First evaluation in multiple sclerosis using PET tracer [¹⁸F]3F4AP demonstrates heterogeneous binding across lesions

Abstract

Purpose

Multiple sclerosis is characterized by demyelination and axonal loss in the central nervous system. While magnetic resonance imaging (MRI) is sensitive to myelin changes, conventional sequences lack specificity and cannot reliably distinguish demyelinated lesions with varying axonal integrity. We evaluated the potential of [¹⁸F]3F4AP, a novel positron emission tomography (PET) tracer and a radiolabeled derivative of 4-aminopyridine, to provide unique insights into demyelination in multiple sclerosis.

Methods

Three people with multiple sclerosis and three healthy controls underwent 2 h dynamic PET scanning with arterial blood sampling after administration of 296 MBq of [¹⁸F]3F4AP and 3T MRI. MRI sequences included T1 Magnetization Prepared Rapid Gradient Echo (MPRAGE), T2 Fluid-Attenuated Inversion Recovery (FLAIR), and Myelin Water Imaging (MWI). Kinetic modeling assessed tracer delivery and binding in brain regions and lesions. PET findings were compared to MRI and correlated with myelin content.

Results

[¹⁸F]3F4AP rapidly entered the brain, showing 14% higher accumulation in gray and white matter in multiple sclerosis than controls. Tracer kinetics were best described using a two-tissue compartmental model (2TCM). Logan graphical analysis provided excellent correlation with 2TCM volume of distribution (V_T) and late Standardized Uptake Value (SUV) images showed good correlation to V_T within subjects and within groups. Lesions showed more variable tracer binding compared to mirrored normal-appearing white matter (NAWM), with some lesions exhibiting high or low uptake despite similar appearance on MRI. A weak inverse correlation between tracer uptake and MWI was observed in controls but not in people with multiple sclerosis.

Conclusion

Despite a small sample size, [¹⁸F]3F4AP showed good reproducibility and demonstrated differences between people with multiple sclerosis and controls. Its heterogeneous binding across lesions suggests that [¹⁸F]3F4AP can differentiate lesions with and without axonal integrity, which could be very valuable for monitoring multiple sclerosis progression and evaluating remyelination therapies.

Trial registration

NCT04699747, Start date March 25^th, 2021.

Supplementary Information

The online version contains supplementary material available at 10.1007/s00259-025-07454-1.

Introduction

Multiple sclerosis (MS) is a chronic, progressive, neuroinflammatory disease of the central nervous system (CNS), characterized by the appearance of focal demyelinated lesions in the brain [1–3]. In the initial stages of lesion formation, the blood-brain barrier (BBB) becomes permeable and allows inflammatory myelin-reactive T-cells and B-cells to enter. These cells attack the myelin and cause demyelination, which impairs propagation of action potentials and leads to neurological symptoms [4]. Lesions can either remyelinate, which can improve clinical symptoms, or undergo irreversible axonal loss and contribute to progressive disability. In order to be able to effectively monitor the disease and develop new treatments, it is critical to be able to image each of these different processes.

BBB damage can be detected using MRI with gadolinium contrast [5]. Neuroinflammation can be quantified, at least some component of it, using TSPO PET [6]. Demyelination can be imaged using Myelin Water Imaging (MWI) [7] or using the PET tracers [¹¹C]PiB [8, 9] and [¹¹C]MeDAS [10, 11] but these methods do not provide information of whether the axons are preserved, which is required for remyelination. In this work, we hypothesized that [¹⁸F]3F4AP, a PET tracer based on the MS drug 4-aminopyridine (4AP) that binds to potassium (K⁺) channels in demyelinated axons, may differentiate MS lesions with and without preserved axons. Potassium channels of the K_v1 family play a crucial role in facilitating axonal signal conduction and are exposed and upregulated due to demyelination [12–15]. The drug 4-aminopyridine is FDA-approved for increasing walking speed in people with MS and has been used to enhance axonal conduction and improve clinical symptoms associated with demyelination [16–18]. When myelin is lost, K⁺ ions leak out of exposed voltage-gated K_v channels in axons, disrupting the ionic gradient that is necessary for proper axonal signaling. By blocking these K_v channels, 4AP prevents leakage and enhances action potential propagation and promotes more efficient signaling with sustained depolarization [19–21].

