Introduction
Historically, the appearance of new white matter (WM) lesions has been the primary metric for evaluating the efficacy of disease-modifying therapies (DMTs) in multiple sclerosis (MS). Highly effective DMTs markedly suppress clinical relapses and the formation of new WM lesions. 1 However, DMTs have limited impact on preventing the neurodegenerative component of MS—reflected by ongoing brain and spinal cord volume loss and the progressive accumulation of irreversible disability. 2,3 Notably, brain atrophy often progresses in the absence of new WM lesions, underscoring the unmet need for DMTs that target the underlying mechanisms of neurodegeneration. 2
WM lesions in MS are pathologically and radiologically heterogeneous. 4,5 Recent focus has shifted toward a subset of lesions characterized by chronic inflammatory activity. These chronic active lesions (CALs) exhibit persistent axonal damage and are believed to drive CNS tissue loss and long-term disability. 6,7 On MRI, CALs exhibit distinct features that may serve as in vivo markers of compartmentalized inflammation and ongoing demyelination. One proposed radiologic surrogate of CALs is their gradual expansion over time, observed on longitudinal imaging—a process for which they are referred to as slowly enlarging lesions (SELs). 8 Another MRI feature associated with CALs is the presence of a paramagnetic rim, visible at the lesion edge using susceptibility-based imaging. 9 These paramagnetic rim lesions (PRLs), such as SELs, are associated with chronic inflammation, progressive neurodegeneration, and disability accumulation. 10 Although both imaging features represent potential radiologic markers of CALs, PRLs and SELs are regarded as 2 distinct radiologic phenomena with only partial overlap. 11
PRLs were first identified in pathologic studies of patients with progressive MS (PwPMS) and initially thought to be associated with later stages of the disease. 9 However, recent evidence shows that the rim in PRLs forms early during lesion development and does not seem in already existing lesions. 12 Because new lesion formation is most common in the early stages of MS, several studies found comparable PRLs prevalence across both relapsing and progressive disease phenotypes. 13 More recent studies suggest a higher prevalence of PRLs in younger individuals and/or earlier disease stages (i.e., radiologically and clinical isolated syndromes, pediatric onset MS). 14,15
On cross-sectional imaging, PRLs show variable rim conspicuity, ranging from well-defined to faint or incomplete peripheral rims, allowing classification as Definite, Probable, or Possible PRLs according to modified consensus criteria reflecting interpretative certainty. 16 PRLs are increasingly explored as potential therapeutic targets for compartmentalized inflammation, wherein attenuation or disappearance of the rim has been proposed as a marker of inflammatory resolution. 17 However, reduced rim conspicuity may also reflect the natural aging of PRLs because studies have shown that rims tend to fade over time. 18 Accordingly, older PRLs with faint rims—less likely to be classified as definite—may represent more structurally damaged lesions.
In this study, we use qMRI—including Neurite Orientation Dispersions Density Imaging (NODDI), Diffusion Tensor imaging (DTI), quantitative susceptibility mapping (QSM), and T1-weighted magnetization-prepared rapid acquisition gradient echo (MPRAGE) signal intensity—to investigate the structural correlates of heterogeneous rim appearances in PRLs among PwPMS, with a specific focus on comparing PRLs classified as Definite or Probable PRLs. Because rim susceptibility tends to decline as lesions age, we hypothesize that, in the context of known disease progression, the likelihood of classifying a PRL as Definite rather than Probable decreases with both disease duration and lesion age.
Recent late-stage setbacks and regulatory uncertainty in progressive MS therapeutics highlight the need for biomarkers that are biologically grounded, longitudinally interpretable, and suitable as trial end points. By linking PRLs' rim conspicuity and classification to quantitative MRI (qMRI) tissue measures and lesion age in progressive MS, this study aims to clarify the biological significance of PRLs' fading or disappearance in MS phenotypes characterized by ongoing clinical progression, thereby informing therapeutic targeting strategies.
Methods
Participant Selection
Participants with MS were retrospectively selected from prospective observational cohort studies conducted at the University of California San Francisco. For each participant, 1 MRI scan acquired from February 2016 to the end of 2024 was quantitatively analyzed.
Inclusion criteria for participants include:
Diagnosed with either primary or secondary progressive (SP) MS at the time of MRI.
Brain MRI, including conventional imaging as well as 3D-EPI and high angular resolution diffusion imaging (HARDI).
Exclusion criterion includes:
Motion affecting MRI scans, particularly the 3D-EPI and/or HARDI images.
A final cohort of 106 PwPMS was selected ( Figure 1 ).
MRI Acquisition and Protocol
All participants were scanned on the same Siemens 3T Skyra scanner. The scan protocol included standard sagittal brain 3D T2-fluid-attenuated inversion recovery (FLAIR) (voxel size 1 × 1 × 1 mm 3 , matrix 256 × 240, 176 slices per slab, TR/TE/TI = 5,000/388/1800 ms, parallel imaging acceleration factor [iPAT] = 3), 3D MPRAGE (voxel size 1 × 1 × 1 mm 3 , matrix 256 × 240, 176 slices per slab, TR/TE/TI = 2,300/2.98/900 ms, flip angle = 9°, iPAT = 2), axial HARDI 2-shell acquisition protocol (simultaneous multislice acceleration factor of 2, anterior to posterior phase encoding direction, 30 directions at b = 700 seconds/mm 2 , and 64 directions at b = 2,000 seconds/mm 2 with TR/TE = 4,300/96 ms, along with 10 b = 0 scans, 1 of which was acquired using a reversed phase-encoding direction; 2.2 mm 3 isotropic voxels; 66 axial slices; field of view = 220 × 220 mm), and sagittal segmented 3D echoplanar imaging (3D-EPI) with 0.65 mm 3 cubic voxel and the following main acquisition parameters: TR/TE = 68/37 ms, EPI factor = 15, anterior to posterior phase encoding, and flip angle = 10°.
Imaging Analysis
Preprocessing of 3D-EPI and HARDI
For all participants, images were inspected and quality ensured, particularly regarding 3D-EPI and HARDI. For 3D-EPI, all images underwent standardized preprocessing. Phase and magnitude images were reoriented to standard space and cropped to remove the neck region using FSL's “fslreorient2std” and “robustfov.” The same preprocessing steps were applied to the T1-weighted and T2-FLAIR images. Brain masks were generated from the magnitude images using FSL's Brain Extraction Tool. For HARDI, susceptibility-induced distortions were estimated from image pairs acquired with reversed phase-encoding directions (resulting in distortions with opposite directions), analogous to Andersson et al., 19 using the susceptibility-induced distortion correction tool (“TOPUP”) tool of the FMRIB Software Library (FSL). 20 In addition, eddy current–induced distortions and head motion were estimated and corrected using “eddy” from the same toolbox; b-vectors were accordingly rotated to account for the corrections.
