Longitudinal stability of progression‐related microglial activity during teriflunomide treatment in patients with multiple sclerosis

The aim was to study brain innate immune cell activation in teriflunomide‐treated patients with relapsing–remitting multiple sclerosis.


INTRODUC TI ON
Multiple sclerosis (MS) is a leading cause of disability in younger adults in the Western world [1,2], and microglial activation is a hallmark of progression-related MS pathology. Chronic slowly expanding lesions are characterized by a perilesional ring of activated microglia, whilst inactive plaques contain low numbers of innate immune cells or other inflammatory cells [3,4]. Quantitative susceptibility mapping (QSM) is an experimental magnetic resonance imaging (MRI) method able to capture susceptibility changes caused by, for example, iron deposition within microglia [5], and iron containing CD68+ microglia and macrophages have been identified around QSM-positive MS lesions with immunohistochemistry [6].
Used in over 500 patients with MS, positron emission tomography (PET) imaging utilizing the 18-kDa translocator protein (TSPO) binding radioligand [ 11 C]PK11195 is an established in vivo method to monitor smoldering inflammation in the MS brain [7,8]. Whilst a subset of astrocytes and endothelial cells may contribute to the TSPO signal, microglia and macrophages constitute the main source of TSPO within the normal-appearing white matter (NAWM) and at the edge of lesions [9][10][11].
In addition to correlating with disability, as measured with the Expanded Disability Status Scale (EDSS) [12], [ 11 C]PK11195 uptake predicts the development of clinically definite MS [13] and disease progression independent of relapse [14]. Interventions with diseasemodifying treatments (DMTs) reduce TSPO binding at chronic MS lesions [15,16] and also in the NAWM [15]. By linking disability progression to microglial activation in the NAWM [17], TSPO-PET has been instrumental in expanding the investigational scope of the pathophysiological process beyond conventional lesions seen with MRI.
Teriflunomide is a treatment for relapsing-remitting MS (RRMS) first introduced almost a decade ago. Of interest is its efficacy in preventing EDSS-measured disability progression in two phase III trials in patients with relatively active MS [18,19]. As the more recent real-world use of teriflunomide has concentrated on milder MS, this non-interventional study enrolled patients at risk of disease progression based on relatively advanced age, which is one of the strongest known predictors of progression [20]. Patients with secondary progressive MS (SPMS) exhibited increased microglial activation compared with patients with RRMS [21]. Thus, it was of interest to assess TSPO binding in a real-world non-interventional setting of first-line DMT use amongst middle-aged patients at progression risk.

Study design
This was an open-label, non-interventional study performed at the and withdrew her consent after the baseline visit. Another subject discontinued teriflunomide due to poor perceived tolerability and was subsequently excluded between visits. Fourteen subjects remained for the entire study duration. One patient was later diagnosed with SPMS from before enrollment and was excluded from the analysis, and another subject was excluded due to failed radioligand synthesis resulting in inadequate injected activity at follow-up imaging. The temporary loss of follow-up resulted in an extended interval of >2 years between the PET scans for one subject, but these data were included in the analysis as an extended follow-up was not expected to produce significant bias. For comparison, 12 age-and gender-matched healthy individuals were imaged at baseline using TSPO-PET and MRI. Toolbox (LST) [22] in SPM, after which these preliminary masks were checked and edited to correspond to T1-hypointense chronic MS lesions according to visual inspection. A chronic lesion was defined as measuring >3 mm 3 voxels in at least one dimension with the most hypointense voxels below the intensity of normal gray matter.

Magnetic resonance imaging and image processing
The perilesional mask used to quantify tracer uptake surrounding chronic lesions was created by dilating the lesion mask by 3 mm and then subtracting the core image from the 3-mm image. for PET assessments.
To estimate rater-independent changes in T2 lesion volume, nicMSlesions [23] was first used to segment the lesions from baseline FLAIR images; then the longitudinal pipeline of nicMS was employed using the baseline masks as input. The threshold was set to ≥9 voxels to exclude small unspecific T2 hyperintensities of <3 mm in any dimension. Whole brain volume normalized for head size was estimated with SIENAX [24], and thalamic volumes were segmented with FSL FIRST [25].

