Early MRI features of new paramagnetic rim lesions in Multiple Sclerosis

Abstract

Objective:

To determine early MRI features of new multiple sclerosis (MS) lesions that will develop into paramagnetic rim lesions (PRLs), which have been associated with progressive tissue injury in MS.

Methods:

New contrast enhancing lesions (CELs) observed on routine clinical MRI were imaged at 7-tesla (T) within 4 weeks of observation, and 3 and 6 months later. The 6-month MRI was used to classify PRL status (PRL or non-PRL). The relationship between early lesion characteristics and subsequent PRL status was assessed using generalized linear mixed effects models. Random forest classification was performed to classify early predictors of subsequent PRL status.

Results:

From 93 CELs in 23 MS patients, 37 lesions developed into a PRL. In lesions that developed into PRLs compared to those that did not, the average lesion T₁ on the initial 7T MRI was 1994 ms compared to 1670 ms (p-value < 0.001) and the average volume was 168.7 mL compared to 44 mL (p-value < 0.001) in lesions that did not. These volume differences were also found on 3T scans (p-value < 0.001) and for intensity-normalized T₁-w (p-value = 0.011) and FLAIR (p-value = 0.005). The area under the receiver operating characteristic curve for the random forest classification with leave-one-out cross validation was found to be 0.86 using initial 7T features.

Interpretation:

New MS lesions that evolve into PRLs can be identified early in lesion evolution. These findings suggest that biological mechanisms underlying PRL development begin early, which has important implications for clinical trials targeting PRLs development and subsequent therapeutics.

Introduction:

Chronic inflammation compartmentalized to the Central Nervous System (CNS) is thought to be an important process leading to neurodegeneration and contribute to progressive disability in Multiple Sclerosis (MS). An emerging MRI biomarker of chronic inflammation in MS is the paramagnetic rim lesion (PRL). PRLs comprise a subset of chronic (> 6 months after lesion development) focal white matter lesions and they are observed to have a peripheral rim of paramagnetism that is best visualized on ultrahigh-field 7-tesla (7T) MRI using susceptibility/T₂*-weighted sequences^4-6. PRLs on MRI spatially co-localize on histopathologic sections with lesions containing a rim of activated, iron-laden macrophages where these lesions are classified as “smoldering” or “chronic-active lesions” (CALs)^1-3.

Clinical correlates of PRLs include an increase in progressive clinical phenotypes (primary-progressive and secondary-progressive MS) and earlier and greater cognitive and motor disability, supporting the contribution of PRLs to progressive CNS injury in persons with MS^8,9. Most studies have focused on the late consequences of PRLs and less is known about the early evolution of new MS lesions that develop into PRLs. Studying the development of PRLs can provide a window into the establishment of chronic inflammation in MS and provide new targets to combat progressive disease.

Prior studies report that approximately one-third of new MS lesions evolve into PRLs and about 50% of people with MS have PRLs^1,4,10. A retrospective study using 3T MRI reported that new contrast-enhancing lesions (CELs) that develop into PRLs are more likely to have ring-like contrast-enhancement and a corresponding rim of apparent diffusion coefficient compared to CELs that do not develop into PRLs¹⁰. In a separate study, CELs that developed into PRLs also had greater normalized-magnetization transfer ratio (nMTR) signal¹¹. Together, these studies suggest that early features on MRI may distinguish new lesions that develop into persistent PRLs.

Prospective studies of CELs with defined imaging time points can aid in identifying early imaging markers specific to the development of PRLs. Additionally, quantitative imaging sequences at 7T can provide new insights into the processes occurring early in new PRL development. Our aim was to prospectively and serially image new CELs first observed on routine clinical MRI using high-resolution quantitative 7T MRI to identify early features within lesions that develop into PRLs.

Methods:

Study Population

Participants were recruited from the Comprehensive Multiple Sclerosis Clinic at the University of Pennsylvania. All participants gave written consent, and the study was approved by the Institutional Review Board of the Hospital of the University of Pennsylvania. Study participants had either an established diagnosis of MS or newly diagnosed MS (including clinically isolated syndrome meeting 2017 MS diagnostic criteria), and had a new CEL observed on a clinical brain MRI. A total of 34 participants were consented and underwent at least an initial 7T MRI. Of those, 23 participants completed the three required 7T research MRIs. Data from participants who did not complete all 3 MRIs are not included in the analysis. Participants did not complete all 3 MRIs for a variety of reasons primarily due to claustrophobia or decreasing exposure to healthcare facilities during the COVID-19 pandemic. Demographics and MS clinical histories were extracted from patient charts including information on disease modifying treatment and if patients received steroid treatment between the clinical MRI and the initial 7T research MRI.

