Intrathecal Inflammatory Profile and Gray Matter Damage Predict Progression Independent of Relapse Activity in Early Multiple Sclerosis.

Introduction

Multiple sclerosis (MS) is an autoimmune demyelinating disease of the CNS, affecting primarily young adults and leading to significant neurologic disability. 1-3

Initial observations from natural history large-cohort studies supported a two-staged disease model, in which most patients initially experience a relapsing-remitting course (RRMS) with prominent focal inflammatory activity, underlying acute neurologic relapses, subsequently followed by a progressive and irreversible disability accumulation (secondary progressive MS). 1-3 However, emerging evidence suggests a biological continuum in which mechanisms underlying relapses and progression coexist from early disease stages, 4-6 arguing against the two-stage dogmatic MS model. Even among young adults with RRMS and no ongoing inflammatory clinical attacks, a “smoldering” biological activity with steady and continuous progression of disability has been shown to be the major determinant of permanent confirmed disability accumulation over time. 2,7-10

Such progression independent of relapse activity (PIRA) represents an unmet therapeutic and clinical need because the underlying processes are only poorly depicted by those MRI measures already adopted in clinical practice and their relation to other surrogate markers is a matter of ongoing debate. 4 Many definitions of PIRA have been proposed. 10,11 In particular, the occurrence of subclinical, MRI-detectable, inflammatory activity has been incorporated in some refined definitions, considering its potential interference with clinical findings. Thus, terms such as “true PIRA,” “nonactive PIRA,” and “progression independent of relapse and MRI activity, PIRMA”) have been suggested, with the latter definition including absence of imaging signs of acute MRI activity close to the PIRA event date. 10

An increasing body of evidence supports that intrathecal meningeal and perivascular inflammation, associated with gray matter (GM) damage and ongoing tissue damage, 11-14 plays a key role in determining the gradual disability accumulation from the earliest stages of the disease. Indeed, previous studies suggested that brain and cortical atrophy are pathologic substrates of PIRA, further highlighting the key contribution of the GM damage to disease progression. 7,15,16 In a previous study, by combining molecular neuropathology of progressive MS cases at postmortem analysis and both imaging and CSF analysis in patients with MS at the time of diagnosis, we demonstrated a common intrathecal (meningeal and CSF) inflammatory profile linked to increased cortical pathology. 17 This profile includes molecules associated with sustained B-cell activity and lymphoid neogenesis and is predictive of subsequent disease activity and cortical atrophy accumulation in early RRMS. 17,18

In this study, we set out to determine a specific CSF profile and MRI measures of intrathecal compartmentalized inflammation that associate with and possibly predict PIRA and PIRMA in early RRMS.

Methods

Study Design

We analyzed clinical, CSF, and MRI data of 80 patients with RRMS, followed prospectively for 5 years from the time of diagnosis, at the MS Center of Verona, University Hospital, Verona, Italy. Inclusion criteria were as follows: (1) a diagnosis of RRMS according to McDonald revised 2017 diagnostic criteria 19 ; (2) the availability of CSF obtained at the time of diagnosis; (3) a 3T brain MRI acquired within 1 week from the lumbar puncture.

The lumbar puncture was performed at the time of diagnosis and, after the diagnosis, all patients were treated with dimethyl fumarate or teriflunomide after study inclusion.

Medical history review and laboratory analysis excluded concurrent infections and systemic disorders.

Clinical Data

Participants underwent a neurologic evaluation using the Expanded Disability Status Scale (EDSS) 20 performed by 2 certified raters (M.C., D.M.) at the time of the diagnosis and every 6 months for 5 years, with additional visits in case of relapses.

Any acute disease activity with particular regard of relapses 21 or new ore enlarging MRI lesions was recorded.

Patients were included in the PIRA group when the following criteria were fulfilled: (1) a baseline/reference score, applied to set a new reference score every time the EDSS score was higher than the previous measure and confirmed at the following visit; (2) a clinically significant increase in the EDSS score of 1.5 points or more from an EDSS score of 0, 1.0 point or more from an EDSS score of 1.0–5.0, or 0.5 points or more from an EDSS score of 5.5 or more; (3) an increase in the EDSS score not determined within 30 days before and 90 days after the onset of an investigator-reported relapse; (4) a confirmation EDSS score assigned at least 6 months after the initial disability increase. 10 The definition of PIRMA further required the exclusion of imaging signs of acute MRI activity (new or enlarging lesions or gadolinium-enhancing lesions) within 180 days before and after the PIRA event. 10,22

CSF Analysis

We collected CSF samples at the time of the diagnosis, at least 1 month after a preceding relapse, and within 1 week from the MRI, according to the Consensus Guidelines for CSF and Blood Biobanking. 23 After centrifugation, we stored the supernatant at −80°C. We analyzed the CSF levels of 68 inflammatory mediators using a combination of immune-assay multiplex techniques based on the Luminex technology (40-Plex and 37-Plex, Bio-Plex X200 System equipped with a magnetic workstation; BioRad, Hercules, CA), as previously described. 18 All samples were duplicated in the same experiment and in 2 consecutive experiments to verify the reproducibility and consistency of results. The CSF concentration of each protein detected during the analysis was normalized to the total protein concentration of each CSF sample (measured by the Bradford assay). 18,19 Finally, CSF/serum albumin ratio and presence of CSF-restricted oligoclonal bands were recorded.

MRI Analysis

MRI Acquisition Protocol

Participants underwent a 3T Philips Achieva MRI acquisition at the Radiology Unit of the University Hospital of Verona at the time of diagnosis and every year, with additional evaluations in case of relapses. The following image sets were acquired: 3D T1-weighted turbo field echo (TFE), with repetition time (TR) = 8.4 ms, echo time (TE) = 3.7 ms, voxel size = 1 × 1 × 1 mm 3 , field of view (FOV) = 240 × 240 × 180, number of excitations (NEX) = 1, and acquisition time = 5:05 minutes 3D fluid-attenuated inversion recovery (FLAIR) with TR/TE = 8,000/292 ms, inversion time (TI) = 2350 ms, voxel size = 1 × 1 × 1 mm 3 , FOV = 240 × 240 × 180, NEX = 1, and acquisition time = 4:48 minutes 3D double inversion recovery (DIR) with TR/TE = 5,500/292 ms, TI = 2,550 ms, delay 450 ms, voxel size = 1 × 1 × 1 mm 3 , FOV = 240 × 240 × 180, NEX = 3, and acquisition time = 10:49 minutes 3D T1-weighted TFE postcontrast sequence with the same parameters of the precontrast sequence. The gadolinium-based contrast agent adopted was gadobutrol, 0.1 mL/kg body weight (0.1 mmol/kg), with an average delay between injection and MRI acquisition of 10 minutes.

