Evaluating the Clinical Utility of TSPO-PET in MS and the Influence of Methodological Heterogeneity

Abstract

Background and Objectives

PET imaging targeting the 18-kDa translocator protein (TSPO) is a widely used method to detect neuroinflammation in vivo and shows promise for detecting innate inflammatory activity in multiple sclerosis (MS). However, TSPO-PET findings in MS vary substantially across studies. We performed a systematic review and meta-analysis to quantify TSPO-PET signal differences between people with MS (PwMS) and healthy controls (HCs) and to evaluate the impact of methodological factors on study outcomes.

Methods

A systematic search of Scopus, MEDLINE, Web of Science, and Embase databases was conducted through September 10, 2025. Original TSPO-PET studies in PwMS were included. Standardized mean differences (SMDs) in TSPO-PET signal between PwMS and HC were pooled using random-effects models with cluster robust variances. Heterogeneity was assessed using I² statistics. Meta-regression and subgroup analyses investigated methodological sources of heterogeneity. Patient-level data were analyzed for associations with clinical and demographic variables. Risk of bias and certainty of evidence were assessed using quality assessment of diagnostic accuracy studies 2 tool and GRADE framework, respectively.

Results

Forty-nine studies met inclusion criteria. Overall meta-analysis revealed significantly increased TSPO-PET signal in PwMS vs HC (SMD = 0.44, 95% CI: 0.27–0.61, p = 0.0002, I²: 97.00%). TSPO-PET signal was also significantly higher in progressive MS than in relapsing-remitting MS, indicating discrimination across MS clinical subtypes. Meta-regression attributed study heterogeneity to tracer (11.47%), PET quantification method (15.37%), and region of interest (19.71%) differences. Significant PwMS vs HC differences were most consistently observed with select tracers ([¹¹C]PBR28, [¹⁸F]DPA-714, [¹¹C]ER176, and [¹¹C]PK11195) and with distribution volume ratio (DVR), nondisplaceable binding potential (BP_ND), and standardized uptake value ratio metrics. Significant effects were found in perilesional, thalamic, cortical, hippocampal, normal-appearing white matter, and T₂-weighted white matter lesion regions. Patient-level analyses, while limited in power, showed significant associations between TSPO-PET signal (DVR) and disability measures (expanded disability status scale).

Discussion

TSPO-PET effectively distinguishes PwMS from HC and differentiates MS subtypes. However, radiotracer, quantification method, image analysis method, reference tissue, and region of interest significantly impact TSPO-PET outcomes and must be considered when designing and interpreting studies. Methodological standardization and improved reporting will be essential to enhance cross-study comparability and clinical interpretability.

Introduction

Multiple sclerosis (MS) is a chronic immune-mediated disease of the CNS characterized by focal demyelinating inflammatory lesions. Historically, MS research and treatment has focused on adaptive immune mechanisms; however, growing evidence highlights a critical role for innate immunity in MS pathophysiology.^1-4 Activated microglia and macrophages are the predominant immune cells in acute and chronic MS lesions⁵ and are closely associated with lesion formation and demyelination.⁶ Notably, increased microglial and macrophage activation has been observed in normal-appearing white matter (NAWM) before lesion formation, as well as in ongoing compartmentalized inflammation manifesting as chronic active lesions, supporting their involvement in both early lesion development and ongoing disease progression.^7,8 Given the central role of myeloid cells in MS pathology, noninvasive imaging is needed to characterize their in vivo distribution and activity in and around lesions.

PET imaging targeting the 18-kDa translocator protein (TSPO) is a widely used method to detect neuroinflammation in vivo, owing to its increased expression across neuroinflammatory diseases.⁹ In the brain, TSPO signal predominantly reflects microglia/macrophages, although expression is also observed in activated astrocytes and endothelial cells.^10,11 Recent human brain tissue studies show that although TSPO expression does not necessarily increase per cell during neuroinflammation, overall TSPO expression closely reflects microglial/macrophage density.^10,12 Nutma et al.¹² found that over 95% of TSPO⁺ cells in the NAWM, active lesions, and rims of chronic active lesions are microglia/macrophages. Notably, interpretation of TSPO-PET signal is complicated by the TSPO polymorphism (rs6971) that influences ligand binding affinity, resulting in low-affinity (LAB), mixed-affinity (MAB), and high-affinity binding (HAB) and contributing to interindividual variability in tracer signal,¹³ which most strongly affects second-generation tracers, although it is not exclusive to them.^14,15 Nonetheless, the relative cellular enrichment and specificity of TSPO supports the use of TSPO-PET for detecting focal innate inflammatory activity in MS. Thus, TSPO-PET can complement conventional imaging, providing insights into MS pathogenesis, progression, and treatment response.^16,17

Although numerous clinical studies have employed TSPO-PET to investigate neuroinflammation in MS, results and approaches vary considerably.^16,17 This variability likely reflects methodological heterogeneity, including differences in radiotracer selection, quantification strategies, outcome metrics, TSPO genotype handling, and patient characteristics. To address these inconsistencies and quantify the strength of the existing evidence, a systematic review and meta-analysis of TSPO-PET studies in MS was conducted. Specifically, the analysis evaluated the ability of TSPO-PET to discriminate MS from healthy controls (HC), how methodological factors influence imaging outcomes, and the association of TSPO-PET with disease severity.

Methods

Study Design and PICO Definition

This study followed Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines¹⁸ (PROSPERO ID: CRD42024510937). We evaluated whether TSPO-PET distinguishes people with MS (PwMS) from HC and differentiates MS subtypes, including relapsing-remitting MS (RRMS) and progressive MS (PMS, comprising primary progressive and secondary progressive MS) from each other and from HC. Patient-level associations between TSPO-PET signal and measures of disease severity and demographics were also examined.

Search Strategy

Keywords related to TSPO, TSPO-PET tracers, and MS were identified using Medical Subject Headings (MeSH) and Emtree databases. A search strategy was developed for MEDLINE (PubMed), Scopus, Web of Science, and Embase (eTable 1), covering studies published through September 10, 2025. Google scholar manual searches and reference screening identified preprint and nonindexed articles.

