Deep cervical lymph node volume decreases following B-cell depletion therapy

Abstract

The deep cervical lymph nodes (dCLNs) are sites of immune presentation and B-cell maturation from the brain, and potentially involved in mechanisms of neuroinflammation. We hypothesized a reduction in dCLN volume following B-cell depletion therapy. In a retrospective cohort, we segmented bilateral dCLN from T2-FLAIR MRI at “prebaseline,” “baseline,” and “post-B-cell depletion” timepoints. Using a multivariable mixed-effect regression model, we find no changes in dCLN volumes between prebaseline and baseline timepoints (p > 0.05), but a significant decrease of 158 mm³ following ocrelizumab infusion (t = −3.3, p = 0.005). Baseline use of a disease-modifying therapy was also significantly associated with a smaller dCLN volume and attenuated the effects of B-cell depletion. These results are congruent with therapeutic mechanisms, although other alternative explanations for reductions in dCLN volumes cannot be ruled out based on this data. Deep CLN represent potential imaging biomarkers of pharmacological or clinical utility and warrant further investigation.

Introduction

Multiple sclerosis (MS) is an immune-mediated, inflammatory, and degenerative disease of the central nervous system (CNS). Although the exact etiopathogenesis of MS is not known, peripheral B-cell depleting agents are effective at reducing the inflammatory component of the disease. ¹ The limited penetration of these depleting antibodies into the CNS suggests a strong role for the peripheral immune system in propagating inflammatory disease mechanisms.

In this context, the deep cervical lymph nodes (dCLN) of the neck—which are situated along with the drainage pathways of the brain and meninges—are of increasing interest as an area of CNS-derived immune presentation and B-cell maturation for the brain. ² Cervical lymph nodes are readily visible on T2-FLAIR MRI sequences of the brain, which are routinely obtained in MS patients, and amenable to volumetric segmentation using threshold-based approaches. The few prior studies examining the size of CLN in MS, however, show discordant clinical associations.^3,4 Here, we hypothesized that treatment with B-cell depleting therapy (BCDT) would reduce the volumes of CLN; this could occur as a direct and/or downstream effect of eliminating (circulating) antigen-presenting cell populations in the peripheral immune compartment. ⁵

Methods

Study design and participants

This was a retrospective analysis of prospectively collected patient data as part of the OPTIMUM study at the University of Massachusetts. All patients provided written informed consent (IRB-approved Protocol # H00016906). We included patients aged 18–75 meeting 2017 McDonald criteria for MS ⁶ who had initiated BCDT within six months after baseline (herein referred to as “T0”) 3T MRI brain scan and also received a follow-up brain scan (“T1”) on the same hardware and software protocol, at a minimum of three months after initiating BCDT. Only MRI brain scans were assessed, as our protocol typically includes a field of view extending inferiorly to the level of the ∼C3 vertebrae. The time period assessed was from January 2016 to August 2022. Of an initial 403 OPTIMUM database subjects, 33 subjects met inclusion and exclusion criteria for preliminary analysis: 130 were excluded for not having longitudinal (≥2 MRI scans) harmonized MRI protocol; 194 excluded for no exposure to BCDT in the observation period; 46 were excluded for not having scans with appropriate timing in relation to BCDT; 7 were additionally excluded for having poor visualization of >1 dCLN (primarily due to extension beyond the MRI field-of-view). Ten of the remaining 26 subjects also had availability of a prebaseline MRI (“T-1”) at least a year before baseline, which allowed these patients to serve as their own control.

Scan image acquisition

Prebaseline (T-1), Baseline (T0), and Follow-up (T1) scans were acquired on a 3T MRI scanner (Signa Pioneer, General Electric Healthcare, Wisconsin, USA) using identical parameters for the sequence of interest (3D sagittal T2 FLAIR). Parameters for this sequence were: TR/TE/TI/flip angle/echo train length = 5400/maximum ∼133/1623/90°/140, frequency/phase = 256/224, slice thickness = 1.6 mm with 50% overlap, FOV = 240 mm, acquired voxel size 0.94 × 1.07 × 1.6 mm, and reconstructed to 0.49 × 0.49 × 0.8 mm.