Positron emission tomography complements MRI by offering molecular-level insights into specific pathophysiological processes. Unlike MRI, which is primarily anatomical, PET can detect molecular markers of disease and provide quantitative measures of underlying pathology. Prior studies have shown that [¹⁸F]3F4AP, a radiolabeled derivative of 4AP, binds to de/dysmyelinated axons and can be used to detect demyelination in rodent models of MS [22]. This tracer has also shown colocalization in demyelinated areas in a rat spinal cord injury model [21], as well as in mice and monkeys with traumatic brain injuries [23, 24]. Furthermore, [¹⁸F]3F4AP has demonstrated suitable pharmacokinetics in the brain of nonhuman primates [23] and favorable dosimetry and biodistribution in humans [25]. In the context of MS, this tracer’s apparent specific binding to axonal K_v channels is expected to result in increased PET signal in demyelinated lesions with preserved axonal integrity and reduced PET signal in lesions with substantial axonal loss. Other tracers, such as the amyloid tracers Pittsburgh Compound B (PIB) [8, 9], [¹⁸F]florbetaben [26], [¹⁸F]florbetapir [27, 28], and the myelin tracer [¹¹C]MeDAS [10, 11] have been proposed to quantify myelin changes in MS [29]. These tracers show decreases in signal in the lesions but do not provide information about axonal integrity. In contrast, [¹⁸F]3F4AP is expected to show increased binding in areas with high concentration of demyelinated axons, as a result of binding to higher numbers of potassium channels that are exposed and upregulated [12, 13, 15, 30, 31].

In this study, we first evaluated the robustness of [¹⁸F]3F4AP as a brain imaging agent in humans, including in people with MS. Our objectives were three-fold: first, to quantify the kinetics and binding of [¹⁸F]3F4AP in the human brain using computational models; second, to determine if [¹⁸F]3F4AP binding differs between healthy controls and people with MS; and third, to assess whether [¹⁸F]3F4AP provides additional information beyond MRI in people with MS. This pilot evaluation of [¹⁸F]3F4AP in people with MS aims to provide initial evidence of whether [¹⁸F]3F4AP may assist in the differentiation between inflammatory and neurodegenerative components in MS and could serve as a valuable imaging endpoint in clinical trials for remyelination therapies.

Materials and methods

Participants and study design

Healthy volunteers were recruited through online ads. People with MS were recruited from the MS Clinic at Massachusetts General Hospital. After expressing interest, subjects were briefed on the study by a clinical research coordinator (AWR) and informed consent was obtained by a physician (ECK). Participating criteria included adults between 18 and 65 years old without neurological conditions (control group) and adults between 18 and 65 years old with a clinically definite diagnosis of multiple sclerosis (MS group) based on 2017 McDonald criteria, and no contraindications to the study such as severe claustrophobia, inability to provide informed consent or accumulated annual radiation dose greater than 30 mSv. No subjects were excluded on the bases of sex, ethnicity, or race. Subjects underwent an MRI and PET within 30 days of each other. Clinical information including medication history and expanded disability status scale (EDSS) score were obtained from their medical record and assessed by the same EDSS-certified rater (ECK).

Radiotracer production

[¹⁸F]3F4AP was produced by the MGH PET Core cGMP radiopharmacy using a Neptis ORA synthesizer as previously communicated [32]. The synthesis method is based on the previous report by Basuli et al. [33]. The tracer was purified using a semiprep HPLC column (Waters XBridge C-18, 5 μm, 10 × 250 mm) using 20 mM sodium phosphate (pH 8) mobile phase containing 5% ethanol at a flow rate of 4 mL/min. The HPLC fraction containing the product (approx. t_R 10–11 min) was diluted with 10 mL of 0.9% sodium chloride for injection, USP, and passed through a 0.22 μm sterilizing PES filter into a vented 30 mL sterile empty vial. The product vial was visually inspected and quality control was performed to FDA and USP standards for chemical identity and purity, radiochemical purity, pH, radionuclidic identity, residual solvents, sterility, and bacterial endotoxins. Identity was confirmed by coinjection of authentic standard on an analytical HPLC column (Phenomenex Gemini C-18, 5 μm, 4.6 × 250 mm) with mobile phase 10 mM sodium phosphate dibasic, 0.25% triethylamine and 5% acetonitrile at a flow rate of 1 mL/min (t_R = 10 min). Purity was assessed by area under the curve (AUC) of the product peak at 239 nm relative to other peaks not present in the blank. Molar activity (MA) at the time of injection ranged from 40.3 to 418 GBq/µmol. The sterile filter used was tested for integrity. The dose was released for injection after passing all quality control tests except for sterility, which was performed after release. All the batches of [¹⁸F]3F4AP used in this study met all product specifications, including sterility test.

Radiotracer administration

Dose was drawn into a syringe, measured, and administered intravenously as a 1-minute infusion through a catheter in the hand or arm. After administration of the dose, the catheter was flushed with 10 mL of saline and the residual activity in the syringe and catheter measured to calculate the injected dose (ID). Vital signs were measured before and after the scan.