7T MRI Imaging Protocol

Participants underwent serial MRI on a Siemens (Terra) 7T whole-body MRI scanner equipped with a single transmit/32-channel receiver array head coil. Sequence parameters are displayed in Table 1.

Table 1:

7T MRI Parameters

Sequence Name	TR	TE	Resolution	Other
3D-MP2RAGE	5000 ms	2.58 ms	0.69 mm3	Ti 1 700 ms Ti 2 2700 ms GRAPPA 3
3D-ME-GRE	40 ms	1st echo = 6 ms echo space = 6 ms	0.8 mm3	6 echos GRAPPA 2
2D-DE-GRE	592 ms	1st echo = 15 ms 2nd echo = 33 ms	0.21 x 0.21 mm x 1 mm	30 slices

Abbreviations: MP2RAGE = magnetization prepared 2 rapid acquisition gradient echo ME-GRE = multi-echo gradient echo, DE-GRE = duel echo gradient echo, Ti = inversion time, TR = repetition time, TE = echo time

MP2RAGE and ME-GRE were acquired with full supratentorial coverage. DE-GRE slabs were placed over regions with known CEL identified on the recent clinical MRI. Magnitude images from the T₂*w sequences were reconstructed using the sum of squares technique and phase images were reconstructed from k-space data using adaptive combine which allowed for improved image quality respectively. T₁ maps (relaxation times) are generated on the scanner as part of the MP2RAGE package¹².

Participants underwent research 7T MRI (initial 7T) a median 22.5 days (range 6 – 38 days) from first observation of the CEL on a clinical MRI and at 3 and 6 months (+/− 2 weeks) after the initial 7T.

A subset of the 7T participants (n=14 participants, CELs = 70) had clinical 3T MRIs performed at Penn including MPRAGE and FLAIR sequences acquired at 1 mm³ isotropic resolution. This subset of participants was used to extract semi-quantitative data including normalized, to normal appearing white matter (NAWM), signal intensity that was used in the 3T analysis.

Image Processing

All images went through a semi-automated image processing pipeline that included rigid registration and N4 inhomogeneity correction using ANTs software (Version 2.1.1)¹³. All sequences acquired within a scan time point were registered to the T₁ image (MP2RAGE-UNI for 7T). T₁w images were registered to the initial 7T MP2RAGE-UNI and the transformation matrix applied to the other sequences. For a subset of clinical scans that were acquired at the University of Pennsylvania under our standard clinical protocol that includes a 3D-T₁w MPRAGE (1 mm³) and 3D-FLAIR (1 mm³), the MPRAGE and FLAIR signal intensities were normalized to NAWM using WhiteStripe¹⁴.

All CELs observed on the post-contrast T₁w image were manually segmented using ITK-SNAP (V3.8) on the MP2RAGE-T₁ map at each time-point and on the 3T FLAIR and MPRAGE. The lesion T₁ relaxation time was averaged across individual lesion voxels and lesion volume was also calculated. For clinical scans obtained at Penn using our standardized clinical sequences, CEL pattern was noted (ring-enhancing, nodular/solid, and faint).

Identification of a PRL was performed using the 4^th echo (~24ms) from the Laplacian unwrapped phase image at each timepoint and classified as a PRL or nonPRL on the 6-month 7T ME-GRE-phase image by a single rater with 8 years of experience in identifying PRLs (MKS). Only lesions that were a CEL on the clinical MRI were assessed for the subsequent presence or absence of a paramagnetic rim. To be classified as a PRL, lesions were required to have a hypointense rim for at least 2/3rds of the circumference, except for when adjacent to the ventricles or cortex where at least 2/3rds of the periphery not adjacent to these structures was required. When available, the DE-GRE phase images were also used to adjudicate the presence or absence of a paramagnetic rim.