Phase Image Processing
The phase images were processed in 3 steps. Phase values were first converted from raw units to radians and unwrapped using FSL's “fslmaths” and “prelude.” In the second step, the background field contribution of the unwrapped phase image was removed through the VSHARP function from the susceptibility tensor imaging (STI) suite. In the final step, the phase contrast image was inverted using FSL's “fslmaths.”
QSM Processing
The QSM was processed from phase, magnitude, and brain mask images using iLSQR QSM inversion from the STI suite. 21 After image coregistration, QSM images were cropped to match the dimensions of the 3D-EPI space images using FSL's “fslroi,” and the spatial headers were aligned using FSL's “fslcpgeom” and “fslswapdim.”
NODDI Processing
NODDI fitting was performed with the NODDI MATLAB Toolbox (version1.0.1; MathWorks) and DTI using Diffusion Imaging in Python, version 0.13.18. The Anisotropic Power Map derived from the high angular resolution diffusion imaging data was then registered to the T1-w MPRAGE images using the Advanced Normalization Tools multidimensional registration tool. 22 The same transformation used to register the Anisotropic Power Map was then applied to each of the NODDI and DTI maps to register the images into the same space. Visual quality assessment of the registration results was performed.
Image Coregistration
T1-weighted and T2-FLAIR images were coregistered to the 3D-EPI space using the APrL Pipeline. 23 This pipeline also outputs a FAST tissue-segmented image aligned to the 3D-EPI space, which was subsequently used to derive a CSF mask for lesion analysis.
PRLs Identification and Categorization
A neuroradiologist with 12 years of experience in imaging of demyelinating diseases (S.S.) examined the processed phase images with a contrast window fixed across all participants and set to maximize the visualization of the paramagnetic rims, identifying 190 WM lesions with susceptibility signal loss within the lesion boundary. The acute nature of these lesions was excluded by either reviewing postcontrast T1 images or, when absent, verifying the presence of the same lesion in a scan acquired at least 1 year before the MRI used for this study ( Figure 2 , diagram of lesion exclusions). Of 190 WM lesions with peripheral susceptibility, 6 were excluded for the presence of enhancement or because they had newly appeared from the prior scan. Eight additional lesions were excluded because of small size (<30 mm 3 ) to avoid imprecision due to partial volume effects. Of the 176 remaining lesions, 114 were classified as Definite PRLs, with a clear prominent rim. 16,24 Sixty-two of 176 lesions showed a less conspicuous paramagnetic rim, often thinner and slightly more discontinuous, resulting in classifications as either Possible or Probable PRL. Lesion terminology here follows the confidence-based classification framework proposed by Hemond et al. in which “Definite” PRLs denote lesions with a prominent, fully visible paramagnetic rim, whereas “Probable” PRLs denote lesions with a less conspicuous but still reliable paramagnetic rim sufficient for independent identification. Regardless of subclassification, all retained lesions independently met international consensus criteria for PRL identification. 24 To retain only Probable PRLs with high diagnostic confidence and exclude lower-confidence lesions with greater diagnostic uncertainty (i.e., Possible PRLs), 3 additional reads of the 62 lesions with less conspicuous paramagnetic rims were performed by 3 readers (S.S., E.L., and O.M.), each blinded to the other readers' assessments, clinical data, and qMRI metrics. Phase images were reviewed using the same fixed contrast window applied during the initial screening. A lesion was classified as a Probable PRL only if the 2 most senior readers (S.S. and E.L.) concurred, achieved either by unanimous 3-reader agreement, or by structured consensus discussion in cases where initial independent agreement was reached by 2 of the 3 readers. Lesions failing to meet this threshold were classified as Possible PRLs and excluded. Twenty-one of the 62 faint-rim lesions did not meet this criterion and were therefore excluded. In total, 155 PRLs were retained for this study: 114 Definite and 41 Probable PRLs ( Figures 2 and 3 ).
Clinical Assessment of PRLs Age
The earliest available clinical scans for all 155 PRLs were reviewed by a neuroradiologist (S.S.). For each lesion, a date of first visualization was established, and the interval to the MRI used in this analysis was computed to quantify differences in lesion age distributions between Definite and Probable PRLs.
Lesion Regions of Interest Subsegmentation
All PRLs were semiautomatically segmented on FLAIR images using Jim software. Lesions were indexed and labeled using FSL's “cluster.” The lesion regions of interest (ROIs) were converted to NIFTI format, and the masks processed to extract 3 lesion subcomponents. For each PRL, 3 types of lesion masks were obtained.
1. “Whole-lesion” mask, as semiautomatically segmented on JIM.
2. Lesion “Core” mask, obtained by eroding 4 voxels from the whole-lesion mask.
3. Lesion “Rim” mask, obtained by subtracting the lesion core mask from the whole-lesion mask (eFigure 1).
Data Extraction
Median values of NODDI and DTI metrics were extracted within each PRL's whole-lesion and core masks. NODDI and DTI metrics were intentionally not extracted from the rim mask because the rim region—defined as the 4-voxel shell between the whole-lesion boundary and the eroded core—is inherently thin and susceptible to partial volume effects from adjacent normal-appearing WM, which would compromise the reliability of diffusion-based quantitative estimates in that compartment. QSM map values were extracted within the rim mask using the 75th percentile value rather than the median, to minimize contamination from the diamagnetic signal of adjacent normal-appearing WM. Rim QSM values were used solely to quantitatively confirm the paramagnetic susceptibility signal visually appreciated on phase images during PRL classification and were not used as an input to the classification process nor as an outcome measure of tissue disruption. The median QSM values were extracted from QSM map even within core and whole-lesion ROIs. The core mask represents the central, demyelinated portion of each lesion, and group comparisons in this ROI specifically capture tissue damage within the demyelinated central compartment. The whole-lesion ROI encompasses the full T2-FLAIR hyperintense lesion volume (rim + core) and therefore reflects a composite signal from both compartments.