Image analysis and cervical lymph node determination

The prebaseline, baseline, and follow-up scans were downloaded in native DICOM format from the hospital picture and communications system and converted to NIFTI files using the “dcm2niix” tool. Preprocessing step included bias field removal using SPM (default parameters), and coregistration/reslicing to the patient's T1-weighted scan, upscaled to 0.8 mm isotropic resolution. Scans were defaced using the “pydeface” toolbox. The FLAIR scans were then visualized and segmented in ITK-SNAP. ⁷ A single rater (NL) was trained by an expert neuroradiologist (SD) to identify dCLN based on their anatomical location, morphology, and T2-FLAIR hyperintense signal characteristics. A second rater (CH)—blinded to the initial rater's assessments—segmented the bilateral dCLN using identical methodology to determine inter-rater Cohen's kappa and dice scores based on the overlap of each CLN. For each side of the neck, the single largest CLN in the jugulodigastric region was identified in the coronal plane. Images were then “presegmented:” this involves the user selecting an intensity threshold in an interactive interface (“speed image”) to maximize contrast between regions of high intensity (the cervical lymph nodes) and surrounding, low-intensity soft tissue. Following presegmentation, seed points are placed within the CLNs after which a semiautomated region-growing algorithm is run, expanding from the seed points to the threshold-defined edges. This tool is available natively in ITK-SNAP. ⁷ A visual overview of this process is provided in Supplementary Figure 1. For each longitudinal timepoint, the same dCLNs were identified based on coregistered images compared to T0, then segmented using identical methodology (see Figure 1). Additional volumetric analysis of submandibular lymph nodes was attempted but abandoned due to limitations from defacing and generally poor visibility.

Figure 1.

Effects of ocrelizumab on deep cervical lymph node volume. Longitudinally aligned T2-FLAIR MRI scans reformatted to the coronal plane. The largest jugulodigastric lymph node on each side of the neck (here showing the left) was segmented using a threshold-based approach and depicted in the zoomed-in portion. The reduction in cervical lymph node (CLN) size is apparent at timepoint T1.

Clinical and demographic data

Patient neurological disability ratings (as rated by the Expanded Disability Status Scale “EDSS”), disease-modifying therapy (DMT) use, weight, and clinical phenotypes were extracted by manual chart review from the electronic medical record and chosen based on the closest proximity to the time of the baseline MRI. All neurological disability scores and MS diagnoses/phenotypes were rated by expert clinical neurologists with dedicated subspecialty training in neuroimmunology. Any steroid use (oral or intravenous) within three months prior was recorded. Any clinical flares or infections within three months prior to, or up to one month after, the MRI were additionally recorded.

Statistical analysis

Using R-studio (www.r-project.org) and the “tidyverse” package, a mixed-effects linear regression was used to determine dCLN volume (outcome) with clinical, MRI, and timepoint data. Uncommon missing data (allowed only if ≤1 dCLN was partially outside the field-of-view) were handled as part of the mixed-effect regression. Variables were visualized using histograms, and their normality was determined by visual inspection as well as measures of skewness, kurtosis, and Shapiro–Wilks testing. Variables that were not normally distributed were analyzed twice, both natively—to facilitate interpretability—and using a log-transformation as a sensitivity analysis.

Results

The overall cohort consisted of 26 patients with MS (mean age at “T0”: 42.9 ± 13.4 years; 96% relapsing MS, 4% primary progressive MS; disease duration 6.2 ± 8.6 years; 62% female) patients; 54% were not on any treatment (see Table 1 for a full breakdown). No patients had any steroid exposure or infections in the month prior to baseline (“T0”) MRI. Six patients had a clinical flare within a three-month window prior to baseline MRI; only one had a flare within the month prior to T0; neither modeling flares at three months nor at one month was significantly associated with dCLN volumes. Univariable analysis of baseline variables and their correlations is presented as a correlation matrix in Supplementary Figure 2. Notably, the unadjusted mean dCLN was decreased by 42% in association with use of (any) baseline DMT (mean untreated = 759 mm³ vs treated = 441 mm³; t = 2.6, p = 0.01). Older age correlated with smaller dCLN volumes (r = −0.37) with a trend toward statistical significance (p = 0.06). None of EDSS, disease duration, weight, or sex was significantly associated with dCLN volumes. A multivariable regression on baseline data that included age somewhat attenuated the association between DMT and average dCLN volumes, but this association remained significant (p < 0.05).

Table 1.

Full cohort characteristics.