PET/CT image acquisition and reconstruction

Imaging was performed on a GE Discovery MI PET/CT 5 ring scanner. Subjects were positioned on the bed of the scanner and a low dose CT of the head was acquired prior to radiotracer administration. Dynamic PET images were acquired from 0 to 120 min with the time bins set to 6 × 10 s, 8 × 15 s, 6 × 30 s, 8 × 60 s, 8 × 120 s, and 18 × 300s. MS002’s acquisition ended at 116 min, and MS003 took a 13-min break at 63 min. Another low dose CT image was acquired at the start of the second part of the MS003’s acquisition for attenuation correction. Dynamic PET images were generated with a 3D time-of-flight reconstruction algorithm incorporating data-driven motion correction in each frame [34]. To improve PET image resolution and reduce PET image noise, a second reconstruction step using the kernel method-based image reconstruction method was added to generate PET images from raw PET list-mode data [35]. For the kernel method, the PET image was represented as a function of features derived from the anatomical image. In our study, the anatomical image came from the MRI MPRAGE sequence. The features of each voxel were the intensity values from the MRI image in a patch centered around the considered voxel. Transformation of the extracted features was computed for each voxel and saved as a kernel matrix that was included in the PET list-mode reconstruction framework. The kernel method applied to dynamic PET imaging has been validated and published [36]. While we focused on results derived from images using the high-resolution kernel method-based reconstruction, in Supplementary Table 1, we show that results obtained from conventional reconstruction and kernel method-based reconstruction in 16 brain regions of various sizes yield similar results. The kernel method-based reconstruction was utilized to generate dynamic images from 0 to 120 min and a late-phase image using the data acquired between 76 and 106 min. For the dynamic images, additional frame-to-frame registration was performed to achieve inter-frame motion correction.

MRI acquisition and processing

Participants underwent MRI on a Siemens 3 T Prisma scanner. MRI sequences included T1 Magnetization Prepared Rapid Gradient Echo (MPRAGE), T2 Fluid-Attenuated Inversion Recovery (FLAIR), and Myelin Water Imaging (MWI) consisting of SPoiled GRadient-echo (SPGR) scans and balanced Steady-State Free Precession (bSSFP) scans acquired over a range of flip angles [37]. Participants with MS also received the gadolinium (Gd) contrast agent Gadavist and underwent post-Gd MPRAGE.

PET/CT and MRI image analysis

MPRAGE sequences were processed using FreeSurfer 7.2 to segment brain regions [38]. We created a composite region for the deep gray matter based on the segmentation of the hippocampus, putamen, caudate and thalamus. Cortex regions were also concatenated to create a gray matter mask. T2 lesion segmentation was done on FLAIR sequences using in-house tools. FreeSurfer-generated white matter mask excluded white matter hyperintensities. It was also eroded by 2 mm to minimize the impact of partial volume effects from the gray matter. All other MRI sequences, such as MWI, were also co-registered to the MPRAGE. In addition, the MPRAGE sequence was transformed into the Montreal Neurological Institute (MNI) template space, and the same transformation was applied to the lesions’ masks. Contralateral/mirrored regions of interest were then mathematically generated by mirroring the coordinates of each voxel belonging to a lesion’s mask in the MNI space. Contralateral regions were discarded when contaminated by white matter lesions or other structures like ventricles or gray matter. The inverted transformation matrix was applied to move the contralateral masks from the MNI space back to the native subject space. Additionally, all lesions and their contralateral masks were moved from MNI template space to individual MR image subject space of healthy controls in order to sample intact white matter with regions of interest (ROI) of the same size and shape as corresponding lesions. The final ROIs were further visually inspected by a board-certified neurologist (NMM) to ensure they accurately represented controls for the lesion masks.

Myelin water fraction (MWF) maps were created using three-pool Multi-Component Driven Equilibrium Single Shot Observation of T1 and T2 (mcDESPOT) as previously described [37, 39]. The method calculates a three-pool water model characterizing water trapped in myelin sheath, extra- and intra-cellular water, and water in the cerebral spinal fluid for correction of partial volume effect in the gray matter. The MWF is estimated as the ratio of water trapped in myelin sheath to the total water and computed using a home-made MATLAB script running parallel computing for efficient reconstruction.

Rigid registration between PET and MRI was performed using ANTs software (6 parameters) [40], with the cost function driven by the Mutual Information metric. The same transformation was applied to all previously generated masks, yielding masks in the PET space for calculating standardized uptake value (SUV) and generating time-activity curves. All SUV reported in this work are standardized by the lean body mass [41] to account for the wide difference in enrolled subjects’ body mass index.

Arterial blood sampling

An arterial line was placed in the radial artery by a licensed anesthesiologist under ultrasound guidance. Arterial blood samples of 1–3 mL were drawn every 30 s immediately following radiotracer injection and decreased in frequency to every 30 min towards the end of the scan. [¹⁸F]3F4AP metabolism was measured from blood samples acquired at 5, 10, 15, 30, 60, 90, and 120 min.