Data Collection and Statistical Analysis
The study team collected clinical and demographic characteristics of participants. Differences in demographics, PRL distribution, disease severity, and treatment use were compared using Wilcoxon rank-sum or χ² tests, as appropriate. Differences in qMRI metrics between Definite and Probable PRLs were assessed with the Mann-Whitney U test. Lesion age was defined as the interval, in years, between the first clinical MRI on which a lesion was detected and the scan used for quantitative analysis. For lesions first detected after a prior scan on which they were absent, this interval was treated as an observed lesion age estimate. For lesions already visible at the first available scan, the recorded interval represents a lower bound on true lesion age because the actual lesion onset predates the earliest available imaging time point. Although this situation is often described clinically as unknown prior onset, in the survival object used here, the outcome was lesion age at the study scan; therefore, lesions present at the first available scan were coded as right-censored observations because true lesion age exceeds the recorded interval. Fifteen lesions with an observed age interval of zero were excluded from Accelerated Failure Time (AFT) models because log (0) is undefined under a log-normal distribution; these lesions were retained in a complementary Kaplan-Meier analysis with log-rank test. Lesion age was modelled using a log-normal AFT model fitted with R's survreg () function, selected based on the approximate normality of log-transformed lesion ages in both groups. Three prespecified AFT models were fitted among the 140 lesions with a nonzero observed age interval: an unadjusted model, a multivariable model adjusted for sex, disease duration, and disease course, and a cluster-robust multivariable model in which within-participant correlation was addressed through sandwich standard errors. Model output is expressed as time ratios (Probable/Definite PRL age) with 95% CIs. The cluster-robust specification was confirmed with a Cox proportional hazards model incorporating a gamma frailty term for participant. A one-lesion-per-participant sensitivity analysis was performed among the 51 participants contributing at least 1 nonzero interval PRL to the AFT models, randomly selecting 1 lesion per participant, to confirm that findings were not driven by participants contributing multiple lesions. A further sensitivity analysis was restricted to the 44 noncensored lesions with a nonzero observed age interval (7 Probable, 37 Definite), for which lesion onset was directly observed on serial clinical MRI and the age estimate is therefore exact; groups were compared using the Mann-Whitney U test. Statistical analyses were performed in JMP (version 18; SAS Institute) and R (version 4.3; survival package). Two-sided p < 0.05 was the primary significance threshold. For qMRI comparisons only, false discovery rate (FDR) correction (Benjamini-Hochberg) and Bonferroni correction (α/k = 0.05/18 = 0.0028) were applied as secondary criteria. Inter-rater reliability among the 3 readers for Probable/Possible PRL classification was assessed using Fleiss kappa. Claude (Anthropic, Claude Sonnet 4.6) was used to assist with restructuring of Section 3.4 , condensing the abstract for submission, and generating the schematic figure for Figure 4 . All AI-assisted content was reviewed, verified for scientific accuracy, and approved by all authors before submission.
Standard Protocol Approvals, Registrations, and Patient Consents
The research protocol was approved by the Committee on Human Research at University of California, San Francisco (UCSF), and informed consent was obtained from all participants before enrollment.
Data Availability
The data that support the findings of this study are available on reasonable request from the corresponding author. The data are not publicly available due to privacy or ethical restrictions.
Results
Demographic
One hundred and six PwPMS were enrolled in the study: 34 (32%) had a primary progressive (PP) course and 72 (68%) had a SP, course. The mean age at the time of the MRI was 56 ± 10 years, and the mean disease duration was 20 ± 11 years. The median EDSS was 5.5(interquartile range 4, 6.5, range 1.5–8.5). Sixty-nine percent of participants were female, and 44% were on DMT at the time of the MRI.
Inter-Rater Reliability for Probable PRL Classification
Of 176 PRL candidates identified during initial screening, 114 demonstrated a prominent paramagnetic rim and were classified as Definite PRLs. The remaining 62 showed reduced rim conspicuity and therefore underwent additional evaluation to distinguish Probable from Possible PRLs, a classification boundary that relies more heavily on rater judgment and represents the greatest source of diagnostic subjectivity. 16 To ensure that retention of Probable PRLs was not driven by any single reader, all 62 lower-conspicuity candidates were independently reviewed by 3 readers. Ultimately, 41 candidates were retained as Probable PRLs and 21 were excluded (i.e., Possible PRLs). Full 3 readers agreement was reached for 28 of 41 Probable PRL. The remaining 13 required a final consensus between the 2 most senior readers (S.S. and E.L.) to be retained as Probable PRLs. Inter-rater agreement across the 3 independent reads, assessed before final consensus, was moderate to substantial, with a lesion-level Fleiss' κ of 0.577. This level of agreement is consistent with, and arguably more conservative than, the κ of 0.53 reported by Hemond et al. 16 for the same categorical boundary, which was calculated at the patient level, rather than the lesion level, between 2 raters after established local consensus.
Distribution of PRLs Across Participants and Relationship to Demographic and Clinical Variables
Among PwPMS, 52% showed ≥1 PRL (mean 1.6 ± 2.4, maximum 14), including either Definite or Probable PRLs. Nineteen percent had at least 1 Probable PRL (9% of PP and 24% of SP; χ 2 p = 0.05). PRLs were more frequently classified as Probable in SP compared with PP (mean number of Probable PRLs in SP = 1.1 ± 2.1 vs PP = 0.2 ± 0.3; p = 0.004). A higher overall total PRL count was associated with younger age ( p < 0.001), whereas no association was found with DMT use. A higher count of Definite PRLs was associated with younger age ( p < 0.0004). Participants with only Definite PRLs (N = 35) had a shorter disease duration compared with those with at least 1 Probable PRL (N = 20) (13 vs 21 years, p < 0.002) and with those without any PRLs (N = 51) (13 vs 24 years, p < 0.0001). To assess the independent contributions of age and disease duration, a multivariable logistic regression model was fitted with PRL presence as the dependent variable and age and disease duration as covariates. In this model, disease duration remained a significant independent predictor of PRL presence (β = −0.057, p = 0.009), whereas age was no longer associated with PRL presence after adjustment (β = −0.034, p = 0.18), highlighting that the negative relationship between PRL presence and disease duration is statistically more important than negative association between PRLs presence and age.
Because the EDSS scale is ordinal, comparisons were performed using a t test on rank-transformed EDSS values. No significant difference in disability was observed between PwPMS with (N = 55) or without PRLs (N = 51) ( p = 0.5). However, participants with only Definite PRLs (N = 35) showed significantly lower disability than those with at least 1 Probable PRL (N = 20) ( p = 0.04).