Statistical Analysis

All statistical analyses and visualization were performed in R (version 4.1.1) (http://www.r-project.org/). Univariate generalized linear mixed effects modeling was performed using the lme4 package (version 1.1.27.1) to assess associations between several different predictors of interest (both at 3T and 7T) and the presence of paramagnetic rims on lesions 6 months from initial 7T. Each model was fit with a subject-specific random intercept to account for multiple lesions per subject and a fixed effect term for each clinical predictor of interest.

Lesion characteristics including total lesion volume, T₁ relaxation time values, and average T₁ and FLAIR after white matter normalization steroid use between 3T and 7T initial scans, disease modifying treatment (DMT) and contrast enhancement pattern at 3T baseline all were used as predictors in subsequent random forest models used to predict PRLs at 6 months. Random forest classification with the default tuning parameters and leave-one-out cross-validation was performed using the randomForest R package (version 4.6.14). Random forest models were fit for 3T and 7T data separately. Additionally, separate models were fit for the 7T initial and 3-month timepoints. Predictive performance was assessed using receiver-operator characteristic (ROC) analysis via the pROC R package (version 1.18.0). Variable importance in the predictive model was assessed using the mean decrease Gini index in the caret R package (version 6.0.90).

Results:

Population demographics and summary MS clinical histories are displayed in Table 2. Participants underwent the clinical MRI where CELs were identified for the following indications: routine surveillance (9 participants), relapse/new symptom evaluation (8 participants), and for diagnostic evaluation (6 participants). Of note, 9 of the participants received pulse dose corticosteroids between the clinical MRI and the initial 7T MRI.

Table 2:

Summary of the demographics and clinical histories of study participants

Participants (N)	23
Median Age (Range)	35 yrs (19 – 61 yrs)
Sex	16 F / 7 M
Median Disease Duration (range)	6 yrs (0 – 36 yrs)
DMT at time of CEL	None: 11
	Glatiramer acetate: 6
	Interferon: 1
	Dimethyl fumarate: 3
	Teriflunomide: 1
	Ocrelizumab: 1
CELs on clinical MRI (range)	Total 93 CELs: Median 3 CELs per participant (1 – 20 CEL per participant)
Median days between clinical 3T MRI and Initial 7T MRI (Standard deviation, range)	22.5 days (10.3 days, 6 – 38 days)

Abbreviations: DMT = disease modifying therapy, CEL = contrast enhancing lesion

From 93 CELs, 37 were classified as PRLs on the 6-month 7T T₂*-phase image. Figure 1 shows an example of the evolution of 2 CELs identified on the clinical MRI, one lesion that evolved into a PRL at 6 months (top row) and the other lesion did not (bottom row).

Figure 1: Examples of MS lesion evolution.

Two examples of CELs identified on the routine clinical 3T MRI with serial 7T imaging. Top row shows a CEL that developed into a PRL at 6 months as seen on the 6 month Phase 7T MRI. Bottom row shows a CEL that did not develop into a PRL at 6 months.

Early 7T imaging markers associate with new PRL formation

Qualitative MRI Features

On the initial 7T MRI, 32 lesions of the 37 PRLs were observed to have a hypointense paramagnetic rim, 3 lesions were observed to have solid hypointense signal throughout the lesion, and 2 lesions were not able to be classified due to lack of acquisition of the T₂*-phase. On the 3-month 7T MRI, all 3 lesions observed to initially have solid hypointense phase signal throughout the lesion were observed to have a hypointense paramagnetic rim.

Of the 56 nonPRLs, on the initial 7T T₂*-phase 4 of those lesions were observed to have a hypointense paramagnetic rim, 14 lesions were observed to have solid hypointense signal throughout the lesion, and 25 lesions were isointense to NAWM. 13 nonPRLs were not able to be classified due to lack of acquisition of the T₂*-phase sequence. On the 3-month 7T MRI, 3 lesions observed to initially have a hypointense paramagnetic rim, were observed to have solid hypointense phase signal throughout the lesion, and 1 lesion continued to have a hypointense paramagnetic rim.