Arterial blood processing and radiometabolite analysis

Radioactivity concentration (in kBq/cc) was measured in whole-blood (WB) and subsequently in plasma (PL) following the centrifugation of WB (1 min at 1,000 g) as previously described [23]. PL samples for radiometabolite analysis were filtered using a 10 kDa centrifugal filter (10 min at 15,000 g) and analyzed using an Agilent HPLC equipped with a coincidence detector as previously reported [42]. HPLC fractions were collected and measured on an automated gamma counter.

Plasma free fraction determination

Plasma free fraction was determined as previously described [23]. Briefly, arterial blood samples drawn before radiotracer injection were centrifuged to separate the plasma. Plasma samples were spiked with [¹⁸F]3F4AP, incubated for 15 min, loaded into 10 kDa centrifugal filters (Millipore Centrifree) and centrifuged at 1,500 g for 15 min at room temperature. Plasma free fraction was calculated as the ratio of radioactivity in the filtrate to the radioactivity in plasma corrected for any radioactivity bound to the filter.

Quantitative analysis of [¹⁸F]3F4AP brain uptake

Regional time-activity curves (TACs) were analyzed by compartmental modeling using the metabolite-corrected arterial plasma input function. One- (1T) and two- (2T) tissue compartment model configurations were investigated while fixing the vascular contribution of the whole-blood (WB) radioactivity to the PET measurements to 5%, as well as using the vascular contribution as a parameter. The 2 T model was also tested in its reversible and irreversible modes. Estimates of the kinetic parameters were obtained using nonlinear weighted least-squares fitting with the weights defined as the frame durations. The regional total volume of distribution (V_T) was calculated as K₁/k₂ for a 1 T compartment model and as  for a 2 T model [43]. In addition, the Logan graphical method [44] for estimation of V_T was performed.

Statistical analysis

Statistical analysis of in vivo PET was performed using MATLAB R2021b, The MathWorks Inc., Natick, Massachusetts. Descriptive statistics including mean and standard deviation (SD) were calculated for each group. Two-group t-tests with a significance level of α = 0.05 were used to assess differences among groups. Grouped data are reported as mean ± SD.

The small-sample corrected Akaike information criterion [45] (AICc) was used to compare the goodness of fit of the considered compartmental models for kinetic modeling. AICc weights were computed for each TAC (white matter, gray matter, and deep gray matter, for all subjects, as well as three representative lesions for each participant with MS) and then summed. The model with the highest sum was deemed the best fit to our data and was used for computing kinetic parameters for group comparisons. The coefficient of determination, R², was used to assess the goodness of fit of linear regression between the compartmental model V_T and simplified methods, and between myelin water fraction quantification and SUV_{76−106 min}. For all linear regression models, we reported the p-value from the F-statistic to assess whether the predictors significantly improve the model compared to an intercept-only model.

Results

Participants characteristics

Three people with MS of different races and ethnic backgrounds and with varying levels of disease severity and duration participated in the study. Two of the participants had relapsing remitting MS (RRMS) and one had secondary progressive MS (SPMS), representing the most common forms of the disease [4]. RRMS is characterized by periods of disease activity resulting in new or worsening episodic symptoms (clinical relapses) followed by periods of partial or complete resolution (remission). SPMS is characterized by a history of relapses with a more recent clinical course of slow worsening of disability over time (clinical progression). All participants were receiving B-cell depleting therapy at the time of imaging. We recruited three healthy volunteers, two of whom were age- and sex-matched to the participants with MS. Participant characteristics, including injected dose, molar activity, sex, age range, body weight, body mass index (BMI), concurrent treatment with 4-AP (yes/no), MS disease subtype, MS disease duration by time of PET, MS disease-modifying therapy at time of PET, and Expanded Disability Status Scale (EDSS) scores are listed in Table 1. All subjects tolerated the study procedures well and there were no adverse events.

Table 1.

Participants characteristics

	HC001	HC002	HC003	MS001	MS002	MS003
Sex	Female	Female	Male	Male	Female	Male
Age range (years)	31–35	46–50	41–45	56–60	46–50	41–45
Weight (kg)	70.3	50.3	126.1	80.7	81.2	86.2
Race/Ethnicity	White	White	White	White	Black	Hispanic
Injected dose (MBq)	285	312	264	377	312	264
Molar activity (GBq/µmol)	167	96.2	418	414	40.3	65.1
BMI (kg/m2)	22.9	19.7	35.7	24.1	29.8	29.8
EDSS	NA	NA	NA	2.5	6.0 (walks with cane)	6.5 (walks with bilateral assistance)
4-AP use? (yes/no)	NA	NA	NA	No	Yes	Yes
MS subtype	NA	NA	NA	Relapsing-Remitting	Relapsing-Remitting	Secondary Progressive
MS disease duration by Time of PET (years)	NA	NA	NA	1.5	5	14
MS treatment at time of PET	NA	NA	NA	Ocrelizumab	Ocrelizumab	Ocrelizumab

BMI Body Mass Index, EDSS Expanded Disability Status Scale., 4-AP 4-aminopyridine, MS multiple sclerosis, NA not applicable (characteristics specific to MS are not applicable to healthy participants)

[¹⁸F]3F4AP displays suitable properties for imaging in the human brain

Two-hour dynamic PET scans with arterial blood sampling were performed in three healthy volunteers and three people with MS.