Literature Screening and Selection Criteria

Two reviewers independently screened titles, abstracts, and full texts against predefined criteria, with discrepancies resolved by a third reviewer. Eligible studies were human TSPO-PET studies in PwMS, excluding reviews, case reports, duplicate studies, conference abstracts, and studies lacking TSPO-PET outcomes. Studies without HC required patient-level TSPO-PET and disease severity outcomes for inclusion.

Data Extraction

Two independent reviewers extracted study-level and patient-level data including patient characteristics, participant demographics, TSPO polymorphism status, and imaging variables. Imaging variables included tracer selection, acquisition timing, correction methods (e.g., partial volume correction [PVC] and plasma correction), analysis methods, PET quantification metric, and region of interest (ROI) definitions.

Extracted PET metrics included nondisplaceable binding potential (BP_ND), distribution volume ratio (DVR), standardized uptake value (SUV), standardized uptake value ratio (SUVR), total volume of distribution (Vt), Vt divided by plasma free fraction (Vt/ƒp), and Glial activity load on PET Z score (GALPz). Data reported as uptake ratio and SUVR_max were pooled with SUVR data for analyses.^19-21

PET analysis methods were classified as: reference tissue kinetic, using reference region data instead of arterial input (simplified reference tissue model [SRTM] and Logan reference tissue model [Logan-Ref]; static/semi-quantitative, based on static single frame measures (e.g., those generating SUV or SUVR) with any “reference” region used only for post hoc normalization; arterial-input compartmental models, using metabolite-corrected plasma input and no reference tissue (2-tissue compartment models [2TCM]); and arterial-input graphical/multilinear models, classified as linearized models (i.e., Logan plot and multilinear analysis 1 [MA1]).

Brain ROI data were collected and anatomically equivalent ROIs grouped together (eTable 2). ROIs reported fewer than 3 times were classified as “other” and excluded from ROI-based meta-regression and subgroup analyses. All perilesional area definitions (i.e., 3 mm,^22-24 3-6 mm,²⁵ 6 mm,²⁶ and 2–6 mm lesion rims²⁷) were combined into a single “perilesional area” ROI. TSPO-PET signal in lesional, perilesional, and NAWM ROIs of PwMS was compared with WM signal in HC. Reference tissue application was characterized as model-embedded or ratio-based normalization for target ROIs. When arterial-input models reported ratios relative to another brain region, that region was classified as a normalization region rather than a kinetic reference tissue.

For group comparisons, sample size, mean TSPO-PET signal, and SD were extracted. Data were obtained from tables, text, supplementary materials, or digitally extracted from figures (using Plot Digitizer software). When only subgroup data were reported, pooled means and SDs were calculated. For longitudinal studies, only baseline data were included. When multiple radiotracers, analysis methods, quantification metrics, and/or ROIs were reported, each was recorded as a separate estimate.

Meta-Analyses

Analyses were performed in R (version 4.1.1)²⁸ using the metafor (version 4.4.0)²⁹ and clubSandwich packages.^30,31 Standardized mean differences (SMDs) were calculated to compare TSPO-PET signal between PwMS and HC. Aggregation was used to generate forest plots for visualization purposes only, with averaged estimates from the same study populations were aggregated, and random-effects model (REM) using restricted maximum likelihood was fitted to obtain pooled SMDs with 95% confidence intervals.

Many studies contributed multiple estimates from the same study participants (e.g., from different ROIs and quantification/analysis methods) and were statistically dependent, violating the independence assumption of standard meta-analysis. To account for this, cluster-robust variance (CRV) estimation grouped estimates from the same participant sample into a single cluster and adjusted standard errors for within-cluster dependence. Overlapping cohorts were assigned to the same cluster (e.g., studies from Herranz et al.³² and Barletta et al.³³), while studies with independent samples were treated as separate clusters.

To explore sources of heterogeneity, meta-regression and subgroup analyses were performed using CRV based on tracer, signal quantification metric, image analysis method, ROIs, reference tissue in algorithm, and reference tissue in ratio, with heterogeneity quantified using R² statistics. Heterogeneity was estimated using the restricted maximum-likelihood estimator,²⁹ Q test,³⁴ and I² statistic³⁵ for analyses with ≥3 estimates. Meta-analyses were repeated using a multivariate CRV-based model to account for all meta-regression variables. Publication bias was assessed by rank correlation tests of funnel plot asymmetry.³⁶

Subgroups with <3 independent estimates were excluded from formal meta-analysis and reported descriptively and heterogeneity statistics were calculated for subgroups with ≥3 estimates. Sensitivity analyses excluding small subgroups were performed to assess robustness.

Individual patient-level data were analyzed to evaluate associations between TSPO-PET signal and clinical disability measures (Expanded Disability Status Scale [EDSS], Symbol digit modalities test Z score [SDMTz], and MS severity score [MSSS]), and demographic variables (disease duration, age, and sex). These represent meta-analyses of patient-level data clustered by study cohort rather than true patient-level mixed models. Linear regression³⁷ with bias-reduced CRV and Satterthwaite degrees-of-freedom correction based on number of clusters was applied in all cases, including sex, to account for the small number of contributing study cohorts. Effect sizes are reported as unstandardized regression coefficients (β) with 95% confidence intervals. Data with fewer than 3 contributing clusters are flagged as hypothesis generating, given the limited reliability of variance estimation in this setting. Sex was coded (female = 1, male = 0) with a positive β coefficient indicating higher signal in women. Sensitivity analyses restricted datasets to tracer and ROI categories with demonstrated group signal differences to minimize dilution of disease-related effects from heterogeneous ROI selection and variable tracer sensitivity.

Quality Assessment

Studies were evaluated using the quality assessment of diagnostic accuracy studies 2 (QUADAS-2) tool,³⁸ assessing bias and applicability across 4 domains: patient selection, index test, reference standard, and flow and timing. The reference standard was any established MS diagnostic criteria (e.g., McDonald criteria³⁹) and the index test was TSPO-PET imaging. Signaling questions were rated as “yes” (low risk), “no” (high risk), or “unclear.” Applicability was evaluated for all domains except flow and timing. Certainty of evidence was assessed using the Grading of Recommendations Assessment, Development, and Evaluation (GRADE) framework for observational studies.⁴⁰ Two independent reviewers conducted the quality assessment, with disagreements resolved by a third reviewer.