Quantitative MRI Features

As depicted in Figure 2, the average lesion volume on the initial 7T MP2RAGE for lesions that became PRLs was 168.7 mL (range: 15.33 – 805.2 mL) compared to 44 mL (range: 2.283 – 261.2 mL) (p-value < 0.001) for lesions that did not (volumes are logged for better visualization purposes). Lesion volume was logged for visualization purposes in Figure 2. The average lesion T₁ on the initial 7T MP2RAGE for lesions that became PRLs was 1994 ms compared to 1670 ms in that lesions that did not become PRLs (p-value < 0.001). Lesions that evolved into PRLs remained different in T₁ relaxation and T₁ volume at 3 and 6 months.

Figure 2: Lesion characteristics at 7T.

The distribution of average T₁ times within lesions and the logged lesion volumes from MP2RAGE 7T images at the initial, 3 month, and 6 month timepoints grouped by whether or not a PRL was present in the lesion at the 6 month timepoint. Lesion volumes are logged for better visualization and to reduce skewness.

Additionally, we analyzed changes in the trajectory of T₁ within lesions between MRI time points (i.e. initial 7T and the 3-month) and (3-month 7T and 6-month) and did not observe any differences in the evolution of signal intensity over time (data not shown).

Clinical 3T imaging markers associate with new PRL formation

A subset of participants (n =14 participants, 70 lesions) whose clinical MRIs had CELs and also had 3D isotropic sequences (MPRAGE and FLAIR 1 mm³) that enabled semi-quantitative analysis were analyzed. As shown in Figure 3, the average 3T FLAIR volume of new lesions that developed into PRLs was 190.7 mL compared to 58.4 mL for lesions that did not develop into PRLs (p < 0.001). Lesion volume was logged for visualization purposes in Figure 3. The average NAWM-normalized FLAIR signal intensity of new lesions that became PRLs was 28.9 compared to 19.8 of lesions that did not develop into PRLs (p = 0.005).The average NAWM-normalized T₁-w values of new lesions that developed into PRLs was −18.1 compared to −11.8 (p = 0.011) within lesions that did not. The average normalized FLAIR signal intensity of new lesions that became PRLs was 28.9 compared to 19.8 of lesions that did not develop into PRLs (p = 0.005). The average 3T FLAIR volume of new lesions that developed into PRLs was 190.7 mL compared to 58.4 mL for lesions that did not develop into PRLs (p < 0.001).

Figure 3: Lesion characteristics at 3T.

The distribution of FLAIR logged lesion volumes and normalized (to normal appearing white matter) FLAIR and T₁ values within lesions from clinical 3T images acquire at first observation of CELs comparing values within lesions that did and did not develop into PRLs at the 6-month timepoint. Lesion volumes are logged to reduce skewness.

Generalized Linear Mixed Effects Modelling

Age of participants and steroid use between baseline 3T and initial 7T MRI were not significantly associated with PRL formation at 6 months. Generalized linear mixed effects models revealed significant associations even after FDR correction for multiple comparisons between the presence of PRLs at 6 months and DMT use at baseline, 3T lesion volume and normalized FLAIR and T₁ values at baseline 3T, initial and 3-month 7T lesion volumes, and initial and 3-month 7T T₁ values (Table 3).

Table 3.

Generalized linear mixed effects models of predictors of new PRL development

Predictor	Log Odds Ratio	95% CI	Odds Ratio	p-value	FDR p- value
Age	0.020	(−0.028 0.078)	1.02	0.433	0.481
Pulse steroids between clinical MRI and initial 7T MRI	−0.24	(−1.25, 0.55)	0.78	0.562	0.562
Log lesion vol (initial 7T)	1.64	(1.01, 2.51)	5.14	<0.001	<0.001
Per 200 ms increase in lesion T1 (initial 7T)	1.44	(0.88, 2.14)	4.22	<0.001	<0.001
Log lesion vol (3 mo 7T)	1.92	(1.17, 2.92)	6.79	<0.001	<0.001
Per 200 ms increase in lesion T1 (3mo 7T)	1.4	(0.80, 2.16)	4.06	<0.001	<0.001
Log lesion volume (clinical 3T)	1.12	(0.56, 1.97)	3.06	<0.001	<0.001
nT1 (clinical 3T)	−0.12	(−0.23, −0.04)	0.89	0.011	0.014
nFLAIR (clinical 3T)	0.10	(0.04, 0.18)	1.10	0.005	0.007
DMT (clinical 3T)	1.32	(0.44, 2.34)	3.73	0.004	0.007

Random Forest Early Classification of New PRL Formation

Random forest regression models fit with lesion characteristics acquired at initial 7T images using leave-one-subject-out cross-validation were able to classify lesions as PRL or non-PRL at 6 months with an AUC of 0.86 (95% CI [0.77, 0.95]), which was similar to using lesion characteristics acquired at 3 months on 7T images (AUC = 0.86, 95% CI [0.78, 0.94]).