Probable PRLs Exhibit Greater Lesion Age When Compared With Definite PRLs
Lesion age distributions differed significantly between Definite and Probable PRLs ( Figure 5 ). In the unadjusted log-normal AFT model fitted among the 140 lesions with a nonzero observed age interval, Probable PRLs had a 4.02-fold longer expected age than Definite PRLs (time ratio: 4.02, 95% CI 2.16–7.51, p < 0.0001). In the multivariable log-normal AFT model—adjusted for sex, disease duration, and disease course and fitted among the 140 lesions with a nonzero observed age interval—Probable PRLs had a 3.36-fold longer expected age than Definite PRLs (time ratio: 3.36, 95% CI 1.81–6.25, p = 0.0001). This estimate was stable after accounting for within-participant clustering using cluster-robust sandwich standard errors (time ratio: 3.36, 95% CI 1.66–6.82, p = 0.0008) and in a multivariable one-lesion-per-participant sensitivity analysis (n = 51 participants; time ratio: 4.62, 95% CI 1.83–11.67, p = 0.0012) indicating that the finding was not driven by participants contributing multiple lesions. The Kaplan-Meier log-rank test independently confirmed a significant difference in age distributions (χ 2 = 18.5, p = 1.7 × 10 −5 ). In the sensitivity analysis restricted to the 44 noncensored lesions with a nonzero observed age interval—for which onset was directly observed on serial clinical MRI and the age estimate is therefore exact—Probable PRLs had a longer mean observed age than Definite PRLs (8.8 vs 5.1 years; Mann-Whitney U test, p = 0.017), confirming that the age difference persists in the subset of lesions free of any censoring assumption.
qMRI Metrics in Definite vs Probable PRLs
Quantitative MRI metrics comparing Probable and Definite PRLs are summarized in Table .
Table
Quantitative MRI Metrics—Whole-Lesion ROI and Core ROI: Probable vs Definite PRLs
Abbreviations: BH = Benjamani-Hochberg; PRL = paramagnetic rim lesion; QSM = quantitative susceptibility mapping. Nonsignificant values after correction are shown without asterisk. Values are median (interquartile range). Mann-Whitney U test (2-sided). Bonferroni and Benjamini-Hochberg FDR corrections applied across k = 18 comparisons. Sixteen comparisons are presented in this table. Multiple-comparison correction was applied across 18 prespecified quantitative lesion-level comparisons, including rim QSM values and lesion size among PRL subtypes (reported in the main text). a Significant after Bonferroni (α/k = 0.05/18 = 0.0028) or Benjamini-Hochberg FDR correction ( q < 0.05).
| Metric | Probable PRLs (n = 41) | Definite PRLs (n = 114) | Unadjusted p value | Bonferroni p Value | BH FDR p Value | ||
| Median | Q1–Q3 | Median | Q1–Q3 | ||||
| Whole-lesion ROI | |||||||
| T1 signal intensity | 224.3 | 194.3 to 244.5 | 241.5 | 201.0 to 296.1 | 0.0320 | 0.5765 | 0.0356 a |
| FA | 0.325 | 0.282 to 0.344 | 0.350 | 0.297 to 0.398 | 0.0105 | 0.1896 | 0.0124 a |
| MD (×10 −3 mm 2 /s) | 1.144 | 0.971 to 1.270 | 0.941 | 0.794 to 1.076 | <0.0001 | <0.0001 a | <0.0001 a |
| AD (×10 −3 mm 2 /s) | 1.533 | 1.385 to 1.702 | 1.317 | 1.144 to 1.680 | <0.0001 | 0.0002 a | <0.0001 a |
| RD (×10 −3 mm 2 /s) | 0.950 | 0.823 to 1.083 | 0.744 | 0.619 to 0.886 | <0.0001 | <0.0001 a | <0.0001 a |
| NDI | 0.197 | 0.153 to 0.277 | 0.315 | 0.250 to 0.388 | <0.0001 | <0.0001 a | <0.0001 a |
| FISO | 0.144 | 0.104 to 0.210 | 0.078 | 0.044 to 0.120 | <0.0001 | <0.0001 a | <0.0001 a |
| QSM (ppb) | −5 | −10 to 1 | 7 | −6 to 20 | 0.0002 | 0.0027 a | 0.0003 a |
| Core ROI | |||||||
| T1 signal intensity (a.u.) | 160.3 | 135.0 to 192.8 | 185.2 | 148.7 to 236.8 | 0.0048 | 0.0871 | 0.0065 a |
| FA | 0.241 | 0.216 to 0.298 | 0.291 | 0.236 to 0.362 | 0.0027 | 0.0479 a | 0.0038 a |
| MD (×10 −3 mm 2 /s) | 1.400 | 1.198 to 1.593 | 1.108 | 0.885 to 1.305 | <0.0001 | <0.0001 a | <0.0001 a |
| AD (×10 −3 mm 2 /s) | 1.773 | 1.359 to 1.970 | 1.487 | 1.229 to 1.726 | <0.0001 | 0.0003 a | <0.0001 a |
| RD (×10 −3 mm 2 /s) | 1.208 | 1.013 to 1.402 | 0.912 | 0.715 to 1.126 | <0.0001 | 0.0001 a | <0.0001 a |
| NDI (neurite density index) | 0.130 | 0.086 to 0.185 | 0.224 | 0.147 to 0.324 | <0.0001 | 0.0005 a | <0.0001 a |
| FISO (isotropic fraction) | 0.194 | 0.137 to 0.263 | 0.081 | 0.038 to 0.152 | <0.0001 | <0.0001 a | <0.0001 a |
| QSM (ppb) | 1 | −5 to 8 | 18 | 4–40 | <0.0001 | 0.0003 a | <0.0001 a |
At a whole-lesion ROI level, compared with Definite PRLs, Probable PRLs showed a qMRI profile suggestive of greater tissue disruption, with higher RD, AD, MD, and FISO and lower NDI, FA (all p < 0.001; all surviving Bonferroni correction aside from FA). Probable PRLs had lower T1 signal intensity within the whole lesion ( p = 0.032, FDR q = 0.036, not surviving Bonferroni). Probable PRLs showed markedly lower rim susceptibility compared with Definite PRLs (rim QSM: 8 vs 22 ppb, p < 0.001, surviving FDR and Bonferroni correction). Probable PRL had significantly lower susceptibility values even in the whole-lesion and core ROIs ( Table ). Probable PRLs were also larger than Definite PRLs (2,556 vs 1,074 mm 3 , p < 0.001, surviving FDR and Bonferroni correction). The same pattern of greater tissue disruption was observed within the core ROI across all diffusion metrics (MD, AD, RD, NDI, FISO, and FA) with all differences surviving both FDR and Bonferroni correction (all p < 0.001). Core T1 also differed significantly between groups ( p = 0.005), surviving FDR (q = 0.007) but not Bonferroni correction.