Standard Protocol Approvals, Registrations, and Patient Consents

Approval was not required for this work as all data were extracted from published articles.

Data Availability

Data available on request.

Results

Study Characteristics

The systematic search identified 1,874 articles. After removal of 677 duplicates, 1,197 articles underwent title and abstract screening, of which 852 were excluded. Full-text review was performed for 345 articles, resulting in 55 articles meeting initial inclusion. Six studies were excluded due to limited or nonstandard reporting (eTable 3).^41-46 Hence, 49 articles were included in the analyses^{19-27,32,33,47-50,e1-34} (Figure 1). Of these, data from 41 studies were included in the meta-analysis comparing TSPO-PET signal between PwMS and HC, 37 studies were included in MS subtype comparisons, and 25 studies were included in patient-level regression analyses. eTable 4 summarizes methodological characteristics in the included studies. Participant clinical and demographic characteristics are summarized in eTable 5.

Figure 1. PRISMA Flowchart of the Systematic Review Article Screening and Inclusion.

Two pairs of studies reported overlapping populations. Barletta et al.³³ referenced population overlap with that of Herranz et al.³² Therefore, data from both studies were retained and grouped for analyses involving different ROIs. Similarly, data from Mantovani et al.^e15 and Pitombeira et al.^e20 were clustered together because Mantovani et al.^e15 described their study as a retrospective analysis of the cohort originally reported by Pitombeira et al.^e20 However, when both studies used identical ROIs and methodologies, only data from Pitombeira et al. were included in the analyses owing to its larger sample size.

Study Level TSPO-PET Signal Comparisons

Across all included studies, a random-effects model with cluster-robust variance (CRV) estimation showed significantly higher TSPO-PET signal in people with multiple sclerosis (PwMS) compared with HC (SMD = 0.44, 95% CI 0.27–0.61, p < 0.001, cluster number (N) = 39, estimate N = 278), with substantial residual heterogeneity (I² = 97.00%) (Figure 2). This CRV-adjusted model, which accounts for multiple statistically dependent effect size estimates within individual studies, was used as the primary analysis. Given this high heterogeneity, meta-regression and subgroup analyses investigated methodological contributors, including radiotracer selection, PET quantification metric, image analysis approach, references tissue application, and ROIs reported (Tables 1 and 2; eTables 6 and 7).

Figure 2. Forest Plot Showing the SMD of the Overall Difference in TSPO-PET Signal Between Patients With MS and Healthy Controls.

For better illustration this figure shows the overall SMD calculated using an aggregation REM. The cluster-robust variance estimate accounting for within-study dependencies (SMD=0.44, 95% CI 0.27–0.61, p < 0.001) is the primary inferential result reported in the text. RE Model = random effects model; SMD = standardized mean difference.

Table 1.

Meta-Regression and Subgroup Analyses of TSPO-PET Signal Between Patients With MS and Healthy Participants by Methodological Variables Including All Tracers

Subgroups	Estimates (N)a	Clusters (N)b	Subgroup analysis				Meta regression
			SMD of effect size (95% CI)	p Value	R2 (%)	I2 (%) (p value)	β coefficient (95% CI)	p Value	R2 (%)	I2 (%) (p value)
Tracer	278	39							11.47	96.57 (<0.001)
[11C]PK11195	142	20	0.42 (0.33 to 0.51)	<0.001	40.99	96.68 (<0.001)	0.42 (0.33 to 0.51)	<0.001
[11C]PBR28	33	7	0.83 (0.16 to 1.50)	0.031	0.00	97.09 (<0.001)	0.85 (0.25 to 1.47)	0.023
[18F]PBR06	48	2	0.18 (−4.06 to 4.42)	0.683	0.00	87.55 (<0.001)	0.19 (−4.14 to 4.52)	0.681
[18F]PBR111	28	3	0.27 (−2.95 to 3.49)	0.604	0.00	91.99 (<0.001)	0.26 (−2.88 to 3.40)	0.604
[18F]DPA-714	19	3	0.75 (0.07 to 1.42)	0.041	41.68	93.99 (<0.001)	0.74 (0.08 to 1.41)	0.040
[11C]DPA-713	5	2	0.72 (−2.69 to 4.12)	0.228	86.64	33.82 (0.198)	0.76 (−2.82 to 4.33)	0.226
[18F]FEDAA1106	2	1	−0.45c	—	—	—	−0.45c	—
[11C]ER176	1	1	0.92c	—	—	—	0.92c	—
Quantification metric	278	39							15.37	96.46 (<0.001)
DVR	135	23	0.58 (0.43 to 0.74)	<0.001	4.6	97.79 (<0.001)	0.58 (0.43 to 0.73)	<0.001
SUVR	65	7	0.31 (−1.02 to 1.64)	0.379	0.00	94.35 (<0.001)	0.32 (−0.98 to 1.61)	0.366
Vt	46	7	0.06 (−0.52 to 0.64)	0.618	94.95	53.75 (<0.001)	0.02 (−0.36 to 0.40)	0.846
BPND	12	5	0.87 (0.35 to 1.38)	0.015	27.37	76.85 (<0.001)	0.87 (0.35 to 1.39)	0.015
SUV	5	4	0.06 (−0.70 to 0.82)	0.824	54.29	75.17 (0.001)	−0.05 (−0.65 to 0.79)	0.774
GALPz	2	1	1.72c	—	—	—	1.72c	—
Vt/ƒp	1	1	−0.32c	—	—	—	−0.32c	—
Image analysis method	278	39							1.59	96.91 (<0.001)
Reference tissue kinetic models (SRTM & Logan-Ref)	113	24	0.57 (0.41 to 0.74)	<0.001	0.00	98.33 (<0.001)	0.57 (0.41 to 0.73)	<0.001
Static/Semiquantitative	72	11	0.33 (−0.67 to 1.33)	0.298	0.00	94.52 (<0.001)	0.34 (−0.65 to 1.32)	0.291
Arterial input graphical or multilinear methods (Logan-plot/MA1)	50	4	0.35 (−0.44 to 1.13)	0.132	56.75	92.36 (<0.001)	0.34 (−0.42 to 1.10)	0.136
Arterial input compartmental models (2TCM)	43	5	0.35 (−0.74 to 1.45)	0.318	18.6	88.26 (<0.001)	0.35 (−0.73 to 1.44)	0.315
Reference tissue within model	113	24							25.73	97.81 (<0.001)
4SCA	104	20	0.48 (0.36 to 0.61)	<0.001	37.14	97.55 (<0.001)	0.48 (0.36 to 0.61)	<0.001
Caudate	5	2	2.34 (−1.50 to 6.20)	0.082	0.00	97.66 (<0.001)	2.25 (−1.30 to 5.81)	0.076
Thalamus	2	1	1.13c	—	—	—	1.13c	—
10SCA	2	1	1.21c	—	—	—	1.21c	—
Reference tissue in ratio-based normalization	104	11							32.95	90.32 (<0.001)
Whole brain	45	1	0.10 (−0.13 to 0.32)	0.404	0.00	85.24 (<0.001)	0.10 (0.02 to 0.17)	0.037
NAWM clusterd	21	3	0.93 (0.47 to 1.39)	0.021	62.9	89.69 (<0.001)	0.94 (0.48 to 1.38)	0.020
4SCA	17	2	0.73 (0.63 to 0.84)	0.007	74.23	89.42 (<0.001)	0.73 (0.64 to 0.83)	0.007
Cortical GM	15	2	0.30 (−0.66 to 1.26)	0.159	68.37	77.57 (<0.001)	0.30 (−0.66 to 1.26)	0.158
NAWM	4	1	0.41 (−0.33 to 1.15)	0.278	0.00	84.04 (0.003)	0.37 (−0.77 to 1.52)	0.151
Cerebellum crus	1	1	0.92c	—	—	—	0.92c	—
GM	1	1	1.40c	—	—	—	1.40c	—