	Baseline (T0)	Follow-up (T1)
N	26	26
Age (years)	42.9 ± 13.4	43.8 ± 13.5
EDSS	1.8 (1.5, 3.0)	1.8 (1.5, 3.0)
Disease duration (years)	6.2 ± 8.6	7.1 ± 8.6
Disease-modifying therapy
None	14 (54%)	0 (0%)
Dimethyl fumarate	4 (15%)	0 (0%)
Teriflunomide	3 (12%)	0 (0%)
Glatiramer acetate	3 (12%)	0 (0%)
Interferon beta-1a	1 (3.8%)	0 (0%)
Fingolimod	1 (3.8%)	0 (0%)
Ocrelizumab	0 (0%)	24 (92%)
Rituximab	0 (0%)	2 (7.7%)
Sex
Female	16 (62%)	16 (62%)
Male	10 (38%)	10 (38%)
MS phenotype
Primary progressive	1 (3.8%)	1 (3.8%)
Relapsing-remitting	25 (96%)	25 (96%)
Left cervical lymph node volume (mm3)	549 ± 331	348 ± 192
Missing/truncated, N	4	3
Right cervical lymph node volume (mm3)	644 ± 443	458 ± 338
Missing/truncated, N	3	3
Mean cervical lymph node volume (mm3)	613 ± 348	415 ± 250

Data are presented as mean ± SD; median (IQR); n (%).

EDSS: Expanded Disability Status Scale; MS: multiple sclerosis.

We then assessed the baseline cohort (N = 26) pre/post data, following exposure to B-cell depletion. The median and mode number of ocrelizumab infusions (600 mg, combining loading doses) was 2 (IQR: 2,2); one person received four infusions, and four others received one interval infusion. No person had an infection or flare proximate to the follow-up time point. Ten persons in this cohort had received 100 mg intravenous methylprednisolone as part of their B-cell depletion infusion in the prior three months (from timepoint “T1”); four had received it in the one month prior. Unadjusted pre-post BCD was associated with a 198 mm³ reduction in average dCLN volume; this was significant (t = 2.35, p = 0.02). We then built a multivariable mixed-effect regression accounting for fixed effects of baseline age (not significant: ns), sex (ns), weight (ns), baseline DMT (any; dichotomous: beta = −192, p = 0.09), and laterality (right dCLN beta = 86, t = 2.2, p = 0.03), number of BCD treatments (beta = −117, p = 0.23), and BCD administration (beta = −186, p < 0.001), with patient identity as a random effect. The full results of this analysis are presented in Supplementary Table 1. Inclusion of the interaction term [(baseline DMT) × (BCD exposure)] showed a beta = 145 and a trend toward significance (p = 0.05). Inclusion of steroid exposure (modeled as either 3 or 1 month beforehand) was not significant and had little effect on the betas of other variables.

Because this cohort had no control comparison, we selected the patients from among this group who had a “prebaseline” MRI and could act as their own controls (N = 10; subcohort data are summarized in Supplementary Table 2). We observed no significant changes in dCLN volumes between their prebaseline and baseline timepoints (p > 0.05), but a significant decrease following ocrelizumab infusion (baseline vs postinfusion: beta = −158, t = −3.3, p = 0.005; see Figure 2). These results were similarly analyzed with a mixed-effect linear regression model adjusted for fixed effects of age (not significant: ns), sex (ns), and lymph node laterality (ns); individually introducing the following, additional variables to the model yielded nonsignificant associations: EDSS (ns), disease duration (ns), patient weight (ns), any prior steroid exposure (either at 1 or 3 months: ns), and any prior clinical flares (either at 1 or 3 months: ns).

Figure 2.

“Spaghetti” plot showing cervical lymph node volumes in the subcohort with three timepoint assessments, in relation to B-cell depletion therapy (BCDT). Significance values are derived from a multivariable mixed-effect model adjusting for age, sex, and lymph node laterality with a patient-specific random effect.

Of note, cervical lymph node volumes from the overall cohort were right-skewed (kurtosis = 3.01, skewness = 1.47, Shapiro–Wilks p < 0.01). Application of a log-transform to dCLN improved the normality of the distribution (yielding a kurtosis = −0.15, skewness = −0.51, Shapiro–Wilks p = 0.02). In case this skew was affecting study results, all of the above analyses were repeated using identically structured mixed-effects models but using a log-transformed dCLN outcome. The only meaningful change in this analysis was that dCLN laterality was no longer significant. Otherwise, all results showed highly similar significance patterns, associations, and magnitude as the nontransformed analysis. Since data interpretation would not substantially change, the non-log-transformed results are presented here for their superior interpretability.

Last, inter-rater assessments were performed in 10 random patients for the determination of the “largest” dCLN on each side; this resulted in an agreement of 19/20 nodes and a Cohen's kappa = 0.90 (assuming an “expected” rate of 50%). Independent segmentation of the 19 agreed nodes yielded a Dice Similarity Coefficient of 0.83 ± 0.09.