3D T1-weighted turbo field echo (TFE), with repetition time (TR) = 8.4 ms, echo time (TE) = 3.7 ms, voxel size = 1 × 1 × 1 mm 3 , field of view (FOV) = 240 × 240 × 180, number of excitations (NEX) = 1, and acquisition time = 5:05 minutes

3D fluid-attenuated inversion recovery (FLAIR) with TR/TE = 8,000/292 ms, inversion time (TI) = 2350 ms, voxel size = 1 × 1 × 1 mm 3 , FOV = 240 × 240 × 180, NEX = 1, and acquisition time = 4:48 minutes

3D double inversion recovery (DIR) with TR/TE = 5,500/292 ms, TI = 2,550 ms, delay 450 ms, voxel size = 1 × 1 × 1 mm 3 , FOV = 240 × 240 × 180, NEX = 3, and acquisition time = 10:49 minutes

3D T1-weighted TFE postcontrast sequence with the same parameters of the precontrast sequence. The gadolinium-based contrast agent adopted was gadobutrol, 0.1 mL/kg body weight (0.1 mmol/kg), with an average delay between injection and MRI acquisition of 10 minutes.

All images were quality-checked for scanner inhomogeneities and artifacts.

Image Processing

White matter (WM) lesions were automatically segmented on FLAIR images using the Lesion Segmentation Tool implemented in Statistical Parameter Mapping software on Matlab 2020b, and T2 hyperintense WM lesion masks, WM lesion volume (WMLv), and number (WMLn) were obtained. We used the T2 lesion masks to fill hypointensities on 3D T1-weighted images. New or enlarging T2 WM lesions were visually evaluated by a neuroradiologist with extensive experience in MS (FBP).

After lesion filling, we automatically segmented and parcellated the brain using multiatlas-based likelihood fusion (MALF) and FreeSurfer software to obtain mean global and regional cortical thickness (CTh) and GM volume. WM lesion filling was performed to reduce the risk of bias and improve the accuracy of CTh measure. 24

Two observers (D.M., M.C.) with experience in evaluation of cortical lesions (CLs) manually segmented cortical lesions (and created a CL mask) and independently evaluated CL number on DIR images, following the recommendations for CL scoring in patients with MS. 25 In case of different evaluations, a decision was taken by consensus of both operators. Owing to the suboptimal performance of the MRI in visualizing subpial lesions, this analysis could have taken into account mainly the intracortical and leukocortical lesions. We used FSL to estimate the CL volume (CLv) using the CL masks.

Statistical Analysis

R (version 4.1) and GraphPad Prism (version 9) were used for statistical analysis.

Clinical Characteristics and Demographics

Mann-Whitney, χ 2 , and Fisher exact tests were used when appropriate to assess differences between participants with and without PIRA. The analyses were repeated to assess differences between participants with and without PIRMA.

CSF and MRI Feature Selection and Prediction of PIRA

Random forest (RF) analysis was used to identify the baseline CSF molecules most associated with and best discriminating participants who experienced PIRA by the end of the follow-up.

RF analysis was also applied to baseline MRI measures, including GM regional thickness and volumes extracted from MALF, WMLn, WMLv, CLn, and CLv, and mean global CTh measures to identify the baseline MRI measures that best discriminate between participants with and without PIRA.

The cohorts were split into training (80%) and testing (20%) sets. The RF (1,000 trees, 8 variables tried at each split) was fitted on the training set, and the testing set was used to evaluate its performance. The abovementioned statistical analyses were repeated to discriminate participants with and without PIRMA.

Univariate estimates of the hazard ratios (HRs) of CSF cytokines and MRI markers identified with the RF analysis were calculated according to Cox regression analysis. Finally, variables significantly associated with the occurrence of PIRA ( p < 0.05) were tested in a multivariate Cox proportional regression model, also including demographic factors (age and sex).

Pathway Analysis

To better investigate the CSF molecular profile associated with PIRA, we used STRING 26 analysis to quantify module connectivity and enriched biological functions. Module connectivity was evaluated by the protein-protein interaction (PPI) enrichment p value. A low PPI p value indicated that the nodes (proteins) were not random and that the observed number of edges (the interaction between proteins) was significant. Strength and false discovery rate measures were used to evaluate how significant the enrichment effect was for each biological process detected by the pathway analysis.

Receiver Operating Characteristic Analysis

The ability of CSF cytokines and MRI parameters evaluated at the time of diagnosis to discriminate between patients with and without PIRA was tested with the area under the receiver operating characteristic curves (AUC-ROCs). The predictive model included the variables that showed the best association with PIRA in the RF model and age. The Youden method was applied to determine the optimal threshold, specificity, and sensitivity. The discriminating ability of a multiparametric model incorporating both CSF cytokines and MRI variables was also tested.

Standard Protocol Approvals, Registrations, and Patient Consents

The study was approved by the Ethics Committee of the University of Verona (MSBio Biological Bank, protocol number 66418), and informed consent was collected from all participants.

Data Availability

Anonymized data not published within this article will be made available by request from any qualified investigator.

Results

Study Cohort

The demographic and clinical characteristics of the 80 patients with RRMS included in the study are presented in Table 1 . During the follow-up, 49 patients experienced disease activity and 23 (28.8%) patients experienced a PIRA event (mean time to event 2.7 ± 1.1 years) with a mean EDSS increase of 1.5 ± 0.6. Of these, 8 developed new or enlarged, asymptomatic brain lesions (mean time to event 3.2 ± 1.5) and 5 experienced a relapse apart from the PIRA event (mean time to event 2.1 ± 2.0).