Abbreviations: 2TCM = 2-tissue compartment model; 4SCA = supervised cluster algorithm with 4 predefined clusters; 10SCA = supervised cluster algorithm with 10 predefined clusters; BP_ND = binding potential, nondisplaceable; DVR = distribution volume ratio; ƒp = plasma-free fraction; GALPz = glial activity load on PET Z-score; GM = gray matter; I² = degree of heterogeneity among the effect sizes of the included studies; MA1 = multilinear analysis; N = number; NAWM = normal-appearing white matter; R² = proportion of heterogeneity that can be explained by the meta regression/subgroup analysis; SCA = supervised cluster algorithm; SRTM = simplified reference tissue model; SUV = standardized uptake value ratio; SUVR = standardized uptake value ratio; Vt = volume of distribution; Vt/ƒp = Vt divided by plasma free fraction; WM = white matter.

^aEach estimate represents TSPO-PET signal comparison between patients and healthy controls (TSPO-PET signal SMD) for a brain ROI using a PET quantification metric, image analysis method, and specific tracer from included articles.

^bEach cluster represents a single article except for the article by Datta^e3 et al., in which case 2 tracers were used, and thus was treated as 2 clusters.

^cR², I², 95% CI, and p values are not reported for rows with less than 3 estimates, and their SMDs are reported as descriptive indicators due to small sample size.

^dClusters in the WM of healthy controls as well as in the NAWM of patients with MS that were within ±0.5 SD of the mean SUV of the global WM in controls were chosen as the reference region.

Table 2.

Meta-Regression and Subgroup Analyses of TSPO-PET Signal Between Patients With MS and Healthy Participants by Brain ROI Including All Tracers

Subgroups	Estimates (N)a	Clusters (N)b	Subgroup analysis				Meta regression
			SMD of effect size (95% CI)	p Value	R2 (%)	I2 (%) (p value)	β coefficient (95% CI)	p Value	R2 (%)	I2 (%) (p value)
All ROIsc	215	38							19.71	95.84 (<0.001)
NAWM	31	27	0.74 (0.46 to 1.01)	<0.001	0.00	98.01 (<0.001)	0.74 (0.48 to 0.99)	<0.001
Thalamus	30	23	0.81 (0.55 to 1.07)	<0.001	0.00	97.28 (<0.001)	0.64 (0.55 to 1.06)	<0.001
T2-WM lesion	19	14	0.33 (0.03 to 0.62)	0.032	10.62	94.95 (<0.001)	0.32 (0.03 to 0.61)	0.033
Perilesional area	18	12	0.35 (0.12 to 0.58)	0.008	47.81	95.00 (<0.001)	0.35 (0.12 to 0.58)	0.008
Cortex	15	14	0.56 (0.20 to 0.91)	0.005	7.08	97.10 (<0.001)	0.56 (0.20 to 0.91)	0.005
GM	13	8	0.24 (−0.32 to 0.80)	0.306	44.22	88.16 (<0.001)	0.22 (−0.34 to 0.77)	0.344
T1 WM lesion	12	11	0.00 (−0.42 to 0.42)	0.864	0	97.58 (<0.001)	0.00 (−0.42 to 0.43)	0.986
Caudate	10	6	−0.06 (−0.39 to 0.27)	0.558	88.45	77.91 (<0.001)	−0.03 (−0.30 to 0.23)	0.692
WM	10	6	0.76 (0.25 to 1.27)	0.020	18.09	95.21 (<0.001)	0.76 (0.25 to 1.27)	0.020
Brainstem	9	4	0.48 (−0.31 to 1.27)	0.114	36.64	92.16 (<0.001)	0.48 (−0.32 to 1.28)	0.115
Cerebellum	9	4	0.19 (−0.61 to 1.00)	0.389	32.80	93.57 (<0.001)	0.20 (−0.60 to 0.99)	0.384
Putamen	8	4	0.64 (0.07 to 1.21)	0.042	50.08	92.84 (<0.001)	0.64 (0.09 to 1.19)	0.038
Whole brain	8	8	0.36 (0.11 to 0.62)	0.013	80.53	88.15 (0.001)	0.36 (0.08 to 0.65)	0.020
Pallidum	7	3	0.55 (−1.67 to 2.77)	0.281	19.27	95.65 (<0.001)	0.55 (−1.67 to 2.77)	0.281
Hippocampus	6	4	0.96 (0.23 to 1.68)	0.027	25.72	87.98 (<0.0001)	0.96 (0.24 to 1.68)	0.026
Striatum	4	3	0.18 (−0.29 to 0.66)	0.218	83.65	65.80 (0.045)	0.20 (−0.33 to 0.73)	0.227
Insula	3	2	0.23 (−2.87 to 3.33)	0.522	58.54	75.40 (0.017)	0.24 (−2.93 to 3.41)	0.513
T2 cortical lesion	3	2	1.34 (−2.67 to 5.34)	0.146	52.50	83.95 (0.002)	1.33 (−2.66 to 5.31)	0.148