Discussion
qMRI was used to investigate the structural correlates underlying the heterogeneous appearance of PRLs in a cohort of people with PwPMS. This data set provides a unique opportunity to examine the lesion composition of distinct radiologic PRL phenotypes—Definite vs Probable PRLs—within the context of established clinical disease progression. In our cohort of progressive MS, PRLs were detectable in only 52% of participants, with Definite PRLs present in just 33%. Younger age was associated with a higher PRLs burden and a greater likelihood of PRLs being classified as Definite. Lesion-level analyses further demonstrated that Probable PRLs were more frequently older than Definite PRLs ( Figure 5 ) and showed greater microstructural disruption on qMRI ( Table ). Overall, these cross-sectional findings are consistent with the interpretation that Probable PRLs with lower rim susceptibility tend to be older, whereas Definite PRLs with a prominent rim are more likely to represent lesions at an earlier age with less underlying tissue disruption.
PRLs are classified as Definite when a clearly visible, continuous peripheral paramagnetic rim is confidently identified around a nonacute WM lesion, with minimal interpretative uncertainty. 16 In line with this definition, Definite PRLs in our cohort exhibited higher rim susceptibility on QSM compared with Probable PRLs, indicating that rim conspicuity and susceptibility magnitude are key determinants of this classification as previously reported. 16 Our data are consistent with the interpretation that PRLs become less likely to be classified as Definite as they age. Paramagnetic rims form early during lesion development after the resolution of acute inflammation and contrast enhancemenand do not arise in established lesions. 12 After the acute inflammatory phase, susceptibility values gradually decline within PRLs, affecting both the rim and the lesion core, as well as non-PRL lesions. 25,26 Consequently, rim visibility on the phase image declines over time. 18 These time-dependent changes are consistent with reports that PRLs are more frequent in younger individuals and at earlier stages of MS. 14,27
The observation that PRLs labelled as Probable can be older and exhibit greater tissue disruption than PRLs labelled as Definite has important implications for the use of PRLs in MS diagnosis and disease monitoring. Definite PRLs were more common in participants with shorter disease duration, whereas Probable PRLs—and even the absence of PRLs—were associated with longer disease duration. These findings are consistent with the notion that PRLs naturally tend to fade and may eventually disappear, even in the presence of ongoing clinical progression. Notably, although overall disability did not differ between participants with and without PRLs, those with Probable PRLs had higher disability than those with only Definite PRLs. Given that Probable PRLs were, on average, older, it is reasonable to infer that PRLs that persist for many years become progressively fainter. 18 Consequently, in progressive MS, PRLs' interpretation becomes increasingly challenging as disease duration increases, rim conspicuity decreases, and disability worsens ( Figure 4 ). The absence of PRLs should not be interpreted as a favorable prognostic indicator in progressive MS.
The paramagnetic rim in PRLs corresponds to iron-laden phagocytes at the borders of WM lesions. 28,29 This peripheral rim of microglia is typical of CAL, but the presence of iron within those cells is not required to label a lesion as CAL. Approximately, only from 25% to 45% of CALs contain iron. 30-32 The defining pathologic feature of CALs is chronically activated microglia at the lesion edge, rather than iron accumulation, which likely reflects only a transient phase of chronic activity. 18,31 It follows that a reduction in rim susceptibility or rim disappearance cannot be reliably interpreted as unambiguous evidence of therapeutic efficacy against compartmentalized inflammation. Our data are consistent with the possibility that rim attenuation may reflect, at least in part, the natural aging of PRLs rather than exclusively representing a response to treatment. However, given the cross-sectional and retrospective nature of this study, we cannot formally exclude treatment effects, and this interpretation remains hypothesis generating.
An important interpretive question is whether Probable PRLs represent a temporally later stage of Definite PRLs or a biologically distinct lesion subtype. Our primary hypothesis is that progressive iron depletion at the lesion rim renders the paramagnetic rim progressively fainter, eventually falling below the threshold for Definite classification. This interpretation is supported by the greater lesion age, larger lesion volume, more advanced microstructural tissue destruction across all qMRI metrics, and the systematic association of Probable PRLs with longer disease duration and SP MS observed in our cohort, all consistent with prior longitudinal evidence that PRL rims may fade over approximately an average of 7 years. 18 We nonetheless acknowledge that a distinct subtype hypothesis cannot be excluded on cross-sectional data alone. Probable PRLs may reflect lesions in which rim susceptibility is intrinsically more heterogeneous or spatially fragmented, independent of their longitudinal evolution. However, regardless of which interpretation is correct, the clinical and trial design implications are identical: Rim conspicuity alone is insufficient to infer the degree of underlying inflammatory activity, and rim disappearance should not be taken as unambiguous evidence of therapeutic benefit without concurrent assessment of tissue integrity. Longitudinal studies tracking individual lesions transitioning from Definite to Probable classification will be required to definitively resolve this question.
A recent longitudinal study reported that PRL disappearance was associated with improved clinical outcomes in a predominantly relapsing-remitting MS cohort. 33 However, rim loss could not be attributed to treatment exposure, suggesting that this association may reflect an epiphenomenon related to the intrinsic biology and natural evolution of PRLs rather than, or in addition to, a direct therapeutic effect, although this cannot be confidently established from short-term observational data alone. 33 PRLs are well recognized as a dynamic biomarker, and the paramagnetic rim typically changes or diminishes as lesions age, a process that may occur independently of treatment. 18 This is consistent with the fact that even the most potent therapies are thought to exert little direct effect on compartmentalized inflammation because they do not adequately penetrate the largely intact blood-brain barrier. 3,34 A longitudinal analysis performed in this same cohort showed that PRLs overlapping with T1 black holes at baseline were more likely to disappear over time, whereas persistent PRLs were more likely to evolve into new T1 black holes. 27 These findings support our hypothesis that rim fading can also reflect progressive tissue degeneration and should not be unequivocally interpreted as resolution of chronic inflammation.
Taken together, our study and the growing body of literature support the need to reconceptualize PRLs as therapeutic targets in clinical trials aimed at decreasing compartmentalized inflammation. Specifically:
1. Disappearance or qualitative attenuation of the paramagnetic rim on serial imaging cannot be reliably interpreted as resolution of chronic lesion activity without concurrent evaluation of qMRI metrics and lesion size to assess underlying tissue damage and exclude SEL behavior ( Figure 6 ).
2. Persistence of the paramagnetic rim may be a more informative marker of ongoing compartmentalized inflammation than rim disappearance is of resolving inflammation. Rim attenuation or loss cannot distinguish between true inflammatory resolution and continued tissue injury, whereas rim persistence more directly reflects sustained iron-associated microglial activation.