Table 1 Demographic, Clinical, and MRI Characteristics at Diagnosis in the Whole Study Cohort and According to Occurrence of PIRA RRMS PIRA No PIRA p Value N (%) 80 23 (29) 57 (71) Age (y; mean [SD]) 39.3 [12.3] 44.0 [10.7] 37.4 [12.4] 0.017 Sex (M/F) 16/64 5/18 11/46 0.768 Median EDSS score [25th–75th interquartile range] 2[1–3] 3[2–4] 1.5[1–2] <0.001 OCBs (positive/negative) 60/20 17/6 43/14 0.99 CSF/serum albumin ratio [SD] 5.02 [1.61] 5.02 [1.66] 5.02 [1.61] 0.99 Whole-brain WM volume (mm 3 ; mean [SD]) 696,300.02 [48,623.0] 675,945.35[56,075.5] 704,033.1 [43,574.6] 0.059 Whole-brain GM volume (mm 3 ; mean [SD]) 787,767 [65,277.39] 748,688.21 [60,539.5] 802,183.4 [61,299.6] <0.001 Mean cortical thickness (mm 3 ; mean [SD]) 2.42 [0.09] 2.37 [0.11] 2.43 [0.08] 0.008 WMLn (mean [SD]) 20.3 [15.8] 27.1 [13.1] 17.5 [16.2] <0.01 WMLv (mm 3 ; mean [SD]) 3.8 [5] 7.54 [5.77] 2.33 [3.79] <0.001 Spinal cord lesions (mean [SD]) 0.49 [0.95] 0.70 [1.18] 0.40 [0.84] 0.311 Gd+ lesions (mean [SD]) 0.16 [0.54] 0.08 [0.29] 0.19 [0.61] 0.751 CLn (mean [SD]) 4.72 [7.67] 6.13 [6.07] 2.91 [3.59] 0.032 CLv (mm 3 ; mean [SD]) 219.8 [399.73] 571.87 [619.36] 260.32 [340.10] 0.039 Abbreviations: CLn = cortical lesion number; CLv = cortical lesion volume; EDSS = Expanded Disability Status Scale; F = female; Gd+ lesions = gadolinium-enhancing lesions; GM = gray matter; M = male; OCBs = oligoclonal bands; PIRA = progression independent of relapse activity; RRMS = relapsing-remitting multiple sclerosis; WM = white matter; WMLn = white matter lesion number; WMLv = white matter lesion volume. Statistical comparisons of baseline characteristics of participants with or without PIRA are shown.

Demographic, Clinical, and MRI Characteristics at Diagnosis in the Whole Study Cohort and According to Occurrence of PIRA

Abbreviations: CLn = cortical lesion number; CLv = cortical lesion volume; EDSS = Expanded Disability Status Scale; F = female; Gd+ lesions = gadolinium-enhancing lesions; GM = gray matter; M = male; OCBs = oligoclonal bands; PIRA = progression independent of relapse activity; RRMS = relapsing-remitting multiple sclerosis; WM = white matter; WMLn = white matter lesion number; WMLv = white matter lesion volume.

Statistical comparisons of baseline characteristics of participants with or without PIRA are shown.

At diagnosis, in comparison with those without PIRA, participants with PIRA were older ( p = 0.017) and had higher EDSS score ( p < 0.001), higher WMLv ( p < 0.001), higher WMLn ( p < 0.01), higher CLn ( p = 0.032), and higher CLv ( p = 0.039). In addition, they were distinguished by baseline lower whole-brain GM volume ( p < 0.001) and lower mean CTh ( p = 0.008) when compared with the no PIRA group ( Table 1 ). Clinical, demographic, and radiologic characteristics of patients without radiologic activity with or without PIRMA at baseline are provided in eTable 1.

CSF Analysis

CSF Markers of the PIRA Group

Compared with those without PIRA, at the time of diagnosis, patients with PIRA had different CSF values of many proinflammatory cytokines/chemokines ( Table 2 , Figure 1 ). The RF analysis revealed the CSF markers that best distinguished patients experiencing PIRA events ( Figure 2A ). The model had an out of bag (OOB) on the training set of 32.8% and achieved an accuracy of 0.69 when applied to the testing set.