Abbreviations: GM = gray matter; I² = degree of heterogeneity among the effect sizes of the included studies; N = number; NAWM = normal-appearing white matter; R² = proportion of heterogeneity that can be explained by the meta regression/subgroup analysis; ROI = region of interest; WM = white matter.

^aEach estimate represents TSPO-PET signal comparison between patients and healthy controls (TSPO-PET signal SMD) for a brain ROI using a PET quantification metric, image analysis method and specific tracer from included articles.

^bEach cluster represents a single article except for the article by Datta^e3 et al., in which case 2 tracers were used, and thus was treated as 2 clusters.

^cROIs reported fewer than 3 times were classified as other (eTable 2) and excluded from ROI-based analyses.

Radiotracer Selection

Meta-regression indicated that radiotracer selection accounted for 11.47% of between-study heterogeneity (Table 1). However, substantial residual heterogeneity remained (I² = 96.57%), suggesting that additional methodological and/or biological factors contribute to variability in TSPO-PET estimates. Several radiotracer categories were represented by fewer than 3 effect size estimates; these were retained in the meta-regression, but their individual contributions should be interpreted cautiously.

Significantly elevated TSPO-PET signals in PwMS were observed only in studies using [¹¹C]PK11195, [¹¹C]PBR28, and [¹⁸F]DPA-714. Subgroup analyses corroborated these findings. However, [¹¹C]DPA-713 was not significant and was based on few estimates, while [¹⁸F]FEDAA1106 and [¹¹C]ER176 had insufficient data for reliable meta-analysis (Table 1).

Assessment of heterogeneity within tracer subgroups revealed significant heterogeneity for most tracers. Heterogeneity could not be reliably detected for [¹⁸F]FEDAA1106 and [¹¹C]ER176 due to insufficient data, whereas [¹¹C]DPA-713 exhibited low and nonsignificant heterogeneity (I² = 33.82%, p > 0.050) (Table 1). These findings highlight substantial tracer-dependent variability in TSPO-PET signal estimation in MS and support consideration of radiotracer selection when interpreting between-study differences.

PET Quantification Metric

Meta-regression identified PET quantification metric as a significant contributor to heterogeneity, explaining 15.37% of variance, with substantial residual heterogeneity (I² = 96.46%, Table 1). Significant differences between PwMS and HC were observed using DVR and BP_ND. GALPz also showed apparent differences; however, with only 2 estimates from a single cluster, this finding could not be reliably assessed. Subgroup analyses yielded consistent results, although substantial residual heterogeneity persisted across most approaches.

To further evaluate the impact of quantification methodology under conditions of demonstratable tracer sensitivity, a secondary sensitivity analysis was performed including only tracers that exhibited significantly elevated TSPO-PET signal in PwMS relative to HC (i.e., [¹¹C]PK11195, [¹¹C]PBR28, and [¹⁸F]DPA-714). In this restricted dataset, PET quantification metric accounted for 13.93% of the between-study variance (R² = 13.93%), with substantial residual heterogeneity remaining (I² = 96.86%, eTable 6). Significant group differences were observed not only using DVR and BP_ND but also for SUVR. Restricting analyses to tracers with demonstrable TSPO-PET signal differences between PwMS and HC showed a marked reduction in heterogeneity for select quantification metrics. Residual heterogeneity was low for Vt (R² = 98.88%, I² = 21.16%) and moderate for BP_ND (R² = 59.65%, I² = 63.25%). By contrast, DVR and SUVR estimates remained highly heterogeneous (R² = 0.0%, I² = 98.30% and R² = 43.15%, I² = 93.98%, respectively).

Image Analysis Method

Meta-regression demonstrated that image analysis method accounted for only a small proportion of study heterogeneity (R² = 1.59%) and exhibited high residual heterogeneity (I² = 96.91%). Significant differences in TSPO-PET signal between PwMS and HC were observed using reference tissue-based kinetic models (i.e., SRTM & Logan reference), but not in static/semiquantitative methods and arterial-input compartmental, graphical, or multilinear models. Subgroup analyses mirrored these results, confirming that significant elevations in TSPO-PET signal were restricted to reference-tissue kinetic model methods (Table 1).

When analyses were restricted to tracers with demonstrable TSPO-PET signal differences, meta-regression again attributed minimal heterogeneity to image analysis method (R² = 3.65%, I² = 97.16%). In this tracer-restricted dataset, significant group differences remained for reference tissue-based kinetic models, with additional significance identified with static/semiquantitative approaches (SUV, SUVR), supported by both meta-regression and subgroup analyses (eTable 6).

Reference Tissue Selection

Although image analysis method explained only a small proportion of heterogeneity, reference tissue selection represents an important methodological consideration because it directly informs the calculation of DVR, BP_ND, and SUVR quantification, metrics associated with between-study variability and group differentiation. Reference tissue effects were therefore examined separately according to whether the reference region was incorporated within quantitative models or applied post hoc as a ratio-based normalization.

Reference Tissue Within Quantitative Models

Meta-regression demonstrated that reference tissue selection within quantitative models accounted for approximately a quarter of heterogeneity (R² = 25.73%, Table 1) in associated studies, with high residual heterogeneity remaining (I² = 97.81%). Meta-regression and subgroup analyses of reference tissue selection within quantitative models revealed that supervised cluster analysis-derived reference regions (i.e., 4SCA and 10SCA) and the thalamus were associated with significant differentiation between PwMS and HC, with the caudate exhibiting a trend (p = 0.075–0.082, Table 1). Notably, estimates were heavily weighted toward 4SCA, with few studies contributing data for alternative reference tissue approaches. Accordingly, findings regarding reference tissue effects beyond 4SCA should be interpreted cautiously.