Iron rim detection using QSM
The QSM images were processed using the Morphology Enabled Dipole Inversion toolbox with automatic uniform cerebrospinal fluid zero reference (MEDI+0) [26] from the multi-echo gradient echo sequence data. The reconstructed QSM images were then coregistered with the T1 images using SPM12. Lesions manifesting a positive susceptibility value indicating iron were preselected. If the QSM signal was morphologically consistent with a lesion-associated ring, this lesion was determined as a QSM iron rim lesion by visual inspection by two experienced raters. Additionally, ITK-SNAP (http:// www.itksn ap.org/pmwik i/pmwiki.php) was used to quantify the signal intensity in the perilesional area and to ascertain the relative hypointensity of the respective core.

[ 11 C]PK11195 radioligand production and PET
To produce the radioisotope 11 C, irradiations were performed with a TR-19 (ACSI, Richmond, Canada) cyclotron by proton bombardment of 14 N utilizing the 14 N(p, α) 11 C nuclear reaction. A target gas mixture of 0.2% oxygen in natural nitrogen produced radioactive CO 2 . Detailed synthesis steps to produce the radiochemical com-

Positron emission tomography data processing and analysis
Image reconstruction was performed with the OP-OSEM algorithm [29] using 17 timeframes. PET image post-processing followed a previously described procedure [15]. Briefly, the dynamic images were smoothed, realigned and co-registered using SPM8. Images were then resliced to match the 1-mm voxel size of the MRI images.
Specific binding of [ 11 C]PK11195 was quantified using distribution volume ratios (DVRs). As no reliable anatomical region devoid of activated microglia (i.e., specific PK11195 binding) exists, time− activity curves representing a region without specific binding were acquired with the MATLAB (MathWorks Inc., Natick, MA, USA) software Super-PK using a supervised cluster algorithm [30] optimized with four predefined kinetic tissue classes [31]. The Logan variant reference tissue model [32] with a 20-to 60-min time interval was applied to the regional time−activity curves.
To determine the proportions of individual active voxels, the mean DVR (± SD) of all voxels in the NAWM of healthy control subjects was first calculated. A 95% confidence threshold was determined with the formula mean + 1.96 × SD, which was used for dichotomous classification of each voxel of all subjects, that is, values above the threshold signified an active voxel. Based on our previous data from healthy controls [33], this threshold was set to 1.56.
Clusters below the three connected voxels were excluded to prevent the inclusion of random peak values. Parametric

Analysis of chronic lesion subtypes according to microglial activation
Lesion phenotyping was performed according to Nylund et al. [33]. Briefly, individual lesions in the T1 lesion masks were classified into "inactive", "overall active" and "rim active" based on the presence and distribution of active voxels in the core versus rim. Chronic T1 lesions with a minimum volume of 27 mm 3 were considered.

Statistical analysis
The statistical analysis was performed using R version 4.

Study subjects
The teriflunomide-treated cohort (n = 12) consisted of three males and nine females of Caucasian descent with an average (± SD) age and disease duration of 46.1 ± 6.2 and 10.8 ± 8.1 years, respectively. The mean (± SD) duration of teriflunomide treatment was 10 ± 3.5 months before baseline imaging. The demographic baseline characteristics are listed in Table 1.

Clinical and conventional MRI parameters at baseline and follow-up amongst patients with MS
The median and interquartile range (IQR) EDSS score was 2 (1.   (Table 2).

Translocator protein PET binding amongst patients versus controls
Comparing TSPO binding between patients with MS and healthy controls, no baseline difference in TSPO availability (measured as DVR) was observed in any examined region (whole brain, NAWM, and thalamus) (Figure 1a-

Longitudinal stability of TSPO availability during teriflunomide treatment
The low level of baseline microglial activation remained stable during follow-up: whilst there was a slight trend downwards, there were no statistically significant DVR changes in the NAWM, thalamus or perilesional area between baseline and follow-up (Figure 2a-d). In the perilesional area, the baseline and follow-up DVRs were 1.15 ± 0.08 and 1.11 ± 0.08 (p = 0.13). In line with the DVR PET results, the proportion of active voxels remained stable during teriflunomide treatment in all examined brain areas (Figure 2e-h).