Table 2 Cytokines and Chemokines Values in the Whole Cohort and According to Occurrence or Not of PIRA Total MS (n = 80) No PIRA (n = 57) PIRA (n = 23) p Value CCL21 1,701.61 ± 1,074.09 1,746.92 ± 1,092.68 1,589.32 ± 1,041.70 0.428 CXCL13 13.90 ± 29.50 9.20 ± 15.01 25.30 ± 48.31 0.144 CXCL5 748.88 ± 610.68 778.57 ± 616.55 680.45 ± 604.85 0.445 CCL11 82.01 ± 85.53 83.01 ± 78.48 79.60 ± 102.57 0.536 CCL24 47.77 ± 44.37 43.69 ± 42.28 54.42 ± 49.35 0.186 CCL26 44.63 ± 54.73 41.15 ± 49.14 52.96 ± 66.74 0.425 CX3CL1 335.28 ± 238.89 320.78 ± 224.21 371.21 ± 273.99 0.750 GMCSF 97.69 ± 98.78 91.36 ± 82.67 112.55 ± 130.05 0.711 CXCL1 102.96 ± 97.52 95.84 ± 91.25 120.63 ± 111.79 0.398 CXCL2 62.35 ± 73.54 61.19 ± 77.55 65.60 ± 62.77 0.560 CCL1 49.83 ± 42.10 43.57 ± 36.38 63.29 ± 50.78 0.091 CXCL10 434.44 ± 539.45 419.97 ± 580.48 470.29 ± 430.54 0.253 CXCL11 78.59 ± 546.82 101.08 ± 647.61 22.84 ± 39.97 0.998 CCL2 499.84 ± 644.05 448.05 ± 653.11 628.19 ± 616.07 0.071 CCL8 222.36 ± 637.78 206.80 ± 685.10 260.95 ± 513.56 0.153 CCL7 53.86 ± 46.59 50.88 ± 42.77 61.53 ± 55.67 0.589 CCL13 31.64 ± 61.02 40.10 ± 68.80 9.49 ± 21.68 0.490 CCL22 44.88 ± 82.17 47.47 ± 93.36 38.95 ± 48.63 0.349 MIF 1,616.21 ± 1,648.13 1,441.24 ± 1,515.04 2,049.81 ± 1906.72 0.165 CXCL9 72.44 ± 69.32 76.22 ± 75.02 63.09 ± 52.96 0.704 CCL3 5.85 ± 5.03 5.14 ± 3.29 7.61 ± 7.68 0.420 CCL15 439.30 ± 397.49 402.37 ± 305.24 530.83 ± 563.85 0.642 CCL20 1.66 ± 1.76 1.38 ± 1.68 2.43 ± 1.81 0.052 CCL19 117.95 ± 103.67 118.08 ± 112.09 117.64 ± 82.33 0.471 CCL23 12.90 ± 17.61 12.83 ± 19.64 13.09 ± 11.29 0.506 CXCL16 1,799.83 ± 1,379.65 1,730.89 ± 1,282.93 1,970.66 ± 1,613.28 0.792 CXCL12 2,162.25 ± 1965.89 1,954.20 ± 1864.28 2,677.86 ± 2,154.28 0.102 CCL25 120.56 ± 74.16 119.43 ± 67.45 123.36 ± 90.31 0.874 TNF 36.02 ± 34.37 34.38 ± 35.53 40.10 ± 31.70 0.277 sTNFR1 4,362.94 ± 3,238.43 3,726.40 ± 2024.07 5,940.43 ± 4,855.88 0.010 sTNFR2 1,032.57 ± 1,089.65 877.62 ± 938.05 1,416.57 ± 1,344.31 0.087 TWEAK 2,134.63 ± 2,242.36 2,009.22 ± 2,133.18 2,445.41 ± 2,516.15 0.472 APRIL 49,874.46 ± 57,941.67 52,461.98 ± 63,825.21 43,461.91 ± 37,938.89 0.823 BAFF 10,596.84 ± 7,175.78 10,287.36 ± 6,315.24 11,363.79 ± 9,084.11 0.874 LIGHT 261.56 ± 428.75 178.97 ± 290.36 455.45 ± 613.10 0.013 sCD30 1,658.38 ± 1,482.91 1,663.60 ± 1,466.19 1,645.68 ± 1,556.26 0.868 IFNg 16.08 ± 21.42 13.63 ± 17.38 21.96 ± 28.52 0.092 IFNalfa2 17.13 ± 18.91 15.60 ± 16.65 20.63 ± 23.38 0.301 IFNbeta 34.44 ± 29.60 31.56 ± 26.42 41.65 ± 36.04 0.366 IL28a 188.38 ± 492.68 200.89 ± 554.11 159.19 ± 315.08 0.211 IL29 40.54 ± 39.65 39.81 ± 39.84 42.19 ± 40.16 0.580 sIL6R-beta 88,235.33 ± 70,647.29 85,402.97 ± 62,638.38 95,257.14 ± 88,672.21 0.866 IL1beta 2.74 ± 3.55 2.55 ± 3.08 3.24 ± 4.57 0.860 IL4 20.57 ± 17.25 17.98 ± 14.53 26.68 ± 21.58 0.148 IL6 25.97 ± 49.78 27.52 ± 55.08 22.12 ± 33.97 0.411 IL8 68.22 ± 83.10 63.17 ± 80.81 80.75 ± 89.12 0.519 IL10 21.41 ± 19.06 21.49 ± 18.53 21.22 ± 20.76 0.784 IL16 108.09 ± 126.80 95.46 ± 107.80 139.40 ± 163.45 0.295 sIL6Ra 3,926.30 ± 2,872.81 3,765.42 ± 2,648.99 4,325.01 ± 3,397.16 0.858 IL11 3.96 ± 13.23 5.06 ± 15.49 1.15 ± 1.28 0.920 IL12 (p40) 18.29 ± 19.87 15.88 ± 17.51 24.57 ± 24.47 0.153 IL12 (p70) 15.80 ± 25.94 21.17 ± 30.08 5.06 ± 7.05 0.007 IL19 69.63 ± 52.26 60.62 ± 41.26 89.79 ± 67.95 0.115 IL20 44.83 ± 68.18 40.57 ± 61.29 54.45 ± 82.39 0.132 IL22 34.68 ± 32.81 29.77 ± 26.32 46.42 ± 43.15 0.056 IL26 925.98 ± 949.61 1,027.41 ± 1,060.73 692.27 ± 576.95 0.351 IL27 192.21 ± 167.77 160.55 ± 145.70 276.61 ± 196.35 0.009 IL32 153.69 ± 179.89 159.61 ± 189.25 139.53 ± 158.30 0.579 IL34 561.51 ± 576.99 616.96 ± 650.27 440.53 ± 353.76 0.486 IL35 270.48 ± 194.98 263.95 ± 189.31 286.68 ± 211.92 0.658 MMP1 860.29 ± 1,029.41 767.02 ± 991.57 1,116.80 ± 1,112.79 0.091 MMP2 1,638.62 ± 4,814.30 2,074.25 ± 5,668.65 577.93 ± 425.77 0.808 Osteocalcin 654.25 ± 408.04 640.80 ± 418.28 687.58 ± 388.45 0.744 Osteopontin 73,914.62 ± 77,298.09 63,978.32 ± 71,800.86 98,539.37 ± 86,273.8425 0.049 Pentraxin 282.49 ± 332.48 295.87 ± 362.17 249.90 ± 249.92 0.783 TSLP 40.10 ± 49.51 39.60 ± 55.09 41.33 ± 32.79 0.114 sCD163 47,917.11 ± 33,673.06 47,618.24 ± 37,419.12 48,657.80 ± 22,510.35 0.408 Chitinase3like1 44,664.38 ± 35,907.8096 41,233.66 ± 33,506.79 53,166.59 ± 40,818.52 0.155 Abbreviations: APRIL = a proliferation-inducing ligand; BAFF = B-cell activating factor; C-C motif chemokine 22; CCL1 = chemokine (C-C motif) ligand 1; CCL13 = chemokine (C-C motif) ligand 13; CCL15 = chemokine (C-C motif) ligand 15; CCL19 = chemokine (C-C motif) ligand 19; CCL2 = chemokine (C-C motif) ligand 2; CCL20 = chemokine (C-C motif) ligand 20; CCL21 = chemokine (C-C motif) ligand 21; CCL22; CCL23 = chemokine (C-C motif) ligand 23; CCL24 = chemokine (C-C motif) ligand 24; CCL25 = chemokine (C-C motif) ligand 25; CCL26 = chemokine (C-C motif) ligand 26; CCL3 = chemokine (C-C motif) ligand 3; CCL7 = chemokine (C-C motif) ligand 7; CCL8 = chemokine (C-C motif) ligand 8; CX3CL1 = chemokine (C-X3-C motif) ligand 1; CXCL1 = chemokine (C-X-C motif) ligand 1; CXCL10 = chemokine (C-X-C motif) ligand 10; CXCL11 = chemokine (C-X-C motif) ligand 11; CXCL12 = chemokine (C-X-C motif) ligand 12; CXCL13 = chemokine (C-X-C motif) ligand 13; CXCL5 = chemokine (C-X-C motif) ligand 5; CCL11 = chemokine (C-C motif) ligand 11; CXCL16 = chemokine (C-X-C motif) ligand 16; CXCL2 = chemokine (C-X-C motif) ligand 2; CXCL9 = chemokine (C-X-C motif) ligand 9; GM-CSF = granulocyte-macrophage colony-stimulating factor; IFNalfa2 = interferon alpha-2; IFNbeta = interferon beta; IFNgamma = interferon gamma; IL10 = interleukin-10; IL11 = interleukin-11; IL12(p40) = interleukin-12 or cytotoxic lymphocyte maturation factor p40; IL12(p70) = interleukin-12 or cytotoxic lymphocyte maturation factor p70; IL16 = interleukin-16; IL19 = interleukin-19; IL1beta = interleukin-1 beta; IL20 = interleukin-20; IL22 = interleukin-22; IL26 = interleukin-26; IL27 = interleukin-27; IL28A = interleukin-28 A; IL29 = interleukin-29; IL32 = interleukin-32; IL34 = interleukin-34; IL35 = interleukin-35; IL4 = interleukin-4; IL6 = interleukin-6; IL8 = interleukin-8; LIGHT = tumor necrosis factor superfamily member 14; MIF = macrophage migration inhibitor factor; MMP1 = matrix metallopeptidase 1; MMP2 = matrix metallopeptidase 2; osteocalcin = bone gamma-carboxyglutamic acid-containing protein; PIRA = progression independent of relapse activity; sCD163 = soluble-CD163 (cluster of differentiation 163); sCD30 = soluble tumor necrosis factor receptor superfamily 8; sIL6Ra = soluble interleukin-6 receptor alpha; sIL6Rbeta = soluble interleukin-6 receptor beta; sTNFR1 = soluble-tumor necrosis factor-receptor 1; sTNFR2 = soluble-tumor necrosis factor-receptor 2; TNF = tumor necrosis factor; TSLP = thymic stromal lymphopoietin; TWEAK = TNF-related weak inducer of apoptosis. Values are expressed as pg/mL; means ± SD are reported. The Mann-Whitney U test was adopted to compare the mean values of CSF molecules between participants with and without PIRA. Significant comparisons ( p < 0.05) are reported in bold.