Restricted tracer analyses demonstrated reduced variance (R² = 16.98%) and showed that 4SCA, 10SCA, and the caudate were associated with significant differences between groups (eTable 6).

Reference Tissue for Ratio-Based Normalization

For post hoc ratio-based approaches, reference tissue selection accounted for nearly a third of heterogeneity (R² = 32.95%, Table 1) in meta-regression analyses, with high residual heterogeneity persisting (I² = 90.32%). The use of NAWM-derived clusters, and 4SCA was associated with significant differences in TSPO-PET signal between HC and PwMS. Subgroup analyses corroborated findings for NAWM-derived clusters and 4SCA, with the addition of whole brain.

Tracer-restricted meta-regression somewhat reduced the attributed variance (R² = 24.32%) and retained NAWM-derived clusters and 4SCA as reference tissues associated with significant group differences. However, subgroup analyses differed slightly with the addition of cortical gray matter (GM) significance (eTable 6). Consistent with the algorithm-based analyses, GM and cerebellum crus regions demonstrated low estimates (n = 1) and should be interpreted descriptively only.

ROIs

To evaluate regional TSPO-PET signal differences between HC and PwMS, meta-regression by ROI was performed (Table 2). ROI-based meta-regression accounted for 19.71% of heterogeneity, with substantial residual heterogeneity remaining (I² = 95.84%). Meta-regression demonstrated significantly higher TSPO-PET signal in PwMS compared with HC in the cortex, hippocampus, NAWM, perilesional area, putamen, thalamus, whole brain, WM, and T₂-defined magnetic resonance imaging WM lesions. Subgroup analyses yielded results concordant with meta-regression findings and similarly demonstrated substantial residual heterogeneity across most ROIs. eFigure 1 illustrates regional TSPO-PET signal differences in brain, with SMDs calculated using aggregation REM. Analyses including restricted tracers showed somewhat similar results with the putamen and T₂ WM lesions showing trends but not reaching significance (eTable 7).

Multivariate Meta-Regression Accounting for Methodological Variables

Meta-analysis of TSPO-PET signal comparison between PwMS and healthy populations was repeated using a multivariate CRV model accounting for key methodological variable including tracer selection, PET quantification metric, image analysis method, reference tissue application, and ROI. In this multivariate model, TSPO-PET signal remained significantly higher in PwMS compared with HC (model intercept = 3.86, 95% CI 2.40–5.33, p < 0.001). Although these methodological variables contributed to 48.27% of variance, residual heterogeneity remained substantial (I² = 93.24%).

TSPO-PET Across MS Subtypes

Pairwise CRV meta-analyses across 35 studies comparing PMS, RRMS, and HC demonstrated significantly elevated TSPO-PET signal in PMS relative to RRMS (SMD = 0.48, 95% CI: 0.31–0.66, p = 0.0002, I² = 93.41%, cluster N: 24, estimate N: 127). Both RRMS and PMS exhibited significantly higher TSPO-PET signals compared with HC (RRMS: SMD = 0.27, 95% CI: 0.06–0.47, p = 0.0161, I² = 94.58%, cluster N: 28, estimate N: 206, PMS: SMD = 0.73, 95% CI: 0.39–1.06, p = 0.0012, I² = 96.33, cluster N: 21, estimate N: 132). These results demonstrate a difference in TSPO-PET signal across phenotypes, with the greatest effect sizes observed in PMS (Figure 3). However, heterogeneity remained high across all subtype comparisons.

Figure 3. Forest Plot Showing the SMDs in Brain TSPO-PET Signal Between (A) Patients With PMS vs Patients With RRMS (B) Patients With RRMS and Healthy Controls and (C) Patients With PMS vs Healthy Controls.

For better illustration, the overall SMD displayed in the following forest plots was calculated using an aggregation random effects model. For better illustration, this figure shows the overall SMDs calculated using an aggregation random effect model. The cluster-robust variance estimates accounting for within-study dependencies are the primary inferential result reported in the text. PMS = progressive multiple sclerosis; RE Model = random effects model; RRMS = relapsing-remitting multiple sclerosis; SMD = standardized mean difference.

Patient-Level Regressions: TSPO-PET and Clinical Measures

When all available data were pooled across all reported tracers and ROIs, no statistically significant associations were found between TSPO-PET signal and clinical or demographic variables. Analyses including restricted tracers and ROIs also did not show statistically significant associations (eTable 8).

Because pooling across numerous heterogeneous ROIs may obscure region-specific and disease-specific associations, additional ROI-stratified analyses were performed in regions implicated in MS pathology. Including all tracers, analyses showed significant but weak associations between DVR and EDSS in NAWM (Table 3). However, the observed β coefficients were small, and several ROI-specific estimates were based on a limited cluster N. Similar results were seen when restricting tracers (eTable 9), with a significant association emerging between DVR and EDSS in NAWM. No data points were available to assess the association between DVR and MSSS in the perilesional area, and the estimates needed to evaluate the association between DVR and SDMT Z-score could not be extracted. Overall, these findings should be interpreted cautiously because the available data for several restricted and ROI-specific analyses were limited in size and completeness, which may have reduced the precision and robustness of the estimated associations.

Table 3.

CRV Based Linear Regression Between TSPO-PET Signal DVR and Disability Scores of EDSS and MSSS in Patients With MS in Select ROIs Including All Tracers (No Data Could be Extracted for SDMTz in the Following ROIs)

Quantification method	Variables correlated with TSPO-PET signal	ROI	Estimates (N)a	Clusters (N)b	β coefficient	95% CI	p Value
DVR	EDSS	NAWM	377	9	0.02	0.01–0.02	<0.001
		Perilesional area	199	2c	0.01	−0.01–0.03	0.088
		Thalamus	209	5	0.02	−0.01–0.05	0.097
		T2 WM lesion	26	3	0.02	−0.10–0.15	0.250
	MSSS	NAWM	153	5	0.01	0.00–0.03	0.092
		Perilesional area	—	—	—	—	—
		Thalamus	42	2^	0.01	−0.04–0.06	0.198
		T2 WM lesion	26	3	0.03	−0.16–0.22	0.307

Abbreviations: CRV = cluster-robust variance; DVR = distribution volume ratio; EDSS = Expanded Disability Status Scale; MS = multiple sclerosis; MSSS = multiple sclerosis severity score; N = number; NAWM = normal-appearing white matter; ROI = region of interest; SDMTz = symbol digit modalities test Z score; TSPO = 18-kDa translocator protein; WM = white matter.