As shown in Fig. 1, [¹⁸F]3F4AP showed a rapid entry into the brain and a fast washout with slightly higher accumulation in the gray matter than in the white matter. All participants with MS displayed higher uptake across the brain as compared to healthy controls illustrated by the higher PET signal in the later frames (Fig. 1a). The time-activity curves (TACs) in the cortex, deep gray matter, and white matter are shown in two age- and sex-matched subjects: HC002 (Fig. 1b) and MS002 (Fig. 1c). In the case of MS002, the TACs of two white matter lesions are also shown demonstrating differences in tracer delivery and binding across lesions.

Fig. 1.

[¹⁸F]3F4AP for imaging the human brain. (a): Series of dynamic PET images for three healthy control subjects (HC001, HC002, and HC003) and three people with MS (MS001, MS002, and MS003). (b): Time activity curves for HC002 in three brain regions. A zoom on the interval [80–115 min] is shown. (c): Time activity curves for MS002 (age- and sex- matched to HC002) in three brain regions and two representative white matter lesions. A zoom on the interval [80–115 min] is shown

Kinetic modeling of [¹⁸F]3F4AP shows high brain penetrance and greater binding in multiple sclerosis

The radioactivity concentration in whole blood also showed a fast washout and was highly consistent across subjects (Fig. 2a). Consistent with prior studies in monkeys [23], the tracer showed rapid and uniform distribution between blood cells and plasma with a whole blood to plasma ratio of 0.9 after 15 min (Supplementary Fig. 1). Analysis of the plasma protein binding and metabolism showed minimal plasma protein binding (plasma free fraction 88.7 ± 2.1%, n = 2) and 40 ± 6% (n = 5) parent compound remaining after 30 min with little variability across subjects (Fig. 2b).

Fig. 2.

Quantification of [¹⁸F]3F4AP via kinetic modeling. (a): Mean SUV lean body mass of whole blood across all subjects with corresponding standard deviation. (b): Mean plasma parent fraction (%) across all subjects with corresponding standard deviation. (c): Delivery rate (K₁) in the white matter, gray matter, and deep gray matter for healthy controls (HC) and people with MS. (d): Volume of distribution (V_T) in the white matter, gray matter, and deep gray matter for healthy controls and people with MS. Asterix (*) indicates a significant difference between HC and MS means with p < 0.05

Using the metabolite-corrected plasma concentration as input function, we analyzed the TACs in different brain regions including the cortex, deep gray matter, white matter and, in the case of participants with MS, representative white matter lesions using compartmental modeling. An example of brain region masks in one subject is shown in Supplementary Fig. 2. A reversible two-tissue compartmental model with variable vascular fraction produced excellent fits in all subjects and all regions assessed by visual inspection (Fig. 1b and c), and Akaike information criterion (Supplementary Table 2). The K₁ values obtained from the model were 0.69 ± 0.09 mL·min⁻¹·cm⁻³ for the cortex, 0.65 ± 0.10 mL·min⁻¹·cm⁻³ in the deep gray matter and 0.27 ± 0.05 mL· min⁻¹·cm⁻³ for the white matter (Fig. 2c), reflecting very high plasma to tissue extraction. There were no statistically significant differences in tracer delivery (K₁) between healthy controls and people with MS.

The V_T values obtained from the model ranged from 1.6 mL·cm⁻³ for the cortex, 1.5 mL·cm⁻³ for the deep gray matter and 1.2 mL·cm⁻³ for the white matter in healthy controls (Fig. 2d). In people with MS, V_T values were about 12% higher in the cortex and deep gray matter (p < 0.05) and 17% higher in the NAWM compared to controls (p < 0.05) (Supplementary Table 3). Among MS participants, the two people with relapsing remitting disease showed greater tracer binding than the person with progressive disease despite their different EDSS scores.

Simplified analysis methods such as Logan or late SUV produce results consistent with V_t.

Logan plots using plasma concentration as input function linearized well after a t* of 30 min and produced V_T values consistent with those obtained via kinetic modeling (Fig. 3a and Supplementary Fig. 3).

Fig. 3.

Simplified analysis methods for [¹⁸F]3F4AP imaging. (a): Correlation plot between V_T from 2-tissue compartmental model and V_T from Logan graphical analysis. (b): Correlation plots between V_T from 2-tissue compartmental model and late SUV lean body mass in healthy controls and MS subjects. Investigated regions include the white matter, gray matter, and deep gray matter for all subjects, as well as three representative lesions per participant with MS

We also evaluated the correlation between late SUVs with V_T and found SUV per lean body mass from 76 to 106 min values showed a positive correlation with V_T (R² = 0.35, p < 0.01). The correlations were stronger when healthy controls (R² = 0.92, p < 0.001) and participants with MS (R² = 0.59, p < 0.001) were assessed separately (Fig. 3b). The differences in gray matter, deep gray matter, and white matter, between the two groups were also stronger when using SUV_{76−106 min} (Supplementary Fig. 4) as opposed to V_t (Fig. 2d).