Cytokines and Chemokines Values in the Whole Cohort and According to Occurrence or Not of PIRA

Abbreviations: APRIL = a proliferation-inducing ligand; BAFF = B-cell activating factor; C-C motif chemokine 22; CCL1 = chemokine (C-C motif) ligand 1; CCL13 = chemokine (C-C motif) ligand 13; CCL15 = chemokine (C-C motif) ligand 15; CCL19 = chemokine (C-C motif) ligand 19; CCL2 = chemokine (C-C motif) ligand 2; CCL20 = chemokine (C-C motif) ligand 20; CCL21 = chemokine (C-C motif) ligand 21; CCL22; CCL23 = chemokine (C-C motif) ligand 23; CCL24 = chemokine (C-C motif) ligand 24; CCL25 = chemokine (C-C motif) ligand 25; CCL26 = chemokine (C-C motif) ligand 26; CCL3 = chemokine (C-C motif) ligand 3; CCL7 = chemokine (C-C motif) ligand 7; CCL8 = chemokine (C-C motif) ligand 8; CX3CL1 = chemokine (C-X3-C motif) ligand 1; CXCL1 = chemokine (C-X-C motif) ligand 1; CXCL10 = chemokine (C-X-C motif) ligand 10; CXCL11 = chemokine (C-X-C motif) ligand 11; CXCL12 = chemokine (C-X-C motif) ligand 12; CXCL13 = chemokine (C-X-C motif) ligand 13; CXCL5 = chemokine (C-X-C motif) ligand 5; CCL11 = chemokine (C-C motif) ligand 11; CXCL16 = chemokine (C-X-C motif) ligand 16; CXCL2 = chemokine (C-X-C motif) ligand 2; CXCL9 = chemokine (C-X-C motif) ligand 9; GM-CSF = granulocyte-macrophage colony-stimulating factor; IFNalfa2 = interferon alpha-2; IFNbeta = interferon beta; IFNgamma = interferon gamma; IL10 = interleukin-10; IL11 = interleukin-11; IL12(p40) = interleukin-12 or cytotoxic lymphocyte maturation factor p40; IL12(p70) = interleukin-12 or cytotoxic lymphocyte maturation factor p70; IL16 = interleukin-16; IL19 = interleukin-19; IL1beta = interleukin-1 beta; IL20 = interleukin-20; IL22 = interleukin-22; IL26 = interleukin-26; IL27 = interleukin-27; IL28A = interleukin-28 A; IL29 = interleukin-29; IL32 = interleukin-32; IL34 = interleukin-34; IL35 = interleukin-35; IL4 = interleukin-4; IL6 = interleukin-6; IL8 = interleukin-8; LIGHT = tumor necrosis factor superfamily member 14; MIF = macrophage migration inhibitor factor; MMP1 = matrix metallopeptidase 1; MMP2 = matrix metallopeptidase 2; osteocalcin = bone gamma-carboxyglutamic acid-containing protein; PIRA = progression independent of relapse activity; sCD163 = soluble-CD163 (cluster of differentiation 163); sCD30 = soluble tumor necrosis factor receptor superfamily 8; sIL6Ra = soluble interleukin-6 receptor alpha; sIL6Rbeta = soluble interleukin-6 receptor beta; sTNFR1 = soluble-tumor necrosis factor-receptor 1; sTNFR2 = soluble-tumor necrosis factor-receptor 2; TNF = tumor necrosis factor; TSLP = thymic stromal lymphopoietin; TWEAK = TNF-related weak inducer of apoptosis.

Values are expressed as pg/mL; means ± SD are reported.

The Mann-Whitney U test was adopted to compare the mean values of CSF molecules between participants with and without PIRA. Significant comparisons ( p < 0.05) are reported in bold.