^aEach estimate here represents TSPO-PET signal for a brain ROI in a single patient using a specific PET quantification metric, image analysis method and tracer from included articles.

^bEach cluster represents a single article except for the article by Datta^e3 et al., in which case 2 tracers were used, and thus was treated as 2 clusters.

^cFew clusters, interpret as hypothesis-generating.

Quality Assessment

Patient selection was rated as low risk of bias in all studies except one, which had unclear risk for both bias and applicability owing to the absence of selection criteria reporting^e14 (eTable 10). Four studies (8.16%) included LAB participants with second-generation TSPO-PET tracers without stratification,^{26,e2,e31,e33} and 1 included patients with clinically isolated syndrome²⁵; thus, these studies were rated as high risk in patient selection applicability. For the index test, 6 studies (12.24%) reported blinded analyses and were designated low risk,^{32,48,e18,e24,e28,e31} whereas the remaining studies classified as unclear risk. Applicability bias for the index test was low across studies. For the reference standard, 34 studies (69.39%) reported MS diagnosis using established clinical criteria (low risk), while 15 studies (30.61%) did not specify diagnostic criteria (unclear risk).^{21,26,32,49,e8-11,e13-14,e16,e21,e25,e30-31} All studies demonstrated low risk of bias for flow and timing.

Funnel plot rank correlation and Egger revealed no substantial asymmetry or publication bias in the overall analysis of MS vs HC or in MS subtype comparisons (eFigures 2 and 3). Certainty of evidence rating was initially low, owing to the observational study design and remained low after upgrading for the multivariate effect and downgrading for substantial remaining heterogeneity. No serious concerns were identified for risk of bias affecting outcome measurement or for indirectness, imprecision, and publication bias (eTable 11). Owing to low number of inputs, publication bias analyses and GRADE rating were not carried out for subgroups. Given the high residual heterogeneity, publication bias tests are limited and cannot exclude its presence. Similarly, GRADE certainty is also limited by the high residual heterogeneity.

Discussion

This study demonstrated that TSPO-PET signal is elevated in PwMS compared with HC, supporting its potential as an in vivo marker of innate inflammatory activity in MS. TSPO-PET also distinguished MS subtypes, with higher signal observed in PMS relative to RRMS, and both showing elevated signal compared with controls. These findings align with pathologic evidence of sustained innate immune activity in progressive disease and support the ability of TSPO-PET to detect biologically meaningful differences across disease states.^e35

Substantial heterogeneity was observed across studies, much of which was attributable to methodological differences. Notably, R² values reported are exploratory and model dependent and do not fully attribute variance to biological vs methodological factors. Radiotracer selection accounted for 11.47% of heterogeneity, consistent with known differences in ligand signal-to-noise ratio and genotype sensitivities.^13,26,e36 Although the rs6971 polymorphism affects second-generation tracers, 20 studies used [¹¹C]PK11195, which is not genotype sensitive. Despite its lower signal-to-noise ratio and higher nonspecific binding,^26,e36 [¹¹C]PK11195 was associated with significantly elevated signal in PwMS. For second-generation tracers, incomplete reporting and stratification by binding class likely introduced additional variability across LAB, MABs, and HABs. Overall genotype effects could not be modeled as few studies reported and stratified data. In addition, tracer and/or tracer generation (first, second, and third) may also contribute to heterogeneity in our data. Therefore, tracer-restricted sensitivity analyses were performed to evaluate how subsequent analytic choices (quantification metric, modeling approach, reference tissue strategy, and ROI definition) behave under demonstrated tracer sensitivity. However, it should be noted that restricted tracer analyses are not representative of the full TSPO-PET literature. They represent signal-enriched conditions and may amplify differences between PwMS and HC.

PET quantification metric also accounted for substantial variance (15.37%). BP_ND and DVR showed significantly elevated signal in MS cohorts in the full dataset. Differences across metrics likely reflect varying sensitivity to modeling assumptions, plasma input requirements, and normalization strategies. SUVR reached significance only under tracer restriction. Under tracer-restricted conditions, heterogeneity decreased for selected approaches, with Vt and BP_ND showing greater stability than DVR and SUVR, likely owing to variability introduced by reference region selection. By contrast, image analysis method accounted for only a small portion of heterogeneity before and after tracer restriction (1.59–3.65%). Significant group differences were seen with referenced-based methods and additionally with static quantification in restricted tracer analyses, including SUVR, suggesting that tracer selection and reference region may be important considerations.

Accordingly, reference tissue and normalization strategy contributed meaningfully to between-study variability. Both within model and ratio-based quantification were attributable to variance. SCA-derived regions are more consistently associated with group discrimination than several anatomically defined reference regions, although evidence for non–SCA-driven approaches was limited. ROI analyses demonstrated consistently higher TSPO-PET signal in MS compared with HC, including in NAWM, perilesional area, T₂ WM lesions, total WM, thalamus, putamen, cortex, and hippocampus. Elevated perilesional signals are consistent with prior pathologic and imaging evidence of rim-associated myeloid activation; other ROI involvement aligns with growing recognition of diffuse and compartmentalized inflammation beyond focal lesions.^{43,e12,e37-38} NAWM, from which many studies have selected a cluster to serve as a pseudo-reference region,^{22,33,e8-9,e11} showed significant MS-control differences in pooled analyses. Similar patterns were observed for the thalamus, cortex, and whole brain, which have also been used as normalization regions and included in analyses here.^e1,e2,e24 These findings suggest that these regions may be better suited as target ROIs for detecting MS-related inflammatory changes, instead of reference normalization regions in MS populations. ROIs reported also accounted for a substantial proportion of between-study heterogeneity, exceeding the contribution of tracer selection and image analysis method, underscoring the importance of anatomical selection in TSPO-PET study design and interpretation. Together, these results support careful selection and validation of reference regions and ROIs in TSPO-PET MS studies and favor disease-sensitive regions as targets rather than normalization baselines.