Even though we were not able to assess intra-subject variability or test/retest variability of each method due to the lack of repeat scans in any of the participants, the inter-subject variability of two paired subjects was found to be small (Supplementary Table 4).

Multiple sclerosis lesions display heterogeneity in PET not visible by MRI

From the late SUV images, it was apparent that [¹⁸F]3F4AP shows heterogenous binding across lesions with some lesions showing very high uptake. To understand lesion heterogeneity, contralateral ROIs were created in the NAWM for each lesion whenever possible. Based on this analysis, 9 out of 10 lesions from MS001 were paired with ROIs sampled in the contralateral NAWM. For MS002 and MS003, these numbers were 17 out of 19, and 28 out of 34 lesions, respectively.

The simplified SUV method is valid for within-subject comparison considering the good correlation between late SUV and V_T within each group (Fig. 3b and Supplementary Table 3). In Fig. 4, we performed our lesions and contralateral ROIs analysis using late SUV. When comparing individual lesions to their contralateral ROI, it was clear that a small subset of lesions (6 out of 9 in MS001, 5 out of 17 in MS002, and 3 out of 28 in MS003) had greater tracer binding (SUV lesion > 1.1x contralateral ROI) (Fig. 4a). A small subset of lesions (1 in MS001, 1 in MS002, and 8 in MS003) had lower tracer binding (SUV lesions < 0.95x contralateral). The remaining lesions (2 in MS001, 11 in MS002, and 17 in MS003) had similar SUVs as their contralateral ROIs. This pairwise comparison suggests that lesions classified as “high” may contain a high concentration of spared demyelinated axons, “medium” may indicate normal levels of Kv channels due to the axons having been remyelinated or there being very low numbers of demyelinated axons, and “low” may indicate axonal loss. We then looked at whether lesions with high, medium, and low PET uptake appeared visually different on T1w or T2w MR images and found no clear differences (Fig. 4b and Supplementary Fig. 5). The high PET lesions do not correspond to acute lesions visible on MRI by gadolinium enhancement, since no Gd-enhancing lesions were present in any of the participants (Supplementary Fig. 6).

Fig. 4.

[¹⁸F]3F4AP for imaging multiple sclerosis lesions. (a): Comparison between lesions and their paired contralateral ROIs sampled in NAWM in three people with MS. (b): Selected lesions shown on MPRAGE and FLAIR sequences. Zoomed-in images of the lesions on MPRAGE and FLAIR sequences, and their corresponding late PET images are shown for each subject

Finally, we compared the quantification of MWF from MWI with the quantification of the PET. Figure 5a shows myelin water fraction, SUV images, and Logan V_T maps from all participants (Sagittal views in Supplementary Fig. 7). We looked at the ROIs sampled in the white matter of the healthy controls group as a proxy of MS lesions or their contralateral controls. Based on 63 lesions and their 54 contralateral masks, we obtained 115, 116, and 114 ROIs for HC001, HC002, and HC003, respectively. A modest inverse correlation (R² = 0.13; p < 0.001) was observed within the ROIs in healthy controls, indicating that myelin content and PET tracer binding were inversely related (Fig. 5b). In people with MS, however, no correlation (p > 0.05) was observed (Fig. 5c, Supplementary Fig. 8, and Supplementary Fig. 9). The absence of correlation between MWF and PET across all subjects suggests that myelin content and concentration of demyelinated axons are not simply inverse metrics of each other.

Fig. 5.

Comparison of [¹⁸F]3F4AP PET findings and myelin water imaging. (a): Myelin water fraction (MWF), late SUV (76–106 min) images, and parametric Logan V_T maps in all participants. (b): Correlation between SUV and MWF in ROIs sampled in the white matter of healthy controls. (c): Correlation between SUV and MWF in MS lesions and contralateral ROIs in the NAWM

Discussion

MS is a heterogeneous disease in which focal demyelinated lesions appear in the CNS as a consequence of autoimmune attacks on the myelin sheath. MRI has been and continues to be an invaluable imaging tool for MS. MRI, which relies on measuring changes in spin-relaxation times of water protons, provides excellent contrast between lipid rich areas (myelinated white matter) and lipid poor areas (gray matter and demyelinated lesions). However, conventional MRI cannot readily distinguish if viable axons are present in demyelinated lesions. Furthermore, MWF and other myelin measurements show low myelin content across lesions, which makes it difficult to assess which lesions may be functionally remyelinated or whether there are spared axons in demyelinated lesions. PET, on the other hand, uses radiolabeled small molecules that specifically bind to proteins or to other biological structures and can provide information on these concentrations. The latest approaches to PET imaging in MS include TSPO and S1PR1 tracers, as well as [¹¹C]MeDAS, which provide information about neuroinflammation and myelin density [6, 10, 46]. In this work, we conducted the first evaluation of a PET tracer targeting voltage-gated K⁺ channels, which are upregulated in demyelinated axons, to image the human brain and to better understand MS.

