophy on MRI. In our cohort, brain volumes correlated with several immune analytes. Patients with increased CD8a exhibited smaller thalamic and caudate volumes, adding further evidence to support the role of CD8 T cell immunity in MS. Notably, higher intrathecal interleukin-33 (IL-33) was broadly associated with atrophy, including diminished cortical gray matter, thalamic volume, and caudate volume as well as larger ventricles and greater total lesion volume.

The IL-33/ST-2 axis is widely present in the brain and in circulating immune cells25, is upregulated in MS26, and may hold a pivotal role in regulating various cell lines important in the disease. Indeed, several studies have demonstrated that IL-33 is elevated in the peripheral blood and brain parenchyma of MS patients relative to healthy subjects25,27,28. However, animal models of experimental autoimmune encephalomyelitis (EAE) have revealed conflicting data regarding IL-33’s role in disease progression29,30,31,32,33, and there are opposing findings regarding its effects on myelination26,34. Our results correlate IL-33 with brain atrophy and radiographic disease accumulation. Several hypotheses may be that IL-33 exacerbates disease burden, that it is a protective cytokine upregulated in response to a more severe disease state, or that it is a biproduct of cell injury and death that is inherent to more advanced disease35.

As a pertinent positive control of our volumetric analysis program, our MRI analyses recapitulated the known relationship between aging and gray matter atrophy (Supplemental Fig. 1). These analyses also demonstrated a positive correlation between total volume of T2 hyperintense lesions and ventricular size (a proxy for central brain atrophy). In a multiple linear regression accounting for age, total T2 lesion volume was still an independent predictor of vCSF (p < 0.01, Supplemental Table 3). Thus, our analyses link both age and cumulative disease burden with loss of central brain matter.

We next analyzed the association between age and immune proteins within our cohort. Analytes upregulated in older age may reflect 1. the natural process of aging, 2. an MS-specific process of aging, or alternatively may be markers of 3. milder or later onset disease. Indeed, within our cohort of only DMT-naïve patients, older patients included those who first presented to a physician for MS symptoms later in their lives, as well as those with who had been diagnosed at younger ages but had mild phenotypes not requiring disease modifying therapy. Except for Subject N who presented with fulminant Tumefactive disease and passed away shortly thereafter, no patient over 45 years old in our study required any form of ambulatory assistance (cane, wheelchair, etc.) at the time of sampling.

Age positively correlated with the following immune markers: LIF (Leukemia inhibitor factor), LIFR (Leukemia inhibitory factor receptor), HGF (Hepatocyte growth factor), CASP-8 (Caspase 8), PD-L1 (Programmed cell death 1 ligand 1), IL7 (Interleukin-7), CDCP1 (CUB domain-containing protein 1), TWEAK (Tumor necrosis factor (Ligand) superfamily, member 12), LAP TGF-beta-1 (Latency-associated peptide transforming growth factor beta 1), and OPG (Osteoprotegerin). In contrast IL-13 (Interleukin-13) a type 2 cytokine, was decreased in older patients. Interestingly, LIF/LIFR36,37,38, HGF39, CASP-840, and PD-L141 all have putative roles in neuroprotection or immunosuppression. Longitudinal sample collection is an important next step to differentiate the physiologic process of aging from potential biomarkers of milder disease.

We then compared immune analytes across sex and disease subtype, both of which are pertinent to clinical disease phenotype.

The relationship between MS phenotype and patient sex is well established, with significant evidence to show that males who develop MS frequently experience more rapid neurologic deterioration42. In our cohort, male MS patients had higher intrathecal FGF-19 (Fibroblast growth factor 19). In a multiple linear regression accounting for age, male sex was still an independent predictor of elevated FGF-19 (p < 0.001). An ROC analysis revealed a sensitivity and specificity of 100% at an FGF-19 NPX cutoff of > 4.691, with AUC of 1.0. Other members of the FGF family of proteins (namely, FGF-1, FGF-2, and FGF-9) have previously been shown to regulate myelination43. Indeed, FGF-9 has been shown to downregulate genes required for axonal transport and synaptic function by neurons in vitro and has thus been proposed as an intrinsic driver of neuronal dysfunction and degeneration independent of inflammation44. Unlike FGF-9, there has been little implication for FGF-19 in MS thus far, however Huang et al. have demonstrated that CSF FGF19 levels in MS patients rise after treatment with Fingolimod13.

On the other hand, females have overall higher susceptibility for developing MS and often exhibit more pronounced inflammatory phenotypes. In our cohort, amongst patients with RRMS, females exhibited greater IL-10RA and IL-20 (Fig. 3B). IL-20 is a part of the IL-10 family of cytokines and has previously been implicated as a pro-inflammatory cytokine in various autoimmune conditions, including psoriasis and rheumatoid arthritis. Dayton et al. suggest a neuroinflammatory and thus pathogenic role for IL-20 signaling in EAE45. IL-10 may exert both pro- and anti-inflammatory functions in MS46.

Finally, we compared analytes across patient disease subtype. Patients whose disease was characterized as Tumefactive expressed higher intrathecal CCL4, a potent chemoattractant, despite a low n of 2. In post-mortem studies of brain tissue, CCL4 is upregulated in Tumefactive demyelinating lesions compared to gliomas47, which often mimic Tumefactive lesions radiologically, and Estevao et al. have proposed that CCL4 increases blood brain barrier permeability and facilitates the transmigration of lymphocytes across endothelial cells48. Of note, one of the two patients with Tumefactive disease in our study passed away shortly after presentation due to fulminant progression. CCL4 may hold a role in fulminant expansion of demyelinating lesions, though a greater number of patients should be assessed in future studies.