3. Quantitative susceptibility measures may offer greater sensitivity to changes in iron content, but validation against normative qMRI longitudinal data and correlation with long-term clinical outcomes, or matched untreated cohorts, are necessary in the context of clinical trials.
Accordingly, future studies should move beyond a binary interpretation of PRL disappearance and instead examine the longitudinal relationships among rim persistence, qMRI-derived tissue changes, and clinical outcomes. Longitudinal analyses are needed to define qMRI trajectories associated with PRL persistence, attenuation, and disappearance.
This study has several limitations. First, qMRI analyses were cross-sectional. The interpretation that Probable PRLs represent a later evolutionary stage of Definite PRLs is a contextualization of our findings within the existing literature, only a hypothesis and not something directly demonstrable from our cross-sectional data. Alternative interpretations—that Definite and Probable PRLs represent biologically distinct lesion subtypes rather than sequential stages—cannot be fully excluded. We note, however, that a sensitivity analysis restricted to the 44 noncensored lesions with directly observed onset confirmed the direction and significance of the age difference between Probable and Definite PRLs (Mann-Whitney U , p = 0.017; eFigure 2). Furthermore, the convergence of multiple independent observations—including the association of Definite PRLs with shorter disease duration and PP MS, the greater tissue disruption seen across all qMRI metrics in Probable PRLs, and independent longitudinal evidence linking rim fading to lesion aging—seem to favor temporal evolution as the most parsimonious explanation. Careful review of longitudinal clinical imaging enabled assignment of a temporal profile to each PRL, strengthening interpretability. Ultimately, whether this cross-sectional difference here detected reflects a temporal trajectory or a biological dichotomy, the clinical and trial design implications are identical: Rim conspicuity alone is insufficient to infer the degree of underlying inflammatory activity. Second, we acknowledge that rim conspicuity on phase imaging, which forms the basis of PRL subtype classification, and rim QSM, which serves as a quantitative descriptor of iron content, share a common biological substrate. However, they are assessed through independent processes: Classification is a binary visual judgment on filtered phase images performed before and independently of quantitative extraction, while rim QSM is derived from a reconstructed susceptibility map through a separate processing pipeline. Crucially, the primary evidence supporting our conclusions rests on the demography of findings and diffusion-based tissue metrics—radial diffusivity, mean diffusivity, fractional anisotropy, and NODDI parameters—as well as lesion size and T1 signal intensity. The QSM rim comparison is presented as mechanistic corroboration of the visual classification, not as independent evidence of tissue damage. Third, inclusion was limited to progressive MS, restricting generalizability to relapsing remitting disease. Although some PRLs in relapsing MS may regress without tissue injury, focusing on progressive MS provides a clearer window into lesions more likely to correspond pathologically to CALs. Fourth, the retrospective analytical framework renders the findings hypothesis generating, with nominal statistical inference requiring independent replication. Finally, the lack of direct histopathologic correlation remains a key limitation, although our results are consistent with recent pathologic evidence, indicating that iron enrichment occurs in only a minority of CALs. 32
Discussion
PRLs remain among the most compelling and widely discussed entities in the MS field. However, amid growing scientific enthusiasm, caution is warranted when interpreting changes in paramagnetic rim susceptibility as evidence of treatment response, particularly in the absence of longitudinal controlled data—especially when such changes are used to evaluate neuroprotective strategies targeting CALs, which do not necessarily harbor iron at the histologic level. Our findings suggest that, in progressive MS, PRLs with lower rim susceptibility on both phase imaging and QSM—and therefore more likely to be classified as Probable rather than Definite—may represent older, persistent lesions characterized by more advanced underlying tissue damage. Because disease duration increases and disability worsens, the detection, classification, and interpretation of PRL pathophysiology become increasingly complex. Changes in the paramagnetic rim observed on serial imaging should not be interpreted in isolation as evidence of resolving inflammation. Persistence of the paramagnetic rim may be a more informative marker of ongoing compartmentalized inflammation than rim disappearance is of resolving inflammation.
Acknowledgment
The authors are deeply grateful to the patients for their long-term participation in the EPIC and ORIGINS studies. The following individuals contributed to data collection and study coordination as part of the University of California, San Francisco MS-EPIC Team: Sergio Baranzini PhD, Richard Cuneo MD, Sasha Gupta MD, Joseph Sabatino MD, Joanne Guo MD, Jill Hollenbach PHD, Sam Pleasure MD, PhD, Anna Sindalovsky, Adam Renschen, Stacy Caillier, Asritha Tubati, Hope Giron, Neville Varghese, Isabella Poppen, Charlese Yeeloy, Ahmed Abdelhak MD, Alyssa Nylander MD, Jacqueline Marcus MD, Ariele Greenfield MD.
Glossary
axial diffusivity
accelerated failure time
chronic active lesion
Diffusion Imaging in Python
diffusion tensor imaging
Expanded Disability Status Scale
echo-planar imaging
false discovery rate
fluid-attenuated inversion recovery
high angular resolution diffusion imaging
parallel imaging acceleration factor
mean diffusivity
magnetization-prepared rapid acquisition gradient echo
multiple sclerosis
neurite density index
Neuroimaging Informatics Technology Initiative
Neurite Orientation Dispersions Density Imaging
primary progressive
paramagnetic rim lesion
people with progressive MS
quantitative MRI
quantitative susceptibility mapping
radial diffusivity
region of interest
slowly enlarging lesion
secondary progressive
susceptibility tensor imaging
repetition time
University of California, San Francisco
variable-kernel sophisticated harmonic artifact reduction for phase data
white matter
Author Contributions
S. Sacco: drafting/revision of the manuscript for content, including medical writing for content; major role in the acquisition of data; study concept or design; analysis or interpretation of data. H. Zhao: analysis or interpretation of data. S. Cheng: analysis or interpretation of data. A.V. Akula: analysis or interpretation of data. E. Luskin: drafting/revision of the manuscript for content, including medical writing for content; analysis or interpretation of data. J. Zhang: analysis or interpretation of data. O. Manouvakhova: analysis or interpretation of data. T.P. Cooper: major role in the acquisition of data. W. Stern: major role in the acquisition of data. V.J. Block: drafting/revision of the manuscript for content, including medical writing for content; major role in the acquisition of data. A. Virupakshaiah: drafting/revision of the manuscript for content, including medical writing for content. R. Gomez: major role in the acquisition of data. A. Santaniello: major role in the acquisition of data. P. Sati: drafting/revision of the manuscript for content, including medical writing for content; study concept or design. N. Papinutto: drafting/revision of the manuscript for content, including medical writing for content; analysis or interpretation of data. R. Bove: drafting/revision of the manuscript for content, including medical writing for content; major role in the acquisition of data. J.M. Gelfand: drafting/revision of the manuscript for content, including medical writing for content; major role in the acquisition of data. D.S. Goodin: drafting/revision of the manuscript for content, including medical writing for content; major role in the acquisition of data. A.J. Green: major role in the acquisition of data. J. Oksenberg: drafting/revision of the manuscript for content, including medical writing for content; major role in the acquisition of data. E. Waubant: drafting/revision of the manuscript for content, including medical writing for content; major role in the acquisition of data. M.R. Wilson: drafting/revision of the manuscript for content, including medical writing for content; major role in the acquisition of data. S.S. Zamvil: major role in the acquisition of data. B.A.C. Cree: drafting/revision of the manuscript for content, including medical writing for content; major role in the acquisition of data. S.L. Hauser: drafting/revision of the manuscript for content, including medical writing for content; major role in the acquisition of data; analysis or interpretation of data. R.G. Henry: drafting/revision of the manuscript for content, including medical writing for content; major role in the acquisition of data; study concept or design; analysis or interpretation of data.