CSF Markers and PIRA

(A) CSF markers increased in patients with PIRA events. (B) Markers differentially increased in the subgroup of patients with PIRMA. CCL20 = chemokine (C-C motif) ligand 20; IL22 = interleukin-22; IL27 = interleukin-27; IL35 = interleukin-35; LIGHT = tumor necrosis factor superfamily member 14; OPN = osteopontin; PIRA = progression independent of relapse activity; PIRMA = progression independent of relapse and MRI activity; sTNFR1 = soluble-tumor necrosis factor-receptor 1. * p < 0.05, ** p < 0.01

CSF Markers Associated With PIRA

(A) Multiway importance plot: most important variables associated with PIRA. Minimal depth and times a root measures are shown. Lower minimal depth values indicate higher predictive accuracy while higher times a root measure indicates a higher predictive power. (B) Multiway importance plot: most important variables associated with PIRMA. CCL2 = chemokine (C-C motif) ligand 2; CCL25 = chemokine (C-C motif) ligand 25; CCL7 = chemokine (C-C motif) ligand 7; CXCL11 = chemokine (C-X-C motif) ligand 11; CXCL13 = chemokine (C-X-C motif) ligand 13; IL35 = interleukin-35; INFbeta = interferon beta; LIGHT = tumor necrosis factor superfamily member 14; MIF = macrophage migration inhibitor factor; PIRA = progression independent of relapse activity; PIRMA = progression independent of relapse and MRI activity; sCD163 = soluble-CD163 (cluster of differentiation 163); sIL6Rbeta = soluble interleukin-6 receptor beta; sTNFR1 = soluble-tumor necrosis factor-receptor 1; sTNFR2 = soluble-tumor necrosis factor-receptor 2.

Univariate Cox regression showed a significant association between higher levels of sTNFR1 (HR 8.37 [1.86–37.55], p = 0.008), sTNFR2 (HR 3.24 [1.07–9.79], p = 0.040), and LIGHT (HR 1.59 [1.00–2.55], p = 0.044) and risk of experiencing PIRA. At the multivariate Cox regression analysis, after adjusting for age and sex, sTNFR1 (HR 10.11 [2.61–39.10], p = 0.001), sTNFR2 (HR 5.05 [1.63–15.64], p = 0.005), and LIGHT (HR 1.79 [1.11–2.88], p = 0.018) were confirmed as independent predictors of PIRA (eTable 2).

All the selected molecules associated with PIRA were confirmed to be involved in chronic inflammatory processes, particularly the recruitment of adaptive and innate immune cells through cytokine and chemokine activity by STRING pathway analysis. The cellular signaling in response to TNF and TNF superfamily molecules and the negative regulation of extracellular matrix protein secretion were highlighted (eFigure 1 and eTable 3).

CSF Markers of PIRMA

Among patients who experienced PIRMA ( marker values in eTable 4), the RF model highlighted those CSF molecules predictive of disability accumulation, which were not correlated with any clinical and radiologic focal inflammatory activity ( Figure 2B ). The model provided an OOB on the training set of 41.7% and achieved an accuracy of 0.56 on the testing set. Univariate Cox regression confirmed sTNFR1 (HR 6.27 [1.02–38.60], p = 0.048) and LIGHT (HR 1.81 [1.03–3.19], p = 0.040) as significantly increased among patients with PIRMA events. Multivariate analysis, adjusted for age and sex, confirmed sTNFR1 (HR 6.91 [1.28–37.27], p = 0.025) and LIGHT (HR 1.91 [1.09–3.34], p = 0.025) levels as independent intrathecal-based molecular predictors of PIRMA.

MRI Analysis

MRI Markers Associated With PIRA

Compared with those without PIRA, patients with PIRA events showed lower baseline GM thickness and volume in many brain areas ( Table 3 ). The RF model was applied to all variables and had an OOB on the training set of 28.3%, achieving a prediction accuracy of 0.75 on the testing set. The most important features in identifying patients with PIRA were the baseline CLn and volume/thickness of several cortical and deep GM regions ( Figure 3A ).

Table 3 MRI Values of Regional Thickness/Volume at Diagnosis in the Whole Study Cohort and According to the Occurrence of PIRA RRMS at diagnosis End of follow-up PIRA No PIRA p Value N (%) 80 23 (29) 57 (71) Cerebellum (mm 3 ; mean [SD]) 3.21 ± 0.48 2.97 ± 0.51 3.29 ± 0.45 0.014 Hippocampus (mm 3 ; mean [SD]) 2.76 ± 0.26 2.77 ± 0.29 2.76 ± 0.24 0.911 Anterior cingulate gyrus (mm 3 ; mean [SD]) 3.51 ± 0.35 3.34 ± 0.33 3.57 ± 0.54 0.019 Anterior insula (mm 3 ; mean [SD]) 3.82 ± 0.37 3.69 ± 0.42 3.88 ± 0.34 0.052 Frontal pole (mm 3 ; mean [SD]) 2.62 ± 0.24 2.56 ± 0.25 2.65 ± 0.24 0.177 Inferior temporal gyrus (mm 3 ; mean [SD]) 3.67 ± 0.33 3.51 ± 0.25 3.73 ± 0.24 0.006 Middle cingulate gyrus (mm 3 ; mean [SD]) 2.56 ± 0.29 2.42 ± 0.29 2.60 ± 0.28 0.018 Medial frontal cortex (mm 3 ; mean [SD]) 3.15 ± 0.38 3.03 ± 0.40 3.19 ± 0.36 0.124 Middle frontal gyrus (mm 3 ; mean [SD]) 2.34 ± 0.21 2.24 ± 0.24 2.38 ± 0.19 0.001 Medial postcentral gyrus (mm 3 ; mean [SD]) 1.32 ± 0.29 1.21 ± 0.33 1.35 ± 0.27 0.037 Medial precentral gyrus (mm 3 ; mean [SD]) 1.80 ± 0.35 1.67 ± 0.39 1.87 ± 0.31 0.053 Medial superior frontal gyrus (mm 3 ; mean [SD]) 2.80 ± 0.34 2.68 ± 0.35 2.84 ± 0.33 0.065 Middle temporal gyrus (mm 3 ; mean [SD]) 3.04 ± 0.29 2.87 ± 0.29 3.10 ± 0.27 0.003 Posterior cingulate gyrus (mm 3 ; mean [SD]) 2.86 ± 0.32 2.70 ± 0.37 2.91 ± 0.29 0.035 Precuneus (mm 3 ; mean [SD]) 2.28 ± 0.33 2.28 ± 0.33 2.47 ± 0.31 0.022 Parahippocampal gyrus (mm 3 ; mean [SD]) 3.00 ± 0.35 2.88 ± 0.36 3.04 ± 0.33 0.139 Posterior insula (mm 3 ; mean [SD]) 3.27 ± 0.39 3.12 ± 0.39 3.33 ± 0.38 0.037 Postcentral gyrus (mm 3 ; mean [SD]) 1.51 ± 0.22 1.43 ± 0.22 1.54 ± 0.21 0.062 Planum polare (mm 3 ; mean [SD]) 1.76 ± 0.29 1.64 ± 0.25 1.80 ± 0.29 0.013 Precentral gyrus (mm 3 ; mean [SD]) 2.04 ± 0.28 1.94 ± 0.31 2.07 ± 0.27 0.041 Planum temporale (mm 3 ; mean [SD]) 1.86 ± 0.29 1.73 ± 0.32 1.90 ± 0.36 0.030 Superior frontal gyrus (mm 3 ; mean [SD]) 2.29 ± 0.20 2.20 ± 0.19 2.32 ± 0.20 0.011 Supplementary motor cortex (mm 3 ; mean [SD]) 2.46 ± 0.34 2.28 ± 0.19 2.52 ± 0.20 0.008 Supramarginal gyrus (mm 3 ; mean [SD]) 2.23 ± 0.21 2.15 ± 0.19 2.26 ± 0.20 0.013 Superior parietal lobule (mm 3 ; mean [SD]) 1.72 ± 0.19 1.67 ± 0.18 1.73 ± 0.19 0.205 Superior temporal gyrus (mm 3 ; mean [SD]) 2.38 ± 0.26 2.26 ± 0.18 2.42 ± 0.19 0.028 Temporal pole (mm 3 ; mean [SD]) 3.75 ± 0.44 3.51 ± 0.44 3.84 ± 0.41 0.006 Inferior frontal gyrus (mm 3 ; mean [SD]) 2.17 ± 0.23 2.05 ± 0.24 2.21 ± 0.21 0.005 Caudate (mm 3 ; mean [SD]) 3,099.58 ± 480.44 2,897.98 ± 618.27 3,171.58 ± 403.29 0.010 Pallidum (mm 3 ; mean [SD]) 1,425.96 ± 158.21 1,425.96 ± 158.21 1,348.90 ± 239.11 0.113 Putamen (mm 3 ; mean [SD]) 4,239.86 ± 547.74 4,010 ± 534.47 4,321 ± 533.38 0.016 Thalamus (mm 3 ; mean [SD]) 7,111 ± 547.74 6,503 ± 1,210.97 7,329.24 ± 855.95 0.002 Abbreviations: PIRA = progression independent of relapse activity; RRMS = relapsing-remitting multiple sclerosis.