Interestingly, we found no intrathecal inflammatory differences between patients with relapsing remitting and progressive disease. This may suggest shared elements of underlying immune pathophysiology of each subtype, despite differences in clinical presentation. It is also possible that analysis of intrathecal proteins is not sensitive enough to capture true immunologic differences between these populations, or that a larger cohort or stricter sample collection during only active relapses is needed to detect them.

Our study has several limitations, including a small sample size, possible selection bias from the lumbar puncture indication, and risk of type I statistical errors. With the exception of our categorical analyses via volcano plot, we did not make attempts to correct for multiple comparisons due to the exploratory nature of this project. Taken together, our data identify potential novel biomarkers that are associated with MS clinical phenotypes and MRI metrics of disease burden. Future studies are needed to validate these biomarkers and test whether they are potential treatment targets and if they have mechanistic involvement in immunopathogenesis.

Materials and methods

Patient selection and clinical samples

This was a retrospective, observational cohort study at a single academic medical center. Inclusion criteria included age 18–65, a diagnosis of MS by 2017 McDonald criteria, and DMT-naïve status at the time of CSF sampling. CSF samples were collected as part of routine clinical care in the UMass Chan Multiple Sclerosis Center. Patients provided written informed consent for storing a portion of CSF in the UMass biorepository. This study was reviewed and approved by the University of Massachusetts ethics board (IRB Protocol 14143). Data collection, storage, and access were in accordance with the Health Insurance Portability and Accountability Act. CSF samples were obtained between 5 years and 8 months prior to this study.

CSF analysis

CSF samples were centrifuged to remove cells and cellular debris and 10uL of the resulting supernatant was shipped in qPCR plates to Olink Proteomics (Cambridge, MA) for analysis using the Olink Target 96 Inflammation Panel (Olink, Boston, MA; Article #95,302) (Fig. 1A). A log-normalized protein expression (NPX or “Normalized Protein eXpression”) file was produced per Olink’s public protocol and utilized for data analysis49. Samples were run on two plates and normalized for comparison via bridging method. Data are available on the Gene Expression Omnibus (GEO) database under accession # GSE256426.

Clinical data

Clinical information for each patient was sourced from the electronic medical record by a medical student or resident, and clinical phenotypes verified by a board-certified neurologist. Patient demographics collected included age, patient-reported race and ethnicity, and sex. Clinical data were collected from the date of the patient’s lumbar puncture, or the nearest date before or after lumbar puncture if not available from the same date. These data included: diagnosis and MS subtype; Expanded Disability Status Scale, timed 25-foot walk, 9-hole peg test, symbol digit modalities test and date and frequency of MS relapses as documented by the patient’s neurologist. Objective metrics used in this study (EDSS, 25-foot walk, 9-hole peg test, SDMT) were obtained within 8 months of initial LP and were only included if they were sustained or worsening over a 6-month period. Patients were classified as either “active” or “inactive” based on evidence of any disease activity within the prior year by NEDA-3 criteria3.

MRI acquisition

All imaging was obtained on 3 T MRI platform (Signa Pioneer, General Electric Healthcare, Wisconsin, USA) using a 32-channel head and neck coil and a consistent acquisition protocol. These sequences consisted of a pre-gadolinium 3D sagittal T1-weighted Fast Gradient Echo (BRAVO) sequence as follows: TR/TE/flip angle = 7.5 ms/3.0 ms/8°, frequency/phase = 220/220, slice thickness = 1.2, FOV = 240 mm; sagittal 3D T2-FLAIR (fluid-attenuated inversion recovery): TR/TE/flip angle/echo train length = 5400/maximum ~ 133/90°/140, frequency/phase = 256/224, slice thickness = 0.8 at 50% resolution, FOV = 250 mm; if gadolinium was administered, an additional sagittal 3D T1-weighted fast spin echo (CUBE) scan was obtained 10 min after gadoterate meglumine infusion as follows: TR/TE/flip angle 435 ms/16.1 ms/90°, frequency/phase = 256/256, slice thickness = 0.6 mm, FOV = 250 mm.

MRI analysis

All T1-weighted scans were manually reviewed for image quality to assure no significant artifacts. Brain masks were obtained using the structural CAT12 pipeline, part of the Statistical Parametric Mapping (SPM, https://www.fil.ion.ucl.ac.uk/spm/) package in Matlab. These masks when then incorporated into the SIENAX tool, a function from the FMRIB software library (FSL, v5.0). SIENAX yields segmented white matter, gray matter, and whole brain tissue volumes, with a normalization multiplicative based on skull registration to a template. FSL FIRST was used to segment subcortical gray matter volumes. T2-hyperintense lesion number and summative volumes (T2LV) were segmented from T2-FLAIR images using the lesion prediction algorithm as implemented in the LST toolbox version 3.0.0 (www.statistical-modelling.de/lst) from SPM with manual adjustment as needed for gross errors in ITK-SNAP v3.8. Only MRIs collected within one year of sample acquisition by lumbar puncture were included in the final analysis.

Data analysis and statistics

Data were analyzed using Olink NPX software, GraphPad Prism software and Microsoft Excel. Bimodal tests were calculated in Olink NPX software for the following parameters: male versus female sex and MS subtypes. For categorical outcomes, statistical thresholds for significance were determined after correcting for multiple comparisons using the Benjamini-Hotchberg method. Clinical metrics (age, EDSS, timed 25-foot walk, 9-hole peg test, SDMT), patient age, and MRI brain tissue segmentation volumes (whole brain, cortical gray matter, whole gray matter, white matter, ventricular CSF as a metric of central brain atrophy, T2-hyperintense lesions, thalamus, and caudate) were analyzed as continuous variables in GraphPad Prism software using correlation analyses. Pearson correlation coefficients and p-values were calculated using Prism.