Random forest models fit with data acquired at baseline on 3T images were able to classify lesions as PRL or non-PRL at 6 months with an AUC of 0.74 (95% CI [0.63,0.86]). ROC curves from each random forest model are shown in Figure 4. Gini-based variable importance measures which can be seen in Figure 5 revealed that lesion volumes and white matter nT₁-w and nFLAIR signal intensity values from clinical 3T images and lesion volumes and T₁ relaxation rates from initial and 3-month 7T images were of greater importance than steroid use, contrast enhancement pattern, DMT use, or steroid use in predicting PRLs in all 3 random forest models.

Figure 4: Random Forest Classification.

ROC curves and corresponding AUC scores after classification using random forest regression with leave one out cross validation.

Figure 5: MRI Feature Importance.

Gini-based importance of features included in the random forest regression model. Lesion volumes and T₁ intensity were about 14-19 times more important than steroid use in both the initial and 3-month models which were fit using 7T data. Lesion volume, and nT₁ and nFLAIR values were around 11 times more important than steroid use in models which were fit using data acquired from 3T baseline images.

Discussion

In this prospective MRI study, persons with MS and new CELs observed on routine clinical 3T MRI underwent serial 7T MRI to identify early imaging features that differentiate new lesions that evolve into PRLs from new lesions that do not. We studied 93 CELs, 37 of which were classified as a PRL on a 6-month 7T MRI T₂*-Phase image. Average lesion T₁ relaxation time and T₁ lesion volume quantified from the initial 7T MRI were significantly greater in new lesions that evolved into PRLs (Figure 2). In a subset of the above patients whose clinical 3T MRI also had standardized T₁ and FLAIR sequences, the lesions that evolved into PRLs had significantly different normalized T₁ and FLAIR signal intensity and larger lesion volume (Figure 3). Random forest regression models fit with the above lesion characteristics performed with an AUC of 0.86 using the initial 7T lesion characteristics and the 7T lesion characteristics performed better than the 3T lesion characteristics (Figure 4).

Emerging MRI data, including from our study, supports the idea that CELs that develop into PRLs differ from those that do not early in new lesion evolution. CELs that develop into PRLs have particular enhancement patterns⁴ and qualitative features (ADC characteristics)¹⁰, differ in semiquantitative MRI techniques including greater nFLAIR and lower nT₁ (Figure 3) and nMTR¹¹, and are larger in volume measured from 3T MRI. Contrast-enhancement within a MS lesion represents breakdown of the blood brain barrier and is the earliest MRI marker of new lesion development. Early in new lesion evolution there is a complex interaction between peripheral immune cells and the intrinsic cells of the CNS. The ability to differentiate lesions that will develop into PRLs from those that do not at this early time could indicate a role for peripheral immune cells in PRL formation. Studies have reported on differences in peripheral immune cells profiles that differentiate MS from healthy controls¹⁵, whether similar differences in immune cell profiles could determine if a new lesion will or will not develop into a PRL is not known and warrants further study. The lack of effect of steroid treatment administered between the 3T and initial 7T MRI and the increased odds of PRL formation in patients on DMT, supports the notion that treatments targeting PRL formation need to be administered very early in lesion development or target different pathways from those the inhibit lesion formation.

While the presence of contrast within a MS lesion is established as an early MRI marker of new lesion development, the actual age of any lesion is impossible to ascertain as contrast-enhancement lasts on average approximately 4 weeks with significant variability between lesions. Imaging characteristics extracted from MRIs with CELs could reflect a lesion that began developing a few days to weeks prior to the incident clinical MRI. This heterogeneity in lesion age may explain the improvement in classifying PRLs on the initial 7T MRI (Figure 4). As more time elapses from lesion development and acute inflammation resolves, the resulting signal change seen on the initial 7T MRI may be more reflective of the differences between PRLs and nonPRLs.