MRI Values of Regional Thickness/Volume at Diagnosis in the Whole Study Cohort and According to the Occurrence of PIRA

Abbreviations: PIRA = progression independent of relapse activity; RRMS = relapsing-remitting multiple sclerosis.

MRI Markers Associated With PIRA

Multiway importance plot with the most important MRI variables associated with PIRA (A) and in a subgroup of patients with PIRMA (B). Minimal depth and times a root measures are shown. Lower minimal depth values indicate higher predictive accuracy while higher times a root measure indicates a higher predictive power. PIRA = progression independent of relapse activity; PIRMA = progression independent of relapse and MRI activity.

Univariate Cox regression analysis of these MRI parameters showed that caudate (HR 0.99 [0.99–1], p = 0.025), putamen (HR 0.99 [0.99–1], p = 0.041), and thalamus (HR 0.99 [0.98–0.99], p = 0.001) volumes, alongside CTh of the middle frontal gyrus (HR 0.03 [0.03–0.38], p = 0.006) and CLn (HR 1.13 [1.03–1.23], p = 0.013), were associated with higher risk of experiencing PIRA. In the Cox multivariate analysis, after adjusting for age and sex, thalamic volume (HR 0.98 [0.97–0.99], p = 0.005), CTh of the middle frontal gyrus (HR 0.05 [0.01–0.72], p = 0.028), and CLn (HR 1.15 [1.05–1.25], p = 0.003) were confirmed as independent predictors of PIRA (eTable 5).

MRI Markers Associated With PIRMA

The abovementioned analysis of the MRI measures was repeated in the PIRMA subgroup (eTable 6). The RF provided an OOB on the training set of 28.1% and achieved a prediction accuracy of 0.5 on the testing set. The model showed that the most important baseline features in identifying participants with PIRMA events were the volume of thalamus and pallidum; the CTh of supramarginal gyrus, planum polare, planum temporale, and posterior cingulate gyrus; and CLn ( Figure 3B ). Univariate Cox regression analysis on the statistically significant features identified by the RF showed that CLn (HR 1.17 [1.06–1.29], p = 0.003) significantly increased in patients with PIRMA. Multivariate analysis, after adjusting for age and sex, confirmed this result (HR 1.17 [1.05–1.30], p = 0.004).

ROC Analysis

The results of the ROC analysis for each CSF cytokine selected by the RF method are displayed in eFigure 2 and detailed in eTable 7. Similarly, eFigure 3 and eTable 8 present the results for MRI parameters. The combined model incorporating CSF markers provided an AUC of 0.82 [0.70–0.94] while the AUC for combined MRI parameters was 0.80 [0.68–0.92]. The model incorporating both CSF markers and MRI parameters achieved an AUC of 0.87 [0.78–0.96] ( Figure 4 ).

ROC Analysis

ROC analysis curves showing combined models with CSF and MRI markers. AUC = area under the curve; CI = confidence interval; CM = combined model; FBR = false-positive rate; ROC = receiver operating characteristic; TPR = true-positive rate.

Discussion

The identification of candidate markers of early PIRA, as well as defining its pathologic determinants, 4 represents an unmet need, hindering the implementation of a personalized therapeutic approach starting at the time of diagnosis. 8

In this study, we compared the demographic, clinical, CSF, and MRI profile at diagnosis between patients with RRMS who experienced PIRA (or PIRMA) in the 5 years after the diagnosis and those who remained free of disability progression. Our analysis revealed that the risk of experiencing disability progression was associated with the altered expression of molecules related to the TNF superfamily and with specific MRI features including cortical lesion number, CTh of the frontal cortex, and volume of deep gray matter structures such as thalamus.

To identify whether molecular markers present at diagnosis can predict later development of PIRA, we assayed a range of inflammatory mediators that are potentially associated with MS pathobiology and progression. The analysis of the diagnostic CSF suggested that mechanisms underlying perivascular and meningeal inflammation are implicated in the development of early PIRA.