Discussion

Here, we report a strong temporal association between the initiation of B-cell depletion and an associated shrinkage of dCLN in relapsing MS. In this study design, a subcohort of patients act as their own control with a “prebaseline” MRI, an analysis that showed stable CLN volumes leading up to B-cell depletion. This study was inspired in part by prior investigators who successfully measured dCLN size (based on diameter) using T2-FLAIR brain scans in MS. ³ We adopted their overall approach, except that instead of measuring diameters, we segmented the CLN using a semiautomated, 3D, threshold-based approach that more accurately reflects lymph node size. Inter-rater agreement using this approach was high. These findings are conceptually concordant with the idea that BCDT interrupts the dynamic B-cell trafficking axis between the brain and CLN,^2,5 but given the small sample size and lack of a direct comparison group, we emphasize that other mechanisms—including natural history—could also explain these observations. The significant association between dCLN volumes and DMT use at baseline also suggests that DMTs in general may be associated with reductions in dCLN volumes and may attenuate the effect of BCD based on the interaction term nearing significance (p = 0.05). Although not directly comparable, in the prior work of Tuulasvaara et al., they showed no difference in cross-sectional CLN volumes between stable patients on high- versus low-efficacy therapy. ³ Notably, their high-efficacy cohort combined sequestering agents (S1P and leukocyte migration inhibitors) with those receiving BCDT ³ ; this, and/or the cross-sectional design, could explain observed differences. Last, we observe a trend between older age and smaller dCLN volume (in our larger preliminary cohort), similar to observations from others.^3,8

A major limitation of this study is the lack of a comparative cohort to rule out other explanations for these results. Alternative interpretations of the data include a generalized dCLN atrophy phenomenon with ongoing treatment, a regression to the mean, or confounded by indication/treatment bias. The latter effect is reduced by having a prebaseline comparison; also, relapses when statistically modeled did not affect CLN volume; this finding is similar to results in one prior study. ³ B-cell depletion treatments are inherently confounded by the administration of steroids as a premedication (typically ∼100 mg intravenous methylprednisolone). However, we did not find any significant effect of steroids when these were included in the statistical model; moreover, steroids have a short half-life and the pharmacokinetic effects on leukocyte populations are likely limited to weeks ⁹ ; here, only one person with MS had exposure to steroids within the month prior to their MRI. Regardless, steroids remain a potential confound, and these results require external validation in independent cohorts. Another practical limitation is the field-of-view of the MRI; 33% of our patients would be excluded for their largest dCLNs being at least partially beyond the field-of-view for clinically acquired FLAIR brain scans; somewhat less than Tuulasvaara et al. who report 36–54% loss. ³ Since we were using a mixed-effect model robust to missing data, we allowed for the occasional, unilateral missing dCLN value so as to maximize sample size. A field of view that cuts off the dCLN is a limitation for clinical translation, although segmentation of dCLN more superior to the jugulodigastric area, could be assessed to overcome this challenge. Last, our dataset is likely too small to demonstrate any clinical disability correlations with CLN volume: prior work from Tuulasvaara et al. has found smaller dCLN in MS versus healthy controls, and that smaller dCLN portends worse MS disease progression. ³ In contrast, a cross-sectional ultrasound study of sequestering agents (fingolimod and natalizumab) showed CLN volumes to be overall larger in MS compared to healthy controls, but similarly did not find differences compared to drug-naive patients. ⁴ The natural history and clinical/pharmacological associations with dCLN represent fruitful ongoing directions for research.

Conclusion

These data preliminarily support the idea that dCLN volumes are lower in patients receiving DMT and further decrease following BCDT; although limited, these results warrant further investigation of the CLNs as novel and potentially useful imaging biomarkers of pharmacological response or disease monitoring in MS.

Supplemental Material

sj-docx-1-mso-10.1177_20552173251371743 - Supplemental material for Deep cervical lymph node volume decreases following B-cell depletion therapy

Supplemental material, sj-docx-1-mso-10.1177_20552173251371743 for Deep cervical lymph node volume decreases following B-cell depletion therapy by Nikhil Lele, Sathish K Dundamadappa and Christopher C Hemond in Multiple Sclerosis Journal – Experimental, Translational and Clinical

Supplemental Material

sj-doc-2-mso-10.1177_20552173251371743 - Supplemental material for Deep cervical lymph node volume decreases following B-cell depletion therapy

Supplemental material, sj-doc-2-mso-10.1177_20552173251371743 for Deep cervical lymph node volume decreases following B-cell depletion therapy by Nikhil Lele, Sathish K Dundamadappa and Christopher C Hemond in Multiple Sclerosis Journal – Experimental, Translational and Clinical