DISCUSS ION
Translocator protein availability remained largely unaltered during the 1-year follow-up in this cohort of teriflunomide-treated patients

TA B L E 2 MRI variables
with RRMS. Furthermore, only a modest difference in brain microglial activation was observed between the RRMS population and an age-matched healthy control population. However, this difference manifested as a statistically significant difference in the proportion of active voxels in the MS versus control brain. No difference in the DVR values in the whole brain, NAWM or thalamus regions of interest was observed between patients and controls. This finding implies that TSPO-PET-based detection of clusters of active voxels in the MS brain may be a more sensitive method for capturing innate immune cell activation compared with quantifying radioligand binding using DVR in various brain regions of interest.
In other MS cohorts, significantly higher DVR values in MS brain NAWM compared to control white matter [13,15,16] have been previously repeatedly demonstrated by ourselves and others. It has During the last decade, MS treatment has shifted towards early escalation and early intense treatment approaches [35]. This shift has happened at the expense of recent phase III trials [36,37], where relatively less active cohorts of patients were enrolled compared with past trials [18,19]. Although the more recent trials successfully demonstrated relative efficacy with limited effect size in RRMS in terms of the annualized relapse rate, prerequisites for successful trials on patients with SPMS should include concentrated efforts to better characterize the population at a high risk of progression.
Although conversion to secondary progression occurs on average at 45 years of age [38], MS is a heterogeneous disease, and, has been previously characterized, and it has been shown that rim active lesion load correlates with disability [33].
It is difficult to estimate exactly what impact teriflunomide treatment had on the longitudinally stable TSPO-PET findings in this cohort. It has been previously demonstrated in an untreated MS cohort with more accrued disability (EDSS = 6) that TSPO binding increased over 1-year follow-up both in the NAWM and perilesional areas [15].
Teriflunomide limits the activation and proliferation of lymphocytes by inhibiting pyrimidine biosynthesis in activated lymphocytes via the mitochondrial enzyme dihydroorotate dehydrogenase [39].
Further to these peripheral effects, effects on resident CNS cells (oligodendrocytes and microglia) have been suggested based on in vitro and experimental animal work [40][41][42]. In rodents, teriflunomide crosses the blood-brain barrier with 1%-2% of serum concentrations (in the range 2.5-4.1 μM) reaching the CNS [43]. Thus, in theory, teriflunomide may have contributed to the unaltered microglial status via either peripheral or central effects.
From a health economic perspective, the correct timing of deescalation or complete cessation of treatment is another unresolved question. Whilst a prior stable disease course does not appear to protect patients from disability after DMT discontinuation [44], discontinuation after a period of inactivity according to conventional clinical parameters was not associated with time to clinical relapse, MRI activity or EDSS increase [45]. The timing of discontinuation is not feasible with PET but based on the presented results and previous evidence PET may be used to better characterize patients at a low risk of progression.
Other potential and more readily available prognostic biomarkers in MS include cortical MRI lesions [46], QSM [6], thalamic volume [47] and neurofilament light chain [48]. Future next steps include cross-sectional comparisons of these biomarkers with PET. Furthermore, a direct effect on brain microglia as a contributing mechanism of action cannot be excluded [40].

ACK N O WLE D G E M ENTS
The authors thank Mikko Koivumäki and Taruliina Parkkali for their valuable support in the execution of this study.

CO N FLI C T O F I NTER E S T S TATEM ENT
Dr Airas has obtained institutional research support from Novartis, Sanofi-Genzyme and Merck and compensation for lectures and advising from Novartis, Sanofi-Genzyme, Merck, Roche and Janssen.
The other authors have nothing to disclose.

DATA AVA I L A B I L I T Y S TAT E M E N T
Any anonymized data used in the preparation of this article will be made available on the request of a qualified investigator.