Study Funding
This study was supported by a grant from the NIH R35NS111644 and by the Valhalla and Spangler Foundations. The UCSF MS Biorepository is supported by a grant from a National Multiple Sclerosis Society (SI-2001-35701). P. Sati and E. Waubant are supported by the Department of Defense (HT9425-23-1-0571), the National MS Society (RG-2110-38526), the Cedars-Sinai Women's Guild Neurology Project, and the Erwin Rautenberg Foundation. A.J. Green was supported by NINDS R01NS105741, R01AG062562, R01AG038791 and NMSS RG-1707-28564. S.S. Zamvil was supported by NIH RO1AI170863, Race to Erase MS foundation and ARPA-H D24AC00084-00.
Disclosure
P. Sati received research support from the NIH, National Multiple Sclerosis Society, Department of Defense, Biogen Idec, and the Erwin Rautenberg Foundation; and received consulting fees from Sanofi and MJH Life Sciences. R.M. Bove received research funding from Biogen, Novartis, Roche Genentech; and Eli Lilly; and received consulting fees from Cadenza Bio, Alexion, Amgen, EMD Serono, Genzyme Sanofi, and TG Therapeutics. A.J. Green received research funding from All May See, Westridge Foundation, Roche, Pipeline Pharmaceuticals, and Cognito Therapeutics. M.R. Wilson received research funding from Roche/Genentech and Novartis; received royalties from CDI labs; received honoraria for lectures form Takeda, Novartis, WebMD, and Genentech; and received consulting fees, stock, and support for attending meetings and/or travel from Delve Bio. S.S. Zamvil received research funding from Genentech, consulting fees from Alexion, Sanofi-Genzyme, and Sixsenth; and received honoraria for lectures from Alexion and Sanofi-Genzyme. S.L. Hauser serves on the scientific advisory boards of Accure, Alector, and Hinge; has previously consulted for BD, Moderna, NGM Bio, and Pheno Therapeutics; previously served on the board of directors of Neurona; and has received travel reimbursement and writing support from F. Hoffmann-La Roche Ltd. and Novartis AG for anti-CD20 therapy-related meetings and presentations. R.G. Henry received consulting fees from Roche, Sanofi, Novartis, Myelin Repair Foundation, QIA, and Boston Pharma; and received research funding from Roche/Genentech and Atara. S. Sacco, H. Zhao, S. Cheng, A.V. Akula, E. Luskin, J. Zhang, O. Manouvakhova, T.P. Cooper, W. Stern, V.J. Block, A. Virupakshaiah, R. Gomez, A. Santaniello, N. Papinutto, J.M. Gelfand, D.S. Goodin, J.R. Oksenberg, E. Waubant, and B.A. Cree have no relevant disclosures. Go to Neurology.org/NN for full disclosures.
References
1. Hartung HP, Benedict RHB, Berger T, et al. Ocrelizumab in early-stage relapsing-remitting multiple sclerosis: the phase IIIb ENSEMBLE 4-Year, single-arm, open-label trial. Neurology. 2024;103(12):e210049. doi: 10.1212/WNL.0000000000210049
2. Schuldesz AC, Tudor R, Cornea A, Geni DN, Lata IN, Simu M. Effectiveness of ocrelizumab on disease progression and disability status in multiple sclerosis patients: a two-year prospective cohort study. J Clin Med. 2025;14(2):553. doi: 10.3390/jcm14020553
3. Maggi P, Bulcke CV, Pedrini E, et al. B cell depletion therapy does not resolve chronic active multiple sclerosis lesions. EBioMedicine. 2023;94:104701. doi: 10.1016/j.ebiom.2023.104701
4. De Groot CJ, Bergers E, Kamphorst W, et al. Post-mortem MRI-Guided sampling of multiple sclerosis brain lesions: increased yield of active demyelinating and (p)reactive lesions. Brain. 2001;124(Pt 8):1635-1645. doi: 10.1093/brain/124.8.1635
5. Sacco S, Caverzasi E, Papinutto N, et al. Neurite orientation dispersion and density imaging for assessing acute inflammation and lesion evolution in MS. AJNR Am J Neuroradiol. 2020;41(12):2219-2226. doi: 10.3174/ajnr.A6862
6. Kornek B, Storch MK, Weissert R, et al. Multiple sclerosis and chronic autoimmune encephalomyelitis: a comparative quantitative study of axonal injury in active, inactive, and remyelinated lesions. Am J Pathol. 2000;157(1):267-276. doi: 10.1016/S0002-9440(10)64537-3
7. Absinta M, Sati P, Masuzzo F, et al. Association of chronic active multiple sclerosis lesions with disability in vivo. JAMA Neurol. 2019;76(12):1474-1483. doi: 10.1001/jamaneurol.2019.2399
8. Preziosa P, Pagani E, Meani A, et al. Slowly expanding lesions predict 9-Year multiple sclerosis disease progression. Neurol Neuroimmunol Neuroinflamm. 2022;9(2):e1139. doi: 10.1212/NXI.0000000000001139
9. Dal-Bianco A, Grabner G, Kronnerwetter C, et al. Slow expansion of multiple sclerosis iron rim lesions: pathology and 7 T magnetic resonance imaging. Acta Neuropathol. 2017;133(1):25-42. doi: 10.1007/s00401-016-1636-z
10. Toru Asahina A, Lu J, Chugh P, et al. Prognostic significance of paramagnetic rim lesions in multiple sclerosis: a systematic review. J Clin Neurosci. 2024;129:110810. doi: 10.1016/j.jocn.2024.110810
11. Elliott C, Rudko DA, Arnold DL, et al. Lesion-level correspondence and longitudinal properties of paramagnetic rim and slowly expanding lesions in multiple sclerosis. Mult Scler. 2023;29(6):680-690. doi: 10.1177/13524585231162262