The PET images produced by the scanner show very high tracer delivery into the brain with little heterogeneity across non-lesioned brain regions. Interestingly, higher PET signal was observed in the brains of people with MS than controls suggesting greater concentration of accessible K⁺ channels in people with MS. This was particularly noticeable in the NAWM, with MS subjects showing around a 17% higher PET tracer binding (quantified by V_T) or 64% higher PET tracer uptake (quantified by late SUV), suggestive that [¹⁸F]3F4AP is sensitive to widespread pathology in MS beyond the white matter lesions. The binding of the tracer to GM, WM and lesions could be accurately quantified using a 2-tissue compartment model as well as using Logan graphical analysis, a simpler graphical analysis method. Furthermore, late SUV images appeared to correlate well with the quantitative measures of binding V_T within each group, suggesting that the simplified SUV measure is sufficient for comparison within diagnosis, in addition to being well suited for within-subject analysis. SUV is a promising simplified quantification measure that, after confirmation in a greater number of subjects, may eliminate the need for collecting a blood input function.

The relatively low variability across subjects suggests that this tracer meets the requirements to become a robust biomarker. While the results in humans were largely consistent with prior results in preclinical animal studies, we observed some interesting differences. For example, greater metabolism was observed in humans than in monkeys, which we attribute to the effect of anesthesia since monkeys were scanned anesthetized and humans were not. This finding is supported by recent in vitro and rodent studies [42, 47] showing that awake mice do not metabolize the tracer as much as anesthetized mice and that both isoflurane and [¹⁸F]3F4AP are metabolized by the cytochrome P450 enzyme CYP2E1 [42, 47]. Despite the faster breakdown of the tracer in humans, this is not a concern given that the parent fraction in plasma (~ 40% after 30 min) was high compared to other tracers (parent fractions for [¹¹C]PiB and [¹¹C]MeDAS were ~ 15% and ~ 10% after 30 min) [10, 48], that the metabolites (Supplementary Fig. 10) do not enter the brain [47], and that the degree of metabolic breakdown was consistent across subjects.

A salient feature of [¹⁸F]3F4AP was the very high initial/first pass signal in the gray matter suggesting a very high extraction fraction (K₁ ranges from 0.22 to 0.82 mL·min⁻¹·cm⁻³). ¹⁸F-labeled tracers with such high extraction fraction are rare [49] and it suggests that [¹⁸F]3F4AP may provide a means to measure blood perfusion throughout the brain including in the white matter. This would be an important application given that [¹⁵O]H₂O is difficult to obtain (t_1/2 = 2.04 min) and other methods such as dynamic susceptibility contrast (DSC) perfusion-weighted imaging (DSC-MRI) have limited sensitivity and resolution [50]. Further efforts are warranted to explore and validate this application.

By quantifying the binding to the gray matter and white matter, we observed a higher signal in the NAWM and the gray matter of MS subjects than controls both using quantitative binding measures such as V_T and semiquantitative measures such as late SUV. This finding suggests that [¹⁸F]3F4AP is able to capture widespread pathology in the brain. Interestingly, the participants with RRMS showed greater signal than the one with SPMS despite their difference in EDSS score. Further validation of this finding in a larger number of subjects, as well as exploring tissue reference quantification methods, are important next steps to enable the wide adoption of this tracer.

Despite the limitation of having a small number of subjects (3 healthy controls, and 3 people with MS), we opted to report the results from this small cohort to address the growing interest in this tracer and demonstrate that [¹⁸F]3F4AP has robust and reproducible properties. We demonstrated higher uptake of this tracer in participants with differing subtypes of MS, including both RRMS and PMS, which may provide insight into comparative remyelination potential and disease pathology. Although a true test/retest in the same subject is needed to truly assess the reproducibility, the low variability across subjects found here, combined with the < 7% test-retest variability found in NHPs [23], suggests that this tracer will be highly reproducible. An important limitation was the lack of Gd-enhancing lesions within the small cohort, which prevented us from assessing the model fit in this type of lesions and the correlation of the PET signal with BBB damage [51, 52]. Other limitations include the limited exploration against state-of-the-art MRI advances in MS. For example, we did not evaluate correlations with paramagnetic rim lesions [53, 54], gray matter lesions [55] or diffusion MRI (DTI, NODDI or other multi-compartment models) [56, 57], which may provide information about persistent inflammation [58], cognitive symptoms [59] and axonal integrity [60, 61], respectively [62]. We also did not attempt to categorize lesions based on the size on T1w and T2w, which have been suggested to provide information about demyelination and remyelination [10]. These will need to be addressed in the future. Finally, two of the three MS participants were on dalfampridine (4-aminopyridine, 4AP) prior to participation in the study. While previous studies in monkeys showed that 4AP does not block the tracer binding, and instead may increase tracer binding due to compensatory mechanisms in the availability of K⁺ channels [23], this will need to be further investigated in future studies in humans.