Our study provides new information of the early evolution of CELs that develop into PRLs in the form of T₁ relaxation times (i.e. T₁ mapping). The T₁ map calculated on the 7T MP2RAGE sequence has been shown to correlate with myelin content measured on histopathology¹⁶. Our initial 7T MRI was on average 3 weeks after CEL identification, and while post-contrast images were not obtained at 7T, by 3 weeks many lesions would have expected to have stopped enhancing with contrast. As such, the signal intensity within lesions on the T₁ map at the initial MRI time point likely reflects tissue loss, more so than acute inflammation. The higher T₁ within lesions that develop into PRLs indicates greater early tissue damage compared to lesions that do not develop into PRLs. This relatively greater T₁ value persisted at 6 months, in keeping with prior studies. The processes that drive these early differences within lesions remain unclear, but these MRI data suggest that need to study early mechanisms in lesion development to identify the mechanisms involved in PRL development.

Early identification of which lesions will evolve into PRLs could be of value to clinical trials targeting these lesions. Given that PRL incidence and prevalence associate with greater tissue focal tissue injury and earlier and more severe clinical disability, stopping the formation of these lesions could have important long-term clinical impacts. While current clinical trials in MS primarily aim to abolish new lesion formation and, thus, would limit future PRL development, current DMTs may increase the risk for PRL formation. Our study suggests DMT effects PRL development as our data showed patients that formed new lesions while on DMT had an increased odds of developing a PRL than those that on DMT. Our sample size is too small to determine if any specific DMT influences PRL development, but most participants were on low to moderate efficacy DMTs which could hypothetically inhibit the formation of MS lesions that are a less severe phenotype (i.e. nonPRL). This finding also suggests that any treatment targeting whether a new lesion evolves into a PRL will be different from targeting new lesion development.

The improvement in PRL prediction at 7T over 3T could reflect technical factors such as greater spatial resolution or better biological specificity of quantitative mapping compared to semi-quantitative measures derived from clinical sequences. Alternatively, the difference in prediction may reflect the difference in timing of when lesions were imaged (~ 3-week delay between clinical 3T and research 7T MRI). Lesions imaged on the earlier MRIs have more heterogenous signal within the lesion as there is likely more acute inflammation making prediction more difficult. It will be essential to perform same day 3T and 7T at early time points to address the question of whether 7T is necessary, and to identify the best timing, to predict new PRL development.

Machine learning is a powerful tool that can help identify which variables are important in identifying an outcome, including new lesion outcome (PRL or non-PRL). In our study, we used a leave-one-out cross-validation paradigm and had a larger, relative to prior published studies, number of CELs that developed into PRLs, both of which help limit model over-fitting. While our study was underpowered to assess if an acute treatment intervention (i.e. steroid administration) or chronic MS treatment (i.e. DMT use) could affect the PRL or non-PRL outcome, these treatments were not nearly as important as lesion volume and signal intensity in our random forest classifier analysis. The factors and processes, and their timing, which may lead to an acute MS lesion evolving into a PRL remain unclear and warrant further study.

Summary for Social Media If Published.

1. If you and/or a co-author has a Twitter handle that you would like to be tagged, please enter it here. (format: @AUTHORSHANDLE).

   None

2. What is the current knowledge on the topic?

   Paramagnetic rim lesions (PRLs) develop from a subset of acute MS lesions and associate with more severe clinical phenotypes in MS. Identifying which new lesions develop into PRLs can aid clinical trials targeting these clinically relevant lesions.

3. What question did this study address?

   Do new lesions that develop into PRLs differ from new lesions that do not and when do these differences occur?

4. What does this study add to our knowledge?

   This study shows that new lesions that develop into PRLs are different from new lesions that do not on early clinical MRI and on early quantitative 7T MRI. The increased T₁ measured on early 7T MRI suggests greater tissue damage early in lesion evolution in new lesions that develop into PRLs

5. How might this potentially impact on the practice of neurology?

   The data in this study suggests that clinical trials and therapeutics that are targeting newly developing PRLs may need to focus on very early processes in new lesion development and evolution in order to have an effect.