12. Clark KA, Manning AR, Chen L, et al. Early magnetic resonance imaging features of new paramagnetic rim lesions in multiple sclerosis. Ann Neurol. 2023;94(4):736-744. doi: 10.1002/ana.26727
13. Tranfa M, Cantoni V, Green R, et al. Prevalence and incidence of paramagnetic rim lesions in multiple sclerosis: a systematic review and meta-analysis. Radiology. 2025;316(3):e243718. doi: 10.1148/radiol.243718
14. Klistorner S, Usnich T, Clarke MA, et al. The presence of a paramagnetic phase rim in multiple sclerosis is linked to lesion age: an exploratory study. Mult Scler J Exp Transl Clin. 2025;11(3):20552173251378788. doi: 10.1177/20552173251378788
15. Sacco S, Virupakshaiah A, Papinutto N, et al. Susceptibility-based imaging aids accurate distinction of pediatric-onset MS from myelin oligodendrocyte glycoprotein antibody-associated disease. Mult Scler. 2023;29(14):1736-1747. doi: 10.1177/13524585231204414
16. Hemond CC, Dundamadappa SK, Deshpande M, et al. Paramagnetic rim lesions are highly specific for multiple sclerosis in real-world data. Brain Commun. 2025;7(3):fcaf211. doi: 10.1093/braincomms/fcaf211
17. Marrodan M, Yañez P, Calandri IL, et al. Impact of oral cladribine on paramagnetic rim lesions of multiple sclerosis patients. Mult Scler Relat Disord. 2025;96:106339. doi: 10.1016/j.msard.2025.106339
18. Dal-Bianco A, Grabner G, Kronnerwetter C, et al. Long-term evolution of multiple sclerosis iron rim lesions in 7 T MRI. Brain. 2021;144(3):833-847. doi: 10.1093/brain/awaa436
19. Andersson JL, Skare S, Ashburner J. How to correct susceptibility distortions in spin-echo echo-planar images: application to diffusion tensor imaging. Neuroimage. 2003;20(2):870-888. doi: 10.1016/S1053-8119(03)00336-7
20. Smith SM, Jenkinson M, Woolrich MW, et al. Advances in functional and structural MR image analysis and implementation as FSL. Neuroimage. 2004;23(suppl 1):S208-S219. doi: 10.1016/j.neuroimage.2004.07.051
21. Li W, Wang N, Yu F, et al. A method for estimating and removing streaking artifacts in quantitative susceptibility mapping. Neuroimage. 2015;108:111-122. doi: 10.1016/j.neuroimage.2014.12.043
22. Avants BB, Tustison NJ, Song G, Cook PA, Klein A, Gee JC. A reproducible evaluation of ANTs similarity metric performance in brain image registration. Neuroimage. 2011;54(3):2033-2044. doi: 10.1016/j.neuroimage.2010.09.025
23. Lou C, Sati P, Absinta M, et al. Fully automated detection of paramagnetic rims in multiple sclerosis lesions on 3T susceptibility-based MR imaging. Neuroimage Clin. 2021;32:102796. doi: 10.1016/j.nicl.2021.102796
24. Bagnato F, Sati P, Hemond CC, et al. Imaging chronic active lesions in multiple sclerosis: a consensus statement. Brain. 2024;147(9):2913-2933. doi: 10.1093/brain/awae013
25. Chen W, Gauthier SA, Gupta A, et al. Quantitative susceptibility mapping of multiple sclerosis lesions at various ages. Radiology 2014;271(1):183-192. doi: 10.1148/radiol.13130353
26. Zhang S, Nguyen TD, Hurtado Rúa SM, et al. Quantitative susceptibility mapping of time-dependent susceptibility changes in multiple sclerosis lesions. AJNR Am J Neuroradiol. 2019;40(6):987-993. doi: 10.3174/ajnr.A6071
27. Reeves JA, Bartnik A, Mohebbi M, et al. Determinants of long-term paramagnetic rim lesion evolution in people with multiple sclerosis. Ann Clin Transl Neurol. 2025;12(2):267-279. doi: 10.1002/acn3.52253
28. Bagnato F, Hametner S, Yao B, et al. Tracking iron in multiple sclerosis: a combined imaging and histopathological study at 7 tesla. Brain. 2011;134(Pt 12):3602-3615. doi: 10.1093/brain/awr278
29. Dal-Bianco A, Oh J, Sati P, Absinta M. Chronic active lesions in multiple sclerosis: classification, terminology, and clinical significance. Ther Adv Neurol Disord. 2024;17:17562864241306684. doi: 10.1177/17562864241306684
30. Calabrese M, Preziosa P, Scalfari A, et al. Determinants and biomarkers of progression independent of relapses in multiple sclerosis. Ann Neurol. 2024;96(1):1-20. doi: 10.1002/ana.26913
31. Popescu BF, Frischer JM, Webb SM, et al. Pathogenic implications of distinct patterns of iron and zinc in chronic MS lesions. Acta Neuropathol. 2017;134(1):45-64. doi: 10.1007/s00401-017-1696-8
32. Najm L, Chomyk A, Cyncynatus K, Davalos D, Mahajan KR, Trapp BD. Demyelinated lesion associated compartmental inflammation in progressive multiple sclerosis brains. Acta Neuropathol Commun. 2025;13(1):204. doi: 10.1186/s40478-025-02122-9
33. Reeves JA, Bartnik A, Jakimovski D, et al. Associations between paramagnetic rim lesion evolution and clinical and radiologic disease progression in persons with multiple sclerosis. Neurology. 2024;103(10):e210004. doi: 10.1212/WNL.0000000000210004
34. Okuda DT, Punnen TG, Moog TM, McCreary MC, Burgess KW, Lebrun-Frénay C. Understanding the therapeutic impact of disease-modifying therapies or no treatment on enlarging T1-Weighted hypointense lesions and total T1-Weighted hypointense lesion volume in MS. AJNR Am J Neuroradiol. 2026;47(3):843-851. doi: 10.3174/ajnr.A8994
Associated Data
Data Availability Statement
The data that support the findings of this study are available on reasonable request from the corresponding author. The data are not publicly available due to privacy or ethical restrictions.