Perhaps the most remarkable finding of our study which, if confirmed, could change the way we understand and monitor the heterogeneity of lesions in MS, was the different binding of the tracer across MS lesions. Around 26% of lesions showed very high binding and around 19% of lesions showed very low binding in pairwise comparisons against contralateral ROIs in the NAWM. Interestingly, the people with RRMS (MS001 and MS002) showed a high number of lesions with high tracer binding whereas the person with PMS (MS003) showed a high number of lesions with low tracer binding. This finding is consistent with a high proportion of lesions in RRMS containing spared demyelinated axons and a high proportion of lesions in PMS having undergone axonal loss. Importantly, this lesional heterogeneity was not seen by MRI. Comparing the quantitative MRI measure of MWF with [¹⁸F]3F4AP PET, we found a weak inverse correlation in healthy volunteers but not in people with MS, suggesting that reductions in myelin fraction do not correlate with the concentration of demyelinated axons. This unique distinction from MRI makes [¹⁸F]3F4AP already a promising tool for MS imaging. Our interpretation of this finding is that lesions containing a high percentage of spared demyelinated axons show high tracer binding, lesions with extensive axonal loss or necrosis show a lower-than-baseline binding level and lesions with binding levels similar or slightly higher than NAWM are those with a low level of demyelination or partial remyelination. This finding is very intriguing as it may provide a means to select candidates for remyelinating treatments and monitoring for the effect of those treatments. Even though non-neuronal cells in the brain also express certain Kv channels, we favor the hypothesis of axonal Kv channels are the primary target since these are the most abundant [63] and the only ones known to increase upon demyelination [12–15]. It is important to replicate this finding in larger cohorts as well as through histopathology examination in postmortem MS brains or postmortem studies in animal models that recapitulate lesional heterogeneity such as the EAE (experimental autoimmune encephalomyelitis) marmoset model [64]. Finally, future studies comparing [¹⁸F]3F4AP with other PET tracers will be instrumental to gain a full understanding of the tracer’s specificity to demyelinated axons and its utility in MS.

Electronic supplementary material

Below is the link to the electronic supplementary material.

Author contributions

Conception and design: Amal Tiss, Nara M. Michaelson, Brian Popko, Eric C. Klawiter, and Pedro Brugarolas.

Acquisition and data analysis: Amal Tiss, Nara M. Michaelson, Andrew W. Russo, Karla M. Ramos-Torres, Yang Sun, Nicole E. DaSilva, Jacqueline M. Noel, Fang Liu, Kuang Gong, Susie Y. Huang, Suzanne Baker, Eric C. Klawiter, and Pedro Brugarolas.

Drafting: Amal Tiss, Nara M. Michaelson, Suzanne Baker, Eric C. Klawiter, and Pedro Brugarolas.

Review and approval of final manuscript: ALL.

Funding

This work was supported by the National Institutes of Health (K99/R00EB020075, R01NS114066 - PB), an Innovation Fund Award from the Polsky Center for Innovation and Entrepreneurship at The University of Chicago (PB, BP), Massachusetts General Hospital Executive Committee on Research Physician Scientist Development Award (KMRT), and Translational Neuroscience Innovation/Anne B. Young Fellowship (NMM).

Data availability

The code and data used in this work will be made available upon reasonable request to the corresponding authors, contingent on the obtention of a formal data/code sharing agreement.

Declarations

Competing interests

Pedro Brugarolas and Brian Popko are named inventors on patents related to [¹⁸F]3F4AP owned by the University of Chicago. Dr. Brugarolas’ interests were reviewed and are managed by Massachusetts General Hospital (MGH) and Mass General Brigham in accordance with their conflict-of-interest policies. All other authors have no potential conflicts of interest to report.

Ethics approval

Studies in human research subjects were performed in line with the principles of the Declaration of Helsinki. Prior approval was granted by the Institutional Review Board (IRB) at the Massachusetts General Hospital (IRB# 2020P002459, PI: Brugarolas). The study was registered in ClinicalTrials.gov (NCT04699747, Sponsor: Massachusetts General Hospital, Responsible Party: P. Brugarolas). [¹⁸F]3F4AP was administered under an investigational new drug (IND) authorization from the U.S. Food and Drug Administration (FDA) (IND # 135,532, Sponsor: Brugarolas).

Consent to participate

Informed consent was obtained from all individual participants included in the study.