When accounted for all mentioned heterogeneity sources using a multivariate model with CRV, an even greater difference between PwMS and HC in TSPO-PET signal was found, accounting for 48.27% of the initial heterogeneity. However, significant heterogeneity remained. Thus, additional unaccounted sources of heterogeneity were explored, one of which was MS disease course.

Individual participant data analyses were limited by small numbers of independent cohorts, constraining statistical power and precision. The effect sizes were small, and the absence of consistent or stronger associations should not be interpreted as definitive evidence of no relationship but in the context of limited patient-level data. Similarly, ROI-stratified correlations involve multiple testing and limited data. Therefore, these findings should be regarded as hypothesis generating, with additional data needed to better understand these relationships across studies.

Several limitations should be considered. In addition to patient-level data availability, disease activity status at the time of imaging was inconsistently reported, restricting assessment of active vs inactive inflammatory states. Data on disease-modifying therapies exposure were incomplete and heterogeneous, with many studies not reporting patient-level data, precluding formal evaluation of treatment effects on TSPO-PET signal despite the likelihood that immunomodulatory therapies may influence TSPO-PET signal differently.^e25 TSPO binding polymorphism status was also not consistently reported, preventing assessment of genotype-related signal variability. Only one study presented separate results for Gd enhancing (Gd+) T₁ lesions²⁵ and one for Gd+ T₂ lesions,^e11 while most did not make a distinction. Thus, all lesions were pooled together as T₁ or T₂-weighted in analysis, causing a possible source of heterogeneity in our results. Additional limitations related to data availability and reporting are summarized in eTable 3. PVC practices were inconsistently reported, and thus, PVC effects were not examined. Potentially relevant clinical and biological covariates including comorbid conditions (e.g., depressive symptoms), which may affect TSPO-PET signal were not analyzed due to limited reporting across studies.

In conclusion, TSPO-PET distinguishes MS from HC and across subtypes, with regional indication of some correspondence with disability. However, findings are strongly influenced by methodological heterogeneity, sparse patient-level reporting, and inconsistent outcome measures, highlighting the need for greater standardization in study design and reporting, raw data transparency, and larger multimodal studies to clarify its relationship to disease progression and treatment response.

Glossary

2TCM

2-tissue compartment model

4SCA

supervised cluster algorithm with 4 predefined clusters

10SCA

supervised cluster algorithm with 10 predefined clusters

BP_ND

binding potential, nondisplaceable

CRV

cluster robust variance

DMT

disease-modifying therapies

DVR

distribution volume ratio

EDSS

Expanded Disability Status Scale

ƒp

plasma free fraction

GALPz

glial activity load on PET Z score

GD+

gadolinium enhancing

GD-

gadolinium nonenhancing

GRADE

grading of recommendations assessment, development, and evaluation

GM

gray matter

HAB

high affinity binding

HCs

healthy controls

I ²

degree of heterogeneity

IDIF

image-derived input function

LAB

low affinity binding

Logan-Ref

Logan referenced tissue model

MA1

multilinear analysis 1

MAB

medium affinity binding

Max

maximum

MeSH

medical subject headings

Min

minimum

MS

multiple sclerosis

MSSS

multiple sclerosis severity score

N

number

NAGM

normal-appearing gray matter

NAWM

normal-appearing white matter

PICO

population, intervention, comparison, outcome

PMS

progressive multiple sclerosis

PRISMA

Preferred Reporting Items for Systematic Reviews and Meta-Analyses

PVC

partial volume correction

PwMS

patients with multiple sclerosis

QUADAS-2

quality assessment of diagnostic accuracy studies 2

R ²

coefficient of determination

REM

random-effects model

ROI

region of interest

RRMS

relapsing-remitting multiple sclerosis

SCA

supervised cluster algorithm

SDMTz

symbol digit modalities test Z-score

SMD

standardized mean difference

SRTM

simplified reference tissue model

SUV

standardized uptake value

SUVR

standardized uptake value ratio

TSPO

18-kDa translocator protein

Vt

volume of distribution

Vt/ƒp

Vt divided by plasma free fraction

WM

white matter

Author Contributions

A. Madani Neishaboori: drafting/revision of the manuscript for content, including medical writing for content; major role in the acquisition of data; analysis or interpretation of data. M. Agarwal: drafting/revision of the manuscript for content, including medical writing for content; analysis or interpretation of data. A. Kalinsky: major role in the acquisition of data; analysis or interpretation of data. J.P. Lofaso: major role in the acquisition of data; analysis or interpretation of data. G. Rampilla: drafting/revision of the manuscript for content, including medical writing for content; analysis or interpretation of data. E.A. Kras: drafting/revision of the manuscript for content, including medical writing for content; analysis or interpretation of data. S.C. Nagy: drafting/revision of the manuscript for content, including medical writing for content; analysis or interpretation of data. J.G. Cooper: drafting/revision of the manuscript for content, including medical writing for content; analysis or interpretation of data. S. Ai: drafting/revision of the manuscript for content, including medical writing for content; analysis or interpretation of data. C.K. Zhang: drafting/revision of the manuscript for content, including medical writing for content; analysis or interpretation of data. K. Friedrichsen: drafting/revision of the manuscript for content, including medical writing for content; analysis or interpretation of data. M. Brier: drafting/revision of the manuscript for content, including medical writing for content; analysis or interpretation of data. A.M. Chaney: 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 work was supported by the NIH under award number (R00AG070105). This manuscript is the result of funding in whole or in part by the NIH. It is subject to the NIH Public Access Policy. Through acceptance of this federal funding, NIH has been given a right to make this manuscript publicly available in PubMed Central upon the Official Date of Publication, as defined by NIH.

Disclosure

The authors report no relevant disclosures. Go to Neurology.org/NN for full disclosures.