In particular, a specific CSF profile that included molecules of the TNF superfamily was found to be the best surrogate marker of PIRA, with LIGHT (also called TNFSF14) and sTNFR1 emerging as the most important predictors of occurrence of disability accumulation not related to acute attacks. LIGHT is a protein that coexists in both a membrane-bound and a soluble form. Although its functional role in CNS inflammatory response is only partially understood, the membrane-bound form may exert proinflammatory activity, including enhanced survival of CD4 + memory T cells 27 and the costimulation of T cells, 28 while, after shedding, it decreases stimulatory signals and acts as a T-cell inhibitor by stabilizing the cis HVEM:BTLA complex. 29 Accordingly, a LIGHT gene variation was shown to increase the MS risk in a genome-wide association study 30 with increased serum levels among patients with disease activity and decreased levels among those treated with natalizumab. 31 Soluble LIGHT, through the interaction with the lymphotoxin-beta receptor (LTbetaR), seems to exert a proinflammatory role in an experimental hepatitis model 32 and has also been reported to contribute to the maintenance of lymphoid microenvironment including B-cell follicles in secondary lymphoid organs. 33 The results are in line with our previous studies, 17,18 where CSF LIGHT was found to be strongly related to the severity of cortical thinning over time. 18 Furthermore, molecules related to lymphoid recruitment, such as CXCL13, which we similarly found associated with cortical atrophy accumulation, were confirmed as candidate biomarkers of PIRA, further supporting the role of intrathecal inflammation, possibly related to the presence of meningeal tertiary lymphoid-like structures, as one of the principal determinants of disability progression that is manifest in the first years of the disease.

The STRING pathway analysis confirmed an inflammatory profile characterized by high and dysregulated levels of the TNF superfamily. Data on the role of TNF in the pathogenesis of MS and associated neuronal loss are convincing and consistent. 34 Patients with progressive MS are characterized by higher TNF levels that correlate with the degree of disability progression. 35 Accordingly, increased TNF detected in meninges and CSF highly correlated with GM demyelination at both the time of diagnosis and death, 18 with unbalanced levels of their receptor soluble molecules (sTNFR1 and sTNFR2), a finding also confirmed in patients at the time of diagnosis. 36 Gene expression profiling of both cortical lesions and normal-appearing GM confirmed that increased meningeal inflammation is associated with a shift in the balance of TNF signaling away from TNFR1/TNFR2, the first preferentially expressed on neurons and oligodendrocytes and the latter more expressed by astrocytes and microglia. 37,38 Finally, the persistent meningeal production in the animal model of TNF and IFNγ led to both subpial demyelination and neuronal loss, corroborating all these proteomic, transcriptomic, and pathologic evidence. 39

In contrast with the increasing amount of evidence on the role of a dysfunctional TNF pathway in mediating MS progression, so far, therapies targeting TNF failed in halting disease activity. One possible explanation is the nonselectivity of the treatment, with absence of stimulation of TNFR2 that potentially leads to immune regulation and tissue protection. Efforts are then needed to develop new therapeutical anti-TNF strategies that could target the detrimental aspects of TNF pathway and be beneficial since early disease stages.

From the imaging point of view, the baseline number and volume of CLs were higher in those patients who developed PIRA events. This is not unexpected because several previous studies suggested a strong association between CLs and disease progression. 16 Moreover, in a 30-year longitudinal study, the number of CLs was the best marker of EDSS accumulation. 40

Similarly, baseline deep GM atrophy emerged as a significant predictive factor in participants with PIRA. Basal ganglia atrophy and particularly thalamic atrophy have been repeatedly suggested as a key finding in MS pathology, associated with cognition, fatigue, and disability progression. 41-44

Such cortical and deep GM involvement could represent a surrogate marker of chronic compartmentalized inflammation and of the widespread MS neurodegeneration processes. Certainly, compartmentalized CNS inflammation, linked to the persistence of adaptive immune infiltrates in niches such as perivascular and meningeal spaces, in combination with glial cell activation and the presence of chronic expanding lesions, represents a key driver of disability accrual. 14,17 Particularly, subpial cortical demyelination, which represents the largest and most frequently observed lesion type of MS, is associated with B-cell and T-cell infiltrates in the adjacent meningeal sulci, 12 a key hallmark of compartmentalized inflammation. Gradients of tissue damage in the subpial tissue, a site adjacent to the CSF-filled space, are also seen at other structures such as the cerebellum and deep GM, including the thalamus. 45,46 These findings are in line with the hypothesis that inflammatory factors diffuse into the underlying tissues, either by influencing the activation status of CNS parenchymal cells to mediate tissue injury or by directly promoting a cytotoxic effect in the adjacent GM. 47 To evaluate the possible role of subclinical MRI activity, we performed an additional analysis on a group of patients with PIRMA. Notably, results from the CSF and MRI marker analysis of participants with PIRA were confirmed in patients with PIRMA, suggesting a common physiopathologic trait that underlies the risk of attaining PIRA regardless of acute inflammatory activity that is discernible on MRI. 48

The work has a number of limitations. Beyond the early accumulation of focal and diffuse GM demyelination, 11,13 many other pathologic processes that have not been evaluated in our study may concur with PIRA, including the presence of demyelinating activity in chronic lesions, 48,49 spinal cord lesions, and atrophy 48 and possible altered remyelination mechanisms, all leading to premature and irreversible neuroaxonal damage. However, despite this, we still noticed a significant impact of our measured markers on PIRA.

The relatively low number of patients with PIRA and the absence of a validation cohort may limit the generalizability of our conclusions. Notably, we noticed a high proportion of patients with PIRA, 50 particularly if considering an early MS cohort. This finding could be explained by the relatively high age at which patients underwent the diagnostic lumbar puncture. Furthermore, the selection of a cohort of patients treated with 2 not high-efficacy treatments, potentially led to a homogeneous sample of participants but with a relatively higher risk of experiencing PIRA. 51

Finally, a composite measure that includes upper limb function, walking speed, and cognitive testing would provide additional value to the clinical definition of PIRA. 10

With the abovementioned limitations, we can suggest that (1) a specific intrathecal inflammatory and MRI profile present at diagnosis characterizes the subsequent risk of early PIRA and (2) anti-inflammatory therapies capable of modulating intrathecal inflammation may be effective in preventing PIRA events and should be subjected to further careful investigations.